[R-sig-ME] Standard Error of a coef. in a 2-level model vs. 2 OLS models

[Simon Harmel]

Hi Harold,

I improved my question, and asked it on CrossValidated: (
https://stats.stackexchange.com/questions/487363/using-variance-components-of-a-mixed-model-to-obtain-std-error-of-a-coef-from-a
)

I would appreciate your answer, either here or on CrossValidated.

Many thanks, Simon

On Wed, Sep 16, 2020 at 4:45 PM Harold Doran <
harold.doran using cambiumassessment.com> wrote:

> This is not how standard errors are computed for linear or mixed linear
> models. I'm not sure what you're goal is, but the SEs are the square roots
> of the diagonal elements of the variance/covariance matrix of the fixed
> effects.
>
> See ?vcov on how to extract that matrix.
>
> -----Original Message-----
> From: R-sig-mixed-models <r-sig-mixed-models-bounces using r-project.org> On
> Behalf Of Simon Harmel
> Sent: Sunday, September 13, 2020 7:51 PM
> To: r-sig-mixed-models <r-sig-mixed-models using r-project.org>
> Subject: Re: [R-sig-ME] Standard Error of a coef. in a 2-level model vs. 2
> OLS models
>
> External email alert: Be wary of links & attachments.
>
>
> Just a clarification.
>
> For `ols1` model, I can approximate its SE of the sector coefficient by
> using the within and between variance components from the HLM model:
> sqrt(( 6.68  + 39.15  )/45)/(160*.25))
>
> BUT  For `ols2` model, how can I approximate its SE of the sector
> coefficient by using the within and between variance components from the
> HLM model?
>
> On Sun, Sep 13, 2020 at 6:37 PM Simon Harmel <sim.harmel using gmail.com> wrote:
>
> > Dear All,
> >
> > I have fit two ols models (ols1 & ols2) and an mixed-effects model (m1).
> > ols1 is a simple lm() model that ignores the second-level. ols2 is a
> > simple
> > lm() model that ignores the first-level.
> >
> > For `ols1` model,  `sigma(ols1)^2` almost equals sum of variance
> > components in the `m1` model: 6.68 (bet.) + 39.15 (with.) For `ols2`
> > model, I wonder what does `sigma(ols2)^2` represents when compared to
> > the `m1` model?
> >
> > Here is the fully reproducible code:
> >
> > library(lme4)
> > library(tidyverse)
> >
> > hsb <- read.csv('
> > https://raw.githubusercontent.com/rnorouzian/e/master/hsb.csv')
> > hsb_ave <- hsb %>% group_by(sch.id) %>% mutate(math_ave = mean(math))
> > %>%
> > slice(1) # data that only considers grouping but ignores lower level
> >
> > ols1 <- lm(math ~ sector, data = hsb)
> > summary(ols1)
> >
> > m1 <- lmer(math ~ sector + (1|sch.id), data = hsb)
> > summary(m1)
> >
> > # `sigma(ols1)^2` almost equals 6.68 (bet.) + 39.15 (with.) from lmer
> >
> > But if I fit another ols model that only considers the grouping
> > structure (ignoring lower level):
> >
> > ols2 <- lm(math_ave ~ sector, data = hsb_ave)
> > summary(ols2)
> >
> > Then what does `sigma(ols2)^2` should amount to when compared to the
> > `m1` model?
> >
>
>         [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>

	[[alternative HTML version deleted]]