To explore which (if any) of the independent variables
had an effect on any breastfeeding at 6 months we used a
logistic regression model. Each variable (of those noted
above that were adjusted for in the primary outcome) was
assessed for its association with any breast milk feeding at
6 months. Where there was an association between breast
milk feeding at 6 months and the variable being tested at
the univariate level, the variable was entered into the
model, along with the primary exposure variable (feeding
only directly at the breast prior to recruitment) and all
other variables that were associated with any breast milk
feeding at 6 months at the univariate level. Variables were
then removed one at a time until only those that showed
an association remained (the likelihood ratio test was used
after each variable was removed to assess that its removal
made no significant difference to the model).