To take a step back, people identifying moving from nonbinary / trans identities to cis identities isn't necessarily a concern. In the trans community, there's a tongue-in-cheek meme called "cis+", but the central premise is that people shouldn't feel bad for questioning their gender and coming out of that with a stronger understanding of their identity (likened to unlocking a new level of cis). You don't have to be trans to experiment with gender, and it's possible many "cis+" people bring elements from their exploratory phase into their new (final?) identity.
Anecdotally, all the "cis+" people I know have done something like this, separating a typically-gendered element from the concept of gender (e.g. men wearing makeup, women wearing binders).
Strange to see breast-binding spoken of as analogous to face-painting! Makeup washes off at the end of the day and doesn't restrict activity (swimming aside, I suppose). Meanwhile it's tough to do so much as run for the bus when your ribcage is constricted.
Which leaves me curious, as gender is clearly important to you, despite it being a socially constructed set of labels and descriptions. Meanwhile, it feels entirely meaningless to me, as 100% of my "identity" is entirely subjective to me and almost inexplicable....so that said, what percentage of how you relate to yourself do you think is partly through the eyes of others and what percentage is instead wholly subjective and internal?
I did go through a whole questioning phase and wound up being more conservative if anything. This is probably one of these things that vary hugely from person to person.
In my opinion, comparing age stratification to change over time doesn’t really work because they’re asking different questions. Age stratification shows how different age groups compare at a single time—for example, whether 16-year-olds differ from 30-year-olds right now. Change over time looks at how the overall population shifts across multiple time points. An age-based difference at one point doesn’t tell you how attitudes or behaviors are trending over time, and a trend over time doesn’t prove that one age group is driving it. So the two analyses can’t really be used to confirm or refute each other.
Also, the small uptick you mentioned in one age group hasn’t been tested for statistical significance, and there’s no margin of error reported. Without that, it’s impossible to know whether the difference is meaningful or just sampling noise.
That said, I also have doubts about the representativeness of Erik Kaufmann’s data. A lot of the attention it received seems driven more by political confirmation bias than real informative value.
Agreed. If nonbinary identification is decreasing compared to previous highs, you would want to show that by polling people in eg 2022 vs now. With your current data it's entirely possible that you have a lot of people who responded that they identify one way and have since changed their minds.
Yea i am def doing a single snapshot vs. the other paper's tracking over time. I'm operating on the assumption that the effects I'm seeing in the snapshot are going to carry over as people age; if this turns out to not be the case I'll publish another post saying 'oops'.
And If you're talking about the uptick at 19, it's almost certainly not sampling noise; the sample is like i think 150k males for that line? Another indicator of high certainty is how smooth the line is. I didn't include CI also cause I don't think there's a reasonable computation for that when I'm doing weighted avgs. I usually do try to include shaded CI ranges when I'm working with just the raw data.
I wonder if the minorities in your daya-set are more likely to know who you are though.
As in, if 90% of your data-set say they don't know who you are; would the fraction of for example trans people in your data-set who know who you are be *higher* than 10%?
In response to this selection comment -- just to say some obvious things:
(1) Selection is inevitable in survey data, even in the most well-resourced economic policymaking surveys like CPS or SCF, because one can't force people to answer one's questions. If you are concerned about selection because of aesthetic horror, avert your eyes from survey data.
(2) If your objection is practical rather than aesthetic, the correct response to selection is to take survey data seriously when the questions asked of the data are along dimensions where selection is unlikely to vary.
(3) With respect to this post, and plots of % non-cis vs. age: You can tell a very plausible selection story that shifts the whole curve up or down (as the author clearly notes, and ballparks the likely magnitude of the shift). You could tell a maybe-plausible story where selection scales monotonically with age. But I do not think you could tell any reasonable selection story to generate the shape of the curves in Figure 5 (the one with n=541,782).
(4) so, tl;dr: selection is not much of a concern in this analysis. (Not because "selection doesn't exist" but because the author asked questions of the data in such a way that selection is unlikely to confound the answers.)
I think that subscribers to Aella are just a very very narrow subset of people; the sheer largeness of the numbers just doesn’t matter very much.
To quote the article: "Selection bias is disastrous if you’re trying to do something like a poll or census” though "Selection bias is fine-ish if you’re trying to do something like test a correlation.” This obviously depends on degree but I really think this is a very very selected population, right? I mean, this feels obvious to me, but please tell me what I’m missing.
You’re right; I should rephrase. The types of people that would have seen/responded to these surveys are likely a very particular subset. Social media platforms don’t just show your content to random people. Do note that she is using this data not sparingly but to refute a different study, which doesn’t have this feature. This seems very weird to me.
The 'particular subset' is mostly 'people who like taking personality surveys', which tends to be young liberal western women. I have a friend who runs an old personality testing website, and the demographics of those who answer his surveys line up almost exactly with mine.
You can look at what groups are more likely to answer your data and get a pretty good sense at exactly how and in which ways selection bias is impacting your results. You'll notice in the above survey I weight for politics and anxiety, two of the biggest things that skew my results differently from representative samples.
Additionally, the original survey I'm responding to is just "people in college", which you could argue is perhaps an even more selected sample than mine, which isn't limited to a specific socioeconomic strata.
You are completely correct and I strongly agree with you. This data is extremely skewed towards those who would ever take online surveys and who would even bother to respond to questions of this sort, as well as being skewed towards those who follow Aella (even if those who don't follow her also respond).
To me all this data is showing me is people who are more online are more likely to identify as trans. Nothing more or less.
The point of the numbers is that her followers are only a small portion of the people responding. Depending on the platform she has 100-200k followers, and only some portion of those took the survey.
I do survey research for a living. There are many sources of error in survey research (coverage, measurement, etc). It is very hard to write survey questions that measure what you think they are measuring. Even if you succeed in writing a 'good' question, that question is still embedded within a larger instrument, so the 'same' question surrounded by other, varying questions, can produce different results in different surveys. We usually put the demographic section (age, sex/gender, education, income) last to avoid breaking rapport. Americans will willingly tell you all about their kinky sex practices, but asking them about income is seen as just too invasive. Right now, I am doing a large, repeated cross-sectional survey for a 'federal agency' (I don't want to get too specific about the client because I am worried about the black helicopters showing up over my house if I appear too woke). When we started a few years ago, during the approval process, we proposed using a simple, one response only, sex/gender item - it was more or less "What are you: male, female, transgender, do not identify as male, female, transgender"? The client objected to this and said we must use the then-current Census approved version, which split it into two questions: (1) "What is your gender identity (multiple response allowed): male, female, nonbinary, questioning, prefer not to answer, other"? (2) What is your gender assigned at birth: male, female, X, unsure, prefer not to answer, not listed"? Note that the word 'sex' does not appear in either version - only 'gender'. When Trump was elected, we were first required to change back to a single item: "What is your sex: male, female, prefer not to answer". Apparently there were some nervous discussions within our client group, and, before we ever fielded this new version, we were again told to drop the 'prefer not to answer' option. Note that this question (web survey) was never mandatory, so a respondent was always free to skip it. Hence the 'prefer not to answer' option was really unnecessary, However, the client apparently did not want to even suggest to respondents that there was any fuzziness about this issue. And, the word 'gender' disappeared to be replaced with 'sex'. The point of all this is that Aella may have difficulty drawing conclusions from these surveys because of the large potential for measurement error. Another point: whatever we are writing in these comments is probably being scraped off the web by Palantir, etc., and incorporated into our secret 'social credit' scores. So, watch out for the black helicopters and slaughterbot drones.
I do pay quite a lot of attention to survey wordings, and for years have experimented with wording (70+ surveys, plus countless twitter polls I use to run beta tests on how wording impacts reults). I've found it matters less than most people think, except in some specific situations where it matters a lot. I have used gender vs sex before. My current gender question is a final evolution of a lot of honing and people yelling at me lol.
Have you written a post on how wording influences results? That would be very interesting.
Relatedly, does your survey put questions about sexual attraction near the gender identity questions, and are there variations in the answers if you do or don't?
I did not mean to denigrate your work in any way - it's truly amazing and better than many PhD dissertations I have read. My comment was more about the current frantic MAGA zeitgeist than gender/sex issues.
I haven't tested order too much I really. The identity and sexual questions are relatively far apart in the survey. You can Google Big Kink Survey to see!
I do keep meaning to write a post about survey wording yeah
Just to be clear, since you naturally didn’t hit the point too hard, this is overwhelmingly strong evidence for the social contagion theorists and for there being a big, big problem with how teenage girls are experiencing themselves in our society. There’s just no way that anywhere within screaming distance of 15% of AFAB people are really men and this has never been noticed in history before. For comparison to the best-known modern society with a more stabilized trans identity option, it’s something like 0.5% of AMAB people are trans according to the gender structures available there in Thailand.
Thai culture is not very accepting of trans people. It's a well known location for transitioning surgery for some interesting historical reasons, but the culture and laws are not accepting. You can't legally change your gender identify for example.
As near as I can tell from surveys, trans people are somewhere between 2-5% of the total population. Widespread social stigma likely suppresses that to some extent, but not enough to justify a 14% difference.
I think this is almost certainly an artifact of who is answering Aella's post, given that I can find no other sources saying anything close to even 10% of AFAB people are identifying as trans men.
Every single other survey including the ones Aella cites have significantly lower rates. All clustering around the 2-5% rate.
When one poll has an extremely dramatic outlier to all other polls, I don't think that's evidence of social contagion. I think that's evidence that something weird is going on in this one poll.
You seem to be ignoring the entire point of this data which is the slope of the curve. Even if Aella’s sample is biased you have to explain why those in her biased sample are getting far, far less cis at younger ages, and to assume that has no relationship with the whole population you’d have to make some really bold epicycles. Note that her rates for 30-35 year olds are quite similar to population data! (I also have no idea where you’re getting 5% numbers for the total population, which would certainly imply stats something like Aella’s for teenagers, again since the rates are aggressively slopes downward with age. The first few data I find say 1-2%, with some 3/4 of these under 35. There are very, very few trans Xers, Boomers, and Silents.)
And it is exactly the difference between the youth rates and older rates that makes me suspect the actual population is closer to 2-5% range. There was extremely strong, state sponsored violence and repression of transgender individuals, combined with far less general knowledge of or support of transitioning.
Similar to the "spike" in people suddenly becoming left handed, it is not shocking that youth rates are substantially higher than older folks and there are very strong reasons to believe that the actual rates are closer to the youth rates or even slightly more than the current youth rates as there are still strong levels of repression in both legal and cultural ways.
You're doing some kind of complicated metaphysical claim, here, about how many people are "really" trans even if they would never identify that way. There is no such number. Modern trans identity is a very new construct; gender-nonconforming people before about the 1960's had completely different ways of conceptualizing their nonconformity, older conceptualizations which have only been slowly replaced during the lifetime of older people today, and it's an act of epistemological violence to imagine that you can say "2-5% of them were really trans, they just weren't allowed to come out" and have the sentence even be meaningful, let alone true. I see no meaningful way to measure how many people are trans other than how many people will identify, anonymously, as being trans. Of course reducing discrimination will tend to lead to larger numbers of such identifications, but again, it's an act of epistemological violence to say that those people were really trans, just secretly (from themselves?) rather than that we now have a different set of social constructs into which people are comfortable classifying their experience.
You are accusing me of making metaphysical claims while throwing around claims of me doing epistemological violence?
There is different societal ways of expressing gender non conformity - but there is also accounts that there were people identifying as other genders even in societies with extreme prejudice and no general concept of transitioning.
This isn't coming from nowhere or general navel gazing. This isn't something that is dreamed up. There are biologically identifiable differences in the brains of trans people that match the cisgender people they are transitioning to.
Let me know how the r^2 on predicting trans identification via brain scanning is doing. I'll wait.
(Actually, that's not really the point: of course trans people are different than cis people in ways that will show up on a brain scan, even though I would bet 100:1 odds that you can't reliably predict trans identification from brain scan alone. But the brain scan doesn't prove that that difference is somehow metaphysically correctly interpreted as 'being the gender opposite your birth sex', as opposed to the any number of ways other societies have interpreted it. These categories, again, are socially constructed and constantly in flux.)
The 2-5% rate is the surveyed rate for trans youth from the various surveys I have seen. There is a relatively wide range, but they all cluster around that range, which is enormously lower than this sample.
I don't know the explanation for why Aella's survey differs so dramatically. It's pretty standard when doing a meta analysis to discard significant outliers though. When you have ten studies that say one thing, and another study that says something wildly different, that's not a reason to say the 10 studies must clearly be wrong or missing something.
It's not bold to suggest that this one outlier survey is not representative or has some other flaw that is making it substantially different than every other survey.
Well, OK, but what is "youth", and why did you say "total population" when you meant "youth"? Pretty sloppy!
Anyway, Aella herself adjusted her estimate for college-age people to 5.5%, which is perfectly in line with what you're saying. You haven't actually shown that Aella's results are even an outlier in any detail, let alone the really significant detail on the massive non-cis identification in young high school-age AFABs, and you're still itching to simply throw out the one thing that she's in a better position to measure than almost anyone, and which would a priori be independent of any sampling bias, which is the slope of the curve of trans identification down to young ages. That might fly if you were simply trying to publish a meta-analysis, but I think we're interested in actually figuring out the truth here.
I looked at the CDC survey Aella cited, and interestingly, it does *not* seem to find any substantial slope in trans identification among high schoolers; it shows a rate of 7-10% (depending on how you count "questioning") among AFAB high schoolers pretty steadily from freshmen to seniors. So, that does at least suggest that Aella might somehow simply be wrong, though intuitively it seems impossible that freshmen aren't more trans-identifying than seniors, given we know that high schoolers are much more trans-identifying than college students than general 20-somethings. But in any case, does anyone other than these two studies have a sample size of tens of thousands of high schoolers answering this question? I kind of doubt it, in which case I hardly think we could call Aella an outlier.
As for the slope of the curve, it seems most likely to be that the survey might have become viral and shared in a more widespread way among a subgroup of the population.
I referenced the same CDC dataset that you reference a study based on [a decade-old version of], which in the recent version Aella relies on in the post, involved 20000 subjects. What is your point, here, exactly?
Yes, of course, it is possible that the survey just happened to become viral specifically in highly gender-nonconforming segments of specifically just 14-16 year olds. But you have no basis on which to throw it out just because some extremely particular kind of measurement error is theoretically conceivable.
I didn't say it "just because it's theoretically possible."
It's because you have one survey which has an extremely divergent result from every other survey, from a variety of sources. When you have one survey that has an extremely different result from literally dozens of other surveys that all cluster around the same range, it's fair to consider reasons why that one survey might not be as reliable. This isn't about malice or not trusting. I like Aella and find her surveys fascinating.
It's well known that weird outliers can happen in individual surveys or studies. That's exactly why it's important to do so many and repeat them. What it's important not to do is take odd outliers as stronger evidence than multiple other surveys. See XKCD 882 for a silly example of why slicing data sets to fine or overlying on one individual survey can lead you to faulty conclusions.
I'm not going to say more than three times that you've completely failed to produce more than one survey comparable to Aella's. I do not think you actually have dozens (your own citations bewail the shortage of data on these questions in high schoolers), but if you do, feel free to clue me in. I am perfectly well aware of all the concerns you note: you should weight one (extremely large) study exactly as much as it deserves, no more *or* less. In this case, all I have in front of me is Aella's data vs the CDC's, and I would weight the CDC perhaps twice as heavily. You are, for some reason, saying I should weight it roughly infinitely more heavily by throwing Aella out as an outlier. You haven't provided any reasons at all to do that.
> Thai culture is not very accepting of trans people. It's a well known location for transitioning surgery for some interesting historical reasons, but the culture and laws are not accepting. You can't legally change your gender identify for example.
--
This seems incorrect. In international surveys such as the yearly Ipsos Pride report, Thailand seems to consistently give, by a wild margin, the most trans-positive responses to a variety of trans-related cultural questions.
See: https://www.ipsos.com/sites/default/files/ct/news/documents/2025-06/ipsos-pride-report-2025.pdf - Thai responses rank first, generally *by far* on all questions, for support for teenage transition, access to single-sex facilities corresponding to gender trans people identify with (vs birth sex), support for the addition of a third gender option on government documents, support for insurance coverage of transition, support for trans athletes competing in sex categories corresponding to gender they identify with (vs birth sex), and support for anti-discrimination laws for trans people.
I wonder how much of that represents recent changes. I might be going off old data. That being said, that same survey says 68% of Thai people believe trans people face a fair amount of discrimination, and laws do actually matter.
We might wonder if it’s like the rise in autism rates. In the case of autism, the leading theory is that the actual rate of autism has stayed about the same, and the apparent rise is due to including a bunch of borderline cases that previously wouldn’t have been counted.
It might be the same here: a constant rate of transgender, but the statistics have been inflated by including a whole load of non-binary or vaguely trans cases that wouldn’t previously have been counted.
In theory, the gender clinics are supposed to diagnose whether you really need hormones etc.
But
- In UK, waiting lists are insanely long, so patient never gets seen by a specialist to work out of they’re really trans or not
- From reading the stories of a lot of trans people, I get the impression that the specialists have totally no clue whether you’re actually trans or not, so even if you get to the end of the waiting list, it doesn’t help
The UK has, officially, banned puberty blockers for minors with gender dysphoria (still allowed for precocious puberty, cancer treatment, etc)
From trans people’s accounts, it would appear that it has also got a lot harder for adults to get hormones, though there is no official ban.
The problem with banning treatment for everybody, rather than making an effort to work out which of these patients is actually trans, is that the adults at least will just get hormones on the black market.
Autism "rise" is not due to "borderline cases", it's due to the fact it was very poorly defined before (of course, because it was defined by literal nazis), and the science has gotten better at defining what it is.
That's it.
Spoiler : they did that by... listening to autists. Sensory hyper/hyporeactivity was added to the DSM only in 2013, which is absolutely wild considering how constitutive of the autistic experience it is.
Also, thinking it's for a specialist to say of you're trans or not is incredibly condescending, transphobic, and frankly disgusting.
Unless you think someone else than you can decide what's happening your mind, which is a whole other level of dystopia 🤷♂️
> we don’t yet know how to separate these people from those who would be harmed
Your operative word being "yet". I don't think there's a "yet" here. I suspect that there's simply no way to do that. Teen brains are much too … fluid, in-the-process-of-rewiring-themselves, impressionable, take your pick … for that to work.
Heck even the percentage of adults who end up regretting transitioning are way larger than statistical error.
If you're interested in sex and psychological profiles in detransition, which you allude to, check out this post (spoiler, we found a greater portion of afabs detransitioning due to changes in identity several years post-transition):
Sorry if I am being stupid here, but are you reading the time trend by looking at the age profile? As in: more young people than old people are trans -> upward trend in transgender identity?
Doesn’t that conflate age effects with time\cohort effects? Is the assumption that gender identity is “set” at age 14?
The main conclusion I would draw from this is that a 14 year old is so precociously interest in sex and gender that they are both likely to find and complete a survey on it, is disproportionately likely to identify as some form of non-binary
Yes, yes, selection bias always exists, but the these biases in this case are especially self reinforcing, and probably explains the steepness of the slope
Nevertheless, I share Aellas concern that there has been an emergent exposure that is increasing in intensity and/or prevalence, that is inducing “transness” in young people, and particularly in young AFABs, who given the same underlying predisposition to transness, would not have developed it if they had existed in a different environment
How is 14 year old precocious for interest in sex ? At this age, most teenagers are interested in sex, and as for all new things they're more likely to research it. Especially when they have all the world knowledge in their pockets.
The world is a big place, and people are interested in travel. But there is a certain kind of person that ends up at an Auyuhasca retreat in the mountains of Peru.
"I haven’t checked to see if there’s other surveys that disagree, so in theory he could have been cherry picking, but I doubt it"
Why ? Anyone supporting the anti-science party need to be triple checked for bias. I mean, in the vast majority of cases, they're not following scientific principles, so on the contrary I'd assume he cherry picked.
"I think it’s unlikely that 11.5% of afabs are actually trans men in a way that would last through adulthood."
Why ? It does seem high to me too, but then again, 5% would have seemed high 10 years ago. There's no particular reason to think this or that percentage is too high, it's kind of arbitrary.
There’s something I noticed about this when I first read it, and I see other people have noticed it too.
There is a difference between measuring how the same cohort changes over time, vs different cohorts at the same point in time.
What the data here seems to be showing is people born later are more likely to identify as trans at the same point in time as those born earlier.
But the question is really, have the same people become less likely (or more) likely to identify as trans due to political changes over time?
It is a basic of survey methods, that is explained in almost every undergraduate psychology text book, that people sometimes lie on surveys, particularly if you ask them about something that society views negatively. So a decrease in trans identification on surveys might be more people refusing to tell you they’re trans. But here, we’re just not measuring that.
The other things that’s interesting here is that the moral panic in the media is almost entirely about trans women (women’s sports, women only spaces etc) but the data suggest that the increase is mostly AFAB.
It's actually pretty easy to detect abnormal responses in this specific dataset. One reason why is that it's a 30-40 min survey to complete, and you can look at how long it takes people to complete it. Maybe in theory someone is running bots very slowly to replicate human response times, but they would have to do a whole lot of that to displace my current organic sample at 850k and I'm not sure why they would be motivated to do so.
And there's further ways of detecting false response sets even *if* they get past that check.
Thanks! I posted a link to me and my team's detransition study above, and below is a link to our sampling methodology paper in which we developed a protocol to identify and remove bots and real human fraudsters (who we ended up interviewing as part of the screening process). It was a heavy lift! But a step my team took took because of how controversial and contested the topic is. You'll see in this paper the challenges occurred even in advertising and recruiting for the study on social media. I think you'll find this paper interesting (I am a queer trans man/transmasc person BTW, and am interested in this topic because several of my own friends I transitioned with 15 years ago have since detransitioned, and it seems to disproportionately affect queer AFAB folks). Sampling paper and bot/fraud removal protocol: https://www.jmir.org/2025/1/e63252
Oh yeah I would be way more afraid of bots and malicious responses if I were doing a trans focused survey. In fact the one time I have detected en mass bad data in my surveys was specifically for a survey about trans people, where it was advertised as such!
To take a step back, people identifying moving from nonbinary / trans identities to cis identities isn't necessarily a concern. In the trans community, there's a tongue-in-cheek meme called "cis+", but the central premise is that people shouldn't feel bad for questioning their gender and coming out of that with a stronger understanding of their identity (likened to unlocking a new level of cis). You don't have to be trans to experiment with gender, and it's possible many "cis+" people bring elements from their exploratory phase into their new (final?) identity.
Anecdotally, all the "cis+" people I know have done something like this, separating a typically-gendered element from the concept of gender (e.g. men wearing makeup, women wearing binders).
Edit: Minor details
Strange to see breast-binding spoken of as analogous to face-painting! Makeup washes off at the end of the day and doesn't restrict activity (swimming aside, I suppose). Meanwhile it's tough to do so much as run for the bus when your ribcage is constricted.
I've worn a binder for five years and this is just not true. I regularly run and hike while binding. It makes it slightly harder, but not much.
Are you speaking from imagination experience
Well-written and measured comment!
Which leaves me curious, as gender is clearly important to you, despite it being a socially constructed set of labels and descriptions. Meanwhile, it feels entirely meaningless to me, as 100% of my "identity" is entirely subjective to me and almost inexplicable....so that said, what percentage of how you relate to yourself do you think is partly through the eyes of others and what percentage is instead wholly subjective and internal?
I did go through a whole questioning phase and wound up being more conservative if anything. This is probably one of these things that vary hugely from person to person.
Aella, I’m a big fan of your surveys—I even used your public data to look at rates of paraphilic interests (thehumanconditionrevisited.substack.com/p/how-unusual-are-unusual-fantasies).
In my opinion, comparing age stratification to change over time doesn’t really work because they’re asking different questions. Age stratification shows how different age groups compare at a single time—for example, whether 16-year-olds differ from 30-year-olds right now. Change over time looks at how the overall population shifts across multiple time points. An age-based difference at one point doesn’t tell you how attitudes or behaviors are trending over time, and a trend over time doesn’t prove that one age group is driving it. So the two analyses can’t really be used to confirm or refute each other.
Also, the small uptick you mentioned in one age group hasn’t been tested for statistical significance, and there’s no margin of error reported. Without that, it’s impossible to know whether the difference is meaningful or just sampling noise.
That said, I also have doubts about the representativeness of Erik Kaufmann’s data. A lot of the attention it received seems driven more by political confirmation bias than real informative value.
Agreed. If nonbinary identification is decreasing compared to previous highs, you would want to show that by polling people in eg 2022 vs now. With your current data it's entirely possible that you have a lot of people who responded that they identify one way and have since changed their minds.
Yea i am def doing a single snapshot vs. the other paper's tracking over time. I'm operating on the assumption that the effects I'm seeing in the snapshot are going to carry over as people age; if this turns out to not be the case I'll publish another post saying 'oops'.
And If you're talking about the uptick at 19, it's almost certainly not sampling noise; the sample is like i think 150k males for that line? Another indicator of high certainty is how smooth the line is. I didn't include CI also cause I don't think there's a reasonable computation for that when I'm doing weighted avgs. I usually do try to include shaded CI ranges when I'm working with just the raw data.
Aren’t you surveys super selected based on who reads your content??? Maybe I’m missing something here...
I do discuss the types of sample I get and how this differs from other data in my post.
But also, the data I'm drawing from is almost entirely people who don't know who I am. I regularly post drawing from this dataset.
I wonder if the minorities in your daya-set are more likely to know who you are though.
As in, if 90% of your data-set say they don't know who you are; would the fraction of for example trans people in your data-set who know who you are be *higher* than 10%?
In response to this selection comment -- just to say some obvious things:
(1) Selection is inevitable in survey data, even in the most well-resourced economic policymaking surveys like CPS or SCF, because one can't force people to answer one's questions. If you are concerned about selection because of aesthetic horror, avert your eyes from survey data.
(2) If your objection is practical rather than aesthetic, the correct response to selection is to take survey data seriously when the questions asked of the data are along dimensions where selection is unlikely to vary.
(3) With respect to this post, and plots of % non-cis vs. age: You can tell a very plausible selection story that shifts the whole curve up or down (as the author clearly notes, and ballparks the likely magnitude of the shift). You could tell a maybe-plausible story where selection scales monotonically with age. But I do not think you could tell any reasonable selection story to generate the shape of the curves in Figure 5 (the one with n=541,782).
(4) so, tl;dr: selection is not much of a concern in this analysis. (Not because "selection doesn't exist" but because the author asked questions of the data in such a way that selection is unlikely to confound the answers.)
> My sample now is ~850,000 people, median age 22, drawn from many corners of the internet (viral on multiple platforms, very good google SEO)
You also might be missing
https://www.astralcodexten.com/p/selection-bias-is-a-fact-of-life
I think that subscribers to Aella are just a very very narrow subset of people; the sheer largeness of the numbers just doesn’t matter very much.
To quote the article: "Selection bias is disastrous if you’re trying to do something like a poll or census” though "Selection bias is fine-ish if you’re trying to do something like test a correlation.” This obviously depends on degree but I really think this is a very very selected population, right? I mean, this feels obvious to me, but please tell me what I’m missing.
Most survey participants are not subscribers, but found the survey where it went viral online ("see how you compare to other people!")
You’re right; I should rephrase. The types of people that would have seen/responded to these surveys are likely a very particular subset. Social media platforms don’t just show your content to random people. Do note that she is using this data not sparingly but to refute a different study, which doesn’t have this feature. This seems very weird to me.
The 'particular subset' is mostly 'people who like taking personality surveys', which tends to be young liberal western women. I have a friend who runs an old personality testing website, and the demographics of those who answer his surveys line up almost exactly with mine.
You can look at what groups are more likely to answer your data and get a pretty good sense at exactly how and in which ways selection bias is impacting your results. You'll notice in the above survey I weight for politics and anxiety, two of the biggest things that skew my results differently from representative samples.
Additionally, the original survey I'm responding to is just "people in college", which you could argue is perhaps an even more selected sample than mine, which isn't limited to a specific socioeconomic strata.
Seems right. Thanks for clarifying (and I’m a big fan!)
You are completely correct and I strongly agree with you. This data is extremely skewed towards those who would ever take online surveys and who would even bother to respond to questions of this sort, as well as being skewed towards those who follow Aella (even if those who don't follow her also respond).
To me all this data is showing me is people who are more online are more likely to identify as trans. Nothing more or less.
The point of the numbers is that her followers are only a small portion of the people responding. Depending on the platform she has 100-200k followers, and only some portion of those took the survey.
See my response to MugaSofer.
I do survey research for a living. There are many sources of error in survey research (coverage, measurement, etc). It is very hard to write survey questions that measure what you think they are measuring. Even if you succeed in writing a 'good' question, that question is still embedded within a larger instrument, so the 'same' question surrounded by other, varying questions, can produce different results in different surveys. We usually put the demographic section (age, sex/gender, education, income) last to avoid breaking rapport. Americans will willingly tell you all about their kinky sex practices, but asking them about income is seen as just too invasive. Right now, I am doing a large, repeated cross-sectional survey for a 'federal agency' (I don't want to get too specific about the client because I am worried about the black helicopters showing up over my house if I appear too woke). When we started a few years ago, during the approval process, we proposed using a simple, one response only, sex/gender item - it was more or less "What are you: male, female, transgender, do not identify as male, female, transgender"? The client objected to this and said we must use the then-current Census approved version, which split it into two questions: (1) "What is your gender identity (multiple response allowed): male, female, nonbinary, questioning, prefer not to answer, other"? (2) What is your gender assigned at birth: male, female, X, unsure, prefer not to answer, not listed"? Note that the word 'sex' does not appear in either version - only 'gender'. When Trump was elected, we were first required to change back to a single item: "What is your sex: male, female, prefer not to answer". Apparently there were some nervous discussions within our client group, and, before we ever fielded this new version, we were again told to drop the 'prefer not to answer' option. Note that this question (web survey) was never mandatory, so a respondent was always free to skip it. Hence the 'prefer not to answer' option was really unnecessary, However, the client apparently did not want to even suggest to respondents that there was any fuzziness about this issue. And, the word 'gender' disappeared to be replaced with 'sex'. The point of all this is that Aella may have difficulty drawing conclusions from these surveys because of the large potential for measurement error. Another point: whatever we are writing in these comments is probably being scraped off the web by Palantir, etc., and incorporated into our secret 'social credit' scores. So, watch out for the black helicopters and slaughterbot drones.
I do pay quite a lot of attention to survey wordings, and for years have experimented with wording (70+ surveys, plus countless twitter polls I use to run beta tests on how wording impacts reults). I've found it matters less than most people think, except in some specific situations where it matters a lot. I have used gender vs sex before. My current gender question is a final evolution of a lot of honing and people yelling at me lol.
Have you written a post on how wording influences results? That would be very interesting.
Relatedly, does your survey put questions about sexual attraction near the gender identity questions, and are there variations in the answers if you do or don't?
I did not mean to denigrate your work in any way - it's truly amazing and better than many PhD dissertations I have read. My comment was more about the current frantic MAGA zeitgeist than gender/sex issues.
I haven't tested order too much I really. The identity and sexual questions are relatively far apart in the survey. You can Google Big Kink Survey to see!
I do keep meaning to write a post about survey wording yeah
Just to be clear, since you naturally didn’t hit the point too hard, this is overwhelmingly strong evidence for the social contagion theorists and for there being a big, big problem with how teenage girls are experiencing themselves in our society. There’s just no way that anywhere within screaming distance of 15% of AFAB people are really men and this has never been noticed in history before. For comparison to the best-known modern society with a more stabilized trans identity option, it’s something like 0.5% of AMAB people are trans according to the gender structures available there in Thailand.
Thai culture is not very accepting of trans people. It's a well known location for transitioning surgery for some interesting historical reasons, but the culture and laws are not accepting. You can't legally change your gender identify for example.
As near as I can tell from surveys, trans people are somewhere between 2-5% of the total population. Widespread social stigma likely suppresses that to some extent, but not enough to justify a 14% difference.
I think this is almost certainly an artifact of who is answering Aella's post, given that I can find no other sources saying anything close to even 10% of AFAB people are identifying as trans men.
Every single other survey including the ones Aella cites have significantly lower rates. All clustering around the 2-5% rate.
When one poll has an extremely dramatic outlier to all other polls, I don't think that's evidence of social contagion. I think that's evidence that something weird is going on in this one poll.
Where are you getting the 14% number? My data has roughly 3-4% of people identifying as transgender (after weighting for politics/anxiety).
My estimate is 11% of fourteen year old females, not females overall. I recommend reading the full post where I describe this.
I did read it. The person I replied to said 14%. I assumed it was somewhere in the hard data post that I didn't have the time or bandwidth to read.
You seem to be ignoring the entire point of this data which is the slope of the curve. Even if Aella’s sample is biased you have to explain why those in her biased sample are getting far, far less cis at younger ages, and to assume that has no relationship with the whole population you’d have to make some really bold epicycles. Note that her rates for 30-35 year olds are quite similar to population data! (I also have no idea where you’re getting 5% numbers for the total population, which would certainly imply stats something like Aella’s for teenagers, again since the rates are aggressively slopes downward with age. The first few data I find say 1-2%, with some 3/4 of these under 35. There are very, very few trans Xers, Boomers, and Silents.)
And it is exactly the difference between the youth rates and older rates that makes me suspect the actual population is closer to 2-5% range. There was extremely strong, state sponsored violence and repression of transgender individuals, combined with far less general knowledge of or support of transitioning.
Similar to the "spike" in people suddenly becoming left handed, it is not shocking that youth rates are substantially higher than older folks and there are very strong reasons to believe that the actual rates are closer to the youth rates or even slightly more than the current youth rates as there are still strong levels of repression in both legal and cultural ways.
You're doing some kind of complicated metaphysical claim, here, about how many people are "really" trans even if they would never identify that way. There is no such number. Modern trans identity is a very new construct; gender-nonconforming people before about the 1960's had completely different ways of conceptualizing their nonconformity, older conceptualizations which have only been slowly replaced during the lifetime of older people today, and it's an act of epistemological violence to imagine that you can say "2-5% of them were really trans, they just weren't allowed to come out" and have the sentence even be meaningful, let alone true. I see no meaningful way to measure how many people are trans other than how many people will identify, anonymously, as being trans. Of course reducing discrimination will tend to lead to larger numbers of such identifications, but again, it's an act of epistemological violence to say that those people were really trans, just secretly (from themselves?) rather than that we now have a different set of social constructs into which people are comfortable classifying their experience.
You are accusing me of making metaphysical claims while throwing around claims of me doing epistemological violence?
There is different societal ways of expressing gender non conformity - but there is also accounts that there were people identifying as other genders even in societies with extreme prejudice and no general concept of transitioning.
This isn't coming from nowhere or general navel gazing. This isn't something that is dreamed up. There are biologically identifiable differences in the brains of trans people that match the cisgender people they are transitioning to.
...Yes, that is what I'm doing?
Let me know how the r^2 on predicting trans identification via brain scanning is doing. I'll wait.
(Actually, that's not really the point: of course trans people are different than cis people in ways that will show up on a brain scan, even though I would bet 100:1 odds that you can't reliably predict trans identification from brain scan alone. But the brain scan doesn't prove that that difference is somehow metaphysically correctly interpreted as 'being the gender opposite your birth sex', as opposed to the any number of ways other societies have interpreted it. These categories, again, are socially constructed and constantly in flux.)
The 2-5% rate is the surveyed rate for trans youth from the various surveys I have seen. There is a relatively wide range, but they all cluster around that range, which is enormously lower than this sample.
I don't know the explanation for why Aella's survey differs so dramatically. It's pretty standard when doing a meta analysis to discard significant outliers though. When you have ten studies that say one thing, and another study that says something wildly different, that's not a reason to say the 10 studies must clearly be wrong or missing something.
It's not bold to suggest that this one outlier survey is not representative or has some other flaw that is making it substantially different than every other survey.
Well, OK, but what is "youth", and why did you say "total population" when you meant "youth"? Pretty sloppy!
Anyway, Aella herself adjusted her estimate for college-age people to 5.5%, which is perfectly in line with what you're saying. You haven't actually shown that Aella's results are even an outlier in any detail, let alone the really significant detail on the massive non-cis identification in young high school-age AFABs, and you're still itching to simply throw out the one thing that she's in a better position to measure than almost anyone, and which would a priori be independent of any sampling bias, which is the slope of the curve of trans identification down to young ages. That might fly if you were simply trying to publish a meta-analysis, but I think we're interested in actually figuring out the truth here.
I looked at the CDC survey Aella cited, and interestingly, it does *not* seem to find any substantial slope in trans identification among high schoolers; it shows a rate of 7-10% (depending on how you count "questioning") among AFAB high schoolers pretty steadily from freshmen to seniors. So, that does at least suggest that Aella might somehow simply be wrong, though intuitively it seems impossible that freshmen aren't more trans-identifying than seniors, given we know that high schoolers are much more trans-identifying than college students than general 20-somethings. But in any case, does anyone other than these two studies have a sample size of tens of thousands of high schoolers answering this question? I kind of doubt it, in which case I hardly think we could call Aella an outlier.
The CDC behavioral risk factor surveillance system surveys hundreds of thousands of people each year.
This study using the results discusses some of the factors.
https://williamsinstitute.law.ucla.edu/wp-content/uploads/Age-Trans-Individuals-Jan-2017.pdf
As for the slope of the curve, it seems most likely to be that the survey might have become viral and shared in a more widespread way among a subgroup of the population.
I do have an entire paragraph in my post above comparing cdc data to mine
Yes. The other commentor seemed to think that other sources had fewer people surveyed, but that one has a significant amount more than they realized.
I referenced the same CDC dataset that you reference a study based on [a decade-old version of], which in the recent version Aella relies on in the post, involved 20000 subjects. What is your point, here, exactly?
Yes, of course, it is possible that the survey just happened to become viral specifically in highly gender-nonconforming segments of specifically just 14-16 year olds. But you have no basis on which to throw it out just because some extremely particular kind of measurement error is theoretically conceivable.
I didn't say it "just because it's theoretically possible."
It's because you have one survey which has an extremely divergent result from every other survey, from a variety of sources. When you have one survey that has an extremely different result from literally dozens of other surveys that all cluster around the same range, it's fair to consider reasons why that one survey might not be as reliable. This isn't about malice or not trusting. I like Aella and find her surveys fascinating.
It's well known that weird outliers can happen in individual surveys or studies. That's exactly why it's important to do so many and repeat them. What it's important not to do is take odd outliers as stronger evidence than multiple other surveys. See XKCD 882 for a silly example of why slicing data sets to fine or overlying on one individual survey can lead you to faulty conclusions.
I'm not going to say more than three times that you've completely failed to produce more than one survey comparable to Aella's. I do not think you actually have dozens (your own citations bewail the shortage of data on these questions in high schoolers), but if you do, feel free to clue me in. I am perfectly well aware of all the concerns you note: you should weight one (extremely large) study exactly as much as it deserves, no more *or* less. In this case, all I have in front of me is Aella's data vs the CDC's, and I would weight the CDC perhaps twice as heavily. You are, for some reason, saying I should weight it roughly infinitely more heavily by throwing Aella out as an outlier. You haven't provided any reasons at all to do that.
> Thai culture is not very accepting of trans people. It's a well known location for transitioning surgery for some interesting historical reasons, but the culture and laws are not accepting. You can't legally change your gender identify for example.
--
This seems incorrect. In international surveys such as the yearly Ipsos Pride report, Thailand seems to consistently give, by a wild margin, the most trans-positive responses to a variety of trans-related cultural questions.
See: https://www.ipsos.com/sites/default/files/ct/news/documents/2025-06/ipsos-pride-report-2025.pdf - Thai responses rank first, generally *by far* on all questions, for support for teenage transition, access to single-sex facilities corresponding to gender trans people identify with (vs birth sex), support for the addition of a third gender option on government documents, support for insurance coverage of transition, support for trans athletes competing in sex categories corresponding to gender they identify with (vs birth sex), and support for anti-discrimination laws for trans people.
(Likewise for 2024: https://www.ipsos.com/sites/default/files/ct/news/documents/2024-05/Pride%20Report%20FINAL.pdf)
It is true that legal documents cannot be changed, but it is important to distinguish current legal status from what is culturally supported.
I wonder how much of that represents recent changes. I might be going off old data. That being said, that same survey says 68% of Thai people believe trans people face a fair amount of discrimination, and laws do actually matter.
Those numbers are crazy high.
We might wonder if it’s like the rise in autism rates. In the case of autism, the leading theory is that the actual rate of autism has stayed about the same, and the apparent rise is due to including a bunch of borderline cases that previously wouldn’t have been counted.
It might be the same here: a constant rate of transgender, but the statistics have been inflated by including a whole load of non-binary or vaguely trans cases that wouldn’t previously have been counted.
In theory, the gender clinics are supposed to diagnose whether you really need hormones etc.
But
- In UK, waiting lists are insanely long, so patient never gets seen by a specialist to work out of they’re really trans or not
- From reading the stories of a lot of trans people, I get the impression that the specialists have totally no clue whether you’re actually trans or not, so even if you get to the end of the waiting list, it doesn’t help
Is this really any different than I'm a vegetarian? I'm gluten intolerant? The viral trend that teen girls fall into every generation?
The UK has, officially, banned puberty blockers for minors with gender dysphoria (still allowed for precocious puberty, cancer treatment, etc)
From trans people’s accounts, it would appear that it has also got a lot harder for adults to get hormones, though there is no official ban.
The problem with banning treatment for everybody, rather than making an effort to work out which of these patients is actually trans, is that the adults at least will just get hormones on the black market.
Autism "rise" is not due to "borderline cases", it's due to the fact it was very poorly defined before (of course, because it was defined by literal nazis), and the science has gotten better at defining what it is.
That's it.
Spoiler : they did that by... listening to autists. Sensory hyper/hyporeactivity was added to the DSM only in 2013, which is absolutely wild considering how constitutive of the autistic experience it is.
Also, thinking it's for a specialist to say of you're trans or not is incredibly condescending, transphobic, and frankly disgusting.
Unless you think someone else than you can decide what's happening your mind, which is a whole other level of dystopia 🤷♂️
I think people have to figure out for themselves if they’re trans or not, and what treatment, if any they want.
It seems likely that the alleged experts don’t have better insight into whether a patient is trans or not than the patient themselves.
Of course, gender clinics also have a financial motive to diagnose people to get $20k surgeries.
> we don’t yet know how to separate these people from those who would be harmed
Your operative word being "yet". I don't think there's a "yet" here. I suspect that there's simply no way to do that. Teen brains are much too … fluid, in-the-process-of-rewiring-themselves, impressionable, take your pick … for that to work.
Heck even the percentage of adults who end up regretting transitioning are way larger than statistical error.
If you're interested in sex and psychological profiles in detransition, which you allude to, check out this post (spoiler, we found a greater portion of afabs detransitioning due to changes in identity several years post-transition):
https://www.psychiatrymargins.com/p/understanding-detransitions-in-transgender
Interesting results, nice work!
Sorry if I am being stupid here, but are you reading the time trend by looking at the age profile? As in: more young people than old people are trans -> upward trend in transgender identity?
Doesn’t that conflate age effects with time\cohort effects? Is the assumption that gender identity is “set” at age 14?
The main conclusion I would draw from this is that a 14 year old is so precociously interest in sex and gender that they are both likely to find and complete a survey on it, is disproportionately likely to identify as some form of non-binary
Yes, yes, selection bias always exists, but the these biases in this case are especially self reinforcing, and probably explains the steepness of the slope
Nevertheless, I share Aellas concern that there has been an emergent exposure that is increasing in intensity and/or prevalence, that is inducing “transness” in young people, and particularly in young AFABs, who given the same underlying predisposition to transness, would not have developed it if they had existed in a different environment
How is 14 year old precocious for interest in sex ? At this age, most teenagers are interested in sex, and as for all new things they're more likely to research it. Especially when they have all the world knowledge in their pockets.
The world is a big place, and people are interested in travel. But there is a certain kind of person that ends up at an Auyuhasca retreat in the mountains of Peru.
"I haven’t checked to see if there’s other surveys that disagree, so in theory he could have been cherry picking, but I doubt it"
Why ? Anyone supporting the anti-science party need to be triple checked for bias. I mean, in the vast majority of cases, they're not following scientific principles, so on the contrary I'd assume he cherry picked.
"I think it’s unlikely that 11.5% of afabs are actually trans men in a way that would last through adulthood."
Why ? It does seem high to me too, but then again, 5% would have seemed high 10 years ago. There's no particular reason to think this or that percentage is too high, it's kind of arbitrary.
There’s something I noticed about this when I first read it, and I see other people have noticed it too.
There is a difference between measuring how the same cohort changes over time, vs different cohorts at the same point in time.
What the data here seems to be showing is people born later are more likely to identify as trans at the same point in time as those born earlier.
But the question is really, have the same people become less likely (or more) likely to identify as trans due to political changes over time?
It is a basic of survey methods, that is explained in almost every undergraduate psychology text book, that people sometimes lie on surveys, particularly if you ask them about something that society views negatively. So a decrease in trans identification on surveys might be more people refusing to tell you they’re trans. But here, we’re just not measuring that.
The other things that’s interesting here is that the moral panic in the media is almost entirely about trans women (women’s sports, women only spaces etc) but the data suggest that the increase is mostly AFAB.
Interesting. Thanks for sharing. How do you screen for and remove bots from your data?
It's actually pretty easy to detect abnormal responses in this specific dataset. One reason why is that it's a 30-40 min survey to complete, and you can look at how long it takes people to complete it. Maybe in theory someone is running bots very slowly to replicate human response times, but they would have to do a whole lot of that to displace my current organic sample at 850k and I'm not sure why they would be motivated to do so.
And there's further ways of detecting false response sets even *if* they get past that check.
Thanks! I posted a link to me and my team's detransition study above, and below is a link to our sampling methodology paper in which we developed a protocol to identify and remove bots and real human fraudsters (who we ended up interviewing as part of the screening process). It was a heavy lift! But a step my team took took because of how controversial and contested the topic is. You'll see in this paper the challenges occurred even in advertising and recruiting for the study on social media. I think you'll find this paper interesting (I am a queer trans man/transmasc person BTW, and am interested in this topic because several of my own friends I transitioned with 15 years ago have since detransitioned, and it seems to disproportionately affect queer AFAB folks). Sampling paper and bot/fraud removal protocol: https://www.jmir.org/2025/1/e63252
Oh yeah I would be way more afraid of bots and malicious responses if I were doing a trans focused survey. In fact the one time I have detected en mass bad data in my surveys was specifically for a survey about trans people, where it was advertised as such!
> [...] but I guess that’s why we’re all here so yolo.
<3