Statistics: mean v. median

[Function]

Hello everyone

 

Again for my paper, I was wondering for a specific variable, if it would be more interesting to report mean or median:

 

The duration of pregnancy is not distributed conform the Gauss normal distribution (significant Kolmogorov-Smirnov test, p < 0.001).

When comparing the durations between 2 sample sizes (they do differ significantly when it comes to pregnancy duration, p = 0.002), is it then more interesting to mention mean (SD) or median duration? So do I say the mean or median durations do significantly differ?

 

Or doesn't this matter?

 

Another problem: apgar score measured at 1 minute v. at 5 minutes postnatal: they are both ordinal variables. It is of course not interesting to mention the mean apgar score, but can it be advised to mention the median for both?

 

Thanks!

 

F

Edited October 27, 2016 by Function

[Prometheus]

So you are comparing 2 samples with a null hypothesis that they are drawn from the same, non-Gaussian, distribution. You will then be performing a non-parametric test on this hypothesis (barring anything weird or wonderful). Generally, non-parametric tests test differences in medians, so this is what you should report, along with inter-quartile ranges to describe the spread of the data. You should first consider transforming the data (log or root transformations are most common) and if this data is normally distributed perform standard parametric tests.

 

How much of this stuff are you covering on your course?

 

The Apgar problem is quite interesting; there's not a consensus. Some argue ordinal data is qualitative and the mean should never be calculated and some argue that reporting the mean, so long as certain criteria are met, is fine. Unless you know exactly about these criteria, median is the safe bet. It will be interesting to see which school of thought you are taught - let me know.

[Function]

So you are comparing 2 samples with a null hypothesis that they are drawn from the same, non-Gaussian, distribution. You will then be performing a non-parametric test on this hypothesis (barring anything weird or wonderful). Generally, non-parametric tests test differences in medians, so this is what you should report, along with inter-quartile ranges to describe the spread of the data. You should first consider transforming the data (log or root transformations are most common) and if this data is normally distributed perform standard parametric tests.

 

How much of this stuff are you covering on your course?

 

The Apgar problem is quite interesting; there's not a consensus. Some argue ordinal data is qualitative and the mean should never be calculated and some argue that reporting the mean, so long as certain criteria are met, is fine. Unless you know exactly about these criteria, median is the safe bet. It will be interesting to see which school of thought you are taught - let me know.

 

We're not taught log or root transformations yet, but hey, I might as well look into it and see what it gives. I'll let you know later.

 

I think I'll go with the median of the apgar scores (interesting: all medians are 9, big surprise). This is what we were taught: that medians can also be used for ordinal data, but of course not for nominal data, given the presence of ranks in ordinal variables

 

EDIT: I performed log and sqrt transformations on the duration of pregnancy, but the transformed data are also not distributed normally; shall continue with nonparametric test.

 

I'd share my final report, but it's in Dutch, so ...

Edited October 27, 2016 by Function