Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Supported by

NAN in post-hoc test

Dear JASPers, I am doing a repeated measures ANOVA analysis, but for some reason many (but not all) of my post-hoc tests show a mean difference, a SE, a statistic, but then a NaN for the p-value (see below).

Could you please help me understand why there is a NaN?

thank you

martin



Comments

  • sorry, I now saw a similar post. I ran my analysis on 0.9.1, so I'll try it on 0.9.2 and re-post if needed (sorry!) :-)

  • Dear All,

    still got the NaNs above in 0.9.2. Any advice?

    Cheers

    Martin

  • Hi Martin,

    I'll ask the expert in our team. We are nearing a new version, so it would be great to see this fixed (if it is a bug).

    Cheers,

    E.J.

  • Hi Martin,

    Due to the Scheffe and Tukey corrections not being suitable for RM factors, we have decided to not display those and clarify this with a footnote in the table. However, I will change the way it is being displayed, since NaN is ususally an indicator that something went wrong in the analysis.

    For now, I would advise you to use another p-value correction, such as Bonferroni or Holm.

    Kind regards,

    Johnny

  • Hi All. I've encountered similar issue of NAN when running Games-Howell post-hoc on a one-way ANOVA using JASP 0.9.2. However, in my case this only seems to be associated with some values and with this particular post-hoc. All data has the same number of repetitions (only 2 data points for each group). As can be seen below, some pairwise comparisons come back with NAN while others are returning a value. Both Dunn and Tukey give values for all comparisons.


  • Hi HMSTAT,

    To calculate the p-value, JASP uses the ptukey function from R (see also here: https://rpubs.com/aaronsc32/games-howell-test).

    It seems that for most groups, the df is very low due to only 2 observations per group, which breaks the ptukey function. If you want, you can send me your data set and I will look into this a little bit more (you can send it to j.b.vandoorn at uva.nl), but I think your n per group is simply to low.

    Kind regards,

    Johnny

Sign In or Register to comment.