Timeline for REML or ML to compare two mixed effects models with differing fixed effects, but with the same random effect?
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 7 at 0:04 | comment | added | Ben Bolker | Hmm, yes, that makes sense. | |
| Jan 6 at 23:46 | comment | added | Nate | Thanks! Would just like to add that I thought this answer contradicted this with "...and when comparing models with different fixed effects (as here) where those terms aren't fully penalized, then you shouldn't use REML so ML is the way to go (at least for the comparisons)". However, I think the key phrase is "fully penalized" and this thing about not using REML only applies to parametric terms (i.e. "process" variable in the OP ->stats.stackexchange.com/questions/578023/…). | |
| Jan 6 at 23:23 | comment | added | Ben Bolker | As the answer to that question says: "when using REML smoothing selection, at least according to my reading of ?logLik.gam, the AIC is using the penalized maximum likelihood estimates regardless of whether smoothness selection is done using ML or REML. So the general problem of comparing models using likelihoods from models fitted using REML doesn't seem to apply to GAMs fitted by {mgcv}." (This reasoning should apply to any likelihood-based model comparison ...) | |
| Jan 6 at 22:42 | comment | added | Nate | Thanks @BenBolker, I saw this post! Not sure if it means one can compare ANY two models with REML in AIC, or just those ones where the fixed effects are the same. In Pedersen, E. J., Miller, D. L., Simpson, G. L., & Ross, N. (2019). Hierarchical generalized additive models in ecology: An introduction with mgcv. PeerJ, 7, e6876. doi.org/10.7717/peerj.6876, they used REML and seemed to include a different fixed effect in "bird_modGS" vs. "bird_modG", namely the variable "species". | |
| Jan 6 at 22:30 | comment | added | Ben Bolker | @Nate, See stats.stackexchange.com/questions/591153/… | |
| Jan 6 at 21:27 | comment | added | Nate | Does anyone know if this holds for generalized additive (mixed) models as well? | |
| Nov 6, 2020 at 22:32 | history | edited | Ben Bolker | CC BY-SA 4.0 |
added 307 characters in body
|
| Sep 8, 2020 at 12:58 | comment | added | Myriad | @BenBoker following up on your response, would it be acceptable to use the anova (LRT) to compare model these models with different fixed effects if REML = FALSE? | |
| Sep 26, 2014 at 12:04 | comment | added | It Figures | Thanks @janhove, AdamO and Ben Bolker. I also found this link from Aaron to be helpful in answering this question. It says, "The REML likelihood depends on which fixed effects are in the model, and so are not comparable if the fixed effects change. REML is generally considered to give better estimates for the random effects, though, so the usual advice is to fit your best model using REML for your final inference and reporting." | |
| Sep 26, 2014 at 11:43 | vote | accept | It Figures | ||
| Sep 25, 2014 at 20:09 | history | answered | Ben Bolker | CC BY-SA 3.0 |