The GLIMMIX procedure as implemented in SAS (Version 9.2) was used to estimate negative binomial models.
However, this problem is overcome by estimating the negative binomial regression (NBR) model, in which a cross-section heterogeneity is naturally formulated by introducing an unobserved effect into the conditional mean (Greene 2002).
All rates and rate ratios estimated by negative binomial regression.
These patients will be evaluated again at the end of the study and the incidence of exacerbations and associated relative risks will be estimated by negative binomial regression.
Model results are shown in Table 9.> -wrap-foot> Model 1 estimated using ordinary least squares; Model 2 estimated using negative binomial; * <0.10 **<0.05 ***<0.01 The relationships between functional limitations and indirect costs were determined by least squares and negative binomial regression models.
Table 2 summarises the impact of selected structural factors associated with variation in HIV research productivity in EU member countries, as estimated by a negative binomial regression model.
The dispersion parameter estimated by the negative binomial regression was 3.7, with a 95% confidence interval (1.9, 7.5), providing additional evidence that the negative binomial regression was the appropriate model.
Rate of hypoglycaemia was estimated by a negative binomial regression model, in which the number of episodes/patient-year of exposure (events/patient-year) was adjusted for country, sex, age and HbA1c at randomization.
To facilitate the interpretation of results comparing adults with diabetes to those without, we present odds ratios, estimated by the two logistic components and relative risk ratios of conditional means (given vi = 1 and then yit > 0) estimated by the negative binomial components truncated at zero.
The rate of hypoglycaemic episodes during the exposure to trial insulin was estimated by a negative binomial regression model in which the number of episodes per patient year of exposure (episodes/patient year) was adjusted by country, sex, age and HbA1c at randomisation (21).
