It turns out that a quick review of the actual data reported in the study demonstrates that it would take a pretty impressive (not to mention highly, make that incredibly, unlikely) deviation at the top end of the number of partners for women to have the average number of partners come out even. The reported data breaks down this way: 0-1 partner: men 16.6%, women 25.0% 2-6 partners: men 33.8%, women, 44.3% 7-14 partners: men 20.7%, women 21.3% 15 or more partners: men 28.9%, women 9.4% The trend is pretty clear - women are overrepresented in the groups with the fewest partners and men are overrepresented (by a very wide margin) in the group with most partners. Nearly half the men reported 7 or more partners, while nearly 70% of the women reported 6 or fewer. Given these facts, to get the average number of partners for women to equal the average number of partners for men, you have to start making some pretty amazing assumptions, which break down more or less as that men are extremely disproportionately at the bottom ends of the ranges and women are extremely disproportionately at the top ends. If you just assume that everyone is at the middle of the ranges, men would average something like 80% more partners on average than women. And one more thing - statistics tells us that, as the sample size gets larger in a randomly-distributed sample the chance that the mean and the median deviate by a meaningful amount gets lower and lower. This survey covered more than 25,000 people over a period of 4 years. With a sample that big, the likelihood that the mean and the median differ enough for it to matter is pretty darned low. If you want to see the data for yourself, here's a link: <link>

My apologies the run-on paragraph - I keep forgetting that Slate did not bother to see how The Fray works in Safari and that the formatting I see on my computer is not what shows up when you submit a post.

The median and mean do not always get closer with large numbers. For example, you could look at the median and average income for ALL American households. That is a pretty darn big sample but the median is far lower than the mean. There are several factors that can contribute to this, not the least of which is that there is no anti-Bill Gates with a negative 3 billion income (they don't get counted anyway). Similarly, you cannot have a negative number of sex partners. So all the outliers in the distribution are on the promiscuous side (if the average is 7, then 0 is much closer to the average then 56). That is why the median is used more often in social context because we are usually not interested as much in the outliers. I sympathize with you on the browser/Fray inconsistencies.

It is true that gaps persist, particularly when one tail of a distribution is much longer than the other one, although I would guess that the range in the number of sexual partners is much smaller than the range in income levels.

Anyway, what I was trying to say is that you're less likely to have differences between the mean and the median that are related to random variations in the sample as the size of the sample goes up. In other words, if one percent of the population has an annual income of $1 million, the likelihood that those individuals will be over- or underrepresented in the sample - and therefore push the mean away from the true mean for the entire population - is much higher when you have a sample size of 100 than when you have a sample size of 10,000. (And, of course, when the sample size is below 100, you have a 100% chance that the 1% or smaller groups will not be represented at the right level.) In the end, this is just a variation on the law of large numbers.