Skip to main content
19 events
when toggle format what by license comment
Sep 21, 2020 at 7:18 vote accept camhsdoc
Sep 20, 2020 at 21:00 history tweeted twitter.com/StackStats/status/1307786623741026306
Sep 20, 2020 at 20:25 history edited Robert Long
edited tags
Sep 20, 2020 at 18:35 comment added Robert Long @whuber Thanks for reopening. I've posted an answer. Hopefully it's what the OP is looking for, but I would be interested in your comments, if you have any !
Sep 20, 2020 at 18:30 answer added Robert Long timeline score: 16
Sep 20, 2020 at 15:28 comment added Robert Long @whuber ......basically the steps for a mixed model are to compute $X\beta$ as above, simulate some random effects $u$ based on the number of groups, their variances and covariances (some people find this a bit tricky), then form the model matrix $Z$ for the random effects (which is the very tricky part), then compute $Zu$ and add the residuals. I guess this kind of answers the question, but since they want to do it from scratch some details about how to form $Z$ when there are random slopes and random intercepts will need quite a lot of thought.
Sep 20, 2020 at 15:23 history reopened mdewey
mkt
Robert Long
whuber
Sep 20, 2020 at 15:20 comment added Robert Long @whuber I agree that the question could do with some more focus. I don't really see the list you quoted as steps though, I see them as parts of a fitted model. They need to take the parts and put them together, which requires statistical understanding. If it were a linear model they would specify a model matrix $X$, a fixed effects coefficient vector $\beta$, compute $X\beta$ then simulate the outcome by adding residuals drawn from some distribution. With a mixed model it's similar but the presence of the random effects makes it quite a bit harder to do from scratch...
Sep 20, 2020 at 14:58 comment added whuber @Robert Could you explain what might be missing from the steps listed in "specify the fixed effects, number of groups, sample size, variances of the random effects (and the correlation between them), and simulate a dataset accordingly"? That might help give this question some focus.
Sep 20, 2020 at 13:11 review Reopen votes
Sep 20, 2020 at 15:23
Sep 20, 2020 at 12:51 history edited camhsdoc CC BY-SA 4.0
Explaining that I needed to know the steps of how to do the simulation
Sep 20, 2020 at 12:48 comment added camhsdoc Yes this is correct. I don't know what the steps are for doing it.
Sep 19, 2020 at 19:22 comment added Robert Long @whuber they appear to not know what the steps that you refer to are. They know about the components of a mixed model, but they don't seem to know how to formulate the steps. The steps are based on a statisitical understanding of mixed models, so it seems like a completely statistical question to me. I'm not sure how it is off topic ?
Sep 19, 2020 at 14:01 history closed whuber Not suitable for this site
Sep 19, 2020 at 14:01 comment added whuber Your question answers itself: you simply execute all the steps you list in R. For help with any particular step, work out the code and if it fails, post it on Stack Overflow with a description of what's going wrong and what you need it to do.
Sep 19, 2020 at 12:42 comment added camhsdoc Yes I want to do it from scratch. I am using R if that is relevant.
Sep 19, 2020 at 12:12 comment added Robert Long What do you mean by "do it myself" ? This kind of simulation needs a quite complicated design matrix for the random effects. Do you want to form that matrix by yourself or are you happy using a mixed model package to do that part for you ? If the former, then it's quite easy, but the latter you will be in a world of pain.
Sep 19, 2020 at 9:26 review First posts
Sep 19, 2020 at 9:28
Sep 19, 2020 at 9:24 history asked camhsdoc CC BY-SA 4.0