STA 290 Seminar: Dalia Ghanem

Dalia Ghanem

Event Date

Location
Mathematical Sciences 1147 (Colloquium Room)

SPEAKER:  Dalia Ghanem; Agricultural & Resource Economics, UC Davis

TITLE:  “Causal Inference using Panel Data with Applications to Experimental and Environmental Economics”

ABSTRACT:  By providing repeated measures of the same cross-sectional units over time, panel data allow empirical researchers to control for individual and/or time heterogeneity, thereby requiring weaker assumptions for causal identification.  Previous work establishes nonparametric identifying assumptions for average partial effects in fully nonseparable panel models.  These assumptions fall under two categories, nonparametric correlated random effects and time homogeneity, which restrict individual or time homogeneity and are thereby suitable for causal inference questions using quasi-experimental and observational data, respectively.  Ghanem (2017) derives the testable restrictions of these assumptions on the outcome distribution and proposes a bootstrap procedure to obtain critical values for the resulting Kolmogorov–Smirnov and Cramer–von-Mises statistics.  Using this framework, Ghanem, Hirshleifer and Ortiz-Becerra (2019) examine the testing procedures used in the field experiments literature in economics to test attrition (non-response) bias and find no consensus in the literature. Identifying assumptions for treatment effects of both the respondent subpopulation and the study population (which fall under the correlated-random-effects category) are then established and their sharp testable restrictions are derived.  This paper also demonstrates that the most commonly used test does not control size in general when internal validity holds.  Finally, panel data are used extensively to estimate weather and climate change impacts on various outcomes.  Cui, Ghanem and Kuffner (2019) formalize the model selection problem in this empirical setting and provide conditions for model selection consistency of Monte Carlo cross-validation and information criteria.  Previous work has established that the Bayesian information criterion (BIC) can be inconsistent for non-nested model selection. This paper illustrates that the BIC can also be inconsistent in this framework, when all candidate models are misspecified.  These results have practical implications for empirical conventions in climate change impact studies. Specifically, they highlight the importance of a priori information provided by the scientific literature to guide the models considered for selection.

* * * * *

DATE:                    Thursday, November 14th, 4:10pm

LOCATION:          MSB 1147, Colloquium Room

REFRESHMENTS: 3:30pm MSB 4110 (4th floor lounge)