Analyzing Likert Data

When our goal is to survey a group of individuals opinion toward a certain item or concept we can use scale-based approaches, the most common of which is Likert scale questions. Likert question an individual’s attitude toward a number of specific items along a 5 or more point scale with choices ranging from negative to positive attitudes toward the topic. Commonly, Likert scales follow the following syntax:

• Strongly agree = 5
• Agree = 4
• Neither agree nor disagree = 3
• Disagree = 2
• Strongly disagree = 1

Note: The wording of each question is important. A negatively worded question (such as including the word ‘not’) may cause a respondent to answer in reverse to if the question was worded positively (not including a ‘not’ or other negative term), and thus during analysis its scoring should be flipped (i.e. a ‘Strongly agree’ response to a negative-worded question would be scored as a ‘Strongly disagree’ response from a positive-phrased question)

Often it can be useful to also include a “Not applicable (N/A)” response so that if a respondent does not have an opinion or feels the question is outside of their ability to answer they are not forced to answer untruthfully.

Although the questions are on a scale, where “Strongly agree” responses are rated higher than “Agree”, which are rated higher than “Neither agree nor disagree”, and so forth, we cannot assume that the difference between the scores is the same. That is to say, if one respondent answers “Strongly disagree” to a question, we cannot conclude that they view that topic twice as negatively compared to a second respondent that only answered “Disagree”, only that the first individual has a stronger negative view. Therefore, we can treat Likert responses as ordinal data, where there is an ordered ranking among the responses but not equal separation between them.

Likert-type Scale vs. Likert Scale

One important consideration before performing your analysis, and one that should be considered prior to and while building your survey, is to determine what research question(s) you plan to analyze statistically and whether you will need to analyze Likert-type scale or Likert scale data (or both) to answer those research question(s). Likert-type scale refers to analyzing responses to any individual question. Conversely, with Likert scale analysis a series of questions that share a specific theme of interest will be analyzed together.

While both approaches can be very useful, the major quantitative difference between the two is the type of statistical analyses that are appropriate. For Likert-type scale data non-parametric tests that assess the median response, such as the Mann Whitney and Kruskal Wallis tests, and ordinal regression can be applied. Parametric methods, such as the t-test, ANOVA, and logistic regression, are instead applicable for Likert scale analysis.