3
$\begingroup$

When asking heterosexual men and women anonymously about the number of sex partners they had, the result is quite apparent.

The mean for men is absurdly higher than the median for women (e.g. here).

This obviously cannot be, there is a bias.

The first thing that comes to mind, is that men write a higher number than women.

I was thinking of normalizing the data by forcing the men distribution to have the same median as women.

However, this will not account for selection bias (for example if virgin men chose not to participate in the survey).

What is the common practice to cleanup / align this kind of sensitive data ?

Improve this question
$\endgroup$
8
  • 5
    $\begingroup$ Best practice is to seek to understand the data rather than to "clean it up" to conform to one's preconceptions. Consider a small closed society including $n$ women in which $k \ge 1$ women are involved in sexual relationships with every man and the other $n-k$ women are not in any such relationship at all. When $n-k$ exceeds $k$, the median for women is zero while the mean for men is $k$. A mixture of such cliques is a crude but plausible approximate model for real societies. $\endgroup$
    – whuber ♦
    Commented Oct 2, 2015 at 17:39
  • 2
    $\begingroup$ Why do you assume there is a lot of bias simply because the numbers are vastly different? It's no secret that men tend to have a much larger number of sex partners than women as measured by many national health surveys. One thing that seems obviously problematic to me about the results, is that there didn't seem to be a response option for 0 (i.e. the respondent didn't have sex with anyone). But this seemed to be problematic for both men and women. Keep in mind too that there are few responses at this time as so, your estimates aren't probably stable yet. $\endgroup$
    – StatsStudent
    Commented Oct 2, 2015 at 18:30
  • 4
    $\begingroup$ I think you might want to re-consider the title of this post :) $\endgroup$
    – C8H10N4O2
    Commented Oct 2, 2015 at 18:40
  • 1
    $\begingroup$ If the sample was big enough, how is it possible for men and women to have different means ? That is the reason I assume there's a bias $\endgroup$
    – Alejandro Rodriguez
    Commented Oct 3, 2015 at 5:58
  • 2
    $\begingroup$ "mean for men is absurdly higher than the median for women" --- did you really intend in your question to compare apples with oranges there? a difference between mean of one and median of the other might simply indicate skewed distributions. I assume you really intend to compare mean with mean or median with median, in which case you should amend your question. $\endgroup$
    – Glen_b
    Commented Oct 7, 2015 at 1:09

1 Answer 1

Reset to default
2
$\begingroup$

This TED talk has a plausible explanation. See ~5:00 in.

The gist of it is that women and men who have had many sexual partners use different counting techniques. Women tend to enumerate (and thus forget about certain partners, biasing the answer downward), and men tend to approximate (X per year for Y years, etc.) which for reasons not clear to me supposedly bias the answer upwards. (They may also round up -- in one particular survey, 80% of males' reported partner counts were multiples of 5.)

The comment from @whuber is certainly a plausible explanation of the difference in median -- but the skew can't explain a difference in mean.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.