Models with assumptions about the underlying dependency structure.

For log-linear models a selection was made out of all hierarchical models. For the multinomial models a selection was made out of all possible models (including models with probabilities conditional on covariates, mixtures and previous detections). p1 = p(sentinel), p2 = p(hospital), p3 = p(NRC), distance in meters, age in days, q = referrals, r = (re)detection probability conditional on previous detection, p3s11 = probability of detection in sample 3 conditional on detection in sample 1, p3s10 = probability of detection in sample 3 conditional on absence in sample 1, c = covariate-effect (notation in S1 Appendix).