Autogynephilia is a sexual interest in being a woman. According to the theory of erotic target location errors, autogynephilia is connected to gynephilia (attraction to women) in some sense, being a sort of inversion of it. The idea being that in most cases, gynephiles have something that prevent them from ending up autogynephilic, but in some cases they don’t.
In the weaker forms endorsed by Julia Serano, attraction to women is merely one contributing factor that isn’t strictly necessary to end up autogynephilic. I consider this speculation to be quite likely; e.g. it aligns with my findings that self-reported autogynephilia is 2-3x rarer and weaker in gay men than straight men. In the stronger forms endorsed by Kay Brown, Rod Fleming, and Michael Bailey, the theory asserts that attraction to women is in some sense1 a necessary factor for becoming autogynephilic. I believe that this theory is unsupported by evidence, and in fact even contradicted by some evidence that I have collected.
But this blog post isn’t about the distinction between these two theories. Rather, it’s about the fact that most of the studies cited for either of them are, from a Blanchardian perspective, totally trash and of no evidentiary value at all. At least for the ETLE question; there may be other questions that they can tell us something about.
To give an example, let’s have Kay Brown present the evidence:
So, back to autogynephilia. This is an observed phenomena. NOT a theory. I’ve already explained how one hypothesis has met several tests… that there are two types of transsexuals who have been shown statistically to have certain common traits with-in each type and two correlated and mutually exclusive traits. As a reminder, those traits that correlate are gynephilia and autogynephilia. The traits that are anti-correlated are exclusive androphilia and autogynephilia.
This correlation leads to proposing another hypothesis, namely, that for autogynephilia to exist, there must be an underlying gynephilia. For someone who is androphilic, there is no existing gynephilia to lead to autogynephilia. The data would support this hypothesis extremely well. In fact, as I have shown in my earlier post, analyzing Leavitt and Berger’s study, that correlation is very, very high… perhaps showing an effect size that is higher than many experimental results in psychology that are accepted and not considered in any way controversial. Just to remind ourselves just how strong the effect size is, let’s revisit the data, looking at only the issue of reported autogynephilia and a history of sex with females:
AGP: 6.7% 33.3% 50%
Sex w/f: 0% 33.3% 58%Phenomena and Theory… …or how to confuse fact and fiction, Kay Brown, On The Science Of Changing Sex
So Kay Brown’s argument is that autogynephilia requires gynephilia because in trans women, autogynephilia is highly correlated with attraction to women.
Blanchard advanced a similar argument:
The first hypothesis suggested by Hirschfeld’s observation – that autogynephilia is a misdirected type of heterosexual impulse – predicts that one should find higher levels of autogynephilia in heterosexual – or at
least nonhomosexual – men than in comparable homosexual men. This prediction has been supported by the results of a study by Blanchard. The subjects in this study were 212 adult male-to-female transsexuals. These were divided into four groups: one homosexual (attracted to other males) and three nonhomosexual (attracted to females, to both sexes, or to neither sex).
The measure of autogynephilia used in this study was called the Core Autogynephilia Scale, or CAS for short. Most of the items in this multiple-choice questionnaire measure ask whether the respondent has ever become sexually aroused while picturing himself with various features of the female anatomy (e.g., breasts). Therefore the CAS is primarily a measure of anatomic autogynephilia.
The four transsexual groups were compared on the CAS (and on several other psychosexual variables that are not immediately relevant). As predicted, all three categories of nonhomosexual males were more likely to report sexual arousal in association with fantasies of womanhood than the homosexual males. This finding supports the view that autogynephilia is, as Ellis put it, “really a modification of normal hetero-sexuality.”Clinical observations and systematic studies of autogynephilia, Ray Blanchard
So that’s a big part of the standard argument from the Erotic Target Location Error side: You see a strong correlation between autogynephilia and gynephilia in trans women, which would seem to support ETLE theory. And indeed this seems to be quite a consistent finding of studies in trans women. So how could I possibly call this of no evidentiary value? The issue is an elementary statistical error:
To understand the problem, it’s time for a lesson on my favorite statistical paradox! Berkson’s paradox! Stealing the explanation from this twitter thread, let’s consider some important questions: Why are handsome men jerks? Why don’t standardized test scores predict university performance great? Why are movies based on good books usually bad? Why are smart students less athletic? Why do taller NBA players not perform better at basketball?
If we filter our sample on the basis of some set of variables, then that filtering introduces a ton of spurious correlations between all of the variables that are upstream of our filtering. The usual pattern will be negative correlations between the causes of the thing you are filtering on, but we might have other things going on, depending on the specific details of the distribution.
In the case of autogynephilia and sexual orientation, the studies considered so far all filter on transsexuality. Thus, regardless of the true correlations in the general population between the variables, these studies will tend to find spurious negative correlations between the causes of transsexuality. And what are some causes of transsexuality? Autogynephilia and exclusive androphilia!2 So regardless of what the actual true correlation is between autogynephilia and exclusive androphilia, by studying it in trans women, we would expect to observe a spurious negative correlation. This makes it of no evidentiary value to find the correlation in trans women.
It’s true that there’s a robust negative correlation between autogynephilia and exclusive androphilia in trans women, but that is a spurious result. This eliminates most of our evidence about negative relationships between autogynephilia and exclusive androphilia. It is for this reason that, when I reviewed the evidence on whether autogynephiles can be exclusively androphilic, I mostly ignored the studies in trans women. Instead, to figure out the association, one has to do one of four things:
- Get a large unfiltered sample of males, and look at the correlation between autogynephilia and exclusive androphilia directly.
- Compare an unfiltered sample of males to a sample of autogynephilic males, and see if they differ on their rates of exclusive androphilia.
- Compare an unfiltered sample of males to a sample of exclusively androphilic males, and see if they differ on their rates of autogynephilia.
- Mathematically adjust for the effect of filtering, e.g. adjusting for the fact that autogynephilia and exclusive androphilia are both causes of transsexuality in the original studies.
Option 4 seems tricky; it requires specific knowledge of small nuances in how autogynephilia and exclusive androphilia contribute to transsexuality. Options 2 seems finicky; if filtering for autogynephilia goes wrong, then it might also accidentally filter for or against exclusive androphilia, which would mess things up. And similarly for option 3.
Option 1 at first seems more promising, but it faces the problem that both autogynephilia and exclusive androphilia are rare, and so maybe it requires a sample size on the order of 5000 or more. Plus with effects that subtle, even tiny measurement errors can cause big problems.
My solution resembles option 1 the most, but with a caveat. I usually do my studies on reddit, and reddit is highly autogynephilic and also somewhat more gay than average. This possibly exposes me to the same sorts of Berkson’s paradox that I talked about before – but not necessarily. I think there’s a good case to be made that I’m not facing it, but I’ll banish that point to a footnote.3 Regardless, the high rates of autogynephilia on reddit makes me need much less sample size to investigate it, and with it I have found some evidence of a negative connection, but also some evidence of the existence of exclusively androphilic autogynephiles.
Illustrations in a simulation
In case the whole argument was a bit unclear, let me give some hypothetical numbers to illustrate how filtering for transsexuality could lead to a spurious connection.
Imagine that autogynephilia is uncorrelated with androphilia, that 20% of males are autogynephilic and 20% of males are androphilic, and further that autogynephilia and androphilia each lead to a 50% chance of transitioning. Obviously these numbers are way too high, but I pick them so that we can have a reasonably small simulated sample size.
Suppose we have 100 males that are representative for these numbers. I will use lowercase letters to represent cisgender men, and uppercase letters to represent trans women. I will use the letter “x” to represent non-autogynephilic heterosexuality, “a” to represent autogynephilia, “h” to represent exclusive androphilia, and “b” to represent both autogynephilia and exclusive androphilia (i.e. “GayGP”). The sample might look as follows:
| non-androphilic exclusively androphilic -----------+----------------------------------------------- | non-AGP | xxxxxxxxxxxxxxxx | xxxxxxxxxxxxxxxx hhhhhhhh | xxxxxxxxxxxxxxxx HHHHHHHH | xxxxxxxxxxxxxxxx | | AGP | aaaaaaaa b | AAAAAAAA BBB
In the general population, we see no correlation between autogynephilia and exclusive androphilia. For instance, among non-androphiles, we have 16/(16+64)=20% who are autogynephilic. Meanwhile, among androphiles, we have 4/(4+16)=20% who are autogynephilic. And vice versa.
On the other hand, if you restrict yourself to trans women, you get the sample:
| non-androphilic exclusively androphilic -----------+----------------------------------------------- | non-AGP | (none) HHHHHHHH | AGP | AAAAAAAA BBB
Here, among non-androphilic trans women, we have 8/8=100% who are autogynephilic. Meanwhile, among androphiles, we have 3/(3+8)=27% who are autogynephilic.
This is a huge correlation, comparable to that in the classic studies on trans women! And the correlation gets even stronger as you reduce the prevalence of androphilia and autogynephilia, to more realistic levels. There’s a lot of additional nuances here, where the results can shift around a lot depending on how you assume the factors lead to transsexuality. But I think the most important lesson to draw is that you should be very careful in what you conclude about autogynephilia from studies of trans women – you might get surprised. In particular, such studies are of no evidentiary value for the purpose of ETLE theory.
1. This runs into a slight nuance due to a phenomenon related to asexuality in AGPs. Specifically, it is believed that high degrees of autogynephilia overshadow and outcompete gynephilia, which can make the ordinary gynephilia disappear, even if it really did cause autogynephilia. I am skeptical of this theory, but it is a bit separate for this blog post – maybe it will come up in some later blog.
2. There’s a fun element here. Usually ETLE is considered to be a Blanchardian theory, in opposition to gender identity theories. However, while it is invalid for Blanchardians to cite these sorts of studies in support of ETLE, it is perfectly valid for gender identity theorists – because they don’t believe autogynephilia causes transsexuality! And indeed ETLE provides an alternative explanation to Blanchardianism for a number of phenomena related to Blanchard’s typology.
3. To understand why it might work fine for me, let’s consider the other name of Berkson’s paradox, collider bias. It is named so due to what it looks like when you draw the conditions for it up in a graph. Consider for instance the studies of autogynephilia and sexual orientation in trans women. We might draw the following graph to illustrate the causes of transsexuality:
autogynephilic → trans ← homosexual
I.e. autogynephilia and homosexuality both cause transsexuality. This pattern, where one thing has multiple causes, is called a collider. Berkson’s paradox only applies to colliders. So the question then is whether my surveys face a similar collider, e.g.
autogynephilic → redditor ← homosexual
This is conceivable, but I don’t expect this to be the pattern. While reddit could plausibly select for homosexuality since it is well-known for being progressive, it is not a site for autogynephiles and there’s not much reason to expect explicit selection for it, especially because the effect is so incredibly big that it is hard to believe.
Instead, it seems that reddit is much more paraphilic than the general population. This raises the question of whether there maybe is a collider of the shape:
autogynephilic ← general factor of paraphilia → redditor ← homosexual
If there is, it doesn’t seem to have much effect; when investigating it, homosexuality seemed uncorrelated with the general factor of paraphilia. On the other hand, maybe it is positively correlated in the general population, and the collider bias just eliminated it on reddit. But this would still mean that we at least wouldn’t expect a spuriously negative correlation between autogynephilia and paraphilia.
Of course it’s always hard to say these sorts of things for sure. It’s not the strongest case. Getting unfiltered data is impossible, and figuring out what ways it has been filtered is nearly impossible. Maybe we should just pack up and give up on social science.