Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Doug Hess <douglasrhess@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: margins command: adjusting predictions with means of a group not esimated |
Date | Thu, 18 Aug 2011 09:54:55 -0400 |
It was hard to come up with a succinct title for this email, and it may be hard to write about this problem (writing about complex adjusted margins is like thinking through one of those variants of chess where you play on multiple inter-related boards...challenging in a way that is fun up until you become frustrated and want to jab forks in your eyes), but here goes: I have a logistic model of an outcome (0=nonevent, 1=event) on the household level with predictors at the household level and (U.S.) state level (set aside clustering issues for the moment). Most of the state level predictors are continuous, but all the household-level ones are binary. After running the model, I want to say what state A's rate of the event is if it had the population characteristics of State B (the state with the lowest prevelance of the outcome), first holding state A's state-level variables as they are, and then also giving state A the state-level variables of state B. However, other than filling in lots and lots of -at(x=(...))- options, I am not sure if there's a way to do this. The problem seems to be that since I don't have the states in the model I cannot use them to control the means assigned to a state (I guess I could add them as indicators, but there are reasons not to have fixed effects). The -over( )- option doesn't seem to do all of this, but perhaps I am missing a creative way to use some of the more complex contrast commands or operators, or other options. Perhaps just doing completing lots of -at()- options is the only way to go? -Doug * * 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/