Accuracy and Error in Electoral Forecasts: The Case of Mexico

In: International journal of public opinion research, Band 11, Heft 2, S. 115-134

Abstract

It is commonly assumed that, in countries where democratic practices are not equal to those of the developed countries (authoritarian, in transition, etc), the errors in predicting electoral outcomes with the help of preelectoral polls result mostly from contextual effects related to the authoritarian nature of the political systems. Forecast errors in developed democracies have been explained by flaws in sampling (GB), last minute changes in preferences (Spain, US, & France), & a social desirability bias in the US. Here, it is shown that, in Mexico, both the accuracy & the sources of error in predicting electoral results are very similar to those confronted by pollsters elsewhere: mostly sampling & allocation of nonrespondents & those who are undecided. An extensive data set of preelectoral polls conducted in Mexico shows an average error of 3-5 % points. Some of this error could be related to socially desirable answers derived from contextual effects, but this effect is small & must not be generalized. 6 Tables, 2 Figures, 25 References. Adapted from the source document.