Search

We study the finite sample behavior of Lasso-based inference methods such as post double Lasso and debiased Lasso. We show that these methods can exhibit substantial omitted variable biases (OVBs) due to Lasso not selecting relevant controls. This phenomenon can occur even when the coefficients are sparse and the sample size is large and larger than the number of controls. Therefore, relying on the existing asymptotic inference theory can be problematic in empirical applications. We compare the Lasso-based inference methods to modern high-dimensional OLS-based methods and provide practical guidance., Comment: Final author version, accepted at The Review of Economics and Statistics

We theoretically analyze the problem of testing for p‐hacking based on distributions of p‐values across multiple studies. We provide general results for when such distributions have testable restrictions (are non‐increasing) under the null of no p‐hacking. We find novel additional testable restrictions for p‐values based on t‐tests. Specifically, the shape of the power functions results in both complete monotonicity as well as bounds on the distribution of p‐values. These testable restrictions result in more powerful tests for the null hypothesis of no p‐hacking. When there is also publication bias, our tests are joint tests for p‐hacking and publication bias. A reanalysis of two prominent data sets shows the usefulness of our new tests.

We propose a robust method for constructing conditionally valid prediction intervals based on models for conditional distributions such as quantile and distribution regression. Our approach can be applied to important prediction problems, including cross-sectional prediction, k-step-ahead forecasts, synthetic controls and counterfactual prediction, and individual treatment effects prediction. Our method exploits the probability integral transform and relies on permuting estimated ranks. Unlike regression residuals, ranks are independent of the predictors, allowing us to construct conditionally valid prediction intervals under heteroskedasticity. We establish approximate conditional validity under consistent estimation and provide approximate unconditional validity under model misspecification, under overfitting, and with time series data. We also propose a simple "shape" adjustment of our baseline method that yields optimal prediction intervals.

In many health care markets, physicians can respond to changes in reimbursement schemes by changing the volume (volume response) and the composition of services provided (substitution response). We examine the relative importance of these two behavioral responses in the context of physician drug dispensing in Switzerland. We find that dispensing increases drug costs by 52% for general practitioners and 56% for specialists. This increase is mainly due to a volume increase. The substitution response is negative on average, but not significantly different from zero for large parts of the distribution. In addition, our results reveal substantial effect heterogeneity.

This paper analyzes estimators based on the instrumental variable quantile regression (IVQR) model (Chernozhukov and Hansen, 2004, 2005, 2006) under the local quantile treatment effects (LQTE) framework (Abadie et al., 2002). I show that the quantile treatment effect (QTE) estimators in the IVQR model are equivalent to LQTE for the compliers at transformed quantile levels. This transformation adjusts for differences between the subpopulation-specific potential outcome distributions that are identified in the LQTE model. Moreover, the IVQR estimator of the average treatment effect (ATE) corresponds to a convex combination of the local average treatment effect (LATE) and a weighted average of LQTE for the compliers. I extend the analysis to more general setups that allow for partial failures of the LQTE assumptions, non-binary instruments, and covariates. The results are illustrated with two empirical applications.

Quantile and quantile effect functions are important tools for descriptive and causal analyses due to their natural and intuitive interpretation. Existing inference methods for these functions do not apply to discrete random variables. This paper offers a simple, practical construction of simultaneous confidence bands for quantile and quantile effect functions of possibly discrete random variables. It is based on a natural transformation of simultaneous confidence bands for distribution functions, which are readily available for many problems. The construction is generic and does not depend on the nature of the underlying problem. It works in conjunction with parametric, semiparametric, and nonparametric modeling methods for observed and counterfactual distributions, and does not depend on the sampling scheme. We apply our method to characterize the distributional impact of insurance coverage on health care utilization and obtain the distributional decomposition of the racial test score gap. We find that universal insurance coverage increases the number of doctor visits across the entire distribution, and that the racial test score gap is small at early ages but grows with age due to socio economic factors affecting child development especially at the top of the distribution. These are new, interesting empirical findings that complement previous analyses that focused on mean effects only. In both applications, the outcomes of interest are discrete rendering existing inference methods invalid for obtaining uniform confidence bands for observed and counterfactual quantile functions and for their difference -- the quantile effects functions., Comment: 38 pages, 6 figures, 4 tables

Factorial designs are widely used to study multiple treatments in one experiment. While t-tests using a fully-saturated “long” model provide valid inferences, “short” model t-tests (that ignore interactions) yield higher power if interactions are zero, but incorrect inferences otherwise. Of 27 factorial experiments published in top-5 journals (2007–2017), 19 use the short model. After including interactions, over half of their results lose significance. Based on recent econometric advances, we show that power improvements over the long model are possible. We provide practical guidance for the design of new experiments and the analysis of completed experiments.

Factorial designs are widely used for studying multiple treatments in one experiment. While “long” model t-tests provide valid inferences, “short” model t-tests (ignoring interactions) yield higher power if interactions are zero, but incorrect inferences otherwise. Of 27 factorial experiments published in top-5 journals (2007–2017), 19 use the short model. After including all interactions, over half their results lose significance. Modest local power improvements over the long model are possible, but with lower power for most values of the interaction. If interactions are not of interest, leaving the interaction cells empty yields valid inferences and global power improvements.

This paper proposes a practical and robust method for making inference on average treatment effects estimated by synthetic control and related methods. We develop a $K$-fold cross-fitting procedure for bias-correction. To avoid the difficult estimation of the long-run variance, inference is based on a self-normalized $t$-statistic, which has an asymptotically pivotal $t$-distribution. Our procedure only requires consistent (in $\ell_2$-norm) estimation of the parameters, which can be verified for synthetic control and many other popular estimators. The proposed method is easy to implement, provably robust against misspecification, more efficient than difference-in-differences, valid with non-stationary data, and demonstrates an excellent small sample performance.

This paper studies the estimation of quantile treatment effects based on the instrumental variable quantile regression (IVQR) model ( Chernozhukov and Hansen, 2005 ). I develop a class of flexible plug-in estimators based on closed-form solutions derived from the IVQR moment conditions. The proposed estimators remain tractable and root-n-consistent, while allowing for rich patterns of effect heterogeneity. Functional central limit theorems and bootstrap validity results for the estimators of the quantile treatment effects and other functionals are provided. Monte Carlo simulations demonstrate favorable finite sample properties of the proposed approach. I apply my method to reanalyze the causal effect of 401(k) plans.

This paper provides a review of methodological advancements in the evaluation of heterogeneous treatment effect models based on instrumental variable (IV) methods. We focus on models that achieve identification by assuming monotonicity of the treatment in the IV and analyze local average and quantile treatment effects for the subpopulation of compliers. We start with a comprehensive discussion of the binary treatment and binary IV case as for instance relevant in randomized experiments with imperfect compliance. We then review extensions to identification and estimation with covariates, multi-valued and multiple treatments and instruments, outcome attrition and measurement error, and the identification of direct and indirect treatment effects, among others. We also discuss testable implications and possible relaxations of the IV assumptions, approaches to extrapolate from local to global treatment effects, and the relationship to other IV approaches.

We study the finite sample behavior of Lasso-based inference methods such as post double Lasso and debiased Lasso. Empirically and theoretically, we show that these methods can exhibit substantial omitted variable biases (OVBs) due to Lasso not selecting relevant controls. This phenomenon can be systematic and occur even when the coefficients are very sparse and the sample size is large and larger than the number of controls. Interestingly, we also show that the OVBs can remain bounded even if the Lasso prediction errors are unbounded. We compare the Lasso-based inference methods to modern high-dimensional OLS-based methods and provide practical guidance.

The instrumental variable quantile regression (IVQR) model (Chernozhukov and Hansen (2005)) is a popular tool for estimating causal quantile effects with endogenous covariates. However, estimation is complicated by the nonsmoothness and nonconvexity of the IVQR GMM objective function. This paper shows that the IVQR estimation problem can be decomposed into a set of conventional quantile regression subproblems which are convex and can be solved efficiently. This reformulation leads to new identification results and to fast, easy to implement, and tuning‐free estimators that do not require the availability of high‐level “black box” optimization routines. Instrumental variables quantile regression contraction mapping fixed‐point estimator bootstrap C21 C26

We extend conformal inference to general settings that allow for time series data. Our proposal is developed as a randomization method and accounts for potential serial dependence by including block structures in the permutation scheme. As a result, the proposed method retains the exact, model-free validity when the data are i.i.d. or more generally exchangeable, similar to usual conformal inference methods. When exchangeability fails, as is the case for common time series data, the proposed approach is approximately valid under weak assumptions on the conformity score.

We introduce new inference procedures for counterfactual and synthetic control methods for policy evaluation. We recast the causal inference problem as a counterfactual prediction and a structural ...

The paper develops estimation and inference methods for econometric models with partial identification, focusing on models defined by moment inequalities and equalities. Main applications of this framework include analysis of game-theoretic models, regression with missing and mismeasured data, bounds in structural quantile models, and bounds in asset pricing, among others.

This chapter reviews instrumental variable models of quantile treatment effects. We focus on models that achieve identification through a monotonicity assumption in the treatment choice equation. We discuss the key conditions, the role of control variables as well as the estimands in detail and review the literature on estimation and inference. Then we consider extensions to multiple and continuous instruments, to the regression discontinuity design, and discuss the testability of the assumptions. Finally, we compare this approach to the alternative instrumental variable approach reviewed by Chernozhukov et al. (2016). Two open research problems are highlighted in the conclusion.

This paper studies estimation of conditional and unconditional quantile treatment effects based on the instrumental variable quantile regression (IVQR) model (Chernozhukov and Hansen, 2004, 2005, 2006). I introduce a class of semiparametric plug-in estimators based on closed form solutions derived from the IVQR moment conditions. These estimators do not rely on separability of the structural quantile function, while retaining computational tractability and root-n-consistency. Functional central limit theorems and bootstrap validity results for the estimators of the quantile treatment effects and other functionals are provided. I apply my method to reanalyze the effect of 401(k) plans on individual savings behavior.

We use the 2011 Fukushima accident to evaluate the impact of the perceived risks of nuclear power plants on apartment rents in Switzerland. Using online advertisements over 12 years and a difference-in-differences approach, we find a 2.3% price discount after the accident for apartments near nuclear power plants.

We estimate the impact of qualitative job characteristics on self-rated general health and the prevalence of specific illnesses. Using panel data and controlling for individual- specific effects and pre-existing health, we find that the perceived risk of unemployment, but not unemployment per se, and an imbalanced work-life have major detrimental effects on health. We calculate compensating income variations ranging from about 10 to 50 percent for a negative work condition. Our results call for an intervention at the workplace to reduce the number of ill-days and prevent long-term health damages.