Joseph,
Thanks for the quick & reasonable response.
The suggestion of a one-sample test restricted to pre-intervention
ADOPT=NO crowd makes sense. I think you are also sneakily suggesting
that the most obvious null hypothesis -- "H0: p = 0" is not a good
choice; there would probably be some adoption even in the absence of the
intervention, and the intervention probably cannot be called a success
unless the proportion of adopters exceeds a minimum cost/benefit
threshold. Instead, I could choose, e.g., "H0: p < .25" (a one-tailed
test). That seems reasonable.
Thanks.
Michael
Joseph Coveney wrote:
Michael I. Lichter wrote:
Greetings! I was asked this afternoon about the most appropriate way to
test of significance for a comparison of proportions before and after an
intervention within a single population, given that the underlying
dichotomous variable can only change from NO to YES and not from YES to
NO (so that the proportion can only increase from pre to post). I wasn't
sure about the answer ...
Is a two-sample test of proportions as in -prtest- reasonable under
these circumstances, given a reasonable N? (The actual N=270, which
should be OK for -prtest-'s asymptotic statistics, I think.) I think
McNemar's test and Cochran's test are inappropriate (like the chi-square
test) because they are ultimately tabular and there cell where pre=YES
and post=NO is empty.
The specifics are that a group of physicians was presented with an
educational intervention intended to prompt adoption of a set of
behaviors. Some physicians had already adopted those behaviors, so their
pre status (ADOPT = YES) did not change after the intervention; only
physicians who had not adopted at the pre-test could change adoption
status after the intervention.
There must be some fairly standard way of handling this in epidemiology
where for some conditions people can go from disease-free to diseased,
but not from diseased to disease-free.
--------------------------------------------------------------------------------
I don't know whether there is a standard way in epidemiology, but one approach
that comes to mind is first to determine your null hypothesis (the hard part),
and then use a one-sample test to evaluate whether the observed proportion of
adoptees (restrict analysis to those physicians whose baseline status is ADOPT =
NO) is different from your hypothesized proportion under the null.
Joseph Coveney
*
* 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/
--
Michael I. Lichter, Ph.D. <mlichter@buffalo.edu>
Research Assistant Professor & NRSA Fellow
UB Department of Family Medicine / Primary Care Research Institute
UB Clinical Center, 462 Grider Street, Buffalo, NY 14215
Office: CC 125 / Phone: 716-898-4751 / FAX: 716-898-3536
*
* 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/
