Revealing Hidden Curvilinear Relations Between Work Engagement and Its Predictors : Demonstrating the Added Value of Generalized Additive Model (GAM)
Tanskanen, J., Taipale, S., & Anttila, T. (2016). Revealing Hidden Curvilinear Relations Between Work Engagement and Its Predictors : Demonstrating the Added Value of Generalized Additive Model (GAM). Journal of Happiness Studies, 17 (1), 367-387. doi:10.1007/s10902-014-9599-z
Published inJournal of Happiness Studies
© Springer Science+Business Media Dordrecht 2014. This is a final draft version of an article whose final and definitive form has been published by Springer. Published in this repository with the kind permission of the publisher.
Previous studies measuring different aspects of the quality of life have, as a rule, presumed linear relationships between a dependent variable and its predictors. This article utilizes non-parametric statistical methodology to explore curvilinear relations between work engagement and its main predictors: job demands, job control and social support. Firstly, the study examines what additional information non-linear modeling can reveal regarding the relationship between work engagement and the three predictors in question. Secondly, the article compares the explanatory power of non-linear and linear modeling with regard to work engagement. The generalized additive model (GAM), that makes possible non-linear modeling, is compared with the widely used simply linear generalized linear model (GML) procedure. Based on the survey data (N = 7,867) collected in eight European countries in 2007, the article presents the following main results. GAM clearly fitted the data better than GLM. All investigated job characteristics had curvilinear relationships with work engagement, although job demands and job control relationships were almost linear. Social support had a clear U-shaped curvilinear connection to work engagement. Interactions between the three job characteristics were also found. Interaction between job demands and social support was curvilinear in shape. Finally, GAM proved to be a more practical and efficient tool of analysis than GLM in situations where there are reasons to assume curvilinear relationships, complex interactions effects between predictors. ...