Content area
Full text
Abstract
The data scale for Likert scale questionnaires is important in the determination of the statistical analyses that can be used to analyze the data. This paper describes a pilot study of a methodology intended to experimentally determine whether experimental subjects tend to give an ordinal or an interval answers in response to questions asked on Likert scales. Subjects answered 10 questions: once on a Likert scale without integer anchors and once on a Visual Analog Scale (VAS) with linguistic anchors. Hypothesis testing was performed to determine if the Likert levels were all the same size and evenly spaced as would be expected for an interval scale.
Keywords
statistical methods, data scale, categorical data, hypothesis testing
1. Introduction
Likert invented the Likert scale in the 1930's and it is still one of the most used methods in social, psychological, and human-factors research. There is a long-standing question as to whether Likert scales produce ordinal or interval data. The data scale is important in the determination of the statistical analyses that can be used to analyze the data. If the Likert scale produces interval responses then it is possible to use powerful analysis techniques such as parametric statistical methods, scatter plotting, correlation, and regression. However, if the responses to a Likert scale are ordinal then nonparametric statistics would be more appropriate and correlation, regression, and scatter plots lose their face validity. [1, 2, 3, 4, 5, 6, 7, 8].
The purpose of this study is to experimentally determine if research subjects tend to provide ordinal or interval-level data in response to questions asked on 5-level Likert scales without integer anchors. This will be examined by determining the range of VAS (ratio data) responses the subjects associate with each of the five levels of the Likert scale. It is hypothesized that subjects will provide ordinal level data. Specifically, we hypothesize that there will exist statistically significant differences between the sizes or spacing of adjacent Likert levels as demonstrated by the inter-quartile ranges or central tendencies of the VAS scores associated with each Likert level.
2. Methodology
This study experimentally examined the hypotheses by asking a set of 10 arbitrary questions to a moderate-sized set of subjects using both Likert and visual analog scales to record the...