Dear list,
I am fitting a logistic multi-level regression model and need to test the
difference between the ordinary logistic regression from a glm() fit and the
mixed effects fit from glmer(), basically I want to do a likelihood ratio test
between the two fits.
The data are like this:
My outcome is a (1,0) for health status, I have several (1,0) dummy variables
RURAL, SMOKE, DRINK, EMPLOYED, highereduc, INDIG, male, divorced, SINGLE,
chronic, vigor_d and moderat_d and AGE is continuous (20 to 100).
My higher level is called munid and has 581 levels.
The data have 45243 observations.
Here are my program statements:
#GLM fit
ph.fit.2<-glm(poorhealth~RURAL+SMOKE+DRINK+EMPLOYED+highereduc+INDIG+AGE+male+divorced+SINGLE+chronic+vigor_d+moderat_d,family=binomial(),
data=mx.merge)
#GLMER fit
ph.fit.3<-glmer(poorhealth~RURAL+SMOKE+DRINK+EMPLOYED+INSURANCE+highereduc+INDIG+AGE+male+divorced+SINGLE+chronic+vigor_d+moderat_d+(1|munid),family=binomial(),
data=mx.merge)
I cannot find a method in R that will do the LR test between a glm and a glmer
fit, so I try to do it using the liklihoods from both models
#form the likelihood ratio test between the glm and glmer fits
x2<--2*(logLik(ph.fit.2)-logLik(ph.fit.3))
> ML
79.60454
attr(,"nobs")
n
45243
attr(,"nall")
n
45243
attr(,"df")
[1] 14
attr(,"REML")
[1] FALSE
attr(,"class")
[1] "logLik"
#Get the associated p-value
dchisq(x2,14)
ML
>5.94849e-15
Which looks like an improvement in model fit to me. Am I seeing this correctly
or are the two models even able to be compared? they are both estimated via
maximum likelihood, so they should be, I think.
Any help would be appreciated.
Corey
Corey S. Sparks, Ph.D.
Assistant Professor
Department of Demography and Organization Studies
University of Texas San Antonio
One UTSA Circle
San Antonio, TX 78249
email:[EMAIL PROTECTED]
web: https://rowdyspace.utsa.edu/users/ozd504/www/index.htm
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
