Timeline for minimum amount of prior to get a mixed-model to converge (in R)
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 15, 2021 at 15:00 | history | tweeted | twitter.com/StackStats/status/1350095410883399685 | ||
| Oct 4, 2020 at 22:54 | answer | added | Erik Ruzek | timeline score: 0 | |
| Oct 4, 2020 at 8:05 | comment | added | Robert Long |
Maybe yes, or maybe some combinations of the factors don't exist. I don't know what the exact problem is, but you need to fix that singular model matrix for fixed effects, becuase that is basically what is used when you fit random slopes for the fixed effects, so that might be the cause of the singular VCV of random effects. So once you can fit the lm() model with sch.id as a fixed effect you can move on to the mixed model.
|
|
| Oct 4, 2020 at 7:51 | comment | added | rnorouzian | @RobertLong, OR you mean here we have collinearity issue? | |
| Oct 4, 2020 at 7:39 | comment | added | Robert Long |
You have to fit lm(math ~ ses*sector + sch.id, data = hsb) %>% summary() with schi.id as a factor to see the problem.
|
|
| Oct 4, 2020 at 7:35 | comment | added | rnorouzian |
@RobertLong, good catch! But I see no change after turning sch.id into a factor.
|
|
| Oct 4, 2020 at 7:29 | comment | added | Robert Long |
OK there is something a bit strange with your data, which may have something to do with the problem. if you convert sch.id to factor and fit an lm model with sch.id as a fixed effect, the model matrix is singular. I would fix this problem first.
|
|
| Oct 3, 2020 at 22:12 | comment | added | rnorouzian | @RobertLong, can you please check this one out? | |
| Oct 3, 2020 at 18:59 | comment | added | rnorouzian |
@RobertLong, no Rob, this is a completely different (and real) dataset. Here ses is a level-1 predictor. sector is a level-2 predictor.
|
|
| Oct 3, 2020 at 17:40 | comment | added | Robert Long | Is this the same data as before? If so then I thought we had established that random slopes over schools for a variable that is constant within schools doesnt make sense. | |
| Oct 3, 2020 at 16:29 | history | asked | rnorouzian | CC BY-SA 4.0 |