Likert scale is an ordinal scale , therefore kruskall-wallis should be ideally used but why do researchers use ANOVA ?
Is there any condition either when to apply ANOVA or Kruskal-wallis on analyzing Likert scale?
Hi, Ashish.
The LS is often defined as ordinal, but it is commonly used as if it were a quantitative scale (under some assumptions). It can even be treated as nominal as well. The Likert scale (LS) is associated with a type of the variable and a postulate (item) *. In each study, it should be defined and justified. I don't want to argue if it is mathematically correct or not or if it is Likert or Likert-type, Etc; but I can suggest practical options:
a) If the LS is considered ordinal or nominal, the homogeneity hypothesis tests for contingency tables (Chi-square) may be useful.
b) If the LS is considered quantitative (discrete or continuous), use ANOVA, provided that the sample is very large or you have evidence of the normality of the variable within each of the factors **.
c) If the LS is considered quantitative (discrete or continuous), but your sample has a skewed distribution (no-normal), use Kruskal-Wallis. Consider that the distribution of the variable must be similar in each factor **.
*Examples.
-Item: The quality of the RG portal is very good.
Type of variable: Ordinal. > Slightly agree, Agree, Totally agree.
Type of variable: nominal> Disagree, neutral, Agree.
-Item: Evaluate the quality of the RG portal, from 1 to 10, where 10 is the maximum positive evaluation.
Type of variable: quantitative discrete> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10.
Type of variable: continuous quantitative> Indicate the quality position with an "x"
1 -----------------------------------------------10
** You can always do numerical transformations to modify a distribution. You can make it normal or skewed on purpose to meet the assumptions of each test.
Hi, Ashish.
The LS is often defined as ordinal, but it is commonly used as if it were a quantitative scale (under some assumptions). It can even be treated as nominal as well. The Likert scale (LS) is associated with a type of the variable and a postulate (item) *. In each study, it should be defined and justified. I don't want to argue if it is mathematically correct or not or if it is Likert or Likert-type, Etc; but I can suggest practical options:
a) If the LS is considered ordinal or nominal, the homogeneity hypothesis tests for contingency tables (Chi-square) may be useful.
b) If the LS is considered quantitative (discrete or continuous), use ANOVA, provided that the sample is very large or you have evidence of the normality of the variable within each of the factors **.
c) If the LS is considered quantitative (discrete or continuous), but your sample has a skewed distribution (no-normal), use Kruskal-Wallis. Consider that the distribution of the variable must be similar in each factor **.
*Examples.
-Item: The quality of the RG portal is very good.
Type of variable: Ordinal. > Slightly agree, Agree, Totally agree.
Type of variable: nominal> Disagree, neutral, Agree.
-Item: Evaluate the quality of the RG portal, from 1 to 10, where 10 is the maximum positive evaluation.
Type of variable: quantitative discrete> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10.
Type of variable: continuous quantitative> Indicate the quality position with an "x"
1 -----------------------------------------------10
** You can always do numerical transformations to modify a distribution. You can make it normal or skewed on purpose to meet the assumptions of each test.
The explanation by Francisco Javier Urcádiz Cázares is appropriate.
Good luck.
Thanks Francisco sir
What are the assumption under which likert scale is considered a quantitative scale ?
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA