A comparison of nonparametric analyses conducted in SPSS, SAS, Stata, and R

[Bruce Weaver]

I just received the latest TOC alert for Behavior Research Methods, and this article caught my eye:

Researcher degrees of freedom in statistical software contribute to unreliable results: A comparison of nonparametric analyses conducted in SPSS, SAS, Stata, and R

I've not had time to read it yet, but judging from a quick glance, I wonder if the main "problem" might be that users do not always take time to RTFM* and therefore, do not understand what their software is doing? In any case, I thought some members of this forum might be interested.

Cheers,
Bruce

* RTFM = Read The Fine Manual

[Nick Cox]

Thanks for the reference. There are many possible reactions in general and in detail. One is that the tests concerned are often far from the state of the art which in many fields supported by Stata consists of modelling an outcome given predictors.

My eye fell on one topic I have played around with

Finally, unbiased calculations of sample skewness and excess sample kurtosis (where the expected value for kurtosis of a sample with a normal distribution is 0) are only used by SPSS and SAS. Despite the bias inherent in the use of population estimates, Stata does not provide any other built-in option for the way in which these measures of normality are estimated for a sample of data. Additionally, it is important to note that R has no built-in functionality for calculating skewness or kurtosis.

The fact is that there are many different measures of skewness and kurtosis and rather than fetishize particular estimators by far the most important thing is to appreciate that the variety and the limitations of all, as conveyed for example by articles in the Stata Journal.