Development of Norms for Verbal Fluency Test in Bilinguals Sample: a Generalized Linear Mixed Model with Poisson Distribution Approach

Abstract Objective To generate norms for monolinguals (MO) Spanish speakers, Basque bilinguals (BI) and Catalan BIs on verbal fluency tests (VFT). Method 89 MOs, 139 Basque BIs, and 132 Catalan BIs completed phonological and semantic VFT in Spanish and Basque or Catalan. Majority of the sample was women (62.2%) with age of 48.5 ± 18.2 and education 13.1 ± 3.8. The participants completed the task in two languages. Two generalized linear mixed models with Poisson distribution (GLMM) were used to evaluate the logarithm of the expected value using fixed effects (region, language, age, age2, education, and sex) and random effects (type of letter/categories, and participant). GLMM was conducted using a long data format with the total number of words as outcome variable. Results A GLMM Poisson for phonological VFT showed a quadratic age, logarithmic of education, region and language effects (ps < 0.001). The random intercepts for type of letter were significant (variance = 0.00875, p-value<0.001), indicating significant differences in total words through letters. Model fit was good (AIC = 24,521, BIC = 24,578). A second GLMM Poisson for semantic VFT showed a quadratic age, logarithmic of education (ps < 0.001) and sex effect (p < 0.01) on the total number of words. The random intercepts for type of category were also significant (variance = 0.02583, p-value<0.001), indicating significant differences in total words through categories. Model fit was good (AIC = 12,128, BIC = 12,185). Conclusions GLMM Poisson assumes that the response variable follows a Poisson distribution, as in VFT case, and it controls random effects, such as participant and range of letters in a given language. This approach could be adequate for generating norms in neuropsychological tests.