[R-sig-ME] glmmTMB: dispformula for mixed beta regression

[Timothy MacKenzie]

Dear Mollie,

Thank you for your clarification. So, I can conclude that X2 has a
logarithmically inverse association with residuals' variance. But then
for interpretation purposes, should I say that: 0.147987 for X2 in the
dispersion model is the slope of that association when X2 is 0?

OR should I say that: for each unit of increase in X2, residuals'
variance decreases by exp(0.147987) = 1.16?

Thank you,
Tim M

On Mon, Jan 30, 2023 at 6:09 AM Mollie Brooks <mollieebrooks using gmail.com> wrote:
>
> Dear Tim,
>
> For the beta family, the conditional variance is mu*(1-mu)/(1+phi) (i.e., increasing phi decreases the variance.) This is in the helpfile ?sigma.glmmTMB. The dispersion model is for log(phi). So higher X2 values give lower variance.  Does that agree with your data?
>
> Cheers,
> Mollie
>
> > On 28 Jan 2023, at 06.22, Timothy MacKenzie <fswfswt using gmail.com> wrote:
> >
> > Dear All,
> >
> > I have a mixed beta-regression model whose residuals don't spread
> > evenly across its fitted values unless I add X1 and X2 (two numeric
> > predictors) to the "dispformula" argument (see below).
> >
> > Based on the Dispersion model results below, can I say the higher the
> > value of X2 (0.147987), the more spread-out the residuals but not so
> > much so for X1 (0.003548)?
> >
> > In other words, can I think of the variables in dispformula as
> > "variance covariates"?
> >
> > glmmTMB(y ~ X1 + X2 + (1 | id), family = beta_family("logit"),
> > dispformula = ~ X1 + X2)
> >
> > Dispersion model:
> >   (Intercept)           X1                   X2
> >      0.752923        0.003548        0.147987
> >
> > Thanks,
> > Tim M
> >
> > _______________________________________________
> > R-sig-mixed-models using r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>