Chaos Factor

Big Kink Survey
Oct 2026

Testing HardweightThe Big Kink Survey overstates how common kink is. Hardweight tries to correct that. Here's how close it gets, checked against census and probability-sample data on questions the weights never touched.

The Big Kink Survey's raw numbers overstate how common kink is, because the people who take a long kink survey aren't typical. "Hardweight" is the attempt to correct that. It finds things that go along with kink and whose real population rates are known, like mental health, sexual assault, partner count and porn use, and reweights the sample to match on all of them. Matching the targets is easy, since weighting always hits its targets. The real test is everything else. This report checks 125 questions the weights never aimed at.

125 checks of hardweight against a published benchmark, US respondents aged 14-42. On target = inside the benchmark range or within 25% of it; close = within 50%; off = further. Many checks are the same question in different age bands or sexes. Another 19 rows are weighting targets and aren't counted.

01What hardweight is, and how to test it

The Big Kink Survey is a self-selected online survey that spread virally, with over a million respondents. Its raw kink rates are inflated: people who are into kink are more likely to take and finish it. Hardweight corrects for that the only way it can. It finds things that predict kink and whose true population rates are known, and makes the sample match on them. It takes the respondents aged 14-42 from the US, Canada, the UK and Western Europe and rakes them to match:

• sex × age × LGBTQ
• sex × politics, LGBTQ × politics, age × politics
• sex × childhood sexual abuse, sex × adult sexual assault
• raised in a religion
• age × ethnicity, age × BMI
• sex × age × lifetime partner count
• autism, ADHD, depression
• married, by sex × age (US vs elsewhere)
• porn-use frequency, by sex × age (for 14-17, only use vs no use)
• current religiosity (US) and anxiety among women (US)

Weights are capped at 30. The effective sample size is about 32,000 out of 565,000 rows. A few of these need explaining:

• Ages stop at 42. Respondents over 42 are scarce in this survey. Under 3% of the sample would have to stand in for about a fifth of the population, so a few thousand people would carry huge weights.
• Porn use. The people who take this survey use far more porn than everyone else, and porn use goes along with kink. The targets come from private online probability panels (AEI 2022, Relationships in America 2014). For men they include a small allowance for under-reporting. For women they're raised by about a third, because BKS asks about watching or reading erotic content and the panels only ask about watching. The 35-42 group falls a few points short of its target: there are too few non-users at that age to stretch further under the weight cap. For 14-17 there's no good frequency data, so the target is just whether teens use porn at all (35% of boys and 65% of girls don't). That figure is provisional.
• Marriage, by sex and age, from the 2024 American Community Survey in the US and census figures elsewhere. Without it, the survey's women in their 30s are far too often single.
• Current religiosity and anxiety in women (US only), held at their benchmarks: the GSS for religiosity, and the top of the range of population estimates for anxiety. Without these, weighting on porn pushes "very religious" to 15% (GSS: 11%), and women's anxiety ends up above its range.

Hitting those targets proves nothing, since a rake always hits its targets. So this report mostly looks at holdout checks: questions in the survey that were not used for weighting but have a known population answer. These come from the census, the GSS, NSFG, NHANES, BRFSS, birth records, and published probability samples on sex and kink. If hardweight fixes the sample in general, these should land close. A leave-one-out test (section 07) checks how far the weighting stretches to things it doesn't target.

All holdout checks use US respondents only (n=372,614, effective n about 22,000), compared against US benchmarks and matched on age and sex as closely as the benchmark allows. Each chart shows the raw sample next to hardweight, so you can see whether weighting helped. It helped in 87 of 125 checks and made things worse in 25.

02Demographics and family background

Most ordinary demographic facts come out right:

• Left-handed: men 10.6%, women 9.3%, matching large handedness studies (10.5-11.6% and 9.5-9.9%).
• Height: women 5'0" or under, 7.1% vs 7.2-9.1% self-reported in national surveys. Men 6'0" or taller come out at 38% vs 30-31%, so the men round up a bit more than usual. BMI is a weighting target and height feeds into BMI, so this check is only partly independent.
• Circumcised: 74% of men, a little under the 79-87% from NHANES.
• Mother's and father's age at your birth: each age bracket is within about 3 points of US birth records for 1980-2010. Older parents are slightly overrepresented (father 36+: 22% vs 16-18%).
• Both parents present growing up: 72%, against 61% (GSS, lived with both at 16) to 69% (Census). The BKS question asks about presence in your life, not who you lived with, so it should run a little high.
• Supernatural "plausible": 67% agree. Pew's closest item ("spirits or unseen spiritual forces exist") gets 61-66%, and broader items run higher.
• Currently religious is a weighting target in the US, so it isn't counted as a check.

Two things are off. More respondents say they grew up below middle class (46% vs 35% "below average income" in the GSS). The class question has more rungs below "middle" than the GSS one, so this could be partly the wording. Only children are also overrepresented: 12% vs 6-10%.

Firstborns

This check needs no outside data. If you have one sibling, there's a 50% chance you're the older one. With two siblings it's 33%, and so on. Among US respondents aged 25-42, firstborns are overrepresented in every family size, and weighting doesn't fix it.

03Marriage and kids

Marriage is a weighting target, so married rates match the census at every age by construction.

The useful check is the things that go with marriage. Women aged 30-39 with no steady partner (single or "casual relationship") come out at 29%, against 23% in the GSS. At 40-42 it's 25% vs 22%. Men 30-39 match (29% vs 31%). Young men come out too partnered: 46% of men 18-29 are single, against 58% in the GSS.

Having kids runs low, more so for women. That's worth remembering for any finding about women over 30.

04Health

The survey asks "Do you have any of the following," with a tip to only check if it's "moderate to severe." That's self-identification, not a diagnosis. Population surveys measure diagnosable disorders, so each benchmark is a range: roughly the 12-month rate of moderate or worse cases at the low end, and the lifetime rate at the high end.

Weighting on depression, ADHD and autism does most of the work here. In the raw data, 67% of women ticked anxiety. It's a weighting target, held at 36%, the top of its range. Men's anxiety is a check, and it lands in range at 18%. OCD is the clearest miss: 8% of women and 5% of men tick it, against 2-3% and 1-2%.

• Hormonal birth control (cis women 15-42): 21.9%, right inside the NSFG range of 18-25%. It matches in most age bands too, except teens, who come in low (14% vs 22%).
• Spanked as a child: 79-85% depending on age, against 75-86% in a large YouGov poll.
• Childhood abuse: not usable as a check. BKS asks about physical and verbal abuse only after a gate question ("Were you, at least semiregularly, abused as a child?"), so it measures something stricter than the BRFSS "ever" questions. As expected, it comes in below them (physical: 13-16% vs 22-30%).

05Sex and porn

Paying for sex. 12% of hardweighted men aged 18-42 have hired a sex worker, inside the 6-15% range between the GSS (whose question also counts being paid) and an older ABC News poll. By age it's off: 18% at 30-39 and 22% at 40-42, against 7-9% in the GSS. Older men in BKS are much more likely to have paid for sex, and weighting upweights them.

Porn use is a weighting target, so it lands where the targets put it. The chart below shows those targets against the range of national estimates.

06Kink

This is the check that matters most, and the hardest to do well. Very few probability samples ask about kink, and none ask the BKS questions. So every comparison here is one BKS question against one question from another survey, chosen to be about the same thing. There are no combined measures.

The main benchmark is Herbenick et al. (2017), a US probability panel of 2,021 adults. It listed 50+ sexual activities and had people rate each one: very appealing, somewhat appealing, not appealing, or not at all appealing. The matches:

The wording still differs in one important way. BKS asks how arousing something is, and its kink section reminds people to "answer for your fantasies; we know what you do in real life may be different." Herbenick asks how appealing something is "as part of sex." Plenty of people find things arousing in fantasy that they wouldn't want to do, so even a perfectly representative BKS sample should come out somewhat above Herbenick. Pushing the other way, BKS only shows these questions after you tick the parent category (e.g. "Sadomasochism") as something that arouses you, and tells you to tick only what you're "definitely into." Everyone who didn't tick it counts as not aroused.

The BKS rating scale is Not / Slightly / Somewhat / Moderately / Very / Extremely arousing. Below, "Very" or "Extremely" is compared to "very appealing," and anything from "Slightly" up to "very" + "somewhat appealing."

The two BKS spanking questions disagree with each other. "I find spanking to be" (only asked after ticking Sadomasochism) gives 28% of men and 36% of women with any arousal. The sex-acts checklist, asked of everyone ("Which of the following sex acts do you find significantly erotic?"), has 62% and 70% ticking Spanking. Same sample, same weights, same topic, and a 2× difference. Where in the survey a question sits matters about as much as who took it.

The Herbenick sample also includes people over 50 (average age 47), who find most of this less appealing. That makes BKS's 18-42 look worse than it is.

Allowing for wording and under-reporting

The comparison above takes the benchmark at face value. Two things say a fair benchmark for the BKS question should be higher:

• Wording. More people fantasize about something than find it appealing to actually do. In studies that asked both questions of the same people, fantasy or desire runs about 1.5-1.6× "appealing" for bondage, pain, exhibitionism and voyeurism, and about 1.45× for spanking (Joyal & Carpentier 2017, a Quebec population sample; Joyal et al. 2015; Herbenick et al. 2017). Among people with any interest, about a quarter are at the strong end on either kind of scale, so the same factor is used for "very."
• Under-reporting. Herbenick's panel is private but not anonymous: the survey company knows who its panel members are. When people answer more anonymously, reports of stigmatized sexual things go up. Same-sex experience rises about 1.5× when a phone interviewer is replaced by an automated self-interview (Villarroel et al. 2006, a randomized study), and 1.3-1.5× when British survey respondents answer again on the web (Burkill et al. 2016). In Quebec, lifetime masochism came out 1.7× higher online than by phone (Joyal & Carpentier 2017). Nobody has measured this for kink in a private web panel, so the factor used here is 1.15×, with a range of 1-1.5×.

Multiply the two and you get an adjusted benchmark. For example, 8% of men in Herbenick found being tied up "very appealing." Adjusted, a representative, anonymous survey asking the BKS question should find about 15% (range 10-24%). Hardweight shows 29%.

Two things come out of this:

• At "any arousal," the gap mostly disappears. Light bondage and exhibitionism land within about 15% of the adjusted benchmark, and receiving pain about 1.3× above it. Spanking and voyeurism land below it. For spanking that fits the gate: you only see the question after ticking Sadomasochism. The BKS voyeur item is also narrower ("watching others sexually, publicly or without permission") than Herbenick's "watching others do sexual things."
• At the strong end, BDSM still runs high. "Very arousing" light bondage and receiving pain stay about 2× the adjusted benchmark (1.8-2.4×), and spanking about 1.5×. Exhibitionism is 1.3-1.5×. Even with the most generous adjustment, bondage and pain for women stay 1.3-1.4× high.

The same pattern shows up across every kink benchmark we could match one item to one item: 191 matches to 119 questions from 15 surveys in Europe, Canada and the US.

BKS hardweight, ages 18-42; US respondents for US benchmarks, all countries for the rest. "After" divides by the central wording factor (1.6× for appeal, 1.3× for behavior, none for arousal questions) and the 1.15× reporting factor. Most benchmarks include people up to 60-90, who report less of everything, so these ratios lean high. Sources include Herbenick et al. 2017, Bártová et al. 2021, Joyal & Carpentier 2017, YouGov GB 2025, Paarnio et al. 2022 and Project SEXUS (Denmark).

So once you allow for how the questions differ, the hardweighted BKS gets the share of people with some interest in these things about right. What it still overstates is how many are strongly into them, by about 2×. That's what you'd expect if the survey draws in people whose interest is intense, which no weighting on demographics, health or porn use can fully see.

Both factors are rough. The wording factor comes from a handful of studies, and for group sex and fetishes it rests on comparisons across different samples. The under-reporting factor has never been measured for kink; it's borrowed from research on same-sex experience and partner counts. The bars in the chart use the low and high ends of both.

Against a question about arousal

Bártová et al. (2021), a Czech national quota panel of 10,044, does ask about arousal. It describes each interest and asks "How sexually arousing would you find it?" from 1 ("It is unpleasant to think about it") to 5 ("Highly arousing"). Three of its items have a single BKS match.

Does the gap grow with age?

Herbenick doesn't report appeal by age. The only age-split benchmark with a single matching item is a 2025 YouGov poll of British adults (opt-in panel, n=8,110), which asked how often you'd want spanking in your sex life. It asks about each role separately. BKS asks one question for both.

BKS comes out at or below YouGov at every age, and the gap doesn't widen with age. YouGov's base includes people who aren't sexually active and the 11-23% who chose "prefer not to say," so its numbers are, if anything, low. On this one item, BKS isn't inflated, and there's no sign that older respondents are more selected.

What survives

Among the four "very" items, the sex differences are similar in both sources:

Women are more into receiving pain in both sources, and men are more into being seen in both. Comparing kinks to each other, or men to women, is safer than reading off what percent of people have a kink.

07Can the weights stretch?

A different way to test the weighting: pretend one target was never measured, re-run the weighting on the rest, and see how much of that target's raw gap gets closed anyway. If the others close nearly all of it, weighting generalizes well to related questions. If they close little, each target is only fixing itself.

So the realistic expectation for a question the weights never touched is that weighting fixes about half of its bias, or less if it isn't related to any weighting target. Porn use is the clearest case. The other 17 targets close only 11% of its gap for adults and 4% for teens, which is why it has to be a target in its own right. That fits the kink results above: the share with any interest comes out about right, but the strongly interested are still overrepresented, and nothing in the weighting can see that directly.

08How to use hardweight

• Fine for: demographics, family background, belief, most health questions, sex history. These mostly land within the range of real-world estimates.
• Kink levels, with care: against benchmarks as published, hardweight runs about 1-2.5× at "any arousal" and 1-4× at "very." Allowing for wording and under-reporting, the share with any interest is about right, while the share strongly into BDSM items is about 2× too high. Treat "very" numbers as upper bounds.
• More trustworthy: comparisons. Which kinks are more common than others, and men vs women, hold up better than the levels.
• Ages 14-42 only. For anything about people over 42, use Softweight or Raw, and remember they aren't corrected for kink.
• Be careful with paying for sex after 30, and with kids in the late 30s.
• Porn use, marriage, religiosity and anxiety in women are weighting targets, so hardweight can't tell you anything about their overall rates. It's fine to use them to split other questions.

09Methods and caveats

• Data: the shipped hardweight file (564,964 rows before cleaning, ages 14-42; 372,614 US). "Raw" is the unweighted pool hardweight was built from (all respondents aged 14-42 in the same countries with complete demographics: 1,044,164 rows, 693,819 US). CIs use the Kish effective sample size.
• US only. Nearly all good benchmarks are American. Weights were raked on all countries together, not per country, except religiosity and anxiety, which are US-only targets.
• Ages. Every check is clipped to ages 42 and under on the BKS side. Many benchmarks cover wider age ranges (Herbenick goes to 91), which is noted in the table below.
• Benchmarks were taken from the published tables or computed from public microdata with survey weights (NSFG 2017-19 and 2022-23, GSS 2018-2024, BRFSS 2021-25, NHIS, NHANES, ACS 2024, NCHS birth files 1980-2010, Relationships in America 2014). When sources disagree, the chart shows the range. Kink benchmarks: Herbenick et al. 2017 (PLoS ONE 12(7):e0181198, Table 5), Bártová et al. 2021 (J Sex Research 58(1)), YouGov GB 2025.
• Kink comparisons are item to item. No combined measures.
• Leave-one-out re-implements the hardweight rake exactly (same targets, order, iterations and cap 30); the re-implementation reproduces the shipped weights with r = 0.9998. Gap = total variation distance from the target.
• Kink adjustment. Wording factors (fantasy or desire vs "appealing") come from Joyal & Carpentier 2017 (J Sex Research 54(2)), Joyal et al. 2015 (J Sexual Medicine 12) and Herbenick et al. 2017. Reporting factors come from Villarroel et al. 2006 (Public Opinion Quarterly 70(2)), Burkill et al. 2016 (PLoS ONE 11(2)), Coffman et al. 2017 (Management Science 63(10)) and Joyal & Carpentier 2017. The broader table uses every benchmark in our kink file that matches a single BKS item, taking the closest BKS item for each.
• Wording never matches exactly. Every check has some mismatch: self-labeling vs diagnosis, arousal vs "appealing," "penetrative sex" vs "vaginal sex," presence vs co-residence. The directions are noted in the table below. For kink, wording probably explains part of the gap (see section 06).
• Late-added questions have smaller samples: handedness (May 2026), kids (Dec 2025), parents' ages (Mar 2026), current religion (May 2026), paying for sex (Nov 2024), birth control (Nov 2024).
• Porn, marriage, religiosity and anxiety targets. Porn (none = "I don't," "less than once a year" or "once a year"; some; weekly or more) by sex and age. Marriage by sex and age, US vs elsewhere. Religiosity (37% not religious, 11.4% very religious; GSS) and anxiety in women (36.4%) for US adults. Each is held within its own cells, so the other targets aren't disturbed. Because ages stop at 42, the sex × politics targets were refit to the younger age mix.
• Quirks of the weighting. Respondents whose partner count couldn't be read as a number (12,521, about 2%) end up with about a third of the average weight, because the rake computes shares over all rows. And country isn't a target. Hardweight is 72% American, while Americans are about 45% of this population. Every check here is within the US, so this doesn't affect them, but all-country numbers lean American.

10Every check

US respondents, ages 14-42. Raw = unweighted pool; Hard = hardweight with 95% CI. on target close off weighting target (not counted). Childhood-abuse rows aren't counted either, because the BKS question is gated.