Transcript 0:00 [upbeat music] Hello, and welcome to Explaining Brazil, the podcast by The Brazilian Report. 0:13 We are a journalistic institution that covers Brazil for foreign audiences. And today, we're going to talk about Brazil's 2026 election, which is just days away. And the presidential race is in a statistical tie. 0:29 With the first round less than two weeks away, the latest surveys put President Lula and Senator Flávio Bolsonaro within each other's margin of error, both in the first round and in a probable runoff. 0:43 From us, every new survey has made the headlines. For voters, it means a flood of numbers that often seem to contradict each other. 0:51 One pollster says Lula has ticked up, another says the president is ticking down, and a third may show stability, and so on and so forth. 1:01 Then social media turns each release into proof that pollsters showing the other side ahead are rigging the numbers. So how should we read these polls? What separates a good poll from a bad one? 1:14 And after the misses of recent elections in Brazil and elsewhere, how much weight should we even be giving polls at all? 1:24 And to help us sort everything, uh, out, I'm joined by Rafael Nishimura, director of sampling operations at the University of Michigan's Survey Research Center. 1:36 Rafael is one of the sharpest voices on survey methodology in the Brazilian public debate, and one of the most patient when it comes to explaining to people why a poll is not a forecast. 1:48 Rafael, uh, welcome to Explaining Brazil. I'm very glad to have you. Yeah, thank you. Uh, thank you for inviting me. Uh, thank you for having me here. It's a pleasure. And let's talk about polling, my favorite topic. 2:01 Yeah. And, and just before we begin, a quick word for our listeners and people watching us on YouTube. Explaining Brazil is an independent podcast, and it survives on your support. 2:12 So if you get value from this show, you can support us by going on buymeacoffee.com. And if you're watching on YouTube, uh, buymeacoffee.com/brazilianreport. 2:23 If you're watching this on YouTube, the QR code is right there in the upper right corner of your screen. 2:29 And this week, I want to thank a listener called Michel Didier for donating two cups of coffee, and this is the kind of support that keeps this microphone on. Thank you very much. 2:41 And now without further ado, uh, Rafael, let's start with the basics. When Brazilians see a new poll, they rarely know how the sausage is made. [laughs] We have multiple types of polling. 2:54 We have, uh, uh, surveyors going to people's homes. We have them finding, uh, respondents on, uh, public squares, on p- uh, places where there's, uh, a lot of people. We have online polling. Uh, we have phone polling. 3:11 Um, can you talk a little bit about the main types of poll? Uh, which are the trade-offs of each one? And why does the survey method matter so much to the results, uh, we get? Yeah, for sure. 3:26 So first, uh, to start with, like, what you, uh, were talking in the beginning that, right, many people, uh, when they see a poll result, they don't s- they, they don't really understand or know how the sausages are made [laughs]. 3:42 And I have to tell you that sometimes, I mean, in general, I, I say that, um, for polling, transparency is upmost important because, uh, it's really understanding the methods that you understand why the results of different polls, uh, are 4:02 potentially different, right? But also sometimes it's actually better that [laughs] you don't see how the sausages are made because, uh, if you actually see how polls are, are collected, um, 4:15 I think someone who hasn't been working with polls, uh, for, uh, for quite a bit would think, "Well, th- this thing cannot possibly work." 4:25 Like, I usually like to say that there is a, there's a life cycle, uh, for someone who works with polling. 4:33 Uh, the, the first stage, where is most people, where most people are, is, like, of ske- skepticism in the sense of how come a sam- a, a poll with only 1,000, right, like, respondents can extrapolate to the whole population, right? 4:48 Like, a lot of people, uh, don't really understand that. They're quite skeptical that's possible. 4:54 Then there's the second stage of poll, uh, of, of this life cycle where you learn about the methods, you learn about the statistical meth- uh, methodology and theory behind, like, what we do. And so, yeah, of course. 5:06 That's perfect. That's so beautiful. Like, it, it, it, it, it all makes sense, right? 5:12 Uh, but then the third stage, which I think is one of the most interesting, is when you actually go in practice and do the polling yourself, and you see interviewers in the field, in the streets, in the households conducting the polls, or you're listening over the phone, like, the interviewers, uh, doing phone interviewings, right? 5:32 Or even, like, online, you see, like, some of those, uh, results that see- uh, that are, they are getting and y- and you're back to the skepticism in a way [laughs]. There's no way this thing can work. 5:42 Like, how, like, all this, like, chaos and mess. It's very messy. Uh, but I, I like to think of that I'm on, on the fourth stage in a way, that, like, well, we're not really trying to get the most perfect, uh, poll. 5:56 That, that would be impossible. We're trying to, um- Get the best that we can with the resources that we have, right? Now, back to your question now, right? 6:05 Uh, yes, i- in Brazil, there are primarily three, uh, modes of data collection. 6:11 There are different methodologies that different polls use, and, like, under this umbrella of methodologies, there are, like, different aspects. 6:17 But I would say that the main aspect in terms of methodological differences is the way the, uh, data is collected. 6:25 Uh, there is, I think, the most, uh, historically, the, the one that's been used the most is face-to-face data collection, right? With an interviewer going to the field, interview, uh, people. 6:38 And even on that mode of data collection, there are a couple of different ways that, uh, it's currently done in Brazil. 6:43 You either, uh, send interviewers to these blocks, what we call, um, census sectors, in Portuguese, setores censitários, which is like, are these geographical units that, uh, IB- IBGE, IBGE, the official statistics office in Brazil, uh, creates to, 7:03 uh, when they're conducting the census, right? So, like, there's, there's kind of like city block, uh, geographical units. And so, uh, pollsters, they select a sample of those, uh, at ran- randomly in a way, right? 7:17 And then send interviewers to the field. Uh, and, uh, in some cases, they, uh, select also s- a sample of households in those blocks, in those areas. 7:28 In some other cases, some other, another methodology of face-to-face is you actually select what they call, 7:35 uh, in Portuguese, as pontos de fluxo, which are like these locations in, uh, certain, in a sample of cities that there are a lot of, like, people going around, w- walking around, right? 7:48 And you ask for the interviewers to stop people in the street to, to answer a survey, right? Like, there's, there's two different mode, e- even, like, methods in face-to-face data collection. 7:57 Then there's cell phone sampling, which, like, I, I'm sure that, like, a lot of the audience here is familiar because that's the-- It used to be, at least in the US, the primarily mode of data collection of polls, uh, in the US, right? 8:07 Like, where you select a random sample of, uh, telephone numbers in a way, and you call people to, uh, answer a survey, right? 8:15 And I'm sure that there's a lot of people here, uh, that's already thinking, "Well, I never answer my phone to, about, uh, from nu- numbers that I don't know, uh, th- that, an income, uh, a number, an income call from a number that I don't know," right? 8:28 We'll talk a little bit about that. Uh, and then the third one, the more recent, right? And by the, by the way, I should say telephone surveys in Brazil, they're not as old as it is in the US, right? 8:37 In the US, telephone surveys- Yeah... they've been conducted since the '70s, '80s. In Brazil, it really only started getting adopted in the mid to 2010s. Re- Like, I mean, I, I, I used to work in Brazil. 8:51 Like, my, the, the, my early career was in Brazil, uh, back in the mid-20- 2000s up to 2010. I was conducting some telephone sampling, uh, telephone surveys, but not a lot really. 9:04 It was really, like, in the mid-2010s, late 2010s that, that started really. 9:08 And, and if I may, uh, just for a piece of context, that's also because until the 1990s when, uh, we privatized telephone services in Brazil, actually owning a telephone landline was, uh, incredibly expensive for most people. 9:25 It was so expensive that you would declare in your income taxes, like, "I own a house, two cars, and three telephone lines," and people would rent their telephone line. So, uh- Yeah... 9:36 just, uh, to explain why this kind of, uh, sampling, this kind of polling is so recent in Brazil. Yeah. Phone penetration really, like, uh, pi- spiked, uh, with cell phones, right? 9:49 Like, so it was really, like, in the 2010s that, like, phone penetration started getting to anywhere near, like, 80 to 90%, where it start, it started to make sense to do a telephone survey. 10:00 Uh, then more recently, right, uh, th- there's, there are, there are some pollsters also doing, um, web surveys, right? So c- conducting... Which is, like, almost, like, now the norm in the US, uh, for post polling. Now, 10:16 there are pros and cons in each approach, right? Um, I, myself think of, like, face-to-face data collection, uh, kind of like a standard, a, a gold standard in a way, uh, because it has very good properties. 10:32 It can cover the population really well, right? Like, you can, you can reach essentially or theoretically anyone in the population. Now, whether you actually reach or not, that's another question. 10:42 I'll, I'll mention about that in a second. But has good properties. Also, having an interviewing, the, an interviewer there to try to, uh, convince people to answer a survey is very helpful as well. 10:53 Because if you just, like, let people, um... 10:57 If you, if you just select people who are willing to take the survey, like, promptly n- with no trying to, you know, like, um, convince them, uh, you would get, like, a different type of respondent, a respondent that is more interested in politics, more interested in, like, uh, surveys, that sort of thing. 11:15 Uh, but there's some cons about face-to-face. Like, uh, I said theoretically you can, uh, reach everyone, but in practice, right, like, it's very hard to, uh, to reach people who live in, 11:28 in, um, buildings with limited access, right? Uh, so buildings that, like, you have to, uh, go through a gatekeeper, right, like, to, to get access to people inside or, uh, closed communities. 11:42 Another pl- another, uh, issue is, um, certain communities, uh, dominated by, uh, the traffic in Brazil. That in some places this is, like, a, a real problem, right, in favelas as, or slums, right? Um, so there's that. 11:57 Uh, telephone survey on the other hand, uh, can reach out better this type of population, but as I was saying- 12:04 Not a lot of people are answering the phone nowadays [laughs] from numbers, numbers that they don't recognize, recognize, right? So, um, it's- a- and it's really inefficient in a way because of that. 12:16 Just so we have an idea, in order to get a, a one completed interview, you really need to call 100, 200 telephone numbers. Wow. Yeah. 12:25 It's, it's really, it's quite inefficient, uh, to get, like, a, a phone, uh, su- survey. And that's true not only in Brazil. It's also true in US, also true in, like, many, uh, other places in the world. 12:38 Sorry, did you wanna- Yeah, and these tend to be also long interviews, right? Because you have to l- uh, stay, uh, in the line and listen to all the options that, uh, uh- Oh... the recording's giving you, so. G- g- good. 12:51 Thank you for reminding me of that. So there are also, in telephone, just as in s- in, in face-to-face, there are two modes, right? Like, two kind of methodologies. In telephones also there are two methodologies. 13:01 One is the one that you are, uh, referring to. It's what we call IVR, interactive voice, um... Recording? Yeah. No, I always, I always forget what the R is, but it's something recording or response. 13:14 I think it's response. 13:15 Um, where you have, right, like, um, uh, a recording, uh, or a bot kind of like a- asking the questions and then, like, they, yes, they list the options, and for each option they, they put a number and then you, uh, respond, you answer with the number, right? 13:30 Like, that you c- you click the number that corresponds to your answer. But, uh, that's not very often, uh, used. There are s- a couple of pollsters in Brazil that use that. That's not the primary mode, I would say. 13:41 Like, most, most of the telephone surveys in Brazil are done with interviewers, actually human beings, like, calling and, like, talking to people, right? 13:51 Uh, which again is helpful because i- in, in the case of someone actually answering the phone [laughs] the, the interviewer can, you know, like, talk to the person, try to explain what it is, uh, uh, convince them to answer the survey, right? 14:05 That's, uh, uh, that's an advantage. 14:08 Uh, disad- disadvantage is that, uh, not everyone answer the phone, and, like, if you, if the person's not answering the phone, there's nothing really you can do there to, uh, convince them to answer the survey, right? 14:19 Uh, finally, the web survey, um, it doesn't suffer this problem necessarily [laughs] of, like, uh, trying to convince people to answer the survey, but it suffers of another, another problem. 14:30 So many of the survey, many of the polls in Brazil using web, they are recruiting their sample by posting, uh, ads in social media, right? 14:41 So you, if you go to Instagram, uh, in Brazil, you're gonna see, like, some ads saying, "Hey," in Bra- in the US as well you can see that, uh, "Hey," this is, um, an ad of, like, answer this poll or answer this poll about, like, presidential elections, et cetera. 14:54 The problem there is that this kind of methodology suffers of something we call self-selection. 14:59 It's really only the people who are interested in, like, answering a survey or interested in that, like, particularly, particular topic about the survey that they're gonna go there, click on the ad, be redirected to a poll or to a survey, and they're gonna answer the survey. 15:14 What we know from methodo- from, um, from methodological research, uh, from the literature is that usually those people are more politically engaged. They, uh, have a higher sense of civic duty, for example. 15:29 They are demographically, socially and, and demographically different somehow from those that do not click on the ad, answer the ad. 15:38 But more importantly, we also have some, uh, reason to believe that they are also politically different from, uh, people who don't answer the polls in the web. Why do you k- why do you know that? 15:48 Well, if you look at, for example, the voting intention, right, like the v- vote choice question, uh, of face-to-face and telephone surveys, there is always at least about 5 to 10% of people who are still undecided who they're going to, who are they going to vote. 16:05 They don't know. Because those are people who are not really thinking about the election. They are not really politically engaged. They don't... 16:11 They're gonna decide their vote the day before the election, the day of the election, right? If you go to the o- online poll, online polls, the web surveys, that number of undecided voters goes down to zero. 16:22 It's like if it's 1%, that's too, too, too m- uh, too high already, right? Yeah. 16:27 Because those answering the poll, polls are those that are, uh, engaged poli- politically, uh, those who really kn- know about politics, uh, what you call the polit- um, uh, what's the expression you use in English? 16:43 Is the... I forgot, but, but anyway. Yeah. They- Those that- They are deeply engaged voters and, uh- Yeah... 16:50 I mean, we, we have seen, uh, Renan Santos, a far-right candidate here in Brazil, and his supporters, uh, uh, teaching other supporters how to game social media algorithm. 17:02 Like, "Oh, uh, uh, you have to engage with the pollster's, uh, Instagram account so you- Yeah... you're going to get more ads from them, and then it's going to boost our voting intentions- Yeah... 17:13 when the results come," so. Exactly. 17:15 And the social media algorithms, right, like in Instagram, for example, they, they don't cop- cooperate with, uh, polls because, like, their algorithm is not made to select people for polls. 17:25 They're made to maximize the click rate. So the algorithm is going to deliver also those ads exactly for the people who are more likely to click on those ads [laughs] and answer the poll, right? 17:35 Which again can be systematically different from those who don't, uh, click on the ads and answer the polls, right? Spe- specifically politically, right, like, they can be different. 17:47 Now, uh, uh, Rafael, you say, and you often say this, that polls are not crystal balls. And I, I think just yesterday you published a, a new thread on X, uh, explaining that. 18:00 And that's something that pollsters themselves say sometimes, and they have said that especially over the past few years after big misses. 18:08 Now the pushback that I would have, uh, is that they only say that when they miss the results. 18:13 Because when they get it right- They are the first ones to use the proximity between their results and the actual, uh, voting numbers as proof that they are incredible and that, uh, they are reliable. 18:29 So I'm not criticizing you for saying it, but in pollsters case, it seems like they want to have their cake and eat it. [laughs] Yeah. Two different standards, right? Like, for, uh, for that. 18:41 Um, I will say that I think especially given the most recent, um, developments in Brazil, like not only, like recent as of, you know, one- a month or two months ago with the, 18:57 uh, S- uh, with the TSE, right? Like the, um, how do you- how- how would you- The Superior Electoral Court Yes. Thank you. Um, putting a proposal to, uh, create a seal of accuracy, uh, a polling accuracy, right? 19:12 Like, but also going back to 2022, where there were- when there were some, uh, congressmen, uh, proposing, um, uh, some, um, some, what's the word? Uh, projeto de lei, right? Some restrictions. Yeah. A bill. 19:30 Yeah. Bill. A bill. Thank you. Yes. A bill. 19:32 Proposing a bill to, um, penalize, uh, pollsters for- if they didn't get the result right, to the point of like even putting in prison s- uh, the pe- the- the person responsible for, uh, the, the, the, the poll, right? 19:51 Uh, and so because of those recent developments, I think that at least a lot of them pollsters are- have been a little bit more careful in, uh, in presenting that, that like polls are not predictive of the elections. 20:07 They are like a snapshot in time. 20:10 Uh, and I hope that even if, right, like some of those pollsters now coming, uh, to election day, the first round at least, even if they are relatively close to the election results, I hope that they don't use that as a, uh, as a way to promote that. 20:28 Because I agree with you. If it's- it, it, it goes both ways, right? Like polls, they are snapshot of the electorate. Like, uh, and that's true whether they get- they, they are close to election results. 20:42 That is also true if they are not close to election results. I, I, I, I quite, I, I, um, absolutely agree, and it's, it's something that really 20:51 I have been trying to push back a little bit fr- uh, to the pollsters that say, "Hey, you shouldn't also say that if you, if you get-" Yeah, no. And it's-... And it's, and it is the same thing. 21:02 I remember going to a debate with, uh, three major pollsters, and this was pre-pandemic, and, uh, two legacy ones and, uh, a, a more modern one, which did online polling and phone polling. 21:17 And the legacy ones were dismissing that. "Oh, this is just like when you put on the home si- the, the web, um, the homepage of a website," like, "Do you like, uh, strawberries or not?" Mm-hmm. 21:29 This is as scientific as that. And then the pandemic came. Everybody started doing phone and online polling, and nobody was batting an eye. 21:38 Uh, so it, it, it, that, that thing that g- uh, wanting to have it both ways, it- that's what bothers me. But, uh- Me too... my theory, though, a- a- about polls is not their predictive power. 21:51 And this is something we always tell our subscribers, is what I find it interesting is how they shape political behaviors. Because parties change their strategies, candidates change their strategies based on the polling. 22:04 Some even drop out. Uh, do you think this is a fair assessment? 22:08 I think the, the power of polling is not, uh, uh, how they assess voter mood, but actually how they end up being part of, uh, how political decision-making is done. Mm-hmm. Yeah. 22:23 I think there are two different perspectives from there, right? I think that like, yes, for campaigns, uh, polls are very, are very po- are a very powerful tool, 22:34 uh, for them to gear what direction they should go with their strategies. Uh, and I think as, as a to- a completely le- legitimate use of the, uh, of, of polling, right? Uh, 22:48 but have to keep in mind that mo- a lot of the campaigns, they have their own internal polls, um, not only to have their numbers because they don't- I mean, there's- Which cannot le- legally be, uh, published, right? 23:00 Exactly. These daily polls, uh, they, they are barred from publication. Yeah. That's what they call trackings, right? Like, they are like, uh, daily doing, uh, conducting polls. 23:09 But, uh, it's not only for them to, you know, like, have an idea of what direction things are going. 23:15 And, and it's not- A lot of people say, "Oh, no, the, the, the internal polls, those are the ones that are really showing the real numbers," right? Like, no. In, in general, they should be showing something similar. 23:27 The thing about internal, uh, these trackings is that they tend to be q- very quick. 23:31 And because they're very quick, they also tends to be, uh, with a smaller sample size, uh, also lower quality, uh, because they have to do, like, something really quick compared to, like, the, the polls that are published in the media, right? 23:45 Um, but apart from that, they use those polls to also, uh, ask other questions, uh, to gear their strategies. 23:54 So there are questions that are ask- they're asking in those polls, polls that are not asked in the public polls. 24:00 Because, like, the, the campaign wants to know how the voters are thinking about certain things for them to, uh, make some decisions about advertisements, uh, or, or strategies, right? 24:12 So, um, I think there's that per- that, that perspective. I like to also think about from the voter perspective, uh, right? Like, I think for the voters, really important to understand what- 24:25 Their pe- their peers are thinking, right? Uh, not necessarily to- A-a-and do you think, do you think the bandwagon effect is, uh, uh, a real thing? And, uh, how, how, how big is it? Yeah. So that's a good question. 24:40 I, I don't think we have very good research about that in Brazil to say whether there is or there is not, or what is the magnitude if there is. 24:49 There is some research scattered, uh, around the world about that with mixed results. Like, some points out that there is some bandwagon, uh, effect and it can be, like, moderate. 25:02 Some that points out that, no, there isn't, like, really a-any bandwagon for that due to the, due to polls, right? 25:09 Now, regardless of whether there is or there is not, I think that it's important for the voters to understand. 25:16 Because o-otherwise, especially on this environment that, where we live in, right, like these online bubbles that we are, um, it, it's, 25:27 it's- it feels to you that, like, if you don't have access to the polls, that, like, you are- It's a lot of people, especially those who are, like, very, um, very online, [laughs] right? 25:39 It feels that, like, uh, you are in a different world from, like, what actually the, the, the, the nation, uh, is going, right? Like, depending on- Right... what kind of bubbles are. 25:48 So, like, I think it's good for, uh, to have, for the voters, public polls so that, like, they have an understanding how the nation is thinking about elections. Um, 26:02 but also, right, like, I, again, related to the ba-bandwagon effect, right? 26:08 Like, I don't think, uh, myself personally, that's my own opinion, I don't think people should decide their vote based on the results of the polls. But some people might want to do that. They have the right to do that. 26:22 They, you know. So, uh, presenting that with the best method that we have, I think that's also, uh, a service, uh, to those people. And just, like, a, a small note about that. Um, 26:36 the way Congress, the, the congressman who was proposing that bill in 2022 justify it, uh, justify the bill was that, oh, voters are using polls to make their decisions about the- Mm-hmm... uh, about their vote, right? 26:54 Well, if, if that's- Let's assume that is true. We don't know. We don't have evidence that it's true. But if that's really true, it's impossible for a poll 27:02 to, uh, predict the election because the voters are using the r- the, the poll, the results of the polls to change their votes on the, on the election, on the election day. 27:14 So the, the things don't- It, it doesn't add up, right? Like, you cannot- Mm-hmm Like, uh, the, the result of the poll is changing, assuming that, that's true, is changing the, the way voters behave, right? 27:29 Now, is there, um, is Brazil- Does Brazil pose specific challenges to pollsters? I'm talking about inequality. 27:40 Uh, uh, you mentioned phone penetration was a thing that, uh, uh, prevented phone polling from, uh, being, uh, uh, uh, uh, really important, uh, for decades. Uh, the s- the sheer size of Brazil. 27:54 It is expensive to, to do a, an in-person polling. Uh, a-as opposed to other countries, uh, i-is Brazil s-specifically hard to poll? 28:09 Um, not s- I, I'm not sure if it's particularly hard to poll. Like, I mean, it, it does, it does have all those challenges you're talking about, right? 28:18 Especially for in-person interviewing, face-to-face interviewing, like the, the, the continental size, uh, of the country. 28:25 That's, uh, a really- Es- And especially because, like, in Brazil, there are some areas where, like, you- it's really difficult to access, right? Like the Amazon, for example, right? 28:35 Like, there are some communities that if you're selecting a sample randomly and you select, like, a community, like, it takes, like, two through three, two through, two through, two to three days just to reach out to, uh, that community. 28:45 So, um, I- I'd say that, yes, there are some challenges but, um, there are a couple of things that makes, I think, uh- I mean, I like to compare a lot with the US, right? 28:59 Like, I'm here in the US, so what are the differences here? So one is that, um, it's this kind of, like, almost this connection between vote in Brazil is mandatory, 29:12 but at the same time, effectively it's kind of not, right? Like, because, um, the, the, the cost of not voting is, for a lot of people, is not really that high. 29:25 And so historically we've been having, over the last few elections, uh, a turnout rate of 80%, right? Like, 20% doesn't show up. Uh, which is, for a country where- And it's going up. 29:36 That, that, that, uh, share of people who skip going to the polls entirely is going up and up, uh, every new election cycle, right? Yeah. I mean, it's, it's not a lot. Uh, and it changes a little bit, right? 29:50 We had, uh, it- it's just like 2020 was a we- weird year. No, not '20, uh, 2022 was a weird year 'cause, like, kind of, like, pandemic, uh, almost kind of during or after the pandemic a little bit. 30:03 But yeah, it's, it's, uh, compared to way before, yes, it's, it, it's going up a little bit. Um, but, 30:12 and, and like, so it makes it, I think, challenging in the sense of for pollsters because in the US it's very clear, right? Not mandatory. So you need to somehow account for that. 30:24 The way pollsters account for that is through, like, what they call likely voter models, kind of like algorithms or statistical models that tries to account for the fact that, uh, some part of the electorate is not going to vote. 30:36 In Brazil, there is this kind of, like, weird thing where in theory it's mandatory but- A lot of people don't. And because of that, a lot of pollsters 30:45 kind of assume that most people or all of people, all, all the voters will show up. That's not the case. Um, and, uh, let me think. Um, 30:58 yeah, I, I, I think that, like, the online is still, I wouldn't say a problem or challenge, it's just, like, the way it's implemented is very different from the way in the US is implemented, and I think we have to overcome a little bit of that, uh, find better ways to poll online in Brazil. 31:15 Now, uh, one thing that we, uh, have seen in over the past few years is pollsters saying that public opinion has become extremely volatile because of social media, because of how we consume information- Mm-hmm... 31:31 and that, uh, one poll may be 100% accurate, but two days later the picture changes entirely. Uh, does the evidence back that up, or is it just an excuse for misses? [laughs] So, um, 31:50 I, I, I don't think that throughout the campaign we see, or at least the last election or two, this, this election at least, and even the previous. I don't think we saw... 32:04 Uh, in 20- 2018 we, we did see a little bit of that, uh, especially with the a, a life attempt against Bolsonaro. Like, we, we saw, like, some effects there. 32:13 But, like, uh, during the campaign we, uh, we haven't really seen a lot of, like, um, movements, like, given what's happening in the- Yeah, but, but for instance, in 2022, we saw, uh, 32:26 renowned pollsters missing Bolsonaro's, uh, figures by nine points, which is- Yeah, so and that's what I'm saying during the campaign, right? 32:36 Like, which is, like, different from what's happens then in during election day. Um, why? Because as I was saying before, there's a good part of the voters in Brazil that they make their minds. 32:51 They make- they decide who's they're gon- who they're going to vote, uh, the day- the night before or even the day of the election. 32:57 Even for presidential election, which is, like, one that, like, uh, people are already know who they're gonna vote more in advance, right? 33:05 So there was a poll, uh, there was a study that estimated that about 10% of the voters in Brazil decide who they're gonna vote for president in the day of the election. 33:17 And another 10% who says that they decide who they're gonna vote for president, uh, the days before, right? Like, one, two, three days before the election, right? 33:29 Now, if that's true, right, if th- if those numbers hold, that's about 20% of the electorate making their mind on the last few days of the election. And why this is important is because 33:43 polls, right, as I, as I always say [laughs] snapshot in time, right? 33:47 So one th- one thing that people don't really, I think, realize a lot is that in Brazil at least, but, like, that's also true in other places, in the US as well. Uh, if a poll is published the night before the election, 34:01 it doesn't mean that it was conducted the night before of the election. The poll started at a collection one, two, maybe three days, depending on the mode of the collection, uh, before it was published. 34:13 Which means that, uh, it's still, uh, capturing vote intentions from people that may or may not change their vote on election day or before that day. Yeah, it's always a moving target, right? Yeah. 34:28 So that's why I think it's not really fair. I would, I, I, I usually say don't compare election polls with election results because, like, it's, the, the, it's not, it's not a, a direct apple to apple... 34:43 Sorry, yeah, apple to apple comparison. I like to make, uh, um, uh, I have a metaphor that say, "Well, you're not comparing, um, apples to apples," right? 34:55 Or, yeah, what, what you're doing, you're kind of comparing apples to apples in a way, but you're comparing a ripe apple with a green apple, right? 35:04 Like, in, in the sense like it- it's the same population, but, like, the states where they are are, are different, right? Like, in, in, in their, uh, opinion or in their voting intention in this case. So, um, 35:17 there can be some differences. Now, it doesn't mean that all the differences that you find between a poll and, uh, the elections is due to this change in the electorate. Uh, there can be some differences. 35:30 I, I don't want to say that pollsters, they don't have any role in those differences. There can be also differences due to meth- methodological, uh, problems in the polls. I'm not saying that they're reason. 35:43 I'm just saying that, like, comparing directly, there's a part of, uh, that difference that is not due to the pollster's fault, but it's just due because, uh, uh, voters are making their mind, voters are changing their mind, especially in Brazil at least, that, uh, a lot of the electorate, like, like 10, 20% are, are, are deciding who they're gonna vote on the day or the days before the election. 36:07 Now, we've started this conversation saying that, uh, uh, different, different methods, uh, may yield different results, and, uh, there are so many variables that can factor in. 36:20 Whether it's, uh, if we're talking about in-person, uh, interviews, it's a female person doing the interview or a male person. Uh, uh, like, many things can change the results one way or the another. 36:36 And still, despite the vastly different methods, uh, pretty much all polls are bringing up the same results Um, how do you explain that? 36:47 Do we have a-- And, and I mean, in 2024 in the US, uh, they mentioned the, uh, uh, the possibility of herding, which is, uh, uh, it was the pollster's fault of picking the same kinds of people rather than accurately, um, 37:03 uh, uh, uh, giving a snapshot of the population. Like, how, how, how do you explain so many different polls bringing up essentially stable results for months? So okay. Uh, let me start with the herding, uh, part, right? 37:20 Like, uh, so for those who don't know, what, what, what, what is herding in polling, right? Like, so herding in polling is actually when a pollster 37:30 or a group of pollsters, right, like I'm thinking about a pollster, uh, it somehow, and there are, like, there are legitimate and illegitimate ways to do that, but somehow, uh, alter its polling results to be closer to the average of the polls, right? 37:46 So you have polls being conducted, right, like every day. 37:50 So you-- And, and nowadays, like it's, this is even easier because there are, like polling aggregators that, uh, they compute already what's the av-average of the polls. 38:00 And let's say a pol- a pollster, uh, for one reason or another, like got a, a, a, a, a result, a poll that, like deviated a lot from that average, right? 38:11 So herding would be the pollster, again, using some sort of strategy. And as I said, like the different strategies, there are legitimate ast- strategies like, uh, statistical adjustments. 38:22 There are statistical adjustments you can make to, uh, make the results go a little bit like more in one or another direction, right? Certain decisions are made. Or like really, 38:32 uh, illeg- illegitimate methods of like really like somehow putting your thumb on the scale and like changing the number, right? Uh, which I definitely-- I, I mean, I, I don't-- I, I wouldn't... 38:43 Uh, herding is a bad practice either way, right? But, uh, so that's what's herding, right? 38:49 Like a pollster kind of like changing its results in a way to be closer to the average so that- of the polls, so that like it doesn't, um, if it misses, right? 39:00 Like [chuckles] it, it, it's not the one who's missing in a way, right? Yeah. Uh, and in 2024 here, here in US, I think there's a little bit of a myth that there was, there was some herding. 39:11 And I, I, I have some discussions with pollsters about that. Uh, I th- and I think it all started in a way with Nate Silver, uh, raising that concern. 39:21 He said, "Oh, look-looks like there's some herding happening here," because like all these polls are very close to what you assume if they were independent polls, like, uh, being conducted independently at random, et cetera. 39:34 Well, that's not particularly true. I mean, that would be particularly true if that was the case in a way, if they were all like doing just like random, um, uh, samples and without any sort of like adjustment in a way. 39:48 But like i-in the US, a, a lot of the pollsters use, uh, re- heavily rely on something called weighting, right? 39:56 Like, which is you, you take your sample, and depending on the distribution of your sample by gender, age, education, uh, region, et cetera, you compute weights for the respondents so that like that sample 40:10 reflect the population with respect to those characteristics, right? And, uh, pollsters in the US uses a lot of weighting. And, 40:19 uh, lately, they also have been including on their weighting scheme something we call recall vote weighting, which is like you ask for the response, "Who you voted for in the last election?" 40:30 So in, in the, in 2024, "Who you voted in 2020?" Right? And they make sure that the sample in 2024 match what was the result in 2020, uh, according to the answer that they give. 40:44 And that's actually a very powerful, uh, tool because we know that vote between elections is highly correlated, right? So if you're going, if you're getting a good, uh, a, a good, um, 40:56 portrait of the electorate, uh, according to 2020, chances are that, uh, the results you're gonna get for 2024 in that case would be high quality. 41:06 And what is happening there is that because a lot of pollster- pollsters are doing that, um, the results, the, the, the, the variability of the estimates of the polls tends to shrink as well. 41:17 And so it looks like it's herding, but it's not really herding. It's just like a byproduct of all these adjustments that they are making their sample, right? Now to Brazil. Is there evidence that it's happening? 41:28 So in Brazil, it's a little bit more complicated because in Brazil, for some reason that I don't quite understand, uh, a lot of pollsters, they are very resistant in using this sort of like, uh, adjustments, this kind of what I call weighting, right? 41:43 Um, they usually prefer to use something else called quotas, which is like when they go to the field, they say, "Oh, if the population is 50 males, 50, uh, 50% males, 50% females, let's make sure that in our sample we have 50% males, 50. 41:56 50% females by like controlling that." So they keep like track. They keep a tally of males, females while they're collecting. 42:03 And when they say, "Oh, we are gonna- are planning to do 1,000 interviews," w- as soon as they got 500 males, okay, any, uh, surveys that like, uh, uh, come in with males- No more males. Yes. 42:15 We just go for the females, right? So in a way, that- that's, that's sort of similar to what weighting is doing, but like on the upfront. Uh, but there's things that weighting can do that, uh, quotas cannot do. 42:29 So, um, I still don't quite understand why pollsters in Brazil are resistant in, uh, using, uh, um, using more of the statistical tool that like is available for them to, uh, refine their estimates. Uh, and so 42:42 it's a little bit surprising to me that we see results as close as they are. But, uh- When we look into the, 42:52 um, into, into more, more details of like the poll, not only of, uh, vote choice, but some other characteristics like, uh, approval, uh, presidential approval, right? Uh, or, um, how, uh, 43:10 how, uh, familiarity of the candidates, right? 43:14 Like, uh, voters familiarity with each candidate or what's the support that they give, like in terms of like whether they would, uh, ever consider, uh, voting for them or not. 43:22 When you look into that, like the results tends to not be as close, as similar as, uh, as, as they are amongst each o- among each other. So I don't know, maybe they are... 43:34 And, and to be fair, I, I don't think that they are also that close. Uh, I think we have a little bit more variability 43:42 in the vote choice, uh, estimates, uh, across polls than we used to, than we had like in the US in 2024. Um, the, the, the thing is that, 43:54 uh, we have also f- much fewer, fewer polls, pollsters in Brazil than we have in the US, right? Mm-hmm. Like, in the US like we really have like literally, um, dozens of pollsters, like dozens. 44:10 Like, uh, and like can get up to, depending on what they're looking at, I don't know. I never made, I never did a, I've never counted, but can be like more than like be- something between 50 to 100, uh, pollsters. Oh. 44:23 In, in, in Brazil, yeah, well, but that's because a, a byproduct also like how elections are in, in the US, right? Like which, um, 44:31 statewide, uh, races like, uh, polls are i- as important if, if not even more important than national polls. Yeah. So there's also a lot of small pollsters, uh, on different states, right? 44:44 In Brazil, we don't have much of that. Um, but yeah, in Brazil, we, we, we don't have more than like a dozen, like large, like poll- pollsters. Like maybe there are some minor ones, like, but it's not really that many. 44:57 No, and by the way, uh, talking about, uh, the difference between national polls and statewide polls, um, one thing that has become a real trend in Brazil is when we try to slice, uh, the poll and look And, and there, there was even a video, a humoristic video that- Yeah... 45:16 ca- became viral in Brazil a couple of weeks ago. Like, okay, uh, the, uh, the president's approval has gone up between females called Stephanie that are, have green eyes and are blonde, et cetera. Yeah. 45:29 And I mean, you are a vocal critic of crosstab diving. Yes. 45:34 Uh, which is, so when people dig into a poll's subgroups, uh, like evangelicals in the Northeast, young women who are beneficiaries of Bolsa Família in São Paulo, and they treat those numbers as findings. 45:49 Uh, why, in your opinion, is that so problematic? And, uh, how do you assess the way that the media is, uh, treating that? And that is important because the media are the biggest sponsors, uh, of polls. 46:06 So they want to milk the, the news value of, uh, of a poll as much as they can, right? Until the get the last drop of, o- o- of news cycle engine- uh, uh, generating headline they can. Right. Yeah. So, 46:26 and this is true, I should say. Not only Brazil, in the US as well, like there's a lot of, uh, crosstab diving, uh, going on, especially the social media, right? 46:34 Like in the, in the Twitter sphere, sphere, Blue Sky sphere, like there's a lot of like, of that going on. 46:41 So the reason why, uh, some people who works with like, uh, right, like polling, uh, and survey methodology, we say, "Don't go cross dive- cros- crosstab diving," is because, well, there's a couple of reasons. 46:54 First, um, as, because the, the sub- So crosstab diving is thing, or crosstab, uh, estimates essentially are subgroup estimates, right? 47:03 Like, so you are analyzing the results br- uh, broken by, uh, gender, by age, education, region, et cetera, right? Now, the problem is that like, uh, the margin of error that are reported in the surveys, right? 47:15 Usu- usually, right, like, uh, one, 2% margin of error. Uh, for those, uh, subgroup estimates, they're larger because the sample size in those groups are also smaller, right? 47:27 So that's the first thing that you need to account for. That like depending on the size of that group, that, that the margin of error could be, uh, substantially larger than one, 1, 2%. 47:38 Could go up to like, uh, five, six, eight, 10%, right? Depending on the size of the group. 47:44 So I, I usually say be careful with that because maybe some differences that you are, uh, you're comparing groups and some differences you're finding are not really like, uh, significance, uh, statistically significant differences. 47:59 They're just like noise, right? Bec- this- Mm-hmm... it's all, it's all within the margin of error, right? 48:03 So what, what, what, what annoys me a little bit is that sometimes there's some headlines saying, "Oh look, this candidate, Lula or, or Flávio," like they, uh, on this group, they, uh, had like a, a boost, uh, uh, compared to the last round of, uh, poll, right? 48:21 Well, actually, when you look into like, if you account for the margin of error, not really. [laughs] Like it's all, it's all statist- it's all like random fluctuation. Uh, or can be considered random fluctuation. 48:32 You can't really distinguish a random fluctuation from an, an actual increase of, uh, support, for example, right? 48:38 But the other thing that is not, um, talked as much, and I'm trying to also raise, uh, the point there, is that when a poll is, is, is done, right, like as I said- 48:51 You can, you can adjust the- you can control, uh, the, the, the res- uh, the poll, uh, controlling the sense of, like, having a, a good, like, quality poll by either using those quotas or- Mm-hmm... 49:04 by using weights, right? Weighting. Or both. I- ideally, I'd like, I'd like to boost both, right? Like, not a lot of them use both. Um, however, uh, this, uh, 49:14 the way you're controlling the sample to get a good, uh, estimate there is done primarily thinking on the national level, right? 49:22 You- so you're adjusting, you're, you're, you're adjusting, you're controlling your sample at, at the national level to get good estimates at the national level. 49:30 Typically, pollsters are not, like, doing these adjustments, the statistical, right, like, adjustments, et cetera, uh, down to each one of the subgroups now. 49:39 Which means also the quality of those subgroup estimates are not the- sorry, I should say this, those subgroup estimates are not calibrated to as well as the national sample. So what do I mean by that? 49:50 I'm, I'm gonna give an example, right? Yeah. It's like- And that's true... a state has 60% male and 40% female. It's not gonna necessarily do the quotas for those specific like, oh, 23% are evangelical women, et cetera. 50:07 So i-it's so, uh, uh, there's so many- Exactly... different groups that it gets hard, right? Yeah. Yeah. And like, um, just as an example, I was just listening yesterday, right? 50:18 Uh, there was a poll in which, like, in the southeast of Brazil, uh, there was, like, compared to the previous poll, uh, Lula, I think, uh, it, uh, increased, uh, I don't know, 10, 12%, right? 50:32 Like, how, how, how, how come, uh, that happened? Well, I mean, not only there's the margin of error, but in this case, only the margin of error doesn't explain. But also, right, like, the poll is not calibrated 50:45 well enough for southeast respondents there, for the southeast voters, voters. 50:51 Uh, such that, like, certain swings, large swings can happen just because, like, the, the sample is not calibrated well enough for those- Mm-hmm... subgroups, right? 51:02 This is something that, like, uh, American, uh, uh, political scientists and, um, pollster Doug Rivers. Some people who follow, like, uh, polls in US might have heard of his name because he works at YouGov. 51:17 Uh, he calls it the phantom swing, right? Like, where, uh, you see large swings in polls because the sample is not well-calibrated, uh, to, uh, to avoid th- these kind of things. 51:31 Now, uh, before I let you go, I have two questions, and the first one, uh, is, uh, with all of what that, that we have been talking for the past 15 minutes, um, how do you rate electoral polling in Brazil? 51:47 Do you think we are in a good state of polling? Has it gotten better? Has it gotten worse? Uh, and like- Mm... 51:55 what are- what, what would, would you recommend political wonks watching this and listening to this podcast that they should bear in mind when looking at a poll? Mm-hmm. I, I would say that, um, 52:09 I s- I would give- I have mixed feelings about the state of the, the polls in Brazil. Like, on one hand, I think that, uh, they are in a good quality. I, I don't think that, like, they are doing a bad job. 52:19 I, I think that they are doing, like, a very competent- at least the, the, the largest players, right? 52:24 I'm not gonna talk about, like, like, a pollster that, like, released one poll in the entire cycle in the, in, uh, in a state in Brazil, right? 52:32 Like, so the largest, uh, players in this industry, I think that they're doing, like... 52:36 And I know, like, a lot of them, people who work, uh, in those, um, in those, uh, polls, uh, polls, uh, in those, uh, polling organizations. Think they're doing a, a decent job. 52:46 Uh, but the part that, like, I feel a little bit like, uh, is I, I feel also that they are a little bit, uh, many of them, not all of them, but many of them, are a little bit, like, stuck in time, right? 52:58 Like, they've been doing kind of like the same thing over the past, like, several years. Like, not just say, like, uh, 10, like, over 10, like, 20 years, right? 53:09 Uh, and I, I don't see a lot of, um, innovations, uh, happening. I mean, there's one or another. There's, they are trying innovations, and I praise them for that. 53:19 But I would like to see more innovation, especially because, as you're mentioning, right, like, society's changing, right? 53:25 Like, the, the whole, like, uh, thing of being, uh, a lot of, uh, being online a lot, like, change, uh, the, how people, how people view the world, like, how people also interact with each other. So 53:38 I, I'd like to see a little bit of more innovations, uh, on- Mm... um, from the part of the pollsters to adapt, to, to make sure that, like, they don't really, um, 53:48 they don't get into a problem that we observe, like, uh, actually here in the US [laughs] uh, in the 2022. 53:55 Like, maybe some of, some, some of the listeners, some of the viewers here might remember, uh, the case of Iowa in, uh, 2022, where, like, uh, this very, um, it's almost legendary pollster, uh, Ann Selzer, who having, uh, cycle after cycle, right, have been, like, getting very good results, like, for, I don't know, five or s- four or five election cycles. 54:19 Then, uh, happens, like, that she gets, like, a terrible, like, performance in 2022. And the problem there was because she kept using the exact same method since, like, early 2000s. And, like- Mm-hmm... 54:33 she didn't adapt her methods to account for this new political environment. And like, uh, honestly, it took too long for her to, like, to, to get it. 54:41 I think Iowa was a little behind in, like, the, the way how things were moving, uh, to, to catch up to the rest of the country, right? 54:49 So that's, that's the kind of thing that, like, I would like pollsters to, you know, in Brazil, to make sure that they're adapting their methods to catch up with, like, how society is changing. 55:00 So you are in the university, so let's try, uh, uh, I w- I'm gonna ask you for a grade. From one to 10, uh, what are the grade of Brazilian polling, uh, that you, you would give? I would give, um... 55:17 I would give it, like, a, a eight. I think, like, they're doing, they're doing- Okay... they're doing a good job. It's a, it's a good job. Uh, but there's still, there's still room for improvement. 55:26 I think there's still, uh, some room for improvement there for, for sure. Now, uh, one last question is, um, something that has become seen as a sort of rival of, uh, traditional polling is prediction markets. 55:42 And I mean, they have been banned in Brazil, but you can use a VPN and, uh, and, uh, take a look at Polymarket, Kalshi, and others. Uh, how do you see this kind of thing? Is there any scientific value? 55:56 There are people who say that, uh, because there is a financial commitment to people who are, uh, answering this, they would sometimes be more accurate in, uh, a, 56:10 a, taking the pulse of how people are feeling about an issue or in an election. 56:15 Uh, uh, do you see any sci- scientific value, anything that we should be looking at these platforms to, to try to imagine the results of a, of an election? Okay. 56:27 I'm gonna answer that, but, like, before I answer that, like, it just reminds me of, uh, one last thing that you asked me to answer in, in your last question, which was what I would recommend to viewers, uh, how to do seed polls. 56:38 And my recommendation- Oh, yes... actually is, like, don't focus so much in one or another poll, right? Like, polls are noisy. 56:45 Polls, like, even though we're talking about, like, how they are similar to each other this cycle, like, there is, like, you know, some room for that, like, there's gonna be some random fluctuations. 56:54 Rather than focusing too much on a poll, look at the averages, right? 56:58 So there are some aggregators, some poll aggregators that, like, uh, are computing averages and, and they're weighting also, uh, these different polls by different characteristics. 57:07 So that, that would be my recommendation. And I'm, I'm, I'm mentioning that to then segue with the, with the prediction markets, right? 57:17 Because honestly, for me, in a way, prediction markets, it's almost like a different type of aggregation, right? 57:26 Where you're aggregating, aggregating, uh, uh, betters, right, people who are betting, like, on, on those, uh, on those, on those markets', uh, points of views or, like, uh, beliefs or information, right? 57:41 Now, the thing is, I think there is a little bit of circularity there with polls because a lot of those, um, those people, uh, putting money on those platforms, they're using polls to inform [laughs] how they should, uh, for like, I mean, right? 57:56 Like, I, I, I think it's very rare that the, those markets are sh- are, are, are pointing to something radically different from what polls... 58:07 I mean, some people say, "Oh, no, in 2022, 2024, the markets were showing, like, something different from the polls." Not really. 58:14 Like, what the polls were showing 2022, if you- 2024 in, in US, if you really take them, like, seriously in, in terms of, like, not only their estimates but also, like, the uncertainty that they had, was that- Mm-hmm... 58:25 it would be a, a close race. Like, especially in the battleground states. And, and that's what happened. It was, like, a close race. Like, I know that for a lot of people it looked like it was, you know, Trump won, like, 58:37 um- Like handedly, yeah... you know, like- Big mandates. Yeah. But yeah. But, but not really. The reality was much, was much more nuanced than that. Yeah. 58:46 And, and, like, of course, yes, those markets, like, they were showing, like, Trump, like, ahead, uh, giving, giving a higher probability than ahead. 'Cause, like, that's another thing, right? 58:54 Those prediction markets, I don't like to compare prediction markets with polls because they are doing different types of things, right? 59:00 Polls, again, they're portraying their, a snapshot of the electorate at a certain time. 59:07 Those prediction markets, what they are doing, they are really, they, they are literally saying they are trying to predict what's going to happen in the future. That's not what polls are trying to do. 59:15 Like, that's not, like, at least in Brazil, most pollsters are saying. They're not saying that they are want- they want to predict the election, right? 59:22 So I don't even like that, um, that, that kind of a comparison to, to begin with. But, uh, what should we, how should we take those prediction markets? I would say that 59:33 I'm as, as a statistician, I like to use different sor- sor- uh, sources of information. 59:38 And I, I do think that there is some value in, in terms of, like, prediction markets in the sense of, like, it's another source of information, right? It's providing another type of information. 59:48 It's providing a probability, uh, that different candidates have of winning the election. 59:52 Um, how predictive it is, I mean, there are analysis that is pointing out that, like, well, they seem to perform, like, pretty well in the sense that, like, if they're saying that a candidate has 60% of a chance of winning, in 60% of similar races, that's what it would happen, right? 1:00:09 Like, that's the way you would kind of, like, assess that sort of thing. But, uh, there's also some... You have to keep in mind that's a probability, right? 1:00:18 [laughs] Like, the, it's not because a probability saying it's 60% it's going to happen, or even 70%, or even 80%, right? Like, uh- Right... there was this case recently here in the US. Uh, there was a race. 1:00:30 Uh, what was the race? Uh, there were two races. Here in Michigan, actually. And, um- No, no, it wasn't, it wasn't in Michigan. It was in Wisconsin, right? Uh, the go- the primary for Democrats there, uh, for governor. 1:00:42 And there was another one, I forgot. But, uh, that one, it was, like, putting, uh, the more progressive candidate, right, like, in, in the, in prediction. They giving a much higher probability. 1:00:53 I think it was on 90% or 95% probability of her, uh, winning that, uh, primary, and it was the mo- most more moderate candidate, that one. What was, what was the reaction of the, the prediction markets? 1:01:05 Well, they, they said- and like- I, I, I totally agree with that. Say, well, we told them that, we told you as, uh, uh, the platform says like 90%, not 100%. There was a 10- Yeah... 5, 10% chance of that happening. 1:01:19 That's true, but [laughs] but at the same time, um, when you get it right, so it's the double standards again, right? Like- Yeah... when they get it right, they, they don't say that. 1:01:30 They don't say like, "Oh, you know, uh, we, we said that 60% of it happens, but look, there was a 40% chance of that, uh, that, that could happen." They don't say that too, right? 1:01:38 So I also- I'm also a little bit annoyed by that. Um, but, uh, anyway, there are- there are gonna be some races that this sort of thing's going to happen. 1:01:48 And, um, you just have to understand that that's what probability is, right? Like, the probability is not really telling you what's going to happen. It's telling you what's more or less likely to happen. 1:01:59 And there are some races that- and, and by the way, aggregators are also, also compute that. But aggregators, unlike the platforms, they don't use like betting, right? Like to compute that. Mm-hmm. 1:02:11 Aggregators, what they use is like they use mod- statistical models. They use polls as one source of information. 1:02:17 They also use, uh, other data like how economy is doing, whether the candidates are, uh, re- re- rerunning for election or not, how is the approval rate of the current, uh, uh, president, for example. 1:02:33 So they're using all this information to then come up with a also probability of, uh, w- uh, winning probability of each candidate, right? 1:02:41 And there are some races that it's completely legitimate that, and that's what happened like in 2024, that the probability is going to be almost 50/50, maybe like 52/48. Uh-huh. Like, 1:02:52 you know, uh, and it's, it's a f- it's a coin flip. And could, could this happen in Brazil? Do you think- I, I don't think- That, that would be my question. Do you think that we are in that coin flip stage? 1:03:04 Or, uh, talking about probabilities, would you dare, uh, giving one probability that based on all the assessment of polls that you have been doing over the past few months, would you dare to say a probability t- uh, uh, of the outcome of this race? 1:03:21 Yeah, I mean, as, so as a statistician, and like I follow a tradition of stat- sta- sta- statistics that thinks that probability, like in, in, in essence is it's personal, it's objective. 1:03:33 Each person has their own probability of something, right? There's no such a thing as like a true probability, right? 1:03:39 Uh, and, and yeah, I mean, I, I wouldn't say a number, but I would say that like Brazil, based on what we're seeing on the poll- from the polls, is, uh, towards, in my opinion, uh, in my own subjective probability, uh, going towards like, uh, um, something close to a flip- a, a coin flip, uh, kind of like e- e- elections in a way. 1:04:00 Um, I maybe I would put, depending on the day [laughs] uh, depending on the results, uh, Lula a little slightly ahead, uh, also or Flávio is slightly ahead, but like honestly, it will all range. 1:04:14 It will be on that range of 40 to 60%, which again like- Right... if you're in the r- in the range of 40 to 60%, it's like al- it's almost like a, a coin flip in a way. So that's what I would say. Ravel, 1:04:27 Ravel Nishimura, thank you very much. It was a pleasure having you. Thank you for inviting me. It was a pleasure talking to you too. 1:04:34 Ravel Nishimura is Director of Sampling Operations at the University of Michigan's Survey Research Center. And that's it for this week's episode. Explaining Brazil is produced by The Brazilian Report. 1:04:45 If you enjoyed this conversation, subscribe wherever you get your podcasts, leave us a review, and consider supporting the show at buymeacoffee.com/brazilianreport. 1:04:56 And for our full 2026 election coverage, head to brazilian.report. I'm Gustavo Ribeiro, and thanks for listening. See you next time
