Minority Drift

In my frequent debates on transgender acceptance across the web (don’t worry, this isn’t actually an article on trans issues per se), I get asked a lot about why the suicide rate for trans people is so high (and, in particular, why it’s elevated post-transition). For the most part, I tend to avoid this question in a remotely political debate because it tends to be used as a derailing tactic to avoid the fact that transitioning greatly improves social, medical, and psychiatric outcomes for trans people, but it is kind of troubling on the surface.

The natural answer, and the one most people give, is that discrimination is so bad that it makes people suicidal. There is some truth to this – parental support is a very strong predictor of suicidality and rates have declined over time – but this is a theory that proves too much. It would suggest, for example, that you’d expect to see higher suicide rates among (say) blacks than among whites, but in fact the exact opposite is true: blacks, Hispanics, and Asians commit suicide at less than a third the rate whites do (and at almost exactly the same rate as one another), and minority suicide rates have been remarkably stable for at least 15 years while rates for whites steadily rose. This being the internet, I am sure someone will come along and claim that everyone just hates whites that much, but I think we can safely dismiss that as a long shot.

A more compelling claim is that parental support is more important than societal support, but that doesn’t really explain the elevation for trans people either because the rates remain elevated even when looking at those accepted by their families (and it would be odd if the economic differences between, say, Hispanics and Asians didn’t produce some differential in parental support). It’s also interesting, given that discrimination apparently doesn’t cause direct problems, that suicide rates have dropped in lockstep with social acceptance over the past 30 years, and that gay men had a similar elevation and show a similar lockstep decline. So what is different about LGBT people, as opposed to, say, racial minorities that can explain why gay men and trans people move in similar patterns?

My hypothesis: selection bias is a hell of a thing.

To demonstrate my point, I need to make a couple of (I think reasonable) assumptions.

One, some gay men/trans people are more likely to seek male partners/transition than others. You have people who leave successful careers, or knowingly leave their families, to seek treatment, and you have people for whom hormone therapy is too much of a drive away to bother. This includes both personality type (willingness to contravene social convention) and strength of desire (exclusively gay man who can’t find any fulfilling sexuality with a woman vs. flexible bisexual man who can make do)

Two, people perform, at some level, a sort of cost/benefit analysis when facing potential societal backlash. In other words, they consider whether the potential gains of their unorthodox decisions are going to be worth the level of hatred/punishment/loss of opportunity/whatever they expect to get from society. Of course these estimates can be way off, but at the end of the day most people have some idea what they’re doing in their local social environment.

Three, the level of backlash against LGBT people has declined over time, in a way clearly visible to everyone involved. This one is so obvious and easily demonstrated that it barely qualifies as an assumption at all, but it’s here for completeness.

And four, that suppressing one’s deep personal desires is bad for one’s mental health. Given that the psychiatric community generally views quote-unquote ‘conversion therapy’ as a very bad idea, this seems reasonable – if one could suppress one’s desires at no cost, perhaps that would be appropriate therapy. (Though it’s probably worth noting that there are occasional opponents who don’t like this assumption – Kenneth Zucker comes to mind.)

So. Rewind to the 1960s. Discrimination against both black and LGBT people is extremely high, to the point that almost anyone who can avoid it will do so. However, your typical black man can’t avoid that discrimination, because he cannot reasonably hide his skin color (which immediately marks him as a member of a particular social class at the time). Your typical gay man, however, can do that: he can hide his feelings entirely, or share them only with extremely close confidants.

In general, to be an openly gay man in 1960 requires either (a) an extraordinary drive to be openly gay, one so strong as to overcome the very high costs or (b) an extremely high tolerance for contravening social norms. People in group (a) suffer years of suppression, and simply by virtue of being extremely unusual on one axis are likely to be extremely unusual on another (say, the presence of mental illness), both of which mean that people in group (a) are likely to be disproportionately unhealthy. Similarly, those in group (b), who are actively violating major social norms, are likely to be pariahs for many reasons other than being gay.

Other gay men, whose desires are more ‘manageable’ or who are less willing to violate social norms, stay in the closet, out of reach of researchers (and potentially even of themselves: unthinkability bias is powerful in its own right). So a survey of openly/self-identified gay men at this time finds only the extremely non-random group from the previous paragraph and, lo and behold!, the study ostensibly finds that gay men have poorer mental health than the general public.

Fast-forward 40 years, however, and the story is very different. Being gay may not be entirely accepted, of course, but it’s well past the point of police raids on gay bars being a thing most people are okay with. Gay men who are more, for lack of a better word, ‘normal’ find their cost/benefit analysis tipping the other way. Bisexual men who might, in past times, have suppressed attraction to men to settle down with a woman are now able to explore themselves further. In short, you no longer have to be crazy (in a sense) to be openly gay, and as a result surveys of gay people no longer find them to be crazy.

Now, let’s talk about evaporation.

As a kid, I thought the boiling point of a liquid was when it went from liquid to gas. But that obviously isn’t the case: water on a countertop will evaporate in a matter of hours despite being nowhere near boiling. So what gives? The answer, as it turns out, comes from statistics.

The temperature of a fluid is, roughly speaking, an average of the kinetic energies of the particles in the fluid. This kinetic energy is exchanged constantly between the particles as they bump into and slide around one another: in general, a faster particle imparts energy to a slower one, causing them to roughly average out their speeds, but the random nature of this process prevents things from averaging out entirely.

As a result, not all particles are average: some move faster than the average speed, and some move slower. (In an idealized gas, the distribution of the speeds of the gas particles can be computed exactly; for a liquid, things are a bit harder – but the exact distribution isn’t important to the example). So at lower temperatures, a small number of particles (namely, those at the very high end of the distribution) have enough energy to break their bonds with the rest of the liquid even though the average particle does not.

The particles that have enough energy to break free do so. Any particle below that threshold does not. This means that high-energy particles leave the liquid preferentially, and so the average energy of the particles – the temperature of the water – declines. And so we get evaporative cooling as a result of a weird sort of non-random selection on the molecules in the water.

But this should sound a bit familiar, shouldn’t it? We have two examples where we select for only the most extreme members of a distribution, and it should be no surprise to us that this creates a paradoxical result. In the same way that boosting particles to high energy manages to cool our hypothetical liquid down, the population of openly gay men could, at one point, be unhealthier than the population of not-openly-gay men even though being out in and of itself is assumed to be a positive.

My hypothesis, a bit more clearly stated, is this: a marginalized group in which visible membership is chosen (as opposed to being based on some commonly-visible inherent trait like skin color) and which is highly stigmatized by the general public will – under these assumptions and no others – have elevated levels of most negative traits relative to the public. When only ‘crazy’ or ‘weird’ (in some sense) people are willing to be members of a group, the group will necessarily be composed of ‘crazy/weird’ people, even if nothing about the group’s category implies craziness or weirdness in and of itself. In more sophisticated terms, such groups self-organize into a Simpson's paradox.

This is the first good explanation I’ve come up with to explain why comparatively mild anti-trans discrimination seems to produce worse results than comparatively intense past anti-black discrimination. A ‘normal’ black man cannot ‘opt out’ of being black, and so this statistical bias never rears its head within the black community. But a ‘normal’ trans person can opt out, and many do. I almost did myself, and spent a lot of time computing the utilities of each choice during lulls at work with equations that included terms for “number of years left in my life” and “probability of social collapse”.

This concept can generalize to a number of other known phenomena:

· Urban decay: once your neighborhood is in a bad enough state, anyone who can leave does leave. This leaves you with a group disproportionately unable to leave, with elevated rates of poverty, criminal histories, disabilities, mental illness, etc. This legitimately makes the neighborhood worse, which further contributes to the decay in a feedback loop. Even if the neighborhood was only believed to be bad at first, it will soon become actually bad.

· Fashion (in clothing or otherwise): once a trend is unpopular enough, anyone who has the knowledge and resources will avoid it. This means that the trend becomes primarily – and rightly! - associated with people who don’t have the knowledge and resources, which accelerates its unpopularity further.

· Education: many people work in jobs totally unrelated to their degree, but having a degree is still required. Why? Because everyone knows that you’re ‘supposed’ to have a degree, so most people who can get one do get one, which makes the non-degreed population disproportionately stupid/low-income/low-motivation/etc, which makes a degree more valued, and so on.

· Corruption: once an institution is known to be corrupt, honest people tend to avoid it, either for fear of looking corrupt themselves or because they find dealing with corruption frustrating. This means the corrupt group now samples from a disproportionately corrupt population, further increasing its level of corruption.

· Third parties in American politics. Once a party is known for having fringe crazies, it can’t make any progress, so non-crazy people tend to avoid it. This further concentrates and empowers the crazies within the party, creating a cycle that results in Jill Stein.

These trends are legitimately hard to fight. Once the cycle is established, the group in question actually is worse in some sense than the general public, so any decision that depends on whatever better/worse axis is under consideration should disfavor that group on pure self-interest grounds. A hiring manager, even if they are absolutely certain that a college degree provides no benefit whatsoever to the position, is still incentivized to favor hiring people with college degrees!

But it gets still worse, because this hypothetical hiring manager probably isn’t absolutely certain that a college degree provides a benefit. How could he tell? Since we can’t just admit random people into colleges, most studies are going to be observational: look at people who graduate, look at people who don’t, control for things that you can easily control for, and compare the results. But it’s highly unlikely that a study can control for every possible general trait (say, personality or family support) that favors both academic success and professional success, and as a result, such a study will show that having a degree does produce benefits – even when it doesn’t. So not only is our manager incentivized to ignore good non-degreed candidates, he probably even thinks that he’s right to do so!

None of this requires any sort of intentional abuse. In fact, it’s entirely possible for this process to take place with zero awareness on the part of anyone involved.

If people are aware of it, it can be further exacerbated by intentional signaling, where “has the resources to signal” becomes one of the favorable traits (and having resources is beneficial in nearly any endeavor). This can lead to absurd scenarios where a parent is paying tons of money for a kid to prepare for a test that provides no actual benefit to them to go to college for a degree that will never provide any benefit and a hiring manager knows very well that all of this is going on and is still incentivized to hire them.

I also wonder if this is perhaps why educational methodology seems to matter so little in comparative studies. It seems a bit counterintuitive that education itself provides a huge boost to life outcomes, but method of education seems to do nothing – unless the value of an education is simply to drive out anyone who won’t put up with the hassle, in which case everything makes perfect sense. The education may not be providing any real benefit – or may even be harmful – but that lack of benefit is compensated by a powerful sampling bias to select people who (for reasons totally unrelated to education) are more likely to be ‘useful’ in some sense later.

When I started writing this particular essay, I was only trying to answer the question, posed at the beginning, of why LGBT people had elevated suicide rates but racial minorities didn’t. In the process of writing, however, I seem to have stumbled onto a larger principle, and I feel like I should at least conclude with a few thoughts on what to do about it.

I’d like to say something like “now that we think this thing is happening and much of the rest of the world doesn’t, we should see if we can achieve some sort of competitive advantage by it” – but as we saw with the hiring manager, this isn’t the case. The incentives all run in the wrong direction, so trying to fight the issue locally just gets you eaten by Moloch.

I’d also like to say that we should be careful of fields in which we can only do observational study, but that doesn’t really work either, because we can’t just do random selection for, say, medical treatments or education. Or at least, we can’t do so without living in a way more totalitarian society than we do right now.

In fact, I worry that by writing this at all I’ve generated a Fully General Counterargument with which to deny any negative traits associated with a marginalized group. One could, for example, say that the reason (say) homeopathy enthusiasts seem to be a generally crazy bunch is that only crazy people are willing to say they’re homeopaths, and that there are in fact many well-reasoned homeopaths hiding all around us. In fact, this theory would say that that might even be true – perhaps the reputation homeopaths get isn’t entirely deserved, and they are more reasonable than they look. But of course that doesn’t mean that homeopathy is a good idea, and yet someone somewhere would read this article and go “oh, that’s why everyone thinks I’m wrong!”

For the moment, all I can suggest is to (a) remember that statistics is really, really hard to do properly and (b) keep this pattern in mind as a potential matching-scheme for the future. As usual, the problems are rather easier to find than the solutions.