2
$\begingroup$

I'm using the Jonckheere-Terpstra test to confirm whether an interval variable, distributed across five levels of an ordinal variable, display a significant trend. I'm using the clinfun package, developed for R.

The resulting JT statistic is 8965.5 and has a significance of around 0.02 (depending on the value I choose for the "nperm" parameter). I'm not sure how the "nperm" value should be determined. The author of the clinfun package used a value of 5000 in his example, so I've been trying similar values.

The value of Kendall's tau for the same two variables is 0.098 and has a significance of 0.08.

I'm aware that these two procedures (Jonckheere's test and Kendall's tau) are closely related. However, I have now stumbled upon a number of (perhaps unreliable) sources claiming that the two procedures should display identical p-values.

Should the p-values be identical, and if so, how can I accurately determine the significance of Jonckheere's test?

Thanks.

$\endgroup$
  • $\begingroup$ In SPSS, for example, p-value for Jonckheere-Terpstra is the same as p-value for Kendall's tau-b in Nonparametric correlations procedure, but is somewhat different from Kendall's tau (b or c) in Crosstabs procedure. One has to dig why it is so; it appears that tau's significance can be computed in different ways. $\endgroup$ – ttnphns Oct 15 '14 at 15:33
  • $\begingroup$ Yes, one of the examples I found reproduced the same value on SPSS (using Nonparametric correlations). Unfortunately I cannot validate my Jonckheere test result on SPSS, since that feature is not part of my SPSS subscription... Anyway, do you think it's worth worrying about the validity of the R Jonckheere test result? $\endgroup$ – VDW Oct 15 '14 at 15:40
  • $\begingroup$ I'm sorry, I can't say anything about this test in R. In SPSS Algorithms document it must be described how p-value is found. $\endgroup$ – ttnphns Oct 15 '14 at 15:44
  • $\begingroup$ I have no answer to your question, but I'm interested in the way you calculate the significance of your JT statistics. Thanks! $\endgroup$ – user72558 Apr 2 '15 at 10:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.