Hi all,
I am trying to use SPSS to replicate some analyses that were done previously in
Stata (which I don't have access to or experience with). The data is from a
within-subjects dose-response study - each subject had their outcome (FMD)
measured at 4 different doses (on the log scale) of the intervention.
In Stata the data was analysed using both 'xtgee' and 'regress' with the
'cluster(subjects)' option. Parameter estimates from the two models are the
same, but the regress analysis gives robust standard errors.
When I try and replicate this in SPSS with a GEE using GENLIN and either
COVB=MODEL (to match the xtgee analysis) or COVB=ROBUST (to match the regress
analysis), I only get similar results when dose is entered as a factor. When
dose is entered as a covariate, the p-values are similar but the parameter
estimates are quite different.
Does anyone with experience of these procedures know if SPSS is using a
different parameterisation to Stata, or something? Is it possible to use SPSS to
get similar results to Stata?
Stata commands:
> .xtset Subject
> .xtgg FMD logdose
Stata output:
Param Coef SE p
logdose .421 .151 .006
_cons 6.111 .722 .000
SPSS command:
GENLIN FMD WITH Log_Dose
/MODEL Log_Dose INTERCEPT=YES DISTRIBUTION=NORMAL LINK=IDENTITY
/REPEATED SUBJECT=ID WITHINSUBJECT=Log_Dose CORRTYPE=EXCHANGEABLE COVB=MODEL
/PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION.
SPSS output:
Param Coef SE p
logdose 1.429 .512 .005
intercept 4.102 1.05 .000
Thanks,
Kylie.
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
