The Analysis of Time-location Sampling Study Data*

Time-location sampling (TLS) is used to collect information from hard-to-reach populations by sampling persons at locations at which they may be found. Epidemiologic studies using TLS have often been analyzed ignoring both clustering within locations and the differential probabilities that persons are sampled. I propose a weighted analysis reflecting approximate differential sampling probabilities that may permit generalization to the population of persons attending locations in the sampling frame. I illustrate the effects of clustering and weighting using TLS data from a study of men who have sex with men. Some design effects are large. In designing a TLS, investigators should think carefully about collecting information about attendance at sampling locations, so that it may be possible to estimate a participant’s probability of being sampled.