Political Scientists routinely rely on self-reports when investigating the effects of political stimuli on behavior. An example of this is found in the American politics work addressing whether campaign advertising mobilizes voters. Findings appear to vary by methodology and are based on varying degrees of self-reports; yet, little attention is paid to the furtive complications that arise when self-reports are used as both dependent and independent variables. In this paper, I demonstrate and attempt to account for the correlated yet unobservable errors that drive self-reports of advertising exposure and political behavior. The results are from a randomized survey experiment involving approximately 1500 respondents. Before the 2002 elections, I showed a professionally developed, non-partisan, get-out-the-vote advertisement to a random subset of a randomly drawn national sample via televisions in their own homes. The analysis shows a great divide between the true effect (using assigned treatment and validated vote) and results using respondent recall of these activities.