Subjects were selected from the 1987
Adolescent Health Survey database. The
survey was administered to grades 7 through
12 of Minnesota public schools, which were
selected by modified stratified cluster sampling.
The sampling design, data management,
psychometrics, and response rate have
been described in detail elsewhere.''58 The
overall participation rate among school districts
exceeded 95%, with nonparticipating
districts replaced by randomly sampled
schools within the same geographic and
population-size stratum. Fewer than 5% of
students refused to participate. The final
sample of 36 254 students composed
approximately 10% of public junior and
senior high school enrollees and closely
resembled the demographic composition of
the Minnesota student body. Ninety-six percent
of the participants completed the item
on self-identification of sexual orientation.
The survey of students was approved by theUniversity of Minnesota Committee on the
Use ofHuman Subjects.
The index subjects in the present study
were those students who described themselves
as "bisexual" (131 males and
144 females) or "mostly/100% homosexual"
(81 males and 38 females). For each
bisexual/homosexual respondent, the preceding
respondent of the same gender was
selected for comparison. Those who were
"unsure'' of their sexual orientation were
excluded, yielding a final comparison group
of 184 males and 152 females who identified
themselves as "mostly/100% heterosexual."
Since subjects in the datafile were
ordered by school, grade, and classroom,
the selection of the bisexual/homosexual
and heterosexual comparison groups in
proximity to each other was considered the
best way to match their sociodemographic
characteristics. This method has been
described as a "specialized cohorts
design,"'9'20 whereby comparison groups
are drawn from a similar social context.
The design assumes that the comparison
groups resemble each other, except for the
index characteristic (in this instance, sexual
orientation). A priori selection of matching
variables, as in classic case-control designs,
is not required. Also, the method yields
more comparable comparison groups thanis typically obtained by random selection.