Document View

Traditionally, researchers quantify customers' opinions using rating scales. Customers rate their overall satisfaction with an organization. Once the rating scale data have been obtained, many analysts use multiple regression to determine which subcomponent areas drive customer satisfaction. However, a binary overall customer satisfaction variable follows the logistic distribution and allows for the use of another regression technique: logistic regression.

Customer satisfaction research is supposed to determine which factors under the organization's control have the most impact on satisfaction. By understanding what variables drive satisfaction, the organization can change its business practices to create more satisfied customers, leading to increased sales and longterm viability.

The backbone of the process is data collection and analysis. Traditionally, researchers quantify customers' opinions using rating scales. Customers rate their overall satisfaction with the organization. In addition they usually rate the organization in several areas that directly affect the overall satisfaction rating, such as delivery time, out-of-stock conditions, helpfulness of sales personnel, accuracy of billing, etc.

Once the rating scale data have been obtained, many analysts use multiple regression to determine which subcomponent areas drive customer satisfaction. The overall rating is used as the dependent variable, the variable that is to be predicted. The ratings on the subcomponent areas are used as independent variables in the multiple regression model. These models predict the customer satisfaction rating based on customers' ratings of the subcomponent areas.

Multiple regression requires the research analyst to treat the overall customer satisfaction rating as continuous data. Also, multiple regression assumes that the ratings are normally distributed on a bell-shaped curve.

It is well-documented, however, that customer satisfaction ratings obtained on ratings scales are not normally distributed, but are skewed toward higher scale values. The main drivers of an organization's customer satisfaction are determined by how much each area contributes to the overall explanation of the multiple regression model.

In practice, customers and organizations do not view overall customer satisfaction ratings on a continuous basis. Ultimately, customers are either satisfied or not satisfied. Organizations routinely set criteria for determining who is a satisfied customer and who is not. On a 10-point scale, customers who give an organization an overall rating of 9 or 10 may be considered satisfied, while the rest of the customers are considered not satisfied. Viewing customer satisfaction ratings in this manner (are/aren't) invalidates the use of multiple regression because the dependent variable is not continuous, but binary.

However, a binary overall customer satisfaction variable follows the logistic distribution and allows for the use of another regression technique: logistic regression. The use of logistic regression is well-documented in predicting response to direct marketing campaigns. Like multiple regression, logistic regression allows the analyst to determine which area affects the prediction of a satisfied customer through the logistic regression coefficients and their associated "logodds." Log-odds specify the direct association between the independent variable and the dependent variable. In addition, logistic regression calculates the probability of each customer being satisfied or not.

To determine whether logistic regression could perform as well as or better than multiple regression in predicting satisfied customers, logistic regression models were compared to multiple regression models developed for customer satisfaction data for several companies in differing industries.

To create the binary dependent customer satisfaction variable for the logistic regression, customers were denoted as satisfied or not satisfied based on the companies' rules for determining satisfaction. In general, this rule is the top two boxes of the rating scale, or a 9 or 10 rating on a 10-point scale. The logistic and multiple regression models were developed on the customer satisfaction data provided in the surveys.

Since model "fit" statistics for multiple regression and logistic regression are not similar, the two techniques were compared to determine the percentage of survey participants correctly classified as either satisfied or not satisfied. Logistic regression consistently outperformed multiple regression in denoting satisfied and not satisfied customers in the nine studies undertaken.

In the first, for example, logistic regression correctly categorized 81 % of the participants, compared with 57% using multiple regression.