[R-sig-ME] Using broom.mixed library with lme4

[Phillip Alday]

Why do you think ubar is for the random effects? In my very quick skim
of the documentation, I didn't see anything indicating that. Looking at
the structures in `fit`, I see:

> tidy(fit$analyses[[5]])
# A tibble: 4 x 6
  effect   group    term            estimate std.error statistic
  <chr>    <chr>    <chr>              <dbl>     <dbl>     <dbl>
1 fixed    NA       (Intercept)        4.91     0.0855      57.4
2 fixed    NA       sex                0.853    0.0350      24.4
3 ran_pars school   sd__(Intercept)    0.820   NA           NA
4 ran_pars Residual sd__Observation    0.767   NA           NA

None of the random effects line up with the output of pool(). Moreover,
the pool() documentation notes that it needs the standard error of each
estimate, but lme4 doesn't produce those (for good reason) for random
effects, so pool() won't produce pooled estimates for the random effects.

The pool() documentation mentions the mipo class, so I looked at ?mipo
and found this:


       ‘estimate’  Pooled complete data estimate
       ‘ubar’      Within-imputation variance of ‘estimate’
       ‘b’         Between-imputation variance of ‘estimate’
       ‘t’         Total variance, of ‘estimate’
       ‘dfcom’     Degrees of freedom in complete data
       ‘df’        Degrees of freedom of $t$-statistic
       ‘riv’       Relative increase in variance
       ‘lambda’    Proportion attributable to the missingness
       ‘fmi’       Fraction of missing information


So `ubar` and `b` are perhaps random effects, but not in the sense
you're thinking of, but rather the random effects that go into
imputation procedures (this is a guess on my part). I don't know much
about imputation, but I suspect this is analogous to the parallels
between mixed models and meta-analysis
(http://www.metafor-project.org/doku.php/tips:rma_vs_lm_lme_lmer). But
again, this is rapidly getting out of my area of expertise and into the
expertise of other members of this list (e.g. Wolfgang Viechtbauer for
meta analysis).

Phillip

On 10/11/20 7:17 am, Simon Harmel wrote:
> Dear All,
> 
> Belwo, I've used library `broom.mixed` and imputed some data with library
> `mice` to then fit a "random-intercept" `lmer()` model.
> 
> BUT I wonder why after I `pool()` my analyses, there is an extra "ubar"
> (random-effect) for slope (`sex`) which is not even in the model?!
> 
> library(mice)
> library(lme4)
> library(broom.mixed)
> 
> imp <- mice(popmis, m = 5) # `popmis` is a dataset from `mice`
> 
> fit <- with(data = imp, exp = lme4::lmer(popular ~ sex + (1|school)))
> 
> pool(fit)
> 
> ### `ubar` is the random effect for intercept (0.007524509) BUT WHY we see
> a ubar ALSO for `sex` (0.001177781)?
> 
> Class: mipo    m = 5
>          term m  estimate        ubar            b           t dfcom
>  df
> 1 (Intercept) 5 4.9007789 0.007524509 0.0004845564 0.008105977  1996
> 547.44383
> 2         sex 5 0.8617941 0.001177781 0.0015867795 0.003081916  1996
>  10.33653
>         riv     lambda       fmi
> 1 0.0772765 0.07173321 0.0751060
> 2 1.6167147 0.61784141 0.6751515
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>