Prediction in HRM research–A gap between rhetoric and reality

• Published date: 01 April 2022
• Author: Marko Sarstedt,Nicholas P. Danks
• Date: 01 April 2022
• DOI: http://doi.org/10.1111/1748-8583.12400

Received: 13 November 2019

-

Revised: 14 May 2021

-

Accepted: 22 June 2021

DOI: 10.1111/1748-8583.12400

REVIEW/PROVOCATION

Prediction in HRM research–A gap between

rhetoric and reality

Marko Sarstedt

1,2

|Nicholas P. Danks

3

1

Otto‐von‐Guericke‐University, Magdeburg,

Germany

2

Babeș‐Bolyai University, Cluj‐Napoca,

Romania

3

Trinity Business School, Trinity College,

Dublin, Ireland

Correspondence

Marko Sarstedt, Otto‐von‐Guericke

University Magdeburg, Faculty of Economics

and Management, Universitaetsplatz 2,

39106 Magdeburg, Germany.

Email: marko.sarstedt@ovgu.de

Funding information

WOA Institution: OTTO VON GUERICKE

UNIVERSITAET MAGDEBURG Blended

DEAL: Projekt DEAL.

Abstract

There are broadly two dimensions on which researchers

can evaluate their statistical models: explanatory power

and predictive power. Using data on job satisfaction in

ageing workforces, we empirically highlight the importance

of distinguishing between these two dimensions clearly by

showing that a model with a certain degree of explanatory

power can produce vastly different levels of predictive

power and vice versa—in the same and different contexts.

In a further step, we review all the papers published in

three top‐tier human resource management journals be-

tween 2014 and 2018 to show that researchers generally

confuse explanation and prediction. Specifically, while

almost all authors rely solely on explanatory power as-

sessments (i.e., assessing whether the coefficients are sig-

nificant and in the hypothesised direction), they also derive

practical recommendations, which inherently result from a

predictive scenario. Based on our results, we provide HRM

researchers recommendations on how to improve the

rigour of their explanatory studies.

Abbreviations: AIC, Akaike Information Criterion; ARIMA, Autoregressive integrated moving average; AUC, Area under the curve; CFI, Comparative fit

index; CV, Cross‐validation; EP, Explanatory and predictive; ETS, Error trend seasonal; HRM, Human resource management; ISSP, International Social

Survey Program; KNN, K‐nearest neighbours; LOOCV, Leave‐one‐out cross‐validation; MAE, Mean absolute error; RMSEA, Root mean square error of

approximation; RMSE, Root mean square error; SMAPE, Symmetric mean absolute percentage error; SRMR, Standard root mean square residual; USA,

United States of America

This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which per-

mits use and distribution in any medium, provided the original work is properly cited, the use is non‐commercial and no modifica-

tions or adaptations are made.

© 2021 The Authors. Human Resource Management Journal published by John Wiley & Sons Ltd.

Hum Resour Manag J. 2022;32:485–513. wileyonlinelibrary.com/journal/hrmj

-

485

KEYWORDS

explanation, explanatory power, generalisability, prediction,

predictive power, relevance

1

|

INTRODUCTION

Socio‐technical systems within the purview of human resource management (HRM) research are inherently

complex due to their intricate underlying causal interactions and processes. Models are formal quantitative

representations of theories and hypotheses that are devised to offer partial explanations of these complex systems

(Lauenroth, 2003). HRM researchers build models to approximate these processes and apply inferential statistical

methods to data in order to test the relationships between the assumed causes and effects. Specifically, the

researchers' focus is on assessing whether the model fits the data and whether the model coefficients are signif-

icant and in the hypothesised direction. This approach, which is widely known as explanatory modelling, differs

from predictive modelling. Predictive modelling applies estimated model parameters, generated from a sample at

hand, to generate predictions for other observations not used in the model estimation. These observations can be

kept separate from the main sample or they can be collected at a future point in time or even in another context

(Shmueli et al., 2016).

Practitioner notes

What is currently known?

Explanation and prediction are two distinct statistical concepts.

Establishing a model's predictive power is central to deriving practical recommendations.

Research has brought forward numerous procedures for evaluating a model's explanatory and pre-

dictive power.

What this paper adds?

An empirical example demonstrates the interplay between explanatory and predictive power.

A review of prior human resource management (HRM) research shows that few researchers consider

the predictive power of their models.

HRM researchers generally confuse the concepts of explanation and prediction.

The paper offers concrete recommendations on how to routinely implement predictive model evalu-

ations in statistical analyses.

The implications for practitioners

Results of research papers should be interpreted in the context of the analyses (explanation vs.

prediction).

Managerial recommendations derived from purely explanatory analyses should be challenged in terms

of their predictive relevance.

Predictive analyses should supplement explanatory modelling in making research practical and relevant

to practice.

486

-

SARSTEDT AND DANKS