Timeline for What is the right way to analyze a nested design in R?
Current License: CC BY-SA 3.0
14 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 10, 2014 at 16:11 | comment | added | Livius |
@SpookyFM @jona The documentation for clmm() on CRAN is out of date, make sure to take a look over at the package on R-Forge. See this message thread from the R special interest group for mixed models.
|
|
| Jul 22, 2014 at 14:09 | answer | added | jona | timeline score: 2 | |
| Jul 22, 2014 at 7:30 | comment | added | SpookyFM | @jona Thanks for the reference to Stan which I did not know. I am unsure if I am willing to learn yet another language just yet but maybe your code can still be enlightening to me. I have also updated the question statement so if you could provide any feedback on my model specification (especially your thoughts on what the maximal model would look like in this case), I would be very grateful! | |
| Jul 22, 2014 at 7:29 | history | edited | SpookyFM | CC BY-SA 3.0 |
Question update
|
| Jul 21, 2014 at 11:55 | comment | added | jona | I can share my Stan code with you if you want to try that, although I can't promise it's any good. | |
| Jul 21, 2014 at 11:17 | comment | added | SpookyFM | @jona You are right, it computes and even converges for me (which lmer doesn't) but it only gives the parameter estimates, no standard errors, z values or probabilities... However, it does give the error message "Variance-covariance matrix of the parameters is not defined" which I can't interpret. | |
| Jul 18, 2014 at 12:43 | comment | added | SpookyFM | Hm, it says so in the documentation. Maybe it's outdated? I'll give it a try, thanks! | |
| Jul 18, 2014 at 10:21 | comment | added | jona | Wait ... are you sure clmm doesn't do random slopes? Because this code worked for me (although it failed to converge): cmod = clmm(a ~ OA + (OA|S), df, link = c("logit")) | |
| Jul 18, 2014 at 10:13 | comment | added | jona | I just analyzed a Likert ratings experiment using ordered logistic regression in Stan, which let me freely estimate slopes for subjects and items. However, setting up a Stan model is a somewhat involved procedure. Alternatively, binomial regression in lmer gave fairly similar results compared to the ordinal regression - maybe you can try doing the biggest mixed model you can build in clmm and compare its results to binomial and/or linear regression using lmer. | |
| Jul 18, 2014 at 9:44 | comment | added | SpookyFM | @jona Thanks for the advice! I had found that package too when googling but wasn't sure about it. I have now tried it but, referring to your comment in the emudrak's answer, it doesn't let me specify random slopes... Do you know any solution for this? | |
| Jul 17, 2014 at 16:35 | comment | added | jona | The package "ordinal" does Ordinal Logistic Regression with mixed effects (clmm), which may be apt for Likert ratings. | |
| Jul 17, 2014 at 16:18 | answer | added | emudrak | timeline score: 0 | |
| Jul 17, 2014 at 9:51 | review | First posts | |||
| Jul 17, 2014 at 9:59 | |||||
| Jul 17, 2014 at 9:49 | history | asked | SpookyFM | CC BY-SA 3.0 |