However, the title of this post could also refer to the recent migration wave that has driven over a million Cubans to leave in the last few years, sparked by the severe economic crisis on the island. Yet this latest wave, with its “last one out, please turn off the lights“ feel, does not look like an exodus of the upper classes. It should remind us, though, that economic crises are often the root cause of major political upheavals and migration waves.

Let’s take a closer look at the sixties wave then, but perhaps a different century.

During the reign of Queen Isabella II, between the years 1866 and 1868, the Kingdom of Spain suffered a severe economic crisis that prepared the ground for the major political shift that followed2:

…the trigger for the 1866 financial crisis was the losses suffered by railway companies, which dragged down banks and credit societies with them… The financial crisis of 1866 was compounded by a severe subsistence crisis in 1867 and 1868, caused by poor harvests during those years. Those affected were not businessmen or politicians, as in the financial crisis, but the working classes, due to the scarcity and high cost of basic goods such as bread. Popular riots broke out … The subsistence crisis was exacerbated by the rise in unemployment caused by the economic crisis triggered by the financial crisis

At the time, the Spanish Kingdom was organized as a constitutional monarchy with a functioning parliament and regular elections, although electoral practices were a far cry from what we now consider democratic standards. In the years leading up to 1868, this already imperfect parliamentary monarchy had become increasingly authoritarian, persuading even the monarchist opposition leaders that maybe Spain should remain a monarchy, but with a different monarch.

On September 18, the Spanish Navy fleet in Cádiz revolted. This was hardly surprising given that the Spanish military had been regularly intervening in politics since the end of the Napoleonic Wars. Soon, revolutionary councils were formed in most major cities across Spain, reflecting the popular support for the insurrection, and after the loyalist army suffered a cleat but indecisive defeat on September 28, Queen Isabella went into exile, clearing the way for opposition leaders (and generals) to seize power and organize a new government.

News of the successful uprising soon reached Cuba, Spain’s sugar-producing colony on the other side of the Atlantic, where independence-minded members of the Cuban elite had already been plotting rebellion against Spain for some time.

With Spanish leaders busy organizing a new political system and dealing with the demands of a restless population, this was the perfect time to revolt. One of the loudest and most often repeated demands of the Spanish lower classes was the end of forced conscription (quintas). If approved - as Cuban rebels probably hoped - it would have drastically reduced Spain’s chances of suppressing the Cuban rebellion by severely limiting the number of troops that could be sent to the island. Encouraged by the events taking place in Spain, the Cuban conspirators launched their own uprising on October 10, 1868, near the town of Manzanillo in eastern Cuba3, followed by a second uprising on November 4 near Camagüey in central Cuba.

The rebels published a Manifesto of the Revolutionary Council of the Island of Cuba and convened a constitutional assembly in the town of Guáimaro with the goal of giving their revolutionary republic a formal organizational framework, while simultaneously facing the Spanish army in a series of skirmishes. Soon, large areas of eastern Cuba began to suffer the effects of the guerrilla tactics employed by the rebels, including the burning of many sugar mills and other economic infrastructure4:

[General Fernández-Cavada Howard] spawned the idea of the burning torch to destroy the properties of landowners who supported Spain, and he systematically applied this method to economically weaken the forces opposed to the freedom of Cuba, which led to him being known by the nickname “General Fire” (General Candela).

The Spanish successfully contained the rebellion to the eastern half of the island by building a fortified defensive line across its center, blocking the highly mobile rebel forces from invading the prosperous western half of Cuba, where Havana is located. Yet the huge amount of resources committed to the building of this barrier, and the vast additional human and material resources devoted to suppressing the rebellion, couldn’t prevent the widespread devastation that eastern Cuba would suffer during the appropriately named Ten Years’ War (1868-1878). When rebel forces couldn’t hold the cities they captured (liberated from their point of view), they sometimes burned them to the ground5:

Three days after the event, the embers were still smoldering. The Spaniards couldn’t believe it, even though they had already seen the gigantic fire, which foretold the terrible outcome: they would find nothing in Bayamo, only ashes…Why did they decide to set fire to such a beautiful city? Since October 20, 1868, Bayamo had been experiencing the fever of a nascent Revolution, the same one that ignited a hymn, a plaza, a City Hall… after more than 80 days without foreign tutelage, [the city] the concept of freedom ingrained in their hearts.

After burning down Bayamo in 1868, the rebels went on to burn the town of Guáimaro (1869) and the city of Las Tunas (1876).

All this burning and destruction eventually persuaded many Cubans to leave the island in search of greener, more peaceful pastures; many others left to escape reprisals by the Spanish authorities against those who supported the rebellion, with the majority heading to Jamaica, the United States, and the Dominican Republic. A large share of these migrants belonged to the upper classes of Cuban society, as recorded by a contemporaneous Jamaican newspaper6:

The French Transatlantic Steamer ‘Tampico,’ in one day from Santiago de Cuba, arrived here yesterday with upwards of one hundred passengers, among them many ladies and gentlemen, who are fleeing to a place of safety ere the worst comes about in a serious crisis which appears to be looming up, if indeed, it has not already begun, in the island of Cuba.

Many others did not7:

The schooner Almogabar in six days from Manzanillo de Cuba, arrived here [at Kingston, Jamaica] on Tuesday evening with thirty passengers, chiefly small farmers, who have been compelled to abandon their country in consequence of the insecurity of life and property.

The social background of Cuban migrants depended partly on the economic opportunities available at each destination. For example, the picturesque American city of Key West, Florida, attracted many tobacco workers, drawn by the establishment of several cigar factories in the island city. It would be inaccurate to describe these workers as unskilled though; cigar makers were among the best paid laborers in Cuba and the U.S.8 at the time (and likely still are today). And most founders of these cigar factories were also Cuban migrants, who took advantage of Key West’s favorable conditions: they imported high-quality tobacco leaf from nearby Cuba, and then produced cigars for sale tariff-free in the American market.

So Florida got Cuban human capital, financial capital, Cuban know-how, and an entire new industry, while Cuba got a revolution. Sounds familiar.

Key West was not the only major destination of Cubans fleeing to the United States. The 1870 U.S. census, taken before the peak of Cuban migration, records around 12,000 Cuban-born residents: 4,500 in New York, 3,000 in New Orleans, and 2,000 in Key West. I’ll come back to these numbers later.

Meanwhile, in the Dominican Republic, a small Cuban community of no more than 1,200 immigrants9 had settled in the northern port of Puerto Plata by 1875. As I mentioned in my post on Haiti and the Dominican Republic, the Dominican sugar industry was in its infancy at the time, and these Cuban refugees played an important role in launching Dominican sugar production. They did not limit themselves to sugar, though, becoming involved in many other enterprises10.

There were several Cuban establishments [in Puerto Plata]: the sugar mills of Carlos F. Loynaz11; the Camagüeyan Pastry Shop, owned by Cirilo Recio; the La Habana Restaurant, owned by Lino Castro; the Dominican-Cuban Hotel Unión de Cuba, later Las dos Repúblicas, owned by Francisco Moroní; the saddlery shop of Juan Anido; the photographic gallery of Narciso Arteaga; the tailor shop of Manuel Miranda; the bookstore, an agency of the one Néstor Ponce de León had in New York, was managed by Fernando Cisneros y Correa.

In Jamaica, highly skilled12 Cuban emigrants fleeing from the Ten Years’ War also gave a big push to certain economic areas of the smaller island. Just like in Florida, they revolutionized Jamaica’s tobacco industry, while also dominating less capital-intensive trades such as tailoring and barbering13. And just as they did in the U.S. and the Dominican Republic, many (though not the majority) remained in Jaimaica after the war in Cuba had ended14.

That these Cuban refugees were not a random sample of the Cuban population seems to be confirmed by their names. Howard Johnson, whose “Cuban Immigrants in Jamaica, 1868–1898“ is one of the main sources for this post, identifies by name just over 40 Cuban migrants. Among them, six had French surnames, one was of Scottish origin, and another (Englishman) of Jewish origin. Most Cubans of French origin descended from white French colonists who had left Haiti after the Haitian Revolution and found refuge in Cuba, where they generally prospered and integrated into the upper classes of Cuban society. Their overrepresentation in Johnson’s sample further confirms the upper-class skew of Cuban emigrants.

The Cuban community in Jamaica was somewhat larger than the one in the Dominican Republic, reaching around 1,500-2,000 at its peak15. Combining this number with the ones recorded for the U.S., we can estimate that the total number of Cuban emigrants was 14,000 at the very least. This minimum does not include emigrants to other countries and makes no adjustment for the increase in U.S. residents between the 1870 census and the peak of migration in 1872-1873.

The true number is substantially higher than the 14,000 minimum, but far lower than the unsubstantiated 100,000 figure cited by some historians16. We should also consider one estimate for the number of Cubans outside the island on the eve of the 1895 Independence War17, which puts the figure at 40,000-50,000, and the number of Cubans residing in the United States according to the 1880 census18 (after the peak). Taking everything into account, my best guess for the peak of emigration during the Ten Years’ War would be 20,000 to 30,000 refugees, or about 1.4% to 2% of Cuba’s population at the time. And given what we have seen about their background, these emigrants accounted for a far higher share of Cuba’s most skilled population.

Although the 1868-1878 Cuban exodus involved only a small share of the island’s total population, it was not an insignificant brain drain. It deprived Cuba of many of its most talented and enterprising people and, more importantly, it was the first of many brain drains.

But that’s not the point. The point is that Cubans like baseball, and they are very good at it. As many as 30 Cubans are currently playing in the American Major League Baseball (MLB), many of them after departing the island illegally.

More than a million African slaves were imported into Cuba before Spain finally abolished the institution in the 1880s, leaving a heavy legacy that still weighs on Cuban society until this day. Partly due to that legacy, baseball and other sports serve a greater purpose in Cuban society than just keeping Cubans entertained. In contrast to competition in areas like politics and business, sports provide a public stage where Afro-Cubans outperform other Cubans, often holding the top national positions. Star MLB players like Yordan Álvarez, Yennier Canó, Randy Arozarena and Aroldis Chapman illustrate this point.

Well, in the cases of Álvarez and Chapman, their names also illustrate another aspect of Afro-Cuban history. Álvarez’s full name, following Spanish naming customs, is Yordan Rubén Álvarez Cadogan. Rubén is his second given name while Cadogan is his maternal surname, inherited from his mother’s family (Chapman’s maternal surname is De la Cruz). What are the odds that two prominent Black MLB players from a Spanish-speaking country would have British surnames? Turns out, not as low as you might think. Maybe close to the odds of finding two British-surnamed ministers among the very few Afro-Cubans who serve on Cuba’s Council of Ministers1.

During the late 19th and early 20th centuries, Cuba became a major destination for Spanish migrants. These migrants were generally unwilling to work in the island’s sugar fields, so after the banning of the slave trade, and especially after slavery itself was abolished in 18862, the sugar industry began actively looking for new sources of cheap labor to complement, and later partly replace, the island’s enslaved workforce.

The first source that Cuban sugar producers drew upon was China. Starting in 1847, more than 100,000 Chinese indentured workers3 were recruited in Macao to cut cane in the sugar fields. This flow lasted until 1974, when the Chinese government sent a fact-finding commission to investigate the living conditions of Chinese laborers in Cuba. What they found in Cuba’s sugar plantations was widespread abuse and brutal working conditions, and their report directly led to the end of Chinese migration to Cuba.

The rapid expansion of sugar production in the 1910s, brought about by a wave of American investment following Cuba’s independence, reignited the search for cheap labor. This time, sugar workers came from nearby lands: the neighboring Caribbean islands of Jamaica and Haiti. Between the years 1913 and 1921, around 75,000 Haitians and 76,000 Jamaicans entered Cuba legally4. They continued to arrive in Cuba in large numbers, both legally and otherwise, until the Great Depression caused sugar prices to collapse in 19305.

Not all of these Caribbean immigrants remained in Cuba. Some of them worked only seasonal jobs, returning to their home islands regularly. Many stayed permanently and integrated into Cuban society, as you can see by the West Indies surnames - more common in the British West Indies than in the UK6 - carried by their prominent descendants. Among West Indian immigrants, Jamaicans in particular appear to have achieved relative economic success. At the time of their arrival, the Jamaican per capita income was half that of Cuba, yet by the time of the Cuban Revolution average Jamaican incomes had overtaken those in Cuba (and yes, that’s another piece of evidence against the myth that Cuba ranked “near the top“ in terms of development).

Then there are the Spanish migrants. As I’ve mentioned before, they tended to avoid work in the sugar mills and fields, instead preferring to try their hand at commerce or small scale agriculture. We can learn more about these Spanish migrants too, by examining the surnames of Afro-Cubans who have excelled at the top tier of baseball (nobody is perfect), and maybe other fields I actually care about.

I’ve already droned on and on about the large share of Canarian migrants among Spaniards who came to Cuba. Nonetheless, other Spanish regions also contributed large numbers of migrants to Cuba, particularly Galicians and Asturians during the late 19th and early 20th century.

The following bar chart shows the percentage of Spanish nationals residing in Cuba in the year 2010 by region of birth. Mind you, this is not a perfectly representative sample of 20th century Spanish migration to Cuba, being obviously skewed towards the 1940s and 1950s rather than the more significant early 20th century, and likely including a substantial number of young expats.

A more accurate breakdown of Spanish regional ancestry in Cuba can be obtained by a surname analysis of the current Cuban population, assigning each of the 350 most common Cuban surnames to the Spanish region it originated from7. This is the same methodology I used in a previous post, and it groups Spain’s official autonomous communities into larger regions for easier comparison (e.g. Cantabria-Rioja-PV-NV corresponds to four northern autonomous communities: Cantabria, La Rioja, Basque Country and Navarre).

According to this analysis, around two thirds of Cuban Spanish ancestry comes from the Canary Islands, Galicia and Asturias, with most of the remainder originating in Andalusia and Castile. No surprises there.

Let’s now repeat the analysis with a sample of surnames from Afro-Cubans. Since I don’t have a list of the most common surnames among Afro-Cubans, I’ll just use the surnames of the most successful Afro-Cuban athletes (according to Wikipedia) in two sports where Cuba stands out: baseball and boxing8.

This is a very rough analysis9, but despite its limitations, it clearly reveals major differences in the regional distribution of Spanish surnames between the general Cuban population and Afro-Cubans:

• Among Afro-Cubans, the highest share of surnames (27.6%) originates from the region that includes Catalonia and the Balearic Islands (Aragon-Catalonia-Balearic).

• The Central-Northern region of Spain, encompassing Cantabria, La Rioja, the Basque Country and Navarre, increases its share to nearly a quarter of all surnames among Afro-Cubans (22.4%).

• The share of surnames from Andalusia, the Canary Islands and Galicia-Asturias decrease substantially, compared to the general population. Other regions show smaller decreases or increases, or their share is so small that any change might just be noise.

Why does the regional distribution of surnames among Afro-Cubans differ so much from the Spanish regional distribution among the general population?

Well, in a slaveholding society where the main economic activity relies on the effort of an enslaved workforce, economic success was largely measured by having a bigger piece of that activity’s economic pie, the sweet sugar pie. Economic success usually meant owning more and larger sugar plantations, and owning more slaves to work them. If certain groups of Spanish immigrants achieved greater economic success than others, it’s not surprising these groups eventually controlled a larger share of the Cuban sugar industry and owned more slaves.

The surnames of these successful Spaniards were inherited by Afro-Cubans, who now carry in their names a vivid testimony of Cuba’s economic history. Catalans and Basques were particularly successful in 19th-century Cuba, far above other Spanish immigrant groups, and their wealth was often based on sugar and on the enslavement of the Africans who worked on their plantations. A striking example of this success was the Basque immigrant Juan Zulueta10:

Zulueta was born in Anucita, an obscure agricultural village in the Basque province of Alava. The son of a common laborer with no education and no money, he followed the example of thousands of Basques and Spanish of his age and circumstances: he migrated to Cuba. By his own account, Zulueta’s goal was to save approximately 2,500 pesos to return to Alava and become a middle-level proprietor, owning anywhere from 10 to 100 hectares. Once in Havana, Zulueta entered the slave trading operation as a middleman, and within twenty years had become one of the richest and most powerful men in Cuba.

Another notable example is Facundo Bacardí, of Bacardí rum fame, who was born in Sitges, Catalonia11:

[Facundo Bacardí] arrived in Santiago de Cuba in 1838, a poor, struggling Catalán, apprenticed to a family firm. Experimenting in the kitchen of his patron’s house, Bacardí succeeded in distilling a rum superior in body, flavor, smell and taste to any previously offered on the market. The light amber color and taste, and the virtual absence of ammonia smell, made addicts of an ever-widening circle of aficionados. From Santiago the fame of Bacardí’s kitchen distillery spread to Havana; from Havana to Madrid, New York, Philadelphia, Baltimore, London and Shanghai.

The success of these Catalans and Basques - and also Cantabrians and Baleares - compared to other Spanish immigrants (Canarians, Galicians) was so pronounced that it distorted the regional distribution of surnames among the descendants of slaves.

There’s also a corollary to the high rate of Basque and Catalan surnames among Afro-Cubans. This result suggests that the regional distribution of surnames among White Cubans is even more skewed towards the Canary Islands, Andalusia and Galicia-Asturias than the distribution among the general population.

The general population distribution somewhat averages the regional frequencies of surnames among Afro-Cubans and White Cubans. Given that Basque and Catalan surnames are so much more frequent among Afro-Cubans, we should expect the share of these surnames among White Cubans, which I have no direct way of observing, to be even lower than the 5%-6% seen in the general population.

That is, by Latin American standards, very low.

(Credit for this post’s thumbnail: Wikipedia. The guy hitting a ball with a wooden stick is Yordan Rubén Álvarez Cadogan)

Have you noticed that all books and articles about the voyages of Christopher Columbus to the New World seem to include a map like this one?

Well, that dot labeled “Palos“ corresponds to the town of Palos de la Frontera, in Huelva province. And if we zoom in on the coastal area where Palos is located, it looks like this (modern map, so ignore the roads and highways):

Revealing that this Palos de la Frontera port from which Columbus departed is not actually located on the coast but on a riverine estuary, offering good access to the sea while at the same time serving as a well-protected harbor. Estuaries make excellent locations for maritime ports, and some of the world’s most important ports (Shanghai in the Yangtze estuary, Sidney Harbor, Antwerp in the Scheldt, San Francisco Bay) are located in them. As a history nerd, of course I wanted to visit Palos de la Frontera simply because of its role in the early voyages to the Americas, but its peculiar geographic location made the trip even more appealing.

If we zoom in even further on a satellite image of the Palos area (credit Google Maps), you can see the location of the old 15th-century port, to the left of the label “La Fontanilla“. You can also see the many, many silvery, unnatural-looking structures surrounding the town of Palos (locals don’t call it “de la Frontera”).

Those structures are not fake plastic trees3; they are very real plastic structures to support the cultivation of real trees (and shrubs). I was struck by how ubiquitous these greenhouses are throughout the province when I traveled from the city of Huelva (capital of the province) to the town of Moguer, passing through Palos along the way. They are everywhere.

Moguer is also closely tied to Columbus’ voyages - it was the port where the ship La Niña was built - so I was very keen to visit both places. After arriving in Moguer, the second thing I noticed (or maybe third, after “there aren’t many Columbus-related sights here; maybe I should have gotten off the bus at the La Rábida monastery like the other tourists…“) was how many foreign-looking immigrants there were in this small Spanish town. They are everywhere.

Granted, I live in the Spanish province with the second-lowest percentage of foreign-born population in the entire country, and a large share of those living here are kind-of, sort-of Spanish-looking Latin Americans like me, who don’t stand out (much). Walking around my hometown one might get the impression that recent reports about Spain’s very large immigrant population are grossly exaggerated4. In Huelva province, on the other hand, walking around Huelva city (10.2% foreign-born) didn’t feel particularly different from any other mid-sized Spanish city, yet arriving at Moguer, where 26.6% of its 22,000 population are foreign-born5, felt very different.

This begs the question: why is there such a high share of foreign-born population in the towns of Moguer and Palos (21.6%)?

Or, more generally, why do people migrate?

I’ve written before about how geography and weather can influence migration patterns, particularly in the case of Canary Islanders:

The fact that the island’s climate allowed for the cultivation of tropical crops was a distinct advantage not shared by mainland Spaniards, making emigration to the American tropics, where similar crops could be grown, even more appealing to [Canary Islanders]

These factors, together with the availability of economically exploitable resources in the receiving regions, explain a large part of the settlement patterns of the Americas. They explain why African populations were forcibly settled in the tropical lowlands, where a plantation economy used slave labor to produce tropical crops for European markets. They also explain why Canary Islanders settled in both tropical regions and sparsely populated border regions of Spanish America. More generally, they help explain why sparsely populated parts of the continent received more European immigration than areas which already had large indigenous populations before the Spanish arrival.

In economic terms, most free migrants moved to regions of the continent where their skills were more valued and wages were higher; or where other production factors - land in particular - were more abundant and cheaper, allowing for greater returns to their labor. And even in the case of unfree labor, African slaves were transported to areas where their labor was more valuable, though the slaves themselves clearly derived no benefit from this.

Today, in a world where almost every piece of land with good agricultural potential is already under intense cultivation, and where many middle-income and poor countries struggle with high levels of unemployment, migrants no longer move in search of land. They move in search of jobs and higher wages.

Back to Moguer and Palos

So, why is there such a high share of foreign-born population in the towns of Moguer and Palos?

Before answering that question, let’s go back to my home province of Córdoba, which has several mostly agricultural towns of a similar size to Palos and Moguer (10k to 25k population).

There’s even a specific Spanish term for these towns: Agrociudad (Agricity). It denotes a small city whose economy is based almost entirely on agriculture and services, with no industry. The term is not only useful for characterizing the economy of such cities, but more broadly as a way to portray the economy of Southern Spain - where almost all of these cities are located - as a region where very little industry exists outside of mid-sized and large cities.

All of these small cities in Córdoba province have relatively low immigrant populations (less than 5%6), and an African-origin immigrant population of less than 2%. The overwhelming majority of small cities in Huelva and Córdoba are Agrociudades, heavily dependent on agriculture, yet for some reason far more migrants come to work and live in Huelva than in Córdoba.

Both provinces suffer from high unemployment, but the rate is slightly higher in Córdoba (17.0%) than in Huelva (15.6%)7. It doesn’t seem like migrants choose Huelva simply because jobs are easier to find there, so why do they flock to agricultural cities in Huelva but not in Córdoba?

The answer is unskilled labor demand. This is what skilled agricultural labor looks like:

And this is what unskilled agricultural labor looks like:

Córdoba province, being far from the coast, does not enjoy the mild weather that is ideal for growing garden vegetables and many fruits, but instead it produces a lot of wheat, over three times as much as Huelva8. Wheat farming is highly mechanized though, requiring little labor compared to other crops. If we want to find an agricultural product grown in Córdoba that requires a large workforce, we should look at other crops. The following map, showing the main crops in every region and province of Spain, should give us a good idea of what other major crops are grown in Córdoba.

Although the map is a bit crowded, you can probably see that olives (mostly for oil), sunflower, and grapes (for wine) are Córdoba’s main crops besides wheat9. And a quick internet search will confirm that sunflower farming is just as mechanized as wheat farming.

That leaves us with olives and grapes. Somebody has to pick those fruits by hand, and countless hectares (acres) of olive trees and vineyards require countless unskilled laborers, right? Well, not necessarily. In case you are not up to date with the latest in agricultural machinery, let me present to you the mechanical tree shaker:

(I hate embedding videos, but there’s no better way to understand how shakers work without seeing one in action (or maybe visiting a friend who owns an almond farm). Also, this should be a caption but Substack doesn’t allow video captions…)

These modern agricultural machines have made it possible to harvest olives, and even grapes, employing little labor compared to traditional hand harvesting. And although in the case grapes the varieties grown in Córdoba10 are overwhelmingly harvested by hand, the wine industry does not generate a substantial demand for labor due to the limited vineyard area - less than 6,000 hectares11 - and the seasonal nature of the work.

In the words of the manager of one of the major vineyards in Córdoba12:

It’s a very difficult situation. Currently, viticulture is the most labor-intensive agricultural activity, and there’s a huge problem finding workers. One of the few positive aspects we see is the increased trend towards consuming young white wines.

In summary, Córdoba’s agriculture employs relatively little unskilled labor, and the seasonal nature of the work fails to draw migrants to the province.

In contrast, Huelva province has a much milder maritime climate. The kind of weather favored by German and British tourists, as well as by strawberries and cranberries. While it’s not much sunnier than Córdoba, the milder temperatures make it almost ideal for growing strawberries in greenhouses, and as a result Huelva accounts for over 90% of Spain’s strawberry production (did you notice the strawberry drawn over Huelva province on the map above?).

Strawberries are planted by hand and harvested by hand.

The same is true for cranberries and raspberries. And because these crops are harvested at different times of the year, they ensure a steady demand for unskilled labor, all year round.

Huelva’s ideal conditions for greenhouse farming create a strong demand for unskilled labor, so migrants naturally move to where the jobs are, which is where the berries are. Mind you, not all of Huelva province shares the same climate. The further inland you go, away from the sea’s moderating influence, the less economically viable greenhouse farming becomes. This geographical pattern of Huelva’s agriculture is mirrored by a geographical pattern of immigrant settlement, as you can see in the map below.

The central and northern areas of the province, as well as the eastern area near the Seville border, have attracted relatively few immigrants compared with coastal Huelva. And all because of the weather.

Even in this day and age, in our modern world, geography and weather still play a major role in determining who migrates and where.

I have to admit that the controlled craziness of Carnival has a certain appeal. And I’m not just saying it because a guy in a black widow costume called me handsome. There’s something to be said for the rule-breaking spirit of Carnival and the blowing off of steam in massive public celebrations.

On the other hand, twice in the last month I’ve been screamed at in public for no particular reason. Canarians have a deserved reputation for being very talkative, and after a couple of years living in Andalusia I’ve grown used to strangers striking up conversations with me, but I could do without the shouting.

After these events, and a few others that would take too long to recount1, I decided to take a look at the mental health statistics for the Canary Islands. I didn’t believe Carnival madness was really an expression of the revellers’ mental health, but the shouting did make me contemplate the possibility of an underlying cause.

Well, according to the 2023 Spanish National health system annual report (as reported by the Spanish press), the mental health of Canarians is the worse among all Spanish regions2:

The [Spanish] age-adjusted prevalence [of behavioral and mental disorders] was 331 cases/1,000, which is 4.7% higher than in 2019.

By region, the highest figures were in the Canary Islands (518.5); Valencian Community (466.8) and Murcia (424.3), while Castilla-La Mancha (103.4); Extremadura (132.9); Andalusia (135.4) and the Basque Country (299.4) did not reach the national average.

Notice that the regions of Andalusia, Extremadura and Castilla-La Mancha show substantially lower rates of behavioral and mental disorders than the Spanish average. This is particularly interesting because these three regions form what is commonly understood as Southern Spain3, and the Canary Islands are usually grouped together with them, due to a similar culture and the Andalusian origin of a large share of the islands’ early colonists. And yet, remarkably, the Canaries differ widely from Southern Spain in terms of mental health.

This high rate of mental disorders is not a recent phenomenon. As you can see in the following chart, this has been the case for more than a decade - though notice that the Valencian Community shows a higher rate in the chart4.

And this pattern aligns with other mental health indicators, such as the suicide rate per 100,000 people. While the Canary Islands do not display the highest raw suicide rate in the next chart, this is likely a result of not adjusting by age structure, as Asturias (highest) and Galicia (second-highest) have the oldest populations in Spain.

In contrast to Asturias and Galicia, the Canaries have a relatively young population structure, which lowers its raw suicide rate compared to older regions since suicide rates usually peak in older cohorts. Using a crude age adjustment5, I confirmed that age-adjusted male suicide rates in the Canary Islands were roughly the same as those in Asturias and Galicia in the years before 2020 (prior to Covid-19), and that a substantial gap exists between the archipelago’s high rate and the (average looking) rates seen in the other Southern Spanish regions.

Furthermore, a 2017 Spanish health ministry’s breakdown of mental health diagnoses6 shows that rates of mental disorders among older people are not exceptionally high in the archipelago7. Disorders affecting the young seem to be more prevalent than the Spanish average8, while those not specific to young or old exhibit exceptionally high rates: phobias, schizophrenia, other organic psychosis9, etc. The only exception in this last category is affective psychosis, with a rate roughly half the national average10.

Lastly, we can examine the rates of accidental poisoning deaths by psychotropic drugs and drugs of abuse, a specific cause of death recorded by Spain’s National Institute of Statistics (INE). Although this indicator may not track mental health as closely as the others, the Canary Islands again show up among the highest ranked regions.

What about Latin America?

I believe the previous figures clearly establish that Canarian mental health indicators diverge from the rest of Spain, but I must acknowledge that I have no evidence this pattern has existed for longer than the last couple of decades. Asking whether this also holds in Hispanic Latin America among populations of Canarian descent compared to those from other Spanish regions implicitly assumes that this is not a recent phenomenon, but a long-lasting one that is endogenous to the Canarian population. Yet, I can’t find any way to measure the mental health of Canarians and other Spaniards in the 17th or 18th century. While I examine Latin American mental health indicators in the following paragraphs, please keep in mind that if Latin American regions with heavy Canarian settlement have worse mental health than other regions there’s a substantial probability this is just a coincidence.

Nevertheless, it might be worthwhile to check whether this mental health pattern holds on the other side of the Atlantic, at least to test the hypothesis that Canarians’ poor mental health is a long-lasting phenomenon which traveled across the ocean along with Canarian culture.

Let’s start with suicide rates. The countries we want to focus on are the ones that received the largest Canarian migration flows and therefore have the highest share of Canarian ancestry (what I call Canari America):

[Canarians] played a significant role in shaping the contemporary societies of a few countries: Cuba, the Dominican Republic, Puerto Rico, Uruguay, and Venezuela11.

Well, Uruguay, Cuba and and Venezuela had the three highest suicide rates in Latin America in 2021, with 24.8, 13.8 and 8.6 suicides per 100,000 people respectively.

Those suicide rates tentatively suggests the pattern holds, but let’s continue by looking at other death rates related to mental health. For this I’ll rely on data from Our World in Data, which compiles death rates by country based on World Health Organization (WHO) statistics. These are the rates published by Our World in Data, along with the ranking of our countries of interest within Latin America12:

• Alcohol and drug use disorders (2021): Cuba (third), Uruguay (fifth).

• Mental health and substance use disorders deaths: Cuba (third), Uruguay (fifth).

• Drug use disorder death rate: Uruguay (fifth).

Some of these rates clearly overlap each other, so don’t take them as three separate pieces of evidence, but rather as one additional piece of evidence suggesting poor mental health in the countries of Canari America, even if not very strongly (notice none of the countries ranks first or second).

Also, even though a country’s overall mental health is just one of several factors influencing the number of psychiatrists in that country, I should point out that Uruguay has the highest ratio of psychiatrists per 100,000 people, and Cuba the third-highest in Latin America, according to Our World in Data13.

Taken together, these figures imply that the Canarian mental health pattern may hold in Latin America, at least in some countries and at least recently. In the case of Venezuela though, only its high suicide rate suggests poor mental health, and I couldn’t find other recent data that would support this hypothesis. But I did find one interesting piece of not so recent data.

The 1941 Venezuelan census

People in 1941 were not as politically correct as they are today. They had no compunctions about calling the intellectually disabled “idiots”, and those suffering from mental health disorders “crazies”. So it’s not surprising that the designers of the 1941 Venezuelan census thought it was a good idea to record the number of people suffering from physical and mental defects, including those classified as idiots and crazies, allowing us to examine the rates and figures of people classified as such by census takers in each Venezuelan state at that time.

You might recall that I’ve written before about the cultural differences between Venezuelan regions:

And how Venezuela’s Caribbean coastal region seems to be more culturally similar to the Canary Islands, while its Andean region differs from both.

Well, the 1941 census allows us to test whether cultural differences between these two regions also extend to mental health differences by looking at the rates of crazies per 100,000 people in each Venezuelan state14.

The three Andean region states (Mérida, Táchira and Trujillo) show low rates of “crazies”, but not exceptionally low. And, of course, it’s also possible that census takers did not evaluate mental health in a uniform manner across Venezuelan states.

You’ll also notice that I marked Barinas state with a question mark in the above chart. Although it’s not geographically part of the Andes, Barinas borders all three Andean states and, as shown in the election maps from my previous Venezuela post15, it can be grouped with the Andean states when it comes to political leanings (UPDATE: In fact, more than 10% of its 1941 population had been born in an Andean state).

To account for the possibility that our politically incorrect census takers applied different thresholds when classifying mental disabilities across Venezuelan states, let’s examine the ratio of crazies to all mental disabilities (crazies plus idiots)16.

That looks like stronger evidence that Andean Venezuelans suffer from lower rates of mental health disorders than other (more Canarian) Venezuelans17. Or at the very least, that craziness was understood in a different way in one cultural region than in the other. This is not definitive proof that the Canarian mental health pattern also holds in Latin America, but it’s suggestive enough that the possibility should not be dismissed.

So to sum up, the Canary Islands do not indisputably have the poorest mental health in Spain, but they clearly rank among the regions with the worst mental health indicators. Notice, though, that other regions with similarly poor mental health indicators (Murcia, Valencia) contributed little to the Spanish settlement of the Americas, so perhaps the Canary Islands have the worst mental health profile among the regions that made a substantial contribution to the peopling of Latin America.

Secondly, there is at least some evidence - though not conclusive - that countries and regions of Latin American which experienced heavy Canarian settlement show worse mental health indicators than the rest of Latin America. This is hardly definitive, but it is another factor worth considering when trying to explain why Canari America diverges from the rest of Latin America.

15

Here’s one of those maps (2013 presidential election), followed by a Venezuela map showing the location of Barinas, in case you don’t want to click on that link:

At the time, I wasn’t particularly interested in Israeli fertility, so I chose France and Germany as comparison countries1. They both have a large number of billionaires, which made it easier to gather a sufficiently large sample of billionaires with publicly available information on their number of children. Their fertility rates also provided a contrast to South Korea’s: France has relatively high fertility for a developed country, while Germany has an average rate for a developed country.

The conclusion I reached from that comparison was:

…Korean billionaires have a significantly lower fertility than their French and German peers.

The values of the fertility gaps between South Korean billionaires and French or German billionaires are 0.71 and 0.62 respectively2. Those same values for the general population are 0.92 and 0.64, a wider but similar range.

This suggests the existence of a factor, specific to South Korea, a non-economic issue that lowers fertility for both Korean billionaires and the general population.

Assuming that billionaires are largely unaffected by economic pressures, and that fertility is primarily affected by economic and cultural factors, the cross-country comparison of billionaires and general population fertility suggests that culture’s effect on fertility is substantial, as you can see in the following chart.

Now that I’ve become particularly interested in Israeli fertility, I decided to take another look at billionaire fertility, this time including Israeli billionaires. Israel has a large number of billionaires for such a small country (in case you are not aware of the stereotype, Jews make lots of money), which not only allows me to calculate a credible billionaire fertility rate but also to restrict this analysis to Secular billionaires, given that Secular Israeli fertility has been the focus of my recent fertility posts.

Among other things, I want to check whether the Israeli billionaire fertility rate can tell us anything about the Mimetic theory of Israeli fertility: the theory that Secular Israelis have a higher fertility rate than the (secular) population of any other developed country due to the memetic influence of high-fertility religious Israelis:

What these groups do is provide a one-way channel of influence between Israeli Charedim and the rest of Charedi society, a sort of valve in which Charedi fertility memes can spread outwards, without allowing fertility decreasing memes inwards….

Around 25% of the Religious Zionist [Dati] community is actually barely religious at all and very comfortable freely mixing with ‘traditional’ Israelis who don’t keep the Sabbath, but ‘respect’ it…Finally, these traditional Jews mingle with secular Jews, who pick up the last scraps of the Charedi fertility meme, enough to keep their demographic head above water.

The basic model here is that, unlike anywhere else in the world, there is no hard cultural barrier (at least in one direction) between Charedim and the rest of society, and enough mimetic stepping stones that almost everyone is influenced to some degree.

The billionaire samples

Before I continue, let me remind you of how I constructed the billionaire samples used in these comparisons:

• All living individuals listed as billionaires in the respective Wikipedia entries for each of the three countries are considered for inclusion (in the case of Israel, which has less billionaires than France and Germany, I also used two other sources3).

• Many billionaires place a high value on their privacy for understandable reasons, so those for whom no information on the number of children could be found were not included in the sample (this time, I also used Grok to help me with this task).

• Didn’t include anyone bellow the age of 43 because younger people have probably not reached their completed fertility (the youngest Israeli billionaires I could find were 49 years old).

• Didn’t include billionaires who hold the nationality of a country yet have spent the majority of their lives, particularly their formative years, outside that country. The sample is supposed to be a good representation of the general population of the country in terms of culture.

• Did not include widowers of billionaires, but did include children of billionaires even if they themselves do not appear in a billionaire’s list.

• I also included billionaire families, whose individual members do not necessarily meet the billionaire threshold on their own. These are families (meaning siblings, cousins) who collectively own billions (e.g. the Hériard-Dubreuil family, majority owners of Rémy Cointreau group, valued at more than 9 billion euros).

The purpose of including close family of billionaires and billionaire families is to have a larger sample of very rich people, especially people under 70 years old, who were raised in the same families as billionaires and experienced the same absence of economic hardship. Billionaires tend to be old for obvious reasons, and I wanted to have as many younger people on the sample as possible.

Also, billionaire families include individuals whose net worth is not billions, but only hundreds of millions. They are in the same category as billionaires from the point of view of not having economic constraints when deciding to have children.

The results

To reduce the effect of decreasing fertility over recent decades on the billionaire sample, I chose a maximum age threshold of 70 years. This ensures the fertility comparison is between the general population and “younger“ billionaires, rather than “older” billionaires. In the case of Israel, I raised the threshold to 72 years of age so its billionaires would be comparable to the billionaires I examined in 2024. And I also excluded Israeli billionaires from Mizrahi or Sephardic backgrounds, because I suspect most of them are Traditional rather than Secular4.

And now, let’s take a look at the estimated fertility rate for under-72 Israeli billionaires and how it compares to the Secular Israeli fertility rate of 1.96 2.00 children per woman.

(UPDATE: I should have used the 2019-2021 Secular Israeli fertility rate (2.00) instead of the one for 2021-2023 (1.96). This doesn’t change the results though. The value of M for Israel drops from 1.72 to 1.69)

Just as Secular Israelis have higher fertility rates than the populations of other countries, Secular Israeli billionaires have higher fertility (3.38) than billionaires from other nations. I guess this isn’t surprising, and it could be interpreted as supporting evidence for the Mimetic theory: once economic constraints are removed, Secular Israeli fertility climbs even higher, reducing the gap with the fertility rates of religious Israelis (Haredi and Dati). But let’s leave the Mimetic theory aside for a moment.

While comparing the fertility gaps between general population and billionaires across countries, I noticed the Israeli gap did not appear to align with those of other countries. But the ratio of billionaire fertility rate to general population fertility rate in Israel did fall within the range of the other countries (except South Korea).

(To give an example, if the billionaire fertility rate were 3.0 children per woman and the general population fertility rate were 1.5, the ratio would be 2.0)

Setting aside South Korea, the fact that countries with fertility rates as low as Italy’s5 (1.28 children) and as high as Israel’s show similar ratios suggests that the same fertility-raising mechanism (removing economic constraints) acts over the baseline fertility rates of different countries, multiplying fertility in roughly the same proportion.

This is very much speculative, so don’t quote me saying the ratio of billionaire fertility rate to general population rate is roughly similar across most developed countries. I’m not completely sure that the pattern is real. An obvious way to settle this would be to calculate billionaire rates for additional countries (I actually mentioned Italy, Japan and Spain in my previous billionaire fertility post), and I might just do that in the near future.

Going back to the Memetic mixing theory of Israeli fertility, the fertility gap between Secular Israelis and (non-billionaire) religious Israelis narrows among billionaires; but is that what the theory would predict? In a trivial sense, yes. Secular Israeli billionaires have higher fertility than other billionaires.

Let’s call the factor that raises Secular Israeli fertility M. This factor supposedly multiplies (maybe 1.2X, 1.4X) a baseline Secular Israeli fertility rate (R), which otherwise wouldn’t be much different from the rates of other developed countries. Similarly, let’s call the factor that raises billionaire fertility above the general population baseline factor B (which falls roughly in the 1.6 to 1.9 range, as you can see in the chart above). Under the Memetic theory, the Secular Israeli fertility rate results from multiplying the baseline rate by M, while the Secular Israeli billionaire fertility rate is the result of multiplying the same baseline rate R by M and then by B:

Secular Israeli rate = RxM ~= 1.96
Secular Israeli billionaire rate = RxMxB ~= 1.96xM ~= 3.38

This implies that Secular Israelis in general, and Israeli billionaires in particular, are mimetically influenced by religious Israelis (factor M). But if so, why would Secular Israeli billionaires be influenced by religious Israelis? Or, more narrowly, why would they be influenced even by relatively more religious Traditional Israelis?

In terms of status, billionaires rank among the highest-status groups in every developed country I can think of. They are literally at the very top of one of the main measures of status recognized in every society. They have no reason to imitate non-billionaire religious Israelis, whether in aspects of their family life or any other area of life.

If Israeli billionaires are not influenced by religious Israelis, that leaves us with two possibilities: either Israeli billionaires do not follow the pattern followed by billionaires in other countries (factor B), or perhaps factor M is not connected to Secular Israelis imitating religious Israelis’ family life (maybe it’s More-analysis-is-needed factor??).

I’m not certain the former explanation is true; I would need to calculate the billionaire fertility rates for more countries to be more confident that factor B is real and applies to Israeli billionaires. And even then, South Korea already provides a clear counterexample. That said, I’m definitely skeptical about the latter, and the more I look at Israeli fertility data, the more convinced I become that “factor M” has little to do with mimetic mixing.

(Credit for this post’s thumbnail goes to the Berlusconi Flying Circus blog. If you think Silvio Berlusconi’s life is one of the most entertaining recent counterexamples to Noblesse oblige, you might enjoy reading Berlusconi Flying Circus’ article on the late Italian billionaire’s family life. Berlusconi and German billionaire Ferdinand Piëch might end up inspiring a future post on fertility.)

That’s until a few days ago, when I encountered this tweet by George Mason University economist Vincent Geloso (who seems very smart and well-informed):

The tweet refers to a 2025 paper by Geloso and Jamie Bologna Pavlik which uses a clever econometric method, synthetic controls, to estimate what the trajectory of Cuban per capita GDP could have been absent the 1959 Revolution, and concludes that the main cause of Cuba’s current poverty is not the American embargo but revolutionary policies.

I have no issue with the paper in general, but I have an issue with this paragraph from the paper’s A.1 appendix “How Rich was Cuba Pre-1959“ which attempts to provide context on Cuba’s economic history:

By 1913, this was no longer the case, as [Cuban] income per head had reached approximately 60% of the level observed in the United States. By 1925, Cuba’s income per capita placed it roughly on par with America’s poorest states (e.g. Alabama, Mississippi). Nevertheless, this remained higher than in all Latin American countries except Argentina, which stood at 80% (Bulmer-Thomas, 2003, p. 492). By the mid-1950s, this relative position persisted (Bulmer-Thomas, 2003, p. 331). Other development indicators—such as car, television, and radio ownership, infant mortality, life expectancy, and literacy—also placed Cuba near the top, at times rivaling European levels (Ward and Devereux, 2012, p. 115)… In other words, no one disputes the claim that Cuba was ahead of most of Latin America around the time Castro arrived in power.

I’m now going to pedantically review each phrase in that paragraph, and dispute Geloso and Pavlik’s claim.

(I shouldn’t need to say this, but given that the only like my tweet/response to Geloso received came from someone who commented “people don’t eat GPD…Cubans are free, americans are slaves“, here it goes: the goal of both this post and the previous one is to place Cuba’s supposed pre-revolutionary (relative) prosperity in the proper perspective. Arguing for a correction of estimated pre-revolutionary living standards downwards SHOULD NOT BE TAKEN as an argument in favor of the Revolution, just like arguing for a correction upwards should not be automatically interpreted as an argument against the Revolution. I do not have a Fidel Castro or Che Guevara poster on my wall)

Let’s start with the first sentence. Regarding Cuban per capita income in 1913, it wasn’t even close to 60% of U.S. levels. In fact, it was roughly 21% of American income, very similar to Mexico’s, as you can see in the following chart.

By 1925, the picture hadn’t changed much. Cuba’s per capita income was still close to 21% of that of the U.S. level, roughly in the same range as four other Latin American countries: Mexico, Venezuela, Costa Rica and Panama (which actually saw a substantial relative drop from 1913 to 1925).

The gap between the American average income and Latin American incomes in the 1920s was so wide that even Argentina, the wealthiest country in the region, reached only around 60% of the U.S. level (not 80%), slightly above the states of Alabama and Mississippi.

Other indicators

When commenting on other development indicators, Geloso and Pavlik cite (in addition to Ward and Devereux) Xavier Tafunell’s 2007 article “On the Origins of ISI: The Latin American Cement Industry, 1900–30“, mentioning that, according to Tafunell “Cuba was also responsible for one quarter of Latin America’s capital formation in machinery and equipment in 1920“. This is not correct. Tafunell (2007) does not say anything of the sort.

I’m glad the authors cited Tafunell (2007) though, because reading that paper allowed me to gather statistics on early 20th century cement production for several Latin American countries, including Cuba. Tafunell collects production data only up to the year 1929, but thanks to the United Nations’ 1959 Statistical Yearbook we can supplement Tafunell’s data with late 1940s and 1950s data. With this longer data series we can compare relative cement production levels across countries and examine historical trends.

Here’s a chart of per capita cement production in several Latin American countries based on the UN’s and Tafunell’s data.

Notice that Cuba (blue line) starts out with an exceptionally high level of per capita cement production in the 1920s, only to end up in a better-than-average but unremarkable position by the 1950s, on the eve of the Revolution. By 1957-1958, Cuban production was roughly on par with that of the Dominican Republic, a country no one would describe (am I sure?) as having been “near the top” in 1950s economic indicators. More on this later.

Near the middle of the paragraph, Geloso and Pavlik seem to imply that by the mid-1950s Cuba’s per capita income “remained higher than in all Latin American countries except Argentina“. I have three issues with this claim:

• I can’t find any statement in Bulmer-Thomas’ book (searched from pages 329 to 333) saying that Cuba had the second-highest Latin American income in the 1950s1.

• As I’ve shown in the previous charts (1913 and 1925), Cuban per capita income in the early 20th century was not higher than Uruguay’s, and very likely not even higher than Chile’s. It was not the second-highest.

• This last point is not that important. What is far more important though, is that Cuba’s relative income position did not remain the same in the 1950s as in the early 20th century.

This third point is illustrated in the following chart: while many Latin American countries significantly increased their per capita income between the early 20th century and the 1950s, Cuba’s income barely grew around 40% (incomes measured in constant dollars). The only other country in the sample with a similarly disappointing performance is Panama.

This is similar to the trend we saw in the cement production chart. You could say that Cuba began the century “near the top”, yet by the 1950s its economic indicators are no longer among the Latin American leaders, with many countries rapidly closing the gap or even surpassing Cuba.

UPDATE: The previous chart was not clear enough, so I’ve now created a bar chart showing the increase in per capita income from the second period (1924-1926) to the third period (1955-1959). You can clearly see the very disappointing performance of Cuba in this chart.

The deaths of infants and the death of credible statistics

Let’s move on to the last sentence in the paragraph: “Other development indicators—such as car, television, and radio ownership, infant mortality, life expectancy, and literacy—also placed Cuba near the top, at times rivaling European levels (Ward and Devereux, 2012, p. 115)“. Regarding infant mortality, I’ll quote myself (sorry about that):

Not only is the 38.9 death rate declared as unreliable, but we now have an explanation for the marvelous Cuban rate: the rate was calculated excluding live-born infants dying within 24 hours of birth. The only rate Cuba reported to the United Nations can’t be compared to other countries’ rate, and is significantly lower than the real infant mortality rate.

I didn’t try to estimate Cuba’s true infant mortality rate in my previous post, but I’ll give a try now. In Cuba, all infant deaths within 24 hours of birth were as stillbirths, so knowing that reported yearly stillbirths ranged between 6,500 and 5,500 in the years 1938-1949, we can try to estimate how many of those are “excess” stillbirths, assuming we know what the expected number of yearly stillbirths is, of course2.

My rough estimate, based on the stillbirth numbers from other Latin American countries is 2,500-3,000 expected yearly stillbirths in Cuba3. This would imply around 3,000 “excess” stillbirths. Next, we need to divide these excess stillbirths by the total number of births. Well, this is problematic.

During the 1938-1949 period, Cuba’s number of reported yearly births varied wildly, from 69,492 to 174,526 (maybe I should write a post trying to square a country’s inability to properly record births with a high rate of car and radio ownership…). Taking the second-highest number of births (148,917 births in 1949) yields roughly 20 excess stillbirths per 1,000 births. And adding that number to Cuba’s reported infant death rate produces a “corrected” infant death rate of around 60 per 1,000.

I don’t trust that number. But it at least gives a rough idea of how distorted the 37-38 rate is, and it puts to rest the claim that Cuba had a lower infant mortality rate than many European countries in 1950s.

Regarding life expectancy, there are three different sources, none of which provide data on Cuban life expectancy during the 1950s4.

The first source is the Economic Commission for Latin America and the Caribbean (CEPAL), cited in Felipe Sixto’s “An Evaluation of Four Decades of Cuban Healthcare” (2002), who’s in turn cited by Ward and Devereux (2012), who are in turn cited by Geloso and Pavlik (G and L). The second source is the Cuban Health Ministry (MINSAP), also cited by Felipe Sixto. And the third source is the UN’s 1970 Demographic Yearbook.

You can see the figures from CEPAL and MINSAP here:

Notice three things:

• The first MINSAP figure corresponds to the year 1960 (and it happens to be a round number).

• The life expectancy estimates from CEPAL and MINSAP do not coincide: MINSAP figures are higher.

• The MINSAP numbers suggest a sharp increase in life expectancy of 1.6 years between 1969 and 1970.

Before drawing any conclusions from these numbers, let’s take a look at the figures from the third source, the Demographic Yearbooks. Life expectancy data was published in most UN Yearbooks since at least 19625, yet no figures for Cuba appear in any Yearbook edition until 1970. The 1967 Demographic Yearbook in particular includes not only the latest life expectancy figures for all available countries, but also historic figures. There was no historic life expectancy data for Cuba at the time.

When the first life expectancy figure for Cuba is published, in 1970, it’s just an estimate produced by the UN’s Population Division, not a figure reported by the Cuban health system. And it lacks a breakdown by age cohort or by sex.

The only other Latin American countries for which the UN Population Division had to provide a rough life expectancy estimate were Honduras, Nicaragua, Paraguay and Brazil (though Brazilian data was published in the 1962 Yearbook). All of these countries, except Brazil, were at the bottom of the regional development scale at the time.

If we compare the UN’s 1965-1970 estimate of 66.8 years to the MINSAP and CEPAL figures, it does seem to match CEPAL’s 1965 estimate of 65.4 years (roughly, most underdeveloped countries increased their life expectancy at a rate of 3 years every half decade during that period). What it does not match, are the MINSAP’s 1960 or 1965 figures, which seem consistently inflated6. We can now calculate a 1950s life expectancy estimate, based on CEPAL and the Demographic Yearbooks.

The blame for misrepresenting life expectancy in 1950s Cuba does not lie with Geloso and Pavlik, but rather with Ward and Devereux (2012), who claim that “Life expectancy at birth, circa 1955 was 64 years in Cuba“ citing the same sources we’ve just examined. This is simply wrong. The most likely figure for Cuba around 1955 would be 58-60 years, putting Cuba on par with Costa Rica.

This is good, but not exceptional, and clearly lower than European levels at the time7.

Read the trend

The illiteracy data supporting Geloso and Pavlik’s claim that Cuba “placed… near the top, at times rivaling European levels“ comes from Ward and Devereux, who in turn use the MOxLAD (OXLAD) Database of Latin American statistics as their source8.

And I have to agree that illiteracy in 1950s Cuba was very low by Latin American standards, around 21% in 1955, placing it near the top of the region.

But I do take issue with the claim that Cuban literacy rivaled European levels (CAVEAT: maybe G and L meant European levels of radio ownership, car ownership, etc… but not literacy). I also have an issue with the implication that Cuba’s low level of illiteracy indicates the country was doing great in the 1950s.

First, let’s look at Europe. The 1964 UN Statistical Yearbook is the first edition to include illiteracy rates9, and even though its sources appear to be the same as those used by MOxLAD (reported rates for Latin American countries mostly match), there is no Cuban illiteracy data in that Yearbook10. So, assuming rates are comparable between both sources, how does Cuba fare compared to the few European rates provided by the Yearbook?

You can see that Cuba’s literacy rate was higher than those of Albania and Portugal11, and comparable to those of Greece and Yugoslavia12. We can safely assume that nearly every other European country not in the above chart had illiteracy rates below 10%13. Cuban literacy did not rival European levels.

The second point I want to make about Cuban literacy, relates to the cement production and income trends we saw above. You could say that newly independent Cuba in 1898 “inherited” a relatively high literacy rate from Spain, and if so, its position near the top of Latin America in 1910 or 1920 should not be surprising.

But how was it doing by the 1950s?

Most Latin American countries were rapidly reducing illiteracy during the mid-20th century, as you can see in the following table from Rosemary Thorp (1998).

And the biggest drop in illiteracy happened from 1950 to 1960: 11.5 percentage points in a decade.

Now, let’s look at the trends in illiteracy rates during the early and mid-20th century in Cuba and its peers.

Cuba’s illiteracy rate decreased by less than two percentage points in the 1950s, and barely 1.6% in the 1940s. This is the smallest reduction in literacy among its peers in the 1940s and 1950s.

I initially thought of ending the chart in 1960, right after the Revolution, but then added the 1970 illiteracy rates for all countries with higher illiteracy than Cuba. These 1970 data clearly suggests that reducing illiteracy does not become more difficult below a certain threshold (30%?). Colombia, Mexico, Panama and Venezuela continued to reduce illiteracy rapidly while catching up with Cuba.

Cuba didn’t slow down on illiteracy reduction during the 1940s and 1950s because progress became more difficult. It simply did worse than its peers14.

Conclusion (for now?)

I have to thank Geloso and Pavlik for prompting me to do the research I didn’t do in my previous post, and allowing me to create a more complete picture of Cuba’s socioeconomic performance in the 1950s.

Cuba’s 1950s indicators were certainly not bad, but the country was not near the top of Latin America. More importantly, the data I’ve reviewed shows a picture of stagnation, or at best slow improvement, much slower than the progress seen in its peer countries. Given its earlier standing, Cuba in the 1950s can be described as disappointing.

In the words of Victor Bulmer-Thomas, the author repeatedly cited by Geloso and Pavlik15:

Nevertheless, the Cuban economy was in serious difficulties by the mid-1950s and had gone backwards in many ways. As Table 16.5 shows, Cuban income per head was now below the Latin American average and was ranked only ninth among the twenty republics. Even among the Caribbean countries, the Cuban ranking was in decline… Why, therefore, is there still a myth of a Cuban golden age in the 1950s ruined only by Fidel Castro? It has, of course, been perpetuated by the exile community for obvious reasons, but it has also been defended by serious scholars.

I couldn’t have said it better.

I wrote about how this ranking is deeply flawed: the index it’s based on does not truly measure child poverty in wealthy countries. Unfortunately, it’s much better known than Eurostat’s more helpful Severe Material and Social Deprivation index, which more accurately gauges whether children experience material deprivation in their households.

A limitation of the Eurostat index though, as its name suggests, is that it only publishes data for European countries, and this makes it unsuitable for comparisons among countries whenever one of the countries involved is not part of Europe. Not a big deal. I didn’t intend to compare child poverty rates between poor or middle-income countries anyway.

Yet in a later comment to the post, I noted that the Oxford Poverty and Human Development Initiative (OPHI) at Oxford University, in partnership with the United Nations Development Programme (UNDP), produces a multidimensional child poverty index (MPI) that includes material conditions as one of its dimensions and is not restricted to a sample of wealthy countries. In fact, OPHI produces the index for 109 countries in total, including 14 Latin American ones; it comes as no surprise that Our World in Data relies on it as a source on poverty.

The index is calculated by grading households on a scale of deprivation and classifying everyone, adult or child, living in a household above a certain deprivation threshold as poor. The MPI value for a given age group can then be obtained by multiplying the share of people living in poverty in that cohort (e.g. children) by the deprivation intensity of the households they live in. Therefore, we can not only compare child poverty rates across countries surveyed by OPHI, but also compare child poverty rates to adult rates within each country, which is precisely what I did.

The outcome of this comparison confirms a result presented by UNICEF’s Innocenti report - maybe the report wasn’t all bad after all: child poverty is more prevalent than adult poverty. Moreover, the average MPI poverty index value for children aged 0-9 is 56% higher than for those aged 10-17. A sad state of affairs, given that children are the most vulnerable members of society.

Here are some selected MPI values, so you can get an idea of how poverty varies across age groups and countries.

Notice how some regions of the world tend to have higher MPI values than others (Sub-Saharan Africa in particular). And this same regional pattern holds when we examine only the share of population living in poverty (no deprivation intensity) in each age cohort for the same set of countries.

Although you can barely see Serbia’s percentage bars in the above chart, the country’s child poverty in the 0-9 cohort is 6.7 times higher than its adult poverty rate. This is an extreme case, but illustrative of the large gap between child poverty and adult poverty that the MPI reveals.

What does multidimensional mean?

I’ve already mentioned that one of the three dimensions comprising the Multidimensional Poverty Index is material conditions of the household (living standards). The other two dimensions, each contributing one third to the MPI index, are health and education.

In turn, both health and education dimensions are made up of two indicators, each weighing one sixth, which are defined thus:

• Nutrition: Any person under 70 years of age for whom there is nutritional information is undernourished.

• Child mortality: A child under 18 has died in the household in the five-year period preceding the survey.

• Years of schooling: No eligible household member has completed six years of schooling.

• School attendance: Any school-aged child is not attending school up to the age at which he/she would complete class 8.

Notice that these four indicators are evaluated at the household level, which partly explains why child poverty rates are higher than adult poverty ones: the larger a household, the more likely that at least one member will be considered deprived in any of the four indicators. Since households with children typically have more members, they are more likely to be classified as poor, all else being equal1.

So maybe the gap between child poverty and adult poverty is not as wide as it first appears.

Let’s now take a closer look at the education and health indicators, beginning by child mortality, and see if we can understand child poverty rates a little better. First, the definition of child mortality used by MPI is not the definition you might be familiar with. The standard definition counts yearly deaths of children under the age of five, while MPI counts all deaths under the age of 18 in the last five years. Since children are more likely than adults to live in households where other children also live, it should come as no surprise that children will be more likely to live in a household where an under-18 child died in the last few years. As a result, the child mortality indicator is consistently higher for children than for adults: as high as 4.35% for Argentine children aged 0-9, 4.4% for Serbian children (0-9), and 12.77% for Nigerian children (0-9).

Secondly, while child mortality is clearly correlated with poverty, this correlation is far from perfect. Children under five may die from causes wholly unrelated to poverty, decreasing the usefulness of this indicator as a good measure of deprivation, and the expanded age range used by OPHI (18 years) amplifies this problem by capturing even more deaths unrelated to poverty.

To illustrate this effect, consider murder rates. Latin American intentional homicide rates are some of the highest in the world, with Ecuador currently holding the top spot: 45.7 intentional homicides per 100,000 population. According to the MPI2, 0.75% of Ecuadorian children aged 0-9 live in a household where an under-18 child died in the past five years, while the same rate for children aged 10-17 is 0.67%, and for adults (18-59) is 0.62%.

The small drop from the 0-9 rate to the 10-17 age group, and then further to 18-59 adults, is unsurprising3. Nothing unusual with Ecuador’s child mortality indicator, apparently.

Before examining the MPI child mortality rates in other Latin American countries, let’s see what the United Nations Office on Drugs and Crime (UNODC) reports about murder in the Americas in its 2023 Global Study on Homicide.

The overall homicide rate decreases with age, with the sex disparity increasing sharply after the age of 14 years. The preponderance of male victims starts to become apparent from 10-14 years of age in the Americas, while in Europe this is the case from 18-19 years of age... In the Americas, young men are particularly at risk of homicide related to organized crime and gang-related violence.

In the Americas, murder rates by victim age sharply increase several years before reaching adulthood, which means many homicides are correctly classified as child murders. This is not directly related to poverty. As you can see in the following charts, the gap between child homicide rates, particularly among boys, and adult rates in the Americas is relatively small compared to the larger gap between children and adults in Europe.

Teenage boys in Latin America are being killed at a strikingly high rate compared to boys in other regions of the world.

This unfortunate state of affairs is apparently reflected in the MPI child mortality index for the 10 Latin American countries with the highest murder rates. Brazil, Costa Rica and El Salvador show higher child mortality rates among children aged 10-17 than children aged 0-9. Conversely, three other countries (Honduras, Colombia and Guatemala) show the expected rate drop from aged 0-9 to 10-17, while the four remaining countries were not included in the MPI or did not report child mortality data.

These seemingly surprising 10-17 child mortality rates are not so surprising if we assume that children in that age range are more likely lo live in a household with another 10-17 child than with a younger 0-9 child. In other words, children aged 10-17 have a higher likelihood of living in a household where another 10-17 child was murdered in the past five years4.

Let’s go back to Ecuador.

As you can plainly see in the chart of the 10 Latin American countries with the highest murder rates, I cheated: Ecuador is not among them. And what’s with El Salvador? Didn’t the Central American country experience a dramatic drop in its murder rate in the past few years?

The reason you see El Salvador in the 10 highest murder rates country list, and not Ecuador, is that the list I’m using is Wikipedia’s List of countries by intentional homicide rate, compiled in 2022 based on data from 2016-2019 [UPDATE: I should have probably used the Wikipedia list from 2019, based on homicide data from 2016-2017. It differs from the list I used in that Uruguay (not in MPI) is replaced by the Dominican Republic]. And the 2025 MPI index uses survey data collected between 2014 and 2019. If I relied on more recent murder rates, I would be comparing murders that took place in a given time period with their presumed effect on child mortality rates of a different period.

And my motivation for drawing attention to the child mortality rates of Ecuador and El Salvador is not simply to highlight that MPI uses not so recent data. The more important point is that in a future update of the MPI Ecuador and El Salvador might switch places, at least on the child mortality indicator. The spike in murders might show up as a child poverty increase in Ecuador, while the sharp drop in El Salvador would likely decrease the MPI value for that country. A well-constructed child poverty index should not behave like that.

Younger and older parents

After looking at the years of schooling data from every angle I could think of, I have only one comment. According to OPHI, a household where all children are too young to have completed six years of schooling and where no adult has completed six years of schooling is classified as deprived on this indicator5. This definition produces a consistent effect: in every single country with a non-negligible share of households (> 1%) classified as deprived on this indicator, the share of deprived children aged 0-9 is higher than the share of deprived children in the 10-17 age group. This is partly caused by the classification of children as deprived if their parents did not complete primary education, regardless of the children’s own access to education.

I presume OPHI would agree that education has become more widespread in recent decades, and that the average years of schooling among citizens of low-income countries have risen, increasing the expected years of schooling of younger adults as compared to older adults. Assuming that younger children tend to have younger parents and older children older parents, why then are households with younger parents (the ones with children aged 0-9) more likely to have no member who has completed six years of schooling?

The likely answer is connected to the “non-negligible share of households classified as deprived“ phrase that I used earlier. The fewer than 20% of countries in the MPI that have a negligible share (< 1%) of deprived households on this indicator are countries that achieved near-universal primary education several decades ago6. Only four of them7 have a higher rate of years-of-schooling deprivation for children aged 10-17 than children in the 0-9 age range.

I’ll try to explain what I think is going on here. For simplicity, let’s assume all households consist of two parents and one child8. In a country which has had near-universal primary education for decades, the probability that both younger parents lack six years of schooling is very low - these would be very unusual parents - and also slightly lower than the probability for two older parents. Once we factor in whether the household’s child has completed six years of schooling, it’s difficult to predict which household, younger or older, will have a higher overall probability of years-of-schooling deprivation. We know that the probability of a younger child (0-9) not having completed six years is higher, close to 100%, because younger children are not old enough to have completed six years of schooling. Obviously, the probability of non-completion for an older child (10-17) is lower than that (most are old enough to have completed six years), but since these are very unusual parents the overall probability for the older household might still end up higher, or lower, than the overall probability for the younger household.

On the other hand, in a country which hasn’t had near-universal primary education for decades, the probabilities of years-of-schooling deprivation for younger and older parents are significantly higher, and parents who didn’t complete six years of schooling are not so unusual. These are the large majority of countries included in the MPI.

Let’s see how our very simplistic three-member households do in this scenario. Suppose younger adults (aged 30-39?) have a 20% probability of not having completed six years of schooling and older adults (40-49?) have a 30% probability of the same. There’s a 4% (0.2*0.2) probability that in a younger household both parents lack six years of schooling, while in an older household the same probability is 9%. After factoring in the probabilities of the younger (0-9) and older (10-17) children, the overall probability of the younger household remains close to 4%, but what’s the overall probability for the older household?

According to MPI data, most countries outside of Sub-Saharan Africa have rates of school attendance higher than 85% for children aged 10-17. The 10-17 child in the older household has roughly a 75% chance of having reached the age where, assuming she has attended school regularly9, she would have completed six years of schooling. So, this 10-17 child has very roughly a 36%10 probability of not having completed six years.

Applying this 36% probability to our previous 9% probability of years-of-schooling deprivation for older parents gives us an overall probability for the older household of 3.26%, lower than the 4% of the younger household. In the overwhelming majority of countries outside of Sub-Saharan Africa, school attendance rates are higher than 90%, and therefore the gap between higher years-of-schooling deprivation in younger households and lower deprivation in older households should be wider11.

I deliberately excluded Sub-Saharan Africa from the previous calculations, but let’s now apply the same logic to SSA, a region where fewer adults have completed six years of schooling.

Assume that the probability of both parents lacking six years of schooling is 20% for younger households and 30% for older household. If the probability that children aged 10-17 haven’t completed six years is 50%, then the overall probabilities of deprivation for the younger and older household end up being 20% and 15% (0.3*0.5) respectively. This means higher years-of-schooling deprivation in younger households. Even if we use a value of 40% or 60% for the children aged 10-17, younger households still show higher years-of-schooling deprivation (only with a value higher than 67% the reverse is true).

If the probabilities for parents are 20% (younger) and 40% (older), then a non-completion rate higher than 50% for 10-17 year olds would produce higher years-of-schooling deprivation in older households. But those rates would imply a very rapid increase in schooling going from the older parents generation to the younger parents generation, followed by a sudden stagnation or even reversal in the current 10-17 year olds generation. That seems very unlikely.

These are all back of the envelope calculations, but they lead me to think that higher years-of-schooling deprivation in children aged 0-9 than children aged 10-17 is likely just an artifact of the way OPHI chose to define and disaggregate this indicator (six years of schooling, 0-9/10-17 age cohort split). I could probably confirm (or refute) this if I had access to data broken down by shorter cohorts, but unfortunately this is not the case.

(If any of these calculations seem wrong to you, feel free to correct me in the comments).

Medians, distributions and standard deviations

Since I don’t have anything to comment on School Attendance deprivation (it just tends to be higher in the 10-17 age cohort than in the 0-9 one), I’ll move on to the last of the four health and education indicators: Nutrition.

As I’ve already mentioned, the MPI classifies a household as deprived in the Nutrition indicator if any member of the household under 70 years of age is undernourished. Notably, its definition of undernourished differs by age: for children under 5 years, a child is considered undernourished if their height-for-age or weight-for-age falls below two standard deviations from the median of the reference population.

Let me try to explain what height-for-age, weight-for-age and reference population mean. Height-for-age and weight-for-age are measurements used to assess the growth and health of a baby or toddler. Since babies can’t even stand up for themselves (they are cute though), they are measured lying flat on an infant measuring board. That’s why we talk of length-for-age instead of height-for-age for children younger than two years old.

By taking repeated height (and weight) measurements at different ages, you can plot a child’s growth curve, tracking their progress over time and allowing comparison of their height or weight against what should be normal at each age (say, normal height and weight at 18 months). Here you can see the length-for-age WHO plot for girls, from birth to 2 years, including not just the normal (median) growth curve, but also 2 and 3 standard deviation curves below and above the median.

Once you have a large set of height and weight measurements for both boys and girls, you can construct normal distributions of both measurements at each age. The normal distributions of length (height) for girls are the ones used to create the pink plot above. At any given age, say 10 months, you could check the normal distribution and use it to classify a girl with a length above 76.5 centimeters as very tall, and a girl below 66.5 cms as stunted (or undernourished, in the case of weight). These thresholds, 66.5 and 76.5 centimeters in length, correspond to two standard deviations below and above the median of the 10-months normal distribution.

Where does the large set of height and weight measurements used to construct the normal distributions that the MPI uses as reference for children under 5 years come from? That question brings me back to the reference population I mentioned above. The MPI uses the normal distributions published in 2006 by the World Health Organization (WHO) as the Child growth standards.

From 1997 to 2003 the WHO conducted the Multicentre Growth Reference Study (MGRS), collecting primary growth data and related information from 8,440 healthy breastfed infants and young children in different countries: Brazil, Ghana, India, Norway, Oman and the U.S. In the WHO’s words12:

The growth standards developed based on these data and presented in this report provide a technically robust tool that represents the best description of physiological growth for children under five years of age. The standards depict normal early childhood growth under optimal environmental conditions and can be used to assess children everywhere, regardless of ethnicity, socioeconomic status and type of feeding.

The MGRS excluded children with illnesses that affect growth13 (approximately 2%), and included only children from high socioeconomic status families in the surveys conducted in the four low- and middle-income countries: Brazil, Ghana, India and Oman14.

This reference population was further selected for high compliance15:

leaving a sample of 1737 children (894 boys and 843 girls). Of these, the mothers of 882 children (428 boys and 454 girls) complied fully with the MGRS infant-feeding and no-smoking criteria and completed the follow-up period of 24 months (96% of compliant children completed the 24-month follow-up). The other 855 children contributed only birth measurements, as they either failed to comply with the study’s infant-feeding and no-smoking criteria or dropped out before 24 months.

This is the reference population.

Besides the under-5 definition of undernourishment, the MPI uses two other definitions: one for children aged 5-19 years, and another for adults in the 19+ to 70 years age range. The 5-19 year group definition is similar to the under-5 one but using age-specific Body Mass Index (BMI) normal distributions, and a threshold of two standard deviations below the mean. The adult definition classifies as undernourished any adult whose BMI falls below 18.5 kg/m2 (which is close to two standard deviations below the mean).

To summarize, the MPI classifies adults aged 19-70 as undernourished if their BMI falls below the the widely used underweight threshold of 18.5 kg/m2; classifies 5-19 children as undernourished by comparing their BMI to the normal distributions for their age and sex; and classifies children under 5 as undernourished by comparing their height (length) and weight to the normal distributions constructed from the reference population of the World Health Organization’s MGRS.

Well, that last part isn’t quite right. The normal distributions published by the WHO in its Child growth standards were not directly constructed from the raw measures taken during the MGRS. The raw distributions were first truncated16:

To avoid the influence of unhealthy weights for length/height, observations falling above +3 SD and below -3 SD of the sample median were excluded prior to constructing the standards. For the cross-sectional sample, the +2 SD cut-off (i.e. 97.7 percentile) was applied instead of +3 SD as the sample was exceedingly skewed to the right, indicating the need to identify and exclude high weights for height.

Between 0.7% (boys) and 0.4% (girls) of all observations were dropped in the longitudinal part of the study. I don’t know enough statistics to evaluate the effect of these exclusions on the final height and weight distributions, but I find it hard to believe they had no impact at all17.

Assumptions

Now that we have a clearer picture of where the WHO’s Child growth standards data comes from, let’s take a look at a few assumptions underlying the MPI’s use of this data to estimate child undernourishment, and consequently child poverty:

• Children from high socioeconomic status families have the same genetic growth potential as children from lower socioeconomic status families18.

• For a child of a given age, having a height, weight or BMI below two standard deviations from the mean of their age’s distribution is likely a consequence of experiencing poverty. There should be a positive relationship between higher likelihood of falling below two standard deviations from the mean and higher likelihood of experiencing poverty.

• The height and weight distributions of the populations selected for the MGRS are representative of the distributions that would be seen in all countries of the world in the absence of health or environmental constraints to growth.

The first assumption can be tested by looking at Genome-Wide Association Studies (GWAS) and data from biobanks. Biobanks are large repositories of biological samples, including genomes, often encompassing data from hundreds of thousands of individuals, and they can be used to answer questions such as: do height and BMI have a causal role in socioeconomic status? Which is precisely what a study by Jessica Tyrrell et al. did in 2016, using data from the UK Biobank.

By disentangling the effect of genetic variants known to affect height from other factors, the study concluded that part of the reason higher-income populations are taller is a genetic predisposition to greater height19:

In the UK Biobank study, shorter stature and higher BMI were observationally associated with several measures of lower socioeconomic status… Genetic analysis provided evidence that these associations were partly causal. A genetically determined 1 SD (6.3 cm) taller stature caused a 0.06 (0.02 to 0.09) year older age of completing full time education (P=0.01), a 1.12 (1.07 to 1.18) times higher odds of working in a skilled profession (P=6×10⁻⁷), and a £1130 (£680 to £1580) higher annual household income (P=4×10⁻⁸).

Granted, this study was conducted in the UK, and you could argue that high-income populations in other countries differ from the British. But the MGRS itself argues against this by noting that there’s little difference in height between the selected high-income groups from Brazil, Ghana, India20 and Oman, and the reference groups from Norway and the United States21.

If the purpose of the Child growth standards is to have a benchmark against which we can compare the growth curve of every child in the world, it seems like a bad idea to construct that benchmark from a selected group that is taller and bigger than most populations.

If the justification for using a benchmark to assess undernourishment is that any child falling more than two standard deviations below the distribution’s mean is likely well below their genetic growth potential, it seems like a bad idea to use a benchmark constructed from a population with a substantially higher mean.

I suspect that the MPI wrongly classifies many children as undernourished by incorrectly assuming their height or weight is well below normal, when in reality their seemingly very low values reflect the use of the wrong benchmark.

There are a few facts we should carefully consider before judging the WHO’s MGRS benchmarks or the MPI Nutrition indicators too harshly. On the one hand, when the MGRS was published in 2006, biobanks were still in their infancy, and the WHO could argue there was no reason to think that children from higher-income families were genetically predisposed to be taller (though this is clearly no longer the case today). On the other hand, there was plenty of data at the time showing that higher-income populations were, and still are, taller on average even in countries with very low levels of poverty. In defense of the WHO, you could propose other causal mechanisms for shorter stature, distinct from poverty, but somehow correlated with levels of income. But then this last hypothesis wouldn’t excuse the use of short stature as a proxy for poverty by the MPI.

Richer and thinner

Regarding the second assumption, I haven’t seen any data that directly contradicts the presumed relationship between poverty and very low height or weight. But I can’t say the same thing about BMI, which the MPI uses to determine undernourishment in everyone aged 5 and older.

I took advantage of my familiarity with the Spanish National Institute of Statistics (INE) website to quickly check the data on children’s BMI, binned by occupational skill level of mother or father. This is a rough way of looking at children’s BMI by household income level, assuming the six skill categories roughly map to household income22. Here’s a chart showing the share of boys and girls aged 2 to 17 who are underweight, by household income level.

The highest percentage of underweight boys is found in the highest skill (highest income) category, while the highest percentage for girls occurs in the “Supervisors and workers in skilled technical occupations“ skill level (middle income?).

I might be mistaken (I’m open to see evidence to the contrary), but my impression is that the same pattern we see in Spain, of higher shares of underweight children in higher income households, is mirrored by most middle-income countries.

If that is the case, then it’s wrong to assume that a higher likelihood of child’s BMI falling below two standard deviations from the mean indicates a higher likelihood of experiencing poverty, at least in middle-income countries.

Rugby against poverty

Finally, regarding the third assumption, there’s plenty of data from rich countries, where health and environmental constraints to grow should be minimal, showing that average heights vary substantially. Norway, one of the countries selected for the MGRS, has an average male height of nearly 180 centimeters, while Japan, Singapore and Taiwan show averages between 170 and 172 cms.

If these large gaps in average height stem from differences in genetic growth potential between wealthy populations23, shouldn’t we expect to see similar differences among low- and middle-income countries also?

To explore this idea, I decided to chart Nutrition deprivation in the 0-9 age group against GDP per capita for all countries in Latin America, Europe, and the MPI-defined East Asia and the Pacific region24.

I’ve colored groups of countries that stand out due to their populations being exceptionally tall or short. The populations of South East Asia (green on the map) are noticeably shorter, just like the indigenous populations of the Americas. By contrast, the populations of the Western Balkans (red on the map) are exceptionally tall, as you can see in the following map of average male height by country.

Notice also that I marked Guatemala in the chart, the Latin American country with the highest indigenous ancestry in the sample, with an arrow. Guatemala has by far the highest reported rate of Nutrition deprivation among 0-9 year olds, even though its income level is near the average for this sample of countries.

Nevertheless, I believe the most striking country in the chart is Tonga. I colored the country yellow, together with other two rugby-friendly countries of Oceania: Fiji and Samoa. The three island nations are well known for their tall (and big) populations and their unsurprising fondness for Rugby, a sport in which they often hold their own against countries several times their population (not body) size. Tonga’s males, the tallest of the three populations, average a height of 177.9 centimeters.

It sure seems like taller countries have a lower rate of Nutrition deprivation than expected, and shorter countries have a higher rate of Nutrition deprivation than expected.

Notice that if height in middle-income countries25 is substantially determined by genetic factors, using BMI as a reference for undernourishment shouldn’t produce misclassification due to taller or shorter averages. BMI measures weight relative to actual height, and therefore shorter children are expected to weight less. But if children 0-9 are being misclassified as undernourished, this also affects the Nutrition indicators of 10-17 children and adults, since at least some individuals in those age groups live in the same households as children aged 0-9.

Putting it all together

To summarize: the Child mortality indicator is not just imperfect, but in regions of the world with high levels of violence, it’s likely a better index of exposure to violence than of poverty. I also suspect that the Years of schooling indicator is misleading as a measure of poverty for children in general, and younger children in particular, usually appearing to show a trend - higher child poverty among younger children - that is artificial.

Yet my biggest qualm is with the Nutrition indicator. I can’t think of any good reason to use the high-income populations of the six countries selected by the MGRS as a reference against which to measure the growth of every child on the planet. I can understand that in low-income countries, a share of population substantially higher than 2% falling below the mean for height, weight or BMI strongly suggests undernourishment in the population. In such cases, I think the MPI Nutrition indicator can provide meaningful information on child poverty, but I doubt that it does for middle-income countries.

I think it’s telling that, in middle-income countries with above-average height, the share of households with at least one presumed undernourished member seems to fall between 6% and 10%, which is roughly the percentage you would expect by random chance in households with three to five members.

Looking at middle-income countries that lack the Nutrition indicator is quite revealing: Vietnam and Indonesia, both lacking Nutrition data, show nearly the same overall MPI for the 0-9 and the 10-17 age groups. And Brazil has a lower overall MPI for children aged 0-9 than for those aged 10-17. If we examine the MPI Indicator contribution charts for these countries, and those of other comparable middle-income countries26, the large contribution of the Child mortality indicator to their overall MPI values stands out very clearly, particularly in urban areas.

In the absence of Nutrition data, Uzbekistan’s MPI is almost completely explained by Child mortality. In urban areas, Brazilian and Indonesian MPI values are roughly two-thirds attributable to Child mortality, and urban Colombia’s MPI is around half Child mortality.

This suggests that for many middle-income countries, Child Mortality and Nutrition explain most of their MPI value in urban areas; and in some cases, in both rural and urban areas. This is worrisome, given what I’ve explained about these two indicators.

You can see this also in the case of Mexico: it has Nutrition data yet lacks Child mortality data, and its overall MPI is almost 80% explained by the Nutrition indicator. Besides Mexico’s chart, the following graph includes the MPI Indicator contribution charts for Argentina, Kazakhstan and China.

As you can see above, Kazakhstan’s overall MPI is almost completely explained by Child mortality and Nutrition, and Argentina’s is roughly 70% explained by those two indicators.

This post may sound as an overly harsh critique of the MPI and OPHI’s work. It’s not intended as such. I truly believe that the Oxford Poverty and Human Development Initiative does great work, and that its researchers make their best effort to measure international poverty, and child poverty in particular.

But I do believe that the MPI, and any similar index, should be scrutinized and analyzed in an effort to improve its validity and strengthen its use as a tool in international poverty comparisons and trend analyses. I hope the creators of the MPI and those who disseminate it (such as Our World in Data) take this post in that spirit.

Yes! I’m finally here in the Canary Islands.

And I regret to report that the Spanish Canarian accent is slowly dying - faster in the cities than in the countryside (or what passes for countryside with so much tourism in every corner of the islands). In a way, finding so many Venezuelan and Cuban immigrants around here makes me happy because at least I get to hear the closest thing to an authentic Canarian accent that you can find.

The connection to Canari America can be seen everywhere, from the countless Cuba streets, Uruguay squares and Venezuela avenues, to the frequent references to the “Indies commerce” in museums and historically significant sites.

The focus of this post, though, is not the natural beauty of the Canaries or the long relationship between the archipelago and its American offshoots. What this post is really about is a recently published book called Marriages of the parish of Our Lady of the Immaculate Conception of Agaete 1614-1925, by Javier Gil Pérez.

Note that I haven’t read the book yet. I’ve only read a short interview with the author, recently published by a Canarian newspaper, yet I thought the piece was so interesting and illuminating that it merits its own post.

The book collates a list of 3,153 marriages which took place in the parish church of the town of Agaete between 1614 and 1925. This large dataset enabled the author to analyze the recorded marriages in detail, finding patterns, some expected and some surprising, which are likely applicable to most Canarian towns during the relevant historical period. Among the expected, a high degree of endogamy in the town:

The statistical analysis shows a high rate of dispensations, which are permissions granted by the Bishopric through the parishes, when the spouses were cousins, up to the fourth degree. That is, when they shared common great-great-grandparents. And in a town as geographically isolated as Agaete was 120 years ago, endogamy was very common within the municipality.

The author also notes that certain surnames of spouses and spouses’ parents show up more often, and lists the ten most frequent ones. Some of these are very common patronymics found throughout Spain (e.g. García, González), but others can be described as distinctively Canarian given their much higher frequency in the archipelago than on the Spanish mainland: Armas1, Rosario, and Sosa.

He then talks about the relative decline, and even disappearance, of certain surnames in Agaete2, while also touching on the history of Canarian migration to the Americas:

The surname Palmé, which was once considered unique to Agaete, has disappeared. In the book you’ll see its origin: a man from La Palma was known in the town as “Palmé,” and this name was passed down to his descendants as a surname. There’s a curious story from the early 20th century about the wealth of a married couple from Agaete who emigrated to Cuba and both died, without children. A call went out in Agaete for relatives to come forward and claim their inheritance. Many changes were then requested at the Bishopric to add the surname Palmé to their genealogy.

Also tangentially related to Latin America is his description of how indigenous Canarians adopted Castilian surnames:

The indigenous [Guanche] people were baptized with surnames that were mostly Castilian. For example, a surname like Lugo, which has been mistakenly thought to derive from the Adelantado Alonso Fernández de Lugo, actually comes from baptized indigenous people. Hernández, from Fernando Guanarteme, is also an example of this. His direct descendants, nephews, and cousins ​​take the surname Hernández. Another surname very common in Agaete is Sánchez, through a relative of Fernando Guanarteme, Hernán Sánchez de Bentidagua, who was the royal mayor of Agaete in the early stages of colonization.

Yet, in my opinion, the most interesting bit comes near the end of the interview, when Gil Pérez talks about single motherhood and foundlings in Canarian society:

It also strikes me that, regarding marriages, there are over 300 spouses who were children of single mothers, in a society so closed and influenced by religion. We [also] have documented cases of foundlings or Santana children, who were placed in cradles, and these help us understand the social reality of women and children in those centuries. When mothers were single, the children took their mothers’ surnames. There were certain social stigmas. But, in my opinion, the connection with the port of Las Nieves, open to Europe through the export of sugar from the mills, made Agaete, despite its limitations, a more open town than those in its surroundings.

That is 300 spouses out of more than 6,000. Very roughly a 5% illegitimacy rate, which doesn’t seem remarkable in the Spanish context3. But his mention of foundlings and Santanas did caught my attention. Spanish foundlings - newborn children abandoned by their mothers under dire circumstances - were usually given the surname Expósito (exposed) by the hospitals and religious institutions that raised them. And, in addition to Expósito, different regions of the country had their own local variants of foundling surnames. In the case of Gran Canaria, where Agaete is located, the surname Santana was commonly given, taken from Saint Anne, patroness of the cathedral church of Gran Canaria’s capital, Las Palmas.

If we use these two surnames as a proxy for the prevalence of foundlings in different Spanish regions, the Canary Islands does seem exceptional, showing by far the highest frequency of the surname Expósito4. The surname Santana is of course more common in the Canaries5, where it originated, than in other regions, but we can tentatively compare its frequency there to that of other local foundling surnames within their respective regions6:

• Santana (Canary Islands, especially Gran Canaria): 1 in 45.

• Deulofeu (Catalonia, especially Barcelona and Girona provinces): 1 in 13k.

• Gracia7 (Aragón): 1 in 98.

• Trobat8 (Balearic Islands, especially Mallorca): 1 in 4.7k.

As you can see, only the Aragonese surname Gracia comes close to the high frequency of Santana. Yet, at the same time, Aragón has the third-lowest prevalence of the Expósito surname among Spanish regions, which makes me think that illegitimate births - which were likely highly correlated to the number of foundlings - were substantially lower in Aragón than in the Canary Islands.

In fact, I believe the rates of foundlings and illegitimate births in the Canary Islands were likely the highest in Spain, just like today’s Canarian out-of-wedlock rate, as I noted in my post about the Canari American marriage pattern:

And that would be the end of this post (don’t worry, there’s yet another pretty picture at the end), if I hadn’t remembered some figures from the 1941 Venezuelan census. As you might recall from my post on Venezuelan election results, the regions of Venezuela that experienced the heaviest Canarian settlement show political leanings similar to those of the Canary Islands within Spain, in contrast to the Andean region, which traditionally leans right.

The migration of Canarians to Venezuela did not exactly mirror that to Cuba, due to differences in their geographic characteristics. While Venezuela boasts a lengthy Caribbean coastline, it also features a temperate mountainous region in its southwest…While most Spanish settlers favored the temperate regions (circled in yellow in the [following] map) over the tropical coastal areas (circled in red in the [following] map), Canarians predominantly migrated to the Venezuelan coast and nearby valleys.

The three states that comprise Venezuela’s Andean region (circled in yellow in the previous map) are Táchira, Mérida and Trujillo. I suggested in that previous post that they tend to vote differently from the rest of Venezuela because they are less Canarian, in contrast to the rest of Venezuela, which to a significant degree inherited its political culture from the Canary Islands. This begs the question of whether we can say something similar regarding marriages and births. And at least in the case of marriages, we can.

These are the 1941 percentages of married individuals (aged 15 and older) for each Venezuelan state:

Excluding the federal district of Caracas - by far the largest city in the country - and Zulia State (the center of the oil industry), the three Andean states have the highest rates of married individuals in Venezuela9.

Regarding the relationship between marriages and births, the Andean region had a slightly higher birth rate than the country as a whole10. This fact, combined with its higher percentage of married individuals, strongly suggests that a larger share of births happened within marriage in the Andean region compared to the rest of the country. In other words, it’s very likely that the rate of out-of-wedlock births in the Caribbean regions of Venezuela was substantially higher than in the Andean region.

And yes, I believe this is because Venezuela’s heavily Canarian-settled regions inherited these marriage and birth patterns from the Canary Islands, just like they inherited other cultural patterns, in contrast to Andean Venezuela.

I now leave you with a lovely picture of Garachico, where in the late 16th century some of the ancestors of Simón Bolívar, the Venezuelan-born liberator of Spanish South America, were born.

Instead of quoting the report directly, I’ll quote one of the well-meaning organizations that helped to disseminate it (in fact, I think it was their mention of the Canary Islands that first caught my attention, because, well, almost anything that mentions the Canary Islands grabs my attention).

According to recent data, about 1 in 3 children in Spain, or around 34%, live at risk of poverty or social exclusion. In late 2023, a UNICEF report named it the worst in the EU for child poverty, ranking it 36th out of 39 EU and OECD countries.

Save the Children has published similar findings - for example, that 1 in 3 Spanish children live in homes that cannot afford a summer holiday. Ironically, almost half of them live in the Canary Islands, one of Europe’s major tourist destinations1.

This casts Spain in a pretty negative light; the higher in the ranking, the higher level of child poverty. Although, if we take a look at the ranking table we’ll see that one of the three countries below Spain is Colombia (39th), which is unsurprising as it is the poorest country in the report2; another one is Turkey, also poorer though less so3; and the third one is the United Kingdom (37th), a fairly rich country… so there’s that. And despite all the talk about British decline, the UK is richer than Spain. So, why do these two wealthy countries have such high levels of child poverty?

Innocenti’s ranking methodology partly explains the UK’s position. Each country’s ranking is “based on [its] latest available rate of child income poverty and [its] success in reducing child income poverty”, allowing a country like Romania, with a higher rate of child poverty than Spain and the UK, to end up in a better position in the composite ranking - the UNICEF Innocenti rank - thanks to an impressive reduction in child poverty in the last few years4.

Notice that the ranking table above shows (surprisingly?) other rich countries, like the United States, France and Italy also near the bottom of the composite ranking. And before considering any recent changes in child poverty, Italy and the U.S. still rank very low on the child poverty index itself (33 and 35), nearly as low as Spain.

So maybe the United Kingdom’s 28th position in the child poverty index is not shocking (though 28 out of 39 still seems low), and its low rank in the composite ranking can be explained by a lack of progress in reducing child poverty in recent years. But I still think the positions of Italy, Spain and the U.S. are, if not shocking, at least surprising.

Given that all countries in the report are either high income or upper middle-income (EU or OECD members), perhaps small variations in the child poverty rate among countries that are on a similar (?) wealth level produce large variations in the ranking?But the child poverty rate varies widely between the reported countries: from a low of 9.9% in Denmark to a high of 28% in Spain5. This is clearly visible in a scatter plot of child poverty rate vs GDP per capita (PPP).

It’s hard to see any trend in that plot6. There is a weak correlation between child poverty and GDP per capita of approximately -0.30, but a very noisy one (R² = 0.08). As the report says: “there is only a weak tendency for richer countries to have lower child poverty rates“.

I wouldn’t just call it weak, but suspiciously, surprisingly weak. What’s going on?

A small clue to this puzzle is the fact that smaller countries dominate the top of the child poverty ranking: Denmark, Slovenia, Finland, Czech Republic, Norway, Iceland, etc. It’s as if a smaller population size has an effect in reducing child poverty. A regression model to predict child poverty from population size seems to confirm this intuition: the correlation coefficient is 0.33 (R² = 0.11), suggesting that population size explains slightly more of the variation in child poverty across countries than do variations in wealth.

What’s going on?

Well, after reading the whole report I realized the mystery was not so mysterious after all. The main source for the child poverty rankings is the Eurostat Database, maintained by the European Statistical Office (a department of the European Commission).

And the Eurostat index interpreted by UNICEF as child poverty is the at-risk-of-poverty rate from the Income and Living Conditions dataset7, which uses a threshold of 60% of median equivalized income after social transfers. Children living in households whose “per capita“8 income falls below 60% of the national median income (after social transfers) are classified as living in poverty9.

National just means that each country’s median income is calculated at the national level (more on that later), and equivalized means that all income sources (even social benefits) are added, while all taxes and social contributions are subtracted10.

So, what UNICEF is really saying when it ranks Spain 36th (or the U.S 35th) in child poverty among 39 high income or upper middle-income countries is that Spain has the 36th fourth largest share of households with equivalized income below 60% of the national median. This is a measure of inequality, not of poverty.

If median household income is $1,000, any child living in a household with an equivalized income below $600, regardless of actual material conditions, is considered poor.

To illustrate what “regardless of material conditions” implies, think of two countries: one with a median income of $1,000 and the other, relatively richer, with a median income of $1,500, yet both have identical income distributions. Depending on the income distribution (variance, skewness), most children classified as “poor” in the richer country could enjoy better material living conditions than any of the children in the (relatively) poorer country.

If we understand poverty as the lack of essential economic resources, then UNICEF’s Innocenti Report Card child poverty ranking does not measure poverty. And this is not just my opinion but also what Eurostat (the data source used by UNICEF) says:

The at-risk-of-poverty rate is the share of people with an equivalised disposable income (after social transfer) below the at-risk-of-poverty threshold, which is set at 60 % of the national median equivalised disposable income after social transfers. This indicator does not measure wealth or poverty, but low income in comparison to other residents in that country, which does not necessarily imply a low standard of living.

Why does UNICEF present an inequality index as if it were a measure of child poverty? I truly don’t know.

You might argue that, given the same level of wealth, a more unequal country should have more poverty than a less unequal one. And you would be right. I know this because Eurostat also collects data on severe material and social deprivation in European countries, and the correlation between percentage of children (under 18 years) experiencing severe deprivation and the share of children in households below the 60% threshold is 0.65 (R² = 0.43), a correlation that strongly suggests more unequal countries tend to have higher levels of child poverty when measured by actual material conditions.

On the other hand, if there is reliable data on child poverty measured by actual material conditions, why doesn’t UNICEF publicize it?

To clarify how Eurostat measures material and social deprivation, here are the 13 items it assesses for their presence or absence in households:

• capacity to face unexpected expenses.

• capacity to afford paying for one week annual holiday away from home.

• capacity to being confronted with payment arrears (on mortgage or rental payments, utility bills, hire purchase installments or other loan payments).

• capacity to afford a meal with meat, chicken, fish or vegetarian equivalent every second day.

• ability to keep home adequately.

• have access to a car/van for personal use.

• replacing worn-out furniture.

• having internet connection (individual level).

• replacing worn-out clothes by some new ones (individual level).

• having 2 pairs of properly fitting shoes (including a pair of all-weather shoes, individual level).

• spending a small amount of money each week on him/herself (individual level).

• having regular leisure activities (individual level).

• getting together with friends/family for a drink/meal at least once a month (individual level).

The percentage of children in poverty is defined by Eurostat as the proportion of underage population lacking at least 7 out of the above 13 items. On this index, which actually measures material poverty, Spain ranks 13th and Italy 17th in child poverty among 34 European countries. Not great, but hardly headline grabbing.

More importantly, this measure of child poverty shows that fewer than 1 in a hundred Spanish (or Italian) children live under conditions of severe material and social deprivation. Child poverty is uncommon in rich countries.

Very uncommon.

But that’s not the whole story. Remember I said smaller countries appear to experience lower child poverty under the flawed 60% income at-risk-of-poverty index? That was just a product of lower economic inequality in smaller countries, which tend to have more homogeneous populations.

Now, let’s think of large countries as composed of smaller countries. An homogeneous large country is composed of several similar regions, with similar populations, similar average incomes and income distributions. Its 60% of national median income doesn’t differ much from the 60% of median income in each of its regions. On the other hand, a large heterogeneous country is comprised of multiple regions with not-so-similar populations, with differing average incomes and income distributions.

And which European countries are large (by European standards) and heterogeneous? That’s right: Spain and Italy.

Back to Spanish regional differences (yes, again)

This brings us back to the Canary Islands - and also to Southern Italy. Does the actual child poverty index show the same huge differences in poverty between Spanish and Italian regions as the flawed index did?

in Italy, there are long-standing disparities in wealth between the north, the centre and the south. This means that, in 2021, the highest regional child poverty rate in Campania in the south (39 per cent11) was four times as high as the lowest rate in Trentino-Alto Adige in the north (9 per cent12).

Although Eurostat doesn’t provide data on severe material deprivation among children broken down by European regions, it does provide such data for the overall population. And the correlation between severe material and social deprivation in children and the overall population is higher than 0.8, so we can use the general population percentage as a good approximation for the underage percentage.

Here are the percentages of the population suffering severe material and social deprivation in each Spanish and Italian region:

So yes, when using a proper poverty index, the Canary Islands exhibit the highest level of overall poverty in Spain - around three times higher than the lowest regions - and therefore likely the highest child poverty rate.

The gap between poor regions and rich regions is even wider in Italy: ignoring Bolzano’s exceptionally low rate, Campania and Calabria have poverty rates more than 10x those of Italy’s richest regions. One would expect to see these kinds of poverty gaps between countries rather than between regions of the same country.

Spain and Italy are very unusual European countries indeed.

Well, the data on divorce rates shows no evidence of the proposed mimetic influence.

However, that’s just one aspect of Israeli family life that I’ve examined. I’m not saying that divorce is a minor feature, but it’s just one among several measurable aspects of family life that could (or should) be impacted by the theory’s mechanism.

If we want to take a look at aspects more closely related to fertility, then choosing divorce as the first one to analyze was a mistake. Before divorce comes marriage; in fact, before having children, most couples marry. Perhaps I should have looked at marriage data before reviewing the data on divorce rates. So, let’s do that now.

But before I continue, a caveat: the rest of this post assumes you are familiar with the four religiosity groups in Israel’s Jewish population. If you are, you can safely skip the following note. If not:

(Note: The four Jewish Israeli religiosity groups are, from most religious to least: Haredi, Dati, Traditionals and Seculars.

The Haredi - also known as Ultra-Orthodox - usually live in close-knit communities, devote much of their lives to the study of the Torah and generally follow a very conservative lifestyle. They constitute roughly 8%-9% of the adult population of Israel, but around 13% of its total population due to their exceptionally high fertility.

The Dati, also know as Religious Zionists (classified as Religious / Very Religious by the CBS), practice their Jewish faith rigorously yet do mix and mingle with other groups far more than Haredis. Approximately 10% of Israel’s population is Dati.

Traditional Israelis are neither strictly religious nor secular, following Jewish religious precepts to varying degrees. They are predominantly of a Mizrahi (Middle Eastern) origin rather than an Ashkenazi origin.

Finally, Secular Jews are, as their name implies, non-religious, and comprise the largest group among Jewish Israelis.)

Haredi and Dati marriage

Haredis (Ultra-Orthodox) Israelis give huge importance to marriage. In fact, the emphasis they put on marriage seems to be increasing over time, to the point where the proportion of women who have never married is higher in older age cohorts than in younger ones.

It’s very rare for a Haredi woman to reach old age without having married, and it appears to be getting rarer. On average, in overwhelmingly Haredi localities, the percentage of women and men 65 and older who have never married is close to 1%. Not zero, but pretty low.

I won’t show you the table with never-married percentages of 65+ women in overwhelmingly Haredi localities because it adds little of value. But I will show you the corresponding Dati table.

A caveat: data on never-married Israelis aged 65 and older in local councils is much harder to find than divorce statistics, forcing me to drop local councils from the Dati sample1. To compensate for this I had to lower the “overwhelmingly Dati” threshold (again), so take these percentages with a big grain of salt. The new threshold is 60% of households or more.

That being said, the above table suggests large variation across localities in the number of never-married 65+ Datis, while implying that unmarried men and women in the oldest cohorts are rare (1.33% for 65+ women), though not as rare as among Haredis. It appears that marriage is almost as important for Datis as it is for Haredis.

Yet, I haven’t explicitly stated what is the yardstick against which I’m measuring Haredi and Dati marriage. Is a 1% or 1.33% never-married rate truly low? And compared to what?

To answer those questions, let’s move on to the next two religiosity groups in Israeli society: Traditional and Secular. As in my previous post, I won’t be able to estimate the Traditional never-married rate directly, but we should get a rough idea of its magnitude, along with a more precise estimate of the Secular rate.

Traditional is marriage

Just like I did in my analysis of divorce rates, I assembled two samples of localities that are overwhelmingly Secular and Traditional2; one of them is more Traditional-shifted, with a larger share of Traditional households (28% or more)3, while the other is more Secular. However, as I’ve already mentioned, the localities for which Israel’s Central Bureau of Statistics (CBS) publishes marital status data are not the same as those for which it publishes divorce data4. As a result, both samples are larger than the ones I used in the divorce post.

First, let’s see how the Traditional-shifted sample does, and please note that the average never-married percentage shown below is unweighted (not adjusted for population sizes).

These percentages of never-married 65+ women and men are significantly higher, often double or more, than the percentages for the Haredi and Dati communities. But are they as high as those seen in most developed countries?

Here are a few never-married rates for selected countries:

• United States: 6.4% of women aged 65+ and 6.9% of men aged 65+ have never married. Source is 2022 American Community Survey, and it’s corroborated by the census’ Current Population Survey (see Table A1 “Marital Status of People 15 Years and Over, by Age and Sex: 2022“).

• Germany: I can’t find the exact numbers, but according to the Federal Office of Statistics it appears to be between 5% and 6% for both sexes.

• England and Wales: Roughly 5.1% of women aged 65 and older and 7.8% of men. Source is 2021 census (see Figure 2 and accompanying Excel file).

• Japan: 4%-6% of 65+ women and 6%-8% of men. Source: Grok. [UPDATE: Japan’s rate is 5.2% for both sexes.]

• Canada: 7.9% for women and 8.5% for both sexes according to Statistics Canada data.

The answer is no. The never-married rates in the Traditional-shifted localities sample are substantially lower than rates in developed countries. And of course, the Dati and Haredi rates are extremely low compared to developed country rates.

Regarding divorce rates, I didn’t see any evidence for mimetic influence on the Traditional community, but now I tentatively do for never-married rates. Since there are no overwhelmingly Traditional localities whose marriage data I can examine, I must rely on the mostly Secular yet Traditional-shifted localities in the sample above to get a sense of what the Traditional rate might be. And it’s almost certainly lower than the average rates of developed countries, which aligns with the predictions of the mimetic theory.

So, one point for the mimetic theory.

Now, let’s check the second sample, comprised of Secular-shifted localities.

Well, that’s certainly higher. My very rough estimate puts the never-married rate for 65+ women in overwhelmingly Secular localities at 5.48%, well within the developed country range. Though the men’s rate appears to be lower than the typical percentage of never-married men in developed countries.

There’s a couple of points I want to make about these numbers. First, these “overwhelmingly” Secular localities are not 90% Secular; not even 80% Secular. The true Secular never-married rates are likely higher than these. Also, the gap between the >= 28% Trad sample and the <= 22% Trad sample is substantial, two and a half percentage points, which suggests that the Traditional rate is quite low, but also that a hypothetical 100% Secular population (0% Trad) would show a never-married rate significantly higher than 5.48%.

Second, the Secular rate for men is clearly lower than for women5, contrary to what appears to be the case in most developed countries. I’m not sure what to make of this, and I hesitate to link it to the mimetic theory, partly because if the true male rate is actually higher, it might very well fall within the developed country range.

So maybe the Secular men’s never-married rate lends some support to the theory; but maybe not.

Did you happen to notice that each of the three never-married rate tables (Dati, Traditional-shifted, Secular-shifted) includes two additional columns besides the ones for 65+ women and 65+ men?

These additional columns show the percentages of never-married women and men between the ages of 55 and 64. By comparing them to the values in the 65+ columns we can try to determine whether never-married rates are increasing or decreasing over time in the Dati and Secular communities.

In contrast to what I observed for the Haredi, Dati never-married rates appear to be slightly increasing in younger generations, though that’s far from certain. But in the case of Secular Israelis the trend is very clear: men and women in the younger cohort (55-64) are substantially more likely to have never married than those in the 65+ group6.

This also challenges the mimetic theory, since never-married rates in developed countries appear to be increasing, just like Secular Israeli rates, rather than decreasing like the Haredi ones.

From no marriage to yes marriage

Since we are already on the subject of marriage, let’s look at another marriage index: median or mean age at first marriage7. I apologize in advance for switching between mean ages and median ages; international data on first marriage is not as standardized (or abundant) as I would like it to be.

To estimate the median age at first marriage for each Jewish Israeli religiosity group, I used the same methodology as before, calculating the average age over a sample of localities where the overwhelming majority of households belong to that group. I’ll skip the tables (you can find them in the footnotes8) and go straight to the results.

If you are familiar with marriage ages in developed countries, you are probably thinking now that Secular Israeli men and women marry very young (27.9 and 25.5) based on the above chart. There are two problems with that conclusion though. The first is that, same as before, the “Secular sample” is not 100% Secular, and therefore the true Secular median ages are higher than 27.9 and 25.5 (notice the sizeable age gaps between Traditional-shifted and Secular-shifted samples: 1.4 for men and 2.1 years for women).

The second issue is that the CBS reports the median age at marriage for the whole population, not just those who married during the census year (2022). Since marriage ages in most developed countries (including Israel) were lower in previous decades, these figures are biased downward by the earlier marriages of older people.

So, what is the true mean age at first marriage of Secular Israelis? The United Nations Economic Commission for Europe (UNECE) publishes age at first marriage data for Israel and several other developed countries, allowing us to treat the Israeli figure as a weighted average across all four religiosity groups9, and compare it with other countries.

The national Israeli mean is slightly below average, but well within the developed country range. And we’ve already seen that Secular Israelis marry substantially later than Haredi and Dati (and very likely Traditionals), so we can safely estimate their mean age at first marriage as higher than 30.6 years for men and 28.2 for women.

To summarize, Haredi men and women marry very early, much earlier than the typical marriage age in developed countries; Dati also marry early, though not quite as young as Haredi; Traditional Israelis appear to marry somewhat early by developed country standards (though I can’t be 100% certain); and finally, Secular Jews do not marry early at all. The Secular do not differ from populations in other developed countries when it comes to marriage age.

Three makes a crowd?

I’ve now documented three measurable aspects of family life where the data on Secular Israeli Jews does not appear to support the mimetic theory of Israeli fertility: divorce rates, never-married rates, and age at first marriage.

In all three cases the data could support Haredi and Dati influence over Traditional Jews, but Secular Jews show no such influence when their marriage and divorce rates are compared to those in other developed countries. This can be interpreted as evidence that the mimetic theory is correct when applied to Traditional Israelis, but that was not the main motivation to put forward the theory. The main motivation was to explain the exceptionally high Secular Israeli fertility rate.

Maybe fertility is a unique aspect of family life, and Secular Jews are influenced by religious Israelis in that particular aspect and no other. Perhaps the three aspects I’ve examined are not the most relevant ones, and I should take a look at another one(s) (suggestions are welcome).

I don’t know whether this analysis will persuade anyone for or against the mimetic theory, but I do hope it will move the discussion forward.

8

Tables of median marriage age for the four samples:

In fact, second lowest in the world, according to Wikipedia. Here’s a theory:

Haredi Jews (the ultra-orthodox) have exceptionally low divorce rates wherever they live. For any reason that a married couple might cite for divorce, Haredi society will counter it with a better reason to stay married. Thing is, in Israel, there are other religious communities - less religious than Haredi - who draw inspiration from the Haredi and are influenced by them. These communities, like the Dati, provide a one-way channel of influence between Israeli Haredi and the rest of Israeli society, a sort of valve in which Haredi marriage/divorce memes can spread outwards, without allowing pro-divorce memes inwards1.

This theory - let’s call it mimetic theory of Israeli divorce rates - explains not only Israel’s very low overall divorce rates, but also why even the most secular group in the country, Secular Jews, have lower divorce rates than those seen in other developed countries. Anti-divorce memes propagate step by step, from the Haredi community all the way down to the Secular community.

Now that we’ve got a theory, what should we do with it?

I now want to test the theory against the available data. If the theory successfully predicts what the data shows, that would strongly suggest it’s right. If it can’t predict correctly, that would make me doubt it’s validity.

But before we proceed, let me clarify that the old 27.69% divorce rate (year 2009) shown in the image above is very similar to the current rate. This rate is calculated as the number of divorces that took place in a given year, divided by the number of marriages2.

Once again, I will use data from Israel’s Central Bureau of Statistic (CBS) to estimate divorce rates, taking advantage of the very different distributions of religiosity groups across Israeli cities, just like I did in the last couple of posts. To arrive at the divorce rate of a city in a certain year (let’s say a Haredi city), I’ll just divide the number of divorces recorded that year by the number of marriages in the same year.

Actually, the CBS also publishes marriage and divorce numbers for local councils - an administrative unit for smaller towns - and regional councils (grouping several minor localities), allowing us to broaden our search for overwhelmingly Secular localities, overwhelmingly Dati localities, etc. This will enable us to examine a larger sample and calculate a rate for each locality in the sample, beginning with overwhelmingly Haredi localities.

So, is Haredi divorce as rare as the theory says?

These are the average divorce rates (years 2021, 2022) for the five Israeli cities and local councils where more than 80% of households are Haredi:

So, the Haredi divorce rate is somewhere around 11.6%3. That’s higher than Vietnam, yet far below the divorce rate of any developed country. So far, so good. Let’s move on to the Dati.

Since the Dati are (apparently) less averse to mixing with other Israelis than the Haredi, it’s harder to find localities that are overwhelmingly Dati. This means I had to relax the requirement for what constitutes “overwhelmingly”, and accept any locality with more than 70% Dati households. And here are the results, averaged across 2021 and 2022.

For comparison, North Macedonia and Ireland seem to be the only European countries which show divorce rates clearly lower than the Dati one. This is precisely what the mimetic theory predicts. The Dati community has a very low divorce rate, around 20%, but not as low as the Haredi community.

Ideally, we should now check the next step in the mimetic transmission chain: the Traditional religiosity group, constituted by those who are partially observant yet not secular. But, as I mentioned in my previous post, there are no overwhelmingly Traditional cities and I couldn’t even find local councils or regional councils that could be described as such. So, I’ll skip the Traditional group, and go straight to the least religious group, Secular Israelis.

One step forward, one step backwards

Let’s now see the table of divorce rates for overwhelmingly (more than 70% of households) Secular administrative units.

35.56% is certainly higher than the Dati rate, but still lower than almost every developed country, except Japan (divorce rate 35.42%). Japan is weird, right? That’s it. The mimetic effect of low-divorce Haredis, and Datis, is strong enough to produce a weirdly low divorce rate in Secular Israelis…

…yet, a few localities look suspiciously high-divorcy. In fact, most localities have a suspiciously high rate, and the average total rate is mostly a product of Tel Aviv-Yafo having the most population by far.

At this point I feel obliged to note that 2021 was not a normal year4. It was a Covid year. Maybe the large variation in rates across localities is driven by the disruptions that took place in 2021. Just to be on the safe side I decided to calculate the divorce rates again, expanding the time period to four years: 2019-2022. With the addition of 2019 and 2020, the average divorce rate now raises to 40.59% - you can see the table in this footnote5. But instead of gloating over the higher rate (which would be highly inappropriate; that’s a lot of divorces), I’ll try to do something more interesting.

While I was updating the divorce rates table with 2019-2020 data, I remembered that Tel Aviv-Yafo is, strictly speaking, a mixed city. Mixed cities are those with more than 10% of the population registered as “Arabs” and more than 10% of the population registered as “Jews”. And even though Tel Aviv’s current Arab population is no more than 5% of the total (higher historically), it has such a high concentration of “Others” (non-Arab, non-Jewish residents) that the Jewish population of the city is only 84.2% of the total. And while most of that 84.2% are Secular Jews, not all of them are, reducing the Secular Jewish share to maybe 60%-65%. Tel Aviv has a lot of non-Secular or non-Jewish population.

But rather than continue by excluding Tel Aviv from the sample, I decided to go one step back. Remember how we skipped the analysis for the Traditional group? I now wanted to estimate what the Traditional divorce rate might be, even if very roughly.

First, I calculated the divorce rate for cities6 where 90% or more of households are either Secular or Traditional, with a Traditional share below 24%, to ensure the sample is overwhelmingly Secular. Then I assembled a group of cities where 85% or more of households are either Secular or Traditional, with at least 26% Traditional. This second sample is shifted towards Traditional Israelis and therefore allows us to see whether divorce rates differ from the Secular sample and, if so, how big is the difference.

And here are the tables for the first and second group of cities, with their computed divorce rates:

There’s a gap between average divorce rates for the Secular-shifted and Traditional-shifted samples, yet it’s a small one.Admitting the limitations of this approach to estimate divorce rates, I believe the Traditional rate does not fall below 40%. And as for Secular, it’s almost certainly above 50%.

Conclusion

As you probably know by now, this post is not truly about Israeli divorce rates, but about Israeli fertility rates. I made up the “mimetic theory of Israeli divorce” as a parallel to the mimetic theory of Israeli fertility. The feature of family life this theory addresses is different - divorce as measured by the divorce rate instead of fertility as measured by TFR - but the proposed mimetic mechanism remains the same, emerging naturally from the structure of Israeli society and the interactions among its religiosity groups.

I want to know if the mimetic mechanism is real, and after examining how much interaction there is between religiosity groups in my last post, I’ve now tried to see the outcome the mechanism should produce in family life, focusing on a metric other than TFR. And I can’t see it.

Haredi and Dati divorce rates are as low as you would expect them to be, but not the Secular one, and likely not the Traditional one either. If the mimetic mechanism truly exists, it has no effect on Secular divorces even though it raises fertility. It’s as if divorce outcomes are divorced from family fertility outcomes.

Of course, this isn’t a final refutation of the mimetic theory of Israeli fertility. It could be that some features of family life are affected by it while others aren’t. And it could be that the mimetic effect reaches down to the Dati and Traditional communities, but isn’t strong enough to influence Secular Israelis.

As of now, I’m less confident of the veracity of the mimetic theory of fertility than I was before. And I’ll keep digging into the data.

5

Average divorce rates of Secular cities, local councils and regional councils (2019-2022):

When most people think about the topics Fertility and Israel, they most likely do not think of the fertility of Arab Israeli citizens, but of Jewish Israeli fertility, and so my two last posts discussing Arab Israeli fertility and religiosity might have come as a surprise, or even a disappointment.

However, not only do I think those articles are cool1 and a small contribution to the discussion of the “religious hypothesis”, I also learned a lot about working with CBS data. And now I want to use what I’ve learned to write about the actual thing most people have in mind when Fertility and Israel come up for discussion.

While nearly every developed country has seen its birth rate fall below replacement level in the last few decades, Israel stands out as an exception to the rule, with a Total Fertility Rate (TFR) of around three children per woman. This remarkable fertility rate has prompted numerous theories trying to explain why Israelis, and Jewish Israelis in particular, have so many children.

Part of the puzzle can be easily explained by the very high fertility rates of Israel’s most religious communities. Haredi Jews (like the family in the image above) are the most religious and have the most children - approximately 6.5 children per woman; then come Dati, the Religious/Very Religious (in CBS’s classification), with roughly 4 children per woman; and finally the two least religious groups, Traditional and Secular Israelis, with birth rates below 3.

While the high fertility of the two most religious groups mirrors the high birth rates seen in other highly religious communities, the Traditional and especially Secular groups defy the trend towards lower birth rates that we see in the (mostly secular) rest of the developed world, with Secular Israelis maintaining a TFR of around 2.0 children per woman - close to the replacement level.

This Secular birth rate is exceptional when compared to those of the United States (1.62), the UK (1.54), Germany (1.46), Japan (1.23) or Poland (1.31)2, and therefore it’s no wonder that explanatory theories have focused on the Secular group. What are some of these explanatory theories?

• Repeated existential threats foster a cultural norm of high fertility.

• Pro-natalist government policies. These include everything from subsidized In vitro fertilization (IVF) to affordable childcare.

• Strong family culture.

Those seem to be the most commonly cited ones. And I want to focus on the third, strong family culture, which was very widely discussed online following the publication of this article some time ago3:

NonZionism

Why is Israel fertile?

It’s quite well known by now that, unique among economically developed countries, Israel’s TFR is robustly above replacement levels and shows no signs of dropping. Low and declining fertility across non-s**thole countries is a serious problem, threatening, at a minimum, chronic pension crises, and, in the most extreme cases, the actual extinction of ent…

Read more

2 years ago · 193 likes · 157 comments · משכיל בינה

My goal here is not to discuss NonZionism’s article in detail4. But when I started digging into the CSB data - particularly the kind of data that could support or contradict the cultural explanation - his post immediately came to mind. In NonZionism’s words, this is how strong family culture works in Israel:

What these groups do is provide a one-way channel of influence between Israeli Charedim and the rest of Charedi society, a sort of valve in which Charedi fertility memes can spread outwards, without allowing fertility decreasing memes inwards.

With a lot of mixing and mingling (emphasis mine):

Around 25% of the Religious Zionist [Dati] community is actually barely religious at all and very comfortable freely mixing with ‘traditional’ Israelis who don’t keep the Sabbath, but ‘respect’ it…Finally, these traditional Jews mingle with secular Jews, who pick up the last scraps of the Charedi fertility meme, enough to keep their demographic head above water.

The basic model here is that, unlike anywhere else in the world, there is no hard cultural barrier (at least in one direction) between Charedim and the rest of society, and enough mimetic stepping stones that almost everyone is influenced to some degree. A very simplified model looks like this:

The mixing of religious groups that NonZionism describes is what I want to test with the CBS data. Remember we have data on the share of households in each religiosity group (Haredi, Dati, Traditional, Secular) for every Israeli city with a population above 10,000 and we’ve analyzed these data before. It looks like this:

You can see in the screenshot above that the CBS also reports the share of Mixed households and Other households in each city. Mixed is the one I’m interested in, because, as its name implies, it directly measures the mixing of Jewish religious groups in a very consequential setting: living together as a household.

Before we continue, a caveat. To simplify calculations, throughout this post I assume that mixing occurs between “contiguous” groups (e.g. Traditional and Dati) rather than between non-contiguous groups (e.g. Traditional and Haredi) because the likelihood of the former household setting is far higher than that of the latter.

Digging into the data

Let’s start with the raw average share of Mixed households. The average value across 86 Israeli cities is 2.46%; it drops slightly to 2.33% if we weight cities by their population. Is that high or low? Unless we have another mixed households value to compare it to (maybe from another country?), it’s hard to tell whether Jewish Israeli society has few or many Mixed households.

Well, we don’t have a comparable figure from another country, but we do have the closest thing to it: the percentage of Mixed households in Arab Israeli cities. As I’ve mentioned before, apart from a few exceptions, Jewish and Arab Israelis do not generally live in the same cities. Thus, the Mixed households percentage of an Arab city exclusively reflects the degree of mixing within Arab Israeli society.

And here’s a first (naive) comparison of the Jewish Israeli percentage to the Arab one:

At first sight Jewish Israelis are substantially less prone to mixing than Arabs (3.67% unweighted).

But, are we sure that we are comparing the same thing?

Mixing always involves exactly two groups5, so in the case of Jewish Israelis the four religiosity levels yield three possible mixed household settings. But there is no Arab equivalent to the Ultra-Orthodox/Haredi community in the CBS data. There are only three Arab Israeli religious groups and consequently only two possible mixed household settings6.

This difference in the number of possible mixed settings for Arabs and Jews implies that if both were equally prone to form mixed households, Arabs cities would show roughly 16% more Mixed households than Jewish ones - if this sounds counterintuitive, please read this footnote7.

That still leaves a large gap unexplained. Arab Mixed households are 50% more common than Jewish ones, not 16%. Jewish Israelis still seem significantly less inclined to mix across religiosity boundaries than Arabs.

Plotting the mixing

While trying to figure out if there’s any more information that can be extracted from the data on religiosity, I decided to take a look at the scatter plots of Mixed household share vs the share of each religiosity group. And at first glance, they look completely random8.

The only thing that caught my attention are the six obvious outliers far from the shapeless main cloud of data points. Five of those six outliers are Bedouin cities - Israeli Bedouins are really weird. I then recreated the two scatter plots excluding Bedouin cities, but I’ll save you the effort of looking for any pattern in those plots: there is none.

Where I did see some kind-of sort-of pattern emerge is in the scatter plot of Arab Mixed household share vs Secular household share, after excluding Bedouin cities.

The pattern I highlighted in the plot may not be real (really; just two points in the right half of the plotted area and somehow there’s supposed to be a downward curve there), but it persuaded me to create the same scatter plots for the four Jewish Israeli religiosity levels.

The Mixed vs Traditional and Mixed vs Secular plots seem to reveal an inverted-U curve - though in the case of Mixed vs Traditional only the left half of the curve is visible for lack of data points (cities with more than 50% Traditional households). To understand why this is expected, consider a city where the population of one religiosity group, say Secular, is close to zero; assuming that Secular individuals have a constant propensity to mix (P), independent of group size, the number of Mixed households including a Secular individual will necessarily approach zero9.

At the other end of the scale, if the Secular share of population approaches 100%, the number of individuals in other religiosity groups will be very few, and the number of those prone to mixing will be even fewer. In graphic terms:

So, when the share of religiosity group X (Secular, Traditional, Dati or Haredi) approaches 0% or 100%, the percentage of Mixed households will tend to be lower than when the share of X is farther from the extremes.

But wait. The Dati and Haredi scatter plots don’t look like an inverted-U curve10. And both plots include data points near the 100% mark, so the explanation can’t be that we are missing the right half of the curve, as happened in the Traditional households plot. To be fair, if I squint enough, I can see a curve in the Dati plot, though the right half is made up of just two points. Yet the Haredi plot (lower right) does not look like a curve, no matter how much I squint. In fact, if not for the widely dispersed data points in the extreme left (with near 0% Haredi share), the plot would look like a downward straight line.

Most cities represented by those scattered points on the left have a relatively balanced distribution of Secular, Traditional and Dati households, tough a few are overwhelmingly Secular or overwhelmingly Dati cities. And since I’m very uncertain about the effects of a high (but not near 100%) Secular or Dati share, I decided to exclude from the plot all cities with more than 50% Secular or 50% Dati households.

That looks like a very convincing downward straight line to me11. The more Haredi population in a city, the less mixing there is. And the fewer Haredi households, the more mixing takes place in that city. That sounds like the opposite of what I would expect if other religiosity groups looked up to Haredis and drew inspiration from them.

If low rates of mixed households imply little interaction, I can’t see how Haredis are supposed to influence the broader Israeli population. There’s no reason to think they exert influence through economic status - they are the poorest community among Jewish Israelis. They lack military prestige (big in Israeli society12) that could raise their status - Haredis are the only group which does not serve in the Israeli armed forces. And they dedicate their lives to an endeavor that has close to zero appeal for Secular Israelis13.

Maybe Secular and Traditional Israelis look up to Haredis only regarding matters of family life, even though that seems somewhat contrived to me. Or maybe the memetic theory does not require Haredi involvement. Perhaps only the Dati influence Israeli society at large. The Dati show reduced mixing as their share of households increases, but the effect is weaker than for Haredis, and not enough for me to conclude they mix very little with other groups14.

To summarize, I find no evidence for what I consider a precondition of this theory in the Haredi case: social interaction between them and the rest of society. There seems to be some mixing between Datis and other Israeli communities, but I’m not sure if it’s enough to influence how other Israelis form and grow their families.

Moreover, I don’t think I’ll be able to confirm or reject the theory by further probing the proposed mechanism of mimetic transmission. Instead, I’ll shift my focus to the outcomes of this mechanism, which will be the subject of my next post.

[UPDATE: I’ve published three follow-up posts focused on other measurable aspects of Israeli family life and how they relate to the mimetic theory:

• The memetic theory of Israeli divorce.

• Matchmaker, matchmaker, make me a match: Israeli marriages.

• Another issue with the Mimetic mixing theory of Israeli fertility: Israeli billionaires.

You can think of this post as addressing the mechanics of the theory and the following posts as addressing the expected outcomes of the theory]

14

This is the Dati scatter plot excluding cities with more than 60% Secular or 60% Haredi households, and adding a few more datapoints (quarters):

Looking at the first and third regression models, the ones that include the Religious / Very Religious variable, we first see no effect (p = 0.815) and then what appears to be a highly significant effect (p = 0.001). But this seeming highly significant effect is the wrong sign. If religiosity increases fertility, the coefficient for the religious variable in the third model should be positive, not negative. This would appear to suggest that greater religiosity lowers fertility.

Let’s recall that Israel’s Central Bureau of Statistics (CBS) classifies Arab Israelis into three levels of religiosity: Religious / Very Religious, Traditional and Secular. In my previous analysis I tried to find a connection between the population share of each of these religiosity categories in a city and the fertility rate (TFR) in that city. If the “religious hypothesis“1 is true - greater religiosity is correlated with (and possibly causes) higher fertility - then we should have seen the following as a result of that analysis:

• A higher share of Secular population lowers the city’s fertility.

• Higher Religious / Very Religious share of population should be positively correlated to higher fertility.

• A higher share of Traditional raises fertility only if the Religious share shows a fertility-raising effect at least as large as that of Traditional.

I did observe some evidence for the first proposition (Secular lowers…) and no evidence for the second. The third one I intentionally wrote as a conditional statement because, despite noticing a positive effect of the Traditional share on fertility, I did not see the even greater effect that the “religious hypothesis“ would predict for the Religious share of population.

This is why I admitted I was confused by the results of my last analysis. I was unable to decide whether the data supports or refutes what I call the “religious hypothesis“.

A closer look

But after perusing the data more closely, I realized that cities with higher Secular population share usually have a substantial Arab Christian minority. These Christian communities are concentrated in the Northern region of Israel, especially in Nazareth and its environs. Could it be that the presence of Arab Christians in a city - with their own distinctive culture - has a confounding effect?

We already know from CBS data that Christians have a lower TFR than Arab Muslims2. And this holds even after excluding Bedouins (there are no Arab Christians in Bedouin cities as far as I’m aware). It wouldn’t be surprising that the apparent effect of the Secular share is in fact an effect of the Christian population share in each city.

Once I realized this pattern existed, I did the obvious and excluded the six Arab cities with a Christian share of population above 10%3, leaving a total of 38 overwhelmingly Muslim cities to be analyzed using the same regression models from the last post.

And the results of this new analysis are:

The Secular share shows no effect whatsoever now. The only remaining apparent effects are: the Traditional share seems to increase fertility, but only when the other religiosity level variable in the model is Secular; the Religious share of population appears to decrease fertility, but only when paired with Secular4.

I don’t believe these two effects are real, mainly because they both disappear when the second religiosity variable changes. Yet seeing these results convinced me that throwing away the six Arab cities with sizeable Christian populations was a bad idea. There is certainly confounding between the share of Christian population and the percentage Secular5, but dropping six data points from my already small sample risks obscuring the effect of the Secular share of population on fertility.

So, instead of dropping data points, let’s expand the dataset. Although the CBS only publishes TFR values for the 44 Arab Israeli cities with over 10k population, it does provide a children per woman (CPW) figure for smaller towns, allowing us to increase the sample size if we switch our fertility measure from TFR to CPW6.

The three new models for predicting CPW use a dataset of 52 towns and the same three pairs of religiosity variables as before. And here are their results.

The Secular share of population is now a strong predictor of fertility: the higher the share, the fewer children per woman.

And the Traditional and Religious shares only appear to be good predictors when the Secular variable is omitted from the model, which makes me think their predictive power is only due to their sum (Traditional+Religious) being strongly negatively correlated to Secular.

Also, if you were paying very, very close attention (I’ll admit that black-background screenshots are not the best way to convey information) you might have noticed that the north variable changed signs. It used to be negative, suggesting the Northern region, where Arab Christians live, has lower fertility. But now it’s positive, suggesting that (maybe) Northern Arab towns have slightly higher fertility.

Conclusion

In short, religiosity does appear to affect Arab Israeli fertility but only in a binary way: religious households have more children than non-religious (Secular) ones. Yet the level of religiosity, Traditional or Religious / Very religious, of practicing Arab Israelis doesn’t have any impact on fertility. In contrast to Jewish Israelis, who exhibit a clear increase in fertility for each increase in religiosity, Arab Israelis show no additional fertility increase beyond the religious-vs-Secular divide.

Regarding the “religious hypothesis“, these results cast some doubt on the idea that there’s a simple relation between fertility and religiosity, with the former being a linear function of the latter.

I intend for my next post to address the idea that, in the case of Jewish Israelis, a culture of high fertility rubs off from the most religious group (Ultra-orthodox) to less religious groups, increasing the fertility of each group by the cultural influence of the one above it.

At first glance, a Total Fertility Rate (TFR) of 3.0 doesn’t seem so high - it ranks only 59th worldwide - but it’s unheard of for a developed country. This unsurprisingly elicits curiosity and has spawned many theories trying to explain Israel’s remarkable fertility, some of them connected to the conspicuous religiosity of certain Israeli communities.

The significantly higher fertility of Ultra-Orthodox (Haredi) Israelis compared to the Very Religious Dati, who in turn have a substantially higher fertility than secular Jews, seems to imply a causal role of religiosity, with greater religiosity positively correlated to increased fertility. You might even say there’s a weak “religious hypothesis“ which states that higher religiosity is correlated to higher fertility, and a strong “religious hypothesis” stating that greater religiosity causes higher fertility.

Luckily, Israel’s Central Bureau of Statistics (CBS) keeps track of the religiosity of the population for every locality, allowing us to see the share of households that are secular, traditional, religious or very religious, and ultra-religious in each Israeli town. The meaning of each of these four categories is fairly well-established, along with their applicability to Jewish and Muslim citizens. For instance, the ultra-religious category only applies to Haredi Jews, not to Muslims or Christians.

Leaving aside the few mixed towns, almost all Israeli cities are either overwhelmingly Jewish or overwhelmingly Arab2. This partly explains why Arab Israeli cities concentrate in three clearly separate regions: a northern one in the historical region of Galilee (Haifa and Northern districts), the centrally-located Triangle along the West Bank border, and a southern one inhabited by Israeli Bedouins3.

In total, 44 Arab cities with more than 10,000 population are spread across these three regions4, with TFRs ranging from 1.74 to 5.83. I now want to examine the relationship between religiosity and fertility using the Total Fertility Rate and religious composition statistics that CBS publishes for these 44 cities5.

Recall that the ultra-religious category is only used regarding Jews, not Muslims, thus the three categories which CBS applies to the population of Arab cities are (in decreasing order of religiosity): Religious / Very Religious, Traditional and Secular. Households with members from different religiosity categories are rare, as you can see in the following screenshot.

First and second tries

My first attempt to explain the fertility rates of Arab cities naively assumed that TFR depends on the percentage of Secular, Traditional and (Very)Religious households in each city. The multiple regression model I ran, with the three percentages as independent variables, was disappointing: none of the three variables had p < 0.056, which suggests fertility can’t be explained by religiosity.

This is not the end of the story, though. A very noticeable pattern in the data is that Bedouin cities show substantially higher fertility rates than Northern or Triangle cities (roughly 4.7 children per woman7), which may explain why we don’t see an effect of religiosity on fertility. It might be that the effect of religiosity is smaller than that of (high fertility) Bedouin culture, and we need to explicitly include Bedouin culture as a variable in the model to separate the two effects.

To ensure the model also captures any fertility-influencing cultural differences between Northern and Triangle cities, I defined a second binary (dummy) variable that distinguishes between them (1 = Northern, 0 = Triangle or Bedouin) and re-ran the model with these two new binary variables. The coefficient of determination (R²) increased from 0.211 to 0.9178, indicating the new model is pretty good at explaining fertility across Israeli Arab cities.

The results, unsurprisingly, show that the Bedouin variable has a significantly larger effect on TFR than the Northern variable. They also reveal that religiosity has no predictive power on fertility even after controlling for regional differences in culture (no statistical significance). Maybe I should add another variable to the model?

In comes the income

As you are probably aware, there is a very well-established negative relationship between income and fertility. As Wikipedia says9:

There is generally an inverse correlation between income and the total fertility rate within and between nations.

Maybe the fact that I haven’t yet controlled for income is masking the effect of religiosity on TFR?

The CBS reports two median income values for each city: one for the self-employed and one for employees. Since self-employed median income doesn’t have any effect on fertility while employee median income does, the latter is the one I’ll use from now on.

But before talking about the effect of median income on fertility, I should mention that including three variables (Very Religious, Traditional, Secular) that consistently add up to “nearly” 100%10 might be a bad idea because it can produce multicollinearity, and multicollinearity is bad.

So, instead of showing you the results of an updated model with all three religiosity variables plus median income, I’ll show you the outcomes of the three possible updated models that result from dropping one religiosity variable while keeping two. Here they are, with p-values (“P>|t|“) and coefficients for each of the five variables (two religiosity, two regional, and median income).

You can see that one religiosity variable, percentage of Traditional households, reaches statistical significance in the two models where it is included11, suggesting there might be a real effect there (I’ll discuss the other two variables shortly, but for now let’s focus on Traditional).

The higher the share of Traditional households in a city, the higher the TFR. Going from a 25% share to a 75% share Traditional would raise TFR by roughly 0.8 children. And given that Traditional people are more religious than secular people, this confirms that greater religiosity raises fertility, right? Well, before definitely concluding that religiosity increases fertility, we should take a look at the other two religiosity categories.

Recall that the most religious category is Religious / Very Religious. If greater religiosity increases fertility, as the previous result would seem to imply, then an increase in the share of Very Religious households should have an effect at least as large as that seen for an increase in the Traditional share.

Looking at the first and third regression models, the ones that include the Religious / Very Religious variable, we first see no effect (p = 0.815) and then what appears to be a highly significant effect (p = 0.001). But this seeming highly significant effect is the wrong sign. If religiosity increases fertility, the coefficient for the religious variable in the third model should be positive, not negative. This would appear to suggest that greater religiosity lowers fertility.

Even weirder, the Secular share of households predicts lower fertility, but only in the third model. This seems to contradict what the religious variable suggests in that same model.

Not really a conclusion

I’m not sure what to make of this. What is clear is that I find no robust relationship between religiosity and fertility. Maybe because my models are incomplete? maybe I don’t have enough data? If there was a clear and strong link between higher religiosity and higher fertility across Arab Israeli cities, I would have expected to see it in these models. But I don’t.

What I can conclude is that culture does have a strong effect on fertility. Beyond the well-known negative effect of income on fertility, it’s Bedouin culture that clearly explains the vast majority of the remaining variation in TFR among Arab cities.

Given the results I’ve just presented - and the fact that I’m not an statistics expert - I would greatly welcome feedback regarding the model I used and any obvious blunders I might have made.

I intend to continue exploring this data for a while, and if I do find anything else of interest, I’ll report it in a follow-up post.

(UPDATE: After digging deeper into the data, I arrived at something more concrete:

The Secular share of population is now a strong predictor of fertility: the higher the share, the fewer children per woman.

And the Traditional and Religious shares only appear to be good predictors when the Secular variable is omitted from the model, which makes me think their predictive power is only due to their sum (Traditional+Religious) being strongly negatively correlated to Secular.

Being religious or non-religious does seem to impact Arab Israeli fertility: religious households have more children than secular ones, after controlling for obvious factors. But the level of religiosity among religious Arabs does not have any effect on their fertility.)

The people inhabiting these Caribbean scenes usually display traits associated with one of the two main population groups that settled the region after 1492: Hispanic and African. Oftentimes they show a blend of these features, reflecting their mixed background.

The substantial African ancestry is explained by the deep impact of slavery, with millions of people forcefully brought from the other side of the Atlantic to work on the region’s plantations, while the sizeable Hispanic ancestry is of course explained by Spain’s main role in colonizing the Americas, particularly the Caribbean. And within that Hispanic ancestry, which originates from several distinct regions of Spain, a substantial portion comes from the Atlantic archipelago of the Canary Islands.

I’ve written several posts describing the region of Latin America that I call Canari America, detailing the cultural traits that distinguish it from the rest of Latin America, and arguing that these traits derive from the Canarian migrants that heavily settled this region.

The geographic area of this region largely, but not entirely, coincides with the Caribbean area. Some countries are widely regarded as part of the Caribbean yet clearly not part of Canari America, such as Haiti and Jamaica, while other countries with significant Canarian influence, like Uruguay, are located far to the south of the Caribbean sea.

Given all this, it’s not surprising that a common objection to my characterization of Canari America is that the region I’m really describing is the Caribbean, and the cultural traits I’ve attributed to Canarian influence are actually Caribbean cultural traits. Specifically, if Hispanic and African ancestry influences are so dominant in the Caribbean basin, maybe I’m presenting certain cultural attributes as Canarian when they are really characteristic of Afro-Caribbean culture.

The cultural traits - I’ve been harping on about them in my previous posts - are three:

• Canarians (and Canari-Americans) show low religiosity as measured by the Catholics per priest index (CPP); a measure of how many priests are required by a diocese’s flock.

• Canarians have a particular marriage pattern, with higher cohabitation-to-marriage ratios1 than other Spaniards. This is based on World Values Survey data on marriages.

• Canary Islanders (and Canari-Americans) consistently lean left in their political preferences.

I now intend to investigate whether these “Caribbean“ traits can be presumed to stem from Afro-Caribbean culture, thereby casting doubt on their Canarian origin, or whether any connection to African ancestry can be clearly rejected.

Certain regions of the Hispanic Caribbean offered almost ideal conditions for cultivating cash crops like sugar and cocoa, enabling the emergence of a thriving plantation economy based on enslaved labor. In contrast, areas less suitable to the profitable exploitation of forced labor experienced limited slave importation and relied more heavily on free labor. Eastern Cuba, with its focus on sugar cultivation2, saw the arrival of countless African slaves whose descendants to this day inhabit the region, while Western Cuba, with an economy mainly centered on tobacco and a greater reliance on free labor shows a lower proportion of African ancestry today.

These regional differences in African ancestry offer an excellent way to test whether certain cultural traits are Afro-Caribbean or not. Is Eastern Cuba less religious - a trait I’ve linked to Canarian culture - than the rest of Cuba? If the answer is yes, I’d be slightly more inclined to believe I might be conflating Afro-Caribbean and Canarian culture. If no, that would make me slightly more inclined to believe that lower religiosity in Cuba is truly a Canarian trait.

So, which one is it?

To answer this question, I’ll use the Catholics per priest (CPP) metric I mentioned above. A higher CPP ratio indicates lower religiosity, while a lower CPP indicates higher religiosity. You can see below the 2020 CPP ratio for each Cuban diocese, ordered roughly from west to east.

The two rightmost dioceses (marked by black arrows) correspond to Eastern Cuba’s Santiago de Cuba and Guantánamo provinces, and as you can see, they don’t have particularly low or high ratios, indicating their religiosity doesn’t differ much from the Cuban average3.

This evidence favors the non-conflation hypothesis, but it’s just one piece of data4 among many to weigh before choosing to believe the non-conflating or the Afro-Caribbean hypothesis - or perhaps some kind of synthesis of both. And it relates to only one of the cultural traits I’m interested in. I’ve already mentioned another trait, political culture, which I believe leans left in populations of Canarian ancestry (though some might attribute this political leaning to Afro-Caribbean influence5).

I have to ignore (for the time being!) the political hue of the government that has ruled Cuba for the last six decades, since it reveals little about the political leanings of Afro-Cubans. Yet, while searching for data on Afro-Cubans’ political preferences, I did find two pre-revolutionary elections with provincial-level results, allowing us to compare Oriente Province (roughly equivalent to Eastern Cuba) with the rest of the country. These are the 1944 and 1948 general elections.

The 1944 Cuban general election pitted an alliance of the Authentic Revolutionary Party and Republican Party against a coalition of the Liberal Party, Communist Party6, Democratic Party (right-wing) and ABC Party7. The Authentic-Republican alliance, dominated by the left-wing Authentics, obtained 60.87% of the Senate vote in Oriente Province and 55.39% of the national presidential vote.

In the 1948 Cuban general election, the Liberal Party8 achieved 30.38% of the national vote and 27.29% in Oriente. While on the left, the Communist (PSP) party obtained 7.47% of votes in Oriente, barely above their national 7.25%, and the extreme-left Orthodox Party, led by Eduardo Chibás, got 16.46% nationally and 20.35% in Oriente9.

In contrast to the previous CCP ratios, these figures slightly favor the Afro-Caribbean hypothesis, hinting that the question of which hypothesis is correct is not easy to answer. Also, notice that I’m comparing along three different dimensions, which are not necessarily correlated with each other. It could be that the answer is: 1 out of 3 cultural traits supports hypothesis X, or 2 out of 3 cultural traits support hypothesis Y.

To reach some sort of conclusion - even a partial one - we need to look at plenty more data from other Caribbean countries besides Cuba. That is what I’m going to do next, and it’s going to be a long and dry (and boring for some) examination of CPP ratios, election results and cohabitation rates across three additional Caribbean countries. If you are not particularly interested in reading several pages of data, I recommend you skip to the conclusions.

The Free Associated State

Puerto Rico’s population of African ancestry is not as numerous as Cuba’s, but like Cuba, it is concentrated in a specific region: the northeast of the island. This region, centered on the town of Loíza10, is the one whose metrics (CPP ratios, etc) I’m going to compare with those of the rest the island.

Let’s start with the CPP ratio. The diocese of Fajardo-Humacao is the most African diocese in Puerto Rico, encompassing Loíza and the towns of Fajardo, Humacao, Naguabo, Ceiba, Canóvanas, Luquillo and Río Grande (some of these towns will show up again later). Its CPP ratio of 3,340 is the lowest among Puerto Rico’s six dioceses, suggesting it’s the most religious diocese in the island11.

Now, let’s look at Puerto Rico’s cohabitation-to-marriage ratio according to wave 7 of the World Values Survey (WVS). The ratio for Puerto Rico as a whole is 0.32912, while Eastern Puerto Rico’s ratio is 0.283, second-lowest among the six regions defined by WVS. There’s a caveat though: the WVS does not sample every town in Puerto Rico13; for Eastern Puerto Rico it only sampled Canóvanas (23.4% Black or African American), Juncos (8.9% Black) and Río Grande (18.7% Black). As a result, some sampled towns in Eastern Puerto Rico may have a low share of Black or African American population (Juncos?), while certain towns outside Eastern Puerto Rico could have a high level of African ancestry.

To address that issue, I calculated the correlation coefficient (Pearson’s) between cohabitation-to-marriage ratios and percentage of Black or African American population across all the sampled towns. And that coefficient is -0.15, implying (maybe) a very weak inverse correlation between the two variables: higher African ancestry slightly predicts a lower cohabitation-to-marriage ratio. However, as you can see in the following scatterplot, where it’s hard to visualize any trend, it’s very doubtful there’s any correlation, positive or negative, at all14.

For reference, 10.9% of the total population of Puerto Rico is Black or African American.

Finally, before tackling the third cultural trait of interest to us, political culture, please notice that, unlike U.S. polls, which often break down results by respondents race, Latin American polls do not typically ask for the respondent’s race. It would be rare to find a Latin American political survey that breaks down data by racial identity. To test any correlation between African ancestry and political leaning, the best I can do is look for correlations between high-African-ancestry districts or regions and electoral results.

The earliest meaningful Puerto Rican election results I can find are those from the 1952, 1956, 1960 and 1964 elections. At the time, the current right-wing New Progressive Party (PNP) did not exist, and its predecessor - the Partido Estadista Republicano - was crushed in every general election by Luis Muñoz Marín’s dominant Popular Democratic Party (PPD). In these early results, there are no geographic patterns that I can discern.

Moving on to the 1970s and 80s15, the PPD won the 1972 election by a wide margin, but Fajardo and Loíza were among the sole six municipalities that backed the right-wing PNP16. The 1984 and 1988 elections show no clear preference for the PNP in northeastern Puerto Rico, but the 1976 and 1980 elections do (except for Canóvanas).

The general elections of 1992, 1996 and 2000 show every northeastern Puerto Rico municipality favoring right-wing candidates. The only exceptions were Canóvanas (1992 and 2000) and Luquillo (only 2000), which voted for PPD candidates17. In the 2004 and 2008 gubernatorial and Resident Comissioner elections, all northeastern Puerto Rico municipalities voted for right-wing PNP candidates.

The 2012, 2016, 2020 and 2024 gubernatorial and Resident Commissioner elections show the same recurring pattern, except for Río Grande (2012 and 2020 gubernatorial; 2012 and 2024 Resident Commissioner) and Juncos (2024 Resident Commissioner).

Lastly, regarding recent Puerto Rico House of Representatives elections, the PNP won in the three northeastern districts (36, 37, 38) in 2020 and 202418, despite the overall defeat of the PNP in 2020.

To summarize, the region of Puerto Rico with the most African ancestry appears to consistently lean right.

Colombia

Our next stop is Colombia, a country with a substantial number of Canarian descendants, especially in its Caribbean coast, and also a large population of predominantly African ancestry. The Afro-Colombian population is concentrated near the Caribbean and Pacific coasts, with peaks in the departments of Chocó and San Andrés and Providencia - a Creole-speaking Caribbean archipelago located 775 kms away from the Colombian shore.

Unfortunately, the Protestant persuasion of the Afro-Colombian population of San Andrés and Providencia means that calculating a CPP ratio for the archipelago is useless. But we can instead take a look at the ratios of Chocó’s two dioceses and a few others along the Pacific Coast and compare them against the CPP for the whole of Colombia: 5,605.

This, intriguingly, reveals two distinct patterns. The Chocó dioceses - Quibdó (CPP 6,360) and Istmina-Tadó (CPP 2,202) - have CPP ratios comparable to or below the national average, suggesting Afro-Colombians have higher religiosity than average. But then the Pacific coast dioceses of Buenaventura, Guapi and Tumaco have ratios ranging from 7,300 to 12,300 - substantially higher than average - suggesting lower, not higher, religiosity than Colombia as a whole19 and contradicting the previous pattern.

To make matters worse, among the three dioceses corresponding to Colombia’s three main Caribbean cities - Barranquilla, Cartagena and Santa Marta - the most African one, Cartagena, has the lowest CPP, even though all three significantly exceed the national average20. These figures seem a bit contradictory. Is the religiosity of Afro-Colombians higher or lower than that of other Colombians? I don’t intend to give a final answer yet - before examining more countries - so I’ll leave that question open for now.

Now, regarding cohabitation ratios, four municipalities with a high share of Afro-Colombian population21 took part in the WVS wave 7: Nuquí (Chocó), Tumaco (Nariño, Pacific coast), Cartagena and Cali. All four surpass Colombia’s average of 1.582 cohabitations for each marriage, with ratios of 6.0, 1.75, 3.4, 1.86 respectively22. That suggests higher African ancestry is correlated with higher prevalence of cohabitation, and I’m not going to dispute that23.

Lastly, let’s check whether Afro-Colombians have any political bias compared to other Colombians. I examined the vote share for left-wing 2022 presidential candidate (and eventual winner) Gustavo Petro in each municipality, and then compared that share to the proportion of self-identified Afro-Colombians to detect any effect on the first round left-wing vote. I only included in this analysis the 11 Colombian departments with the highest African ancestry: seven from the Caribbean coast and four (including Chocó) from the Pacific coast24.

What are the results?

The correlation coefficient (Pearson’s r) between Afro-Colombian population percentage and Petro’s vote share is positive (0.27); higher African ancestry is linked to higher support for Petro, but the connection is very weak (R² is 0.071). And the results are very much the same if I restrict the analysis to just the three Pacific coast departments (coefficient = 0.28, R² = 0.077), or the Pacific coast plus Chocó (coefficient = 0.29, R² = 0.071)25.

I then restricted the analysis to the seven Caribbean coast departments. They show no correlation (-0.023) and a value of R² close to zero. The conclusion I must draw from this is that the vast majority of Afro-Colombians (see footnote 24) exhibit no proclivity to vote left or right compared to other Colombians.

What about Panama?

The Isthmian country’s population of African descent has an interesting origin story. Alongside its Afro-Panamanian community of colonial origin, similar to that of Colombia, Panama is home to an Afro-Antillean population descended from West Indian workers who were recruited to build the Panama Canal. Many of these workers settled in the Canal Zone, the area of Panama that was under American sovereignty from 1903 to 1979, or close to the American-administered canal, including the city of Colón and its surrounding region, which now constitute the Province of Colón.

This is the main reason why Colón Province has the highest share of Afro-Panamanians among all of the country’s provinces26. Approximately 61% of Colón’s population is Afro-Panamanian, compared to a share ranging from 17% to 37% in every other province27. If the cultural traits I’ve been analyzing are Africa-Caribbean in origin, they should be more prominent in Colón than everywhere else in Panama.

Let’s start with the CPP rate by looking at the rates of all Panamanian dioceses. The diocese of Colón also includes the territory of the Kuna Yala Indigenous Comarca, but since the comarca has barely a ninth of Colón Province’s population, we can just treat the diocese’s CPP as a good measure of the province’s religiosity and disregard the effect of Kuna Yala’s population on the rate.

These are the CPP rates of every Panamanian diocese:

Colón Province, with a rate of 8,166 Catholics Per Priest, is more religious than Panama as a whole (CPP 8,488).

Next, let’s look at marriage patterns in Panama. Since the country did not take part in wave 7 of the World Values Survey, I’ll use out-of-wedlock births as a proxy for cohabitation28, excluding children born to mothers in civil unions. This is a cultural aspect in which Colón Province stands out from all other provinces.

Colón’s 62% of out-of-wedlock births is by far the highest among all provinces.

Finally, regarding the political leanings of Afro-Panamanians, I couldn’t find provincial election results prior to the year 2009, so that year’s general election is the earliest I could analyze. And there wasn’t much to analyze. The right-wing candidate won 60% of the vote nationally and a majority in every province but four29, so his victory in Colón doesn’t reveal much.

The next presidential election, in 2014, saw a victory for the left (PRD party) in Colón Province, while the following 2019 election brought a defeat. And as for the legislative results, I found the 2014 vote percentages for left-wing parties but only found the number of seats won for 2019 (two out of four). Here are the 2014 results.

These results are similar to the ones for the 2024 legislative election, where again Colón Province voted slightly more left than the country as a whole30, but not remarkably so.

In fact, in the 2024 presidential election, Colón Province voted slightly less left than the national average, with 5.81% of votes cast for left-wing candidates31 compared to 6.64% nationwide. Overall, these results show that Colón, the most Afro-Panamanian province in the country, does not lean more left than Panama as a whole, suggesting that Afro-Panamanians are not particularly left-leaning.

To summarize, Afro-Panamanians are more religious than the national average, have significantly higher rates of out-of-wedlock births, and don’t seem particularly left-leaning or right-leaning within Panama’s political spectrum.

Conclusions

I wish I had more data on which to base my conclusions32, but for now this is all I have. So, to summarize, of the three cultural traits that I’ve characterized as Canarian, two of them show strong evidence against an Afro-Caribbean origin, making what I called the Afro-Caribbean hypothesis highly unlikely. These are low religiosity and a political bent to the left, both of which now look even more likely to be Canarian in origin.

The third one, higher cohabitation-to-marriage ratios, does appear to be connected to Afro-Caribbean culture, conflating the supposed Canarian origin of higher cohabitation in Canari-American countries with a possible Afro-Caribbean one. I have now updated the probability I assign to an Afro-Caribbean origin of this trait from low to not-so-low33.

To help you better evaluate the data, and decide whether you agree with my conclusions or not, here’s a summary table of all the evidence I’ve examined in this post, arranged by country and cultural trait.

When trying to explain what makes some Latin American countries different from others, what characterizes them, and analyze the cultural attributes that shape their societies, I don’t believe that just a simple “they have a lot of Canarian ancestry, so obviously those cultural traits come from the Canary Islands” is enough. Questioning what might seem logical at first sight is a healthy exercise - and I’m glad that I now have to revise my claim that high cohabitation ratios in Latin America are clearly Canarian in origin.

I’m also glad that I’ve addressed what I see as the strongest objection to a Canarian origin for the two other cultural traits - successfully I hope; if you are still not convinced, please let me know in the comments. Now I look forward to moving on to (hopefully) less dry, less data-intensive topics related to Canari-America.

It also reveals the Caribbean region as particularly irreligious within Latin America. For Latin America as a whole there are 5,159 Catholics per priest, whereas the four Hispanic Caribbean nations average a much higher CPP ratio of 8,324.

The previous chart, contrasting the four Hispanic Caribbean countries that form the core of Canari America with other similarly Hispanic countries, makes the lower religiosity of this region of Latin America conspicuously evident.

The chart I was referring to, showing the number of Catholics per priest in the year 1950 across 13 Latin American countries is this one:

And if you contrast that chart, where Uruguay appears to have the highest religiosity in Latin America, with my latest post, The Canari American marriage pattern, where I wrote: “I feel compelled to highlight aspects of Uruguayan culture that may show Canarian influence… Uruguay’s cohabitation ratio is almost triple that of most neighboring subnational entities“, it may seem like there’s a contradiction between the two posts, or at least there’s a valid yet overlooked piece of evidence challenging the notion that Uruguay is very Canarian.

To refresh your memory, the Catholics Per Priest ratio (CPP) is calculated by dividing the Catholic population of a diocese by its number of priests, and my interpretation of this ratio is that the higher a country’s CPP is, the lower its religiosity. And in the same November post, I showed that the Canary Islands had the highest CPP ratio among Spanish regions, and presumably the lowest religiosity, in 1950.

That’s why it seems contradictory that I would attribute the high CPP ratio and low religiosity of countries like Cuba and Venezuela to Canarian influence, while Uruguay, a nation which also received a large number of Canarian immigrants, has the lowest CPP ratio among all 13 countries.

A plausible explanation might be that although Uruguay welcomed a large number of Canarian settlers, the influx was not as substantial as that received by Cuba and Venezuela. Based on my admittedly rough estimation of Canarian descendants in the new world, around 40% to 45% of European ancestry in Cuba and Venezuela stems from the Canary Islands, while only 30% of European ancestry in Uruguay originates in the archipelago1. So, if Uruguay is less Canarian, it should be less culturally Canarian.

Yet, it’s somewhat inconvenient for my thesis that Uruguay not only appears more religious than Cuba and Venezuela, as shown by its low CPP value, but seems to be the most religious country in all of Latin America.

The exception to the rule?

I must admit my main focus of interest when I started researching the religiosity of Latin American countries was Cuba. The reason I chose the year 1950 for my initial research was simply to avoid the effects of the 1959 Cuban Revolution on the CPP ratio and on Cuban religiosity itself2.

It should come as no surprise that the number of priests in modern-day Cuba is rather small, and the country’s CPP is very high, after more than six decades of official atheism and religious repression3. It would be absurd to ascribe Cuban low religiosity to Canarian cultural influence. If we want to measure religiosity in Latin America in the 21st century, Cuba’s CPP is not going to be very informative, so we might as well ignore it. But setting aside Cuba’s data, looking at more recent Latin American CPP values can still be informative, and that is exactly what I did.

But before we take a look at recent CPP data, specifically from the year 2020, there are two issues I want to address.

First, I’m not using the exact 2020 CPP values calculated by catholic-hierarchy.org. Since my goal is to gauge the religiosity of the general population, if only 50% of the population of a traditionally Catholic country identifies as Catholic in 2020, I want to preserve the information this percentage provides. To give an example, if a population of 4,000 includes only 50% who declare themselves Catholics and two priests serve that population, catholic-hierarchy.org reports a CPP of 1,000 (2,000 divided by 2), while I will calculate a CPP of 2,000, implying substantially less religiosity4. This wasn’t an issue when dealing with 1950 CPP data, as nearly all Latin American countries were close to 100% Catholic at the time.

Second, regarding the 13 “other similarly Hispanic” countries I mentioned, the justification for restricting the previous analysis to them was:

To assess the specific impact of Canarian heritage on Latin American religious culture, as opposed to other regional Spanish heritages (like Andalusian, Basque, Castilian, etc.) in the former Spanish colonies, I will exclude from this analysis the aforementioned countries where overall Spanish ancestry - from any region of the mother country - is not predominant.

In 2020, just like in 1950, the excluded countries5, characterized by higher Indigenous American ancestry, exhibit higher CPP values. This aligns with the observation that European Catholic nations remain more religious than Latin American countries, which are only partially European, in 2020.

I’m keeping the same criteria, thus including the same 13 countries in this analysis (plus Brazil).

And now, without further ado, let’s see if we can find anything interesting in the 2020 Catholics Per Priest data for Latin American, whether related to Uruguay or otherwise.

Well, from 1950 to 2020 Uruguay has moved from exceptionally high to fairly average religiosity. Other changes in the rank of religiousness surprised me (see Chile6 and Paraguay7), but I can’t say that Uruguay’s shift did.

This is still a bit disappointing from my point of view, because Uruguay doesn’t have a high CPP ratio even when compared to its neighbors Argentina and Brazil. But I’m a stubborn man, so I decided to keep digging and delve into the regional 2020 data for Argentina and Brazil.

The bar chart I’m now going to show you displays the CPP values for seven Argentine provinces and Brazilian states bordering or near8 Uruguay. These subnational entities are mostly comparable to Uruguay in size and population, and just like Uruguay, they received a substantial number of European migrants during the late 19th and early 20th century. You could say they are nearly as European as Uruguay, but without a significant Canarian influence.

Turns out that compared to similarly European regions of Argentina and Brazil, Uruguay actually looks less religious. It’s Catholics per priest ratio is unusually high.

So, there you have it. The 2020 CPP data doesn’t conclusively prove that Uruguay is a low religiosity country, but it does imply an unexpectedly lower religiosity than what 1950 data suggested. Uruguay may not be the most culturally Canarian country in Latin America, but it’s far from being the least.

To get a sense of which Latin American countries and regions are encompassed in Canari America, here’s a map showing the main migration routes from the archipelago to Latin America taken from my old August 2024 post:

As you can see on the map, most areas colonized by Canarian immigrants are located in the Caribbean. Some of these settlement areas are easily recognizable as modern Latin American countries, like Cuba and Uruguay, while others correspond to certain areas within countries, such as Colombia’s Caribbean coast or Venezuela’s Caribbean region (home to the majority of the country’s population).

The cases where certain regions within a country were settled by Canarians while others were not are particularly interesting, allowing us to compare Canarian vs non-Canarian political culture, religious culture, etc, within the same country, free1 from confounding factors that might affect cross-country comparisons.

This is what I did in August 2024, comparing the voting patterns of Colombia’s Canarian-settled region, as a proxy for its political culture, to those of Colombia’s (non-Canarian) Andean region. That analysis, and a similar comparison between the Canarian and non-Canarian (Andean) regions of Venezuela, led me to conclude that areas colonized by Canarian Islanders exhibit distinct voting patterns compared to regions colonized by other Spaniards2.

So, does my hypothesis that Venezuelan regional voting patterns, and other cultural differences, derive from the cultural differences between Canarian settlers (a major component of the colonization of the Caribbean coast) and those from other Spanish regions hold up when applied to the case of Colombia?

…Colombia mirrors Venezuela in that its Caribbean and Andean regions exhibit distinct voting patterns….Historically, in most Colombian elections, the Caribbean region has shown a stronger preference for the left-wing candidate (typically the Liberal party candidate) than the Andean region, mirroring the electoral trends for Venezuela.

I also compared religious culture across countries and found a consistent pattern of lower religiosity in Latin American regions with more Canarian influence3, mirroring the lower religiosity of the Canary Islands compared to the rest of Spain.

So, dividing Colombia into a Caribbean and a non-Caribbean region seems to corroborate the relationship between Canarian influence and lower religiosity. While applying the same division to Venezuela, although it doesn’t contradict the premise, does not provide such a strong support for the relationship.

Regardless of whether you agree with my view that the CPP ratio is a reliable indicator of religiosity, the fact stands that it captures a core feature distinguishing the various Catholic cultures within Latin America. A feature that is consistent over time, and correlates very strongly with Canarian influence and culture.

These are only two of the several (many?) cultural facets in which Canari America differs from the rest of Latin America, usually mirroring cultural differences between the Canari Islands and mainland Spain. I’ve been intending to write about other cultural aspects where Canarians significantly differ from other Spaniards for quite some time - but other topics got in the way4. Then, as luck would have it, I stumbled upon an X post mentioning the exceptionally high rate of cohabiting unions in Latin America.

After reading Yeyo’s tweet I thought to myself, “maybe I should take a moment to research this and see if I find anything interesting“. Next thing I knew, I was diving into the World Values Survey data - the source for the cohabitation rates in the table above - trying to replicate the figures for each Latin American country.

The World Values Survey (WVS) conducts its surveys in waves, each spanning several years during which samples of residents from many countries will be polled using the same questionnaire. The last wave - number seven, referenced in the tweet above - took place from 2017 to 2022 and is the main data for this post.

I had never examined data from the World Values Survey (WVS) before5, but I soon realized it’s very easy to work with, and since I’m only interested in one of the many questions asked by the survey, Are you currently living together as married?, made the process even simpler6.

The table above doesn’t show the rate of cohabitation for each and every Latin American country included in the WVS data7, so my first task was to calculate the rates for all available countries. Here it is:

I’m not going to show you how these results look on a map just yet. First, I want to address the issue of how different population structures affect the percentages of cohabiting survey respondents. Consider Guatemala, which has the second youngest population in Latin America and also the lowest cohabitation rate (8.62%). A younger country may have more single individuals8, making the rate of all non-single categories (married, cohabiting, etc.) seem lower than that of an older country, even though its people are no less inclined to marry or cohabitate.

I can avoid this distorting effect by calculating a ratio of cohabiting respondents to married respondents, thus quantifying the propensity to cohabitate only among those who are in a stable relationship. And after calculating this ratio, Guatemala’s aversion to cohabitate - relative to the rest of Latin America - appears less exceptional, with a ratio of 0.282 compared to Argentina’s 0.287.

And now that we have a more robust index, more closely linked to couples’ propensity to cohabitate, let’s take a look at a colored map of Latin America, showing cohabitation ratios for all twelve countries with available data.

Colombia and Venezuela have the highest ratios by far (1.58 and 2.06), with many more individuals cohabiting than married, but Peru and Nicaragua also have fairly high cohabitation ratios. Also notice that Uruguay does not look very reddish on the map, but its ratio is rather high (0.889, fifth highest) compared to its neighbors.

My hypothesis is, of course, that these results follow a pattern of higher cohabitation ratios in countries with greater Canarian influence, with Peru and Nicaragua as notable exceptions to this rule. I know the data I’ve presented so far is too sparse to confirm this hypothesis, but for now please note that’s what I’m aiming for.

Uruguay is a bit of an anomaly among Canari American countries as the only nation outside the Caribbean basin which received a large flow of Canarian immigrants, challenging the apparent climatic connection between the semi-tropical Canarian archipelago on one side of the Atlantic and the tropical sea on the other side of the ocean. That is why I feel compelled to highlight aspects of Uruguayan culture that may show Canarian influence, or at least appear more Canarian than those of the average Latin American country. If these deviations from the Latin American norm were confined to Canarian regions around the Caribbean, one might be tempted to ascribe them to “Caribbean culture” as the most straightforward explanation for this pattern9. Uruguay provides strong evidence that this explanation is wrong.

Perhaps “Strong” sounds like a strong word when based solely on the colored map above. Uruguay appears only a shade redder than Brazil, so at first glance it doesn’t look very exceptional among Southern Cone countries. But this first impression results from both the color scale I (naively) chose and the substantial variation in cohabitation ratios among Brazilian states.

Let me try another colored map, but this time showing subnational entities and using a color scale representing values from 0.20 to 0.90.

Now that’s very red. As you can see in the left side legend of the map, Uruguay’s cohabitation ratio is almost triple that of most neighboring subnational entities10, exposing its distinctiveness within the River Plate basin of South America.

Zooming in on Colombia and Venezuela

Given the insights we gained by zooming in on Uruguay and its neighbors, revealing the distinct Uruguayan marriage pattern, perhaps we should now shift our focus to Colombia and Venezuela, the two countries with the highest cohabitation ratios according to WVS data.

The idea is to compare Andean against Caribbean regions of both Colombia and Venezuela, just like I previously did when analyzing political culture and religious culture in Latin America. The WVS collects data from most Colombian departments and Venezuelan states, though in some cases the samples are too small (10 or 20 respondents) to produce reliable results. To avoid this pitfall, I grouped together Colombian departments by region, Andean and Caribbean, and calculated a cohabitation to marriage ratio for each. Then I repeated this process with Venezuela’s states to obtain a ratio for its Andean and Caribbean regions.

And here are the results, displayed in a colored map as usual.

The deep reddish tone of Venezuela’s Caribbean region clearly indicates its high cohabitation ratio, yet its striking prominence - compared to the milder purple of the Colombian Caribbean - suggests there might be administrative or legal factors, related to Venezuela’s bureaucracy or legal system, which cause the high ratio.

These hypothetical non-cultural factors might be raising Venezuela’s ratio or conversely lowering Colombia’s, but I’m skeptical of such non-cultural explanations because:

• Venezuela looks very red on the map, but not entirely so. The country’s Andean region, home to roughly 10% of the population, has a substantially lower cohabitation to marriage ratio (1.24), despite sharing the same legal system as the rest of the country.

• Colombia appears mostly blue on the map, but that is largely because I chose to focus on its Caribbean and Andean regions while leaving the rest of the map blank. However, Colombia’s eastern half seems to have a very high cohabitation ratio11, perhaps higher than Venezuela’s, despite sharing the same legal system as the rest of the country.

In brief, the most plausible explanation I can think of for the marriage patterns in Colombia and Venezuela, characterized by significantly higher cohabitation in Caribbean regions than in Andean regions, appears to be cultural. And one plausible origin for this cultural disparity is the high level of Canarian settlement in the Caribbean regions of both countries.

But why would Canarian settlement lead to this particular marriage pattern? Correlation does not imply causation. The high cohabitation ratios in regions with greater Canarian settlement doesn’t mean the latter causes the former.

To answer that question we must look at the source. The source of Canarian migration that is.

Use the source

If a deep cultural divide exists between Canarians and mainland Spaniards, so deep in time that those (mostly) descended from Canarians who migrated to the Americas centuries ago still show the effects of this cultural divide when compared to descendants of other Spaniards, then we should expect to see its effects in modern Spain. The Canary Islands should exhibit a higher cohabitation ratio than mainland Spain if they are the origin of the marriage pattern we’ve uncovered in Canari America.

So let’s look at the WVS data, the same I’ve been using to analyze Latin American countries, and see if the Canary Islands appear culturally Canarian within Spain. There’s just one problem: Spain did not take part in wave 7 of the WVS. That leaves me no choice but to look at the previous survey, wave 6, conducted in 2011.

Even though the question code in wave 6 (V57) differs from that in wave 7, the question itself and possible answers remain identical, which gives me confidence in the comparability of results from both surveys.

So here are the cohabitation ratios - no map this time - for each Spanish region (Autonomous Community):

As you can see in the chart, the Canary Islands (indicated by the arrow) had a higher than average cohabitation ratio (0.26) in 2011, though not exceptionally so. Both the Balearic Islands and the Basque Country have similar or higher ratios12. This is not strong evidence for this whole Canarian culture hypothesis.

Let’s try another thing. I know I said that Spain did not take part in wave 7 of the WVS, but it did participate in a similar survey called the European Values Survey (how original), which took place in 2017, around the same time wave 7 of the WVS began.

We can examine the 2017 European Values Survey (EVS) data, but there’s a catch. The equivalent to our familiar question Q273 in the 2017 EVS survey is question X007: “What is your current legal marital status?” - notice the word “legal”.

Then there’s the issue of answer number 2, “living together as married“ (no reference to legal status) in the WVS, which becomes “registered partnership“ in the EVS, implying a particular form of cohabitation under a certain legal framework. This is not the same thing as ordinary cohabitation. Nevertheless, it might be instructive to see how the Canary Islands compare to other Spanish regions using the somewhat different cohabitation ratio defined under the legally recognized form of a registered partnership.

So here they are; “legal cohabitation” ratios for each Spanish region:

This doesn’t look promising. The Canaries show a higher than average ratio, but so do three other regions.

Let’s step back from the cohabitation to marriage ratio for a moment. Maybe, within the Spanish context, we are not measuring the same thing as in the Latin American context. If that is the case, using an alternative index or variable that is closely related to people’s predisposition towards cohabitation, or aversion to legal marriage, could be a valid way to measure that predisposition (or aversion) without relying on the World Values Survey.

Given that a common consequence of couples choosing to live together, whether married or not, is producing children, it should come as no surprise that Latin America stands out not only for its high rates of cohabitation but also for its high percentage of out-of-wedlock births compared to the rest of the world.

Although the strength of the correlation between a predisposition to cohabitation and out-of-wedlock birth rates likely varies across countries, we can generally expect some degree of correlation.

I should also mention, from a reliability perspective, out-of-wedlock birth rates have one clear advantage over the cohabitation ratios I’ve presented thus far. Unlike the ratios, they are not estimated from a survey, but computed from official birth records of the state we are interested in, which in this case is Spain. So let’s look at Spain’s 2023 birth data, provided by the Spanish statistical institute INE, with a focus on out-of-wedlock births by region (Autonomous Community).

To analyze births outside of marriage, I’ll introduce a new ratio: an out-of-wedlock birth ratio, defined as the number of births to unmarried mothers divided by the number of births to married mothers. While not equivalent to the cohabitation ratio, the out-of-wedlock birth ratio has some properties that make it intuitive if you are already familiar with the previous metric. Its magnitudes are similar to those of the cohabitation ratio, with a value of 1 representing parity between the numerator (out-of-wedlock births) and denominator (births to married mothers); and higher values also indicate, implicitly, a lower propensity to marriage.

And here is the bar chart with the out-of-wedlock birth ratios for each Spanish region.

I omitted the arrow pointing to the Canary Islands bar because it’s unnecessary. The Canary Islands bar is the one towering over every other bar. The second-highest region, Extremadura, has a ratio of 1.25, which is 35% lower than the Canaries’.

This provides much stronger evidence that the disparity in marriage patterns between Canari American countries and the rest of Latin America has a mirror image in Spain, between the Canary Islands and the rest of the country. I strongly believe this disparity stems from a lower propensity to marriage among Canarians than among other Spaniards.

So there you have it: political culture, religious culture and marriage patterns; all these cultural facets make Canari America stand apart from the rest of Latin America.

11

The ratio for Caquetá, Meta and Putumayo departments of Colombia is 3.43 cohabiting individuals per each married one.

Therefore, I was happy to see Alan Perlo - who writes a very interesting Substack of his own - raise a good point regarding my American soft power revisited post, where I stated that American film directors are less appealing to international audiences than non-American directors:

As far as I know, the top-grossing films that do historically do well internationally are action and franchise movies where dialogue, social commentary and well-crafted humor are less important, because executing these things well depends a lot on the background and tastes of a specific national audience and can be literally "lost in translation". To most moviegoers, what sways them in watching a film at the cinema is what piece of IP it is based on, not whichever director was hired to direct that specific franchise installment. With that being said, the graph you have does show a trend of international directors outperforming American ones. Perhaps making it to Hollywood for a foreign director requires more initiative and skill than reaching it as an American director.

This is a sensible objection to my argument that the higher box office of movies directed by non-Americans compared to those directed by Americans might be due to cultural differences.

If foreign directors face greater challenges when trying to break into the largest and most dynamic cinema industry in the world, then only a small and very skilled fraction of non-American directors, a highly selected group, will make it.

My first intuition when pondering this argument was that foreign directors make more money for Hollywood studios in international markets but less money in the domestic (U.S. and Canada) market, therefore they are not necessarily more skilled than their American colleagues.

Then I realized that foreign directors’ higher international revenue share (a relative measure) doesn’t necessarily result in lower domestic revenues than American directors in absolute terms. Maybe foreign directors are more skilled, and Hollywood hires them because they can generate the same (or more) revenue domestically while at the same time increasing Hollywood’s international box office compared to what an American director would bring in.

As Alan puts it:

It would be interesting to look at how American directors do compared to foreign directors in the domestic market box office. I have a feeling that say in the 90s, when comedy and mid-budget movies had a larger share of the box office, American directors might have brought in more money than they do today because they specialized in movies that only did good business domestically.

This made me realize I needed to compare the domestic box office of American and foreign directors to evaluate whether differences in skill could explain the gap in international share of revenue between one group of film directors and the other.

Let’s remind the gap. Here’s the chart showing the percentage of total revenue that comes from international markets for both groups.

As we’ve seen before, non-American directors consistently derive a higher percentage of revenue from international markets, though this pertains to the share of the total (domestic and international) amount.

Now, let me show you a similar chart, this time plotting the average domestic revenue of films directed by American and foreign directors.

Foreign directors consistently bring in less revenue from the domestic market (U.S. and Canada) than American directors. Furthermore, the gap between American and non-American directors appears to have widened in the last few years, post-2016.

This doesn’t refute the hypothesis that foreign directors could be highly selected. Even if they make less money in the domestic market, their higher international revenue share means they could make up for it in the international market, generating more revenue than American directors overall.

Perhaps Hollywood selects foreign directors who will maximize revenue regardless of which market that revenue comes from?

We should be able to determine whether that is true by looking at the total average movie revenue for American and foreign directors.

Well, there’s no discernible pattern in the above chart. Averaging across the whole time period (21 years) reveals that films directed by Americans earn slightly more than those directed by non-Americans, 261MM compared to 248MM; thus, if anything, the total (domestic and foreign) average revenue figures suggest American directors are more profitable for Hollywood. And yet, around 30% of films are directed by non-Americans.

Why do Hollywood studios frequently choose non-American directors over American ones? Because there’s a lot of variance in the amount of revenue generated within each group, and studios probably base their selections on the individual merits of each director, disregarding any potential skew toward international revenue that non-American directors might cause.

That seems to me the most plausible explanation.

This result makes me think that the selection process for foreign directors is no more rigorous - beyond the obvious, such as English proficiency - than for American ones.

Which is what I expected. I would be very surprised to find a result implying Hollywood studios are systematically making bad decisions and losing money as a consequence. Studios selecting and employing directors with the goal of maximizing profits aligns with my priors.

For each year, from 2007 to 2024, I calculated the ratio of international box office revenues - excluding Canada and the United States - for the top 200 highest grossing films of that year (numerator) divided by the domestic box office revenues for the same films (denominator).

I observed that the trend of a decreasing international box office share, beginning around 2019, also holds for franchises like the Spider-Man and Transformers series of movies. Since franchises, characterized by their common subjects and shared cast of characters, form a series of closely related entertainment offerings, the declining demand for them in international markets suggests a change in the tastes of international movie audiences.

Hollywood keeps offering a similar product, yet over time, non-American moviegoers become less and less interested, likely reflecting cultural changes in international markets.

I also noted that movies directed by American directors are less appealing to non-American audiences, implying some kind of (higher?) cultural mismatch between American directors and foreign audiences.

You can see in the chart that the average international revenue share of non-American directors is consistently higher than the international share of American directors (aside from 2007).

I now want to add another important piece to the puzzle of the disappearing international movigoer1. But first I need to address a small issue with the general trend of rising international share before 2014 and declining share post-2019.

Hollywood goes to China… and then leaves

In response to my previous post, several people pointed out the effect of the Chinese market on Hollywood’s box office performance. Perhaps the rising trend in international revenue share before 2014 was caused by China’s rapid transition from poverty to middle-income status and its consequent exploding demand for Hollywood’s films?

China’s box office did explode, going from $16 million in 2003 to $1.9 billion in 2014, so the pre-2014 rising trend can be attributed, at least in part, to the huge expansion of the Chinese market. At its peak, China represented nearly 10% of Hollywood’s total box office.

And then, following a deterioration of relations between the United States and China in 2018-2019, China’s appetite for Hollywood movies collapsed.

The pre-2014 rise and post-2018 collapse align with the trend I described in my previous posts2, making it reasonable to assume that the Chinese film market might fully explain the observed trend.

Yet it doesn’t. The trend is still there after excluding China from the total international box office and the shape of the curve is roughly the same:

The updated curve now shows two peaks, in 2011 and 2017, rather than just in 2017 when including China, suggesting that my previous assumption of a 2014 start date for the inflection period - when the curve transitioned from a rising trend to a declining trend - may be wrong. The inflection period could have begun earlier than 2014.

China’s case is particularly interesting because it offers a clear candidate for the cause of the declining box office and it suggests where to look for potential causes of other box office declines. It would be interesting to see if any other country shows a revenue drop that coincides in time with a deterioration in its relations with the United States.

Could I do a similar (admittedly superficial) analysis for other major international markets, such as Europe, Latin America or Asia? Well, not for the time being, but I hope to do that analysis in the future3.

Let’s now move on to the actual subject of this post: what else can I say about Hollywood’s decline in international markets besides the fact that American directors are less appealing to international audiences than non-American ones.

Categorizing directors as American or foreign is just one of many ways in which we can partition the set of all Hollywood filmmakers. Because the data I’ve used spans from 2003 to 2025, directors from the early years of this time period comprise a distinct subset from the directors of the 2020s. Some are too young to have been active in 2003, while others, born in the 1940s or before, are too old to remain active in the later years.

This generational churn creates an opportunity to examine the international box office data in a new way. Old Hollywood directors may not have been active during the same period as those from the younger Hollywood age cohort, but I can define an additional middle-aged cohort consisting of directors who worked alongside the older generation. And maybe some of these middle-aged ones were still active when the younger generation became dominant in Hollywood.

In fact, I can define four or five different age cohorts and, for each, compare various characteristics of the films they made against those of other cohorts - and yes, the one characteristic I have in mind is the international share of revenue.

I can’t define too many age cohorts because then each one would have so few directors as to make any comparison meaningless. Therefore, I made sure each cohort directed at least 10 movies each year so as to make comparison with another cohort meaningful. This approach allowed me to calculate a yearly average international share of revenue for at least two cohorts (almost) every year.

I also excluded non-American directors from this analysis to isolate the effect they have on international revenue, and tried to have at least three distinct American cohorts to work with.

But ended up defining four.

(I did not use the conventional American generational labels of Boomers, Millennials, Generation Ç, or whatever Americans call them. They are arbitrary and they mostly exist so that lazy journalists and editors can save some time when writing catchy - usually inaccurate - titles for their next piece. Believe me. I can be a lazy writer. I know what I’m talking about)

The four cohorts are:

• Directors born between 1948 and 1957. The oldest cohort.

• Directors born between 1958 and 1967.

• Directors born between 1968 and 1977.

• Directors born after 1977. You could call them the youngest cohort.

(I first tried to use 1980 - a nice round year - as a boundary year, but there are too few directors born after 1980)

What does the average international share of Hollywood revenue for each of these four age cohorts look like?

You’ve certainly seen prettier charts.

Except for the years 2018 and 2019, there are enough movies (10 or more) to calculate the average for only one or two cohorts each year. We can compare the oldest cohort against the 1958-1967 cohort, 1958-1967 vs 1968-1977, or 1968-1977 vs the youngest.

I’d rather not comment on the first of those three comparisons for now, given that the oldest cohort has only two data points, but I will comment briefly on it in the conclusions.

The 1958-1967 (red) age cohort is largely characterized by a higher international box office share than the younger 1968-1977 (yellow) cohort. Only in two years (out of twelve) did films made by the 1968-1977 cohort achieve a higher international share. Thus, although the limited available data (movies per year, directors) is not enough to assert with certainty that the pattern is real, it does suggest American directors from the 1968-1977 cohort, currently aged between 47 and 57, are less appealing to non-American audiences than their older counterparts.

Finally, the comparison between the 1968-1977 (yellow) age cohort and the youngest cohort shows a similar pattern. Movies directed by the older cohort (1968-1977) show a higher international box office share than those by directors from the youngest cohort. From 2018 to 2025 (six years), only the year 2024 deviates from this pattern.

Being shifty

Before drawing any conclusions from the preceding observations, I wish to underscore the limitations or weaknesses that these comparisons between cohorts suffer from.

Even though Box Office Mojo, my source for movie box office figures, provides data for 200 films each year, the usable subset of Hollywood movies is much smaller than 200. And only some of those movies are directed by Americans or Canadians4.

Also, some directors’ Wikipedia pages do not record a year of birth or birthplace, or they list this information in a non-standard way that makes parsing difficult (and I’m too lazy to put in the effort to deal with those cases).

In brief, the dataset I’m using is smaller than ideal.

Another possible weakness I should discuss is, I think, pretty obvious: did I cherry pick the four age cohorts defined above to fit my narrative?

The short answer is: no, trust me.

The long answer is that I made sure the results are not sensitive to the boundary years I chose for the age cohorts I defined.

I carried out two additional calculations of the international share of revenue, spanning the same time period, using two alternative definitions for the four age cohorts. The first shifts all the cohorts to the left, starting or ending one year earlier; for example, the 1968-1977 cohort becomes the cohort 1967-1976. The second shifts all the cohorts to the right, beginning or ending one year later; in the case of the youngest cohort, this means it now starts in the year 1979.

Here are the charts for the two sets of shifted cohorts (same colors as before).

The left shifted chart looks almost exactly as the original chart. The right shifted one shows some differences but not significant enough to invalidate the observations I made based on the original chart. In particular, the right shifted chart seems to suggest that there is little difference in international appeal between directors from the oldest cohort and those from the 1969-1978 cohort.

Given these results, I don’t think the differences between cohorts are just an artifact of the boundary years I chose.

Conclusions

Looking at the trends I presented in my last post alongside the new insights derived from comparing the 1958-1967 cohort to the 1968-1977 cohort, and the 1968-1977 one to the youngest cohort, I have to conclude that American Hollywood directors are less appealing to non-American moviegoers than foreign directors, and the younger the American director the lower the international appeal.

I don’t believe this is a demand-side phenomenon. I can’t think of any good reason why international audiences would consciously prefer movies by older directors over movies by younger directors. Which leads me to think this is not a deliberate choice, but a preference driven by what (American) directors put into their films. It’s a supply-side phenomenon.

To quote myself:

The most likely reason for the higher international share of non-American directors is that they contribute something personal (their culture?) to their work and that something makes their films more appealing to non-American audiences.

You could also argue that American directors contribute something (American culture??) that makes their films less appealing to international audiences.

I’m not yet certain whether the demand-side factor - international audiences changing tastes - or the supply-side factor plays a heavier role in decreasing Hollywood’s international box office share, but I’m now pretty convinced that both factors drive this trend.

There’s another issue I have to mention, and it relates to the diminished appeal of younger directors.

The right shifted chart I presented above, to test the sensitivity of these results to changes in the boundary years, has a much longer oldest cohort curve spanning 12 years. This longer curve clearly shows the close similarity between the international box office share of the oldest cohort from 2002 to 2013 and that of the 1959–1968 cohort during the same period. The oldest cohort has a higher share, though the difference is small enough that I doubt whether it’s real or not.

It looks like the international share goes down with each new generation of American directors, starting with the 1969-1978 age cohort, or maybe even with the 1959–1968 if you think there’s a real difference between the oldest and 1959–1968 cohorts.

In either case, this appears to be a medium-term phenomenon, not a long-term phenomenon. Something changed. And it’s driving non-Americans away from Hollywood.

2

Ratio of international (China included) box office to domestic (Canada, US) box office revenues, from 2005 to 2025:

For each year, from 2007 to 2024, I calculated the ratio of international box office revenues - excluding Canada and the United States - for the top 200 highest grossing films of that year (numerator) divided by the domestic box office revenues for the same films (denominator). For instance, in 2007, international box office revenue reached $11.45 billion, while domestic box office was $9.36 billion, producing a ratio of 1.22.

I speculated that a decreasing share of Hollywood’s international revenue is a sign of declining American cultural influence in the world, though I did not speculate on the reasons behind the noticeable change in trend that happened around 2017-2019.

Since I used only partial data for 2024 in my previous post, I could present just two (and a half) data points following the 2017-2019 inflection period to support my hypothesis. So, I’ve decided to update the international-to-domestic ratio chart, now extended to mid-2025, to check whether the declining trend still holds.

And it does1:

And the answer is yes: the downward trend seems to intensify from 2024 to 2025, though with the caveat that 2025 data is incomplete.

Notice that I also updated the chart’s starting point to 2005, making the pre-2017-2019 rising trend all the more clear. Given the updated results, I’m now more confident that the post-2019 declining trend is real, yet I haven’t proposed any causes that might explain this phenomenon.

In the previous post I noted that:

…this mild shift in trends might be attributed to a changing selection of movie offerings from Hollywood, rather than a change in the reception of Hollywood films by the wider world. Understandably, audiences desire variety and don’t want to watch the same movie over and over again. Hollywood responds to this desire by changing its offerings, which could consequently explain the changing reception of those movies by audiences.

And I tested whether a changing selection of movies could explain the trend by looking at the international revenue of the Spider-Man movie franchise, a series of films connected by common subjects and a shared cast of characters. If the trend holds for a series of closely related movies, it’s very unlikely to derive from a changing selection of movies being offered by Hollywood, rather than by the shifting preferences of international moviegoers.

The trend does show up in the international revenue numbers of Spider-Man, with an inflection period around 2014-2019, but just to be sure I’ve now analyzed the international box office of the Transformers franchise2 (a new Transformers movie was released after my last Hollywood post).

Here’s the chart showing international revenue (as a share of total) for the eight Transformers franchise films released between 2007 and 2024.

The trend is again very clear, with the inflection point occurring somewhere between 2014 (Transformers: Age of Extinction) and 2017 (Transformers: The Last Knight). At its peak, Transformers’ international revenue approached nearly 80% of the total, but has since dropped to around 55%.

This means the decline in Transformers’ international revenue is very similar to that of Spider-Man, a drop of approximately 25 percentage points. It also makes me question my prior view that the inflection point happened around 2017-2019. I now believe there was a longer inflection period going from 2014 to 2019, and any proposed explanations put forward to account for the change in Hollywood’s international revenue trend should fit with that time period.

So, we have a rising trend in Hollywood’s international box office until the 2014-2019 period, followed by a declining trend after that inflection period. And this last trend appears to be driven by a change in the tastes or preferences of international movie audiences (demand side, not supply side).

Direction downward

While trying to figure out what attributes of a movie have an impact on its international revenue, I compared films with a high international-to-domestic ratio to those with a low ratio and noticed an intriguing pattern: the choice of movie director appeared to have an impact. Based on a small sample of movies, it seemed to me that the national origin of a film’s director has an effect on the international revenue share of the film.

To check whether this is actually true, I looked at the nationality of directors for the top 60 highest-grossing Hollywood movies from 2003 to 2025, identifying them as either American3 or foreign based on their birthplace (as noted in Wikipedia). Then, I calculated the average international box office share for movies directed by Americans and those directed by non-Americans each year, excluding movies not made in Hollywood.

So here’s the chart showing international share for American directors and foreign directors.

You can see in the chart that the average international revenue share of non-American directors is consistently higher than the international share of American directors (aside from 2007). It seems that non-American Hollywood directors have a certain quality, perhaps in greater abundance than their American peers, that is valued by international audiences and makes their movies more appealing to those audiences. It could also be that American directors are simply less appealing to international audiences.

Furthermore, these are all Hollywood movies, so there’s no reason to think that the source material for these films (script, original book, comic, etc.) is more appealing to American audiences when directed by an American than when a non-American director is in charge4.

Maybe the share of non-American directors has decreased in the last few years?

Less non-American directors working in Hollywood could bring the average international box office share down. But this has not been the case. There is no decrease in the share of non-American directors in recent years that could account for Hollywood’s declining appeal to international audiences5.

This agrees with the decline in the international revenue share for both American and foreign directors during the last few years, as seen in the previous chart. Non-American directors are not immune to whatever is causing the downward trend of international revenue.

Well, that was all I was planning to write about in this post… until I noticed that Hollywood’s international revenue has become a hot topic in the last couple of weeks, due to the release of Superman and The Fantastic Four: First Steps.

For example, renowned American media commentator Angry Joe says6:

He then goes on to blame Superman’s low international revenue on: ‘American Way’ Reputation is BAD Right Now. Trump Tariffs Era.

While Norwegian youtuber Andre Einherjar says7:

And then proceeds to quote Superman’s director James Gunn who said: And it also affects things that we have a certain amount of anti-American sentiment around the world right now. It isn’t really helping us.

Well, assuming that right now refers to the last couple of years, those views8 are wrong.

The decline in international revenue share began in 2019 at the latest, and given this general decline together with the box office trend of the franchises I mentioned earlier (Transformers, Spider-Man), the low international box office of these recent movies is not surprising.

In the case of Superman, we have figures from two previous Superman films released during (or near) the peak years of international share: Man of Steel (2013) had a 56,6% international share and Batman v Superman (2016) got to 62,2%. The current release, with a 42,6% international share, represents a drop of roughly 20 percentage points from the peak, similar to the decline we saw in the other two analyzed franchises.

I also doubt that Hollywood’s declining appeal derives from any conscious preference by international audiences. When pondering whether to watch a newly released movie, most moviegoers probably pay no attention to the director’s national origin. The most likely reason for the higher international share of non-American directors is that they contribute something personal (their culture?) to their work and that something makes their films more appealing to non-American audiences.

One could also argue that American directors contribute something (American culture??) that makes their films less appealing to international audiences.

Culture and tastes are not static. They are something dynamic. And what used to be appealing a decade ago may be less captivating today.

[If you liked this post, you might also like its follow-up: The declining appeal of Hollywood: another piece of the puzzle.]

5

A chart of foreign to American directors ratio (two-year average):