While I appreciate you sharing your thoughts… it’s uncontroversial to say that certain physical traits are more attractive, and that some of those have an evolutionary basis (associated with health/fertility/dominance; alternate sociological hypothesis have been tested; found to be cross-cultural).
What I do find controversial is dismissing studies on the basis of ‘it must be flawed’ without checking if this is true. I find this mind boggling when I’ve provided citations.
On the other hand, most people don’t have the skills to evaluate studies. I’ve worked as statistician/data analyst so I can hopefully clarify somewhat (especially around how machine learning works):
- If you want to look at the effect of one variable, you only alter this variable e.g. photos of the same subjects digitally altered to vary clothing colour. Their methodology looks good to me: “Each experiment used a single male target and a between-subjects color manipulation (with random assignment) so that participants would not see repeated color presentations that could alert them to the purpose of the experiment.”
- The trick is having enough participants to be able to detect a difference in group mean if there is one — this is determined by statistical power (not by merely looking at the sample size and assuming it’s too small as a layperson). If there’s a statistically significant difference in attractiveness rating and the effect size is large (i.e. the groups’ attractiveness ratings actually differed substantially), that points toward red clothing enhancing people’s perception of your attractiveness.
- Finally, this finding needs to be replicated in other studies (including cross-cultural ones) while testing alternative theories (e.g. ‘red is associated with passion in only this particular culture, and is rooted in sociological rather than physiological factors’) — which has been done for the well-documented phenomenon of red increasing attractiveness.
- If you want to add heaps of confounding variables do an ecological study in a nightclub setting.
- If you want to predict physical attractiveness (i.e. the subject of the post, and not what determines success of approach in a nightclub — much harder), build machine learning models that look at a constellation* of traits. Feature selection means you only include variables that explain a lot of variance in the model, by the way, thousands of columns are not ideal: https://en.wikipedia.org/wiki/Curse_of_dimensionality.
- You can test the accuracy of the model on fresh data (training, validation, and test data sets). In this way it’s easy to determine if your model is not accounting for some important factor, as its predictive ability will be poor.
- *It will always be a combination thereof, that’s how machine learning works regardless of specific algorithm.
I hope this clarifies it, there are dodgy studies out there, but I’m convinced there’s enough evidence for red-clothing enhancing attractiveness to the opposite sex :)