We develop a mathematical framework aimed at analyzing repeat and near-repeat effects in crime data. Parsing burglary data from Long Beach, CA according to different counting methods, we determine the probability distribution functions for the time interval τ between repeat offenses. We then compare these observed distributions to theoretically derived distributions in which the repeat effects are due solely to persistent risk heterogeneity. We find that risk heterogeneity alone cannot explain the observed distributions, while a form of event dependence (boosts) can. Using this information, we model repeat victimization as a series of random events, the likelihood of which changes each time an offense occurs. We are able to estimate typical time scales for repeat burglary events in Long Beach by fitting our data to this model. Computer simulations of this model using these observed parameters agree with the empirical data.