On the Nuisance of Control Variables in Regression Analysis
aa r X i v : . [ ec on . E M ] M a y On the Nuisance of Control Variablesin Regression Analysis
Paul Hünermund † Beyers Louw ‡ This version: 1 June 2020First version: 20 May 2020
Abstract:
Control variables are included in regression analyses to estimate the causal effectof a treatment variable of interest on an outcome. In this note we argue thatcontrol variables are unlikely to have a causal interpretation themselves though.We therefore suggest to refrain from discussing their marginal effects in the resultssections of empirical research papers.
Key words : Multivariate Regression; Research Methodology; Causal Inference;Econometrics
JEL classification : C18; C51
Introduction
Multivariate regression analysis is an important tool for empirical research in strategicmanagement and economics. These methods account for confounding influence fac-tors between a treatment and an outcome by including a set of control variables inorder to obtain unbiased causal effect estimates. Notwithstanding their importancefor causal inference, in practice scholars often overstate the role of control variables inregressions. In this note we argue that, while essential for the identification of treat-ment effects, control variables generally have no structural interpretation themselves.This is because even valid controls are often correlated with other unobserved factors, † Maastricht University, School of Business and Economics. Tongersestraat 53, 6211 LM Maastricht,The Netherlands. Email: [email protected] ‡ Email: [email protected] et al. , 2020). Consequently, researchers need tobe careful with attaching too much meaning to control variables and should consider toignore them entirely when interpreting the results of their analysis.Drawing substantive conclusions from control variables is common however in appliedresearch. Authors frequently make use of formulations such as: “control variables haveexpected signs" or “it is worth noting the results of our control variables" . Based onthe volume of papers published in the last five years in Strategic Management Journal ,we found that 47 percent of papers that made use of parametric regression models alsoexplicitly discussed the estimated effects of controls. Moreover, in our own experienceas authors of empirical research papers, we encountered instances in which reviewersspecifically asked us to provide an economic interpretation of control variable coefficients.The argument was that, although they were not the main focus of the analysis, thecontrols could still provide valuable information to other researchers in the field whoare investigating related research questions. In the following, we will explain why thisapproach is potentially misleading and should therefore better be avoided.
The structural interpretation of control variables
The relationships between the main explanatory variables and the controls in a regressioncan be complex, therefore it is useful to explicitly depict them in a causal diagram(Pearl, 2000; Hünermund and Bareinboim, 2019). Durand and Vaara (2009) were thefirst to introduce causal graphs in the strategic management literature, by arguing theirusefulness as a tool for empirical research. Figure 1a presents a simple economic modelwith a treatment variable X and an outcome variable Y . Both variables are connected We analyzed all research articles published in Strategic Management Journal between January 2015and May 2020 and found that, out of a total number of 458 papers which included parametricregression models, 213 proceeded to explicitly interpret and draw substantive conclusions based onthe marginal effects of control variables. YZ Z (a) X YZ Z Z Z Z (b) Figure 1 by an arrow, denoting the direction of causal influence factor between them. In addition,there are two confounding factors, Z and Z , affecting the treatment and the outcome. Z and Z are correlated, as a result of a common influence they share, which is denotedby the dashed bidirected arc in the graph. The fact that Z and Z are correlated createswhat is known as a backdoor path between the treatment and the outcome (Pearl, 2000). X and Y are not only connected by the genuinely causal path X → Y , but also by asecond path, X ← Z L9999K Z → Y , which creates a spurious, non-causal correlationbetween them.The role of control variables in regression analysis is exactly to block such backdoorpaths, in order to get at the uncontaminated effect of X on Y . For this purpose, it issufficient to control for any variable that lies on the open path. Thus, in the exampleof Figure 1a, the researcher has the choice between either controlling for Z or Z , sinceboth would allow to identify the causal effect of interest. The choice between differentadmissible sets of control variables is thereby of high practical relevance. Researchersoften have a fairly detailed knowledge about the treatment assignment mechanism Z → X , for example, because there are specific organizational or administrative rules thatcan be exploited for identification purposes (Angrist, 1990; Flammer and Bansal, 2017; Technical note: Requiring the path to be previously unblocked rules out that the variable which isadjusted for is a collider (Hünermund and Bareinboim, 2019). A discussion of collider variables incausal graphs goes beyond the scope of this note. Z that aredirect influence factors of Y will likely be quite large. Thus, in practical applicationsit might be much easier to control the treatment assignment mechanism than trying toinclude all the variables that have an effect on the outcome measure in a regression.Nevertheless, although controlling for Z is sufficient to obtain an unbiased estimatefor X , its marginal effect will itself not correspond to any causal effect of Z on Y .This is because Z is correlated with Z and will thus partially pick up an effect of Z on Y too (Cinelli and Hazlett, 2020). The danger of interpreting estimated effects ofpotentially endogenous control variables, such as Z , is referred to as table 2 fallacy inepidemiology (Westreich and Greenland, 2013). A similar point has been made recentlyby Keele et al. (2020) in the field of political science.Figure 1b depicts a more complex example, with several admissible sets of controls,each sufficient to identify the causal effect of X on Y (Textor and Liśkiewicz, 2011).One possibility in this situation would be to simply control for Z , which is the onlydirect influence factor of X , and thus blocks all paths entering X through the backdoor.Similarly, controlling for the direct influence factors of Y ( Z , Z , and Z ) would alsoblock all backdoor paths. A third alternative is to control for the entire set of confounders( Z , Z , Z , Z , and Z ), although this would be the most data-intensive identification To illustrate this phenomenon quantitatively, we parametrize the causal graph in Figure 1a in thefollowing way: z ← u + ε ,z ← u + ε ,x ← z + ε ,y ← x + z + ε , with N = 1000 , and u and ε m being standard normal. We then run a regression of Y on X and Z , which gives us a consistent coefficient estimate for X (= 0.970, std. err. = 0.052; bootstrappedwith 1000 replications), while the effect of Z (= 0.541, std. err. = 0.064) turns out to be biased.By contrast, if we also include Z in the regression, the coefficient of Z drops to zero (= -0.016,std. err. = 0.042), which corresponds to its true causal effect on Y in this example. Epidemiologists usually present the results of multivariate regression analyses right after a table withdescriptive statistics of the data, therefore the name table 2 fallacy . Z ) for identifying the causaleffect of X is often much smaller than the total number of confounding variables in amodel. At the same time, the estimated marginal effects for the control variables onlyhave a structural interpretation themselves if all the direct influence factors of Y (here: Z , Z , and Z ) are accounted for in the regression. As we argued above, this is unlikelyto be the case though, since in many real-world settings the number of causal factorsdetermining Y can be prohibitively large. Implications for research practice
Attaching economic meaning to the marginal effects of biased control variables is prob-lematic, as researchers could develop false intuitions or come to erroneous policy con-clusions based on them. Therefore it is advisable to not discuss the results obtained forcontrol variables in empirical papers, unless the researchers can be sure that they haveaccounted for all relevant influence factors of the outcome in a regression ( all-causesregression ). Since in many practical settings this is unlikely, however, we recommend totreat controls as nuisance parameters , which are included in the analysis for identificationpurposes but are not reported themselves in the output tables (Liang and Zeger, 1995).Our suggestion thereby corresponds to the way control variables are treated by non-parametric matching estimators (Heckman et al. , 1998) and modern machine learningtechniques for high-dimensional settings (Belloni et al. , 2014). These methods similarlydo not report estimation results related to controls, either because there are simply toomany covariates in the analysis, which is the primary use-case for machine learning, ormarginal effects of control variables are not even returned by the estimation protocol,as in the matching case. 5n short, there is no reason to be worried if the estimated coefficients of control vari-ables do not have expected signs, since they are likely to be biased anyways in practicalapplications. Instead, researchers should rather focus on interpreting the marginal effectsof the main variables of interest in their manuscripts. The estimation results obtainedfor controls, by contrast, have little substantive meaning and can therefore safely beomitted—or relegated to an appendix. This approach will not only prevent researchersfrom drawing wrong causal conclusions based on endogenous controls, but will also allowto streamline the discussion sections of empirical research papers and save on valuablejournal space. 6 eferences
Angrist JD. 1990. Lifetime Earnings and the Vietnam Era Draft Lottery: Evidence fromSocial Security Administrative Records.
The American Economic Review , (3): 313–336.Belloni A, Chernozhukov V, Hansen C. 2014. High-dimensional methods and inferenceon structural and treatment effects. Journal of Economic Perspectives , (2): 29–50.Cinelli C, Hazlett C. 2020. Making sense of sensitivity: Extending omitted variable bias. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , (1):39–67.Durand R, Vaara E. 2009. Causation, counterfactuals, and competitive advantage. Strategic Management Journal , (12): 1245–1264.Flammer C, Bansal P. 2017. Does a long-term orientation create value? Evidence froma regression discontinuity. Strategic Management Journal , (9): 1827–1847.Heckman JJ, Ichimura H, Todd P. 1998. Matching as an econometric evaluation estima-tor. The Review of Economic Studies , (2): 261–294.Hünermund P, Bareinboim E. 2019. Causal inference and data-fusion in econometrics, https://arxiv.org/abs/1912.09104 .Hünermund P, Czarnitzki D. 2019. Estimating the causal effect of R&D subsidies in apan-European program. Research Policy , (1): 115–124.Keele L, Stevenson RT, Elwert F. 2020. The causal interpretation of estimated associa-tions in regression models. Political Science Research and Methods , (1): 1–13.Liang KY, Zeger SL. 1995. Inference based on estimating functions in the presence ofnuisance parameters. Statistical Science , (2): 158–173.Pearl J. 2000. Causality: Models, Reasoning, and Inference. Cambridge University Press,New York, United States, NY, 1st ed.Textor J, Liśkiewicz M. 2011. Adjustment criteria in causal diagrams: An algorithmicperspective. In Proceedings of the 27th Conference on Uncertainty in Artificial Intel-ligence, AUAI press, 681–688.Westreich D, Greenland S. 2013. The table 2 fallacy: presenting and interpreting con-founder and modifier coefficients. American Journal of Epidemiology , (4): 292–298. 7 ppendix The following table lists the number of articles per volume and issue, published in the
Strategic Management Journal between January 2015 and May 2020, that were identi-fied to include an explicit economic interpretation of the estimation results for controlvariables.
Total refers to the total number of papers using parametric regression mod-els in the respective issue. We counted articles that either explicitly discuss marginaleffects of control variables (e.g., their sign and significance) with regards to to prior re-search findings or draw substantive conclusions based on them for policy and managerialpractice.
Volume Issue Total Count Articles
41 5 5 2 Sakakibara & Balasubramanian, 2019;Rocha & van Praag, 2020;41 4 7 3 Aggarwal, 2019; Fu, Tang & Chen, 2019 (Ap-pendix); Bonet, Capelli & Homari, 2020;41 3 4 2 Arikan, Arikan & Shenkar, 2019; Agarwal,Braguinsky & Ohyama, 2019;41 2 5 2 Ryu, Reuer & Brush, 2019; Jia, Gao & Ju-lian, 2019;41 1 5 0 -40 13 7 1 Hsu, Kovács & Koçak, 2019;40 12 6 4 Kim, 2019; Petrenko, Aime, Recendes &Chandler, 2019; Guldiken, Mallon, Fainsh-midt, Judge & Clark, 2019; Shi, Conelly,Mackey & Gupta, 2019;40 11 5 2 Woo, Canella & Mesquita, 2019; Zweiger,Stettler, Baldauf & Zamudio, 2019;40 10 6 3 Ridge, Imgram, Abdurakhmonov & Hasija,2019; Gómez–Solórzano, Tortoriello & Soda,2019; Kavusan & Frankort, 2019;40 9 5 0 -40 8 6 1 Barlow, Verhaal & Angus, 2019;40 7 5 2 Corsino, Mariani & Torrisi, 2019; Andrus,Withers, Courtright & Boivie, 2019;40 6 5 2 Hiatt & Carlos, 2018; Piazzai & Wijnberg,2019;40 5 4 2 Hill, Recendes & Ridge, 2018; Yu, Minniti &Nason, 2018;40 4 5 3 Paik, Kang & Seamans, 2018; Bruce, deFigueiredo & Silverman, 2018; Zheng, Ni &Crilly, 2018;8 olume Issue Total Count Articlesolume Issue Total Count Articles
41 5 5 2 Sakakibara & Balasubramanian, 2019;Rocha & van Praag, 2020;41 4 7 3 Aggarwal, 2019; Fu, Tang & Chen, 2019 (Ap-pendix); Bonet, Capelli & Homari, 2020;41 3 4 2 Arikan, Arikan & Shenkar, 2019; Agarwal,Braguinsky & Ohyama, 2019;41 2 5 2 Ryu, Reuer & Brush, 2019; Jia, Gao & Ju-lian, 2019;41 1 5 0 -40 13 7 1 Hsu, Kovács & Koçak, 2019;40 12 6 4 Kim, 2019; Petrenko, Aime, Recendes &Chandler, 2019; Guldiken, Mallon, Fainsh-midt, Judge & Clark, 2019; Shi, Conelly,Mackey & Gupta, 2019;40 11 5 2 Woo, Canella & Mesquita, 2019; Zweiger,Stettler, Baldauf & Zamudio, 2019;40 10 6 3 Ridge, Imgram, Abdurakhmonov & Hasija,2019; Gómez–Solórzano, Tortoriello & Soda,2019; Kavusan & Frankort, 2019;40 9 5 0 -40 8 6 1 Barlow, Verhaal & Angus, 2019;40 7 5 2 Corsino, Mariani & Torrisi, 2019; Andrus,Withers, Courtright & Boivie, 2019;40 6 5 2 Hiatt & Carlos, 2018; Piazzai & Wijnberg,2019;40 5 4 2 Hill, Recendes & Ridge, 2018; Yu, Minniti &Nason, 2018;40 4 5 3 Paik, Kang & Seamans, 2018; Bruce, deFigueiredo & Silverman, 2018; Zheng, Ni &Crilly, 2018;8 olume Issue Total Count Articlesolume Issue Total Count Articles
40 3 3 2 Chatterji, Delecourt, Hasan & Koning, 2018;Bigelow, Nickerson & Park, 2018;40 2 5 3 Criscuolo, Alexy, Sharapov & Salter, 2018;Ren, Hu & Cui, 2018; Boone, Lokshin,Guenter & Belderbos, 2018;40 1 7 4 Haans, 2018 (Appendix); Chatterji, Cun-ningham & Joseph, 2018; Westphal & Zhu,2018; Belderbos, Tong & Wu, 2018;39 13 5 1 Garg & Zhao, 2018;39 12 5 4 Cui, Yang & Vertinsky, 2017 (Appendix);Ranganathan, Ghosh & Rosenkopf, 2018;Arslan, 2018; Asgari, Tandon, Singh &Mitchell, 2018;39 11 8 5 Feldman, Gartenberg & Wulf, 2018;Claussen, Essling & Peukert, 2018; Bur-bano, Mamer & Snyder, 2018; Koch-Bayram& Wernicke, 2018; Mata & Alves, 2018;39 10 8 4 Eberhardt & Eesley, 2018; Hornstein &Zhao, 2018; Kang & Zaheer, 2018; Albino-Pimentel, Dussauge & Shaver, 2018;39 9 6 3 Khanna, Guler & Nerkar, 2018; Hawk& Pacheco-de-Almeida, 2018; Schepker &Barker, 2018;39 8 5 4 Yayavaram, Srivastava & Sarkar, 2018; Gan-dal, Markovich & Riordan, 2018; Manning,Massini, Peeters & Lewin, 2018; Shi & Con-nelly, 2018;39 7 8 4 Byun, Frake & Agarwal, 2018; Mawdsley& Somaya, 2018; Alvarez-Garrido & Guler,2018; Gupta, Mortal & Guo, 2018;39 6 0 0 -39 5 9 5 Chen & Garg, 2017; Kaul, Nary & Singh,2017; Flammer, 2018; Ramírez & Tarziján,2018; Wiersema, Hishimure & Suzuki, 2018;39 4 8 5 Hawn, Chatterji & Mitchell, 2017; Choud-hury & Haas, 2017; Bode & Singh, 2017;Tarakci, Ateş, Floyd, Ahn & Wooldridge,2017; Rhee & Leonardi, 2017;39 3 0 0 -39 2 6 4 Chen, Kale & Hoskisson, 2017; Choi & Mc-Namara, 2017; Deichmann & Jensen, 2017;Pek, Oh & Rivera, 2017;9 olume Issue Total Count Articlesolume Issue Total Count Articles
40 3 3 2 Chatterji, Delecourt, Hasan & Koning, 2018;Bigelow, Nickerson & Park, 2018;40 2 5 3 Criscuolo, Alexy, Sharapov & Salter, 2018;Ren, Hu & Cui, 2018; Boone, Lokshin,Guenter & Belderbos, 2018;40 1 7 4 Haans, 2018 (Appendix); Chatterji, Cun-ningham & Joseph, 2018; Westphal & Zhu,2018; Belderbos, Tong & Wu, 2018;39 13 5 1 Garg & Zhao, 2018;39 12 5 4 Cui, Yang & Vertinsky, 2017 (Appendix);Ranganathan, Ghosh & Rosenkopf, 2018;Arslan, 2018; Asgari, Tandon, Singh &Mitchell, 2018;39 11 8 5 Feldman, Gartenberg & Wulf, 2018;Claussen, Essling & Peukert, 2018; Bur-bano, Mamer & Snyder, 2018; Koch-Bayram& Wernicke, 2018; Mata & Alves, 2018;39 10 8 4 Eberhardt & Eesley, 2018; Hornstein &Zhao, 2018; Kang & Zaheer, 2018; Albino-Pimentel, Dussauge & Shaver, 2018;39 9 6 3 Khanna, Guler & Nerkar, 2018; Hawk& Pacheco-de-Almeida, 2018; Schepker &Barker, 2018;39 8 5 4 Yayavaram, Srivastava & Sarkar, 2018; Gan-dal, Markovich & Riordan, 2018; Manning,Massini, Peeters & Lewin, 2018; Shi & Con-nelly, 2018;39 7 8 4 Byun, Frake & Agarwal, 2018; Mawdsley& Somaya, 2018; Alvarez-Garrido & Guler,2018; Gupta, Mortal & Guo, 2018;39 6 0 0 -39 5 9 5 Chen & Garg, 2017; Kaul, Nary & Singh,2017; Flammer, 2018; Ramírez & Tarziján,2018; Wiersema, Hishimure & Suzuki, 2018;39 4 8 5 Hawn, Chatterji & Mitchell, 2017; Choud-hury & Haas, 2017; Bode & Singh, 2017;Tarakci, Ateş, Floyd, Ahn & Wooldridge,2017; Rhee & Leonardi, 2017;39 3 0 0 -39 2 6 4 Chen, Kale & Hoskisson, 2017; Choi & Mc-Namara, 2017; Deichmann & Jensen, 2017;Pek, Oh & Rivera, 2017;9 olume Issue Total Count Articlesolume Issue Total Count Articles
39 1 8 3 Furr & Kapoor, 2017; Vidal & Mitchell,2017; Jiang, Xia, Canella & Xiao, 2017;38 13 8 4 Chem, Qian & Narayanan, 2017; Rabier,2017; Dorobantu & Odziemkowska, 2017; Li,Yi & Cui, 2017;38 12 5 3 Lee & Puranam, 2017; Werner, 2017; Theeke& Lee, 2017;38 11 8 4 Carnahan, 2017; Kölbel, Busch & Jancso,2017; Bos, Faems & Noseleit, 2017; Li &Zhou, 2017;38 10 8 4 Moeen, 2017; Raffiee, 2017; Jiang, Canella,Xia & Semadeni, 2017; Wei, Ouyang & Chen,2017;38 9 8 5 Souder, Zaheer, Sapienza & Ranucci, 2016;Caner, Cohen & Pil, 2016; Shan, Fu &Zheng, 2016; Wang, Zhao & Chen, 2016; Li,Xia & Lin, 2016;38 8 9 5 Zhou & Wan, 2016; Kulchina, 2016; Kim& Steensma, 2016; Steinbach, Holcomb,Holmes, Devers & Canella, 2016; Makino &Chan, 2016;38 7 10 3 Armanios, Eesley, Li & Eisenhardt, 2016; Ref& Shapira, 2016; McCann & Bahl, 2016;38 6 7 3 Roy & Cohen, 2016; Dowell & Muthulingam,2016; Vanacker, Collewaert & Zahra, 2016;38 5 9 5 Stan & Puranam, 2016; Asgari, Singh &Mitchell, 2016; Kuusela, Keil & Maula, 2016;Girod & Whittington, 2016; Connelly, Ti-hanyi, Ketchen, Carnes & Ferrier, 2016;38 4 8 1 Silverman & Ingram, 2016;38 3 10 4 Bermiss, Hallen, McDonald & Pahnke, 2016;Chatterjee, 2016; Oh & Oetzel, 2016; Blake& Moschieri, 2016;38 2 11 5 Flammer & Luo, 2015; Madsen & Walker,2015; Mackey, Barney & Dotson, 2015; Fonti,Maoret & Whitbred, 2015; Deb, David &O’Brien, 2015;38 1 0 0 -37 13 6 2 Hawn & Ioannou, 2015; Stuart & Wang,2015;37 12 7 3 Wang, Zhao & He, 2015; Easley, Decelles &Lenox, 2015; Wu & Salomon, 2015;10 olume Issue Total Count Articlesolume Issue Total Count Articles
39 1 8 3 Furr & Kapoor, 2017; Vidal & Mitchell,2017; Jiang, Xia, Canella & Xiao, 2017;38 13 8 4 Chem, Qian & Narayanan, 2017; Rabier,2017; Dorobantu & Odziemkowska, 2017; Li,Yi & Cui, 2017;38 12 5 3 Lee & Puranam, 2017; Werner, 2017; Theeke& Lee, 2017;38 11 8 4 Carnahan, 2017; Kölbel, Busch & Jancso,2017; Bos, Faems & Noseleit, 2017; Li &Zhou, 2017;38 10 8 4 Moeen, 2017; Raffiee, 2017; Jiang, Canella,Xia & Semadeni, 2017; Wei, Ouyang & Chen,2017;38 9 8 5 Souder, Zaheer, Sapienza & Ranucci, 2016;Caner, Cohen & Pil, 2016; Shan, Fu &Zheng, 2016; Wang, Zhao & Chen, 2016; Li,Xia & Lin, 2016;38 8 9 5 Zhou & Wan, 2016; Kulchina, 2016; Kim& Steensma, 2016; Steinbach, Holcomb,Holmes, Devers & Canella, 2016; Makino &Chan, 2016;38 7 10 3 Armanios, Eesley, Li & Eisenhardt, 2016; Ref& Shapira, 2016; McCann & Bahl, 2016;38 6 7 3 Roy & Cohen, 2016; Dowell & Muthulingam,2016; Vanacker, Collewaert & Zahra, 2016;38 5 9 5 Stan & Puranam, 2016; Asgari, Singh &Mitchell, 2016; Kuusela, Keil & Maula, 2016;Girod & Whittington, 2016; Connelly, Ti-hanyi, Ketchen, Carnes & Ferrier, 2016;38 4 8 1 Silverman & Ingram, 2016;38 3 10 4 Bermiss, Hallen, McDonald & Pahnke, 2016;Chatterjee, 2016; Oh & Oetzel, 2016; Blake& Moschieri, 2016;38 2 11 5 Flammer & Luo, 2015; Madsen & Walker,2015; Mackey, Barney & Dotson, 2015; Fonti,Maoret & Whitbred, 2015; Deb, David &O’Brien, 2015;38 1 0 0 -37 13 6 2 Hawn & Ioannou, 2015; Stuart & Wang,2015;37 12 7 3 Wang, Zhao & He, 2015; Easley, Decelles &Lenox, 2015; Wu & Salomon, 2015;10 olume Issue Total Count Articlesolume Issue Total Count Articles
37 11 10 7 Ghosh, Ranganathan & Rosenkopf, 2016;Kalnins, 2016; Chang, Kogut & Yang, 2016;Tsang & Yamanoi, 2016; Massimo, Colombo& Shafi, 2016; Chadwick, Guthrie & Xing,2016; Park, Borah & Kotha, 2016;37 10 8 2 Husted, Jamali & Saffar, 2016; Van Reenen& Pennings, 2016;37 9 6 1 Gomulya & Boeker, 2015;37 8 10 5 Fonti & Maoret, 2015; Rodríguez & Ni-etro, 2015; Zhu & Yoshikawa, 2015; Yu,Umashankar & Rao, 2015; Jain, 2015;37 7 12 3 Bennet & Pierce, 2015; Anand, Mulotte &Ren, 2015; Geng, Yoshikawa & Colpan, 2015;37 6 8 3 Smith & Chae, 2015; Klingebiel & Joseph,2015; Karna, Richter & Riesenkampf, 2015;37 5 5 4 Roy & Sarkar, 2014; Lungeanu, Stern & Za-jac, 2015; Tyler & Caner, 2015; Brandes,Dharwadkar & Suh, 2014;37 4 6 4 Adner & Kapoor, 2015; Maslach, 2014;Poppo, Zhou & Li, 2015; Eckhardt, 2015;37 3 9 3 Feldman, Amit & Villalonga, 2014; Pe’er,Vertinsky & Keil, 2014; Barroso, Giarratana,Reis & Sorenson,2014;37 2 8 3 Chen, Crossland & Huang, 2014; Desender,Aguilera, Lópezpuertas-Lamy & Crespi,2014; Kang, 201437 1 6 2 Dezsö, Ross & Uribe, 2015; Ge, Huang &Png, 2015;36 13 8 3 Joseph & Gaba, 2014; Macher & Mayo, 2014;Zhu & Chen, 2014;36 12 7 3 Fuentelsaz, Garrido & Maicas, 2014; Malho-tra, Zhu & Reus, 2014; Chen, 2014;36 11 8 5 Zheng, Singh & Mitchell, 2014; Speckbacher,Neumann & Hoffmann, 2014; Skilton &Bernardes, 2014; Bermiss & Murmann, 2014;Fosfuri, Giarratana & Roca, 2014;36 10 6 3 Kaplan & Vakili, 2014; Chen, Crossland &Luo, 2014; Ang, Benischke & Doh, 2014;36 9 7 4 Chittoor, Kale & Puranam, 2014; Chang &Shim, 2014; Banalieva, Eddleston & Zell-weger, 2014; Hashai, 2014;11 olume Issue Total Count Articlesolume Issue Total Count Articles