It seems like a popular response for this kind of this research is “Well of course - I expected this result” or to point out the murky causality going on when an intelligent technology is trained on the behavior of people using it.
I agree this murky causality is important to discuss and that indeed these results are consistent with other observed distributions of online content (i.e. Pareto/Zipf), but I think excessively focusing on these responses undermines the importance of audit research like this. Audit research is mainly descriptive by design (can’t really conduct an experiment unless you’re Google) and without the descriptive results from audit research we can’t argue empirically about the these topics at all!
I don't think this dismissal is quite fair, especially regarding this work being unscientific or having negative value to humanity!
The paper described here isn't survey-based; it actually seems very quantitative compared to other psych papers (they collected audio for "150,000 recordings" and computationally analyzed the data; the survey part sounds more like diary data collection). But even if this was entirely survey-based, it wouldn't necessarily be useless.
I do think that it would be better to cover trends in the literature instead of individual studies, but in this case it seems the paper has important implications for sentiment analysis and computational social science in general. I currently hope conferences don't go away!