Advocating the use of threshold effects estimation: An illustration using the gender wage gap

Abstract

While linear models remain the backbone of human resource development (HRD) as well as broader social science research, they are limited by their inability to detect curvilinearity or changes in direction and, therefore, may not accurately represent the nature of relationships between variables. This paper advocates the utility of the Andrews' (1993) threshold effect technique in the HRD, work, organization, and employment study literatures and does so using the gender wage gap as an illustration. The analysis compares traditional linear regression, a curvilinear quadratic model, and the threshold effect technique using publicly available wage data. This article demonstrates how the threshold method permits the data to organically determine the location of a critical break point, should one exist, and how it addresses the problem of overlooked relationships due to traditional linear assumptions. An overview of the technique and related coding is provided for researchers interested in broadening their analytical arsenal to enhance the study of the gender wage gap as well as other human‐resource‐related phenomena.

Examining the boundary conditions of tokenism: within-occupation gender wage gaps and female representation in the Canadian labor market

“…The technique has some advantages over traditional linear methods, and even quadratic or curvilinear estimations, as it permits the data to organically determine whether and where a structural break exists (Freeman, 2005; Yerger and Stephenson, 2020). Additionally, should a quadratic model be more appropriate to the data, this method approximates the association by discerning a threshold point where the linear trend changes, suggesting that it will approximate the point of the curve.…”