Researchers often test firm conduct models using pass-through regressions or instrumental variables (IV) methods. The former has limited applicability; the latter relies on potentially irrelevant instruments. We show the falsifiable restriction underlying the IV method generalizes the pass-through regression, and cost pass-through differences are the economic determinants of instrument relevance. We analyze standard instruments' relevance and link instrument selection to target counterfactuals. We illustrate our findings via simulations and an application to the Washington marijuana market. Testing conduct using targeted instruments, we find the optimal ad valorem tax closely matches the actual rate.