Analysis of rating scales: A pervasive problem in bilingualism research and a solution with Bayesian ordinal models

Research in bilingualism often involves quantifying constructs of interest by the use of rating scales: for example, to measure language proficiency, dominance, or sentence acceptability. However, ratings are a type of ordinal data, which violates the assumptions of the statistical methods that are commonly used to analyse them. As a result, the validity of ratings is compromised and the ensuing statistical inferences can be seriously distorted. In this article, we describe the problem in detail and demonstrate its pervasiveness in bilingualism research. We then provide examples of how bilingualism researchers can employ an appropriate solution using Bayesian ordinal models. These models respect the inherent discreteness of ratings, easily accommodate non-normality, and allow modelling unequal psychological distances between response categories. As a result, they can provide more valid, accurate, and informative inferences about graded constructs such as language proficiency. Data and code are publicly available in an OSF repository at https://osf.io/grs8x.