Simultaneous-equation bias adjustment of the binary regression coefficients of farm-level risk factors for lice (Bovicola ovis) in sheep flocks in Queensland, Australia

Abstract

Simultaneous-equation bias adjustment of the binary regression coefficients of farm-level risk factors for lice (Bovicola ovis) in Queensland sheep flocks was undertaken to control for the interdependency between lice prevalence and their control. Using probit analysis and a two-stage estimation procedure, the correlation between the error terms and the explanatory variables was removed — resulting in coefficient estimates that were asymptotically unbiased.

The data for the study were derived from a cross-sectional study of 342 Queensland farms obtained by postal questionnaire in February 1993. The outcome variable of interest was the frequency of flock lice infestation which was classified into two groups: nil versus lice each of the last 3 years. Probit analysis was used to estimate the regression coefficients of the binary explanatory variables chosen for examination: ability to maintain sheep-proof boundary fences and method of post-shearing lice treatment (backliner or dip).

Removal of the simultaneous-equation bias (SEB) importantly reduced the calculated probability of a farm being infested with lice. No change in associated probabilities for farms using backliner or relative to the fencing practice occurred when the SEB was removed. A table is presented that shows all possible combinations of the explanatory variables and can be used to assess the probability of lice infestation.