Yes, Trans is in Decline among Young Educated Americans

A Reply to Critics on both the Left and Right

Has the dramatic rise of transgender identification among young people gone into reverse?

That was my claim, published on Tuesday in a thread on X, at Unherd and in my new Centre for Heterodox Social Science report. The first chart is key: it shows the percentage of students in three surveys identifying as a gender other than male or female. The non-binary share rose across three independent surveys of educated young people between 2020 and 2023, then declined between 2022-23 and 2025. The fall in non-binary identification is quite dramatic, a halving in just two years. This is the first indication that the transgender trend among young people has reversed.

The data is drawn from three sources. First, the massive annual Foundation for Individual Rights in Education (FIRE) survey of American undergraduates. The latest wave covered over 68,000 students at more than 250 institutions (skewed toward major research universities and liberal arts colleges). The FIRE survey can only sample a small fraction of the student body, so I compare this data with two sources from individual institutions where a larger share of the target pool is surveyed. These include Andover Phillips Academy (a private elite prep school) which conducts an annual student survey that is generally answered by more than 3 in 4 students, and the Brown Daily Herald spring and fall polls of Brown University students, which reach as much as 50 percent of the target population. The fact all three tell a broadly similar story is remarkable.

The other major finding in the report concerns sexual orientation. While the more traditional gay and lesbian category had a stable share across 2020-25, the proportion that is of non-conforming sexual orientation - queer, questioning, bisexual, pansexual, asexual or other - has risen and fallen along the same trajectory. This shows up in the non-heterosexual share, as captured in the graph below. There is a rise in most datasets to a peak in 2023, followed by decline. (Bear in mind that data is missing in many series beyond 2023-24). Notice that while the non-binary line above peaks at 9 percent of the total student body, the non-heterosexual share peaks much higher, between 30% and, in the case of Brown students, over 40%, of the total student population. Nearly half of Brown’s undergraduates were LGBTQ+ in 2023, and in some liberal arts colleges such as Smith College, that figure topped 50 percent in the FIRE data.

Just so you know exactly what these lines are measuring, here are the questions that ran on the FIRE surveys in these years (Andover Phillips and Brown used very similar formats):

1. Gender

Which of the following genders do you most identify with?

1) Man
2) Woman
3) Nonbinary
4) Agender
5) Genderqueer or genderfluid
6) Unsure
7) Prefer not to say

2. Sexual Orientation

What is your sexual orientation?

1) Gay/lesbian

2) Straight (heterosexual)

3) Bisexual

4) Pansexual

5) Queer

6) Fluid

7) Same-gender-loving

8) ASexual or aromantic

9) Unsure

10) Questioning

11) Other

The trans proxy encompassed categories 3 through 7 in question 1. The BQ+ proxy for the sexual orientation question amalgamated answers 3 through 11.

The Reaction

The first thing to say is that the tweet thread went viral, retweeted by a number of major conservative accounts, among others, Elon Musk (via Wilfred Reilly), Donald Trump Jr and Matt Walsh. It was also covered widely in the press. The post had 30,000 likes and was seen 6.1 million times. I have been on a number of TV and radio spots since, including Unherd and Fox (Laura Ingraham), with more in the offing.

The Critics

Two main lines of critique soon emerged. The first more conceptual, concerning the operationalization of transgender, the second more technical, revolving around whether my analysis should have used data weights.

1. Non-binary is not Transgender

The first critique is that the survey I responses I used picked up non-binary identity which is not the same as trans. The argument here is that a trans individual identifies with a binary gender (male or female) identity that is different from the one they were born into.

My fellow Vancouverite Billboard Chris Elston (whose work I generally admire), along with some others who are campaigning against gender ideology, are worried that if people feel trans is decreasing they will relax the political campaign to reign in gender ideology in schools, institutions and the law. Chris cited a post from Benjamin Ryan, a centrist writer on this platform, and a Swedish researcher on X.

My first response is that the meaning of the term transgender is not fixed to only refer to binary transgender individuals who identify as male or female. Movement organizations and activists include non-binary under their definition of trans.

Second, even if we are focused only on binary trans individuals, the question is how they would answer the gender question above. Even if most tick male or female (which is a big assumption that I am skeptical of!), so long as the share who tick a non-binary option is non-trivial, the non-binary category will be an indicator - even if somewhat muted - of trans identification. My study was not about providing an exact point estimate of the binary trans population, only the direction and magnitude of the trend over time.

Third, even if every binary trans individual ticked a male or female identity, the non-binary category is arguably the gateway between conventional gender identification and transgender identification. That is, a girl who comes to believe she was born in the wrong body would go through a period of non-binary uncertainty before definitively moving to the other gender. Thus if non-binary is increasing, the pool from which binary trans individuals springs will increase, and the number who will fully transition will increase.

That is, non-binary and binary transgender trends should be correlated over time. Which means that trends in non-binary will reflect trends in binary trans identification.

I said as much in my response tweet on Thursday:

And Benjamin Ryan was good enough to admit that there is no data that can disprove my challenge (I accept that I don’t have the data either, but I think there are strong conceptual reasons to expect a correlation).

Jean Twenge, whose work I greatly respect, then produced an article which argued that while non-binary is not exactly the same as trans, and my data focuses more on elite youth than the wider young population, there is an indication from the US Census Bureau’s Household Pulse survey data that transgender identification rose after 2021 and declined between early and late 2024.

Sadly, as Twenge notes, the Trump administration’s policies have meant that government sources can no longer ask questions such as these, so we won’t get 2025 data. This is one reason why I think people should try to keep facts and values separate: asking about trans is not the same thing as endorsing trans. I have a similar skepticism about the power of census racial categories to shape identity.

Of course it is theoretically possible that you could get a fall in non-binary share but no decline in binary trans identification, or even hardcore (chemical, surgical) transitions. But I think that is unlikely. These phenomena have risen together and it would be surprising if they did not fall together.

2. I Did Not Weight my Data

The second line of attack came more squarely from left-wing accounts. Erin Reed, who runs a popular blog on the left, triumphantly declared a ‘gotcha’ that debunked the entire story from ‘far-right professor Eric Kaufmann.’ This was a more technical claim, which argued that because I did not use survey weights (which are used to make survey responses match the population if some groups such as men or Hispanics are undersampled), the results were invalid. I had not used the weights included in the FIRE data, and therefore my results were ‘dead on arrival’. She managed to find a number of methodologists that endorsed that view. The subtext: I was a methods neophyte who committed a schoolboy error.

The source for Erin’s piece - which was picked up by other sympathetic sources - was blogger Jacob Eliason, whose piece argued that I had made a ‘significant analytical error’ by not using survey weights.

Eliason produced a chart showing that applying weights not only removed any increase in non-binary identification, but actually showed the opposite: a big jump in non-binary in 2024-25. Gender non-confirmity was not in decline, it was surging!

But take a look at Eliason’s graph, comparing dotted (my unweighted data) and solid lines (the weighted data) within each colour band. Warning lights should have been going off. Why? The first thing to notice is that the weighted solid lines for each colour pairing show more volatility over time than the unweighted dotted lines. For instance, even when looking at female share in red, this moves up and down a lot while the unweighted data is more stable.

Now focus on the green lines for non-binary. If Eliason’s account is correct, non-binary share was a stable 2% between 2020 and 2023, then jumped to 9-10% during 2024-25.

What that would mean is that instead of my chart, which shows an alignment between the FIRE data and the Andover and Brown data (where sampling is much less of an issue due to a larger coverage of the target population)…

..the reality is that the FIRE data (watch the orange line) exhibits a diametrically opposing trend from the others, not to mention from the non-conforming sexual orientation trends:

Why is Eliason’s chart wrong? The truth is that I had experimented with FIRE data weights, but past experience had taught me that when looking at small groups, applying weights can distort survey results much more than they can help them approximate the population.

What happens with weighting is that the raw data are ‘corrected’ to match a population (by sex, race, school, and so forth). This makes sense when you are comparing universities for rankings such as FIRE’s free speech rankings. You need to be sure that you haven’t sampled a disproportionately female group of students at Harvard and male students at Yale, knowing that men are more pro-free speech than women, or the rankings won’t be fair. Official population and college data sources provide the demographics to adjust the raw survey data. That’s the right approach, especially to construct rankings in a given year.

But there is a problem with using weights to look at aggregate trends over time. This arises in two ways. First, if there is a change in weighting methodology between years that affects certain questions. This affects small groups like trans where a few individuals may be inflated or deflated several times over in the survey. Second, because the aggregate ‘population’ shifts as some colleges are included in certain years and others in other years, and so the weighting formula to derive nationally-representative samples varies over time.

This volatility means that going with the raw data may be a better strategy than applying weights which introduce more error than they correct for. In this case, I think the evidence is quite clear that weighting is a mistake.

Lo, all of a sudden, after my response to the weighting criticism, there was an ‘update’ from Eliason recognizing (to his credit) that applying weights might not be such a great idea.

It turns out that the gender benchmarks used by College Pulse to weight its FIRE survey were changed in 2023-24, producing the spike in non-binary we saw in the weighted data.

Eliason then pivots to the argument that using unweighted data is a problem, illustrating his claim by showing an unequal distribution of freshmen, sophomores, juniors and seniors across the years.

But when weights cannot be relied upon, there are other strategies one can use. Running a fixed-effects statistical model that controls for each college, as well as demographics such as graduating class, financial aid status, socioeconomic background, race, religion and ideology screens out the distorting possibilities that Eliason asserts. The result of such an exercise, graphed below, confirms the ‘rise and fall’ pattern in my report.

Year of study remains statistically significant in predicting whether a student identifies as neither male nor female even when controlling for many of the things weights correct for, i.e. demographics and the mix of colleges.

To end on a positive note, the debate on X shows that twitter is not a sewer of misinformation and partisan claims, but an ecosystem where claims can be challenged and defended and serious people can assess truth or falsehood.

Trans is either in decline or it isn’t. I believe the data shows that it is, and that this represents a significant change in youth culture. It also has implications for the confidence of cultural progressivism, as rising gender and sexual diversity is a key component of its historicist narrative that humanity is progressively overcoming oppressive traditions.

However, at the end of the day, future data will be the judge of whether this is a blip, the beginning of the end of trans, or a decline that will settle out at a higher steady state than before the trans awakening of the 2010s.