[R-sig-ME] Mixed models with a dichotomous outcome, random slopes, and accounting for autocorrelation

[Ben Bolker]

   [adding r-sig-mixed back to the cc: list]

   This looks mostly right. The only thing you have to watch out for is 
that you want `linear_time` to be *numeric* for the fixed-effect model 
but numFactor()-ized for the ar1() term.  So

dat$linear_time_f <- numFactor(dat$linear_time)

so as not to mess up the original variable, then use linear_time_f 
within ar1().

On 2023-04-04 1:59 p.m., Adam Roebuck wrote:
> Hi again Professor Bolker,
> 
> Thank you kindly for taking the time to respond and even follow up on my 
> original question.
> 
> My original post to the listserv was actually asking about a different 
> project where I had three levels – day, team, and year – without 
> cross-classification. However, my curiosity about handling 
> auto-correlation applies to that project and this new one with 
> cross-classification.
> 
> The new project has a cross-classification of (a) ego and (b) alter 
> nested within a higher level team. Ratings of ego and alter are captured 
> over time, hence the desire to account for auto-correlation. 
> Extrapolating from your email, then, I am guessing the base model would 
> look something like the following:
> 
> dat$linear_time <- numFactor(dat$linear_time)
> 
> mod <- glmmTMB(continuous_outcome ~ 1 + linear_time + Predictor_X +
>      (1|ego) + (1|alter) + (1|team) + ar1(linear_time + 0 | ego:alter)
> 
> Does that seem to track?
> 
> Any additional information would be above and beyond, though highly 
> appreciated.
> 
> Thanks again for the reply and for all you have done for mixed-level 
> modeling.
> 
> My best,
> Adam
> 
> On Tue, Apr 4, 2023 at 9:47 AM Ben Bolker <bbolker using gmail.com 
> <mailto:bbolker using gmail.com>> wrote:
> 
>     *Message sent from a system outside of UConn.*
> 
> 
>        This seems to have slipped through the cracks, sorry about that.
> 
>        lme4 doesn't do correlation structures, but glmmTMB does. The closest
>     analogue would be something like
> 
>     ## this should be observation number within group
>     dat$times <- numFactor(dat$times)
> 
>     model2 <- glmmTMB(continuous_outcome ~ 1 + Predictor_X + Day +
>          (1 + Day | Team / Year) + ar1(times + 0 | Team:Year)
> 
>        (your random effect has two | in it; was the second meant to be a
>     / ?)
> 
>     See https://glmmtmb.github.io/glmmTMB/articles/covstruct.html
>     <https://glmmtmb.github.io/glmmTMB/articles/covstruct.html>
> 
>     On 2023-01-25 5:13 p.m., Adam Roebuck wrote:
>      > Hello,
>      >
>      > I am attempting to set up a multilevel model with two grouping
>     variables
>      > (team and year), random slopes for day of the year, and
>     accounting for
>      > autocorrelation. I have a number of different outcome variables I
>     can use.
>      > When the outcome variable is continuous, I can set up a simple
>     growth model
>      > using the nlme package that looks something like the following:
>      >
>      > model1 <- lme(continuous_outcome ~ 1 + Predictor_X + Day,
>      >
>      > random = ~1 + Day | Team | Year, data = dat, method = "REML",
>      >
>      > control=list(opt="optim"), correlation = corAR1()
>      >
>      > However, I now want to build a model with a dichotomous outcome
>     variable.
>      > As far as I know, nlme cannot account for all of the above and a
>      > dichotomous outcome. I have, however, seen a few references to
>     glmer (in
>      > the lme4 package) and glmmTMB being capable of such. I've gone
>     back through
>      > the archives, though, and am not seeing any clear explanations on
>     how to
>      > set up such a model.
>      >
>      > Using the parameters above and the variables I specified, could
>     someone
>      > please help specify what that model might look like in R or at
>     least point
>      > me in the right direction?
>      >
>      > Thanks,
>      > Adam
>      >
> 
> 
> 
> 
>     --
>     Dr. Benjamin Bolker
>     Professor, Mathematics & Statistics and Biology, McMaster University
>     Director, School of Computational Science and Engineering
>     (Acting) Graduate chair, Mathematics & Statistics
>       > E-mail is sent at my convenience; I don't expect replies outside of
>     working hours.
> 
>     _______________________________________________
>     R-sig-mixed-models using r-project.org
>     <mailto:R-sig-mixed-models using r-project.org> mailing list
>     https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>     <https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models>
> 
> 
> 
> -- 
> *Adam Roebuck*
> University of Connecticut | School of Business
> Doctoral Candidate | Management
> 2100 Hillside Rd., Unit 1041
> Storrs, CT 06269
> (248) 613-9609

-- 
Dr. Benjamin Bolker
Professor, Mathematics & Statistics and Biology, McMaster University
Director, School of Computational Science and Engineering
(Acting) Graduate chair, Mathematics & Statistics
 > E-mail is sent at my convenience; I don't expect replies outside of 
working hours.