Replicate all the important figures and tables in the paper. These include: summary statistics for the data (if available), regression results, and simulations (if any). If there are descriptive figures of the data or descriptive figures of the simulation results, replicate those as well. Why do you need to replicate the summary statistics and descriptive figures in addition to the regression results? If your regression results are off, you need to know whether it was the data or the estimation procedure. If you can eliminate the data (by checking the summary stats and the plots), then you know it's the estimation procedure. And if your model uses simulation, simlarly you have to eliminate the data and regression output as sources of error. If your paper runs 10 specifications and looks at a lot of stars, you still have to replicate those results before attempting to simulate because regression is an important intermediate step to obtaining quantities of interest and you need ot make sure that you hit second base on your way around. (You may omit the stars themselves, if you wish, but should report everything else in these tables.) If your paper address three substantive questions using five separate data sets and sixteen models, you should probably just focus on one of the substantive questions and the data and models assocated with that one question. That would be quite reasonable. Now, why do you need to use R rather than SAS, STATA, or SPSS? The substantive extensions to the paper are limited by the software you use. If you attempt to do the extensions in the canned software, you won't be able to do as much as in R -- where you can/will write the extension yourself. Yours, Olivia. On Fri, 5 Nov 2004 mschoi(a)fas.harvard.edu wrote: the Please, _______________________________________________ gov2001-l mailing list gov2001-l(a)lists.fas.harvard.edu http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l

In Zelig, is there a convenient way to run a model for different subsets of my data? In the manual this is only discussed for "easy" cases such as run the model: by = state. but what about things like run the model: for all obs where covariateX >= 0.5 (or even something like run the model: for all obs for which covariateX=3 & covaraiteV>0.5) Obviously, one way to do this would be to create an indicator variable like ID=1 if covariateX>=0.5 and ID=0 if covariateX<0.5 and then use the usual run the model: by = ID but this is crazy tedious if I want to estimate my model for many different sub samples conditioning on may combinations of specific values of many different covariates. is there a more elegant way to do this? any help is highly appreciated. jens _______________________________________________ gov2001-l mailing list gov2001-l(a)lists.fas.harvard.edu http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l

thanks olivia. just to clarify, so you're saying there is no way to do something like: z.out <- zelig(onsetnew ~ war + gdpcap ,model="logit",subset(data,gdpcap>0.2) or even better: z.out <- zelig(onsetnew ~ war + gdpcap ,model="logit",subset(data,gdpcap>0.2 & year>1976) i alwyas have to create the indicator variables (in this case ID=1 if gdpcap>0.2 and ID=0 if gdpcap<=0.2) outside zelig first and then use these subsets in zelig? best, jens _______________________________________________ gov2001-l mailing list gov2001-l(a)lists.fas.harvard.edu http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l