Not sure about list etiquette regarding answering one's own questions,
but for thread-reading posterity, the answer of how to how to do the
bootstrap that I am interested in appears to be given in the following
reference (in particular, "marginal standardization"):
Berlin JA, Margolis DJ, Localio AR. Relative risks and confidence
intervals were easily computed indirectly from multivariable logistic
regression. Journal of Clinical Epidemiology 2007; 60: 874-882
(the reference has appeared previously on this list).
With interaction terms, this works out to:
. global Xpos _b[_cons] + _b[int_zlog_pos]*zlog + ///
_b[int_zero_pos]*zero + _b[zlog]*zlog + _b[zero]*zero + _b[mbpos]
. global Xneg _b[_cons] + _b[zlog]*zlog + _b[zero]*zero
. program boot_posneg, rclass
. logistic outcome zlog zero mbpos int_z*
. tempvar rrt
. gen `rrt'=invlogit(($Xpos))/invlogit(($Xneg)) if e(sample)
. qui sum `rrt' ,meanonly
. return scalar rr=r(mean)
. end
. bootstrap rr_univ=r(rr), reps(1000): boot_posneg
. estat bootstrap
Which, I believe, provides the 95% CIs for the relative risk
for condition mbpos=1 vs. mbpos=0 in the population as otherwise specified.
(this is somewhat different from what I said I wanted, but it does the job).
Daniel
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
