Timeline for Multiple Nested Random Effects affecting a Mixed Model
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 15, 2019 at 8:40 | comment | added | Unai Vicente | I tested with and without the slope model and made an ANOVA on both fits and even though it was not significantly different I ended up with what I think is, for now at least, the most accurate expression of my experimental design. | |
| Aug 15, 2019 at 8:29 | comment | added | Unai Vicente |
1. lmer(electrode_response ~ TrialRep + (1+TrialRep|CoupleID/SubjectID) 2. lmer(electrode_response ~ participant_adjustment + (1+participant_adjustment|ID/AbsSuj) + (1+participant_adjustment|Trial_all/TrialRep) 3. lmer(electrode_response ~ divergence_withinparticipants + (1+divergence_withinparticipants|ID/AbsSuj) + (1+divergence_withinparticipants|Trial_all/TrialRep)
|
|
| Aug 15, 2019 at 8:22 | comment | added | Unai Vicente | I also tried putting everything transformed into Z scores but contrary to my beliefs, it did not affect the output. At the end, I ended up using the theoretical logic, I studied a mixed model tutorial [bodowinter.com/tutorial/bw_LME_tutorial2.pdf] that really helped me understand better the basis of it and I ended up, even though with a poor explaining of my variability, with these models: | |
| Aug 14, 2019 at 10:03 | history | edited | Unai Vicente | CC BY-SA 4.0 |
edited body
|
| Aug 14, 2019 at 7:56 | history | edited | Unai Vicente | CC BY-SA 4.0 |
deleted 2 characters in body
|
| Aug 14, 2019 at 7:50 | history | edited | Unai Vicente | CC BY-SA 4.0 |
added 27 characters in body
|
| Aug 14, 2019 at 7:45 | comment | added | Unai Vicente |
First of all, thanks for the input @RobertLong , I started doing what you suggested on removing the Block from the random nest structure and fitting it to the fixed effects and the model didn't converge warning with negative eigenvalues. Then I followed the suggestion of removing TrialRep as a random slope and it worked just fine, even though Rsq did not substantially get better. At last, I tried GLMMadaptive but I think I could not figure out how to fix two nested random structures looking at the documentation, I'm sorry I am not a statistician and probably is me not using it properly.
|
|
| Aug 13, 2019 at 14:40 | comment | added | Robert Long |
Try removing Block from the random structure and fit it as a fixed effect. Alternatively, or perhaps as well, try removing TrialRep as a random slope. You might also try using the GLMMadaptive package instead of lme4
|
|
| Aug 13, 2019 at 12:03 | history | edited | Dimitris Rizopoulos | CC BY-SA 4.0 |
deleted 38 characters in body
|
| Aug 13, 2019 at 11:30 | review | First posts | |||
| Aug 13, 2019 at 12:49 | |||||
| Aug 13, 2019 at 11:30 | history | asked | Unai Vicente | CC BY-SA 4.0 |