Timeline for Intuition about parameter estimation in mixed models (variance parameters vs. conditional modes)
Current License: CC BY-SA 3.0
13 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 28 at 7:39 | comment | added | Sextus Empiricus | @Geoff regarding question 1 even though you make an unconditional computation, you are still adding an effect: Why do fixed effects in a logistic regression model differ depending on the presence of a random intercept? | |
| Sep 20, 2023 at 11:47 | comment | added | Geoff | @amoeba, this is a great answer! I was following just until the end, and had two questions. 1) I've seen that random effects can change the value of other coefficients in the model, if the model is estimated unconditionally, how is this possible - I assume it's through respecifying the var-covar matrix, but I don't know how this affects the coefficients. 2) How are each of the c_i's estimated? Is it through BLUPs? | |
| Jun 25, 2018 at 13:54 | comment | added | Sextus Empiricus | @statmerkur In an answer to this question I demonstrate the manual calculation of a mixed effects model (manual in the sense of writing the likelihood function, the optimization is still done by a standard optimization function in R) stats.stackexchange.com/a/337348/164061 | |
| Feb 25, 2018 at 20:52 | comment | added | amoeba | Thanks for accepting my answer and awarding me the bounty @statmerkur, but it's too bad that it remains unclear. I will try to think of an example. I will ping you when I update the answer. | |
| Feb 25, 2018 at 2:42 | history | bounty ended | statmerkur | ||
| Feb 25, 2018 at 2:41 | vote | accept | statmerkur | ||
| Feb 25, 2018 at 2:41 | comment | added | statmerkur | I think I just don't get the integration step. As @Martijn Weterings pointed out a little (R code) example or reference were one can find this would be great! | |
| Feb 24, 2018 at 22:41 | comment | added | amoeba | @statmerkur Tau is a parameter; the last formula in my answer still includes tau. The crucial point is that the last formula does NOT include $c$. We simply combine two equations together such that $c$ falls out of there (technically, we integrate over $c$). Then we fit the model, which means we fit tau and other parameters. | |
| Feb 23, 2018 at 22:37 | history | edited | amoeba | CC BY-SA 3.0 |
added 211 characters in body
|
| Feb 23, 2018 at 21:51 | history | edited | amoeba | CC BY-SA 3.0 |
deleted 2 characters in body
|
| Feb 23, 2018 at 21:45 | history | edited | amoeba | CC BY-SA 3.0 |
added 156 characters in body
|
| Feb 23, 2018 at 16:27 | comment | added | Sextus Empiricus | I like this answer. I also liked the question. Personally, I am still struggling on the mechanism (I actually never cared about it to study the algorithms that solve LMEM's). So I guess that the difference of the random effects is being made by changing from $$\mathbf{y} \sim \mathcal{N}(a + b\mathbf{x}, \sigma^2 I)$$ to $$\mathbf{y} \sim \mathcal{N}(a + b\mathbf{x}, \Sigma) $$ I imagine that a tiny example that works this out might be nice. I am considering to make this myself, but maybe there are resources that already show such examples (anyone?). | |
| Feb 22, 2018 at 22:35 | history | answered | amoeba | CC BY-SA 3.0 |