I'm trying to translate the following line of Stata code into its equivalent SPSS syntax:
mixed outcome i.time c.age i.gender || id:, cov(unstr)
My understanding is that the equivalent SPSS syntax would be:
mixed outcome by time gender with age
/fixed time gender age
/random intercept | subject(id) COVTYPE(UN)
/print = solution
...and indeed, the two commands produce identical parameter estimates for the fixed effects. However, the parameter estimate for the _cons/Intercept term, standard errors and p-values they calculate are completely different. I would expect some variation in how these are calculated between programs, but this is to an unusual extent. Any thoughts?
Hello Alice Wickersham. A couple of possibilities come to mind. IIRC, Stata's -mixed- command uses ML by default, whereas SPSS's MIXED command uses REML. So you'll have to decide which one of those you want to use, and modify one of the commands to use that one.
Also, IIRC, Stata's mixed command does large sample estimation by default--i.e., it reports z-tests and Chi2 tests. If you want it to report t-tests and F-tests instead, you have to use the dfmethod() option.
HTH.
EDIT. PS- The satterthwaite and kroger options for dfmethod() are only available when you fit the model using REML.
The translation of the Stata code to SPSS syntax seems to be mostly accurate, but there may be some differences in how the programs handle the random intercept term and the way the calculations for the intercept term, standard errors, and p-values are done.
One thing to note is that in SPSS, the option COVTYPE(UN) is used to specify an unstructured covariance matrix, whereas in Stata, the option "cov(unstr)" is used. Also, in SPSS, the option "WITH" is used to specify the covariates in the random effects model, whereas in Stata, the option "||" is used.
It is also possible that the two programs are using different algorithms or estimation techniques to calculate the parameter estimates and standard errors, which could lead to some differences in the results. To be sure, it would be best to consult the documentation for both programs or consult with experts in the software to confirm.
Dinesh Kumar, perhaps you did not notice that Alice Wickersham has already found the answer she was seeking, and that with the right settings, the results from SPSS and Stata agree. Here is what she wrote:
Thank you, you're absolutely right, it was a combination of the estimation method and df method. Once I add options to make those consistent between programs I get the same results - perfect!
- Badges
- Science topic