Timeline for Interpreting three forms of a "mixed model"
Current License: CC BY-SA 3.0
23 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 2, 2012 at 15:22 | comment | added | Michael R. Chernick | @Macro Keep up the good work! You didn't anger me often though and I was as guilty as you. Someone wrote (probably a now deleted comment) that they couldn't find your comment about taking a hiatus from CV. Another person commented back that it was in one of those Chernick-Macro arguments that always get deleted! | |
| Oct 2, 2012 at 14:46 | comment | added | Macro | Thanks @Michael and you're welcome. I'm glad to hear that and I can acknowledge that I'm not a angel - I've gone out of my way to needle you from time to time and that's not ok so I'm sorry about that (deleted my comments in the meta thread). See you around. | |
| Oct 1, 2012 at 17:44 | comment | added | Michael R. Chernick | @Macro Thank you! That is so well expressed. I know that you haven't been away that long but I have missed your presence. You may find that my behavior is improving and I am trying to adhere better to the site's policies. | |
| Oct 1, 2012 at 17:23 | comment | added | Macro |
@Michael, I agree with you. In these kinds of hierarchical models, the random effects are defined by a grouping variable (as opposed to other multivariate models such as spatially indexed data sets, where the 'grouping' variable is continuously varying). In the OP's question, Site would be referred to as the random effect, not T or A or anything else. Thinking of it that way, Site's effect clearly could not be both fixed and random, since the two wouldn't be identified from each other. You can have both fixed and random coefficients for a variable, but that's a different question.
|
|
| Oct 1, 2012 at 17:21 | comment | added | Peter Flom | @MichaelChernick Well, the model statement lists fixed effect and the random statement lists random effects. I think there's a lot of terminology confusion here, but I don't know how to clear it up. | |
| Oct 1, 2012 at 17:14 | comment | added | Michael R. Chernick | @Macro Nice to see you helping with this question. I don't understand how a variable can be both a fixed and a random effect. I think somehow my disagreement with Peter and the OP could be about terminology but if i am wrong can you help clear up my confusion. | |
| Oct 1, 2012 at 17:12 | comment | added | Michael R. Chernick | @PeterFlom Okay but how does that make CSES both a random effect and a fixed effect? | |
| Oct 1, 2012 at 17:05 | comment | added | Peter Flom |
@MichaelChernick Using SAS, there is an example on p 12 of this article by Singer where CSES is both in the random statement and the model statement.
|
|
| Oct 1, 2012 at 16:45 | comment | added | Peter Flom | @macro Yes, that's what I meant, sorry. The terminology gets very confusing! That may be why Gelman eschews the terms 'fixed' and 'random' | |
| Oct 1, 2012 at 13:17 | comment | added | Macro | @PeterFlom, re: "if children are nested in classes, you usually want children only as a random effect." I think you mean that class is the random effect. Unless there is further nesting in the data (e.g. repeated measurements on kids) then child level random effects are not identified. | |
| Oct 1, 2012 at 12:40 | comment | added | smillig |
Maybe the example in section 4.2.1 on pages 146-148? The way I read it, the model called fm10rth.lme treats age as both a fixed and a random effect.
|
|
| Oct 1, 2012 at 12:23 | comment | added | Michael R. Chernick | @mark999 I have the book by Pinheiro and Bates and I know Jose Pinheiro very well. A mixed effects model is called that because it contains some covariates that are fixed and some that are random but the same variable can't be both fixed and random. Maybe instead of continually referring to the book you could give me an example from the book that you are referring to. I think this is a misunderstanding of terminology. | |
| Oct 1, 2012 at 11:50 | comment | added | mark999 | @MichaelChernick No, I mean the same variable having both a fixed effect and a random effect in the same mixed-effects model. There are several examples in Pinheiro and Bates. | |
| Oct 1, 2012 at 11:46 | comment | added | Michael R. Chernick | I think you must be misunderstanding or you are not being very clear. I see a variable can be treated either as fixed or random in the ways you describe bith not both ways in the same model??? | |
| Oct 1, 2012 at 11:43 | comment | added | mark999 | @MichaelChernick as I understand it, if you have a fixed effect and a random effect for the same variable, then the fixed effect is the overall effect in the population, while the random effect allows a different effect of the variable for each subject. There are several examples in Pinheiro & Bates. | |
| Oct 1, 2012 at 11:35 | comment | added | Michael R. Chernick | Yes I thought that the OP specified the same variable as a fixed effect and a random effect in the same model. I don't know how that could be interpreted and it seems to make no sense. The variable has to be modelled as one or the other. | |
| Oct 1, 2012 at 11:25 | comment | added | mark999 | Yes Peter, I'm misunderstanding something. The way I read the previous comments, Michael asked whether having a fixed and random effect for the same variable makes no sense, and then you said that it would be very rare. | |
| Oct 1, 2012 at 11:20 | comment | added | Peter Flom | Hi @mark999 if you think I said that, then you are misunderstanding me. In my experience, in a given model, there is usually only 1 effect that is random and not fixed, sometimes there are none, sometimes (with complex nesting) 2. Certainly having something be fixed and random is common. That happens on nearly all these models. | |
| Oct 1, 2012 at 11:16 | comment | added | mark999 | Maybe I'm misunderstanding you, but I would have thought that having fixed and random effects for the same variable was more common than a variable having just a random effect. Having fixed and random effects for the same variable is not uncommon in the Pinheiro and Bates book. | |
| Oct 1, 2012 at 10:55 | history | edited | Peter Flom | CC BY-SA 3.0 |
edited body
|
| Oct 1, 2012 at 10:45 | comment | added | Peter Flom | Well, I wouldn't go so far as to say you never want to put them in the fixed effects too, but I think it would be very rare. You aren't really interested in how Mary did vs. how Joe did, you're just (as you know but some may not) using them as their own controls. | |
| Oct 1, 2012 at 10:42 | comment | added | Michael R. Chernick | But putting it down as both random and fixed makes no sense right? | |
| Oct 1, 2012 at 10:34 | history | answered | Peter Flom | CC BY-SA 3.0 |