3 ms·

It’s great that you actually looked at the paper and you inspired me to do the same. But could you please explain why you think the correlations seem “really re

It’s great that you actually looked at the paper and you inspired me to do the same. But could you please explain why you think the correlations seem “really really small”? To me they actually seem surprisingly high. Some examples: 0.25 for safetyism, -0.47 for resilience, 0.45 for loneliness. Especially the latter two are almost shockingly high given what we see elsewhere in psychology. You can’t expect one multi-faceted construct like social cognitive distortions to fully or almost fully explain some other complex and multifaceted construct like safetyism. If the correlation was super high, that would just mean that we’re likely using two different names for the same thing. Plus it’s not just a single correlation that supports the hypothesis in this paper but the whole pattern of correlations is consistent with it as far as I can tell as a non-expert. I haven’t read the paper closely, and the conclusions may very well be entirely wrong, but I wouldn’t conclude that based on the correlations.

contravariant 5y ago
Asuming they're reporting the pearson's correlation coefficients and not the slope of some linear fit (which they seem to use elsewhere) then a correlation coefficient with an absolute value <0.5 only explains <25% of the variation. Now that isn't actually too bad for a psychological test, so the fact that resilience and loneliness have are related to cognitive distortions is reasonable (and not too dubious a claim). The main problem I have with it is that their main claim "Cognitive distortions predict safetyism" is based on a correlation coefficient of 0.25, this leaves roughly 90% of the variance unexplained which really stretches the meaning of the word "predicts" in their title. Also I don't have enough trust in their questionnaires to be sure that all of the variance in their results is actually directly caused by the thing they claim to be measuring so when you find some correlations it's hard to rule out it isn't caused by some confounding variable. All in all someone's Cognitive distortion score is a pretty bad predictor of their adherence to safetyism belief. There's a nonzero amount of correlation between the results of their questionnaires but given the small effect size I'm dubious about their claim of a causal link between them. I should at this point note that "percentage of variance unexplained" is a bit of a simplification and only really works for linear fits, but I'm using it here as a rule of thumb to reason about the strength of the correlation.

contravariant 5y ago

Asuming they're reporting the pearson's correlation coefficients and not the slope of some linear fit (which they seem to use elsewhere) then a correlation coefficient with an absolute value <0.5 only explains <25% of the variation. Now that isn't actually too bad for a psychological test, so the fact that resilience and loneliness have are related to cognitive distortions is reasonable (and not too dubious a claim). The main problem I have with it is that their main claim "Cognitive distortions predict safetyism" is based on a correlation coefficient of 0.25, this leaves roughly 90% of the variance unexplained which really stretches the meaning of the word "predicts" in their title. Also I don't have enough trust in their questionnaires to be sure that all of the variance in their results is actually directly caused by the thing they claim to be measuring so when you find some correlations it's hard to rule out it isn't caused by some confounding variable. All in all someone's Cognitive distortion score is a pretty bad predictor of their adherence to safetyism belief. There's a nonzero amount of correlation between the results of their questionnaires but given the small effect size I'm dubious about their claim of a causal link between them. I should at this point note that "percentage of variance unexplained" is a bit of a simplification and only really works for linear fits, but I'm using it here as a rule of thumb to reason about the strength of the correlation.