Análisis de Componentes Principales con correlaciones policóricas: Aplicación en consumo de medios


This working paper describes Principal Component Analysis (PCA) as an exploratory and reductive statistical technique and its application with polychoric correlations. These bivariate models operate as an auxiliary technique and allow a better fit with ordinal/quasi-quantitative variables, which is helpful for social science research. A practical application is conducted using national opinion survey datasets with probabilistic sampling to explore media consumption and political participation in Argentina (N = 1,200), Chile (N = 1,200) and Uruguay (N = 1,202). The results allow us to identify three relevant findings. First, polychoric coefficients tend to be more accurate than Pearson’s coefficients for quasi-quantitative variables. Second, various component/factor retention methods are evaluated, and the usefulness of Parallel Analysis (PA) is demonstrated. Third, the extraction of polychoric matrices is not substantially different from that of Pearson. In sum, this paper shows an applied use of PCA with polychoric correlations that could be extended to other social science research.

Tufte Working Papers. OnlineFirst
Bastián González-Bustamante
Bastián González-Bustamante
Post-doctoral Researcher

Post-doctoral Researcher in Computational Social Science at the Faculty of Governance and Global Affairs at Leiden University, Netherlands.