New article published: Honesty, intelligence and race

Together with Sebastian Jensen, I have a new article:

Research shows that honesty is positively correlated with intelligence. Similarly, there are racial differences in honesty, with Europeans more honest than other ethnic groups. It is currently unknown to what extent race differences in intelligence can explain differences in honesty. We investigated this question using data from the National Longitudinal Study of Youth 1997 (NLSY97), a large US longitudinal data set. We replicate previous findings that honesty was correlated with measures of intelligence (r = 0.38, 95% CI [.34, .41]) and that blacks (d = -0.67, 95% CI [-.76, -.59]) and Hispanics (d = -0.4, 95% CI [-.50, -.31]) are less honest than Whites, and this is true whether honesty is measured by self-report, interviewer report, or parent report. Furthermore, race differences in honesty remained between blacks and whites, but not between whites and Hispanics after controlling for intelligence. Differences between Blacks and Whites, but not Whites and Hispanics, were notably lower in self-reports (Blacks: d = −0.18). [-0.24, -0.11]Hispanics: d = -0.24 [-0.31, -0.17]) than parent reports (Blacks: d = -0.43 [-0.52, -0.35]Hispanics: -0.24 [-0.33, -0.15]) and interviewer reports (Blacks: d = -0.7 [-0.75, -0.64]Hispanics: -0.3 [-0.36, -0.25]). Cross-national comparisons were made using national IQ data and Hofstede’s cultural dimensions. The Bayesian model mean suggests that Hofstede’s dimension of individualism (β = .64, PIP = 100%), national IQ (β = .25, PIP = 73.6%) and masculinity (β = -.35, PIP = 100%) predict differences in honesty between countries. Parking violations by diplomat were only predicted by national IQs (r = -0.28, p < 0.001), given that no other variable reached a posterior inclusion probability greater than 0% apart from national IQ. Implications and theories about these findings are discussed.

It is well known that there are racial differences in crime rates, including for fraud and similar crimes involving deception. Similarly, there are large national differences in the level of corruption, with Northern Europeans are the least corrupt:

Another approach to looking at honesty at the country level is to do experiments. A 2019 study had informants pretend to have found a wallet and hand it to hotel receptionists to see how many were returned. The results look like this:

Regarding the differences in this portfolio study, we tried to explain them using national intelligence and Hofstede’s cultural dimensions:

Here we used Bayesian model averaging (BMA) because our number of predictors was small for the number of countries (n = 35). The most important predictor was individualism (beta = 0.64), which is basically what differentiates Northern Europeans from the rest of the world (The strangest people in the world stuff, but see also Kevin MacDonald’s work). There was also a plausible role for national intelligence (beta 0.25) and masculinity (beta -0.35, think Russian-style macho behavior). It should be noted that with this small sample size, these conclusions are not very firm.

Because of the known group differences, one might wonder if there are differences in more general everyday honesty. We examined this question using data from a large US survey (National Longitudinal Study of Youth 1997 (NLSY97)).

Measuring honesty in a survey is tricky. You can’t simply ask a potentially dishonest person if they’re honest or not and expect an honest answer, since they might as well be lying to that question. Nor can they lie, for example, if they don’t even care to lie in general. Alternatively, parents can be asked to rate their children. Again, this isn’t perfect because parents have some level of positive bias toward their kids (“he was a good kid”). A more impartial account can be given by an interviewer, who is neither a friend nor a family member. The limitation with them is that they have not known the person for a long time, but only for the purpose of interviewing them for the survey. Data from all three approaches were available for the study. Here are the race gaps:

The results align with expectations based on the amount of bias for Blacks: self-reports produce the smallest gaps with Whites and have the most bias, parent reports produce the largest gaps and have the weakest bias, and interviewer ratings show the largest gaps with the least amount of bias. For Hispanics, there was no such pattern, which is strange given the results for blacks. Maybe a statistical fluke? It cannot be completely ruled out given the sample sizes.

From a psychometric perspective, we might expect the three measures of honesty to be positively correlated, since they are all indicators of a general honesty factor. They do indeed correlate, but not by much:

The correlation between the 3 different honesty variables was quite low. The tetrachoric correlation between self-reported honesty and interviewer-reported honesty was .07, the tetrachoric correlation between self-reported honesty and parent-reported honesty was .32, and the tetrachoric correlation between parent-reported honesty and interviewer-reported honesty was .18.

Still, it can be investigated with the imperfect scores of this general honesty factor which is (hopefully) less biased and more reliable than each of the indicators:

For Hispanics, the difference is larger on the overall honesty score than on each individual measure, consistent with expectations; but for blacks, the gap is slightly smaller than for the interviewer report measure alone. Perhaps the black parent and parent report biases are strong enough to offset the increased reliability of combining them. The honesty distributions look like this:

Most people of all races are pretty honest, but there is a tail on the left that is very dishonest, and that tail is much larger for blacks than for whites and Hispanics. This is where we expect to find most career criminals, serial welfare fraudsters, etc.

Looking at the national rankings above, we saw that smarter countries are generally more honest, and China is a notable and large negative outlier (presumably related to communism). Similarly, we know from decades of research that criminality correlates negatively with intelligence, so we would also expect honesty to correlate positively with intelligence. In fact, this is what we find:

A correlation of about .40 is pretty big! Bigger than I expected. Theoretically, that is yet underestimated because of the imperfect reliability of the overall honesty score and because of the ceiling effect of the honesty scores. Indeed, based on the low correlations among the measures of honesty, the reliability of the overall score should be quite low — perhaps .50 — so the adjusted correlation between intelligence and honesty should be quite high, perhaps .60. Such a high value is hard to believe, so this requires further study.

On a similar note, one might want to know whether race differences in honesty can be explained by parental socioeconomic status (SES):

Looking at Model 1 and Model 4, we see that parental SES does not predict honesty after controlling for intelligence. But even controlling for intelligence, blacks remain slightly more dishonest than whites, an effect that was not seen for Hispanics. The same pattern of results was found for each of the three methods of measuring honesty, albeit with weaker results (beta for self-reported honesty intelligence 0.09, for parent report 0.24, for interviewer report 0.30).

• Honesty differs by race, probably by a substantial amount.

• Honesty correlates with intelligence, probably not below .40, even if that seems hard to believe.

• Intelligence explains most of the racial differences in honesty, but race likely predicts honesty independently of intelligence. This is probably due to differences in personality traits and culture, mainly individualism.

• The results are potentially relevant to explain the prevalence of scientific dishonesty seen outside the Western world, and by foreigners in the Western world. Given the ease with which scientific dishonesty can be discovered if one cares to look at the articles with a critical eye, I predict that an AI capable of automatically detecting scientific dishonesty will soon be created. Then they will check all the published studies in the world and identify thousands of fraudulent or very incompetent researchers. There will be a large over-representation of non-Europeans among them.