Likert scales (or ratings scales) and how to interpret them
A very common and useful question type on a survey is the Likert scale. That’s the kind that asks the respondent to agree or disagree with a statement, or rate the extent to which they liked something. Here is an example:
How satisfied were you with the variety of produce at our store?
- Extremely satisfied - Satisfied - Not very satisfied - Extremely dissatisfied*
You might end up with some responses like this:
Because there are four responses, and they have a clear order to them from best to worst, it’s very tempting to assign numbers to these answers and then average the responses.
But if you’ve ever done that, you’ve probably been admonished that that’s a terrible thing to do. If you haven’t been so admonished, you’re just lucky. Here’s why you should not do that, and a better way to represent the results of your Likert scaled question.
What’s wrong with the average?
When I assigned numbers, I used numbers that were evenly spaced. Essentially, I made an interval variable. That’s one where the values have an order and the steps between them are all equal. But the spaces between the answer choices in my Likert scale are not even. Is it the same distance from satisfied to extremely satisfied as it is from satisfied to not very satisfied? Nobody knows. In this case, I got an average of 2.77. How do I interpret that? Customers are almost satisfied? It’s not very helpful and actually removes a lot of detail from the data.
The problem is that a Likert scale is actually an ordinal variable, where the responses have an order, but the steps between them are not equal. So, when you average them, you get a nonsensical response.
So, how to present the findings?
In many cases, you’ll want to convert your number of respondents to percentages – that makes it easier for your reader to digest the information. Then, a stacked bar chart is a clear way to present the data. Stephanie Evergreen gives some great tips on how to present that here. My chart below relies heavily on her advice.
Now, you can see that more customers are satisfied than not satisfied. And it would be OK to aggregate those categories and say that 66% of the customers are satisfied or extremely satisfied. We can also see that a lot of our customers are satisfied, but we have some room for improvement. We might use this bar chart to set a goal to increase the number of customers who are extremely satisfied and decrease the number who are dissatisfied.
Here, I used color to distinguish the more positive responses from the more negative
responses and make the chart even easier for my reader to interpret. You might want to do this if you have a lot of categories. In any case, when you present this bar chart, don’t let excel pick your colors for you, but use color to illustrate the order of the response types. I’ve also presented my responses from most positive to least positive, in order to emphasize the positive responses and let us see how well we are doing.
Likerts are a very useful way of gathering information and with these tips, your findings will be even more usable.
*a note on response choices: I like to have an even number of answer choices on a Likert scale and not provide a neutral choice. Not everyone agrees on this, and there are some good reasons to include a neutral choice and to exclude it. One concern about not offering a neutral choice is that it might be uncomfortable for the respondent to be forced to pick an answer, and they’ll stop answering questions or just pick an answer randomly. I think this is mostly a concern if the topic is sensitive. When there is a neutral choice, I think it’s hard to interpret. Do they mean “I’ve never thought about this and don’t have an opinion”, or “I don’t have an opinion because this does not apply to me”, or “I don’t understand this question”? I think it’s better to give respondents more specific ways of expressing those things.