Full likelihood functions for distance sampling methods involve assuming probability models for the encounter rate or animal distribution, for the detection process, for cluster size (if animals are detected in clusters), and possibly for other data. Conventional distance sampling (CDS) methods avoid assuming a probability model for encounter rate by using an empirical estimator of its variance and getting confidence intervals assuming density D to be log-normally distributed. Similarly, a point estimator and sampling variance of mean cluster size, E[s] can be obtained in a regression frame-work, so no probability model π(s) for cluster size s in the population need be assumed. Probability models and likelihood inference are conventionally used only for the distance part of the data.