[R-sig-ME] Correlations among random variables

[Avraham Kluger]

Dear Daniel,

I thank you very much for your note.  In the meantime, my student Limor Borut solved the problem with lme:

mlm <- lme(outcome ~  0 + focalcode + 0 + partcode, random = ~ 0 + focalcode + partcode|focalid/dyadid, data = df)

Best,

Avi

-----Original Message-----
From: d.luedecke using uke.de [mailto:d.luedecke using uke.de] 
Sent: Saturday, January 12, 2019 11:28 AM
To: Avraham Kluger <avik using savion.huji.ac.il>; r-sig-mixed-models using r-project.org
Subject: AW: [R-sig-ME] Correlations among random variables

Hi Avi,

You can find some of the numbers from the covariance parameters from the SPSS output also in the "summary()" from your model. Other parameters don't match, maybe the random effects structure needs to be specified in a different way? However, I'm not sure how to translate the rather "confusing"
SPSS notation into R-syntax.

Best
Daniel

-----Ursprüngliche Nachricht-----
Von: R-sig-mixed-models <r-sig-mixed-models-bounces using r-project.org> Im Auftrag von Avraham Kluger
Gesendet: Samstag, 12. Januar 2019 08:01
An: r-sig-mixed-models using r-project.org
Cc: Michal Lehmann <chikush using gmail.com>; Kenny, David <david.kenny using uconn.edu>; Sarit Pery <sarit using peryjoy.com>
Betreff: [R-sig-ME] Correlations among random variables

Hi,

I am struggling to analyze, in R, MLM models that specify correlations among random variables, as can be done with SPSS, SAS, or MlWin.

Consider the following code in SPSS
-----------------------------
MIXED
   Outcome  BY role  WITH focalcode partcode
   /FIXED = focalcode partcode | NOINT
   /PRINT = SOLUTION TESTCOV
   /RANDOM focalcode partcode | SUBJECT(focalid) COVTYPE(UNR)
   /REPEATED = role | SUBJECT(focalid*dyadid) COVTYPE(UNR).
-----------------------------
And a minimal code (with data) in R

-----------------------------
df <-
read.csv("https://raw.githubusercontent.com/avi-kluger/RCompanion4DDABook/ma
ster/Chapter%2010/Chapter10_df.csv")
head(df)
library(lme4)

mlm <- lmer(outcome   ~ 0 + focalcode + partcode + role +
                       (0 + focalcode + partcode|| focalid/ dyadid),
                       data = df)
summary(mlm)
-----------------------------

These SPSS and R codes produce the same variance estimates.  However, SPSS also produces a correlation among "focalcode" and "partcode."  How can this be done in R?  Is it also possible to produce the correlation among the respective error variances (as in SPSS)?

Additional information


1.       MOTIVATION.  The question arises from David Kenny's work on
one-with-many reciprocal designs (e.g., a manager rate all subordinates, and all subordinates rate the same manager).  These models estimate the variance stemming from the one (e.g., managers) and the many (e.g., subordinates), and the correlation among them (termed generalized reciprocity).  The data and codes for SAS etc. are available at http://davidakenny.net/kkc/c10/c10.htm.

2.       SPSS OUTPUT (download HTML file):
https://www.dropbox.com/s/eqch0kq6djtbsfx/One%20with%20many%20SPSS%20output.
htm?dl=1

Sincerely,

Avi Kluger
https://www.avi-kluger.com/


	[[alternative HTML version deleted]]

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

--

_____________________________________________________________________

Universitätsklinikum Hamburg-Eppendorf; Körperschaft des öffentlichen Rechts; Gerichtsstand: Hamburg | www.uke.de
Vorstandsmitglieder: Prof. Dr. Burkhard Göke (Vorsitzender), Prof. Dr. Dr. Uwe Koch-Gromus, Joachim Prölß, Marya Verdel _____________________________________________________________________

SAVE PAPER - THINK BEFORE PRINTING