I know most people recommend R over SPSS and SAS--particularly for multi-level modeling--but I have yet to read WHY R is superior for these kind of analyses. I am preparing to run some HGLM and any information about specific advantages and disadvantages for statistical packages would be helpful.
I suggest a couple of reasons:
For what they have SAS and SPSS are probably fine. I know of no problems there. However, most new results are generally available in R, in development in R, or available from the original researchers in R. My own work in variable selection, robust logistic regression, cox regression and gradient boosting has been greatly helped by this fact. If you can do some R your pace is greatly improved. Best wishes, David Booth
Thank you David Eugene Booth! This makes a lot of sense. I am not [yet] literate in R. I am trying to weigh my options before I begin studying R and drafting the code for my analyses. Thank you again for your informative response!
In that case, let me share two books that I think all new users of R should have. The first is at this link. The second is attached. Best wishes, David Booth
https://www.amazon.com/Everyone-Advanced-Analytics-Graphics-Addison-Wesley/dp/013454692X/ref=sr_1_1?ie=UTF8&qid=1533577284&sr=8-1&keywords=jared+lander
Thank you so much! I definitely need "R for Dummies"--reflective of how I feel when writing code. Ha!
Have you figured out what package you would use in R? Would it be something along the lines of glmer, glmm, glmmADMB? Or is HGLM something different? Would SAS GLIMMEX be the appropriate proc?
If these are the packages or procedures you might be using, the following link gives some sense of what you'll be looking at. Overall, not too intimidating for the basic model fitting.
https://stats.stackexchange.com/questions/231980/glmer-in-r-to-recreate-glimmix-from-sas
I think probably the main difference to appreciate in using SAS and R is that a SAS proc is pretty much self-contained in that it will probably spit out everything you want it to, with some simple options. (I've never used PROC GLIMMEX, so I can't speak directly to that).
R often isn't like that. Often you have to go to other packages to get certain types of outputs. For example, for most models you can call a summary which will tell you the coefficients, and p-values for those. But if you want an anova-like table where the dummies for factor variables are grouped together, you may need to pass that model object to another package. For glmer model objects, that may be lmerTest::anova, car::Anova, or emmeans::joint_tests. Otherwise, you can build it yourself by comparing nested models, but this isn't as slick as the software spitting it out for you. If the model is supported by emmeans *, that gives you options for performing post-hoc separations of factor effects, which is the kind of thing I always want to do. If you want a pseudo r-squared for a model, you may need to dig or be creative.
If you have access to SAS, my honest advice is start with the example in the link above. Get it running in SAS and R, and get the software to output the things you want it to output. If you can get through that exercise without causing bodily harm to your computer, R may be a candidate for the task. Otherwise using SAS should be pretty straight-forward.
Finally, I'll give you a link to an example I think is pretty clear using lmer. (Mixed model, but not glmm). It will give you some sense of what working with R is like in these circumstances. http://rcompanion.org/handbook/I_07.html .
________________
* https://cran.r-project.org/web/packages/emmeans/vignettes/models.html
Hi Dorlisa!!! So great to hear from you! Thank you for the book recommendation. My colleague and I are reading through the "R for Dummies" book. I feel the same way--R is quite a jump from point-and-click programs. I'm going to put the time in to learn it since so many staticians recommend it. I'll let you know how it goes.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software