Distrust of electoral surveys is common and cuts across different voter profiles. The recurring question in electoral cycles is: how can you trust a survey if the majority of people have never been interviewed? The doubt comes from an intuitive perception — but it ignores the central principle that supports opinion surveys, statistical sampling.
In Brazil, with around 150 million voters, institutes can map trends with around 2,000 interviews in state surveys and up to 5,000 in national surveys. The point is not in the absolute size of the sample, but in the way it is constructed.
“It’s a sample of society by salary, by age, by sex. It’s a cut”, explained Real Time Big Data strategist, Wilson Pedroso, when detailing how the institutes replicate, on a reduced scale, the composition of the electorate.
The logic is similar to that of laboratory tests. It is not necessary to analyze the whole to obtain a reliable diagnosis. What matters is that the sample accurately represents the characteristics of the set.
How the sample is formed
To construct this profile, the institutes use official data, such as IBGE, to calibrate the interviews. This means ensuring proportions similar to the real population in criteria such as gender, income, age and region.
In practice, as these groups are filled, the interviews are closed for that segment. When, for example, the number of women reaches the expected percentage, new responses from this group are no longer incorporated.
Continues after advertising
The process is repeated regardless of the methodology, whether carried out in person, by telephone or over the internet, each with its advantages and limitations, but all follow the same principle of reproducing, on a reduced scale, the structure of society.
Why few interviews are enough
From a statistical point of view, the sample size defines the margin of error and the level of confidence. In general, surveys with around 2 thousand interviews operate with a margin of error close to two percentage points and a confidence level of 95%.
This means that, in 95% of cases, the results will be within this range of variation. Differences between surveys, therefore, tend to reflect this expected fit and not necessarily survey flaws.
“If you do two thousand interviews, you have a margin of error of two percentage points 95% of the time”, explained XP Política analyst Vitor Scalet. 
This detail is central to reading the data. Small fluctuations between institutes or over time do not necessarily indicate a real change in the scenario, but may be an effect of the statistical margin itself.
Between photography and prediction
Another point that tends to generate noise in public perception is the expectation that polls anticipate electoral results. In practice, surveys only capture the moment they are carried out.
Continues after advertising
“It’s a snapshot of the moment,” said Pedroso.
This instantaneous nature explains why changes in the mood of the electorate, political events or campaign movements can quickly change the scenario without representing a methodological error.
Despite the limitations, research has increasing weight not only in the public debate, but also in shaping expectations in the financial market. Investors monitor data to calibrate probabilities of political and economic scenarios.
Continues after advertising
At the same time, campaigns use surveys to adjust strategy, discourse and positioning. The result is a cycle in which polls not only reflect the political environment, but also influence decisions that can, in turn, alter that environment.
The Risk Map, the policy program of the InfoMoneyairs every Friday, starting at 5am, on YouTube and your favorite podcast player.
