Here's the output of comparing two nested models (just setting var = across twin and zygosity...)
> mxCompare(twinSatSub1Fit,c(twinSatSub2Fit)) base comparison ep minus2LL df AIC diffLL diffdf p 1 twinSat <NA> 4 1600 2284 -2970 NA NA NA 2 twinSat twinSat 3 1610 2285 -2960 2.67 1 0.102
But in the actual models (accessed via summary) the values are 1602.957 and 1605.626
> summary(twinSatSub1Fit) # observed statistics: 2288 # estimated parameters: 4 # degrees of freedom: 2284 # -2 log likelihood: 1602.957
> summary(twinSatSub2Fit) # observed statistics: 2288 # estimated parameters: 3 # degrees of freedom: 2285 # -2 log likelihood: 1605.626
The diffLL seems correct (2,67 and not 10 as the table would imply), but the p value is wrong.
> 1-pchisq(1605.626-1602.957, 1)
All done 2010-07-27 under
openmx version number: 0.3.3-1264
I wonder if it is a side effect of mxCI leaving the df and fit changed (as mxCompare sees it) compared to the fitted model values (as summary sees them)?