Paul Seed wrote:
I am interested in a bootstrapped confidence interval for the
difference between two means, and to save time
wish to work out the means using -summ, meanonly-.
I have written a Stata program to do this, which runs perfectly well
on it's own. However, when trying to get -bs- to work out the confidence
intervals, I get an unexplained error message.
Any thoughts on wher ethe problem lies, greatly appreciated.
the program:
cap prog drop myprog
prog define myprog, rclass
summ centile if aps, mean
di "r(mean) = " r(mean)
local m1 = r(mean)
return local m1 = r(mean)
summ centile if control, mean
return local m2 = r(mean)
return local m2_m1 = `m1' - r(mean)
end myprog
called with -bs-
set seed 35435141
bs "myprog" "r(m1)", reps(50)
or
> bs "myprog" "r(m1) r(m2) r(m2_m1)", reps(50)
and output:
. * Now use it for bootstrap"
. set seed 35435141
. bs "myprog" "r(m1) r(m2) r(m2_m1)", reps(50)
command: myprog
statistics: r(m1) r(m2) r(m2_m1)
error in statistic: r(m1)
r(198);
I believe the -bs- command is complaining about the use of macro results placed
in the expression list, i.e. "r(m1)"
It should work if these are instead scalar results. Eg. in the program
-myprog-, replace
return local m1 = r(mean)
with
return scalar m1 = r(mean)
^^^^^^
and likewise for m2, and m2_m1
and it should work.
However, am wondering whether this type of sampling is best for your design?
The command
> summ centile if control, mean
hints that this might be a case-control design?, in which case you might want to
be sampling separately from cases and controls? which can be done, but not with
-bs-.
- Gary
*
* 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/
