ANOVA post-hoc t-value calculation

Comments

• EJ

  February 2020

  Dear AMa,

  Sorry for the super-tardy response. I've forwarded your issue to the team. I think this particular interaction term was still under development, but perhaps Johnny has recently added it...we'll let you know soon

  Cheers,

  E.J.
• JohnnyB

  February 2020

  Hi AMa,

  JASP uses the emmeans/lsmeans package to compute the posthoc results:

  We first create a reference grid and estimate the marginal means with the lsmeans function by providing the ANOVA model and which factors to use for post hoc tests. Then we use that reference grid and give it to the pairs.emmGrid function of emmeans, computing the results for each post hoc correction. Below is a snippet of the relevant JASP code - maybe that helps.

  resultPostHoc <- summary(pairs(referenceGrid[[var]], adjust="bonferroni"),

                           infer = TRUE, level = options$confidenceIntervalIntervalPostHoc)

  This also gives us the t-values.

  Does that answer your question?

  
  

  Kind regards,

  Johnny
• JLeborgne

  February 2020

  I was wondering something very similar! But it's about the fact that post hoc comparisons for a significant interaction (in RM ANOVA) give different results than Bonferroni-corrected paired-sample t-tests in SPSS. Why do you think it is? I don't think it's because JASP uses EMMs. All the RM ANOVA results are the same between JASP and SPSS in my datasets, but the analyses I ran specifically for significant interactions is what differs. Any ideas? :) Thanks