Our statistical model was a negative binomial regression model. The model can be used for overdispersed count data when the conditional variance exceeds the conditional mean [8]. It can be considered as a generalization of Poisson regression in which the unobserved heterogeneity of an establishment is included and assumed to distribute following a gamma function. Our model was very similar to the one adopted by Fenn and Ashby [5], who examined the relationship between union density and workplace accidents. The estimation of our model was done with the maximum likelihood method using Stata version 13.