@josephbenton that means r2 is 0.0144. the best case scenario for this line is barely over 1% of variance lmao
@josephbenton gOOd fucking lOrd this is terrible
@josephbenton That, uh.
That looks like sample size bias to me.
Either that or an uncovered pollock original
@paladin @josephbenton I mean you can literally see that there's a huge clump of twenty somethings and basically no one above 50 in the data
@RussellsBarbershopQuartet @josephbenton exactly. Plus there doesn't seem to be any indication of the various ages of the viewers, which would probably also be a factor, cu ne?
@paladin @RussellsBarbershopQuartet
Here's the full thread if you want to see the full horror: https://twitter.com/baumard_nicolas/status/1308715609795039232
@josephbenton @paladin every time I see someone cite Ekman who clearly hasn't read Ekman I die a bit inside
@josephbenton @paladin @RussellsBarbershopQuartet are they running this analysis on oblivion NPCs, or
@josephbenton @paladin @RussellsBarbershopQuartet omfg it's even worse in context
@josephbenton @paladin @RussellsBarbershopQuartet Wow. Maybe I can find a way to use this twitter thread when I teach machine learning, to illustrate everything *not* to do.
@josephbenton @paladin @RussellsBarbershopQuartet this is truly horrible
@josephbenton @paladin @RussellsBarbershopQuartet "Our results thus show that trustworthiness in portraits increased over the period 1500–2000 paralleling the decline of interpersonal violence and the rise of democratic values observed in Western Europe" how can you just state that seriously, the heck is trustworthiness and oh my god this is bad. This is truly bad
@josephbenton @paladin @RussellsBarbershopQuartet what punches me in the face is the lack of reflexiveness on how their metric is defined (how about non white people, or disabled people), their spurious correlation and how they absolutely do not care about the consequences of their work on a surveillance capitalism setting where facial recognition is the big hyped stuff
@paladin @josephbenton right, but if you do that, then you'd have to multivariate math, and no one wants to see that bc if you're getting fucking r2 = 0.0144 with this many participants, you'd need so so many more to get the appropriate statistical power for what is, again, demonstrably barely even a drop in the bucket
@RussellsBarbershopQuartet @josephbenton you want multivariate? let's throw perceived race and gender presentation of both viewer and viewed, perceived economic status of both, i mean
trying to draw such a conclusion from such facile, noisy data is abject foolishness.
@paladin @RussellsBarbershopQuartet
Nevermind stats, I see the question "define trustworthiness" and I have a nightmare about the millennia worth of philosophical and social science texts one would have to read just to attempt an answer.
@josephbenton @paladin it's a failure on every level
@josephbenton @RussellsBarbershopQuartet Aka why ethics and cultural studies are so goddamn important to doing STEM of any kind
@josephbenton that average line is doing so much heavy lifting
@josephbenton this is basically a red line over a scatter plot
@josephbenton I needed this laugh today, thank you
@josephbenton lmao
@josephbenton ewe
@josephbenton I like the error indicators on the regression because they don't seem to have any correlation to the data or the regression
@josephbenton
relevant xkcd
@josephbenton those correlation coefficients sure are doing a lot of work to try to make their case.
@josephbenton Huh, negative r
@josephbenton as we all know, 99/100 people die before reaching the age of 60,
@josephbenton conclusion: mist faces are 28 years old
@josephbenton Almost got fooled by the trendline apparently arbitrarily drawn through a meaningless field of data points.
@josephbenton r=.12 lol