Statistical control in correlational studies: 10 essential recommendations for organizational researchers

Abstract

Statistical control is widely used in correlational studies with the intent of providing more accurate estimates of relationships among variables, more conservative tests of hypotheses, or ruling out alternative explanations for empirical findings. However, the use of control variables can produce uninterpretable parameter estimates, erroneous inferences, irreplicable results, and other barriers to scientific progress. As a result, methodologists have provided a great deal of advice regarding the use of statistical control, to the point that researchers might have difficulties sifting through and prioritizing the available suggestions. We integrate and condense this literature into a set of 10 essential recommendations that are generally applicable and which, if followed, would substantially enhance the quality of published organizational research. We provide explanations, qualifications, and examples following each recommendation.

How to achieve sustainability?—Employee's point of view on company's culture and CSR practice

“…In both cases, the general results were similar, but Model B fit the data better. The results for both models are presented and discussed (Becker et al, 2016) to explain how company size and employee's position influence perception of CSR practice. The results for both models are presented and discussed (Becker et al, 2016) to explain how company size and employee's position influence perception of CSR practice.…”

Variations in coaching style and their impact on subordinates' work outcomes

“…As recommended by Becker () and Becker et al (), additional analyses were conducted without our control variables to ascertain whether the theory‐based rationale for their use was justified. Although the Level 1 controls did not affect any of our findings, at Level 2, the age and hierarchical position of the supervisor/manager made a difference in two instances.…”

“…As explained earlier, there were theoretical reasons we included a variety of control variables in this study. Nonetheless, for the purposes of comparison, in line with Becker () and Becker et al (), we conducted a series of additional analyses without the controls. First, a model without the Level 1 control variables (i.e., subordinates' age, gender, company tenure, and department tenure) was examined.…”

Race and Gender Differences in Undergraduate Research Mentoring Structures and Research Outcomes

“…The “purification principle” (Spector and Brannick, 2011, p. 288) that underlies the rationale often used to justify the inclusion of control variables has been seriously questioned on both theoretical and empirical grounds (Spector and Brannick, 2011; Bernerth and Aguinis, 2016), and current best practice is “when in doubt, leave them out” (Becker et al , 2016, p. 158).…”