Importance of Scale Size in Satisfaction Surveys

By Dr. Dan Mount

Scale size is a key issue in creating a valid survey that provides meaningful data analysis. I have created, reviewed, and revised satisfaction surveys for years, and have found that the majority of research companies involved in customer and employee satisfaction work, use seven-point scales.

The goal of a properly scaled survey is to have a response pattern that provides the classic “bell curve” we all learned about in Statistics 101. The characteristics of a bell curve are that the curve has one peak (unimodal), responses are not bunched toward one end of the scale (skewness) and that responses are not bunched in one or two response categories resulting in a “steep” peak (kurtosis). If a company is using a methodology that requires more advance statistical techniques, such as “key driver” analysis or “critical issues analysis,” it is of vital importance that efforts be made to attain the bell curve response pattern.

What I have found when analyzing satisfaction data is that the typical response pattern, regardless of the scale size, will be skewed toward the positive end of the response scale, resulting in steeper peaks. It is in the best interest of your company to spread the results out without reaching a point where the scale is too large, resulting in missing data points.

In some instances, when three-point scales are used, the bandwidth is inadequate to get a true statistical read on a survey. With a five-point scale, the response curve is generally highly skewed and highly kurtotic. The resulting distribution curve may not even be a curve; it may start at the highest point of the scale with the most responses, and fall away from there down to the most negative response. Visualize half of a classic bell curve and you get the picture.

I have completed research that asked respondents to complete surveys with various scale sizes. Ten or 11-point scales will help distribute the responses. Respondents that answered with a “5” on a five-point scale ranged their answers from “11” to “9” on an 11-point scale. This helps flatten and lengthen the response distribution curve, but I have found that the lower end of the scale does not provide much insight. In many cases, there are missing or lightly populated response categories, such as a rating of “3” on an 11-point scale. Missing or lightly populated response categories seriously restrict the ability to provide more advanced statistical methodologies. My work has also found that these larger scales may provide curves that are bi- or even tri-modal (two or three peaks in the curve). The author of an article appearing in the 2003 Cornell Hotel and Restaurant Quarterly, also found that a seven-point response scale was the most effective in providing a valid response distribution pattern, thus corroborating my findings.

In summary, my research indicates that a seven-point response scale maximizes the ability to attain the most normal distribution in a response curve. While still slightly skewed toward the positive responses, and still slightly kurtotic, the seven-point scale more frequently provides a unimodal distribution that has sufficient data points in all response categories.

Dr. Dan Mount, CHA from the Penn State School of Hospitality Management, is also an honored UniFocus Advisory Board member.