We propose and develop a model of behavior in threshold public good games. The model draws on learning direction theory and impulse balance theory. We find good support for the model and demonstrate that it can explain the success rates observed in threshold public good experiments. The model is applied in a variety of different settings: we compare games with a full refund to those with no refund, consider changes in relative endowment, and consider changes in the step return and net reward. Theoretically grounded hypotheses are distinctly lacking in the literature and so our approach offers a big step forward in understanding behavior in threshold public good games.