The main objective of this study is to compare the quantitative (correct answers) and qualitative (error types) performance of children belonging to different linguistic groups on a non-verbal reasoning test, Raven's Colored Progressive Matrices Test, after being matched based on level of exposure to poverty, certain individual characteristics and test performance. The sample is representative for Mexico at a population level and consists of children aged 5 to 12 (n = 4644), of which 671 are bilingual in Spanish and an indigenous language, 3970 are monolingual in Spanish and 78 are monolingual in an indigenous language. The results show significant quantitative differences with a lower overall performance in the Raven's test by bilingual children as compared to their monolingual (Spanish only) peers, but no qualitative differences when analyzing their error types. When considering each linguistic group individually, the relative frequency of three error types (Repetition, Wrong Principle, and Incomplete Correlate) is similar in children aged 5 to 8 and in those aged 9 to 12. However, considering the two age cohorts, the results reveal how the intragroup differences in each linguistic group, are only statistically significant in the case of Difference errors, in the group of monolingual children in Spanish.

In addition to practical use that may be potentially derived from this empirical evidence, these results may also be encouraging from a methodological point of view. They demonstrate how the method used, in addition to permitting greater comparison between the experimental groups of a representative sample at a population level, does not present high sensitivity, either for the model used to estimate the Generalized Propensity Score method, or for the specific estimator used.