There are many types of satisfaction scales that are used in market research and beyond. This diversity stems from the fact that there is no magic ruler to directly measure distances associated with attitudes. For the researcher, a major consideration is how is this data to be analyzed?

Four Different Modeling Types

Long ago, you probably learned about four different modeling types for data: nominal, ordinal, interval and ratio. Nominal data simply refers to unordered categories such as: “Red”, “Blue” and “Green.” Analyses for nominal data are limited to counting and filtering. Ordinal data, such as “Detractor”, “Passive” and “Promoter”, have an inherent order and allow additional sort-based statistics such as the median. Interval and ratio data both have the very important concept of distance which allows calculations such as mean, variance and Pearson correlations. Correlations amongst distance variables are fundamental building blocks for many types of advanced analysis

All Scales Are Ordinal in Nature

So you might be saying, “How is this related to labeling and satisfaction scales?” Well, technically all such scales, labeled or not, are ordinal in nature and limited to statistics such as top box and median. (Ok, don’t panic yet –keep reading on.) In Figures 1 and 2, there is an example of a fully labeled scale and an example of an end-point only labeled.

An example of a fully labelled scale

Figure 1 – Full labeled satisfaction scale

End point scale

Figure 2 – End-point only scale

Recommended for You


Webcast, April 4th: How This Startup of the Year Accelerated User Acquisition Growth 18x in 10 Months

Now, I know some of you are saying: “Wait a minute, I’ve seen averages and variances calculated with satisfaction scales.” How can this be? Well, the answer really lies in when you can treat an ordinal rating scale as if it were intervally scaled. Let’s start out with recommendations for the “extreme” situations and then we will look at the in-between situations.

Recommendations:

  • If you have four or less gradations in your rating scale, like in Figure 1, you should treat it like an ordinal scale.
  • If you have 10 or more gradations, that are endpoint only labeled, then you can treat it as if it were interval. This recommendation will save you from some bar fights with statisticians!

Ok, what about the in-between situations? Pay attention.

  • It is usually ok to treat a 5 point scale, fully labeled or not, as if it were intervally scaled with some conditions.
    1. If your responses aren’t too skewed, i.e., everybody gives you a 4 or 5 thus you really have a two-point scale.
    2. If you use a fully-labeled scale make sure the labels are reflective of equidistances (i.e. equal “distances” amongst the labels.) Figure 3 shows a well-constructed scale whereas Figure 4 shows a poorly constructed scale. Do you think that the “distance” between “Terrible” and “Poor” is the same as the distance between “Very Good” and “Excellent?”

Well constructed scale

Figure 3: Well-constructed scale

Poorly constructed scale

Figure 4: Poorly-constructed scale

The recommendations for more than 5 gradations are similar to a 5-point scale. However, for practical reasons, as you get more gradations it becomes harder to “think of” and physically display labels for all gradations. Consequently, end-point only labels are more prevalent though you will see some 7-point fully labeled scales. One could argue end-point only scales are much easier for the respondent to interpret as distances.

Most of these recommendations revolved around analysis – you can’t blame me as I have a statistical bias. However, there are some additional considerations, such as the type of survey and what type of scale is prevalent in your industry. For example, employee satisfaction scales or automotive studies often use fully-labeled 5-point scales. I don’t advise challenging these “norms” unless you think that is easier than a bar fight with a statistician.

A Word of Caution

I’ll leave you with two other cautions. If you are contemplating adding a midpoint label to a scale that is now end-point only, as is done for NPS scales, think twice about that. You may ruin the distance feature – see the blog. Also, don’t use numbers with a fully labeled scale as that adds to respondent confusion as some will focus on numbers and others focus on the labels

If you need more scale advice, feel free to reach out.