So I ran the model with two different R implementations ( gee and geeM). That's still more different than I'd expect, and it's a bad sign that the statsmodels estimate hasn't converged. Here are the fitted values for the four combinations of the two binary variables statsmodels SPSS SPSS has 1 as the reference level of the two variables and statsmodels has 0. They're less wildly different once you correct for the different contrasts the two programs use. I'm much more familiar with Mixed Effect models.but trying those in statsmodels were not replicating the GEE results either (even though in principle they should be similar). I should point out if I run a model with an additional predictor (White*Right+Macro) the results are closer.but still quite a bit different: The results aren't even close (seems statsmodels isn't converging-and I've tried up to 10000 iterations but get the same result). Stim_model_out = GEE_model.fit(maxiter=1000)
GEE_model = smf.gee("DifferenceScore ~ White * Right", groups="ParticipantID",ĭata=stim_df_with_facename,cov_struct=ind, family=fam) PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION.įam = sm.families.Gaussian(link=sm.) CRITERIA SCALE=MLE PCONVERGE=1E-006(ABSOLUTE) SINGULAR=1E-012 ANALYSISTYPE=3(WALD) CILEVEL=95ĬORRTYPE=EXCHANGEABLE ADJUSTCORR=YES COVB=ROBUST MAXITERATIONS=1000 PCONVERGE=1e-006(ABSOLUTE)
MODEL White Right White*Right INTERCEPT=YES GENLIN DifferenceScore BY White Right (ORDER=ASCENDING) Code Sample, a copy-pastable example if possible USE ALL.ĬOMPUTE filter_$=(BehTaskNum = 1 or BehTaskNum = 2 or (BehTaskNum = 3 and BlockNumber = 6)). csv (and I filter using pandas to achieve the same dataframe as in SPSS). The input matrices are being loaded from the same. I even signed up for a free trial of SPSS and can confirm SPSS gives the answers reported in the paper.
I am getting very different answers for what seems like the same input structure. I'm attempting to replicate a GEE model in statsmodels from a published paper that used SPSS ( ).
0 kommentar(er)
