How Many Points On A Likert Scale?

In trying to determine how to adapt a Likert-style survey for quantitative data analysis, I found a research study from 1971 that stated it was the first of its kind to measure how many response options impact validity and reliability.  The article is in the Journal of Marketing Research and written by Jacob Jacoby and Michael S. Matell.  The title?  “Three-Point Likert Scales Are Good Enough.”  So great, there is my answer!  I referenced this article, as have 167 others, and kept on going.  To be sure, I double-checked on the authors.  Jacob Jacoby has a Ph.D. in Social Psychology and teaches Research Methods at NYU.  He wrote many books, articles, chapters, and appears to be a subject matter expert.  I didn’t check on Michael S. Matell at that time, but since then have not found the same level of output.  This study showed that the relationships of validity and reliability to number of alternatives in a survey, display a lack of significant difference between a multi-stepped, dichotomous, or trichotomous questionnaire format.


The following year (1972), the same journal asks “Are Three Point Likert Scales Always Good Enough?”  This title is suggesting to me, “No.”  The author, Donald Lehmann, a professor at Columbia University, is just as distinguished and impressive as three point Jacob Jacoby.  You may agree, have no opinion, or disagree.


In Google Scholar alone, there are 234, 000 hits on “how many points on a Likert scale.”  Consider culture, even vs. odd, the midpoint, and a Monte Carlo Approach.  The Monte Carlo approach is a mathematical model that determines a way around the acknowledged (thank you!) contradictions and confusion about the optimal number of points for reliability.  I did not read this article, but the title is encouraging.


One of the best things I can do here for now is to get out of the 1970s and adjust my research options to at least the past four or five years.  It brings me down to only 15,430 hits.  It should keep me busy, and I will let you know what I end up with.


Even better, would anyone mind letting me know?  How many points do you strongly agree should be on the Likert scale?  Also, could you include a reference?

This entry was posted in Blog and tagged , . Bookmark the permalink.

One Response to How Many Points On A Likert Scale?

  1. Jonathon isham says:

    I just finished a dissertation where I used a likert scale, and I have two suggestions…which may be contradictory (sorry about that).
    First, I was advised that the number of items on a likert scale are governed by the amount of granularity that your research needs. I needed to know if people agreed or disagreed (or were neutral). For my research, there was really no difference for me between agree and strongly agree, so I went with a three point scale. If there is a difference for your research, then go with more. I have seen Likert scales of up to 11 points. Also look to see if you need neutral or do you want to force an opinion.
    Second, I have seen research that discounts the statistical validity of using likert-type data for quantitative analysis because Likert scale data is ordinal, and most quantitative data procedures need at least interval data (standard deviation means nothing if the distance between “agree” and “neutral” are not the same from one person to the next…). One way around this is using Tastle and Weirman’s (2006) consensus measure instead of standard deviation. That one saved my bacon. One caveat, though. the larger the Likert scale, the better the consensus measure is. One of my identified weaknesses in my dissertation was that with a three point scale, the survey answers needed to have almost unanimity to exceed the consensus measure I used. that made it difficult.
    So, conflicting advice…Likert scale amounts are based on your need for data granularity so only use what you need, and the consensus measure works better with more scale points.

    Tastle, W. J., & Wierman, M. J. (2006). An information theoretic measure for the evaluation of ordinal scale data. Behavior Research Methods, 38(3), 487-94. Retrieved from

Leave a Reply