[R-sig-ME] Multi-level Rasch Model Per Douglas Bates' paper

[Simon Harmel]

Ben,

This is exactly what I'm trying to understand for my glmmTMB models in my
original post ('m1' & 'm2').

So to get the item easiness I think we need to use: *easiness *<-
*coef(f11)[[1]][[1]][1]*
*
        persons <- coef(f11)[[1]][[2]][1]*

Am I right Ben?

On Wed, May 13, 2020 at 12:32 PM Ben Bolker <bbolker using gmail.com> wrote:

>     Apologies for not looking this over in great detail, but not sure
> why you're mixing lme4 and glmmTMB here??  Isn't item easiness just
> lme4::coef(fm2)$item ?
>
> easiness <-
>    lme4::ranef(fm2)$item[[1]] +
>    glmmTMB::fixef(fm2)[imap$itype])
>
> On 5/13/20 1:25 PM, Rasmus Liland wrote:
> > On 2020-05-13 18:50 +0200, Rasmus Liland wrote:
> >> Indeed it does work now!  Thanks!
> > Right, so this code reproduces code until the
> > easiness variable on page 15.  Perhaps this
> > is useful?
> >
> > data("lq2002", package="multilevel")
> > wrk <- lq2002
> > # wrk[1:5,]
> > for (i in 3:16) wrk[[i]] <- ordered(wrk[[i]])
> > for (i in 17:21) wrk[[i]] <- ordered(5 - wrk[[i]])
> > lql <- reshape(wrk,
> >    varying = list(names(lq2002)[3:21]),
> >    v.names = "fivelev",
> >    idvar = "subj",
> >    timevar = "item",
> >    drop = names(lq2002)[c(2, 22:27)],
> >    direction = "long")
> > lql$itype <-
> >    with(lql, factor(ifelse(item < 12, "Leadership",
> >      ifelse(item < 15, "Task Sig.", "Hostility")
> >    )))
> > for (i in c(1, 2, 4, 5)) lql[[i]] <- factor(lql[[i]])
> > lql$dichot <- factor(ifelse(lql$fivelev < 4, 0, 1))
> > # str(lql)
> > # summary(lql)
> >
> > ## 3.2 Fitting an initial multilevel Rasch model
> > (fm1 <- lme4::glmer(
> >    dichot ~ 0 + itype + (1 | subj) +
> >      (1 | COMPID) +
> >      (1 | item),
> >    lql,
> >    binomial))
> > rr <- lme4::ranef(fm1, condVar = TRUE)
> > str(rr$COMPID)
> > head(rr$COMPID)
> >
> > qq <- lattice::qqmath(rr)
> > print(qq$subj)
> >
> > ## 3.3 Allowing for interactions of company and item type
> > (fm2 <- lme4::glmer(
> >    dichot ~ 0 + itype + (1 | subj) +
> >      (0 + itype | COMPID) +
> >      (1 | item),
> >    lql,
> >    binomial))
> >
> > (fm3 <- lme4::glmer(
> >    dichot ~ 0 + itype + (1 | subj) +
> >      (1 | COMPID:itype) +
> >      (1 | item),
> >    lql,
> >    binomial))
> >
> > (fm3a <- lme4::glmer(
> >    dichot ~ 0 + itype + (1 | subj) +
> >      (1 | COMPID:itype) +
> >      (1 | COMPID) +
> >      (1 | item),
> >    lql,
> >    binomial))
> >
> > anova(fm3, fm3a, fm2)
> >
> > str(imap <- unique(lql[, c("itype", "item")]))
> >
> > (easiness <-
> >    lme4::ranef(fm2)$item[[1]] +
> >    glmmTMB::fixef(fm2)[imap$itype])
> >
> > Best,
> > Rasmus
> >
> > _______________________________________________
> > R-sig-mixed-models using r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>

	[[alternative HTML version deleted]]