Timeline for Should I use a multilevel model if I have lots of observations?
Current License: CC BY-SA 4.0
8 events
when toggle format | what | by | license | comment | |
---|---|---|---|---|---|
Mar 16, 2020 at 13:04 | comment | added | Erik Ruzek |
Possibly, @BenMann. If you go Bayesian, you need to carefully consider an appropriate prior on the intercept variance. You may in fact try a few different ones, however, as you note you have a lot of data and these models may take a while. You are probably ok if you stick with REML estimation (default in lmer ).
|
|
Mar 16, 2020 at 11:34 | vote | accept | Ben Mann | ||
Mar 16, 2020 at 11:34 | comment | added | Ben Mann |
Thank for your answer, Erik. Given I have a small number of groups, would it make sense to use Bayesian packages in R, such as brms and rstanarm which, from what I understand, are better at accounting for uncertainty with small group sizes? Would the computation simply take too long?
|
|
Mar 14, 2020 at 22:27 | comment | added | Erik Ruzek | Thanks, @IsabellaGhement! I added some clarifications about REML and small numbers of clusters. | |
Mar 14, 2020 at 22:21 | history | edited | Erik Ruzek | CC BY-SA 4.0 |
Provided further information on REML and Kenward-Roger.
|
Mar 14, 2020 at 21:30 | comment | added | Isabella Ghement | Very nice answer, Erik! Can you elaborate a bit more on what makes the use of MLM questionable in this situation and why you recommend REML (e.g., REML deals better with a small number of cities?). Thanks! | |
Mar 14, 2020 at 20:21 | history | edited | Erik Ruzek | CC BY-SA 4.0 |
Fixed mutate code
|
Mar 14, 2020 at 20:04 | history | answered | Erik Ruzek | CC BY-SA 4.0 |