Estimaciones de encuestas preelectorales en México: en busca de las principales fuentes de error

Abstract

In this paper we test different hypotheses that reflect some of the most common sources of estimation error in pre-election polls. We test for questionnaire design effects, sampling effects, interviewer effects, spiral of silence effects, and several contextual effects (such as the perception of safety or danger in a polling point in face-to-face polls). We analyze data from a state-level pre-election poll conducted in the State of Mexico on June 2011, two weeks prior to election-day. This poll included an embedded experiment about the placement of the voting question and recorded several contextual variables that allow us to test for different possible sources of estimation error. In addition, this paper offers a brief review of preelection polling in Mexico during the last two decades, evaluating the polls’ performance in both national and state-level elections. This analysis is part (and certainly the first formal step) of a larger effort by polling firms and public opinion researchers, as well as by the Federal Elections Institute, to determine the most common causes of estimation error in Mexican pre-election polls.

In this paper we test different hypotheses that reflect some of the most common sources of estimation error in pre-election polls. We test for questionnaire design effects, sampling effects, interviewer effects, spiral of silence effects, and several contextual effects (such as the perception of safety or danger in a polling point in face-to-face polls). We analyze data from a state-level pre-election poll conducted in the State of Mexico on June 2011, two weeks prior to election-day. This poll included an embedded experiment about the placement of the voting question and recorded several contextual variables that allow us to test for different possible sources of estimation error. In addition, this paper offers a brief review of preelection polling in Mexico during the last two decades, evaluating the polls’ performance in both national and state-level elections. This analysis is part (and certainly the first formal step) of a larger effort by polling firms and public opinion researchers, as well as by the Federal Elections Institute, to determine the most common causes of estimation error in Mexican pre-election polls.