« What Evidence Filtered Evidence? | Main | Rationalization »

September 30, 2007


It suggests it, but only very weakly. The strangest thing is that it shows that either some people tell the truth and then claim to have lied or that some people lie and then admit to having done so under the slightest of pressures. It basically doesn't even suggest that everyone who lies admits to having done so instantly. Also, I don't know what epistemic class the respondents belong to. It isn't clear to many people, and is in fact a point of serious philosophical contention, whether the fact that dreamers think themselves awake impairs the ability of waking people to rationally and confidently know that they are awake.

The assumption the numbers have to be equal doesn't hold. Imagine that every male has two partners, one of which always happens to be Lady X. In this case the averages would be close to 2:1. If, more realistically, the number of partners follows a power law then power laws with different exponents would create differing averages.

The High School Prom Theorem is still true of course, but if there are rare high-promiscuity people that are hard to find in surveys or there is more bias for them we should expect the surveys to be mismatched. This is still insufficient to explain the discrepancies as far as I can see, so there are real biases *beyond* this.

Anders, we should expect means across men and women to be exactly the same, though variances and medians can differ, and so one might expect median sample means to differ for small samples. Anyone know the sample sizes for these surveys?

Michael, what evidence suggests others are "dreamers" about sex, while you are "awake"?

Certain Doubts

It seems like people argue this out fresh every time it comes up. There are various ways to reconcile the numbers that depend on things like prostitutes etc being rare or correctly marking a box like "over 20". Also it seems plausible that men or women or both lie about sex.

What would happen if the average mundane experiment got as much thought about methods as this topic has? How many of the results would change as a result?

In addition to the counterarguments raised by the other comments, Robin's conclusion probably applies much weaker, if at all, to people with an intact memory that don't remember ever having had sex.
Keeping track of the exact number of partners requires more memory than keeping track of the boolean variable indicating whether you have had any, and consequently there are more bits to get wrong.

Robin: With respect to sex, I just said that the evidence doesn't strongly support the existence of dreamers, just of persistent liars, and I'm aware that I have been one of those in this domain upon occasion, and can also easily think of occasions when I lied about other things, was called on it, and maintained my protests of innocence. Arguable across agents, I have my presence here and my history of successful implementation of the Agreement theorem with select agents who I do treat as epistemic peers to argue for my not treating all humans equally as epistemic peers, but I think I had good reasons for sometimes treating myself as a better epistemic source than at least some other humans before then, or even, as was suggested here earlier, as treating different people as belonging to different observer classes.

More generally I don't accept the assumptions behind your assertion that Uncommon Priors Require Origin Disputes, and, it seems to me, neither does Robin Hanson, the poster of "Disagree With Suicide Rock". Figuring out how this sort of situation should be handled and how impossible possible worlds should be handled seem to be among the essential functions of this blog, and little is to be gained by assuming that we have figured them out already.

One very useful tool, it seems to me, in doing this, is to ask what an idealized AGI would do rather than what we "should" do.

Ok, clearly a proposal like "Lady X" doesn't change the averages (once more: 99 males with 2 partners one being Lady X, 1 male with Lady X as partner; 99 females each with 1 partner, Lady X with 100, resulting male median 2, average=1.99 where female median=1 but average=1.99). However, it does suggest that only Lady Xs need to lie to generate a discrepancy in averages, whereas a lot of males would have to do so to generate an equivalent effect. It seems plausible off-hand that Lady X has a strong motivation to under-count her partners, and that some males will have a strong motivation to over-count theirs. Has somebody proposed a test which distinguishes between (or even finds the overlap between) these motivations as causes? Or one which rules them out?
(Of course the subsample size of Lady Xs would be much smaller than the overall sample size.)

(Of course the subsample size of Lady Xs would be much smaller than the overall sample size.)

Which means you could get the same effect even if all Lady Xs were honest, if they're rare enough and the sample is small enough.

Robin Hanson postulates a reporting bias. Recent research seems instead to support a sampling bias. It appears that men report substantially more sexual partners than women do, not because members of either sex consciously or unconsciously lie about the number of partners they've had, but because female prostitutes are underrepresented in the surveys which register those reports. This is what Devon D. Brewer and his collaborators claim in "Prostitution and the Sex Discrepancy in Reported Number of Sexual Partners" (Proc Natl Acad Sci U S A 2000 October 24; 97(22): 12385–12388). Here's the abstract:

One of the most reliable and perplexing findings from surveys of sexual behavior is that men report substantially more sexual partners than women do. We use data from national sex surveys and studies of prostitutes and their clients in the United States to examine sampling bias as an explanation for this disparity. We find that prostitute women are underrepresented in the national surveys. Once their undersampling and very high numbers of sexual partners are factored in, the discrepancy disappears. Prostitution's role in the discrepancy is not readily apparent because men are reluctant to acknowledge that their reported partners include prostitutes.

estimate 1
Swedish study estimates 50 clients/prostitute/year

estimate 2
1985 west german study estimates licensed prostitutes at 13 clients/week.

estimate 3
2006 Croatia study estimates 5-6 per day.

The best study I found
American analysis of previous data. About 2% of american women said they'd ever sold sex. Estimate of 23 per 100,000 full-time equivalent prostitutes in the US population during 1970-1988, assuming Colorado Springs is typical of the USA. Studies at less typical sites give higher numbers.

"98 adult prostitute women reported a mean of 347 male sexual partners in the last 6 months (median = 103; interquartile range = 11–228; range = 1–5,401). The four prostitutes with the most partners (2,700–5,401) in this period reported very heavy cocaine/crack use. Their high level of reported activity is consistent with field observations of crack-addicted prostitutes (17). We doubled the 6-month mean to obtain an estimated mean of 694 male partners in the last 12 months for these women.

"This doubled figure is consistent with an estimate derived from prostitutes' reported number of male partners in the last 5 years. For the latter estimate, we assumed that the rates of entry into and exit from prostitution were equal for the Colorado Springs cohort of prostitutes, which implies that these women were, on average, halfway through their prostitution careers (18). Because prostitute women in Colorado Springs have a mean career length of 5 years (15), prostitute women in this sample most likely worked as prostitutes for only 2.5 of the last 5 years, on average. Therefore, currently active adult prostitutes' (n = 98) mean of 2,171 reported male partners in the last 5 years corresponds to an estimated mean of 868 male partners per year (2,171 partners per 2.5 years) for working adult prostitutes. We opted for the more conservative estimate based on doubling the 6-month mean."

The authors claim that prostitutes at these rates entirely account for the difference in reported numbers of sex partners between american men and american women. I didn't look at it carefully enough to make sure there were no flaws, but it looks plausible.

I'm a little doubtful. It was published in october 2000. If it actually solved the problem, would people have debated it online more or less continually for 7 years longer?

Thomas: "I'm a little doubtful. It was published in october 2000. If it actually solved the problem, would people have debated it online more or less continually for 7 years longer?"

I don't think this line of reasoning works for determining the quality of a finding. There have been pretty convincing evidence for evolution for many years now, yet people debate it fiercely online anyway. I guess one can refine it a bit by discounting the value of debate if there are emotional or other reasons to assume people are unlikely to want to change their minds or reach a consensus. But sex partner issues are likely in that category, so we cannot apply the debate ending heuristic here anyway.

But it would be interesting to analyse this more in detail. Just how much information about the quality of a finding can we infer from the existence or nonexistence of a debate several years afterwards?

I'll accept that lying prostitutes may explain the difference. I guess I didn't realize how common it was; maybe I'm missing out on a good thing and should look into it? :)

The other thing to note is that you might not be able to leave out the self-identified liars, because they might be a skewed sample. As in, assume the Lady X hypothesis (or prostitution) and then suppose that Lady X is more likely to lie than anyone else in the survey. This will skew the numbers. If she self-identifies as lying and you throw her out, this will probably skew the numbers even more.

Now suppose, as is plausible, that guys with few partners are more likely to lie than guys with many, and women with many are more likely to lie than women with few; then throwing out the self-identified liars could actually make the discrepancy worse, not better (depending on by how much they lie).

This reminds me of an episode of I believe it was Coupling (a British sit-com similar to, but better than Friends in my opinion) where one of the characters brings this up and says that when you ask a man how many people he's slept with you always have to subtract 2 or 3 from their number and when you ask a woman how many people she's slept with you always have to add 2 or 3 from their number. I would suspect that in surveys and studies that ask that sort of question you'll find much of the difference can be explained by the fact that men like to appear like they're studs and women like to appear as if they're not sluts because there exists an odd double standard in many cultures (America especially). I would be more interested to see the result of surveys like this in non-western cultures, like some remote native tribe in South America. Perhaps there are some places where that cultural double standard isn't as strong, is non-existent, or even the reverse. That might lead to less of a reason to lie when asked how much you get around.

Funny thing though, I myself (a man) often lower the number if asked by most people, however I have met girls who are probably lying when they say they've slept with absurdly large numbers of men (25+ !!). Guess not everyone follows the typical pattern.

"I'll accept that lying prostitutes may explain the difference."

It isn't just prostitutes that lie. This particular study suggested that prostitutes in Colorado Springs tended to have about a 5 year career, and during that time they would tend to be underrepresented in polls for various reasons. For example, if you poll by residence, the prostitutes tended to live in places that wouldn't get polled.

With an incidence of less than 3 in ten thousand, an average 5 year career would increase that to only about 3 in a thousand. Not that many studies include ten thousand women.

And the prostitutes on crack brought up the average considerably, and they were much rarer, and also have a much shorter life expectancy. So they would be severely underrepresented in polls.

My conclusion: academics are not getting enough sex.

The number of lifetime partners (for both genders) reported in Robin's original post seems low to me based on personal experience and what close friends have told me. If those numbers are accepted as reasonable by the commenters on this board... then I can only conclude that they need to get out a bit more.

In my experience, there are a few women who are in fact statistical outliers, and the ones I know are not prostitutes. You should call them escorts, by the way. The one friend who finally did progress from trophy wife to stripper to escort (in her 40s!) had passed the triple-digit number of sex partners a decade earlier. She's a welfare mom now, which provides some kind of moral, though I can't think exactly what.

Another potential loophole - men and women could be using different definitions of what counts as a sexual partner. If, for example, men considered act X sufficient for counting someone as a sexual partner and women did not, then that could also cause differences in results. If two people can disagree on whether they were partners, then the High School Prom Theorem doesn't hold even if everyone answers honestly and the sampling rate is 100%.

Doug S, yes, a study I ignored while I was finding the one that looked so good claimed that the women in their sample figured that it only counted when they could remember the man's name.

Doug S wrote:

Another potential loophole - men and women could be using different definitions of what counts as a sexual partner.
David Buss considers these and other potential explanations for the sex difference in reported numbers in the revised edition of The Evolution of Desire but concludes that "the real reason" of the discrepancies is the misrepresentation of prostitutes in sex surveys (p. 259).

misrepresentation => underrepresentation

Bob Knaus, I don't think those numbers reflected academics, it is my understanding they are taken from the population of the country as a whole (correct me if I'm wrong). Those numbers don't seem low at all to me and I know some pretty promiscuous people. I would say if you've reached triple digits you are probably a very very very very very very very very far outlier.

But no one is counting what you do by yourself!

The comments to this entry are closed.

Less Wrong (sister site)

May 2009

Sun Mon Tue Wed Thu Fri Sat
          1 2
3 4 5 6 7 8 9
10 11 12 13 14 15 16
17 18 19 20 21 22 23
24 25 26 27 28 29 30