Prevalence ratios were calculated by using Poisson regression with
robust SEs (30) for all factors that were found to be substantially
different (P < 0.05) between breakfast skippers and breakfast eaters in
Tables 1–3, with the exception of dietary variables because dietary
intake may be influenced by eating breakfast. Model 1 was the
unadjusted univariable analysis. Model 2 included adjustments for all
factors that were significant (P < 0.05) in model 1. The sample size for
this analysis was reduced to 2711 due to the number of women missing
at least 1 covariate. Sensitivity analysis was conducted by using 2
different definitions of breakfast skipping: those who rarely/never ate
breakfast and those who ate breakfast no more than 2 d/wk (i.e.,
combined rarely/never and 1–2 d/wk). All statistical analyses were
conducted by using STATA software (version 12.0; StataCorp).
for the analyses. Thirds were chosen because, although the sample size
was large, the number of breakfast skippers was low and would have
resulted in small cell sizes in multivariable models..