The use of artificial neural networks and multiple linear regression in modelling work–health relationships: Translating theory into analytical practice

Abstract

Although psychological theory acknowledges the existence of complex systems and the importance of nonlinear effects, linear statistical models have been traditionally used to examine relationships between environmental stimuli and outcomes. The way that we analyze these relationships does not seem to reflect the way that we conceptualize them. The present study investigated the application of connectionism (artificial neural networks) to modeling the relationships between work characteristics and employee health by comparing it with a more conventional statistical linear approach (multiple linear regression) on a sample of 1003 individuals in employment. Comparisons of performance metrics indicated differences in model fit, with neural networks to some extent outperforming the linear regression models, such thati? for worn-out and job satisfaction were significantly higher in the neural networks.Most importantly, comparisons revealed that the predictors in the two approaches differed in their relative importance for predicting outcomes. The improvement is attributed to the ability of the neural networks to model complex nonlinear relationships. Being unconstrained by assumptions of linearity, they can provide a better approximation of such psychosocial phenomena. Nonlinear approaches are often better fitted for purpose, as they conform to the need for correspondence between theory, method and data. Modeling the impact of work and organizational characteristics on employee health is important in terms of decision-making for job design, re-design and organizational development. Effective management of work-related health relies on accurately prioritizing potential risks and explaining the highest variability in outcomes from a combination of work and organizational characteristics (Clarke & Cooper, 2004;Glendon, Clarke, & McKenna, 2006). In turn, this traditionally relies on correlational and linear approaches.Although such an approach has historically proven useful, it produces two sources of uncertainty for the assessment of risk to work-related health. First, it implies a linear relationship between work and organizational characteristics and health outcomes, which often conflicts with available empirical evidence (Karanika, 2006). Second, organizations themselves provide a demonstration of complexity (Cox et al, 2007;Schneider & Somers, 2006). It also renders examination of the impact of combinations of risks (as opposed to bivariate relationships) more pertinent to decision-making in relation to risk management for work-related health (Karanika-Murray, Antoniou, Michaelides, & Cox, in press). The present study tested a nonlinear connectionist model vis-a-vis a traditional linear regression model of the impact of work characteristics on work-related outcomes. The remainder of this section discussing two key sources of uncertainty in modeling work-health relationships which set the main arguments for artificial neural networks, before it briefly outlines their principles. Uncertainty 1: Nonlinear Work-H...

‘Suddenly I get into the zone’: Examining discontinuities and nonlinear changes in flow experiences at work

“…Likewise, several studies have found that employee flourishing tends to behave in a nonlinear way Navarro, 2009, 2011;Guastello et al, 1999;Losada and Heaphy, 2004). Ergo, organizational researchers are increasingly favouring a nonlinear dynamical systems approach, which considers nonlinearity and discontinuous change, to study employee happiness and well-being (e.g., Navarro, 2009, 2011;Guastello, 2002;Karanika-Murray and Cox, 2010).…”

Forecasting the influence of customer-related micro-events on employees’ emotional, attitudinal and physiological responses

“…RI is an effect size metric that specifies the importance of a predictor considering all others, providing a description of the weights of all input variables (Lucek & Ott, 1997). A relative importance greater than .10 indicates major contributions, whereas those between .05 and .10 represent moderately, but still important, contributions (e.g., Karanika-Murray & Cox, 2010). As it is reported in Table 4, the high propensity for emotional contagion group was significantly (RI > .05) influenced by approximately all listed events, independently of their valence (nine out of thirteen events).…”

“…For the analyses, the sample was randomly divided in two: 70% assigned for training and 30% allocated for testing. This procedure allows us to track the errors and to prevent over-training (e.g., Karanika-Murray & Cox, 2010). We used batch training because it reduces the total error and it is more accurate to small-medium datasets.…”

Big Data Methods

“…Importantly, the detection of nonlinearity and interactions using these techniques can happen without prespecification of the exact pattern of nonlinearity. Numerous studies have found that artificial neural networks often outperform traditional regression based techniques in terms of their ability to predict counterproductive work behaviors (Collins & Clark, 1993), employee health (Karanika-Murray & Cox, 2010), turnover (Somers, 1999), and job satisfaction and performance (Somers & Casal, 2009). In all of these instances, the superiority of the artificial neural network was attributed to its ability to accommodate nonlinearity, even when compared to regression approaches with nonlinear terms included.…”