Is there a recommended approach to modeling twin data in openmx where the variables are semi-continuous/skewed such that they are exhibiting a ceiling effect? Alternative estimation method perhaps, or is ML robust enough to such violations?
I would probably just make an ordinal variable with perhaps 10 or more categories. The loss of information is relatively minor if the categories each have reasonable representation. ML estimates are fairly robust, but they will depend on the degree of ceiling effect. Truly J-shaped distributions are likely to exhibit bias, and GxE analyses with such data should be viewed with great caution (large pinches of salt).
If the rest of the distribution (below the ceiling) are truly continuous, and you don't want to lose any of the information therein, it would be possible to subdivide the twin groups into 4 (concordant ceiling, discordants, concordant not ceiling) and have a separate likelihood function for each. Harold Snieder and I used this approach in Snieder H, Boomsma DI, van Doornen LJ, Neale M.C.: Bivariate genetic analysis of fasting insulin and glucose levels. Genet Epidemiol. 1999; 16:426-446 http://www.vipbg.vcu.edu/vipbg/Articles/genetep-bivariate-1999.pdf - see the Appendix. The Classic Mx script for this problem has not yet been converted to OpenMx syntax.
Thanks for the pointers and the reference!