ACCURACY AND ERROR IN ELECTORAL FORECASTS: THE CASE OF MEXICO |
| |
| Authors: | Beltran Ulises; Valdivia Marcos |
| |
| Institution: | Ulises Beltran is the Director of the Survey Research Unit at the Office of the President of Mexico, serving two consecutive administrations since 1989. In 1982, he obtained a Ph.D. in Economic History at the University of Chicago and holds a previous degree in Sociology by the Universidad Nacional Autonoma de Mexico. He has taught in various Mexican universities and published several works in Economic History, politics and public opinion. He is a member of the American Association of Public Opinion Research, the World Association of Public Opinion Research and the American Political Science Association.
Marcos Valdivia currently is a graduate student of Economics in the New School for Social Research, New York. He obtained a Master degree in Philosophy of Science 1996) at the Universidad Autonoma Metropolitana, Mexico City, a Diploma in Applied Statistics (1993) by the Instituto Tecnologico Autonomo de Mexico and a B.A. in Economics (1992) at the Universidad Nacional Autonoma de Mexico. From 1991 to 1997 he worked as assistant to the Director of the Survey Research Unit of the Presidency of Mexico. |
| |
| Abstract: | Forecast errors in preelectoral polls are common. Since theNicaraguan Election in 1990, a common assumption is that wheredemocratic practices are not equal to those of the developedcountries (authoritarian, in transition, etc.), the errors inpredicting electoral outcomes with the help of pre-electoralpolls result mostly from contextual effects related to the authoritariannature of the political systems. Forecast errors in developeddemocracies have been explained by flaws in sampling (GreatBritain), last minute changes in preferences (Spain, UnitedStates, and France) and a social desirability bias in the UnitedStates. The intention of this paper is to show that in Mexicoboth the accuracy and the sources of error in predicting electoralresults are very similar to those confronted by pollsters elsewhere:mostly sampling and allocation of nonrespondents and those whoare undecided. An extensive data set of pre-electoral pollsconducted in Mexico shows an average error in the order of 3–5percentage points. Some of this error could be related to sociallydesirable answers derived from contextual effects, but thiseffect is small and must not be generalized. |
| |
| Keywords: | |
| 本文献已被 Oxford 等数据库收录！ |
|
