Hi Ott The reason for calculating would be to add "context" to the OLS estimates (from the probit) - e.g. a 1 year increase in age might increase the dependent variable by 1 unit, but given that the selection model is based on a subset of the full data set, if the probability of reaching the selection criteria falls with age, then a 1 year increase in age will have a slightly lower impact on the outcome dependent taking the two combined.
Agree completely on dummies and factor variables - although a part of me thinks that they shouldn't complicate things too much... Many thanks to you both. I will let you know how I get on! Mark 2010/1/4 Ott-Siim Toomet <ott.too...@ut.ee> > Hi Mark, > > why do you need that? If your task is to estimate how much your y changes > if x change, why not use simple OLS? (Well, right, you should be able to > use sampleSelection as well). > > It shouldn't probably be hard to compute it -- it is just OLS marginal > effect + som kind of derivative of Inverse Mills Ratio. A little more > tricky question is, what to do with dummies and factor variables. > > As Arne told, we are open to incorporate your changes! > > Best, > Ott > > > Hi Mark! > > > > On Sun, Jan 3, 2010 at 9:08 PM, Mark Bulling > > <mark.bull...@googlemail.com> wrote: > >> Hi there > >> > >> Within sampleSelection, I'm trying to calculate the marginal effects for > >> variables that are present in both the selection and outcome models. > >> > >> For example, age might have a positive effect on probability of > >> selection, > >> but then a negative effect on the outcome variable. i.e. > >> Model<-selection(participation~age, frequency~age, ...) > >> > >> Documentation elsewhere describes one method for doing this in Stata > >> based > >> on Sigelman and Zeng: http://polisci.osu.edu/prl/Selection%20Models.pdf > >> - > >> see page 16. > >> > >> I'd like to replicate this in r, but wanted to check I'm not reinventing > >> the > >> wheel, before doing so. > > > > I don't know a function/method that does this in R. So if you want to > > implement this in R, I suggest that you add a "marginalEffects" (or > > similar) method for objects of class "selection" to the > > "sampleSelection" package. You can get (write) access to the source > > code of this package on R-Forge [1]. Please let me (and Ott) know if > > you need any assistance. > > > > [1] http://r-forge.r-project.org/projects/sampleselection/ > > > > /Arne > > > > -- > > Arne Henningsen > > http://www.arne-henningsen.name > > > > > [[alternative HTML version deleted]] ______________________________________________ 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.
