The case for border control

The argument:

1. Intelligence causes economic growth at a national scale.

2. This is largely because of group, not individual level effects.

3. Immigrants resemble their countries of origin.

4. If developed countries open their borders to unselected immigrants from undeveloped countries, the standard of living of the native population will drop.

I will outline the necessary evidence for each statement, as well as rebuttals to the following objections later in the article:

OBJ1: national IQs are not valid measurements of intelligence.

OBJ2: the differences in IQ between nations in intelligence are not hereditary.

OBJ3: racial diversity decreases social trust, which leads to a decrease in support for economic redistribution, which increases economic prosperity (Hanania argument).

OBJ4: “it’s good for the economy”.

OBJ5: immigration does not reduce the smart fraction of the developed world, which are the main determining factor in the average levels of intelligence of a nation.

OBJ6: the criminality and fiscal effects of immigration can be dealt with by changing public policy.

OBJ7: AGI/strong AI makes this all irrelevant.

1. Intelligence causes economic growth between nations.

Consider the relationship between GNI per capita and national IQ:

from here

There are a few notable outliers in the upper direction, though they are all tax havens or oil producing countries; the downwards outliers tend to be former Communist countries. Otherwise, a country’s level of development can be easily predicted by its average level of intelligence.

The relationship is not linear, as can be seen visually or using formal statistical testing.

GNI ~ rcs(NIQ, 5) is the spline model, GNI ~ NIQ is the linear model

Concretely, the relationship is exponential, and an IQ point is associated with a 9.2% increase in the GNI per capita of the country.

(e^.0876-1)*100 = 9.2%. I got this wrong in the initial twitter thread, as multiplying by 100 alone doesn’t work after the effect reaches a certain threshold.

Because the correlation between the variables is so high, the only plausible explanations that can be made are that national IQ causes economic growth or that economic success increases the average IQ of a nation. We know that intelligence is associated with job performance and income within individuals, and that undeveloped regions tend to have poor levels of nutrition and are therefore stunted — so bidirectional causality should be considered the default hypothesis.

Separating the variance into genetic and environmental categories is somewhat nonsensical, as genetic differences between populations will cause them to develop at different rates, so the environmental tax of poverty cannot be considered wholly environmental in origin. That said, North Korean refugees tend to score about 15-20 IQ points below South Koreans (depending on how Flynn Effects are corrected for), and Sub-Saharan Africans score about 15 points below American Blacks, so plausibly some of these effects are due to reverse causality:

A clear example of how a bad environment can hurt IQ can be seen in the IQ scores for sub-Saharan African countries. They average only around 70. In contrast, African-Americans average about 85. It appears unlikely that African-Americans' white admixture can account for most of this 15-point gap because they are only around 17%-18% white on average, according to the latest genetic research. (Thus African-Americans white genes probably couldn't account for more than 3 points of the gap between African-Americans and African-Africans.) This suggests that the harshness of life in Africa might be cutting ten points or more off African IQ scores.

2. This is because of group level, not individual level effects.

There are two reasons that intelligence can cause economic growth at a national level:

1. IQ causes higher levels of income at an individual level.

2. There are group-level effects of intelligence on economic development that go beyond the individual effects.

At an individual level, higher IQ individuals do earn more wages — while nobody in the know denies the causality of the relationship, many dispute the strength of it. For what it’s worth, the effect is 2% in the NLSY79 (born in the late 50s and early 60s) and 1.7% in the NLSY97 (born in the early 80s) — I consider these the most accurate estimates of the effect size in the USA for several reasons:

• The sampling is not restricted to any one region of the US.

• The test involved (ASVAB) is of high quality.

• The sample is broadly representative of the USA’s population

Some other datasets like the WLS have a weaker association, though the sample is not representative as it is a sample of White high school graduates, and the weaker correlation between IQ and income in this dataset appears to be due to this:

I've looked deeper into the question of why we see a weaker IQ/income correlation in the WLS (~0.28) versus the NLSY79 (~0.4).

Two hypotheses: First, it's due to differences in the sample; second, it's due to differences in the IQ test (Henmon-Nelson versus AFQT). From the NLSY79, I created a sample that I think is similar-ish to the WLS. I included only high school graduates and white Midwesterners (Wisconsin was 99% white when the WLS cohort was born). I found quite a bit of attenuation; the correlation dropped from 0.4 to 0.32 (already close to the 0.28 from the WLS)…

The largest meta-analysis to date places the correlation at .22, which is deflated by the fact that many of the studies do not use full scale IQ tests (e.g. the PPVT, WORDSUM) or measure income too early in life, which results in an attenuated correlation.

The national-level correlation, on the other hand, is much stronger, as one IQ point increases income by 9.2% — so about 20% of the relationship between countries can be attributed to effects between individuals.

3. Immigrants resemble their countries of origin in terms of traits

This is true for test scores (this data is from the PISA exams):

from here

And even attitudes towards herbal medicine:

from here

The differences between different immigrant groups in fiscal impact is very similar to the map of IQ by country:

from here

from here.

Differences in income between different American immigrant groups are correlated with the average IQ of the origin countries:

from here

4. If developed countries open their borders to unselected immigrants from undeveloped countries, the standard of living of the native population will drop.

This must be the case if the effect of intelligence on national development is due to group-level effects and not individual-level ones, and that unselected immigrants from undeveloped countries will lower the average IQ of the nation.

#########################################################################

Objections

1. National IQs are not valid measurements of intelligence.

I don’t have a strong opinion with regard to intelligence tests being fair measurements of intelligence across nations, but differences in PISA/PIRLS/TIMSS test scores between countries highly correlate with the differences in intelligence.

Consider the higher learning outcomes database:

HLO database

The database was produced through a large-scale effort by the World Bank to identify, collect and collate student assessment data worldwide. We include seven assessment regimes in total: three international tests, three regional standardized achievement tests and the Early Grade Reading Assessment, which adds 48 countries to the database with at least one data point in the past 10 years, including large developing economies such as Bangladesh, Nigeria and Pakistan. Each test covers between 10 and 72 countries. By combining these assessments and making them comparable, we include countries that represent 98% of the global population. A detailed description of the methodology that we use to develop harmonized learning measures as well as all data included in the database are provided in the Methods and Supplementary Information II.

Their map:

from here

Or, the basic skills dataset:

We separate the estimation of world skill levels into five layers of decreasing reliability that indicate different degrees of certainty and precision in the comparability of available test information. Layer 1 includes countries that have participated in any wave of PISA or PISA for Development (PISA-D) – a total of 90 countries. Layer 2 adds countries that have participated in the Trends in International Mathematics and Science Study (TIMSS) but not in PISA – 14 additional countries. Layer 3 incorporates countries that have participated in regional tests – TERCE and SERCE in Latin America and SACMEQ and PASEC in Sub-Saharan Africa – but not in PISA or TIMSS, an additional 20 countries. Layer 4 adds the two countries – India and China – that have some sub-national test information but that have not fully participated in the international assessments.

Their map, in IQ scale:

from here

The most reasonable objection I could think of is that the rank order of the differences is accurate, but the magnitudes of the differences are exaggerated by test bias or bad sampling. Taking the objection seriously, I created a different set of national IQs, where the absolute differences from 100 were 40% smaller than the observed ones, and ran the log(GNI) ~ NIQ regression again using the data from this paper. When this adjustment is made, the effect of the average IQ on GNI per capita increases — an increase in one point of national IQ is associated with an increase in GNI per capita of 15.6%.

(e^.145-1)*100 = 15.6

2. The differences in cognitive ability between countries are not due to heredity.

They are, but my argument with regard to immigration does not rely on hereditary differences, just the persistence of them across groups.

3. Racial diversity decreases social trust, which leads to a decrease in support for economic redistribution, which increases economic prosperity.

Hanania has made the argument for diversity (and therefore immigration) by arguing that America’s prosperity comes from economic policy, which is downstream from ethnic diversity, noting that individual American states are much wealthier than what would be expected from the average IQ of the United States.

from here

This suffers from what I call the “multiplied effect size” problem, where multiplying multiple effect sizes that have moderate effect sizes leads to a final effect size that has a very small effect. Conceptually, if the effect racial diversity has a standardized effect of -0.4 on social trust, and social trust has a standardized effect of -0.4 on economic redistribution, and economic redistribution has a standardized effect of -0.4 on GDP per capita, then the effect of racial diversity on GDP per capita should be expected to be 0.064.

Empirically, based on Emil Kirkegaard’s analysis, there is no relationship between racial diversity and economic prosperity after accounting for the nation’s average IQ.

from here

Besides that, I don’t actually think racial diversity and social trust correspond that closely — the correlation in the United States arises from the fact that diverse census tracts tend to have more Black and Hispanic people. Controlling for the main effects of each race makes the relationship between ethnic homogeneity and social trust is weak:

from here

As for why the United States has a lot of money, I’m not even sure if it has that much to do with economic policy; the strength of the dollar is an obvious contributing factor, and Americans also work more than what you would expect from the levels of development observed in their economy.

from here

4. Immigration is good for the economy

Economists tend to be strongly for immigration. Why this is doesn’t really matter to me; what does is if they are correct. The basic economic argument for immigration not increasing unemployment or lowering wages goes something like this:

1. Immigrants become employed.

2. They earn wages.

3. They use those wages to buy goods.

4. That increases the demand for goods.

5. The demand for goods causes an increase in the demand for employment.

6. So immigration should not depress wages or increase unemployment.

This is then justified by a bunch of fancy mathematical models that thirty people in the world understand (I am not one of them), which is unnecessary; anybody with a functioning brain can see that the argument makes sense. The problem I have with this argument is that in practice unemployment among immigrants is extremely common in Europe [1] [2], and that it ignores the group-level effects of intelligence on national productivity.

There is a 2nd argument that goes something like this:

1. Low skill immigrants take low skill jobs.

2. Natives are crowded out of the low skill jobs.

3. This frees them up to work in higher skill industries.

4. Therefore, their pay increases.

The economists I cited used their usual fancy mathematical models to argue that the average pay of natives will increase due to low skill immigration. Like in the previous example, I don’t understand the models they used.

And I don’t believe them. Consider that, in white collar jobs, most of the people working are not that productive, and that productivity is pareto distributed. Workers who were previously working in low skill jobs, get crowded out by immigrants, and then apply to high skill jobs are probably going to be less productive on average than the people already working there. Because of that, if they get hired, then the average productivity of the field will lower, and the workers will receive lower wages.

An argument could be made that the new workers could just be paid lower wages than the old ones, but society doesn’t really work that way. Wages are sticky, and differences in pay don’t correspond that well to differences in productivity between individuals (consider that employers are not that good at judging productivity in the first place — inter-rater reliability of productivity ratings is about .6),

5. Immigration doesn’t reduce the smart fraction of a country, which is what decides economic growth

As far as I know, there is no proof that smart fraction theory is true, so it can be dismissed. A priori, given that the effects of intelligence on life outcomes are linear, it should be expected that the effect of average intelligence on development at each part of the distribution of intelligence should be equal.

As smart fraction theory is a common belief in HBD circles, I may as well do the due diligence of refuting it. Starting with La Griffe du Lion, who argued that the relationship between national IQ and GDP per capita in driven by the smart fraction:

from herefrom here

The smart fraction fit looks pretty good. The parsimonious explanation that he ignores is that the relationship between national intelligence and GDP per capita is exponential — note the linearity between log(GNI per capita) and national IQ.

from here. Blue = LOESS, Yellow = OLS

I also tried testing whether the raw correlation between the predicted fraction size of the 125+ population was more predictive of socioeconomic development than the raw average. It was not.

from here

There is the Kirkegaard paper, but this is testing a different theory: that average level of ability at the 95th percentile is more predictive than the average. And it’s confounded by the fact that the 95th percentile of ability might be more predictive of the average than the average itself; some countries cheat with the averages of the PISA/TIMSS tests by excluding low SES or immigrant students, or only sample a particular region of the country. These decisions should have a smaller effect on the 95th percentile than the average.

If I had to guess, a nation with people who only have an IQ of 101 would probably have a worse economy than a nation with a regular distribution that has an average IQ of 95. My rejection of smart fraction theory is more driven because I have a different theory: that the relationship between IQ and income is driven by the average IQ of the ruling group. In the United States, for example, differences between Hispanics and Whites in income vanish after controlling for intelligence. But I doubt the differences between Mexicans and Americans would — that’s because Hispanics in America benefit from the group-level effects of the American economy. It’s also a nice fit for South Africa’s trajectory.

6. Immigration can be dealt with by changing public policy.

This is the objection I see made to the fiscal effects of low skill immigrants that arise due to use of public services and petty crimes. Theoretically this can be done, but policy inevitably decays due to the fact that Cthulhu swims to the left.

7. AGI/strong AI makes this all irrelevant.

I am an AGI denialist, so I’ll focus on strong AI.

This is the only argument that I find somewhat convincing, but I reject it on the grounds that the economic effects of AI are still unknown and potentially not that large. LLMs for instance have not changed the economy very much, despite being used by 200M people.

I could still believe that AI will free up a lot of labour that works in the image generation and customer support businesses. But I find it hard to believe that anything that can be replaced by AI, will be replaced. Machine learning has been competitive with doctors in diagnostic abilities (even out of sample) for a long time. But doctors are not getting replaced by artificial intelligence, so I must infer that the efficient market hypothesis is incorrect with regard to technological replacement.

I would assume that, in the future, a lot of work being done at the moment will be automated. But it’s not clear how fast this would be, and mass third world immigration seems like an extremely unnecessary risk to take.

Appendix

This is something that didn’t fit well into the main article, but I failed to replicate the ‘10 IQ point rule’ in computational code.

#model is GDP/P = e^(m*IQ_mean+b+r)
#m = .088
#b and r are irrelevant
#for the USA, p = 330000000 and IQ_mean = 97 for simplicity's sake
#Fediq is the average IQ of the immigrant
#Expected difference in GDP per capita is 2.71^(0.088*(TIQ+Fediq)/(p + 1)) - 2.71^(0.088*(TIQ)/(p))
#R code:

p = 330000000
TIQ = 330000000*97

daf=data.frame(immiq = 90:108)
for(Fediq in 90:108) {
   daf$econeffect[Fediq-89] = 2.71^(0.088*(TIQ+Fediq)/(p + 1)) - 2.71^(0.088*(TIQ)/(p))
}

> daf
   immiq    econeffect
1     90 -9.237734e-06
2     91 -7.918058e-06
3     92 -6.598381e-06
4     93 -5.278705e-06
5     94 -3.959030e-06
6     95 -2.639353e-06
7     96 -1.319677e-06
8     97  0.000000e+00
9     98  1.319676e-06
10    99  2.639353e-06
11   100  3.959029e-06
12   101  5.278705e-06
13   102  6.598381e-06
14   103  7.918057e-06
15   104  9.237734e-06
16   105  1.055741e-05
17   106  1.187709e-05
18   107  1.319676e-05
19   108  1.451644e-05