Scholars and commentators alike have expressed concern about the risks associated with the election of radical right populist parties to legislatures. I study the effect of populist entry on legislator speech by building a data-driven, low-assumption populism detector pipeline, using document embeddings to encode the features of legislative speech, large language models to annotate training data, and model selection to estimate the relationship between the features and labels. Drawing on the exogenous entry timing of the Alternative für Deutschland into the sixteen German state legislatures and a novel corpus of all speeches given in these legislatures from 2008 to 2024, I find that non-AfD parties adopt more populist legislative speech in response to AfD entry. The effect is positive, moderately large (3 p.p., 61% relative to baseline), and robust to inferential threats, including spillover. These findings substantiate concerns about downstream effects of populist entry and highlight speech contagion as a mechanism sustaining institutional equilibrium. In addition, this article offers a workflow and guidance for integrating artificial intelligence tools into computational text analyses.

Why is there so little money in U.S. politics? The small sums firms give relative to the stakes suggest donations are political "consumption" rather than strategic investment. We argue that resolving this puzzle requires measuring not how much firms give, but how well they target. Existing studies match each industry to one or a few committees the research deems "relevant", underestimating the extent of strategic giving by overlooking a firm's full interests. Linking 1.6 million lobbying reports to 9,345 firms, their PACs, and donations, we trace the bills lobbied by industry peers to the committees that consider them. The result is a data-driven, continuous measure of every committee's relative importance to any firm, computable at the subcommittee level and at any industry aggregation. We find that firm interests span many committees, that conventional measures miss much of this exposure, and that corporate PACs target high-exposure legislators more aggressively than previously estimated.

Recent research demonstrates that national, state, and local election outcomes are becoming closely correlated, raising concerns that political accountability may be undermined by voters making electoral decisions using information irrelevant to subnational governance. The observational and aggregate data used to make these conclusions, however, are consistent with an alternative: the dimensions of national and subnational politics are correlated, and voters make informed decisions with imperfect but useful information. We distinguish between these possibilities using conjoint experiments varying the contested office level and candidate policy positions across national, state, and local politics. We find national issues more strongly affect candidate selection, but do not eliminate the effects of state and local issues. Additionally, national, state, and local issues affect candidate selection for all levels of office. Our results suggest U.S. politics is nationalized, but not fundamentally broken, with meaningful opportunities for the enforcement of political accountability in low-information settings.