Survey analysis using Likert scale

Likert scale
Likert scale is a psychometric scale (i.e. a scale which measures individual differences) that is commonly used in survey research involving questionnaires (i.e. instrument). Each question or statement of the questionnaire forms the “Likert item”. Likert item measures the participants’ level of agreement to a statement, such as “strongly agree” or “neutral” or “disagree” which are orderly numbered. Generally 5 levels of responses are used i.e.   1. Strongly disagree, 2. Disagree, 3. Neither agrees nor disagrees, 4. Agree 5. Strongly agree. However, more than 5 levels i.e. 7 and 9 levels are also sometimes used.

Before analyzing the Likert scale data, the reliability of the instrument or scale is performed. This is achieved by three different ways. First, the uniformity in response within the instrument (i.e. internal consistency) is measured by estimating the Cronbach’s alpha. A Cronbach’s alpha value of ≥0.7 is accepted. Second, the test-retest reliability is calculated.  In SPSS, the test-retest is calculated by bivariate correlation which is denoted by the Pearson’s correlation coefficient (r). Third, the inter-rater reliability is also estimated as test-retest in SPSS.

After determining the reliability of the instrument, analysis of Likert data is carried out. Each Likert item can be analyzed either separately (also called as Likert-type data) or summed to create a score for a group of items (summative scales, Likert scales). Each Likert scale consists of at least four or more Likert-type items, all measuring a single variable.

  1. Likert-type data is an ordinal data, i.e. we can only say that one score is higher than another. Due to the ordinal nature of the data, generally parametric tests (i.e. t-test, ANOVA) are not applied. Rather, non-parametric tests such as Mann Whitney-U test, Wilcoxon signed-rank test, Kruskal-Wallis test should be used. Descriptive statistics used for Likert-type data includes mode or median for measuring central tendency and frequencies for variability. Further analysis appropriate for ordinal scale items includes the chi-square measure of association, Kendall Tau B, and Kendall Tau C.
  2. Likert scale data, on the other hand, are analyzed as interval data. Since the data are of interval, parametric tests are used for analysis of Likert scale. Analysis that can be performed includes mean for central tendency, standard deviations for variability, Pearson’s r for bivariate analysis, t-test and ANOVA for comparing group means, and regression procedures for associations.
  3. When Likert-type or Likert scale data can be reduced to nominal level i.e. yes vs. no, agree vs. disagree, then, chi-square test, Cochran Q test, and Mc Nemar test can also be performed.

Further, Likert scales may be subject to biases from several causes. Central tendency bias occurs when respondents may avoid using extreme response categories; acquiescence bias occurs when respondents agree with statements as presented and social desirability bias occurs when respondents try to represent themselves or their institution more positively. All the biases can be checked by conducting a pilot survey before the actual study. If any biases are observed the questionnaire can be modified accordingly.  Crafting a scale with an equal number of positive and negative statements can counteract the problem of acquiescence bias. Central tendency bias can be avoided by either making the survey questionnaire short or by forcing comparable rating i.e and/or by randomizing the questions. Social desirability bias can be prevented by implementing the all the above methods i.e. to minimize central tendency bias and acquiescence bias and also by making the questions indirect.