We conducted exploratory latent class analyses to
identify dimensions of pregnancy intention using SAS Proc
LCA [17]. The number of classes was determined through
minimization of the Bayesian Information Criterion. The
LCA procedure estimated latent class parameters using the
expectation–maximization algorithm where the conver-
gence criterion was defined as a maximum absolute devi-
ation of 0.000001. We created parsimonious scores to
capture the dimensions identified using the latent class
analysis with a stepwise elimination process. The process
was initiated by quantifying responses to individual ques-
tions and obtaining the sum of all relevant intention vari-
ables to produce a saturated score. Next, we eliminated the
variable which least impacted the observed agreement
between the saturated scores and the latent class dimen-
sions. We repeated the elimination step to minimize the
number of variables in the scores while maintaining a
substantial level of agreement with the latent classes.
Substantial agreement was defined as a weighted kappa
exceeding 0.7.