I am studying the heritability of a variable by utilizing the ACE model, and would like to ask the OpenMX community if i am analyzing my results correctly.
When i run the ACE model, followed by the AE and CE models, i get the following results for 2 different datasets:
COMPARE ALL MODELS: Print Comparative Fit Statistics
(Nested.fit <- rbind(
mxCompare(SatFit, AceFit),
mxCompare(AceFit, AEFit)[2,],
mxCompare(AceFit, CEFit)[2,]))
Dataset 1
Model A (95% CI) C (95% CI) E (95% CI) (-2LL) df AIC ∆-2LL ∆df p-value
ACE 43.68% (0.15 - 0.69) 17.18% (0 - 0.39) 39.15% (0.3 - 0.5) 4869.562 16483 -28096.44 8.57 0.07
CE (-) 48.08% (0.4 - 0.55) 51.92% (0.45 - 0.6) 4878.036 16484 -28089.96 8.47 1 0
AE 63.84% (0.55 - 0.72) (-) 36.16% (0.28 - 0.45) 4871.678 16484 -28096.32 2.12 1 0.15
Dataset 2
Model A (95% CI) C (95% CI) E (95% CI) (-2LL) df AIC ∆-2LL ∆df p-value
ACE 39.2% (0 - 0.61) 9.73% (0 - 0.48) 51.07% (0.39 - 0.65) 1884.466 2668 -3451.53 11.45 0.02
CE (-) 42.3% (0.36 - 0.53) 57.7% (0.47 - 0.69) 1887.154 2669 -3450.85 2.69 1 0.1
AE 49.83% (0.37 - 0.61) (-) 50.17% (0.38 - 0.63) 1884.668 2669 -3453.33 0.2 1 0.65
I just want to ask if my interpretation of the data (below) is correct:
For Dataset 1, removing A (genetics) from the model gave a p-value of 0, meaning that its presence is significant. Removing C (shared environment) however gave a p-value of 0.15, meaning that removing this factor did not have any significance on the model. In conclusion, seeing that removing the C factor proved insignificant, my variable of interest has a heritability of about 63.84%.
With respect to Dataset 2, the p-values for all except the ACE model were insignificant.
Is this interpretation correct?
I am confused as to whether i should just look at changes in p-value alone, AIC measures, or whether i should be analyzing these results in another way.
Thank you very much.
Printer-friendly version
You have a model where A cannot be dropped but C can.
The .15 p-value for dropping C will be associated with a lower AIC for the AE compared to ACE model, but it would be unwise to claim that we now know C is zero, nor that the variance attributed to C in the ACE model should be apportioned (mostly) to A..
More informative would be to report the ACE model with confidence intervals (add mxCI with A, C, and E as parameters). The CI for C is likely quite wide.
That said, it is likewise true that you have found evidence for genetic influences, have not found significant support for family influences. That might be contrary to (some) theories of your trait, and supportive of others, and you should note that.