Una aproximación empírica al error de diseño muestral en las encuestas electorales del CIS

Abstract

Electoral polls conducted by the Centro de Investigaciones Sociológicas (CIS) are fairly thought of as a benchmark in the Spanish polling industry. The CIS uses a multistage stratified cluster sampling procedure, with selection of sampling units (census sections) at random with probabilities of selection proportional to the resident population. The sizes of resident and electoral populations, however, are unequal and their differences are heterogeneously distributed in the electoral territory. This paper quantifies (for a sample of elections) the impact of this discrepancy and suggests using as weighting factor for each census section the product of the number of Spanish residents aged 18 years or more and the section turnout rate in the previous elections. The analyses show that this strategy would significantly reduce the bias introduced in the sampling design.