Content area
Full Text
Partial Least Squares vs. Regression
When assessing customer loyalty modeling issues, Partial Least Squares (PLS) analysis produces accurate results. In fact, the authors argue, PLS works more effectively than Multiple Regression or Principal Components Regression for a typical loyalty model.
Customer satisfaction, loyalty, and retention research programs have moved well beyond the reporting of ratings to the diagnosis of causes and effects. This article assesses two satisfaction and loyalty modeling issues. First, although satisfaction long has been assumed to predict loyalty, a growing movement has emerged to replace satisfaction with value.1 We will argue that loyalty models are inherently complex and suggest a procedure that allows managers to include and assess the relative predictive power of satisfaction and value. Second, regression-based models frequently are used to sort through and quantify the relative impact of posited loyalty drivers. We will use an energy utility example to show that Partial Least Squares (PLS) analysis produces more accurate results than either Multiple Regression or Principal Components Regression (PCR) for a typical loyalty model.2
CUSTOMER LOYALTY MODELS
Loyalty models may contain several salient drivers. For example, automobile customers typically hold perceptions about the product, sales experience, and service experience.3 Although each perception may drive overall satisfaction, value, or loyalty, some undoubtedly have more effect than others. The problem, then, is how to separate the few important drivers from the trivial many. Without the ability to identify strong drivers, managers are apt to waste resources by improving perceptions that contribute very little or miss opportunities to improve loyalty.
In addition to containing many salient drivers, the relationships among the variables in a loyalty model are complex. The model shown on pg. 20 represents a typical set of drivers for a large energy utility. The model's purpose is to diagnose and predict changes in loyalty. Six independent variables, or loyalty drivers (price, core relationships, energy services, corporate activities, and service and billing) and three mediating variables (value, corporate reputation, and overall satisfaction). The italicized text in the boxes beneath the variable descriptions describes the battery of items from the two surveys used in this study. While the model is structured simply for purposes of exposition, in reality it may contain two or three tiers of drivers or sub-drivers.
The analysis...