sem is package for pretty traditional covariance matrix analysis. To deal with other members of exponential family than normal, you can also look at GLLAMM package (http://www.gllamm.org) that works from Stata. Or Mplus. Or you might have some luck building up a mixed model for your purpose, see ?nlme. Even if you get your latent variables from sem, they will be subject to sampling and measurement error, and that would be devastating for logistic regression: on top of the usual attenuation bias typical for measurement error models, you have a wrong scale for your coefficients. Economists have good explanation for those effects in terms of the scaling of the coefficients and errors in nonlinear models; statisticians usually don't care so much about those.
On Sun, Jul 27, 2008 at 1:51 PM, Jeroen Ooms <[EMAIL PROTECTED]> wrote: > > Is it possible to fit a structural equation model with link functions in R? I > am trying to build a logistic-regression-like model in sem, because > incorporating the dichotomous variables linearly seems inappropriate. Mplus > can do something similar by specifying a 'link' parameter, but I would like > to be able to do it in R, ofcourse. > > I have explored the 'sem' package from John Fox, but it does not seem to be > able to fit non-linear relations. Is there some R-package or way to get this > done? I have also considered creating a seperate latent variable in the sem > model for the systematic component of the predictors, but then I still need > a way to fix a non-linear link from the systematic component to the > dichotomous Y variable. > -- -- Stas Kolenikov, also found at http://stas.kolenikov.name Small print: I use this email account for mailing lists only. ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
