Discussion about this post

User's avatar
Emiel de Jonge's avatar

Yeah, unfortunately, once again I completely agree. I would even state that r = 0.2 is not something to be proud of either. I work with quite a bit of statistical results daily and an r of 0.2 does not even get attention. Yeah, I work with well validated and reproducible techniques and methods in medical science. Nonetheless, relationships with a r of 0.2 are so unreliable that it cannot be used to differentiate on variable from another. I work with identifying proteins on cells and the sample sizes are in the thousands to millions. But still all of our dot-plots are with bivariate relationships and sometimes even multivariate ones. And an r of 0.2 is not a big move away from randomness, and the mean squared error would be quite high, which means that predicting variable A with B would not pan out well as the error would generally be very high. Also the r^2 or coefficient of determination would be 0.04 in that case, also not something to be proud of. I'd say that psychology generally has a very low bar for what counts as reliable relationships in statistics. I would say that an r of 0.4 and above is better. Maybe this would not be possible in many cases, but then it would be reasonable that psychologists would just be honest and state that the relationship is very weak and should not be trusted until validated with stricter and better methods.

Expand full comment
2 more comments...

No posts