Sensitivity Analysis

Instrumental variable (IV) analysis relies on the exclusion restriction: the instrument should affect the outcome only through the independent variable of interest. Any other causal pathways are exclusion-restriction violations and invalidate the instrument. Weather has been widely used as an IV across social science to predict many different independent variables. The use of weather to instrument different independent variables therefore represents strong prima facie evidence of exclusion-restriction concerns for studies that rely on weather IVs.

Many researchers treat ordinal scales as numeric measures. Theory guarantees only that estimates keep the same sign under any monotonic increasing transformation of an ordinal scale, which leaves important parameters underidentified. This paper develops a practical way to test how much reported empirical estimates depend on the implicit "cardinal" treatment of ordinal variables.

This paper examines whether mixing different orders of integration on the left- and right-hand sides of a balanced time-series equation prevents researchers from identifying a true relationship. Building on prior work about equation balance, the authors focus on consequences for type II error when using general error correction/autoregressive distributed lag (GECM/ADL) models and consider both the asymptotic case and finite-sample behavior. Their analysis combines theoretical results and simulations to show how model selection and standard assumption checks affect practical inference.

Unsupervised text-as-data methods are widely used in political science, yet systematic attention to preprocessing choices is scarce. Those choices — from how text is tokenized to which features are retained — can have profound effects on the results of real models applied to real data. Substantive theory is often too vague to guide feature selection in these settings, and lessons from the supervised learning literature are not necessarily applicable to unsupervised tasks.

Political science relies heavily on data from surveys, elections, and policy records. However, these sources often contain non-random measurement error, which occurs when variables are measured inaccurately in a way systematically tied to other factors.

Voter file data has revolutionized studying American political participation, but it obscures crucial information about nonregistrants. This creates inferential problems when analyzing turnout among registrants using posttreatment registration data.

Conditioning on observed covariates is a standard strategy for reducing confounding in non-experimental causal inference, but some covariates can increase rather than decrease bias. This article develops a clearer way to think about when and why that happens by decomposing omitted-variable bias into three distinct parts and studying the mechanisms that produce bias increases when conditioning.

Scholarly findings on what makes economic sanctions effective are mixed. Two potential reasons for inconsistent results are examined: the frequent omission of threats (as distinct from imposed sanctions) and sensitivity of results to model specification. The analysis treats threats and imposed sanctions separately and uses a systematic sensitivity-checking approach to reduce model dependence.

This paper tackles a central challenge in political science research: studying causal mechanisms more effectively. It identifies the limitations of common statistical approaches – they often rely on untestable assumptions and don't capture how effects unfold.