Data Envelopment Analysis Journal > Vol 5 > Issue 1

Three-Stage Approach to Analyze Managerial Ability

Rajiv Banker, Fox School of Business, Temple University, USA, , Han-Up Park, Edwards School of Business, University of Saskatchewan, Canada,
Suggested Citation
Rajiv Banker and Han-Up Park (2021), "Three-Stage Approach to Analyze Managerial Ability", Data Envelopment Analysis Journal: Vol. 5: No. 1, pp 27-54.

Publication Date: 22 Jun 2021
© 2021 R. Banker and H.-U. Park
Performance measurement,  Estimation frameworks
DEAefficiencymanagerial ability


Download article
In this article:
1 Introduction 
2 Measurement of Managerial Ability in Academic Literature 
3 Overview of Data Envelopment Analysis 
4 Three-Stage Approach of Analyzing Managerial Ability 
5 Other Alternatives 
6 Conclusion 


Demerjian et al. (2012) provide theoretically and empirically rigorous measurement of managerial ability based on data envelopment analysis. We discuss that the method can provide a consistent estimator and suggest best practices for empirical researchers. The three-stage approach of conducting inference with managerial ability begins with the first-stage estimation of firm-efficiency with inputs and outputs. The second stage removes the impacts of contextual variables on the firm-efficiency to construct managerial ability. The third stage uses the measure as a dependent or an independent variable. We discuss why data envelopment analysis that incorporates production theory and allows multiple inputs and outputs is more appropriate than other methods to measure managerial ability. We then discuss specific choices that researchers need to make in three stages: returns to scale, the number of inputs and outputs, industry-specific inputs and outputs, outlier detection, choice of estimation sample, adjustments to yield a valid measure of managerial ability, choice of contextual variables, the functional form of the second-stage regression, advantage of residual approach, and consideration for the inference with managerial ability as a dependent and an independent variable. Our suggestions allow researchers to apply the rigorous approach of Demerjian et al. (2012) in many contexts yet to be explored.