Identifying the Component Structure of Satisfaction Scales by Nonlinear Principal Components Analysis

Abstract

The component structure of 14 Likert-type items measuring different aspects of job satisfaction was investigated using nonlinear Principal Components Analysis (NLPCA). NLPCA allows for analyzing these items at an ordinal or interval level. The participants were 2066 workers from five types of social service organizations. Our results suggest that taking into account the ordinal nature of the items was most appropriate. On the basis of a stability study, a two-component structure was found, from which we extracted two subscales (“Motivation” and “Hygiene”) with reliabilities of .81 and .77. A Multiple Group analysis confirmed this structure. We also investigated whether workers in the five types of organizations differed with respect to the component structure, employing a feature of the program CATPCA. We found that the organizations did not differ much with respect to the job satisfaction components.

Key Words:

• CATPCA
• Herzberg
• monotonic transformations
• multiple group method
• nonlinear principal components analysis
• satisfaction survey data
• scale construction

Additional information

Notes on contributors

Marica Manisera

Marica Manisera is a researcher at the Department of Quantitative Methods, University of Brescia, Italy. From 2005 to 2009 she was post-doctoral fellow at the same department. In 2005 she received her Ph.D. degree in Methodological and Applied Statistics at the University of Milano-Bicocca (title of thesis: “Measuring job satisfaction by means of Nonlinear Principal Component Analysis”. Advisors: Prof. M. Carpita and Prof. M. Civardi). In 2003 she was lecturer in Statistics at the University of Milano-Bicocca; from 2004 she has been lecturer in Statistics, Data Analysis, and Data Mining at the University of Brescia. Her research interests include the statistical methods and models for the evaluation of the quality of work and customer satisfaction; measuring nonlinearity in data analysis; minimax estimation methods in time series; missing data imputation techniques.

Anita J. van der Kooij

Anita J. van der Kooij is a researcher at the Department of Education, Data Theory Group, Leiden University, The Netherlands. Her research interests are optimal scaling techniques for multivariate analysis, and software development. She can be reached at [email protected]. For publications, see her personal page at www.datatheory.nl/pages/kooij.html

Elise Dusseldorp

Elise Dusseldorp was during the main part of this research Assistant Professor at the Department of Education, Data Theory Group, Leiden University. From September, 2007, she works as a statistician at the Netherlands Organization of Applied Scientific Research: TNO, division Quality of Life. She received her Ph.D. degree in the Social and Behavioral Sciences (2001) at Leiden University. Title of thesis: “Discovering treatment covariate interaction: An integration of regression trees and multiple regression”. Advisors: Prof. J.J. Meulman and Prof. S. Maes. From 2002–2006 she was post-doctoral fellow in the VENI-project “Modeling interaction effects as small trees in regression and classification”. Her research interests are interaction effects (moderators) in regression analysis, classification and regression trees, differential treatment effectiveness, aptitude-treatment-interaction, and optimal scaling. For publications, see her personal page at www.datatheory.nl/pages/dusseldorp.html