Electoral Survey: it went viral and you understood?

The Brazilian electoral cycle occurs every two years, alternating between municipal, state and federal elections. With the elections, electoral surveys appear, trying to predict the results, helping people vote and consolidating the electoral system.

But do you really understand the science behind this subject?

Jokes about polling institutes on voting intentions have gone viral on social media. The difficulty in portraying voters' real opinions is mainly due to the volatility of opinions (today you may prefer one candidate and next week another), but mainly due to the way the sample is selected. If you select your entire sample from an elite neighborhood or all of it from a peripheral area, the conclusions may be different. The art of selecting a good sample is called sampling. The rigor with which this sampling is done determines the reliability of the survey.

But how is it possible, in the presidential case, to portray the opinion of 155 MILLION VOTERS (TSE - 2024) with only a sample of 2,000 voters?

Objective answer: with the quality of the selection method and two indicators of uncertainty: margin of error and confidence level.

Why don't we select more voters in a survey? Basically for two reasons: time and costs.

After the sampling phase, we arrive at the calculation stage. In this stage, we calculate the desired measurement, usually an average, or, in the case of elections, a proportion. It turns out that the greater the variability, the greater the prediction error.

The sample depends greatly on the margin of error, with 2% commonly used in presidential elections and up to 5% being used in municipal elections. This difference occurs mainly due to the investment made in the survey.

There is also the confidence index, which is another measure of prediction uncertainty, usually set at 95%. Ultimately, the situation is aggravated by the fact that any percentage variation can change the election result. Therefore, in this case, precision is essential.

Not limited to the case currently in vogue, the applications of the science of Sampling are practically infinite. Other examples: the extraction of a drop of blood for a health examination, the quality of data from a judicial system, the quality index of a car being manufactured, and people's opinions about services provided by the government, among many others!

I conclude and dare to say: the applications of Sampling are infinite; our knowledge about these possible applications is finite.