Suggestions for anovaRM

Discuss the jamovi platform, possible improvements, etc.
Post Reply
arcaldwell49
Posts: 9
Joined: Wed Oct 11, 2017 7:03 pm

Suggestions for anovaRM

Post by arcaldwell49 »

Hey jamovi team,

I really appreciate everything you all are doing with the jamovi package, and thank you so much for everything you all have done so far.

With that said, I have a few suggestions that would really help the repeated measures anova function. Overall, this function works really well, but I do have a few suggestions.

1) Output for descriptives statistics (i.e. mean and standard deviation at each level and ability to collapse across conditions). Example from a 2-way (within-within) repeated measures ANOVA below.

2) In the post-hocs right now the pooled standard deviation is used. I don't think this inflates the error rate (I haven't look too deep into the literature), but I do not think this a recommended procedure https://www.researchgate.net/post/How_d ... ts_factors

3) Is there anyway you could add the option of having the multivariate output for repeated measures? In many cases, when the sample size is large enough, this procedure is preferred because you do not need to meet the assumption of sphericity.

4) Another analysis of interest would be trend (orthogonal polynomial) analysis. I don't use it often, but it has been of greater interest lately.

5) Also, sometimes I have a few pre-planned contrasts of interest rather than pairwise comparisons. I typically use SAS for this, and the example code is below.

Code: Select all

Title "Low PWV Standing";
proc glm data=WORK.Stif1 ;
model 	LowStandStand	Low1hrStand Low2hrStand
	 	LowStandWalk	Low1hrWalk Low2hrWalk = / nouni effectsize;

repeated trial 2, time 3 / MEAN nom;
manova h=intercept m = 
		0.5*LowStandStand + 0.5*LowStandWalk - 0.5*Low1hrStand - 0.5*Low1hrWalk,
		0.5*LowStandStand + 0.5*LowStandWalk - 0.5*Low2hrStand - 0.5*Low2hrWalk
	prefix=diff /summary;
run;
quit;
Again, great job on jamovi and the jmv package! I just want to provide some suggestions to improve what is already there. jmv is easily the best thing available in R for repeated measures. If you all can get these options put into the package I'll probably never need to use SAS or SPSS again!

If there is anyway I can help get this implemented I'm more than willing to help!
Attachments
Output from SAS proc glm for a repeated measures analysis
Output from SAS proc glm for a repeated measures analysis
descriptives output.png (20.88 KiB) Viewed 4456 times
User avatar
Ravi
Posts: 194
Joined: Sat Jan 28, 2017 11:18 am

Re: Suggestions for anovaRM

Post by Ravi »

Hi,

Thanks for taking the time to make these good suggestions. Here are my replies:

1) Yes, definitely something we want to add. It was a bit difficult to come up with a good way to let users make a selection of which descriptive tables they are interested in (main effects and different interactions). If feel that we are almost there so it should make its way to the RM ANOVA.

2) Thanks for the link, I'll definitely look into this! At the moment we use estimated marginal means using the emmeans package for the post-hoc tests in all ANOVA's. These marginal means are estimated using the model, so a bit different from normal pairwise t-tests (see https://cran.r-project.org/web/packages ... asics.html for a explanation of the procedure in R).

3) Can you make an issue about this on github so I don't forget about this? You can do that over here: https://github.com/jamovi/jamovi/issues.

4) Yeah, this would make a nice module. Do you want to give it a try? We have a tutorial on how to make modules over here: https://dev.jamovi.org/tuts0101-getting-started.html. Making a jamovi module is very similar to making an R package, so if you are familiar with R it's actually not that hard (and it's also a lot of fun). If you have any questions, we are very happy to give you some guidance.

5) Yeah, would like to add this but it has proven difficult to come up with a way to make this fit nicely in the current UI. I will keep pondering about this and hopefully we'll find a good solution at some point.

Cheers,
Ravi
Post Reply