The phantom of the lazy genius
Why the relationship between intelligence and success is stronger than myth makes it
Recently, a flurry of discussion crossed my Twitter feed about the smart kids who don’t succeed, a lot of the discussion being driven by the below graph and a viral post asking people to describe the handful of individuals in the red ellipse. It’s worth noting that this is not a large segment of the population - only a small fraction of the population gets labeled as “lazy geniuses.”
As someone familiar with the guts of social science research and intimately familiar with the trope of the lazy genius, I’d like to talk about lazy geniuses, misfits, IQ, and intelligence for a little bit. After all, I am a bit of an odd duck myself. My IQ was measured at 1411 and I had a GPA of 3.4.2
I was, in other words, at the right-hand edge of the red ellipse on this graph, and it’s not hard to imagine having been near the middle of it.
In high school, I made a fine show out of being a “lazy genius.” In college, it was a little harder to maintain that illusion. What I’m going to talk about today is why the “lazy genius” often is an illusion - there are good reasons for this, even if they aren’t as obvious. The “lazy” part can be an illusion and so can the “genius” part.
Why high IQ scores are unreliable proxies
This is the part of the essay where I talk a little bit about statistics and social sciences, and then build a model. Like most models, the one I’ll present is wrong but useful. The first thing to understand is that IQ tests are imperfect measurements of intelligence. For a typical IQ test, there’s a standard deviation of ±15 points. The variation in IQ scores has three sources:
Variation in general intelligence.3 (Model: ±10 points of score).
Variation in test-specific abilities and education.4 (Model: ±10 points of score.)
Sheer luck.5 (Model: ±5 points of score.)
Second, many people who take at least one actual IQ test that they’re aware of take more than one IQ and almost always report their highest score; I have score reports from three IQ tests sitting in a binder along with numerous scores from other IQ-like tests.6 Usually, people only report their highest test score. I’ll assume that after taking an IQ test, someone is 50% likely to take another test, and that there are three different IQ tests on offer:
One that relies on test-specific factor A (25% of tests).
One that relies on test-specific factor B (25% of tests).
One that relies on a composite of the two factors (50% of tests).
Most people who take IQ tests probably take more than one, but a few of those report their most recent score instead of their highest score. This is probably close enough to the underlying (more diverse) reality of between-test variations for government work. For the purposes of this model, each person will have three underlying scores: General intelligence, IQ-test-specific factor A, and IQ-test-specific factor B.7
What does the model think the relationship between reported IQ and general intelligence looks like?
I’ve added a simple geom_smooth() line (i.e, LOESS regression) on this chart to aid in interpretation, because what’s going on is not quite linear. In fact, in the simulation above with 10,000 simulated honestly (if slightly selectively) reported IQ scores, there is nobody who both has an IQ score three standard deviations above the norm and has an underlying general intelligence three standard deviations above the norm.
Instead, the highest simulated IQ scores in the dataset belong to individuals with above average but more modest general intelligence scores, combined with high test-specific intelligence and either good luck or multiple IQ scores to choose between. If we look at the set of people with at least one reported IQ score of 130 or higher (98th percentile), these people average 2.5 tests each (instead of 2 tests) and have an average underlying general intelligence that is only at the 91st percentage.
Similarly, some of the highest levels of intelligence in the simulation have IQ scores that are more modest - in the above simulation, one simulated subject had a general intelligence score of 160, but simply was unlucky enough to score 144 on an IQ test instead. There are real-life examples of this as well; famous physicist Richard Feynman had a recorded IQ score of 125, but was a rare genius. It is highly likely that Feynman would have scored much higher on a different IQ test.8
If you look closely at the LOESS regression, you can see that 130 (two standard deviations) is roughly the point where things start to visibly break down in reliably relating reported IQ scores with underlying general intelligence. In fact, if we look strictly at scores greater than 130 in the model, general intelligence is almost completely flat relative to reported IQ score.
Based on this simulation, if someone reports an IQ score of 141 in the 99.7th percentile, they are most likely somewhere around the 90th percentile when it comes to general intelligence.9 IQ test scores tend to overstate how much of a genius someone is.10
So, what does that mean for our “lazy geniuses” in the red ellipse? Well … they may have high IQ scores, but that only imperfectly indicates high intelligence. They are likely to score a little lower if they retake the same IQ test and are likely to score significantly lower if they take a different IQ test. If general intelligence is the actual driver of differences in outcomes, correlations with IQ scores will tend to be weaker and also tend to break down outside of ±2 standard deviations from the norm.11
The lazy geniuses aren’t always that lazy
Speaking as someone who often described himself as “lazy” when he was younger, the fact that I got middling grades in high school had nothing to do with actual laziness. The typical American adult reads fewer than a thousand books in their lifetime. I was so lazy that by age thirteen, I had read around four thousand books, averaging significantly more than a book a day after graduating to the adult section of the library at age six.
Ultimately, my high school classmates thought I was lazy because I focused on learning instead of on grades. As long as my grades were good enough, I kept ahead on my relevant reading, and I kept acing quizzes and tests, turning in homework seemed mostly pointless. In high school, I frequently didn’t pay much attention in class because the teacher was covering material I already knew.
Most of my peers had no idea what kind of effort I put into learning outside of the classroom. I was sleeping about four and a half hours a night during the school week. I watched very little TV. Nobody in my family owned a video game console. My social life was limited and tended to involve substantial intellectual activities like running D&D games with similarly intellectual friends.12 Most didn’t notice that I had perfect attendance. In college, I described myself as being so lazy that I had to triple major because I couldn’t be bothered with making a decision between math, physics, and philosophy.
By that point, me calling myself “lazy” was a laugh line for my peers most of the time. Graduating in the middle of the class in high school while acing tests made me look like a “lazy genius” who was underperforming academically, but in truth, I was on track for a high level of academic success even while getting merely average grades … and this became more and more obvious as I moved on to more difficult and advanced classes without my grades dropping.
There are misfit geniuses
Now, while I would contend that the number of lazy geniuses who simply don’t apply themselves is a lot lower than it may look at first glance, there are misaligned geniuses who simply do not have the same priorities as their peers and end up having a great deal of trouble with life. As a general rule, these well-attested genius misfits were obsessive rather than lazy, which is why we have heard about them.
Famous examples include Isaac Newton (who probably died a virgin in the 17th century), the Marquis de Condorcet (who died of an unhealthy interest in French politics in the 18th century), Everiste Galois (who died defending the honor of a prostitute in a duel in the 19th century), and Nicola Tesla (who spent the latter half of his life living in hotels and socializing primarily with pigeons in the 20th century).
Where are the lazy geniuses?
As a former self-proclaimed “lazy genius,” I think that lazy geniuses are even rarer than they seem at first glance, because both laziness and genius can be difficult to measure. Measurement errors will tend to exaggerate the population of lazy geniuses.
Twice, but we’ll talk about that more in a little bit.
For high school and undergraduate work. In grad school, it climbed up a bit. This was basically unrelated to course difficulty - my handful of low grades included a few courses that were easy and a few others were classes that were easy for me, like the D I earned in my high school physics class by getting 100% on every test and failing to turn in any homework assignments.
It is politically fashionable to give a nod to Gardner’s multiple intelligences, along with the accompanying body of bad educational research on multiple learning styles, but see here.
Variation in g is estimated to drive about half the variation in IQ scores.
Test-retest validity.
Some research will use SAT scores, GRE scores, ASVAB scores, et cetera as IQ tests. When applied to their target demographic, older SAT tests and ASVAB tests are comparable in quality to some “real” IQ tests in terms of predicting scores on other IQ tests. The IQ-test correlation has generally declined for most standardized tests in the last couple of decades as test-makers have been removing more heavily g-loaded items.
This is not necessarily the most accurate underlying structure, but it produces a between-test correlation coefficient of, respectively, 0.63 for the least-related tests and 0.77 for correlations with the “better” test.
It’s generally suspected that whatever IQ test Feynman took as an adolescent, it probably relied heavily on verbal tasks, because he was far ahead of his peers mathematically. Personally, I suspect that he didn’t take the test very seriously.
They are still probably smarter than someone with a measured IQ of 121 in the 90th percentile of measured IQ. The relationship between measured IQ and general intelligence is monotonic as far as anyone knows; it’s just that it isn’t a perfect correlation.
The simulation I used assumes symmetric error structures. It is plausible (even probable) that IQ tests are much better calibrated for detecting cognitive deficiency than for detecting cognitive gifts. Selective reporting patterns are likely different for low IQs, especially in a context of eligibility for special education services or supporting a legal defense of mental incompetence.
Not that it’s easy to get good samples that cover IQs past 130 with any significance - remember, that’s only 2% of the population. Convenience samples (such as using MENSA members) tend to run into significant selection effects.
If you look at the half dozen peers that I was most closely connected with during my high school years, three went on to get doctorates and two of the others went to highly academically selective colleges. Our conversations tended to be intellectual exercises.