New York Times Misses the Median vs. Arithmetic Mean Distinction:

The New York Times reports:

In study after study and in country after country, men report more, often many more, sexual partners than women.

One survey, recently reported by the federal government, concluded that men had a median of seven female sex partners. Women had a median of four male sex partners. Another study, by British researchers, stated that men had 12.7 heterosexual partners in their lifetimes and women had 6.5.

But there is just one problem, mathematicians say. It is logically impossible for heterosexual men to have more partners on average than heterosexual women. Those survey results cannot be correct.

Well, heterosexual men generally can't have more partners on average than heterosexual women, where "average" means "sum up the partner counts and divide by the number of people." I say generally because there are possible boundary cases (e.g., if you asked all live men and women in the U.S. how many opposite-sex sexual partners they've had, and they answered completely accurately, the numbers wouldn't match because some of their partners are now dead, or are now outside the country, or because the partners are prostitutes and the study doesn't ask prostitutes or ask them enough), but it's generally so. As the article points out, "invoking women who are outside the survey population cannot begin to explain a difference of 75 percent in the number of partners, as occurred in the study saying men had seven partners and women four." (Perhaps prostitution might explain the difference, but the numbers that I've seen suggests that it doesn't suffice.)

But the medians may well be very different. Just as a sample, imagine a population with men A, B, C, D, E, and women P, Q, R, S, T, in which the sex partners map out this way:

ABCDE
PY
QY
RY
SYYYYY
TYYYYY

The median number of sex partners for the women is 1 (since the women's partner counts are 1, 1, 1, 5, 5); the median number of sex partners for the men is 3 (since the men's partner counts are 2, 2, 3, 3, 3). The arithmetic means for both are 2.6, since there are 13 male-female pairings; but the medians differ substantially.

Now this having been said, the bottom line is likely still correct; as I understand it, there is substantial reason to believe that men overreport their sexual partner counts and women underreport them. And it may well be that the 6.5 and 12.7 numbers are arithmetic means (the article doesn't say), which would be substantial evidence for this theory. But the "men had a median of seven female sex partners" / "[w]omen had a median of four male sex partners" data is extremely weak evidence, if evidence at all; and it surely can't be confounded with the "average" in the sense of arithmetic mean, which the article does.

Thanks to reader John Crawford for the pointer.

Oh My Word:
Stated another way, 10% of the guys get 90% of the nookie. And no, my nerdy academic Volokhites, you ain't in that 10%.

Just in case your day wasn't ruined by this fact of our sexually liberated world, http://www.hotchickswithdouchebags.com will be the salt for your wounds. Yes, it hurts. The pain, the horror.
8.13.2007 2:52pm
Daniel Chapman (mail):
Just another example of that much-lauded editorial diligence that makes the MSM superior to the internet...
8.13.2007 2:57pm
Tomm:
Given that they report the results as 6.5 and 12.7, I think it is safe to assume those are not median figures. Unless of course the study was using fractional fractional values for different types of sex, for example 1 for intercourse, but only 1/2 for fellatio, 1/10 for such and such, and so on.
8.13.2007 3:04pm
jc:
Oh My Word:

Actually, if the median for men is higher (which is what the first survey cited reports), then it would mean a relatively small portion of the women were getting a relatively large percentage of the nookie, not vice versa.

Anyway, I'll continue to run these calculations while you go get some play....
8.13.2007 3:08pm
arthur (mail):
The Times also missed the well-known fact that a much higher percentage of American men than women spent a few years in their late teens or early twenties outside of the United States, having sex with foreigners and serving in the military.
8.13.2007 3:11pm
Eugene Volokh (www):
Tomm: I expect that's right, but I don't think we can be sure, because sometimes one sees reports that use fractional medians (google the term and you'll see, though I'm not sure whether that's the standard term, or whether under any name it is a common practice).
8.13.2007 3:12pm
Benjamin R. George (mail):
Do we think the ‘heterosexual’ designation was a self-identification, or is it more likely to have been given a consistent operational definition.

After all, there are self-identified heterosexuals who have at one point or another had same-sex partners, and there are plenty of self-identified non-heterosexuals who've had heterosexual opposite-sex partners at one point or another (self-identified bisexuals are perhaps worth mentioning explicitly, although they aren't the only relevant group).

If self-identified heterosexual men were inclined to count their former same-sex partners for the total without counting them against their own heterosexuality, or if there were a large group of self-identified non-heterosexual women who were willing to sleep with self-identified heterosexual men, that could push the arithmetic means apart, even if we ignore death, migration, sampling error, and other effects.
8.13.2007 3:18pm
Benjamin R. George (mail):
Oh, never mind, the designations ‘heterosexual men’ and ‘heterosexual women’ were the Times' interpolations, from the looks of it. The actual data appears to be about opposite-sex pairings. That should make my proposed effect irrelevant.
8.13.2007 3:21pm
Sasha Volokh (mail) (www):
I think the 12.7 and 6.5 were exact numbers of partners.
8.13.2007 3:25pm
Hovsep Joseph (mail) (www):
Benjamin, It would be a really sloppy survey design to lump same-sex and opposite-sex encounters together. I'm sure the surveys refer only to opposite sex encounters.

However, a commenter on another site pointed out that there is evidence that men and women define sex differently and that this, in addition to deliberate under-/over-reporting, may partially explain the discrepency.
8.13.2007 3:25pm
Connie:
Reminds me of the memo we received from HR once, complaining about misuse of sick days.

"The median number of missed days is [x]. However, many employees take more days than that."

Yeah, I'm thinking that probably EXACTLY HALF take more than that. But it was from an HR person, so I didn't try to instruct her.
8.13.2007 3:33pm
Oh My Word:
jc, your mathematical point is well taken. Unfortunately, it fails to dull the pain, the pain!!
8.13.2007 3:36pm
frankcross (mail):
Amputees
8.13.2007 3:43pm
srg:
Connie,
It's even worse than that, The MSM is always saying things like "The average stock portfolio is down 10%, but some people are doing much worse than that."
I think I can remember their shock that 20% of the population is in the bottom quintile economically, but maybe I'm making it up.
8.13.2007 3:51pm
happylee:
I have never met a nonreligious man who did not over-estimate/over-state the number of lovers he bagged; nor have I ever met a woman who did not under-estimate/under-state the number of men who bagged her. Traditional truth syrums, such as teguila, will sometime work with the gals, but it only make the guys lie more.

Professor Bernstein might say that my anecdotal data cannot be extrapolated to the population at large, just as my success in using homeopathic remedies must merely be an illusion. And he may be right. But I wonder if anyone has believed it to be otherwise -- and the mathematical impossibility tends to support this, er, theory.
8.13.2007 3:55pm
Tomm:
The median 4/7 study, located here, gives some numbers for the distribution of the number of partners.

For men/women 16.7/25.0 claimed 0-1 partners, 33.8/44.4 2-6 partners, 20.7/21.2 7-14 partners, 28.9/9.4 15+ partners, and a crude median of 6.8/3.7. The first thing to note is that they use a fractional median, and the second is that the distributions are significantly different.

Since the NYTimes didn't actually sight the British research, it is a bit harder to find out what was going on there.
8.13.2007 4:02pm
Tomm:
Oh god, errors abound!

-cite, not sight
-the figures 16.7/25.0 ... figures are in percent, which I neglected to mention
8.13.2007 4:05pm
Harry Eagar (mail):
Shoulda quit while you were ahead, happy, but I'll match you anecdote for anecdote.

It's been my experience that people who claim cures by mens of mystical practices (religion, chiropractic, homeopathy, qi gong etc.) overestimate their illnesses.
8.13.2007 4:10pm
A.C.:
About two minutes of internet research has suggested an answer to the mystery. Women answer the question by counting up their past partners by name, an approach that might lead to underestimation (if they forget one) but is unlikely to lead to overestimation. Men are more likely to make a ballpark estimate of how many partners they think they've had, rather than to count up individual partners by name. This seems to result in overestimates -- when men do it, at least.
8.13.2007 4:17pm
Hattio (mail):
Slightly off-topic, but I remember hearing a advertisement for hooked on phonics or a similar program, back when growing illiteracy was the big problem. Advertisement claimed that over half (I think 60%) of school kids were reading below their grade level. The fact that it was national ad, and apparently successful, made me think that reading might not be where we should focus our attention.
8.13.2007 4:17pm
Justin (mail):
I don't think its weak evidence at all - a median and a (projected) mean that are substantially far apart, with no plausible explanation for such being far apart, would be indeed pretty strong (though hardly conclusive) evidence that the study was flawed.

If women indeed had a median of 6.5 and a mean of, say, 9, and men had a median of 12.7 with the same median, that would mean

1) that there is a large percentage of women who are having an incredible amount of sex with a large amount of partners, and/or a (extreme) relative paucity of women having 1 or 0 partners.

*and*

2) that, at the same time, there is a relatively *small* amount of men who are having large amounts of partners, and/or an (extreme) relative plethora of men having only 1 or 0 partners.

Unless these two statements seem reasonable (and on first glance, while either one by themselves seem reasoanbly possible, both statements being true seems highly unlikely), then even if the mean is undiscoverable by the survey, the difference in median is very strong evidence of the survey's flaw.
8.13.2007 4:31pm
Bill Poser (mail) (www):
Actually, it is perfectly possible for the arithmetic means of the number of sex partners of men and women to be different, so long as the numbers of men and women are different. Consider a population consisting of five men and three women, where every man has had sex with every woman and every woman has had sex with every man. The mean number of sex partners for the men is 3, for the women, 5.

The claim that the two means must be the same assumes equal numbers of men and women.
8.13.2007 4:41pm
jc:
Justin: this is the same as "Oh My Word" - you've got it exactly backwards. If the mean for women was much higher than the median, it would mean a relatively *small* percentage of them were getting a relatively *large* amount of the "nookie" (as in Eugene's example above - where the median for women is 1, but the mean is 2.6 due to two relatively active women).

Bill Poser: this, I think, is what Eugene was getting at with "possible boundary cases."
8.13.2007 4:54pm
Nels Nelson (mail):
Is everybody working with the same definitions? Perhaps women are more likely than men to define "sexual partner" or "sex" to require vaginal intercourse.
8.13.2007 4:55pm
jc:
Bill Poser: I retract what I said - I see it's a distinct point, and valid. (My apoglies.) But of course while a 75% differential is at least conceivable in the medians reported, it is, of course, inconceivable in the means given actual demographics.
8.13.2007 5:02pm
Bill Poser (mail) (www):
jc,

I agree - the magnitude of the disparity in this case is not likely to be accounted for by the difference in male and female populations. I was just making a mathematical point.
8.13.2007 5:04pm
Justin (mail):
No JC, I have it right. For the mean to be higher than the median, the people after 50% need to be numbers heavy relative to the people below 50%.

I.e., in a 10 person world, if you have a data pool of 1, 2, 3, 4, 5, 5, 10, 20, 30, 50), you end up with a median of 5 and a mean of 13. On the other hand, if you have a data pool of 1, 2, 3, 4, 5, 5, 5, 5, 5, 5 then you have a median of 5 and a mean of 4.
8.13.2007 5:21pm
DJR:
I've always wondered how they come up with these "lifetime" numbers. A survey of septugenarians would not accurately reflect the sexual activity of people born in the last 30 or 40 years, so a true proxy for a "lifetime" count would be inaccurate. Just as inaccurate, however, would be an extrapolation based on past activity, since a common pattern is to have relatively more partners when young and then settle down and get married. Of course, it would also be inaccurate, in a country with a 50% divorce rate, to assume that anyone who is married will not have any more partners, or what their behavior would be were they to divorce, not to mention how infidelity would alter the calculus.
8.13.2007 5:38pm
John Armstrong (mail) (www):
Over at God Plays Dice, there's a good discussion of this article by an actual statistician.
8.13.2007 5:42pm
jc:
Justin, I agree with your second post, but your first stated that for such a disparity to exist one precondition would be "a large percentage of women who are having an incredible amount of sex with a large amount of partners, and/or a (extreme) relative paucity of women having 1 or 0 partners." I guess we have different definitions of "large percentage" and "relative paucity." If 90% of women had had six or seven partners (average: 6.5) and 10% had had 31 or 32 (average: 31.5), then you'd have a mean of 9 and a median of 6.5. No women would have 1 or 0 partners; whether 32 partners is a "large amount" or 10% is a "large percentage" (large compared to what?) would then be at issue. I wouldn't call 10% a "large percentage," and I've known social circles where 30 partners is about average. At any rate, these are semantic points - I don't think we disagree on the math.
8.13.2007 5:54pm
jc:
Sorry Justin - I just reread this and realize you're right about the relative paucity. I should stop commenting now.
8.13.2007 5:57pm
Syd Henderson (mail):
Two other possibilities:

(1) The survey didn't specify the partners had to be human.
(2) Alchoholic amnesia.
8.13.2007 7:17pm
aces:
The survey didn't specify the partners had to be human.

I wonder how Amanda Grayson would answer...
8.13.2007 8:17pm
Isabel (mail) (www):
The 12.7 has to be an arithmetic mean; medians are by definition either one of the data points or halfway between two of them, and the data points are integers.

That being said, I think that the median/mean distinction isn't the big problem here; the problem is that people lie about their sex lives, and men and women do it in different ways.
8.13.2007 8:21pm
theobromophile (www):
Anecdotally, I'll cast my vote for the Madonna/whore theory: women have very few partners, or a lot of partners.

It would be interesting to see if age is a factor (could be wrong, but boys often have sex before girls do).
8.13.2007 9:00pm
Eugene Volokh (www):
Isabel: As I mentioned, it probably is the arithmetic mean, but medians are sometimes reported as fractional medians (and not just halves). Also, I agree that as to the bottom line the problem is probably misreporting by respondents -- but the specific supposed proof that the New York Times gave of this is not accurate as to the median study, which is the whole point of this post.
8.13.2007 11:17pm
Mike G in Corvallis (mail):
This all reminds me of a speech by Senator Ted Kennedy back in the mid-1980s. I heard the relevant sound bite on the radio -- he said it was a powerful indictment of American society that fifty percent of the American population has a below-average education.

(Can anyone supply the exact quote? I'd like to keep it and cherish it. He almost caused me to have an accident when I heard it on my car radio.)
8.14.2007 3:13am
Sparky:
Bill Poser is right -- even the means could be different, if the size of the two populations is different.

Of course, everyone assumes that the male and female populations are essentially equal (although this requires inductive reasoning, and is not a "proof" at all). However, there could be significantly different numbers of *heterosexual* men than *heterosexual* women, or vice versa.
8.14.2007 11:54am
Bruce Hayden (mail) (www):
I could definately see how treating partners you know the name of versus those you don't might skew the statistics. Of course, by now in my mid 50s, it isn't hard to say that I don't know the names of at least half my partners over the last 40 years. Whether I didn't know the names at the time is a totally different question.
8.15.2007 2:05pm