Structural Estimation in Social Science : Department Research : Department of Political Science : University of Rochester

Structural Estimation in Social Science

Structural estimation embodies the vision for research in the social sciences laid out in the 1930s by the Econometric Society and the Cowles Commission. It endeavors to use mathematics and statistics to quantify empirical relationships of interest as identified by fully-specified and internally-consistent theoretical models of decision-making. Its modern form has its roots in the field of industrial organization in economics, where the importance of institutional context and strategic considerations drove scholars to ground empirical research in economic theory in order to carefully account for all the direct and indirect (equilibrium) consequences of potential policy interventions. Yet structural estimation has found broad applicability in virtually all fields of economics and related disciplines, including political economy in recent years.

Due to its unified theoretical and empirical framework, the structural approach to empirical research offers several key benefits:

It makes all relevant assumptions (both behavioral and identifying) explicit, facilitating sensitivity analyses and clarifying opportunities for future research.
In observational studies where alternative identification strategies are not available, relying on theory can be a fruitful avenue for nevertheless making progress on important questions.
One of the central objectives of structural estimation is quantifying key unobservables of substantive interest, such as preferences and beliefs.
By design, structural models not only elucidate the theoretical mechanisms underlying empirical relationships but also summarize the weight of the evidence supporting them.
Grounded in fully-specified models of decision-making, structural estimation enables researchers to quantify both the direct and indirect, equilibrium effects of counterfactual policies or institutional reforms.

While strategies such as differences-in-differences, regression discontinuity, and randomized controlled trials have come to dominate empirical research in social science due to a desire for robust identification of causal effects, structural estimation can nonetheless be a powerful complement, helping to disentangle alternative mechanisms. And, in settings where contextual or strategic considerations threaten the validity or generalizability of such causal estimates (e.g., SUTVA violations), the structural approach may be the only suitable alternative.

In Political Science

Structural estimation is severely underrepresented in political science. This stems in large part from misconceptions and methodological debates that the discipline is appropriately still working through. But it is also a result of the simple fact that structural estimation is difficult: practitioners need proficiency in formal modeling, statistical analysis, numerical methods, and computer programming, as well as broad substantive knowledge. Despite these challenges, the approach has seen growing demand in the top journals in the discipline, across a wide range of applications, including inter- and intrastate conflict (Crisman-Cox and Gibilisco, 2018; Gibilisco and Montero, forthcoming), democratization (Abramson and Montero, 2020), and electoral competition (Kalandrakis and Spirling, 2012; Ascencio and Rueda, 2019; Frey, López-Moctezuma, and Montero, forthcoming). In addition, political economy applications of structural estimation are frequently published in the top economics journals (Merlo, Journal of Political Economy, 1997; Diermeier, Eraslan, and Merlo, Econometrica, 2003; Diermeier, Keane, and Merlo, American Economic Review, 2005; Iaryczower and Shum, American Economic Review, 2012; Francois, Rainer, and Trebbi, Econometrica, 2015; Iaryczower, Shi, and Shum, Journal of Political Economy, 2018; Spenkuch, Montagnes, and Magleby, American Economic Review, 2018; Canen, Kendall, and Trebbi, Econometrica, 2020).

At Rochester

The Department is a leader in the dissemination of structural estimation across political science. This is evidenced in our overwhelming share of related publications, by both faculty and graduate students. Going back to Signorino (American Political Science Review, 1999), recent contributions include (PhD alumni in boldface):

Tasos Kalandrakis, with Arthur Spirling, "Radical Moderation: Recapturing Power in Two-Party Parliamentary Systems," American Journal of Political Science, 56: 413-432 (2012)
Casey Crisman-Cox and Michael Gibilisco, "Audience Costs and the Dynamics of War and Peace," American Journal of Political Science, 62: 566-580 (2018)
Sergio J. Ascencio and Miguel R. Rueda, "Partisan Poll Watchers and Electoral Manipulation," American Political Science Review, 113: 727-742 (2019)
Scott F. Abramson and Sergio Montero, "Learning about Growth and Democracy," American Political Science Review, 114: 1195-1212 (2020)
Sergio Montero, with Michael Gibilisco, "Do Major-Power Interventions Encourage the Onset of Civil Conflict? A Structural Analysis," Journal of Politics, forthcoming
Anderson Frey and Sergio Montero, with Gabriel López-Moctezuma, "Sleeping with the Enemy: Effective Representation under Dynamic Electoral Competition," American Journal of Political Science, forthcoming

Furthermore, our PhD program offers virtually unparalleled training in structural modeling and estimation. This builds on our historic strengths in formal theory and statistical methods and is part of the natural evolution of the Rochester Approach. In the first year of our graduate curriculum, four required courses on formal modeling and statistical analysis lay the groundwork, providing students with a degree of sophistication rarely achieved by graduates of more traditional programs. In subsequent years, alongside several advanced theory and methods seminars, two dedicated courses on the structural approach are regularly offered:

PSCI 585 - Dynamic Models: Structure, Computation, and Estimation
PSCI 587 - Structural Modeling and Estimation

Since structural models rarely admit estimation using canned routines in popular statistical software, familiarity with a programming language and state-of-the-art computing resources is indispensable for practitioners. Students in our PhD program acquire the necessary programming skills throughout their graduate training, and they have a wealth of resources at their disposal, including licensed software and sponsored access to BlueHive, the University's high-performance Linux cluster. Detailed information about department computing can be found here.