Timeline for Bootstrap for random effects logistic regression to get CI for difference in proportions
Current License: CC BY-SA 4.0
13 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| S Jul 1, 2021 at 21:55 | history | bounty ended | Björn | ||
| S Jul 1, 2021 at 21:55 | history | notice removed | Björn | ||
| Jul 1, 2021 at 13:37 | answer | added | EdM | timeline score: 4 | |
| Jun 25, 2021 at 18:00 | history | tweeted | twitter.com/StackStats/status/1408485170546724864 | ||
| S Jun 24, 2021 at 7:12 | history | bounty started | Björn | ||
| S Jun 24, 2021 at 7:12 | history | notice added | Björn | Draw attention | |
| Jun 23, 2021 at 14:49 | comment | added | Frank Harrell | Richard McElreath's amazing book Statistical Rethinking has a lot of good wisdom about what to do with mixed effects models once you've estimated the parameters. The book is all Bayesian but may give you some frequentist ideas too. | |
| Jun 22, 2021 at 12:55 | comment | added | Björn | Thanks, @FrankHarrell. I implemented a Bayesian option, seems to work well (now added to question). But, for the Bayesian approach I wonder: would you simulate a new population (random effects drawn based on MCMC samples for the hierarchical scale parameter and covariates distribution using the observed values) or - as I did - use the MCMC samples for fitted random effects (=these patients)?! Is that just philosophical, or is the 1st more dependent on assumptions & 2nd less so (of course still applies shrinkage per normal random effect)? Also still interested how to bootstrap properly. | |
| Jun 22, 2021 at 12:47 | history | edited | Björn | CC BY-SA 4.0 |
Added Bayesian option as suggested by Frank
|
| Jun 22, 2021 at 12:28 | comment | added | Frank Harrell |
BTW logit and inverse logit are built into R and your inverse logit is a terrible implementation. Use plogis and qlogis. But on the bigger picture you are moving towards cruder approximations when you go to the cluster bootstrap. Better would be to use a Bayesian random effects model and get exact inference on any quantity of interest.
|
|
| Jun 22, 2021 at 12:22 | history | edited | Björn | CC BY-SA 4.0 |
Adding parametric bootstrap results
|
| Jun 22, 2021 at 11:03 | history | edited | Björn | CC BY-SA 4.0 |
added simulated example data and code for bootstrapping records
|
| Jun 21, 2021 at 23:53 | history | asked | Björn | CC BY-SA 4.0 |