“Better than chance” doesn’t really mean much. The question is how well, quantitatively. Just briefly scanning over your studies:
The first one seems kinda sketchy because it has a smallish sample size and then it splits it up by sex, which leads to a truly small sample size within each sex. The difference between men and women seems unrealistically big to me, so I wanna say that to get less biased results, we should probably average the answers, yielding a correlation of about 0.24 for appearance alone. Though this is probably an underestimate due to range restriction. Also they didn’t report the raw correlation coefficients so I did some quick kind of sketchy calculations and they might be wrong.
Second study is kind of hard to interpret because they give correlation with sound, without sound, and with only a transcript. Range of correlations go from 0.37 (for video with sound) to 0.04 (for transcript only), but that’s with about 80 participants, so the standard errors are fairly large.
I don’t have time to parse the entirety of the third study because I need to go to bed soon, but it looks to me like the correlations there were about 0.2?
Note that it might be tempting to think that you can just average effects over a longer time to get more accurate results, but that is not always true; because of correlated measurement error, you might get stuck at some accuracy. For instance you could imagine that people would use academic interests as a proxy for intelligence; in such a case, they cannot exceed the prediction accuracy determined by the correlation between academic interests and intelligence.
“Better than chance” doesn’t really mean much. The question is how well, quantitatively. Just briefly scanning over your studies:
The first one seems kinda sketchy because it has a smallish sample size and then it splits it up by sex, which leads to a truly small sample size within each sex. The difference between men and women seems unrealistically big to me, so I wanna say that to get less biased results, we should probably average the answers, yielding a correlation of about 0.24 for appearance alone. Though this is probably an underestimate due to range restriction. Also they didn’t report the raw correlation coefficients so I did some quick kind of sketchy calculations and they might be wrong.
Second study is kind of hard to interpret because they give correlation with sound, without sound, and with only a transcript. Range of correlations go from 0.37 (for video with sound) to 0.04 (for transcript only), but that’s with about 80 participants, so the standard errors are fairly large.
I don’t have time to parse the entirety of the third study because I need to go to bed soon, but it looks to me like the correlations there were about 0.2?
Note that it might be tempting to think that you can just average effects over a longer time to get more accurate results, but that is not always true; because of correlated measurement error, you might get stuck at some accuracy. For instance you could imagine that people would use academic interests as a proxy for intelligence; in such a case, they cannot exceed the prediction accuracy determined by the correlation between academic interests and intelligence.