Contamination of raw poultry meat with foodborne pathogens could occur because of improper handling at primary production and slaughterhouse levels. Low microbial prevalence data often consists of a high amount of non-detections (zero positives), so a flexible framework is required to characterise the underlying microbial distribution and conduct reliable inferential statistics. Thus, the objective of this work was to evaluate the performance of zero-inflated binomial (ZIB) regression models to describe the effects of sampling site (carcass, thigh, breast, wings) on the measured incidences of Salmonella, Listeria monocytogenes and Staphylococcus aureus on chicken meat. For each pathogen, four regression models based on the zero-inflated binomial ZIB (p, w0) distribution were fitted to the presence/absence data with sampling site as covariate and random-effects due to sampling occasion either in the binomial probability(p)or in the extra-proportion of non-detections(w0).Forthe three pathogens, the sampling site exerted a greater effect on w0 than on p itself, with breast bearing the lowest prevalence estimatesofSalmonellaspp.(mean: 0.88%;95%CI:0.02–1.95%)andS.aureus(mean1.48%;95%CI: 0.01–4.00%). The fitting capacity of the models was further improved when random effects due to sampling occasion were placed in w0 (deviances decreased from 146.7–156.7 to 140.2–140.6). This would imply that, theoretically, the variability in pathogens’ occurrence from batch to batch mainly arises from the variability in non-contaminated zones. At any sampling site, the mean prevalence was estimated as 1.35 (95% CI: 0.15 – 2.70) for Salmonella, 2.11 (95% CI: 0.04 – 5.63) for L. monocytogenes and 2.36 (95% CI: 0.04 – 5.12) for S. aureus. Sampling performance analysis showed that wings were mostly suitable to detect Salmonella and S. aureus with higher probability (0.016 and 0.035 respectively), while for L. monocytogenes, sampling of thigh could be more effective (0.032).
