Angrist, J., Pischke, J.-S. 2008. Mostly harmless econometrics: an empiricist’s companion. Princeton University Press, Princeton, NJ.Graduate level book with a somewhat irreverent (i.e., conversational) writing style, yet heavy on theorems and formulas. Begins with a conceptual overview of experimentation and causal inference, then covers ordinary least-squares regression and continues with a strong focus on instrumental variables. Also explores difference in differences, fixed effects, and discontinuity analyses, which are presented as special cases of regression applied as quasi-experiments. The book does not consider interactive effects, mediation, or multilevel issues that may be of interest to micro researchers.Campbell, D. T., & Stanley, J. C. 2015. Experimental and quasi-experimental designs for research. Ravenio Books.Comprehensive resources on all types of experiments. All begin with a discussion of conditions and parameters of causality and then offer discussion of various types of experiments and the various threats to validity therein.Kennedy, P. 2008. A Guide to Econometrics. Wiley-Blackwell.Textbook focused on the question of good estimates of coefficients and organized around ways to address violations of the assumptions of OLS regression. Each chapter is short and math-free followed by general notes and technical notes to fill in the details. Endogeneity is primarily handled in Chapters 9-11 and selection is discussed Chapter 17. The notes on measurement error in chapter 10 are particularly helpful. There is limited coverage of panel data and methods to estimate treatment effects.Wooldridge, J. M. 2002. Econometric Analysis of Cross Sectional and Panel Data. MIT Press.A comprehensive textbook of econometric methods with an emphasis on formulas, explicit assumptions and derivations. Chapter 4 gives an overview of endogeneity. Chapter 5 covers single equation instrumental variable techniques to address omitted variable and measurement error problems. Chapter 8 covers system of equation instrumental variable techniques to address simultaneity. Chapters 10 and 11 cover panel data with an emphasis on omitted variables and unobserved effects. Chapter 19 covers sample selection and Chapter 21 covers methods to estimate treatment effects.
文章
Angrist, J. D., & Pischke, J.-S. 2010. The credibility revolution in empirical economics: How better research design is taking the con out of econometrics. Journal of Economic Perspectives, 24(2): 3-30.Reviews the extent to which greater emphasis on experimental and quasi-experimental research designs in microeconomics studies has dramatically increased the credibility of empirical work. Point out how early studies suffer from a variety of endogeneity biases including omitted variable and reverse causality but that current econometrics is more 'design based' (i.e., focus on design is similar to that of a true randomized experiment). Define quasi-experimental methods as: instrumental variables, regression discontinuity methods, and differences-indifferences-style policy analyses. Review a number of influential studies and designs.Angrist, J. D., Imbens, G. W., & Rubin, D. B. 1996. Identification of Causal Effects Using Instrumental Variables. Journal of the American Statistical Association, 91: 444-455.Widely cited and technical econometrics paper outlining the method for establishing a local average treatment effect, LATE); an instrumental variables estimate of the effect of treatment on the population of compliers. The term 'compliers' comes from an analogy with randomized trials where some experimental subjects comply with the randomly assigned treatment protocol (e.g., take their medicine) but some do not (i.e., defiers or non-compliers). This method requires a number of critical assumptions that have been the subject of debate.Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. 2010. On making causal claims: A review and recommendations. The Leadership Quarterly, 21: 1086-1120.Aimed at micro researchers; the main argument is that causal claims can be made with non-experimental data under certain conditions. Gives a detailed introductory overview of various threats to validity, i.e., endogeneity) and statistical and design approaches that address these concerns. Evaluates 120 management articles; find that a majority, 66%-90%) did not adequately address threats to validity.Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. 2014. Causality and Endogeneity: problems and solutions. The Oxford Handbook of Leadership and Organizations, 93.A micro-oriented review of endogeneity couched in leadership examples. Presents a straightforward explanation of what endogeneity is, why it biases estimates, various threats to validity, i.e., omitting a regressor, measurement error, common method variance, omitting fixed effects, omitting selection, and simultaneity). Then explores various methods to test causal claims in non-experimental contexts with a primary focus on 2SLS.Antonakis, J., Bastardoz, N., & Rönkkö, M. 2019. On ignoring the random effects assumption in multilevel models: Review, critique, and recommendations. Organizational Research Methods, 1094428119877457.Targeted extension of Antonakis et al., 2010) to multilevel modelling techniques with both micro and macro applications. Show that endogeneity threats are introduced into multilevel models when researchers interested in estimating the “within” effect (i.e., the effect of a Level 1 regressor on a Level 1 outcome) do not correctly model the unobserved variation due to the hierarchical structure of the data. Endogeneity and the accompanying biased results arise out of violations of the random effects assumption (see Wooldridge, 2013), which is a testable assumption that is not generally adequately addressed in management articles, demonstrated through a review of a random sample of 204 micro and macro articles). Provide detailed explanations and techniques along with extensive appendices including links to an explanatory video, mathematical derivations, a decision chart for different analyses, and Stata and R code for simulations with different degrees of endogeneity.Arellano, M., & Bond, S. 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The Review of Economic Studies, 58: 277-297.Derivation of a consistent generalized method of moments (GMM) estimator for dynamic panel data models. Uses lagged values of the dependent variable as instruments after a first-differences transformation. The core idea is that the model is specified as a system of equations, one per time period, where the instruments applicable to each equation differ, for instance, in later time periods (additional lagged values of the instruments are available). This the basis of the xtabond and xtdpd command in Stata and the pgmm procedure in R.Bascle, G. 2008. Controlling for endogeneity with instrumental variables in strategic management research. Strategic Organization, 6: 285-327.Overview of endogeneity aimed at macro scholars and building off of Hamilton and Nickerson (2003) and Shaver (1998). Defines three situations that lead to endogeneity, i.e., errors-in-variables, omitted variables, simultaneity) and proposes that selecting between the Heckman two-step procedure and instrumental variables estimations should depend on type of endogeneity present and the emergence of specification issues or violations of specific assumptions, which are reviewed in detail. Provides STATA commands for tests and methods reviewed in the article.Bliese, P. D., Schepker, D. J., Essman, S. M., & Ployhart, R. E. 2019. Bridging Methodological Divides Between Macro-and Microresearch: Endogeneity and Methods for Panel Data. Journal of Management, 0149206319868016.Overview of panel data methods meant to bridge the gap in techniques and terminology between micro and macro researchers. Endogeneity is not a primary topic of the paper, but is discussed as it applies to panel data through two forms of unobserved heterogeneity: 1) across the units of observation and 2) over time. The conclusion is macro and micro researchers tend to use different analytic tools which very in effectiveness for addressing these forms of endogeneity. Provides an in-depth overview of various analytic tools as they apply to modeling frameworks concerning different types of research questions. Review 142 articles that used panel data in leading management journals in 2017 and divide analysis across micro and macro applications.Blundell, R., & Bond, S. 1998. Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87: 115-143.Based on a potential weakness in the Arellano and Bond (1991) approach, that the lagged levels are often rather poor instruments for first differenced variables, this paper builds off of Arellano & Bover, 1995 to present a modification of the estimator that includes lagged levels as well as lagged differences. This addition requires additional restrictions on the initial conditions of the process generating the outcome. This is the basis of the xtdpdsys command in Stata.Blundell R & Dias MC. 2009. Alternative approaches to evaluation in empirical microeconomics. Journal of Human Resources, 44:565–640A descriptive (i.e., non-technical) review of some of the most popular methods used in econometric studies: social experiments, natural experiments, matching methods, instrumental variable methods, discontinuity design methods, and control function methods. Focus on selection models and various parameters of interest including the average treatment effect (ATE) the average effect on individual assigned to treatment (ATT) and the average effect on nonparticipants (ATNT). An overview of technical applications and in-depth reviews of exemplar applications are provided.Certo, S. T., Withers, M. C., & Semadeni, M. 2017. A tale of two effects: Using longitudinal data to compare within‐and between‐firm effects. Strategic Management Journal, 38: 1536-1556.Focuses on the distinction between sample selection and other causes of endogeneity. Reviews 63 articles (SMJ 2005-2014) and notes inconsistencies in how scholars implement and report results, specifically in that the stated endogeneity concern does not align with the type of endogeneity (i.e., sample-induced endogeneity) addressed with Heckman models. Provide detailed descriptions of the sources of endogeneity in such cases and guidelines for when and how to use Heckman models.Chatterji, A. K., Findley, M., Jensen, N. M., Meier, S., & Nielson, D. 2016. Field experiments in strategy research. Strategic Management Journal, 37: 116-132.Propose wider use of experiments in strategy research to address concerns for causality. Discuss advantages and challenges associated with the research design. Propose two strategy-specific types of experiments: strategy field experiments which align with the general conception of a field experiment in that individuals are randomly assigned to groups which receive a particular treatment X and assess changes in an outcome Y and process field experiments in which a researcher manipulates a third variable M to assess the treatment-outcome relationship, XàY. Report results from two field experiments to answer substantive questions.Clougherty, J. A., Duso, T., & Muck, J. 2016. Correcting for self-selection based endogeneity in management research: Review, recommendations and simulations. Organizational Research Methods, 19: 286-347.Focus on self-selection endogeneity in both micro and macro domains. Explain selection bias as a special case of omitted-variable bias as the selection process can be represented by an excluded variable; failure to include that variable is captured in the error term which correlates with the endogenous choice construct. Present selection-based endogeneity as occurring through two forms: sample-selection and self-selection and give an in-depth review of the self-selection endogeneity problem. Review papers in SMJ, 2002-2014, and report that while the number of papers addressing selection effects has risen, the estimation strategies and reporting standards are so poor as to make findings uninformative. Use simulations to evaluate different analytical strategies and recommend that researchers take care in choosing analytical methods as the exact nature of the endogenous self-selection - endogenous treatment (which only involves an intercept effect) and endogenous switching (which involves slope coefficients for the other explanatory variables in addition to the intercept effect) affects the outcome of ultimate interest.Conley, T.G., Hansen, C.B. and Rossi, P.E. 2012. Plausibly exogenous. Review of Economics and Statistics, 94: 260-272.Focus on instrumental variable techniques and the exclusion restriction in particular (i.e., the IV correlates with the endogenous predictor and is only related to the outcome of interest through the endogenous predictor). Define plausibly, or approximatelyexogenous instruments as those where the relationship between the instrument and error term is near zero, instead of exactly zero. Motivated by the fact that 2SLS is sensitive to violation of the exclusion condition, especially when instruments are weak; so providing evidence that strong instruments can yield informative results even when deviating from the exact exclusion restriction is useful. Present four derivations and methods for inference and provide empirical examples to demonstrate.Dehejia, R. and Wahba, S. 2002. ‘Propensity score matching methods for non-experimental causal studies’. Review of Economics and Statistics, 84: 151–61.Foundational article on propensity score matching, see also Dehejia and Wahba (1999). Discusses the use of propensity score-matching methods to correct for sample selection endogeneity due to observable differences between the “treatment” and comparison groups in nonexperimental settings. Propose that this approach is preferable in situations when, 1) there are few units in the nonexperimental comparison group that are comparable to the treatment units, and, 2) there is a high number of pretreatment characteristics available on which to match treatment and comparison groups.Elwert, F., & Winship, C. 2014. Endogenous selection bias: The problem of conditioning on a collider variable. Annual Review of Sociology, 40: 31-53.Argue that selection bias can be difficult to identify. Use causal graphs to illustrate all forms of selection bias that can occur with non-random samples. Gives multiple examples from sociology.Hamilton, B. H., & Nickerson, J. A. 2003. Correcting for endogeneity in strategic management research. Strategic Organization, 1: 51-78.Show that despite the centrality of endogenous choices in our literature, few papers published in Strategic Management Journal (1990-2001) correct for endogeneity. Shows that many strategy research questions are based on estimating the effect of a discrete, endogenous choice on future performance. Explains how the Heckman treatment effect model addresses this kind of question and then shows that panel data may allow for estimation with fewer assumptions.Heckman, J. 1979. Sample selection bias as a specification error. Econometrica, 47: 153–61.Foundational article that characterizes nonrandomly selected sample as an omitted variable problem. Derives a two-stage estimator and shows that it is consistent. The first stage is a model of inclusion in the sample. The invers Mill’s ratio calculated from the first stage is used as a control in the second stage.Heckman, J. 1997. Instrumental variables: A study of implicit behavioral assumptions used in making program evaluations. Journal of Human Resources, 441-462.Argues that instrumental variable methods of estimating treatment effects are built on an assumption that all individuals would receive the same benefit from the treatment. If unobserved variables predict how much an individual will benefit, not just whether they will participate, then instrumental variable methods do not yield “economically interesting” parameters.Imbens, G.W. 2004. Nonparametric estimation of average treatment effects under exogeneity: A review. Review of Economics and Statistics, 86: 4-29.Reviews a number of methods used to estimate treatment effects including matching, propensity score and weighting methods. None of these methods address endogeneity (since exogeneity is assumed) and yet significant estimation issues remain.Imbens, G. W., & Wooldridge, J. M. 2009. Recent developments in the econometrics of program evaluation. Journal of Economic Literature, 47: 5-86.A comprehensive review of the econometric challenges of evaluating programs or policy changes. Many of methods discussed are applicable to organizational research questions. Krause, M. S., & Howard, K. I. (2003). What random assignment does and does not do. Journal of Clinical Psychology, 59(7), 751-766.Discusses the advantages and limitations of using study designs which employ random assignment, particularly as it relates to avoiding confounded estimates at the heart of omitted variables endogeneity.Li, M. 2013a Social network and social capital in leadership and management research: A review of causal methods. The Leadership Quarterly, 24: 638-665A survey of methods used in research on social networks and social capital. Demonstrates that common procedures for network sampling and analysis introduce multiple endogeneity concerns. A review of studies shows that scholars are not using rigorous methods to address for endogeneity.Li, M. 2013b. Using the propensity score method to estimate causal effects: A review and practical guide. Organizational Research Methods, 16: 188-226.Introduces and explains the propensity score matching method. Notes that this method cannot adjust for bias caused by unobserved drivers of selection because it is based on the strongly ignorable assumption. However, the method can be used to create treated and control groups that have similar observable covariate distributions.McNeish, D., & Kelley, K. 2019. Fixed effects models versus mixed effects models for clustered data: Reviewing the approaches, disentangling the differences, and making recommendations. Psychological Methods, 24: 20.Notes that psychology has traditionally ignored endogeneity concerns but argues that this must change as research is becoming more observational and less experimental. Discusses the differences between mixed effects and mixed effects models in addressing endogeneity and omitted confounders.Reeb, D., Sakakibara, M., & Mahmood, I. P. 2012. From the editors: Endogeneity in international business research.Provide an explanation of the endogeneity problem in international business research and give guidelines for achieving the goal of research that approximates a randomized-controlled experiment.Roberts, M. R., & Whited, T. M. 2013. Endogeneity in empirical corporate finance1. In Handbook of the Economics of Finance, Vol. 2, pp. 493-572. Elsevier.Extensive overview of the sources of endogeneity and methods that either rely on an exogenous instrument or event for identification (instrumental variable, difference-in-differences, regression discontinuity) or that rely on other assumptions (matching, panel methods, higher order moment estimators).Rocha, V., Van Praag, M., Folta, T. B., & Carneiro, A. 2019. Endogeneity in strategy-performance analysis: An application to initial human capital strategy and new venture performance. Organizational Research Methods, 22: 740-764.Argue that many organizational decisions cannot be adequately modeled as single choices. Instead, many managers must make simultaneous choices and interdependent choices. Research that focuses on just a single decision, even if it accounts for the endogeneity of the decision, may be biased if it ignores related decisions.Rosenbaum, P. R., & Rubin, D. B. 1983. The central role of the propensity score in observational studies for causal effects. Biometrika, 70: 41-55.The first articulation of the propensity score matching methods as a way to reduce biased estimates of a treatment effect. The method is built on the assumption of strongly ignorable treatment assignment.Semadeni, M., Withers, M. C., & Trevis Certo, S. 2014. The perils of endogeneity and instrumental variables in strategy research: Understanding through simulations. Strategic Management Journal, 35: 1070-1079.Reports the analysis of simulated data with an endogenous regressor. When endogeneity is ignored, OLS coefficient estimates are biased. The use of instrumental variables can eliminate bias. However, instrumental variable estimation reduces statistical power and the use of weak instruments can result in estimates that are inferior to OLS regression.Shaver, J. M. 1998. Accounting for endogeneity when assessing strategy performance: Does entry mode choice affect FDI survival. Management Science, 44: 571-585.One of the first papers to recognize the endogeneity of strategic choices as an important bias. Argues that the choice between acquisition and greenfield is endogenous. An observed survival advantage of greenfield entry disappears when the self-selection of the choice is accounted for.Shaver, J. M. 2005. Testing for mediating variables in management research: Concerns, implications, and alternative strategies. Journal of Management, 31: 330-353.Argues that tests of mediation in field data are often biased by endogeneity. Demonstrates an alternative estimation procedure based on instrumental variables using SEM.Shaver, J. M. 2019a. Causal Identification Through a Cumulative Body of Research in the Study of Strategy and Organizations. Journal of Management, 0149206319846272.Highlights why causal identification is important in organizational research but also why the nature of the available data make identification difficult. Suggests that, in addition to improving research practices, progress will require building evidence of causality through multiple empirical studies.Shaver, J. M. 2019b. Interpreting Interactions in Linear Fixed-Effect Regression Models: When Fixed-Effect Estimates Are No Longer Within-Effects. Strategy Science, 4: 25-40.Fixed effect models are one way to account for unobserved heterogeneity. However, this paper demonstrates that the desirable properties of fixed effects models are eliminated when interaction terms are introduced.Stock, J. & Yogo, M. 2002. Testing for weak instruments in linear IV regression. In Andrews, D. W. K. and Stock, J. (eds.), Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg. Cambridge: Cambridge University Press, 80–108.Weak instruments are variables that are not strong predictors of the endogenous variable. It is well established that weak instrumental variables produced biased coefficient estimates. This paper proposes a quantitative test for weak instruments in models with multiple endogenous regressors.Winship, C., & Morgan, S.L. 1999. The estimation of causal effects from observational data. Annual Review of Sociology, 25: 659–706Start with a review of methods that can be used to establish causal effects from cross-sectional data and then presents estimators that make use of longitudinal data. Focus on interrupted time series design as a potentially useful method for sociology research.Wolfolds, S. E., & Siegel, J. 2019. Misaccounting for endogeneity: The peril of relying on the Heckman two‐step method without a valid instrument. Strategic Management Journal, 40: 432-462.Focus on selection bias as a source of endogeneity and evaluation of the valid exclusion condition in the selection equation for Heckman models, which was reported in only 54 of 165 papers in top journals. Use simulations to show that when using a valid instrument or selection on observables, the Heckman method generally performs better than OLS, regardless of the distribution of the error terms. Yet when the exclusion condition is not met (i.e., one does not have a valid instrument that affects selection but not the outcome), the distribution of errors affects whether the Heckman or OLS method is more accurate. Except in the case when errors are distributed according to a bivariate normal or bivariate lognormal distribution, the Heckman method performs significantly worse than OLS.
Source: Hill AD, Johnson SG, Greco LM, O’Boyle EH, Walter SL. Endogeneity: A Review and Agenda for the Methodology-Practice Divide Affecting Micro and Macro Research. Journal of Management. 2021;47(1):105-143. doi:10.1177/0149206320960533
