Incels Rising

TL;DR

• Analysis of three major US datasets (YRBSS, NSFG, GSS) confirms increasing sexlessness and later loss of virginity among youth.

• Relative sexual inequality is increasing (Gini coefficient) due to an increase in virgins, but absolute inequality is declining.

• Reasons for being a virgin appear to be increasingly involuntary or ambiguous, as instead of voluntary.

Introduction

In the past few years there has been significant discussion of ’the rise of sexlessness’ or on incels broadly. For example, we’ve all seen this plot:

It still generates clicks to this day, despite being outdated and puddle deep analytically. Fortunately the data-driven folks have long since dismantled any debate to be had over this sensationalism.

However, as far as I’m aware, nobody has simultaneously covered all three main datasets in this discussion directly on this topic. It’s either the NSFG or GSS, very rarely the YRBSS.

In an ideal world there would be a team collaborating together meta-analysing these trends internationally, not just on sex but many other milestones like age of first employment, alcohol consumption, etc. Sort of like what Jean Twenge did with her book iGen but bigger.

Since that would be too difficult for me to do I’ve decided to limit my analysis to the US; specifically on sexlessness, without breaking it down by race or social class.

I’ve limited it to the US because they have the most accessible and prevalent data in this discourse. But again, keep in mind the US is not representative of global trends. I personally have not seen that these trends can be applied internationally - in fact what evidence I’ve seen so far suggests that sexlessness is not rising abroad.

It’s often assumed that any socio/political phenomena happening in the US must also be happening in the Anglosphere, and probably in other developed countries (Europe, East Asia). This isn’t without reason for many trends in the US are being replicated elsewhere: falling birthrates, woke culture (Great Awokening), mass immigration, obviously social media, etc. These are just a handful. However, regardless my advice would be to approach such assumptions to other countries with caution.

Overview

The three datasets of focus:

1. The General Social Survey (GSS)
2. Youth Risk Behavior Surveillance System (YRBSS)
3. National Survey of Family Growth (NSFG).

Keep in mind the sample sizes and date range given here are after filtering for the variables we want. As we can see here, it’s not only the sample size that’s important but also the age range and time span across which these surveys were conducted. The GSS is the weakest on these metrics because after we divide by age and number of years available, we are nearly left with nothing. If we naively divide n by (n-Ages * n-Years) we get approximately 15, 118 and 1218 for the GSS, NSFG, and YRBSS respectively. Of course there is obviously an advantage to having the GSS despite nearly being two OOMs behind the YRBSS, since it includes frequency of sex and enables us to take a look at older generations - both of which are invfaluable later on.

Sexlessness in the YRBSS

We will start with the YRBSS. Is sexlessness rising?

Seems like the answer is yes. Interestingly, it seems to be rising faster in males than in females. This is statistically significant (seen below).

Moving on, we can use a spline to predict the % virgin by age, then use that model to predict the exact age at which virginity dips below 50% of the sample.

Let us call this the median age at which virginity is lost:

The betas here are at approximately a 6 month increase every decade.

Inequality in the YRBSS

We can also measure inequality. The two important measurements: absolute and relative.

1. Absolute inequality can be measured by variance, mean squared error, mean absolute error, etc.

2. Relative inequality however is usually measured by these economists’ functions, like the Gini coefficient.

To distinguish the importance between the two I’ll give an example:

Let’s say 10% of the population owned all the wealth in a society. If that 10% doubled their wealth, the Gini coefficient wouldn’t change. However, the absolute difference for that 10% compared to the mean would double. Likewise, if the wealth of the 10% were to halve, the Gini coefficient would still remain unchanged. This is relevant as we will see soon.

Let us apply theory to data. What do you notice here?

This almost looks like exactly the same plot as the one on rising virginity! So, what’s the correlation?

See? There is a near perfect correlation at r = 0.996, near perfection. So, almost all the relative inequality is driven by virginity, not changes among those who have had sex.

Let’s break the lifetime partner counts down by decile, grade, sex and then compute the deviation from the mean for each. Values closer to zero indicate that the mean for that sex/year/decile is less deviated from the overall mean of that sex/year.

As we observe, despite massive increases in relative inequality, absolute inequality is declining across the board.

If you look closely at men in the 10th decile, you’ll see the deviation is increasing for 12th graders. This is an artefact of the data - everyone in the 10th decile answers the highest response (6 or more partners). As a result, when overall sexual activity declines, the 10th decile doesn’t move because it’s still above the ceiling, resulting in the deviation appearing to increase.

Another fun way to visualize the data is to produce a lorenz curve by decile, sex and year.

Now you can see the inequality increase. 10% of the population accounts for 50% of the partner counts in the instance of 12th grade males in 2023. Contrast this to 1991, where it was only ~25%. Even the bottom 30% were getting something back then, now 50% aren’t getting anything!

YRBSS Conclusions

It’s from this simple analysis here that we might be able to steelman the Incel argument. Incels’ grievances might be due to simply rising inequality, regardless of the absolute trends. From the perspective of the average guy (which is now a virgin leaving high-school) a handful of guys are now having most of the fun - whereas this was never happening before.

So overall, we can draw from this that virginity is rising for the youth, resulting in:

1. Rising sexlessness.
2. Rising relative inequality.

However, absolute inequality is declining even when broken down by decile.

The Incel claim that women are unaffected is false; but they are less affected in the YRBSS than men for both sexlessness and virginity.

If we generate a linear model of the summarized data we get the following:

Sexlessness:

> lm(virgin ~ year * sex_coded + year * grade_coded, data = virgin) %>%
+   broom::tidy() %>%
+   arrange(p.value)
# A tibble: 10 × 5
   term                   estimate std.error statistic  p.value
   <chr>                     <dbl>     <dbl>     <dbl>    <dbl>
 1 year                   0.00756   0.000723    10.5   8.27e-19
 2 (Intercept)          -14.5       1.45        -9.97  1.26e-17
 3 year:grade_coded12th  -0.00369   0.000915    -4.04  9.37e- 5
 4 sex_codedMale         -5.21      1.30        -4.01  1.02e- 4
 5 year:sex_codedMale     0.00257   0.000647     3.97  1.20e- 4
 6 grade_coded12th        7.10      1.84         3.87  1.75e- 4 <- Men more affected
 7 year:grade_coded11th  -0.00199   0.000915    -2.17  3.17e- 2
 8 grade_coded11th        3.78      1.84         2.06  4.18e- 2
 9 year:grade_coded10th  -0.000874  0.000915    -0.955 3.41e- 1
10 grade_coded10th        1.65      1.84         0.899 3.70e- 1

Mean Virginity:

> lm(`Virgin Age` ~ year * Sex, data = threshold_ages) %>%
+   broom::tidy()
# A tibble: 4 × 5
  term         estimate std.error statistic     p.value
  <chr>           <dbl>     <dbl>     <dbl>       <dbl>
1 (Intercept)  -69.6     13.8         -5.05 0.0000202  
2 year           0.0428   0.00687      6.23 0.000000730
3 SexMale      -55.7     19.5         -2.86 0.00772    
4 year:SexMale   0.0276   0.00972      2.84 0.00805  <- Men more affected

Although, I wouldn’t call these results ultra-significant. So do they replicate?

Sexlessness in the NSFG

When I first made this post, only data up until 2019 was released. At that point in time the sexlessness trends in the NSFG weren’t spectacular.

That completely changed with the 2023 data release.

Since the NSFG is superior to the GSS in sample size this made it a favorite tool of many bloggers to dismantle Incels’ claims of rising sexlessness.

This is not longer a strong point since:

1. The NSFG is the shortest running survey of both GSS and YRBSS, meaning it’s harder to observe statistically significant trends.

2. The sample size of the YRBSS is larger.

3. It doesn’t matter because the newest NSFG release shows massively rising sexlessness anyway.

Here are similar visualizations:

Inequality in the NSFG

What about inequality?

Moreover, this is now a replication of the YRBSS.

One might argue that these recent declines in the NSFG on every metric should be interpreted cautiously; mostly because this is the first post-COVID NSFG.

No.

The YRBSS (which again has a larger sample) and GSS have been corroborating these trends for decades. It was pretty obvious what was going on before this single survey, this just killed the last argument that could be made.

The rise in virginity is statistically significant across all age groups together (year), but individually, the time series trends for each age group are not (age 20, age 25, age 30).

The widening gap between male / female virginity in the YRBSS was not replicated here.

> nsfg_stnd %>%
+   group_by(age, year, sex) %>%
+   mutate(age = round(age/5, 0) * 5) %>%
+   summarise(virgin = mean(not_virgin == 2, na.rm = TRUE)) %>%
+   filter(age <= 30) %>%
+   mutate(age = as.factor(age)) %>%
+   lm(data = ., virgin ~ year * age + year * sex) %>%
+   broom::tidy()

# A tibble: 10 × 5
   term           estimate std.error statistic   p.value
   <chr>             <dbl>     <dbl>     <dbl>     <dbl>
 1 (Intercept)  -11.4        2.78       -4.12  0.0000659
 2 year           0.00604    0.00138     4.37  0.0000242
 3 age20         -4.64       3.51       -1.32  0.189    
 4 age25          0.547      3.51        0.156 0.877    
 5 age30          4.50       3.51        1.28  0.202    
 6 sexMale       -0.584      2.48       -0.235 0.815    
 7 year:age20     0.00208    0.00175     1.19  0.235    
 8 year:age25    -0.000581   0.00175    -0.333 0.740    
 9 year:age30    -0.00257    0.00175    -1.47  0.144    
10 year:sexMale   0.000295   0.00123     0.239 0.812  

Ironically, again, highlighting the nuance of the absolute / relative distinction, the youngest age cohorts have the highest inequality yet the lowest absolute deviations.

Reasons / Cope for Being a Virgin

Next, we’ll focus in on a set of variables the NSFG includes: reasons for being a virgin, how much the respondent wanted to lose their virginity when it was lost, and sexuality (opposite to same-sex, ‘asexual’).

Quite apparent in this plot is the decline of religion. This debunks IFS’ “we’re having less sex because of religion”. This article argued that the religious were increasingly having less premarital sex, and that this therefore was driving the sexlessness trend. However, if sexual activity is universally declining then one would obviously expect the religious to have less sex! This isn’t because of a change in attitudes (as is visualized above) towards religiosity - rather - the exact opposite is true.

The ‘Incel’ like reasons here are ‘Other’ and “Haven’t found the right person”; both are increasing.

To support the hypothesis that these reasons are more Incel like, we should expect them to increase with age.

As expected, ‘Other’ increases with age, quite starkly. “Haven’t found the right person” is mostly constant; if there were a clear trend we would’ve seen it across the 20s. Religious reasons, as expected, decline after the age the religious would tend to get married. Although, this could just be a cohort/generational effect given declining religiosity.

Unfortunately most of these reasons are voluntary as opposed to involuntary. Only “Haven’t found the right person” could be plausibly classed as involuntary. “Other” is ambiguous. But since most of the big voluntary reasons are already covered, it would make sense that ‘Other’ would mostly be involuntary by exclusion.

Desire to Lose Virginity, NSFG

Another angle from which we can confirm that ‘Other’ is perhaps involuntary is to take a look at the WANTEDSX variable from the NSFG. We can couple this with the age the respondent answered that they lost their virginity. In theory, increasingly Incel and desperate men should be glad that they’re losing their virginities as they age!

The rape curve for females is quite dark. But, moreover, we can see men are mostly constant in their desire. We can quantify this plot into a continuous variable; we’ll set ‘I really wanted’ at 1, the middle response at 0.5 and “I didn’t want” at 0.

It seems women and men converge on their ‘consent’ or desire of having their first sexual experience. If anything, men are just always desperate, and women are increasingly so as they age.

So, this sort of goes against the hypothesis that ‘Other’ is an increasingly desperate and lonely category. Well, except for women.

Is it Asexuality?

What about asexuality? Sexuality is coded like so in the NSFG as the variable ATTRACT.

nsfg code 	label
1 	        “Only Opposite Sex”
2 	        “Mostly Opposite Sex”
3 	        “Equally Opposite and Same Sex”
4 	        “Mostly Same Sex”
5 	        “Only Same Sex”
6 	        “Not Sure”
7 	        “Not ascertained”
8 	        “Refused”
9 	        “Don’t know”

We will assign ‘asexuality’ to 6 through to 9.

Increases with age as expected; however, it can’t be the overwhelming explanation for answering ‘Other’ as a reason for virginity since this only covers 15% of virgins at best, while 60% of virgins are answering ‘Other.’ So most of ‘Other’ must be reasons of Incel. Lets break it down by year and age. Keep in mind n is VERY LOW for this plot. Median for each datapoint is about n = 15.

Seems like there could be a trend here. More on this later though.

INCELS RISING

So now that we’ve gone through the reasons, and possibly explained whether the more ambiguous ones are voluntary / involuntary; we’ll finally classify the reasons into voluntary / involuntary.

nsfg code 	label 	                                                my classification
1 	        “Against religion or morals” 	                        Voluntary
2 	        “Don’t want to get a female pregnant / get pregnant” 	Voluntary
3 	        “Don’t want to get a sexually transmitted disease” 	    Voluntary
4 	        “Haven’t found the right person yet” 	                Involuntary
5 	        “In a relationship, but waiting for the right time” 	Voluntary
6 	        “Other”                                             	Involuntary
8 	        “Refused” 	                                            Involuntary
9 	        “Don’t know” 	                                        Involuntary

Again, many of these involuntary reasons are ambiguously involuntary, but are most likely so. Okay, here goes. Incels by year.

It seems like the more ‘Incel’ reasons for sexlessness have come to dominate men, and shortly women too who enjoy more voluntary reasons for being virgins - in line with Incel theory actually, where female virgins are often blamed as voluntary.

But how is it cut across age?

As expected, the involuntary reasons tend to increase with age.

NSFG Conclusions

So moreover, increases in sexlessness and partner counts are not significant statistically, but they do follow the expected direction.

Despite this, the NSFG does seem to indicate that involuntary reasons are increasing over voluntary reasons. However, the involuntary reasons are ambiguously involuntary.

It’s possible that some of the rise of the feeling surrounding Inceldom can be attributed to declines in religiosity. If you were religious; remaining virgin until marriage was an excuse you could literally believe in. Now with the death of religiosity among zoomers, the culture around sex has flipped - there are no taboos to promiscuity. As a result the natural inclination of men towards lust take over, which isn’t satisfied, leading to more involuntary reasons.

But this is just curious speculation on my part.

Sexlessness in the GSS

If you’re still reading this post then I don’t need to introduce the GSS. The strength of the GSS is its age range, wide array of variables beyond sex, long history, and that it measures sex frequency, not just partner count.

First, we’ll look at changing sex frequency:

Within this plot we can broadly see the lower sex frequency answers increasing in prevalence, while the higher sex frequencies decline. This indicates a shift towards sexlessness. These all follow the expected direction.

What about partner counts in the past year?

A stronger trend for ‘No Partners’ than any other. Disregard anything on the bottom row. The sample size for each datapoint there is ~10.

The GSS has a single variable that can plausibly measure virginity - number of sexual partners since 18. What proportion answer 0?

A far, far stronger increase here. This puts a gravestone in the “GSS doesn’t show declining sexlessness” claims. Following the linear regression this is a doubling over a 30 year period.

Inequality in the GSS

What about inequality? We’ll convert SEXFREQ and PARTNERS from categorical to continuous variables to do this. Sex frequency was quantified to sexual instances per week. Following the respective order you see in the above plots:

code sexfreq    partners
0    0          0
1    1.5/52     1
2    12/52      2
3    30/52      3
4    1          4
5    2.5       7.5
6    5         15.5
7              60.5
8              30

As expected from the above, an increase in relative inequality for across sex frequency.

What about absolute deviation in the GSS?

At least for sexlessness, the first quintile seems to be approaching the mean (0).

Is Everyone Affected?

Given the age range on the GSS, we can check whether everyone is affected by this.

Seems like the relative inequality is up, but it’s hard to tell from the variable trends.

Where are the Incels?

Okay, so sexlessness is up in the YRBSS, GSS and possibly the NSFG. So given this, we should expect more search interest in Incels overall? For this, we’ll focus on the US using Google Trends, using Elliot Rodger as a proxy for Inceldom.

Mostly constant. It could be that Incel history is too short for the above sexlessness trends to be clearly reflected here; alternatively, it could be a selection effect. Incels’ are a highly selected group; they’re mentally ill, disproportionately NEET (Not in Education, Employment or Training) and likely have a higher libido.

Since the proportion of the population fitting into those categories is small, as the population of Incels grows, increasingly normal people are becoming sexless. By regression to the mean, these people must be more normal; employed, happier, less mentally ill, lower libido.

As a result they’re not funnelled into the same terminally online pipelines most older Incels would be. This is good news; because it means that most normal people, deprived of sex, mightn’t care as much as we’d expect by looking at the Incel community.

In a more sophisticated analysis the population of various manosphere communities was measured Ribeiro et al., 2020.

Seems like between 2016-2019 Incels started moving upward; and broadly, it seems the population of the manosphere has grown since its inception. However, there are issues:

• The population of internet users has grown massively. So is the activity relative to the size of the internet any different?

• Data is international (because the internet is). So US trends may not be as relevant here.

• Selection effect; maybe those most likely to be Incel were the first adopters of the internet.

Conclusions

To conclude the trends in the US data:

• Virginity is increasing, sexual frequency and partner counts are declining in the youth.

• The NSFG was an outlier, rather than the GSS, because the YRBSS corroborates the GSS while also being larger in sample size than the NSFG. But with the new 2023 NSFG survey, it appears to have regressed toward reality; NSFG is now in line with YRBSS declines.

• Sexual milestones are still being reached, but at a marginally older age.

• Reasons for being a virgin are increasingly involuntary.

• Relative inequality is increasing.

• Absolute inequality is declining.