SEM-based customer satisfaction measurement: On multicollinearity and robust PLS estimation

Abstract

The performance of companies is traditionally rated by financial performance measures, such as EBITDA, EBIT, ROCE, cash-flow, etc. However, the use of non-financial performance criteria, such as human capital, brand equity, customer asset and environmental performance, as valuation tools additional to the financial performance measurement, has increased over recent years. Information on non-financial performance is used in Balanced Scorecards for corporate control of business units, and is reported internally as well as externally. One way of rating and assessing the customer asset is through customer satisfaction measurements. Some national indices for rating customer satisfaction have been developed over the last decade. One example is the European Performance Satisfaction Index (EPSI Rating), which has conducted surveys since 1997. Within the framework of EPSI Rating, Partial Least Squares (PLS) is applied as the common statistical method. The hypothesis of PLS being robust against various statistical specification problems, such as multicollinearity, is an important argument for using PLS. The aim of this article is to examine the effects on PLS estimates of the inner relations in a Structural Equation Model (SEM), when introducing multicollinearity in the data. This is done through Monte Carlo simulations. Overall, the simulations confirm the robustness hypothesis.

Keywords:

• customer satisfaction measurement
• EPSI Rating
• Structural Equation Model
• multicollinearity
• Partial Least Squares
• robustness
• Monte Carlo simulations