To cut meat, would someone use a spoon? Surely yes. There is always the exception that confirms the rule. But, without a doubt, common sense is different: better a knife for the meat and a spoon to stir the sugar in the coffee.
We should also apply this obviousness when assessing surveys. We demand something from them that they don’t know how to do. They are lightly condemned as useless because they do not perfectly estimate an electoral result, when the juiciest thing about a survey is that it helps us in a good sociological, political and electoral diagnosis.
The surveys are one more input. Complementary to others. Never a substitute for anything. And it is useless as a shortcut to analyze what is happening in a country. Nor are they infallible in accurately estimating voting intention. Polls are by no means an accurate crystal ball in this regard, and those of us who work with them know it. And we tried to explain it, but without success.
Although it will not be for not continuing to try. Here it goes one more time.
Surveys are very useful to detect common senses. ideological cleavages. tensions. Trends. Loyalties and antipathies. Everyday concerns. Likes and dislikes. dominant frames. Public policy evaluations. Levels of knowledge about candidates. conjunctural positions. And much more. That is: in a survey there are many more questions and variables that are analyzed together, and never in isolation. Much less the intention to vote.
This abundant information helps us guide the political-electoral strategy, without this meaning that each decision is made based on what the survey shows us. Because of
follow what the polls say we would be doing something quite different, which would have nothing to do with
Surveys are definitely very useful because they provide a complex picture of a complex reality. For this reason, reducing them in their interpretation to the simplicity of a datum, to an exact number that facilitates the estimation of a candidate’s vote, is to want to eat meat with a spoon.
This erratic reading of the polls also ignores the fact that each study always publishes its margin of error. And this is not a statistical whim. It means that if, in the case of Lula, there were many polls that estimated a vote intention of 50 percent, what they were really saying is that she could get between 48 and 52, which is what happened.
If we continue with this last case, that of Brazil in the last presidential election, we can observe another notable phenomenon that must be considered in each survey: its rate of non-response or rejection. That is to say: to be able to interview 10 people, in many situations you have to try with 100. People do not always want to answer, and less so during electoral times. This happens for many studied reasons, different for each country. In addition, this parameter varies depending on whether it is an IVR telephone, CATI telephone or face-to-face survey. This is an increasingly determining variable in the reading of each study and cannot be ignored. It is very likely that the underestimation of some polls regarding the vote obtained by Bolsonaro is due to this matter.
Nor can it be ignored that most polls end days before the vote. Consequently, it is obvious that they cannot estimate what happens in decisive hours in each election. This is because there is usually a notable percentage that is defined at the end, in the last 24 or 48 hours, either because they are in doubt, because they are not clear or because they choose to change at the last minute, resignifying the usefulness of their vote.
There are many methodological aspects behind a survey. From its extension, its sample, its weights, the number of days it takes, etc. Clearly, they all condition each value that is obtained. It is therefore crucial to know how to read and interpret the numbers assuming each bias. Otherwise, it would be like taking a medicine without knowing its contraindications.
And a final recommendation that is usually helpful to make fewer mistakes when reading the surveys: look at all those that are published and analyze them comprehensively. We have been doing this at Celag and it allows us to have a better thermometer than if we draw hasty conclusions based on the last one published. For example, when doing a fortnightly average of polls published in Brazil, the last data we had was Lula with 44 percent of the total vote and Bolsonaro 36. These data must be projected to a valid vote because, if not, it would fall into another classic mistake when reading survey data. That is, only values that are in an equivalent dimension can be compared. Therefore, the most rigorous thing is to have everything in a valid vote, and in this way it will be possible to compare the day of the vote. So, in the example we have, the biweekly average data for Lula on valid votes is 47.6 percent and Bolsonaro 39.1. In other words, close to reality without being completely precise: Lula’s within the margin of error (he scored 48.43); and Bolsonaro’s a couple of points above his top end if we consider his margin of error (he scored 43.2).
* Doctor in economics, director of Celag