The Effect of Nonlinear Transformations on a Likert Scale

Abstract

The analysis of data collected from behavioral assessment instruments is typically conducted using parametric statistics, with little or no reference given to the underlying nature of the scale being used. If the nature of the distances between the scale points is not understood, the concept of normality of the distribution becomes clouded. An empirical approach to studying this problem was developed, using responses to a clinical performance evaluation instrument that uses a four-point behaviorally-anchored scale. Various combinations of nonlinear transformations were applied to the evaluation responses. The factorial structure of the fifteen items constituting the evaluation form was minimally affected by the transformations, suggesting that parametric statistics can be applied to behaviorally-anchored rating scales.

Ten Common Misunderstandings, Misconceptions, Persistent Myths and Urban Legends about Likert Scales and Likert Response Formats and their Antidotes

“…These data were collected from subjects who ranged from the eighth grade level to adults. The findings from these studies were supported and extended by other researchers [25][26][27] . The Vickers study in particular is an excellent exemplar of how "Likert scale" data can very closely approximate ratio scale data under a particular set of conditions [27] .…”

“…One, therefore, doesn't know how much stronger or how many more units "agree" is than "somewhat agree" (not to mention that a "somewhat disagree" response and "somewhat agree" response could actually represent a differing number of units of "agreement"!). But most of these same points are also true at the macro scale level even if the data are empirically an interval scale at this level, as in the studies done by Carifio [22][23][24] and the subsequent studies cited [25][26][27] . An interval scale has an arbitrary zero point (not a true zero point) so one cannot make ratios or interpret the data in "ratio ways" when the data is interval either.…”

Operationalizing professionalism: A meaningful and practical integration for resident education

“…38 The same is true when evaluating substandard performance. 39 One method that has been used to improve the value of such evaluations is to provide clear behavioral anchors, definitions of what actions align themselves with various ratings. 40 Faculty development programs can provide training to improve the validity and clarity of evaluations.…”

Quality of primary care from the patient perspective in Saudi Arabia : a multi-level study.

“…Third, the scale needs to contain at least 5-7 response categories. Finally, simulation studies showed that parametric analysis for Likert scale (under the above-mentioned conditions) produced consistent results even after manipulating the distances between the data points on the scale response categories (Baggaley, et al, 1983;Lumley, et al, 2002). …”

“…The first school of thought argues that Likert scale is an ordinal-level data and therefore it is most appropriate to conduct non-parametric analysis for this type of data (Gob, et al, 2007). The second school of thought argues that it is appropriate to use parametric analysis for Likert scale data under certain conditions (Baggaley, et al, 1983;Carifio, et al, 2008). First, the outcome variable needs to meet the normality assumptions.…”