+However, the majority of male-to-female trans people in Western countries do not fit this profile. They are attracted to women or are bisexual and, while reporting a desire to be female dating back to puberty or earlier in childhood, they don't exhibit an _unusual_ number of female-typical traits compared to other males. In contrast to the "early-onset", androphilic type, who couldn't fit in to the world as men if they tried, this second group of "late-onset", non-exclusively-androphilic gender-dysphoric males _can_ function socially as men; we[ref]I think I'm justified in counting myself in this taxon even though I'm [choosing not to transition](/2017/Sep/hormones-day-156-developments-doubts-and-pulling-the-plug-or-putting-the-cis-in-decision/).[/ref] just—aspire to a higher form of existence. The covertness of late-onset gender dysphoria explains why someone like [Caitlyn Jenner](https://en.wikipedia.org/wiki/Caitlyn_Jenner) can have a long, successful public existence as a man—winning men's decathalons, racing sports cars, marrying women and fathering children—before eventually deciding to transition at age 65.
+
+This proposed two-type taxonomy of trans women is very controversial, probably in large part because it's part of a theory that claims that the late-onset type is rooted in an unusual sexual interest termed _autogynephilia_ ("love of oneself as a woman"). Anne Lawrence, herself a self-identified autogynephilic transsexual, iconically describes autogynephiles as ["men who love women and want to become what they love."](http://annelawrence.com/becoming_what_we_love.pdf)
+
+A review of the empirical evidence for the two-type taxonomy is beyond the scope of this post. To interested or skeptical readers who only have time to read one paper, I recommend Lawrence's ["Autogynephilia and the Typology of Male-to-Female Transsexualism: Concepts and Controversies"](http://unremediatedgender.space/papers/lawrence-agp_and_typology.pdf); for a more exhaustive treatment, see the first two chapters of Lawrence's monograph [_Men Trapped in Men's Bodies_](https://surveyanon.files.wordpress.com/2017/07/men-trapped-in-mens-bodies_book.pdf) or follow the links and citations in [Kay Brown's FAQ](https://sillyolme.wordpress.com/faq-on-the-science/).
+
+To avoid the main ideas of this post getting mired in _unnecessary_ controversy, I'd like to emphasize that it's possible to reject the hypothesis that autogynephilia is the _cause_ of the second type, while [still agreeing that](https://thingofthings.wordpress.com/2017/04/18/against-blanchardianism/) there observationally seem to be _at least_ two types of trans women, with the late-onset/non-exclusively-androphilic type or types being much less overtly feminine and not sharing the etiology of the early-onset/androphilic type.[ref]To be clear, I _do_ think autogynephilia has a causal role in late-onset gender dysphoria in males, but justifying that can be left to other posts; arguments can only be strengthened by leaving out [burdensome details](http://lesswrong.com/lw/jk/burdensome_details/).[/ref] Between the statistical signal in the psychology literature (I again defer to [Brown's review](https://sillyolme.wordpress.com/faq-on-the-science/)) and study of the public biographies of trans women (the life-arcs of people like Jenner or [the Wachowski sisters](https://en.wikipedia.org/wiki/The_Wachowskis) _look different_ from those of people like [Janet Mock](https://en.wikipedia.org/wiki/Janet_Mock) or [Laverne Cox](https://en.wikipedia.org/wiki/Laverne_Cox)), I think this is hard to dispute.[ref]But for reference, some of the most popular critiques of the typology (often—I claim erroneously—cited as [debunkings](http://slatestarcodex.com/2014/12/13/debunked-and-well-refuted/)) are [Serano 2010](http://www.juliaserano.com/av/Serano-CaseAgainstAutogynephilia.pdf) and [Moser 2010](http://unremediatedgender.space/papers/moser-blanchards_autogynephilia_theory_a_critique.pdf).[/ref]
+
+I _am_, however, supposing that the late-onset type or types is either not an intersex condition, or at _most_, a very mild one: we could perhaps imagine a gender identity "switch" in the brain that can get flipped around (explaining the eventual need to transition) without much affecting other sexually-dimorphic parts of the brain (explaining how transition could be delayed so long, and come as such a surprise to others). This hypothesis is weaker than the autogynephilia theory, but still has implications for the ways in which transgender identity claims might or might not be validated by natural, prediction-motivated categorization schemes. If most trans women's traits are noticeably _not drawn from from the female distribution_, that's a factor making it less practical to insist that others categorize them as women.
+
+To this it might be objected that there are many different types of women. Clusters can internally have many subclusters: Pureto Rican women (or married women, or young women, or lesbians, _&c_.) don't have the _same_ distribution of traits as women as a whole, and yet are still women. Why should "trans" be different from any other adjective one might use to specify a subcategory of women?
+
+What makes this difficult is that—_conditional_ on the two-types hypothesis and specifically gender dysphoria in non-exclusively-androphilic biological males being mostly not an intersex condition—most trans women aren't just not part of the female cluster in configuration space; they're specifically part of _male_ cluster along most dimensions, which people _already_ have a concept for. This doesn't mean that we can't get away with classifying them as women—there's nothing _stopping_ us from drawing the category boundary however we want. But it [isn't an arbitrary choice](/2018/Feb/blegg-mode/)—the concepts of _women_-as-defined-by-biological-sex, _women_-as-defined-by-self-identity, and _women_-as-defined-by-passing are picking out different (though of course mostly overlapping) regions of the configuration space, which has inescapable logical [consequences](http://lesswrong.com/lw/nx/categorizing_has_consequences/) on the kinds of inferences that can be made using each concept.
+
+<a href="/images/genderspace_cluster_choice.png"><img src="/images/genderspace_cluster_choice.png" width="532" height="421" alt="genderspace cluster choice"></a>
+
+<aside class="boxout"><strong>Figure.</strong> A schematic visualization of genderspace using fictitious but hopefully illustrative data <a href="/ancillary/categories-scatterplot-source/">(scatterplot source code)</a>. Suppose that the distributions of cis men (represented by the <span style="color: #1E90FF;">light blue</span> datapoints) and cis women (the <span style="color: #FF1493;">hot pink</span> points) have the same variance, but their means differ by 3.5 standard deviations in each of the <em>x₁</em>, <em>x₂</em>, and <em>x₃</em> variables, and that the distribution of non-exclusively-androphilic trans women (the <span style="color: #B000B0;">purple</span> points) is the same as that of cis men for the <em>x₁</em> and <em>x₂</em> variables, but resembles that of cis women for <em>x₃</em>. People who care more about predicting <em>x₁</em> and <em>x₂</em> have reason to prefer categories and corresponding language that group by natal sex (<span style="color: #0000C8;">blue</span> category boundary); people who care more about predicting <em>x₃</em> have reason to prefer categories and language that group by gender identity (<span style="color: #E00000;">red</span> boundary).</aside>
+
+In less tolerant places and decades, where MtF transsexuals were very rare and had to try very hard to pass as (cis) women out of dire necessity, their impact on the social order and how people think about gender was minimal—there were just too few trans people to make much of a difference. This is why experienced crossdressers often report it being easier to pass in rural or suburban areas rather than cities with a larger LGBT presence—not as a matter of tolerant social attitudes, but as a matter of _base rates_: it's harder to get [clocked](https://www.urbandictionary.com/define.php?term=clocked&defid=4884301) by people who aren't aware that being trans is even a thing.[ref]In [predictive processing](http://slatestarcodex.com/2017/09/05/book-review-surfing-uncertainty/) terms: the prediction errors caused by observations of a trans woman failing to match the observer's generative model of women get silenced for lack of alternative hypotheses if "She's trans" isn't in the observer's hypothesis space.[/ref]