The unit of randomization and intervention was the worksite, while the unit of
measurement was the employee. We com-
puted all analyses by taking into considera-
tion the nesting of employees at worksites.
We used mixed-effects linear modeling to test
hypotheses about the intervention groups,
controlling for the clustering of respondents
at worksites. All analyses were conducted
with the personal computer version of SAS
statistical software.26The linear logistic re-
gression analyses were computed with the
GLIMMIX macro, which uses iteratively
reweighted likelihoods to fit a logistic re-
gression.27Statistical significance of fixed
effects was tested with F tests based on Wald
statistics.