Timeline for Inconsistent results between linear model and linear mixed-effects model
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 13, 2017 at 12:44 | history | edited | CommunityBot | replaced http://stats.stackexchange.com/ with https://stats.stackexchange.com/ | |
| Jul 28, 2016 at 16:38 | history | edited | EdM | CC BY-SA 3.0 | clarified heuristic nature of t-test analogy |
| Jul 28, 2016 at 16:32 | history | edited | EdM | CC BY-SA 3.0 | clarified heuristic nature of t-test analogy |
| Jul 28, 2016 at 16:20 | comment | added | EdM | When you write random=~1|subject in your call to lme, you allow each subject to have its own intercept (value for $y$ when QM = 0) in the relation between $y$ and QM. If there are substantial differences among subjects in terms of that intercept, lme corrects for that and thus minimizes the residual variance in the model. The lm model has no way to take differences in intercepts among subjects into account and thus has much higher residual variance. The relation to paired t-tests was intended to be heuristic; I'll edit the answer a bit to make that clearer. | |
| Jul 28, 2016 at 15:10 | comment | added | bluepole | Thanks for answer. However, I don't see how the inconsistency is related to the issue of pairing. Essentially I'm focusing on the overall effect (or average effect between the two emotions) of 'QM' with the lme model, not the contrast between the two emotions. Similarly for those lm models. | |
| Jul 28, 2016 at 1:17 | history | answered | EdM | CC BY-SA 3.0 |