Pairwise comparisons for Estimated Marginal Means in Linear Mixed Models?
Hi,
I'm using JASP 0.98.1 and have a question about the Estimated Marginal Means output in the Linear Mixed Models module.
After fitting a model, I can obtain the Estimated Marginal Means table, but I cannot find an option to perform pairwise comparisons between the estimated marginal means (e.g. Pre vs Post, Pre vs Follow-up, Post vs Follow-up) with adjusted p-values.
Have I overlooked this option, or is it not currently available in JASP?
Many thanks!
Per the JASP lover😍
Comments
Hi again, just a quick clarification: when I teach LMM to our students, I actually have to switch to SPSS or Jamovi because the pairwise EMM comparisons aren’t available in JASP. It would be wonderful to stay in JASP the whole time.
Per 😍
I 100% agree with you that this is missing. In the data library, they have an example for LMM, and they use contrasts (i.e. you would have to code your pairwise comparisons manually):
setting up contrasts for pairwise comparisons is "easy" (just -1 and +1 for the two means you want to compare and 0s elsewhere), but this is done manually and quite inconvenient. But seeing as that's how they do it in their example, I assume that's how you have to do it in JASP.
I also remember that decomposing interactions in JASP is sometimes not trivial (I think in the regression module).
@patc3
Thank you for the swift response. I really appreciate it. I agree, this is not a very convenient solution. Hopefully JASP will implement this functionality soon.
I am promoting JASP at my university and have started using it in several statistics courses. It is a great piece of software, but features like this are important if it is to become a fully adequate alternative for teaching and applied research.
Thanks again!
Per the JASP lover 😎
@patc3
What was the name of the dataset you refererad to in JASP?
All the best
Per😎
it's the Larks and Owls dataset in the data library in the mixed models folder:
Per, have you made a feature request on our GitHub page?
Just an FYI, I don't think SPSS has ever been adequate for ANOVA-type analysis (I haven't looked at the LMM case). This is because SPSS doesn't do pairwise comparisons relevant to interactions.
For example, if you have a two-by-two design producing Cell A, Cell B, Cell C, and Cell D, SPSS cannot produce post-hoc tests comparing each cell to every other cell. So no post-hoc tests are produced.
Another example would be a two-by-three design (e.g., two 'conditions' by three 'sessions'). SPSS will produce only three post-hoc tests, one comparing the mean of Session 1 to the mean of Session 2, one comparing the mean of Session 2 to the mean of Session 3, and one comparing the mean of Session 1 to the mean of Session 3. That's it.
R
Also, there's a current JASP bug report about contrasts: https://github.com/jasp-stats/jasp-issues/issues/4472
R
@EJ Hi. Yes, I have made a GitHub feature request.👍
Per😎