We gathered >1 million human correlations spanning personality, demographics, behaviors, and beliefs, which enabled us to investigate an interesting question: is everything correlated with everything? In other words, is the so-called “crud factor” real? Here’s our answer.
On average, there is almost no correlation between human variables (see image). The average is only r=0.02, with the median even smaller at r=0.01. But this is misleading because it allows positive correlations to be canceled out by negative ones.
So let’s instead look at the distribution of the absolute values of the correlations. The average *magnitude* is r=0.1, with a median of 0.07. This shows how small a correlation of 0.1 is – it’s about the magnitude you’d get correlating two human variables picked at random!
We see similar results if we look only at positive correlations. The mean and median are still r=0.1 and r=0.07. So it seems that there *is* a “crud factor” – random human things really are correlated (on average)! But not very much, just slightly.
Yet, while there’s some correlation on average, it’s also worth noting that the most *common* correlation is right around 0! So, the correlation mean (r=0.1), median (r=0.07), and mode (r≈0) are all a bit different.
Finally, note that different data sets could find different results depending on what they contain. But, to my knowledge, this is the most diverse and largest test of its kind by far.
But wouldn’t we measure a non-zero correlation even if there were no true correlations, just due to random chance?
The answer is yes, but if there were no true correlations, the average absolute value of correlation would be a lot smaller. I ran a simulation where I generated fake correlations (where the true correlation was zero, any measured correlation was just due to noise arising from a limited sample size). I did this using the real sample sizes from our actual database of correlations.
The result is that we still do measure an absolute value of correlation above zero, as expected, but it’s very small. The median absolute value of correlation is just 0.02 and the mean is r=0.03. We get pretty much the same numbers if we take the median and mean of just the positive correlations as well (instead of the absolute values). So even though a small amount of the crud factor likely is attributable to random noise, we can see that it is substantially bigger than what we’d expect if it was all just due to random noise.
You can explore our 1 million correlations for free with our Personality Map tool: https://personalitymap.io
To learn more about the “crud factor” in psychology, check out this interesting paper:
Crud (Re)Defined.
Comments