Non-response is a large and growing problem in survey research. Weighting addresses non-response associated with measured variables, but may exacerbate non-response bias associated with unmeasured factors. Selection models correct for non-response related to measured and unmeasured factors, but often prove unreliable for conventional survey data when no variable can credibly be assumed to affect response but not the outcome of interest. This paper presents survey techniques that generate the information needed to make selection models function properly. First, by randomly treating potential respondents with questions that affect response propensity, we produce a variable that explains response, but does not affect outcome variables directly. Second, by eliciting behavioral willingness to respond to political questions independent of the content of response, we can directly assess the relationship between response willingness and outcomes. Results from surveys using these tools demonstrate their simplicity and potential for identifying substantial non-response bias, especially among partisan subsamples.