Thread by @davidmanheim

I said last week that Brandon Tuckfield's (@MormonThinker) article on @Quillette about attraction inequality in online dating makes unjustified conclusions based on bad data and worse analysis. Here's the story - quillette.com/2019/03/12/att…

@MormonThinker @Quillette First, a disclaimer; Life isn't fair. I'm not arguing about whether some people get disproportionately more attention in online dating than others, or if that's good or bad. I'm just explaining that the article is bad statistics, bad journalist, and bad social science.

@MormonThinker @Quillette Let's start with the Gini coefficient, which is being used to compare what is supposedly "inequality of sexual attractiveness" to income inequality. (Except, as I'll explain, that's not even what it's being used for.) But first, what does a Gini coefficient mean?

@MormonThinker @Quillette The most critical issue is that we need to define we're measuring the inequality of, and what it means. Financial inequality is often measured with a Gini coefficient. If everyone is middle class and makes $50,000 except one guy who makes $500,000, how unequal are things?

@MormonThinker @Quillette Gini is a measure of inequality in a frequency distribution. If everyone has the same frequency, the Gini index is 0. If one person alone gets everything, Gini is 1-(1/n). That's kind of what we see in the previous example, with $50k vs. $500k, right?

@MormonThinker @Quillette Turns out, not at all. The difference between everyone but the rich family making barely enough to live on, and everyone having nothing, is huge. Gini looks at percentage of total income. If we have 10 poor people plus a guy making 10x as much, he get half, and Geni is 40.9%.

@MormonThinker @Quillette But Gini depends on how many people there are. Instead of 40.9% if we have 11 people, it's 25.4% with 26 people.

For the analyses in the article, this matters. But it isn't the biggest problem with the article - or even the biggest problem with using Gini this way.

@MormonThinker @Quillette For example, let's give everyone in either of our examples a 100% raise. Gini is unchanged, but $100k a year seems a lot less "unfair" compared to $1m than $50k does compared to $500k. This is my problem with using Gini in general.

For online dating specifically, it's worse.

@MormonThinker @Quillette What do we measure about "inequality" in online dating? We noticed that $1 matters differently for wealth depending on total income - but at least it's comparable between people. The article thought it was comparing attractiveness, though.

@MormonThinker @Quillette Attractiveness isn't universal, and different people find different things attractive in different ways or different amounts.

Let's ignore this for a bit to point out a different problem. Pretend everyone's "hotness" can be ranked 0-5, and that's all that matters to anyone.

@MormonThinker @Quillette We'll pretend Tinder has 25 male users, and 25 female users, and they are all similarly moderately attractive - there are 5 "4s", and 20 "3s" for both the men and women. What's the Gini index of this distribution? 5.0% - It's a veritable paradise of attractiveness equality.

@MormonThinker @Quillette Now, we have them use Tinder. We assumed everyone agreed about rankings, so all the men "like" the 5 most attractive women, and the women do the same for the men. What's the inequality of likes? (Which is what the article actually measured.) In the example, the Gini index is 80%!

@MormonThinker @Quillette So likes are set up in a way that exaggerates the inequalities, as the analyses show. But it's not an exaggeration, it's real inequality.

What's the cause? It's a structural problem with @Tinder and similar apps, not a fundamental attribute of human sexual attractiveness.

@MormonThinker @Quillette @Tinder I'm not going to spend more time discussing the structural problems induced by online dating apps, but it should be clear that it might be a big deal, and it has basically nothing to do with the people dating, and everything to do with the design of Tinder and similar apps.

@MormonThinker @Quillette @Tinder Another analysis cited in the article look at actual attractiveness, namely a 2009 @OKCupid analysis: web.archive.org/web/2009112108…

This survey asked people about how attractive the opposite gender is, using integers on a scale of 0-5, with an option for 2.5 for average.

@MormonThinker @Quillette @Tinder @okcupid This kind-of agrees with the story that women are pickier than men - only 8-9% of men are rated above average, i.e. 3, 4, or 5 - but 39% of women are.

How unfair is this? I used the data to compute Geni indexes for a 200 simulated users - it's 27% for women, 42% for men.

@MormonThinker @Quillette @Tinder @okcupid This is unequal, obviously, but it's lower than the analyzed inequality on @hinge in the article, of 32.4% and 54.2% for women and men, respectively. The article compared the values for women to "average" Western Europe versus "kleptocracy, apartheid" in South Africa for men.

@MormonThinker @Quillette @Tinder @okcupid @hinge Our attractiveness calculation moves the country comparisons from Western Europe vs. South Africa to Finland vs. Israel - imperfect, but doing pretty well.

But these are computed values for likes, not attractiveness, and are far lower than our example of "4"s and "3"s. Why?

@MormonThinker @Quillette @Tinder @okcupid @hinge Remember that we assumed that everyone ranks people the same way - that's an unfair and unreasonable assumption that basically created the massive imaginary inequality in likes that I suggested earlier. That should undermine my claims greatly!

Well, yes, except...

@MormonThinker @Quillette @Tinder @okcupid @hinge ...it's exactly what one of the analyses that the article mentions does! This is what led to the claim that 80% of men compete for the bottom 22% of women.

The article missed that point, but the cited analysis clearly admits that "this is the biggest flaw in this analysis."

@MormonThinker @Quillette @Tinder @okcupid @hinge And if people view attractiveness differently, it makes the field much more even.

But the data is discussing Tinder likes, not attractiveness. Maybe despite personal differences in attractiveness, likes are unevenly given? Maybe women only "like" the 8-9% of above average men?

@MormonThinker @Quillette @Tinder @okcupid @hinge Actually, data shows that the opposite is true. Women are happy to message less attractive men. Going back to the OKCupid article, only 19% of women's likes go to those 8-9% of men.

What about men? Men are clearly shallower, and send 17% of likes to the 6% of women who are 5s.

@MormonThinker @Quillette @Tinder @okcupid @hinge But there is something more fundamentally wrong with all of these analyses - they count how often people initiate conversations on a dating app, which is a single mode of interaction, and one that favors those with the best profile pics.

@MormonThinker @Quillette The article's conclusion lines up with the narrative of nice guys and incels, perhaps more than it matches reality. Assumption after assumption, all biased in the same direction, leads to a conclusion that "We may pity the large majority of men who are regarded as unattractive."

@MormonThinker @Quillette If we want, we could reframe this to look at who is getting rejected, how often, and if it matters.

Maybe it does. Or maybe, as @JordanPeterson might say, these men "need to grow the hell up" instead of being bitter about getting rejected on Tinder.

(But I don't care.)

@MormonThinker @Quillette @JordanPeterson I might enjoy going to my favorite hobby-horse - ribbonfarm.com/2016/09/29/sof… - and note that when you pick your metric on the basis of what's easy to measure, you get bad answers. Then I'd talk about how we look at income inequality and ignore inequality of wealth and opportunity.

@MormonThinker @Quillette @JordanPeterson We need to choose what we measure based on what we want to know.

If not likes, what should we measure? If we care about inequality in access to sex, as the article thinks it was measuring, number of partners makes more sense - but they didn't look for that data.

@MormonThinker @Quillette @JordanPeterson But measuring numbers of partners doesn't capture key points. I only have one partner. (I'm happily married.) Many people have a dozen partners per year. That looks unequal. Am I deprived? No. Are Incels? Maybe.

(But again, I don't care - I'm just asking about methodology. )

@MormonThinker @Quillette @JordanPeterson I'm going to wear my social science methodologist hat for now, and point out that useful conclusions require clear definitions of terms and conceptual work on what we mean before we start looking for data.

This is completely missing in both the article and from the analyses.

@MormonThinker @Quillette @JordanPeterson All of the analyses were based on crappy social science, misapplied statistics, and a lack of definitions. The reporters are reporting on "analyses" they don't understand written as blog posts by computer scientists who seem to know nothing about how to do social science.

@MormonThinker @Quillette @JordanPeterson As a matter of methodology, talk about anything, but find relevant metrics FIRST. When you pick your metric on the basis of what supports your narrative, you choose your conclusion - it's not analysis, it's mostly rationalizing beliefs and providing post-hoc justification.

@MormonThinker @Quillette @JordanPeterson The blame can go anywhere - I don't care. Bad science education, lack of rigor in science, or journalists who are paid to write about things they don't understand.

Blame whatever you'd like, but stop promoting analysis like this that poisons the epistemological commons.

https://twitter.com/davidmanheim/status/902523739258179585