Article published by José Alberto León Alonso in El Día on 10/18/2015.
Citizens have been subjected to a bombardment of electoral polls with disparate results regarding the elections to be held in 2015, so they end up concluding that the polls are useless, which is the same as what politicians think...when they are unfavorable to them. Given that at Corporación 5 we have been conducting market research for more than eighteen years through our subsidiary SAO, I will try to clarify doubts about the plausibility of the surveys, as well as describe our own methodology for conducting them.
To determine the reliability of a survey, it is necessary to know its working methodology, and unfortunately most surveys hide or conceal fundamental data to differentiate a technically correct survey from one that is not. Every survey, to be credible, must indicate the number of interviewees, whether the interview is in person or by telephone, direct voting intention, political sympathy, vote recall, response rate, callback strategy or substitution rule. Likewise, the sample must be well segmented by age, sex, residence, beliefs and social status of the voters. These points, for example, are not met in a good number of the polls analyzed for the general elections of December 20. Even in reliable polls, the difference between the direct intention and the estimate of the vote is attributed by many to “cooking” carried out with dark intentions. In reality, all serious surveys correct the results obtained in direct voting intention. Because?
Direct Voting Intention (DIV) is the most immediate and spontaneous response when asking who would be voted for if the elections took place tomorrow. About 60% of those surveyed usually respond to this direct question, but we know that part of the other 40% finally goes to vote on election day. To obtain a faithful representation of the electoral result, it is necessary to assign a probable vote to those who do not answer the direct question. To do this, they are asked about the party with which they feel the most affinity or with which they identify to a greater extent: this response constitutes the “Sympathy” that is added to the IDV. In order for an estimate of results to be extrapolated from this data, the sample (the people surveyed) would have to exactly reproduce the universe of potential voters. Something impossible, since when selecting the sample it is not possible to know, a priori, the vote of the respondents, since it is secret. However, well-conducted surveys also ask about vote recall from past elections, which allows us to detect a non-representative sample (with more or fewer voters from a party than should be), or a hidden vote. In both cases, the results must be treated and corrected so that they resemble reality as closely as possible. This is where the famous “kitchen” of surveys comes in, which, if it is professional, is not only appropriate but essential. As long as an investigation is methodologically sound, with a representative sample and applies the pertinent post-survey adjustments to the data, it is possible to approximate a good forecast of the electoral results.
At Corporación 5-SAO we have our own methodology to make electoral estimates. Given that most polls commit biases and errors introduced by the sampling methodology of each pollster, we use the poll aggregation technique, which allows us to reduce the volatility in the predictions. Overall, the aggregation technique substantially improves predictions of election results. For example, in the recent Catalan elections, polls that used this technique predicted the final results better than 95% of published polls. Likewise, to reduce the systematic biases of each pollster, we work with the unadjusted data, weighting them according to the number of respondents, the proximity to election day and the success of the pollster in the past. The weight given to the date of data collection is particularly important, especially in these days when electoral sentiment is more fickle than ever in our democratic history. Thus, at C5-SAO we have been reducing these weights exponentially over time, so that surveys carried out more than fifteen days ago have a relevance of less than half compared to the closest ones.
With this methodology, after the Catalan elections there is a certain reduction in the vote estimate for PP and PSOE (-0.6 p.p. each), a significant fall for Podemos (-1.6 p.p.), as well as a strong increase in the vote estimate for Ciudadanos (+3.6 p.p. in one month). Ciudadanos improves in all the polls carried out since September 27, benefiting from the well-known “honeymoon” effect, by which the winning parties in an election temporarily improve in the polls. Whether Ciudadanos' progress will be firm or not will be seen over time.
U!
U!
