| This makes a great example of conflating correlation and causation. (I don’t mean this as a criticism of the original post. Rather, I think it’s an opportunity to discuss something that’s fascinating and counterintuitive and that even researchers often get wrong.) The original post says a researcher (Liam Hudson) was surprised to discover that average students seemed to be more creative than high-IQ students. Thus the question arises: Does having a high IQ make you less creative? The problem, however, is that the question is causal whereas the evidence that prompts it is observational and tricky for humans to interpret causally. To see why it’s tricky, let’s imagine a universe in which IQ and creativity are completely unrelated. Nevertheless, if both factors contribute to success (and it’s easy to believe that they do), they will have a negative correlation, conditional upon success. And guess what? Almost all creativity/IQ studies are implicitly conditioned upon success because researchers don’t want to compare “A” students to “C” students, for fear of biasing their results. For example, let’s model our imaginary universe in R. First, let’s assign normalized IQ and creativity scores to 100 students, completely randomly and completely independently: n <- 100
creativity <- rnorm(n)
iq <- rnorm(n)
And now let’s say that success is some increasing (causal) function of creativity and IQ, plus chance: success <- creativity + iq + rnorm(n)
And now let’s pretend that some researchers have captured this very data. We know that IQ and creativity are completely unrelated (because we created the universe), but they don’t. Here’s what happens when they try to tease out the relationship between IQ and creativity, conditioned upon success: Call:
lm(formula = creativity ~ iq + success)
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.08848 0.08042 -1.100 0.274
iq -0.58801 0.09394 -6.259 1.05e-08 ***
success 0.57538 0.05431 10.594 < 2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Note that the IQ coefficient is negative. That is, given knowledge of a student’s success, if we also learn that the student has a higher IQ score, we should update our state of knowledge to decrease our belief that the student has a higher creativity score.I know that that claim may sound counterintuitive, but if you do the probability math, you’ll see that it makes sense. It only seems counterintuitive. That’s because your brain can’t help but try to interpret the relationship causally. That is, it’s all too easy to take this relationship that exists between states of knowledge about the universe and think that it actually exists within universe for real. (This mistake is what E. T. Jaynes called the “mind projection fallacy” in his book Probability Theory: The Logic of Science.) And lots of researchers make this mistake, too. I don’t know for sure if that’s what’s happening here, in the IQ vs. creativity research, but it wouldn’t surprise me to find that it has at least contributed to some of the unexpected findings. |