[R-sig-ME] theoretical question about random effects specifications

[ONKELINX, Thierry]

Dear Dan,

(1|x:y) and (1|x/y) are two different things. (1|x:y) is a random effect for every combination of x and y. Whereas (1|x/y) are two random effects: one for each level of x and one for each level of y nested in x. You can write is as (1|x) + (1|x:y). So note that is not equivalent to only (1|x:y)

Something like (1|listener) + (1|talker) + (1|sentence) + (1|talker:sentence) seems to be reasonable for your design. You could try (1|listener) + (1|talker) + (0 + talker|sentence) This allows for different variances per talker for the random effect of sentence. (1|talker:sentence) assumes that all those variances are equal. But using (0 + talker|sentence) comes at a high price: you'll need to estimate n * (1 + 1) / 2 parameters (with n the number of levels of talkers). So you'll need lots of date to support such a complicated model.

Best regards,

Thierry


ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium
+ 32 2 525 02 51
+ 32 54 43 61 85
Thierry.Onkelinx op inbo.be
www.inbo.be

To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of.
~ Sir Ronald Aylmer Fisher

The plural of anecdote is not data.
~ Roger Brinner

The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey

-----Oorspronkelijk bericht-----
Van: r-sig-mixed-models-bounces op r-project.org [mailto:r-sig-mixed-models-bounces op r-project.org] Namens Dan McCloy
Verzonden: maandag 28 mei 2012 19:07
Aan: r-sig-mixed-models op r-project.org
Onderwerp: [R-sig-ME] theoretical question about random effects specifications

In psycholinguistic research, it seems to be common to have random effects for both subjects (study participants) and items (stimulus conditions).  I have a somewhat more complicated experiment, where each stimulus involves one of 180 different sentences, read by one of
20 different talkers.  It is possible (indeed, likely) that responses will cluster based on who the talker is, and it is also likely that certain sentences are harder than others.  So the experimental design would seem to demand random effects for listener, talker, and sentence.  However, there is a distinct possibility that some sentences are easy with some talkers and hard with other talkers (i.e., due to dialect differences that are particularly strong for certain words).  Question (1):  is it best to account for this possibility by including a random effect for "stim" (i.e., talker-sentence pairing)?  Question (2):  if so, should I include the separate random effects for talker and sentence in addition to "stim", or is that redundant?  Question (3):  I've seen some models referenced on this list and others that use what seem to be interactions between random effects terms:  (1|x:y) or (1|x/y).  Is there a reference somewhere where I could read up on these aspects of model specification?  I tried running one or two of those, and they didn't seem to behave like interactions between fixed effects (which I'm more familiar with).

_______________________________________________
R-sig-mixed-models op r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
* * * * * * * * * * * * * D I S C L A I M E R * * * * * * * * * * * * *
Dit bericht en eventuele bijlagen geven enkel de visie van de schrijver weer en binden het INBO onder geen enkel beding, zolang dit bericht niet bevestigd is door een geldig ondertekend document.
The views expressed in this message and any annex are purely those of the writer and may not be regarded as stating an official position of INBO, as long as the message is not confirmed by a duly signed document.