Scholars have invested a great deal of effort in trying to estimate the impact of political campaigns on the public. While progress has been made, one fundamental problem continues to plague our attempts to study campaigns: the lack of good, detailed data about the behaviour of candidates. In the United States, for instance, the presidential battle is a national struggle that unfolds locally on the stages of fifty states, yet most data collection efforts have treated presidential elections as if they were national contests, conducted (implicitly) in an identical manner across the entire country. Using available (national) data as a baseline and theory to predict plausible variations from that baseline, the authors devise a method for simulating variation in presidential campaigns across states and over election years—one of the crucial missing pieces of the puzzle. Their method generates a range of plausible effects, which is often narrow enough to shed light on important hypotheses. It can be employed whenever data are available at a more aggregate level than is desirable. This method is then applied to assess the debate over the impact of attack advertising on turnout. This approach suggests that campaign negativism stimulated (rather than demobilized) turnout in presidential elections from 1980 through 2000.