Is there a function to get the optimal number of choices in a Likert scale?

Question

I was answering a question today and it reminded me of a discussion I had with a couple colleagues about the optimal amount of choices in a Likert scale. I vouched for the classic 5, one of us stood for 10 and the third person liked a variable approach where 10 options should be used for in depth research with a great deal of fine tuning while fewer choices should be used on superficial research and the extremes of age targets (younger - older). Nevertheless, while I could see some logic in the second guy's approach, it didn't convince me at 100%, maybe because I'd only use a Likert scale if I can't use a Semantic Differential Scale.

Now I'm working with a behavioral psychologist and the first test she did used a Likert Scale with 5 choices. I asked about the number of choices and she told me you can use a variable number of choices, but in the end, when translated to a mathematical matrix, it returns more or less the same median, variance and kurtosis.

However, I'd never pass on a chance of knowing a better way to use a tool, so, In short, my questions are:

• are there any studies that unequivocally demonstrate the optimal number of choices in a Likert scale?
• assuming "variable" is the correct answer, are there any tools or guidelines that allow the researcher to define this optimal number?
• is it correct that 5,7 or 9 options will provide more or less the sam results? Or more choices equal more fine tuning

Please notice I already know about the different theories defending what my colleagues stood for, I'm probably looking for an answer involving maths or some kind of function that statistically shows the answer is correct in as many different scenarios as possible

Answer 1

There seems to be plenty of research around this particular topic, and not surprisingly it has been covered in a number of different psychology and marketing research papers (just google "optimal number of choices in a Likert scale").

Unfortunately, it is not easy to decide on what is the optimal number of choices because there are a number of different factors involved. My understanding is that as the number of choices increase, it becomes more difficult for the user to decide or discriminate between the value assigned to the choices (e.g. what the difference between a 5 and 6 is).

However, if you don't provide enough choices, then it is very likely for responses to be biased towards certainly values, especially when the question is worded in a biased manner as well (e.g. loaded questions that prompt for a positive or negative response).

It also depends on whether the user is being asked for a qualitative or quantitative measure on the Likert scale. So if the values at the ends of the scale are phrase such as "More likely"/"Less likely" you should probably consider less choices as opposed to "Complete"/"Incomplete", which should have more choices because it is easier to determine the degree to which they think something is complete versus something being more likely.

A summary of the answer to the questions are:

• YES, there are studies but it is important to understand whether the studies test the same type of things that you want to find out because the way the test is created and conducted may affect the answer.
• YES, the guideline would be based on the type of question you are asking and the type of response required, specifically relating to the accuracy and precision that the respondent can be expected to provide.
• PROBABLY, I would expect little difference between 5 and 7 or 7 and 9, but I would expect more difference between 5 and 9.

This article gives a good summary of the things to watch out for in Likert scale usage, both around the number of choices and whether you should have an even or odd number of choices.

Answer 2

I agree with Michael in that true validity of using this type of self-administered questionnaire really depends on how focused your questions are. As you've said, the research does not appear to find a clear winner with regards to the number of choices. Interpreting the data can also be done in a number of ways. Sometimes looking at mode instead of the mean or the distribution on a bar chart can help to clarify.