What's really true about intelligence and IQ? We empirically tested 40 claims
- Nikola Erceg, Spencer Greenberg, and Beleń Cobeta
- Sep 20, 2024
- 13 min read
Updated: 7 days ago
You've probably heard lots of claims made about IQ - for instance people saying that it's important and captures most of intelligence, or that it's meaningless or pseudoscientific. Lots of claims about IQ are also made in the academic literature, such as that it can predict a variety of life outcomes. But what's actually true about IQ? We set out to answer this by running a giant study aiming to check many claims about it to see if they hold up. This report contains our findings on 40 such questions, like:
What's the relationship (if any) between IQ and income?
Are people with higher IQs happier or less happy?
Do psychopathic and narcissistic people have higher IQs or lower IQs?
Are there any intelligence tasks that women do especially well on, and are there any that men do especially well on?
Is there any relationship between personality and IQ, and if so, what is it?
Perhaps the most critical claim in the IQ literature is that IQ can be estimated from performance on just about any diverse set of intelligence tasks. For this study, we measured IQ using the following method:
Each participant was randomly assigned to do intelligence tasks from a pool of 62 distinct intelligence tasks (each testing a different skill, such as spelling, math, vocabulary, logic, and so on). On average participants completed between 6 and 7 tasks.
Following a common practice, we converted each person's performance on each task to z-scores (meaning we subtracted and multiplied so as to make each task score have a mean of 0 and standard deviation of 1), and then conducted principal component analysis on these scores to find a "common factor" among the 62 intelligence tasks (after converting task scores to z-scores), and a weight was assigned to each task based on how strongly it correlated with this common factor. Each participant's IQ score was based on a weighted average of their z-scores on whichever of the 62 tasks they completed, where the weights were determined by the correlations with this common factor. IQ scores were also normalized (through simple subtraction and multiplication) aiming to make the average American's score be about 100, with a standard deviation among Americans of roughly 15.
We estimate our IQ test’s correlation with a hypothetical perfect measurement of the general intelligence factor (g) to be between 0.76 and 0.85. To determine the lower bound, we randomly split each participant’s tasks into two independent task sets, generating two separate IQ scores (IQ₁ and IQ₂). The correlation between these scores was .58. Since both are equally representative of g (on average), their correlation to g is estimated to be √.58 ≈ .76. This serves as a (rough) lower bound because each score is based on only half of the available tasks, and so using all tasks to estimate IQ (as we did in this research) will produce a more accurate measurement of IQ. For the upper bound, we re-tested participants from the initial study several months later using a refined version of the test (that, on average, contained few items in common with the original version participants had taken). The correlation between their original and new IQ scores was .76. Assuming the new and old version were equally accurate and independent, this would yield an estimated correlation with g of √.76 ≈ .87. This likely represents an upper bound due to some minor task overlap between the two test administrations, and because the new (refined) version was likely at least somewhat more reliable than the original version that participants had taken. With this in mind, correlations in this report between measured IQ and different outcomes may be approximately 20% smaller than they would be if our IQ measure was a perfect (noise free) measurement of g.
Our sample consisted of n = 3691 participants (61% male, 37% female). The mean age of our participants was 37.4 with a standard deviation of 13.
It's important to note that our participants came from two different sources. A total of 1853 participants came from Positly, which is our platform for recruitment of participants for studies. The rest (n = 1838) of participants were recruited through social media posts (and posts on other sites, such as reddit) calling for participation in the study. Importantly, these two subsamples substantially differed in number of characteristics, the most prominent being age, gender and average IQ. The Positly sample was older than the social media sample (mean ages of 41.7 vs. 33.0) and more gender balanced (the ratio of male participants to female participants in the Positly sample was 0.84 male participants per 1 female participant, while in the social media sample it was 3.95 male per 1 female). Finally, non-Positly social media sample had on average substantially higher IQ estimates than Positly sample (IQ = 120.65 vs. IQ = 100.35).
It's also important to note that given our large sample size, even very low correlations are statistically significant by conventional criteria (i.e., achieving p<0.05) as statistical significance depends on sample size. It does not mean, however, that such results are practically significant. Therefore, whenever the absolute value of correlation was lower than 0.10, we considered it a non-meaningful effect, even if the result was statistically significant. Additionally, the fact that our study showed or failed to show some effect, does not necessarily mean that the effect actually exists or does not exist. We can never completely trust a single study because, although the chances for it are low, the effect could just be a false positive (meaning that we detected the effect in our sample by chance even though it does not exist in a population), just as well as the lack of effect could be a false negative (meaning that we failed to detect the effect in our sample by chance, even though the effect exists in a population). It also could be the case that the effect we found was the result of the particular populations we conducted this research on. To help reduce the chance of that being the explanation, we have controlled for age, gender and data source whenever the hypothesis being tested is not related to one of those variables.
Note that because of the large number of study participants used in this research, correlations reported here will typically be statistically significant any time that the correlation magnitude is at least r=0.1 (so long as it's an analysis that involved a sample size of at least 400 people). With a sample size of at least 400 people, even a correlation of only r=0.15 will have a low p-value of less than 0.003. Therefore we do not bother reporting p-values in this report.
Many of the correlations shown in this report are small or only modestly sized. When referring to the size of the correlations, we are relying on Cohen’s (1988) criteria for small, medium and large effect sizes. Specifically, here is the nomenclature we are using depending on the size of correlation:
Not all of our analyses include the full sample of participants. Given that our full questionnaire would be intolerably long had we given all the questions to all participants, some of our tests and questionnaires were given to only a fraction of participants. So, for these variables, the analyses were conducted on a smaller sample of participants that received that particular test or questionnaire.
While we believe our IQ tasks did quite a good job of measuring IQ for most study participants, due to the nature of our study (with each study participant getting a random sample of the 62 intelligence tasks) more accurate methods of doing IQ testing exist. The more noise there is in a measure of IQ, the lower the magnitude of correlations will be found with all other variables, on average. As such, our analyses are likely to slightly underestimate correlations with IQ compared to more accurate tests.
As a final note, remember that a correlation does not necessarily imply causation. The results below reflect correlations between IQ and many other different variables, but that does not necessarily mean that IQ causes changes in those variables. For instance, suppose we find a link between IQ and the personality trait of "openness to experience." This could be because:
Higher IQs cause greater openness
Higher openness causes higher IQs (e.g., maybe more open people learn more voraciously when young which causes higher IQs)
Some third variable separately causes both higher IQ and higher openness (e.g., perhaps being raised by parents that strongly value education causes kids to both be more open and to have higher IQs)
Higher IQ and greater openness both cause each other (e.g., maybe higher openness leads to more learning in childhood which leads to a higher IQ which leads to a more rewarding experience in school which leads to even more openness)
For more about how to interpret correlations, see our article here.
With these caveats covered, let’s dive deeper into the results.
Summary table
Here is a brief summary of the findings in this report. Click on the corresponding research question to go to that part of the report. Note that all of our analyses involved controlling for age, gender and data source (except for research questions where that would be inappropriate, such as when looking at the relationship between age and IQ).
Research question | Main takeaway(s) |
In our sample, IQ was normally distributed, which agrees with prior studies. | |
There is a positive manifold of intelligence tasks, meaning that performance on nearly all intelligence tasks is positively correlated with performance on nearly all other ones. | |
Higher IQ people are more successful in using their time as they would ideally like to than lower IQ people. | |
To the extent IQ predicted variables that we tested, it did not predict them more strongly on the left side of the IQ distribution. The only notable exception to this was the "good employee self-report score" (an indication of how good people believe they are at their jobs). IQ was more correlated with this score for those with lower IQs than for those with higher IQs. | |
Looking across ages for the population at a single point in time, fluid intelligence seems to rise throughout younger age, then remains stable in adulthood and steadily declines in older age (after 50). Crystallized intelligence also rises during early adulthood, but then plateaus. | |
Our study data matched what is typically referred to as the Dunning-Kruger effect, though the interpretation of such data is complicated, and may not mean what it is generally believed to mean, as we discuss in our report here about the Dunning-Kruger effect. | |
There is little to no relationship between IQ and conscientiousness (other than, perhaps, some relationship with a small number of specific conscientiousness items). | |
IQ has a small positive correlation with the "intellect" facet of openness, but not with other measured facets. | |
There is a small positive correlation between IQ and agreeableness, as well as between IQ and the "empathy" facet of agreeableness. | |
Higher intelligence people were generally less extraverted in our study with correlation sizes ranging from small to moderate. | |
IQ and emotional stability / neuroticism are not related, i.e. the correlations between IQ and emotional stability / neuroticism factor and facets were negligibly small. | |
The relative importance of IQ and personality depends on the outcome: for some outcomes, such as high-school and college GPA, they predict approximately equally well, while for some outcomes personality is much stronger predictors (happiness and life satisfaction). Overall, when compared to all of the Big Five personality traits together, IQ was on average a weaker predictor than personality on the outcomes we tested. The effects of IQ and personality tend to be additive, so using both typically makes predictions more accurate than just using one. | |
We found higher IQ people to be lower on two out of three dark triad traits, narcissism and sadism, but not on machiavellianism (with which there was no correlation). However, meta-analyses appear to find no relationship between IQ and these traits. | |
IQ is linked to higher levels of education obtained and to high-school GPA, but its relationship with college GPA is quite lower than with high-school GPA. | |
IQ predicts better self-reported job performance, but only in a subsample of participants with IQ that is lower than average. However, in that group of people, the link between IQ and self-reported job performance was quite large. | |
Higher IQ is linked to greater income but the correlation is small. Part of this effect may be that higher IQ people can get hired and perform better at some high paid jobs, but that is probably not the full explanation. | |
IQ likely has little to no correlation with either momentary happiness or life satisfaction. | |
IQ was not related to any of the 14 mental challenges we screened in our study, and more broadly there is a lack of consensus on the relationship between IQ and mental health. | |
In our study childhood poverty and low childhood socioeconomic status were not related to IQ, although this contradicts findings from other studies that find a modest negative correlation between these factors and IQ. | |
We found that some aspects of childhood nutrition, specifically self-reports (as an adult) of having enough food in childhood and being breastfed as a baby, had a small positive correlation with later life IQ. | |
IQ differs depending on the family structure in which a person grew up, being highest in those from nuclear families compared to other family types (extended family or stepfamily). | |
In our study, childhood adverse experiences had little to no association with adult IQ, but it must be noted that other studies found that these traumatic experiences are associated with detrimental effects. | |
People who say that they were read to more in childhood also score higher on IQ tests in adulthood, and this effect does not appear to be due to childhood wealth or childhood social class. | |
People that report not having been breastfed in infancy seem to have slightly lower IQ than people that were breastfed. | |
Just like in other studies, in our study measured IQ was moderately related to the self-estimated IQ. Both higher IQ and lower IQ people may have a tendency to estimate themselves closer to the average than they really are. | |
Those with higher IQs tend to have more socially progressive/liberal/left (i.e., less socially conservative/right) views. | |
Higher IQ people would like to see stricter gun laws compared to people with lower IQs. The size of this effect was cut in half when we controlled for political ideology. | |
Higher IQ people are more likely to have tolerance for groups that they politically oppose with correlations between IQ and tolerance being small to moderate. | |
Higher IQ people are more prone to actively open-minded thinking with the medium-large correlation between the two ("the willingness to consider alternative opinions, sensitivity to evidence contradictory to current beliefs, the willingness to postpone closure, and reflective thought"). | |
Grit ("passion and perseverance for long-term goals") and IQ have little to no correlation. | |
IQ is generally positively correlated to a variety of different self-reported behaviors that one might think could be associated with IQ, such as enjoying solving riddles/difficult puzzles, finding that math comes easy, and being interested in science, but there were a few surprising negative correlations as well, such as believing one could figure out solutions for society's big problems. | |
People with lower IQ are more likely to report playing the lotto, watching more TV, keeping up with celebrity gossip, have difficulties filling out complicated forms, and getting bored with just sitting and thinking than people with higher IQ. | |
On average, women appear to perform slightly better than men at verbal tasks related to word production, while men appear to outperform women on spatial tasks. It's unclear, however, why these differences occur. | |
Higher IQ people may be less susceptible to believing that B.S. "pseudo-profound" statements are profound. | |
Lower IQ is related to a higher obsession with celebrities and pathological attitudes toward celebrities. However, this is not true about more typical positive feelings towards celebrities (e.g.,, really enjoying watching, reading or listening to them), in which case there is no relationship to IQ. | |
Both verbal and numerical intelligence is positively correlated with the ability to accurately recognize emotions from facial expressions. | |
In our study, IQ was not related to charitable behavior, though this contradicts typical findings by others on this subject. | |
IQ predicted only two of nine healthy behaviors that we measured. Higher IQ people were less likely to use drugs and to smoke than people with lower IQ. However, other studies have found broader positive links between IQ and healthy behavior that we did not find. | |
Participants who were feeling nervous or anxious both before and while taking IQ tests performed worse on those tests. | |
Room conditions such as temperature or freshness of the air had minimal effects on IQ (though we did not test extreme conditions, just the natural conditions people found themselves in). |
If you'd like to read this full report as one long PDF, you can download it here.
And if you'd like to uncover your cognitive strengths and understand and where your intellectual strengths and weaknesses lie, try the cognitive assessment tool that we developed out of this research:
