[R-sig-ME] Computing reliability for the least squares estimates of each level1 coefficient across the set of J level-2 units

[Doran, Harold]

Correction, I meant use the marginal variances (not marginal reliabilities)

-----Original Message-----
From: R-sig-mixed-models [mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of Doran, Harold
Sent: Monday, January 29, 2018 3:32 PM
To: Blazej Mrozinski <blazej.mrozinski at gmail.com>; 'r-sig-mixed-models at r-project.org' <r-sig-mixed-models at r-project.org>
Subject: Re: [R-sig-ME] Computing reliability for the least squares estimates of each level1 coefficient across the set of J level-2 units

Best to keep the sig on the email thread. Assuming HLM (the software) is actually computing the reliability in the way reported in the book by B & R, then just use the marginal reliabilities provided by lmer in your summary output and plug those into your formulas.

I think we’re deviating from the purpose of this list, however. We’re here to help you use R, not here to help you understand how to replicate what other software is doing. With that said, you can see in the output you’re providing from HLM and from lmer, the marginal variances are the same, and those are the inputs into the “reliability” formula. So, why you’re not replicating HLM isn’t something we can (or should) go much further with on this list.



From: blazko at gmail.com [mailto:blazko at gmail.com] On Behalf Of Blazej Mrozinski
Sent: Monday, January 29, 2018 3:21 PM
To: Doran, Harold <HDoran at air.org>
Subject: Re: [R-sig-ME] Computing reliability for the least squares estimates of each level1 coefficient across the set of J level-2 units

Of course, I'm aware of the summary() function, but that won't get me the reliability I'm trying to match with HLM output.
I must be a special kind of stupid but still can't work it out.
For example -  an unconditional model of sleepstudy data gives the following numbers:
lmer(Reaction ~ 1 + (1|Subject), data = sleepstudy)


Random effects:

 Groups   Name        Variance Std.Dev.

 Subject  (Intercept) 1278     35.75

 Residual             1959     44.26

Number of obs: 180, groups:  Subject, 18

Same unconditional model in HLM replicates random effects estimates:
 Final estimation of variance components:
 -----------------------------------------------------------------------------
 Random Effect           Standard      Variance     df    Chi-square  P-value
                         Deviation     Component
 -----------------------------------------------------------------------------
 INTRCPT1,       u0       35.75385    1278.33765    17     127.94046    0.000
  level-1,       r        44.25907    1958.86519
 -----------------------------------------------------------------------------

and provides the reliability value:
 ----------------------------------------------------
  Random level-1 coefficient   Reliability estimate
 ----------------------------------------------------
  INTRCPT1, G0                        0.867
 ----------------------------------------------------

Blazej Mrozinski

2018-01-29 21:07 GMT+01:00 Doran, Harold <HDoran at air.org<mailto:HDoran at air.org>>:
No, just do summary() as it outputs the variances of the random effects

From: blazko at gmail.com<mailto:blazko at gmail.com> [mailto:blazko at gmail.com<mailto:blazko at gmail.com>] On Behalf Of Blazej Mrozinski
Sent: Monday, January 29, 2018 3:06 PM
To: Doran, Harold <HDoran at air.org<mailto:HDoran at air.org>>
Cc: r-sig-mixed-models at r-project.org<mailto:r-sig-mixed-models at r-project.org>
Subject: Re: [R-sig-ME] Computing reliability for the least squares estimates of each level1 coefficient across the set of J level-2 units

Harold, thank you for your reply.
Problem is, if I knew where to find needed values for those formulas I wouldn't bother anyone with this question.
I guess that `getME()` might be what I need to use, but that's it. I'm stuck.

Blazej Mrozinski

2018-01-29 20:12 GMT+01:00 Doran, Harold <HDoran at air.org<mailto:HDoran at air.org>>:
The formulas you need are in the SO post you put up yourself. So, lmer gives you the output you need to do it, just follow those formulas you have already posted. An object of class mer does not provide the reliability (which I question in terms of usefulness, but that's another issue)



	[[alternative HTML version deleted]]

_______________________________________________
R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models