Our goal was to identify putative risk factors for H5N1
in backyard poultry at the village scale. To accomplish this,
we fit a regression model to predict the occurrence of H5N1 per village based on predictor variables measured by
scan sampling and questionnaires. A village was considered positive for H5N1 if at least one bird from the village
was found to be positive by PCR, serology, or both. The
regression model was a generalized linear model, which
was identical to a logistic regression except that it included
a term representing the number of samples per village to
account for differences in sampling effort. We measured a
total of 289 predictor variables at the village scale. The
initial 289 variables were restricted to those that were
significant by univariate analysis and uncorrelated. This
resulted in the retention of 12 continuous variables
(Table 2). Starting from these 12 variables, we carried
out stepwise selection to add significant variables to the
regression model and remove insignificant ones