Measuring the similarity between languages

Abstract In typology, statistical methods have been successfully used to assess similarities and differences between languages. In creole studies, the use of quantitative methods has been discussed controversially. In the debate many methodological aspects of the statistical models used have been criticized (e.g. Meakins 2022 ; Bakker 2023 ). This paper presents an investigation of two methodological problems that have not been critically looked at so far: the question of which statistical models produce which results, and the question of how the amount of missing values in data sets influences the results. We present a study in which we tested different statistical models on 21 features from two arbitrarily chosen domains (‘Word Order’ and ‘Nominal Categories’) from the WALS ( Dryer & Haspelmath, 2013 ) and APiCS ( Michaelis et al., 2013 ) data bases. It is demonstrated that different statistical methods yield similar results, and that different sample sizes do not dramatically influence the model outcomes.