Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Dan Kimmel <dkimmel@uchicago.edu> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: re:st: pscore question |
Date | Thu, 17 Feb 2011 12:54:57 -0600 |
Hello Ariel (et al), My treatment is binomial -- it is just an indicator of whether the respondent has experienced violence at school. However, since students at different schools have different risks of violence exposure, I estimated the propensity score using an HLM. (I suppose I could do this with a dose-response model, but I'm having enough trouble as it is.) My outcome, however -- the variable on which I want to see if violence exposure has any effect -- is an ordinal measure of general health ("excellent," "very good," "good," "fair," "poor"). I was originally trying to stratify the observations into blocks, balanced on their propensity scores and all other covariates, and then run a simple ordinal regression model with block ID dummies. However, I used quite a number of covariates, and since I don't know of a way to estimate a multilevel model using pscore or psmatch2, calculating the strata by hand became prohibitively difficult. psmatch2 uses propensity scores calculated from an outside program, but (as far as I can tell) seems to assume that the outcome is continuous. I will take a look at the article you recommend. Thanks for the tip. Any further suggestions are still, of course, welcome. Dan -- Daniel M. Kimmel Department of Sociology University of Chicago 917.696.2597 On Thu, Feb 17, 2011 at 12:36 PM, Ariel Linden, DrPH <ariel.linden@gmail.com> wrote: > > Hi Dan, > > Modeling any outcome can be done using propensity score methods. However, > what I am not clear on here is if the treatment variable in multilevel (or > even continuous). > > While not the easiest of reads, I suggest you read: Robins JM, Hernán MA, > Brumback B. Marginal structural models and causal inference in epidemiology. > Epidemiol 2000;11:550–60. > > In particular, section 6 discusses multilevel treatment. In general, you > would use an ordinal or multilogit model to estimate the propensity score, > and then use the estimate corresponding to true level of the treatment. > > For a continuous treatment variable (e.g. a drug with increasing dosage), > modelling the propensity score is even more complex. Fortunately, there is a > user written stata program available called -doseresponse- , (but you really > would only need the sub-routine called -gpscore-) > > This program comes with an accompanying paper in the Stata Journal by > Michela Bia and Alessandra Mattei called "A STATA Package for the Estimation > of the Dose-Response Function through Adjustment for the Generalized > Propensity Score", 2008. Stata Journal Volume 8 Number 3. > > I hope this helps > > Ariel > > From: Dan Kimmel <dkimmel@uchicago.edu> > Subject: Re: re:st: pscore question > > Dear Statalisters, > > Thanks for your suggestions re: my pscore question. I am now > encountering another problem, which was (indeed) my reason for not > simply using psmatch2 in the first place: does anyone know if there is > a way to use psmatch2 (or some other propensity score module) to model > a nominal- or ordinal-scale outcome? I am attempting to predict the > effect of exposure to violence in school on students' health outcomes, > where health is measured with a simple 5-category subjective response. > Thanks, > > Dan > - -- > Daniel M. Kimmel > Department of Sociology > University of Chicago > 917.696.2597 > > > > > * > * For searches and help try: > * http://www.stata.com/help.cgi?search > * http://www.stata.com/support/statalist/faq > * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/