Latest / Drive with John Perlman / How accurate were election predictions?

Transcript

How accurate were election predictions? · Drive with John Perlman

• 0:00
  Speaker 1: Local issues to the bigger picture follow the local government
• 0:03
  Speaker 1: elections on 702 drive.
• 0:04
  Speaker 2: With john perlman Right, let's go back to the numbers
• 0:12
  Speaker 2: and bring into the conversation the Council for Scientific and
• 0:16
  Speaker 2: Industrial Research. They did some heavy-duty predicting of what might
• 0:21
  Speaker 2: happen in the local government elections, and to talk about it,
• 0:25
  Speaker 2: we're joined now by Professor Pravesh Deba, who's Principal Statistical
• 0:29
  Speaker 2: Researcher at the CSIR. Prof, welcome, and thanks very much
• 0:33
  Speaker 2: for giving us your time. Using data for predictions, is
• 0:38
  Speaker 2: it more difficult for local government elections than it might
• 0:41
  Speaker 2: be for provincial and national? Because there, as I understand it,
• 0:44
  Speaker 2: there's a direct relationship between the number and its expression
• 0:49
  Speaker 2: or manifestation in political power. Whereas in local governments, you
• 0:54
  Speaker 2: can get lots of votes all over the place, which
• 0:56
  Speaker 2: makes up a nice number. But it doesn't mean you
• 0:59
  Speaker 2: may not win an award in the process.
• 1:02
  Speaker 1: Thanks, John. And, you know, perhaps first I just want
• 1:05
  Speaker 1: to say that the work that we are doing is
• 1:10
  Speaker 1: in collaboration with the SABC as well.
• 1:12
  Speaker 2: Oh, great.
• 1:13
  Speaker 1: And, you know, what we are doing is we are
• 1:19
  Speaker 1: trying to determine what is the final outcome for each
• 1:23
  Speaker 1: of the political parties with, you know, as little of
• 1:26
  Speaker 1: the voting districts declared. So we try to do that
• 1:30
  Speaker 1: as early as possible. And you are absolutely correct that
• 1:35
  Speaker 1: for local government elections, it's much more difficult to predict
• 1:41
  Speaker 1: what the final outcome would be as opposed to your
• 1:43
  Speaker 1: national and government elections. And there's many reasons for that.
• 1:49
  Speaker 1: And one of the reasons is that it's very difficult
• 1:54
  Speaker 1: to take care of all type of local effects. And,
• 2:01
  Speaker 1: you know, there are so many local effects in the
• 2:04
  Speaker 1: local government elections and you can't possibly model everything. So,
• 2:09
  Speaker 1: you know, there's a lot of assumptions the one is
• 2:11
  Speaker 1: making and, you know, we are using a sort of
• 2:17
  Speaker 1: an overall effect to try and understand what is happening
• 2:20
  Speaker 1: at a local level.
• 2:22
  Speaker 2: So far, from what I've looked at, your earlier predictions,
• 2:25
  Speaker 2: they seem to be pretty much on track. The DA
• 2:28
  Speaker 2: is higher than you guys predicted, but there's still a
• 2:32
  Speaker 2: long way to go. And the expectation, I spoke to
• 2:35
  Speaker 2: Helen Ziller earlier, that there may well be areas, although
• 2:38
  Speaker 2: she actually expected it to balance out. But with the ANC,
• 2:41
  Speaker 2: you're almost bang on.
• 2:43
  Speaker 1: Yes, that is correct. And I think also, you know,
• 2:47
  Speaker 1: the audience need to know that The model actually is
• 2:54
  Speaker 1: easier to predict for a large political party than smaller
• 2:58
  Speaker 1: parties as well. And for ANC, right from the beginning,
• 3:04
  Speaker 1: we did get the percentages correct. For DA, we do
• 3:10
  Speaker 1: tend to have some difficulties, especially in the Western Cape,
• 3:14
  Speaker 1: to do that prediction correctly. Sometimes it doesn't always work.
• 3:27
  Speaker 1: In this election, we also tried to solve that issue
• 3:32
  Speaker 1: by creating a separate cluster for the Western Cape. It
• 3:40
  Speaker 1: certainly did help, but it did not totally resolve the problem.
• 3:45
  Speaker 2: The complexities of trying to reduce human choice and behavior
• 3:52
  Speaker 2: to hard and fast stats. But thank you, and we'll
• 3:54
  Speaker 2: follow with a great deal of interest on the SABC
• 3:58
  Speaker 2: the work of the CSIR. Thank you to Professor Pravesh Deba,
• 4:02
  Speaker 2: who is Principal Statistical Researcher at the Council for Scientific
• 4:06
  Speaker 2: and Industrial Research. We appreciate your time.