Job Market Paper
Abstract: When linkage methods employed in the creation of longitudinal panels are subject to error they can be imperfect representations of reality. In this paper I study survival estimation (e.g. firm death, mortality, or emigration) when missed linkages induce error in the observed lifetime durations, and thus inconsistency in standard survival estimators. Importantly, the error introduced does not take the form of a standard competing risks model, and the methods developed here illustrate that consistent estimation can be restored without correcting the linkages. This work makes three distinct theoretical contributions: First, under a known independent linkage error process I show that the marginal distribution of time to death is non-parametrically identified from linkage error induced durations. Second, when observations on start and end dates are introduced, I show that nonparametric point identification of the joint distribution of lifetimes and linkage error is typically achieved. Third, when no restriction is placed on the dependence structure, I apply partial identification methods to derive sharp informative bounds on the marginal distribution of lifetimes. New estimators and inference methods are established across all scenarios. The methods are then applied to longitudinal business data (where linkage error occurs due to firm relocation) to show that naive estimates of death rates of new establishments can be over-estimated by as much as 10 percentage points. Finally, I discuss additional applications to the estimation of household migration and mortality where linkage error is induced by family name changes at marriage.
"Instrumental Variables with Treatment-Induced Selection: Exact Bias Results" with Felix Elwert, Probabilistic and Causal Inference: The Works of Judea Pearl, edited by Hector Geffner, Rina Dechter, and Joseph Y. Halpern. ACM Books, 2020 (Forthcoming)
Abstract: Instrumental variables (IV) estimation suffers selection bias when the analysis conditions on the treatment. In this paper, we derive exact analytic expressions for IV selection bias across a range of data-generating models, and for various selection-inducing procedures. We present four sets of results for linear models. First, IV selection bias depends on the conditioning procedure (covariate adjustment vs. sample truncation). Second, IV selection bias due to covariate adjustment is the limiting case of IV selection bias due to sample truncation. Third, in certain models, the IV and OLS estimators under selection bound the true causal effect in large samples. Fourth, we characterize situations where IV remains preferred to OLS despite selection on the treatment. These results broaden the notion of IV selection bias beyond sample truncation, replace prior simulation findings with exact analytic formulas, and enable formal sensitivity analyses.
``Elasticity Estimation in Discrete Choice Models with Population Misspecification'' (with Diwakar Raisingh)
``Efficiency in Measurement Error Models: Applications to Twin Studies and Peer Effects''
``Sanctuaries for Immigrants or Criminals? Investigating the Effects of Sanctuary Policies on Crime''
Works in Progress
``Generalized Method of Moments Estimation with Linked Data Sets''
``A New Geography for Research: Describing Heterogeneity in Research Projects via Classification of Expenditure Profiles''
``Identification of Market Size in the Estimation of Discrete Choice Models'' (with Diwakar Raisingh)
``Sufficient Statistics For Correcting Record Linkage Error''