FACULTY Q&A

In just over two weeks, Brazilians go to the polls to elect the next president, in addition to governors, senators and members of congress. Each week, in the period leading up to the 2022 elections, several election polls are released in the country. But what do these surveys show? Are they trustworthy?

Raphael Nishimura, director of sampling operations at the Survey Research Center at the University of Michigan Institute for Social Research, answers these and many other questions.

Can the Brazilian voters believe in the results of the election polls published weekly in the months before the presidential elections?

First, it is important to emphasize that election polls portray the electorate in the period in which their data are collected; therefore, they are not and should not be used as a predictor of electoral results. That said, Brazilian voters can trust most of the election polls published in Brazil, especially those conducted by large polling organizations. These polls are carried out using scientific methods developed and improved over the last 80 years, including by several researchers from the U-M Survey Research Center at the Institute for Social Research, internationally recognized as one of the most important research centers in the world in this area.

What are the implications of different methodologies on survey results?

Different types of methodologies have different implications for survey results. For example, telephone surveys do not cover people without access to a telephone. If this part of the population is systematically distinct concerning the topic being studied from those with telephone access, then the survey estimates may be biased. Therefore, assessing how different methodological aspects of the surveys impact their results is essential.

How should the sample be selected to correctly reflect the opinion of the population of a country or state?

Ideally, in sampling, we would like all population elements to have a known and greater than zero probability of selection and that the sample is drawn randomly—properties that describe what we call probability sampling. In practice, however, this is not always possible due to certain costs and time constraints, as is usually faced by election and public opinion polls. In this case, strategies are used to minimize potential biases associated with a nonrandom sample selection. One of the main strategies adopted in election polls for this purpose is the use of sociodemographic quotas in the selection of respondents, in which the sample is recruited to have a distribution concerning sex, age, education, income and region equal to that of the population.

Do the results of these polls reflect the trends in the electorate or are they predictive?

We usually say that surveys are a snapshot of the moment. That is, they seek to reflect the population in the period in which they are collected. If there are changes in the opinion or preferences of the population at some point after this data collection period, the surveys carried out previously will not be able to capture these changes. That is why we also tend to call attention to the fact that election polls are not and should not be used as predictions. Even if a given candidate is ahead in an election poll carried out days before the election, this should not be seen as a prediction that this candidate will maintain that position in the electoral process. On the other hand, a series of election polls carried out over time can identify specific trends in the electorate, such as a possible increase or decrease in the intention to vote for a candidate.

What is the difference between straw polls and election and public opinion polls?

Election and public opinion polls follow rigorous statistical methods and data collection protocols to allow statistical inferences for the population through data collected from a population sample. On the other hand, straw polls do not use such methods and protocols. They are usually available on websites or social media for anyone to answer them, which can generate bias in their data that does not allow extrapolations of their results to the population.

What is the main difference between the U.S. and Brazilian public opinion polls?

One of the main differences is the data collection method. In the United States, there is a growing trend in polls carried out via the web or using mixed mode designs, which survey respondents by telephone, in person and by mail. In Brazil, telephone surveys—which have been utilized for over 30 years in the U.S.—began to be used more often only recently. However, there is still a predominant use of face-to-face data collection and very few web surveys.