<>
Problem seems to arise when you put tempvars into a regression and then return the coefficients: my guess is that the matrix stripes are messed up by the fact that these tempvar names do not exist outside the program, and when bootstrap gets the coeff vector back its names are poorly defined. Works fine with explicit variables:
webuse sysdsn1, clear
capture program drop proca_reg
program define proca_reg, eclass
tempvar es
tempname beta
mlogit insure age male nonwhite
capt drop p1 p2
predict p1, outcome(1)
predict p2, outcome(2)
// predict `p3', outcome(3) don't need all outcomes in regression
regress patid p1 p2 age male nonwhite
g byte es = e(sample)
mat `beta' = e(b)
ereturn post `beta', esample(es)
end
bootstrap _b, saving(regmlogit,replace) : proca_reg
Note that you do not need all the predicted outcomes in the model, as they are collinear. You can calculate the missing outcome with lincom.
Kit Baum | Boston College Economics & DIW Berlin | http://ideas.repec.org/e/pba1.html
An Introduction to Stata Programming | http://www.stata-press.com/books/isp.html
An Introduction to Modern Econometrics Using Stata | http://www.stata-press.com/books/imeus.html
*
* 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/
