Timeline for In regression, can an effect be random if all group-level population units are sampled (but not all subsamples)?
Current License: CC BY-SA 4.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 8, 2023 at 7:00 | comment | added | Estimate the estimators | I posted my question here, am a bit confused: stats.stackexchange.com/questions/630731/… | |
| Nov 8, 2023 at 2:36 | comment | added | Estimate the estimators | I will review. I guess to me a fixed effects model makes the same specification (varying association by cluster); the difference I assume is in the interpretation of the coefficient and in the modeling assumption of the random effects as jointly multivariate normal to other effects in the model and in the effects having been sampled from a population as opposed to being a set of groups that are pre-selected and not random. I plan to pose a question on this tonight as I am still struggling with the distinction. | |
| Nov 7, 2023 at 23:19 | comment | added | Shawn Hemelstrand | This answer provides a simulated example of random intercepts versus random slopes. The main idea is if you have a random effects cluster (such as subject) and you believe the association between the IV and DV varies by cluster, then a random slope model may be appropriate. The Harrison article also goes into this distinction in a readable way. | |
| Nov 7, 2023 at 19:25 | comment | added | Estimate the estimators | I am reading all your posts and this is gold, thank you. Do you have a good post on when to do random slopes vs. fixed effects? This trade-off seems to get at the heart of understanding when random effects are useful from a variance explanation standpoint. | |
| Nov 7, 2023 at 8:23 | history | answered | Shawn Hemelstrand | CC BY-SA 4.0 |