Adapted from Dr. Claes Fornell's book | June 23, 2017
The Satisfied Customer: Winners and Losers in the Battle for Buyer Preference
Since most markets are fairly competitive and some have an abundance of buyer-choice options, a five-point or seven-point scale isn't enough. We have found that ten-point scales work best.
All measurement instruments, be they cameras, x-rays, or survey questionnaires, require sufficient resolution. If a binary instrument is insufficient, how many scale points should there be in a customer satisfaction questionnaire? Three? Five? Seven? Ten? The most common scales in public opinion polls have two (agree, disagree), five, and seven scale points. But such scales don't have enough resolution for customer satisfaction. Yet, aside from percentages, they are very commonly used by companies today.
So what's wrong with them? The answer has to do with frequency distributions. If one took a random sample of people's opinion of, say, McDonald's, Toyota, or Earl Grey tea, chances are that the responses would be distributed in the shape of a bell curve. That is, the responses would probably be normally distributed, with most of them in the middle and fewer at the end points of the scale. It turns out that a lot of things are normally distributed. It makes statistical analysis a lot easier when that's the case.
But, although people's opinions may be normally distributed in general, customers' satisfaction isn't. And that's exactly the way it should be in a well-functioning economy. But it's not well understood. As a result, inappropriate scales and incorrect statistics are being used. Let me explain.
Consider the tail of a bell curve in which the numbers are low. Here's where the really unhappy customers are. Suppose we are talking about Coca-Cola. One would expect to find very few customers in this tail. If you don't like Coke, you won't be a Coke customer for long.
Bear in mind that we are not measuring the public at large, but customers. We are not measuring people who are not customers and for those who don't like Coke, the cost of going elsewhere is low. There are many other soft drinks available. And if that's not enough, there are other beverages. The buyer-seller exchange is quick, efficient, and not burdened by high costs per unit. On the buyer side, there is no learning cost to speak of, no service is necessary, and there is little buyer risk in moving from one product to another.
On the opposite side of the frequency distribution is where the satisfied customers reside. This tail is fat. People who buy Coca-Cola like Coca-Cola. If they didn't, they wouldn't continue to buy it. Now then, what does this mean for instrument resolution and the number of scale points?
Let's try a scale of five points, with very satisfied and very dissatisfied as the end points. What will we get? To be sure, we're not going to see many ones, twos, or threes. These customers have already left. For all practical purposes, the five-point scale becomes a two-point scale and we're back to the problem of having a binary scale that doesn't have enough resolution.
Would a million scale points be better than 100 scale points? Obviously, the million scale point has greater resolution. A billion scale points would be even better. More is preferable, but always tempered by the respondents' ability to discriminate between scale points and the usefulness (to management) of additional granularity.
So where do we end up? The general principle of non-normal frequency distributions holds in any market where there is sufficient consumer choice and where it's possible to go from one product to another without too much trouble. It's only when dissatisfied customers have nowhere to go or find it too expensive to get there, that we will get closer to a normal distribution. Statisticians would like that, but the rest of us wouldn't. The economist would see a market that's not functioning well. Dissatisfied customers would be locked into something they would prefer to get out of.
Since most markets are fairly competitive and some have an abundance of buyer-choice options, a more granular measurement instrument would be desirable. A five-point scale isn't enough. Even a seven-point scale is questionable. We have found that ten-point scales do well. The responses exhibit a reasonable dispersion and respondents are able to discriminate between scale points. That's not the situation in public opinion polls, where the questions often pertain to things that the respondents are not all that familiar with. The opposite is true, of course, when the topic is one's own consumption experience.
In sum, what this means is that well-functioning markets, where there is enough buyer choice, suggest a scale of about ten points. Since the respondent, by definition, has actual experience, it's also possible to use the same type of scale in markets where there is some degree of monopoly power or in other contexts in which the freedom to choose is curtailed (e.g., government services).