Using ordinal logistic regression to analyse self-reported usage of, and attitudes towards swearwords

Abstract

Likert-type data is commonly used in many research fields in humanities: from gaging the usability of different user-interface designs, to determining users’ likeliness to vote for a particular political party, to evaluation of course materials–to name but a few examples. Despite its prevalence, there is still some disagreement within the statistics community on whether Likert-type scales are true ordinal variables, and by implication whether parametric tests are legitimate to be used in such cases (Endresen &Janda 2017).In this paper, we explore one parametric statistical test, viz. cumulative odds ordinal logistic regression (OLR),as an analysis method for self-reported data in the humanities. For illustration purposes, our focus is specifically on data of users’ self-reported usage of, and attitudes towards swearwords, with the aim of identifying demographic attributes that are predictive of their usage and/or attitudes. After a brief description of the data we’re using, including how the data is being collected, we give a layman’s overview of OLR. Since one of our aims is to demonstrate the usability of OLR, we apply our discussion practically to a step-by-step procedure (based on Laerd Statistics 2015) that could be followed easily. We demonstrate the usefulness of the results in reporting on the usage of, and attitude towards two near synonymous Afrikaans swearwords. We show, amongst others, that the odds ratios that are generated as part of the modelling procedure can be used to draw direct conclusions about specific demographic groups.

The Confluence of Supplemental Instruction (SI) Programme Factors on Selected Student Outcomes in a Historically Disadvantaged University

“…Considering the measurement scales used, [40] maintain that if researchers wish to study the effects of explanatory variables on all levels of the ordered categorical outcomes, an ordinal regression method is more appropriate for valid results. In this instance, both the predictor variables and the outcome variables were ordinal scales (Likert scale data) [41].…”

Perceptions of food environments in the school and at home during Covid-19: An online cross-sectional study of parents, teachers and experts from Latin America

“…In this graph, the independence is presented if the boxes across categories all have the same area [ 25 ]. When there was a statistically significant association in the bivariate analysis, an ordinal logistic regression model was run; and to measure the multivariate associations between the explanatory variables, the probability of the response categories per profile of the adjusted models was determined and plotted, with 95% confidence interval (CI) and respective P-value [ 26 ]. It was considered statistically significant when p<0.05; all analyses were performed using STATA 14 software for Mac.…”