Contributions to public goods simulated in economists' laboratoryexperiments have two peculiarities from the perspective ofstatistical modelling. There is a variety of contributor behaviours(Ledyard, 1995), suggestive perhaps of separate classes ofindividuals, and contributions are doubly censored. We present aneconometric model of contributions in sequential play, which takesinto account the censoring, admits variation both within and betweenindividuals, and allows for the existence of a distinct class offree-riders. The model synthesises the 2-limit tobit analysis ofNelson (1976), the extension of tobit to panel techniques by Kim andMaddala (1992) and the p-tobit hurdle model of Deaton and Irish(1984). We estimate it for panel data from a public good experimentreported in Bardsley (2000). It reveals pronounced inter- and intra-individual variation, and shows significant effects for subjects'order in a sequential game, others' contributions and the position ofthe choice task within the experiment. These effects are plausiblyattributable to egoism, reciprocity and learning respectively. Inaddition, the existence of a distinct class of free-riders, whoconform to a game theoretic prediction of unconditional non-contribution, is confirmed. The model is estimated for tasks inwhich others' behaviour was controlled by the experimenter (butwithout using deception). We compare its predictions for actual play(in which others' behaviour is not controlled) with behaviour in areal game task. The predictions are consistent with the data.