Bibliography

Abadie and Gardeazabal introduce the synthetic control idea in the context of estimating the economic costs of terrorism in the Basque Country. They construct a synthetic Basque Country from other Spanish regions and show that terrorism reduced GDP per capita by about 10 percentage points.

Abadie and Imbens derive the large-sample properties of nearest-neighbor matching estimators, showing that such estimators are not root-N consistent in general and do not attain the semiparametric efficiency bound. Their main practical contribution is a consistent analytical variance estimator that does not require nonparametric estimation of unknown functions. Bootstrap invalidity for matching is established separately in Abadie and Imbens (2008), and the bias-corrected matching estimator is developed in Abadie and Imbens (2011).

Abadie and Imbens show that the standard bootstrap is inconsistent for nearest-neighbor matching estimators with a fixed number of matches, even though these estimators are asymptotically normal. Researchers should use the analytical variance estimator from Abadie and Imbens (2006) instead of bootstrapping.

Abadie, Diamond, and Hainmueller formalize and popularize the synthetic control method, which constructs a weighted combination of control units to approximate the counterfactual for a single treated unit. The application to California's Proposition 99 tobacco control program becomes the canonical example of the method.

Abadie and Imbens develop bias-corrected matching estimators that adjust for the finite-sample bias inherent in nearest-neighbor matching when matching is not exact. Their bias correction uses a regression adjustment within matched pairs and has become a standard recommendation for applied researchers using matching methods.

Abadie et al. distinguish between sampling-based uncertainty (from drawing a sample from a population) and design-based uncertainty (from treatment assignment) in regression analysis. They show that conventional standard errors can be conservative when the sample includes a substantial fraction of the population, providing a rigorous framework for understanding what regression standard errors actually measure. This paper clarifies the conceptual foundations for inference in empirical work and complements their separate 2023 QJE paper on clustering.

Abadie et al. provide guidance on when clustering standard errors is necessary. They show that clustering can be motivated by sampling-based uncertainty (e.g., two-stage sampling of clusters then units) or design-based uncertainty (e.g., treatment assigned at the cluster level), and that whether to cluster, and at what level, is a substantive question tied to the sampling and assignment process — not a purely mechanical rule.

Acharya, Blackwell, and Sen develop a sequential g-estimation approach for estimating controlled direct effects in observational studies, addressing the problem that conditioning on a post-treatment mediator can introduce collider bias. Their method is particularly useful in political science and social science settings where intermediate confounders make standard mediation analysis unreliable.

Adao, Kolesar, and Morales show that standard errors in shift-share regressions are too small when computed with conventional clustering because residuals are correlated across regions that share similar industry compositions. They propose an inference procedure that accounts for this dependence.

Ai and Norton show that the interpretation of interaction terms in nonlinear models like logit and probit is much more complicated than in linear models. The marginal effect of an interaction is not simply the coefficient on the interaction term, a mistake that is widespread in applied research.

Akerlof proposes the gift exchange model of labor markets, in which firms pay above-market wages and workers reciprocate with above-minimum effort. This framework provides a behavioral foundation for efficiency wages and has been tested extensively in laboratory and field experiments.

Allison shows that naive comparisons of logit or probit coefficients across groups are misleading because differences in residual variation across groups rescale the coefficients. He proposes a method to adjust for this confound, which is essential for interpreting interaction effects and group comparisons in nonlinear models.

Altonji, Elder, and Taber develop the idea that if selection on observables is informative about selection on unobservables, one can bound the bias from omitted variables. Their approach becomes the basis for the widely used Oster (2019) sensitivity framework.

Amemiya provides a comprehensive survey of qualitative response models including logit, probit, and tobit. This survey organizes the theoretical properties, estimation methods, and specification tests for binary and multinomial choice models and becomes a standard reference for applied researchers.

Anderson proposes using summary index tests and familywise error rate corrections to address multiple inference in program evaluation. Reanalyzing the Abecedarian, Perry Preschool, and Early Training Projects, he finds that girls garner substantial short- and long-term benefits from early interventions, but there are no significant long-term benefits for boys after correcting for multiple testing.

Angrist uses the Vietnam-era draft lottery as a natural experiment in this landmark application of instrumental variables. He shows that randomly assigned lottery numbers provide an instrument for military service, allowing causal estimation of the earnings effect of military service.

Angrist and Krueger use quarter of birth as an instrument for years of schooling, exploiting the fact that compulsory schooling laws interact with birth timing. This paper is one of the most-taught examples of instrumental variables in economics and also sparks important debates about weak instruments.

Angrist, Imbens, and Rubin formalize the LATE framework — originally introduced in Imbens and Angrist (1994) — within the Rubin Causal Model, providing a detailed treatment of the assumptions required for causal interpretation of IV estimates. This paper introduces the complier taxonomy (always-takers, never-takers, compliers, defiers) that is now standard in the IV literature. The practical implication is that IV estimates should be interpreted as local to the complier subpopulation, not as average effects for the entire population.

Arkhangelsky et al. introduce the synthetic difference-in-differences estimator, which combines the strengths of DID (parallel trends assumption) and synthetic control (re-weighting to improve pre-treatment fit). The method uses both unit weights and time weights to construct a more credible counterfactual, and provides valid inference without requiring a large donor pool.

Arkhangelsky and Imbens develop doubly robust identification strategies for causal panel data models, combining outcome modeling with re-weighting to provide consistent estimates if either the outcome model or the weighting scheme is correctly specified. The framework is broader than synthetic DID specifically but directly relevant to it, strengthening the theoretical foundations for panel-data treatment effect estimation.

Ashenfelter provides one of the earliest applications of the difference-in-differences logic, comparing the earnings of trainees before and after a job training program to a comparison group. The key insight is that differencing removes time-invariant unobserved differences between treatment and control groups. This paper also documents the 'Ashenfelter dip' — the pre-program earnings decline among trainees — which becomes a canonical example of why parallel trends cannot be taken for granted.

Athey and Imbens introduce causal trees, adapting the CART algorithm to estimate heterogeneous treatment effects with valid inference. They propose the honest estimation approach, where one subsample is used for tree construction and another for estimation, ensuring valid confidence intervals.

Athey and Imbens provide a modern, rigorous treatment of the econometrics behind randomized experiments. They cover design, analysis, and inference issues such as stratification, clustering, and multiple hypothesis testing. It is an excellent reference for researchers running field experiments.

Athey, Tibshirani, and Wager introduce the generalized random forest (GRF) framework, which extends causal forests to a broad class of estimating equations including quantile regression, IV, and local average treatment effects. GRF provides the theoretical foundation and the widely used grf R package.

Bach and colleagues develop the DoubleML Python package, providing a user-friendly object-oriented implementation of the DML framework. The package supports partially linear, interactive, and instrumental variable models with a variety of machine learning methods for nuisance estimation. A companion R package is described separately.

Bang and Robins provide an accessible exposition of doubly robust estimators, demonstrating their properties through simulations and clarifying when the double robustness property provides meaningful protection. This paper helps make the method more accessible to applied researchers.

Baron and Kenny introduce the widely used four-step approach to testing mediation, comparing total, direct, and indirect effects using sequential regressions. While later work has identified limitations of this approach, it remains one of the most cited papers in all of social science.

Bartik introduces the shift-share instrument—constructing predicted local employment growth from national industry growth rates interacted with initial local industry composition. This 'Bartik instrument' has become one of the most widely used instruments in labor and urban economics.

Battistin and Rettore propose using ineligible units and eligible non-participants as a double comparison group in regression-discontinuity designs. This specification-testing strategy allows researchers to assess the validity of RDD assumptions by checking whether the two comparison groups yield consistent estimates, strengthening the credibility of RDD-based inference.

Bell and Jones argue that the 'within-between' random-effects model (closely related to the Mundlak approach) can outperform pure fixed effects in certain settings because it allows explicit decomposition of within- and between-unit effects while accounting for unobserved heterogeneity. This approach retains the unbiasedness of the within estimator for time-varying regressors while also estimating between-unit effects that fixed effects discard. The paper provides practical guidance for researchers who need to estimate both types of effects or who have time-invariant regressors that fixed effects cannot identify.

Belloni, Chernozhukov, and Hansen introduce the post-double-selection LASSO method for inference on treatment effects with many potential controls. This paper is a key precursor to DML, demonstrating how regularized selection in both the treatment and outcome equations can yield valid inference.

Ben-Michael, Feller, and Rothstein propose augmenting the synthetic control estimator with an outcome model to reduce bias when the synthetic control does not achieve perfect pre-treatment fit. The resulting doubly robust estimator is consistent if either the outcome model or the weighting is correct, providing a practical improvement for applied synthetic control studies.

Ben-Michael, Feller, and Rothstein extend synthetic control and synthetic DID methods to staggered adoption settings where multiple units adopt treatment at different times. They demonstrate the approach by estimating the effects of teacher collective bargaining laws on school spending across U.S. states, showing how synthetic DID-style reweighting improves counterfactual estimation when treatment rolls out over time.

Benjamini and Hochberg introduce the false discovery rate (FDR) as an alternative to family-wise error rate control. Their step-up procedure for controlling FDR is less conservative than Bonferroni while still providing meaningful protection against false positives, and has become the standard in many fields.

Bertrand, Duflo, and Mullainathan show that standard errors in DID studies are often far too small because they ignore serial correlation within units over time. They propose clustering standard errors at the group level as a simple fix, which is now widely recommended practice in DID applications.

Blomquist, Newey, Kumar, and Liang provide a critical examination of the identification assumptions underlying bunching estimation. They show that the standard bunching estimator identifies the elasticity only under strong assumptions about the functional form of the counterfactual density and the distribution of preferences. Without these assumptions, the amount of bunching is consistent with a range of elasticities. The paper sparks an important methodological debate about what bunching can and cannot identify, and motivates subsequent work on tightening identification in bunching designs.

Bloom introduces the minimum detectable effect (MDE) framework, which reports the smallest effect size a study can reliably detect given its design and sample size. This approach is now the standard way to discuss statistical power in program evaluation and experimental economics.

Bonferroni develops the classical correction for multiple comparisons, which controls the family-wise error rate by dividing the significance level by the number of tests. While conservative, the Bonferroni correction remains widely used due to its simplicity and broad applicability.

Borusyak, Hull, and Jaravel provide an alternative framework where identification comes from the exogeneity of the shocks rather than the shares. They show that with many independent shocks, the instrument is valid even if shares are endogenous, greatly expanding the range of credible applications.

Borusyak, Jaravel, and Spiess propose an imputation estimator for staggered DID that first estimates unit and time fixed effects from untreated observations, then imputes the counterfactual outcomes. This approach is efficient, flexible, and avoids the negative weighting problem of TWFE.

Bound, Jaeger, and Baker demonstrate that instrumental variables estimates can be severely biased when instruments are weakly correlated with the endogenous regressor. They show that with weak instruments, the finite-sample bias of IV approaches that of OLS, and that the standard IV confidence intervals can have coverage far below their nominal levels. The paper motivates the widespread practice of reporting first-stage F-statistics as a diagnostic for instrument strength.

Brinch, Mogstad, and Wiswall show how to estimate the MTE curve semiparametrically even with a discrete (binary or multivalued) instrument, which is a common case in practice. They demonstrate that the local IV approach can be implemented with discrete instruments by imposing additive separability between observed covariates and unobserved heterogeneity along with a parametric structure on the MTE. Applied to the quantity-quality tradeoff of children using twin births and sibling sex composition as instruments for family size, they find that MTE varies with unobserved resistance to having additional children, demonstrating how discrete instruments can recover policy-relevant heterogeneity beyond LATE.

Brown and Warner extend the event study framework from monthly to daily stock returns and examine the statistical properties of various test statistics. Their simulations show that simple methods perform well in most settings, providing practical reassurance for applied researchers.

Bruhn and McKenzie compare different randomization methods—simple, stratified, and pairwise—in practice and show that stratified randomization substantially improves balance on baseline covariates and increases statistical power. They provide practical recommendations for choosing among randomization procedures in field experiments.

Buchanan and colleagues develop inverse probability of sampling weighted (IPSW) estimators for generalizing treatment effect estimates from randomized trials to well-defined target populations, and derive consistent sandwich-type variance estimators. The method models the probability of trial participation as a function of observed covariates, reweighting trial outcomes to represent the target population. Researchers seeking to transport trial results to a broader population can apply IPSW when a probability sample or census of the target population is available for comparison.

Callaway and Sant'Anna propose group-time average treatment effects (ATT(g,t)) that avoid the problematic comparisons in TWFE. Their framework allows for heterogeneous treatment effects across groups and time and provides aggregation schemes for summary parameters.

Calonico, Cattaneo, and Titiunik develop bias-corrected confidence intervals for RDD that address the problem of conventional confidence intervals being invalid when using optimal bandwidth selectors. Their rdrobust software package has become the standard tool for implementing RDD in practice.

Cameron and Trivedi compare Poisson, negative binomial, and other count data models, providing tests for overdispersion and guidance on model selection. This paper helps establish the practical toolkit for applied researchers working with count outcomes.

Cameron and Trivedi develop regression-based tests for overdispersion in count data models, enabling formal testing of whether the Poisson equidispersion assumption holds. Their tests compare the observed variance to the Poisson-implied mean, providing the foundation for model selection between Poisson and negative binomial specifications. Researchers working with count outcomes should use these tests before defaulting to either model.

Cameron, Gelbach, and Miller address what happens when clustering is necessary but the number of clusters is small (fewer than 30-50). They propose the wild cluster bootstrap as a solution, which has become the standard approach when researchers have too few clusters for asymptotic cluster-robust standard errors to be reliable.

Cameron, Gelbach, and Miller extend cluster-robust variance estimation to settings with two-way (or multi-way) clustering. The variance estimator adds the two one-way cluster-robust variance matrices and subtracts the heteroscedasticity-robust matrix.

Card and Krueger compare fast-food employment in New Jersey (which raised its minimum wage) with neighboring Pennsylvania (which did not) in perhaps the most famous DID study in economics. They find no negative employment effect, challenging the standard textbook prediction. This paper popularizes DID as a research design.

Card, Lee, Pei, and Weber formalize the regression kink design, establishing conditions under which a kink in the treatment assignment function identifies causal effects. They show that the estimand is the ratio of the change in the slope of the conditional expectation of the outcome to the change in the slope of the treatment function at the kink point. The paper develops inference procedures and applies the method to estimate the effect of unemployment insurance benefits on unemployment duration.

Cattaneo, Drukker, and Holland extend matching and inverse probability weighting methods to settings with multi-valued (rather than binary) treatments, developing estimators for dose-response functions under conditional independence. Their accompanying Stata implementation made these methods readily accessible to applied researchers.

Cattaneo, Frandsen, and Titiunik develop a randomization inference framework for regression discontinuity designs, exploiting the local randomization interpretation of close elections. They apply the method to estimate party advantages in U.S. Senate elections, demonstrating how Fisher-style permutation tests can provide finite-sample exact inference in RDD settings where asymptotic approximations may be unreliable.

Cattaneo, Titiunik, and Vazquez-Bare provide methods and software for power calculations in RDD, essential for study design and determining adequate sample sizes near the cutoff. The associated rdsampsi command enables researchers to plan appropriately powered RDD studies before data collection.

Cattaneo, Jansson, and Ma propose a local polynomial density estimator for manipulation testing in regression discontinuity designs. Implemented in the rddensity package, it provides a modern alternative to the McCrary (2008) density test with better boundary properties.

Chamberlain extends the fixed effects approach to nonlinear models like logit, showing how to condition out the fixed effects in discrete choice settings. This work is fundamental for researchers who need fixed effects in models where the dependent variable is binary or categorical.

Chernozhukov and Hansen develop an instrumental variable framework for quantile regression to address endogeneity. Proposes the inverse quantile regression (IQR) method that exploits moment conditions implied by the structural quantile model. Provides conditions under which quantile treatment effects are identified with endogenous treatments, extending quantile regression to credible causal inference settings.

Chernozhukov et al. introduce double/debiased machine learning (DML), showing how to combine Neyman orthogonality with cross-fitting to obtain root-n consistent and asymptotically normal estimates of low-dimensional causal parameters while using high-dimensional machine learning for nuisance functions. This paper provides the theoretical foundation for valid inference when first-stage estimation uses flexible ML methods that would otherwise invalidate standard asymptotic arguments. The cross-fitting procedure it introduces is now standard practice for any application combining ML prediction with causal parameter estimation.

Chernozhukov, Wuthrich, and Zhu develop a conformal inference method for synthetic control that provides exact, finite-sample valid p-values and confidence intervals without requiring a large number of control units. This approach offers a modern, robust alternative to placebo-based inference for counterfactual and synthetic control estimators.

Chernozhukov, Escanciano, Ichimura, Newey, and Robins develop locally robust semiparametric estimators that extend the DML framework, demonstrating how automatic debiasing with machine learning first-stage estimates can be applied broadly. Their approach yields root-n consistent estimates of causal and structural parameters even when nuisance functions are estimated with regularized machine learning methods.

Cinelli and Hazlett develop a modern framework for sensitivity analysis based on partial R-squared measures, extending the omitted variable bias formula. Their approach allows researchers to benchmark the strength of hypothetical confounders against observed covariates, making sensitivity analysis more interpretable.

Cinelli, Ferwerda, and Hazlett develop the sensemakr R and Stata package implementing their partial R-squared sensitivity analysis framework. They demonstrate the tool with applications to studies of violence and political attitudes, showing how researchers can benchmark potential confounders against observed covariates to assess the robustness of causal claims from observational data.

Clarke, Romano, and Wolf develop a Stata implementation of the Romano-Wolf stepwise multiple testing correction, which controls the family-wise error rate while accounting for the dependence structure among test statistics via resampling. This correction is more powerful than Bonferroni or Holm procedures when test statistics are correlated, which is the typical case in applied research with related outcomes. The rwolf command provides applied researchers with an accessible tool for rigorous multiple hypothesis testing.

Clarke and colleagues develop the sdid Stata package for implementing synthetic DID, providing detailed documentation and empirical examples. This paper makes the method accessible to applied researchers and demonstrates implementation with real policy evaluation data.

Cohen's foundational textbook introduces the concepts of effect size, statistical power, and sample size determination that becomes standard in the behavioral sciences. He provides power tables and conventions for small, medium, and large effect sizes that remain widely used across disciplines.

Conley develops GMM estimators and nonparametric, positive semi-definite covariance matrix estimators that account for cross-sectional dependence characterized by economic or geographic distance between observations. The approach extends HAC-style inference to spatial settings by allowing error correlations to decline smoothly with distance, and the covariance estimator remains consistent even when distances are imprecisely measured. Researchers with spatially distributed data should use Conley standard errors when observations within a defined neighborhood are likely correlated.

Conley, Hansen, and Rossi develop methods for inference when the exclusion restriction is 'plausibly' rather than exactly satisfied, parameterizing the degree of violation and constructing valid confidence intervals. This approach provides a formal sensitivity analysis for IV estimates, answering the question: how large would the violation of the exclusion restriction need to be to overturn the result? Applied researchers can use these methods to transparently assess the robustness of IV findings to a common critique.

Correia develops an efficient iterative demeaning estimator for linear models with multiple high-dimensional fixed effects that scales to very large datasets. The estimator handles arbitrary numbers of fixed-effect dimensions and supports cluster-robust standard errors. Its implementation as the reghdfe Stata command has become the standard tool for applied researchers working with high-dimensional fixed effects in panel data.

Correia, Guimaraes, and Zylkin introduce the ppmlhdfe Stata command for fast Poisson estimation with multiple levels of fixed effects, making PPML feasible for large datasets with high-dimensional fixed effects. This tool has become standard for applied researchers working with count data in panel settings.

Cox introduces the proportional hazards model with an unspecified baseline hazard, estimated via a conditional likelihood argument (later formalized as partial likelihood in Cox, 1975). The semiparametric approach avoids distributional assumptions on the baseline hazard while allowing covariate effects to be estimated consistently. One of the most cited papers in statistics.

Dahabreh, Robertson, Tchetgen Tchetgen, Stuart, and Hernan develop a formal framework for generalizing causal inferences from randomized trial participants to all trial-eligible individuals in a target population, using baseline covariate data from both randomized and non-randomized individuals. They establish identifiability conditions and propose inverse probability weighting, outcome modeling, and doubly robust estimators for the target population average treatment effect. Researchers conducting trials nested within observational cohorts can apply this framework to estimate treatment effects for the full eligible population rather than only for those who enrolled.

De Chaisemartin and D'Haultfoeuille show that the TWFE estimator can assign negative weights to some treatment effects, potentially producing estimates with the wrong sign. They propose an alternative estimator and a decomposition that reveals which group-time effects receive negative weights.

Dong examines regression discontinuity designs when the running variable is subject to rounding or heaping, a common practical concern. She shows that standard RD estimators can be biased in such settings and derives correction formulas for the resulting discretization bias, extending the applicability of RDD to settings with imperfect measurement of the running variable.

Dong and Lewbel show that the derivative of the RD treatment effect with respect to the running variable at the cutoff is identified. Under a local policy-invariance interpretation, this derivative can be used to evaluate counterfactual policies that shift the eligibility threshold, broadening the policy relevance of RDD beyond the effect at the existing cutoff.

Doudchenko and Imbens place synthetic control within a broader framework that includes DID and regression as special cases, proposing extensions that relax the non-negativity and adding-up constraints on weights. This paper helps researchers understand the connections between synthetic control and other methods.

Dube and colleagues propose a local projections (LP) approach to difference-in-differences estimation that combines LPs with a flexible 'clean control' condition to define appropriate treated and control units. The LP-DiD estimator subsumes many recent solutions to negative weighting problems, accommodates covariates and nonabsorbing treatments, and is simple to implement.

Dunning provides a systematic framework for identifying and analyzing natural experiments across the social sciences. The book covers as-if random assignment, instrumental variables, regression discontinuity, and difference-in-differences through a unified design-based lens, making it essential reading for researchers exploiting natural variation for causal inference.

Fama, Fisher, Jensen, and Roll establish the modern event study methodology by studying how stock prices adjust to stock splits. They develop the framework of measuring abnormal returns around corporate events using a market model to construct the counterfactual return. This methodology has become the standard tool for studying how information events affect asset prices and is used in thousands of subsequent studies across finance and strategy.

Fan and colleagues propose nonparametric estimators for conditional average treatment effects in high-dimensional settings. Their approach uses machine learning to estimate nuisance functions in a first stage, then applies local linear regression for the CATE function of interest, with functional limit theory and multiplier-bootstrap uniform inference.

Fine and Gray develop a regression model for the cumulative incidence function under competing risks. The Fine-Gray model extends the Cox framework to settings where multiple event types compete, allowing estimation of covariate effects on the subdistribution hazard.

Firpo, Fortin, and Lemieux introduce the recentered influence function (RIF) regression for estimating unconditional quantile effects. They show that standard quantile regression estimates conditional quantile effects that do not aggregate to unconditional effects. RIF regression transforms the outcome variable so that OLS on the transformed outcome recovers the effect of covariates on unconditional quantiles. The key innovation enabling policy-relevant distributional analysis.

Firpo and Possebom develop formal inference procedures for the synthetic control method, including sensitivity analysis tools and confidence sets. Their framework provides a more rigorous basis for statistical inference in synthetic control applications beyond the standard permutation-based placebo tests.

Fisher's classic book lays the foundations of experimental design, introducing concepts like randomization, blocking, and factorial designs. The 'lady tasting tea' example from this book remains one of the most famous illustrations of hypothesis testing and the logic of controlled experiments.

Frake and colleagues introduce partial identification methods to strategic management, providing a practical framework for assessing the sensitivity of difference-in-differences and instrumental variables estimates to violations of identifying assumptions. The paper demonstrates how researchers can construct informative bounds on treatment effects when parallel trends or exclusion restriction assumptions are relaxed. It bridges the gap between the theoretical ideal of point identification and the practical reality that identifying assumptions are rarely perfectly satisfied.

Frank develops the impact threshold for a confounding variable (ITCV), which calculates how much bias an omitted variable would need to introduce to invalidate an inference. This approach is widely adopted in education and management research.

Freyaldenhoven, Hansen, and Shapiro study panel event-study designs in which unobserved confounds can generate pre-event trends. They show how causal effects can still be identified by exploiting covariates related to the policy only through the confounds, yielding a 2SLS estimator that remains valid even when endogeneity induces pre-trends.

Frisch and Waugh establish that a coefficient in a multiple regression can be obtained by first residualizing both the outcome and the regressor against all other covariates. The Frisch-Waugh-Lovell (FWL) theorem provides the theoretical foundation for understanding what 'controlling for' means in multiple regression and is the basis for modern fixed-effects estimation.

Gelman and Carlin extend traditional power analysis by introducing Type S (sign) errors (the probability a significant estimate has the wrong sign) and Type M (magnitude) errors (the expected exaggeration ratio of significant estimates). These concepts provide a richer understanding of what happens in underpowered studies.

Gelman and Loken argue that data-dependent analysis creates a 'garden of forking paths' that explains why many statistically significant comparisons do not hold up. They emphasize that researchers' analytical choices conditional on data characteristics inflate false positive rates even without deliberate p-hacking.

Gelman and Imbens show that using high-order global polynomials in RDD leads to noisy estimates, sensitivity to the degree of polynomial, and poor coverage of confidence intervals. They recommend local linear or quadratic fits with appropriate bandwidth selection instead, fundamentally changing best practice for RDD estimation.

Gerard, Rokkanen, and Rothe study regression-discontinuity settings in which the running variable is manipulated, so conventional point identification fails. They show that treatment effects are still partially identified and derive sharp bounds under a general model in which the extent of manipulation is learned from the data.

Glynn and Quinn introduce the AIPW estimator to political scientists, providing intuition, simulation evidence, and practical guidance. This tutorial demonstrates the advantages of doubly robust methods over propensity score weighting or outcome regression alone in social science applications.

Gobillon and Magnac connect synthetic control to interactive fixed-effects models, showing that synthetic control can be interpreted as an estimator that allows for time-varying factor loadings. This paper bridges the synthetic control and factor model literatures.

Goldsmith-Pinkham, Sorkin, and Swift provide a rigorous econometric framework for shift-share instruments, showing that the Bartik instrument can be decomposed into a weighted sum of individual share-based instruments. They clarify that identification requires exogeneity of the initial shares, not the shocks.

Goodman-Bacon decomposes the two-way fixed-effects DID estimator into a weighted average of all possible two-group, two-period DID comparisons, revealing that some comparisons use already-treated units as controls. The decomposition clarifies when already-treated units enter as controls and why this can make the estimator difficult to interpret under treatment-effect heterogeneity.

Gourieroux, Monfort, and Trognon develop the general theory of pseudo maximum likelihood estimation for cases in which the likelihood family may be misspecified. They derive conditions for consistency and asymptotic normality and characterize efficiency bounds in this broader framework. The Poisson PML result — consistency for the conditional mean under misspecification — is a special case that underpins the later widespread use of Poisson regression with robust standard errors.

Grambsch and Therneau introduce the scaled Schoenfeld residual test for the proportional hazards assumption. Plotting scaled Schoenfeld residuals against time reveals time-varying effects. The test is the standard diagnostic in applied survival analysis.

Griliches systematically examines the biases in OLS estimates of returns to schooling, including ability bias and measurement error. This paper is a classic illustration of why researchers must think carefully about omitted variables when interpreting OLS coefficients causally.

Hahn derives the semiparametric efficiency bound for estimating average treatment effects and shows that knowledge of the propensity score does not improve the bound—it is ancillary for ATE. The efficient estimators take the form of sample averages completed by nonparametric imputation. This paper is foundational for understanding efficient semiparametric estimation of treatment effects.

Hahn, Todd, and Van der Klaauw provide the formal econometric framework for both sharp and fuzzy regression discontinuity designs. For the fuzzy case, they show that the treatment effect can be identified as the ratio of the discontinuity in the outcome to the discontinuity in the treatment probability, analogous to a Wald estimator.

Hainmueller introduces entropy balancing, a reweighting scheme that directly targets covariate balance by finding weights that satisfy pre-specified balance constraints while remaining as close to uniform as possible. Entropy balancing has become a popular alternative to propensity score matching because it achieves exact balance on specified moments by construction.

Hamilton and Nickerson warn strategy researchers that naive OLS estimates of the strategy-performance relationship are often biased by endogeneity, because firms that adopt a strategy differ systematically from those that do not. They provide an accessible tutorial on endogeneity and point toward solutions including instrumental variables and Heckman selection models. The paper remains a key reference for understanding why strategic management research requires identification strategies beyond simple regression.

Harrison and List provide an influential taxonomy of field experiments, distinguishing artefactual, framed, and natural field experiments from conventional lab experiments. The paper helps establish field experiments as a mainstream methodology in economics.

Hausman develops a general framework for specification testing based on comparing two estimators: one consistent under a broad set of assumptions and one efficient under a narrower null hypothesis. The test's most well-known application compares fixed effects (consistent if unit effects are correlated with regressors) against random effects (efficient under the null of no correlation), but the framework applies broadly to IV, simultaneous equations, and time-series cross-section models. The test statistic has a chi-squared distribution under the null and remains one of the most widely used diagnostic tools in applied econometrics.

Hausman and Taylor develop an instrumental variables estimator for panel data that allows consistent estimation of coefficients on time-invariant variables even when individual effects are correlated with some regressors. The Hausman-Taylor estimator occupies a middle ground between fixed effects (which cannot estimate time-invariant coefficients) and random effects (which requires strict exogeneity).

Hausman, Hall, and Griliches develop the econometric framework for Poisson and negative binomial regression models applied to count data, using the relationship between R&D spending and patent counts as the motivating application. The paper is a classic early econometric treatment of count-data models in panel settings.

Hausman and McFadden develop a specification test for the independence of irrelevant alternatives (IIA) assumption in multinomial logit. The test allows researchers to assess whether the logit model's restrictive substitution patterns are appropriate for their data, which is critical for applied work with multiple choice categories.

Heckman introduces the two-step estimator for correcting sample selection bias using the inverse Mills ratio. The paper shows that selection bias can be treated as an omitted variable problem, where the omitted variable is the conditional expectation of the error term given selection. One of the most cited papers in econometrics.

Heckman, Ichimura, and Todd develop the econometric theory behind matching estimators, including conditions for identification and the importance of common support. They apply these methods to evaluate job training programs and show when matching works well and when it does not.

Heckman and Vytlacil use the marginal treatment effect (MTE) to connect the treatment-effects literature with structural econometric policy evaluation. A central result is that commonly used treatment-effect parameters (ATE, ATT, LATE, PRTE) can be expressed as weighted averages of the MTE curve, with each estimand using a different weight function. The framework shows how IV estimates with different instruments recover different weighted averages of the same underlying MTE, providing the theoretical foundation for understanding instrument-dependent variation in treatment-effect estimates.

Heß develops the ritest Stata command for randomization inference, providing a practical tool for conducting permutation tests under arbitrary randomization procedures in experimental and quasi-experimental settings. The command accommodates stratified, clustered, and blocked randomization designs, and produces exact finite-sample p-values without distributional assumptions. The paper serves as both a software introduction and a practical guide to randomization inference for applied researchers.

Ho, Imai, King, and Stuart argue that matching should be used as a preprocessing step before parametric modeling, reducing model dependence and improving robustness of causal estimates. This influential paper reframed matching not as a standalone estimator but as a way to make subsequent parametric analyses less sensitive to specification choices.

Hoenig and Heisey demonstrate that post hoc (observed) power calculations are fundamentally flawed because they are a monotone function of the p-value and add no information beyond the test result itself. This paper is essential reading for understanding why power analysis must be conducted before data collection.

Hofmann introduces hierarchical linear models to the management research community, explaining when and why multilevel random-effects models are appropriate for organizational data with nested structures. This tutorial is highly influential in promoting multilevel methods in management journals.

Holland articulates the fundamental problem of causal inference—that we can never observe both potential outcomes for the same unit—and formalizes the Rubin Causal Model framework. His dictum 'no causation without manipulation' shapes how a generation of researchers thinks about the conditions under which statistical associations can be given causal interpretations.

Hollenbeck and Wright introduce the concept of 'Tharking' (Transparently Hypothesizing After Results Are Known), arguing that post hoc analysis of scientific data is valuable when conducted and reported transparently. They distinguish destructive HARKing from constructive post hoc exploration, making the case that management researchers should embrace exploratory analysis in discussion sections rather than disguising it as confirmatory.

Horowitz and Manski extend the bounding approach to experiments with missing data on both covariates and outcomes. They show how to construct valid bounds under different assumptions about the missing data mechanism, providing a principled alternative to complete-case analysis and imputation.

Iacus, King, and Porro introduce Coarsened Exact Matching (CEM), which coarsens covariates into bins and then performs exact matching within those bins. CEM avoids many pitfalls of propensity score matching, such as the need to check balance iteratively, and gives the researcher direct control over the matching quality.

Imai, Keele, and Tingley develop a general framework for causal mediation analysis grounded in the potential outcomes framework. They clarify the assumptions needed for identifying causal mediation effects, particularly the sequential ignorability assumption, and provide sensitivity analyses for violations.

Imai and Kim provide a modern causal-inference framework for understanding when unit fixed effects regression yields unbiased estimates with longitudinal data. They clarify the often-implicit assumptions about treatment history and carryover effects, offering a more rigorous foundation for applied fixed effects analysis.

Imbens and Angrist show that IV identifies the average causal effect for compliers -- the subpopulation whose treatment status is changed by the instrument -- under the monotonicity assumption, in this foundational paper on LATE. This reinterpretation fundamentally changes how researchers understand what IV estimates.

Imbens and Manski develop methods for constructing valid confidence intervals when parameters are only partially identified—that is, when the data and assumptions narrow the parameter to a set rather than a point. This paper provides the inferential foundation for reporting uncertainty around bounds estimates, including Lee bounds.

Imbens and Lemieux provide a comprehensive practical guide to implementing RDD, covering bandwidth selection, functional form, and graphical analysis. Their treatment of fuzzy RDD as a local IV estimator clarifies the interpretation and implementation for applied researchers.

Imbens and Kalyanaraman derive the asymptotically optimal bandwidth for the local linear regression discontinuity estimator and propose a simple data-driven bandwidth selector. The IK bandwidth becomes the standard choice before the Calonico-Cattaneo-Titiunik (2014) refinement.

Jaeger, Ruist, and Stuhler highlight a threat to shift-share instruments in immigration research: serial correlation in immigrant inflows can bias estimates if past immigration affects current outcomes through channels other than current immigration. This paper raises important concerns about the exclusion restriction.

Kaplan and Meier introduce the product-limit estimator (Kaplan-Meier estimator) for the survival function from right-censored data. The KM curve is the standard nonparametric tool for visualizing survival and comparing groups before fitting regression models.

King and Zeng develop a correction for logistic regression when the outcome event is rare. Standard logit underestimates the probability of rare events; their rare-events logit (relogit) applies a correction based on prior information about the event rate in the population. Essential reference for binary outcome studies with highly imbalanced classes.

Kleven and Waseem extend bunching estimation from kinks to notches -- discrete jumps in the tax schedule where the average tax rate changes discontinuously. They develop a structural framework that distinguishes between frictionless and frictional bunching, showing that optimization frictions attenuate observed bunching and cause the naive estimator to understate the true elasticity. Their model identifies both the structural elasticity and the friction distribution from the observed bunching pattern. Applied to Pakistan's income tax notches, they demonstrate that frictions are empirically important and that ignoring them substantially biases elasticity estimates downward.

Koenker and Bassett introduce quantile regression, proposing to estimate conditional quantile functions by minimizing an asymmetric absolute loss (check function), generalizing least absolute deviations to arbitrary quantiles. Establishes asymptotic theory and demonstrates robustness to outliers and heteroscedasticity relative to OLS.

Koenker and Machado introduce a goodness-of-fit measure for quantile regression analogous to the R-squared of least squares, based on the ratio of minimized check functions across restricted and unrestricted models. They also develop related inference processes for testing composite hypotheses about covariate effects over an entire range of quantiles, with asymptotic behavior linked to Bessel processes. Practitioners estimating quantile regressions can use this pseudo-R-squared and joint significance tests to assess model fit across the conditional distribution.