As the prevalence of each event was high, ORs would not provide a good estimate of PRs.: 23 rather, the log-linked binomial model was applied, using the PROC GENMOD procedure in the SAS statistical package (V.9.3) with DIST=BINOMIAL and LINK=LOG options.
Crude and adjusted binomial regression models were applied to estimate the association between the exposure and the outcome analysing men and women separately.
Negative binomial regression models were applied to determine the percentage change in rate and to accommodate for over-dispersion in the data, with the log of NSW population used as an offset to control for the population changes over time.
Pearson's Chi2 test was used to compare proportions of categorical data (e.g., gender differences, medical specialties, etc).. Separate binomial logistic regression models were applied to measure the relationship between multiple attendances, short-stay cases, respectively, and the degree of alcohol consumption and several other clinical characteristics (tables 2, 3).
For the prevalence data, logistic regression models were applied with a binomial error distribution and logit link function.
Subsequently, three separate negative binomial regression (NBR) models are applied to examine the impact of including the access measure versus land-use measures in the models.
Descriptive statistics, chi-square tests, negative binomial and logistic regression modelling were applied to determine: (a) municipality characteristics predicting Portuguese and International physicians' geographical distribution, and; (b) doctors' characteristics that could increase the odds of residing outside the country's metropolitan areas.
Since the likelihood ratio test for alpha = 0 (χ2(1) = 221.60; P = 0.000) indicated over dispersion, a robust negative binomial regression model was applied to predict the number of infections per woman (a count outcome).
