Sessions

Six two-hour sessions: lecture, discussion, and guided STATA lab

Readings marked as Optional are for further study and are not required for class discussion. Slides, do-files and data are added to each session as the course progresses.

Session 1 — Beyond control variables

Regression, identification and selection on observables

What a regression coefficient actually estimates. From “controlling for” to identification: omitted variable bias in explicit form, DAGs as a tool for choosing controls, good and bad controls (colliders, mediators, post-treatment bias). Alternatives to regression under selection on observables: matching, weighting (IPW), doubly robust estimation. What we can say when ignorability is not credible: sensitivity analysis.

Materials

Slides, do-file and data will be posted here.

Slides Session 1

Readings

  • Keele, L., Stevenson, R. T., & Elwert, F. (2020). The Causal Interpretation of Estimated Associations in Regression Models. Political Science Research and Methods, 8(1), 1–13. https://doi.org/10.1017/psrm.2019.31
  • Huntington-Klein, N. (2022). The Effect: An Introduction to Research Design and Causality. CRC Press, chs. 5, 8 and 13. Open access: https://theeffectbook.net

Optional readings

  • Cinelli, C., & Hazlett, C. (2020). Making Sense of Sensitivity: Extending Omitted Variable Bias. Journal of the Royal Statistical Society Series B, 82(1), 39–67. https://doi.org/10.1111/rssb.12348
  • Montgomery, J. M., Nyhan, B., & Torres, M. (2018). How Conditioning on Posttreatment Variables Can Ruin Your Experiment and What to Do About It. American Journal of Political Science, 62(3), 760–775. https://doi.org/10.1111/ajps.12357

Session 2 — Measuring concepts

From exploratory factor analysis to CFA, SEM and measurement equivalence

Measurement error and its consequences for regression estimates. From the exploratory factor analysis covered in the first year to the confirmatory model: specification, identification, fit indices, modification indices and the risk of specification search. Structural equation models with latent variables. The central problem of comparative research: measurement equivalence across countries and over time (configural, metric and scalar invariance), and what to do when it does not hold.

Materials

Slides, do-file and data will be posted here.

Material Session 2

Readings

  • Acock, A. C. (2013). Discovering Structural Equation Modeling Using Stata. Stata Press, chs. 1 and 5.3.

Optional readings

Session 3 — The experimental benchmark

Experiments, survey experiments and conjoint designs

Randomization as the solution to the identification problem: what changes, in a regression, when treatment is assigned by the researcher. Laboratory, field and survey experiments. The main survey-experimental designs in research on public opinion and electoral behavior: vignette, framing, priming, list and endorsement experiments. Conjoint designs: profile randomization, AMCEs and marginal means, and the assumptions that underpin them (stability, no carryover or order effects, the distribution of attributes). Subgroup analysis and effect heterogeneity: interactions, marginal effects, and the most common errors in their specification and interpretation. Statistical power, pre-registration, multiple comparisons. Demand effects, non-probability samples and external validity.

Materials

Material Session 3

–> –> –>

Required readings

  • Mutz, D. C. (2011). Population-Based Survey Experiments. Princeton University Press, chs. 1.

  • Dunning, T. (2012). Natural experiments in the social sciences a design-based approach. Cambrdige University Press, ch. 1.

  • Levy Paluck, E. (2010). The promising integration of qualitative methods and field experiments. The ANNALS of the American Academy of Political and Social Science, 628(1), 59-71. https://doi.org/10.1177/0002716209351510

Session 4 — From fixed effects to difference-in-differences

DiD, parallel trends

DiD as a natural extension of the panel models students already know, and as an attempt to reconstruct the counterfactual that an experiment guarantees by design. The parallel trends assumption: what it implies, how it can be argued for, and why it cannot be tested directly. Event-study specifications and pre-trends.

Materials

Material Session 4

Required readings

Session 5 — Instrumental variables

Endogeneity, relevance, exclusion

Endogeneity as a structural problem and the logic of two-stage least squares. The three assumptions (relevance, exclusion, monotonicity) and their epistemic status: two of them can be defended only through argument. What an instrument actually estimates: the LATE, compliers, and the problem of the target population, with an explicit return to the effect heterogeneity discussed in Session 3. Weak instruments and finite-sample bias. A critical review of the instruments most widely used in political science (weather, historical instruments, shift-share) and of the critiques they have attracted. Experimental encouragement designs and regression discontinuity as special cases.

–> –> –>

Session 6 — From numbers to cases

Multi-method research, mechanisms and transparency

The inherent limits of the quantitative toolkit: what a well-identified estimate still cannot tell us. Nested analysis as a research architecture. Case selection following a statistical analysis: typical, deviant and influential cases, and the logic behind the choice. Process tracing as a tool for investigating mechanisms. Qualitative evidence in support of measurement validity (returning to Session 2), the interpretation of experimental treatments (Session 3), and arguments about identifying assumptions (Sessions 4 and 5). Quant–qual and qual–quant sequences. Transparency, replicability and open data.

The session includes a workshop in which students present the design of their final paper.

–> –> –>