5

http://www.cdc.gov/nchs/nsfg/key_statistics/n.htm

Median number of female sexual partners in lifetime, for men 25-44 years of age, 2006-2008: 6.1

Wouldn't it only be able to be x.5 or x.0?

user248640
  • 51

2 Answers2

2

You are right, then median of a sample is either one of the possible sample values or the arithmetic mean of two sample values. Hence if all sample values are integers, it is an integer or half-integer. Specifically, numbers like 6.1 are not possible as sample median of "number of (whatever)".

Hagen von Eitzen
  • 374,180
  • While I agreed with you, I can't imagine the CDC doesn't know how to do medians? – user248640 Jun 16 '15 at 17:46
  • @user248640: Perhaps they're running their raw data through a process that thinks they are a discretization of continuously distributed samples, and then tries to estimate the median of the underlying smooth distribution. – hmakholm left over Monica Jun 16 '15 at 18:10
  • 2
    @HenningMakholm That would mean they do not only use the wrong tool, but do not even notice that the output is not compatible with what they ask for, so they make, like, three statistics beginner's error at once? – Hagen von Eitzen Jun 16 '15 at 18:38
  • 5
    If you check the references, you’ll find that they are estimating medians (and providing standard errors for the estimates) from weighted data obtained from a complex sample design. – Brian M. Scott Jun 16 '15 at 19:16
1

It is likely that the sample is comstructed out of sets of grouped data rather than individual integer values only. Therefore this figure could legitimately be arrived at by linear interpolation.

For example, it is quite likely that the questionnaire would involve ticking boxes such as "zero", "between 1 and 5" or similar.

After all, would you expect that all respondents be able to give an accurate integer value as a response if that value is large?

David Quinn
  • 34,121