Latest / The Midday Report with Mandy Wiener / CSIR prediction model for 2026 LGE
Transcript
- 0:00Speaker 1: So as many of you continue to cast your ballots today,
- 0:02Speaker 1: we're not looking at the results just yet. Obviously, we
- 0:06Speaker 1: will be doing that at the results center over the
- 0:08Speaker 1: next few days. And one of the people or the
- 0:12Speaker 1: organizations we rely on quite heavily is the CSIR. They
- 0:16Speaker 1: have an excellent prediction model. We saw from the last
- 0:19Speaker 1: elections how accurate it was. So let's just find out
- 0:22Speaker 1: a little bit about that with Professor Pravesh Deba, who's
- 0:25Speaker 1: a researcher and analyst at the CSIR. Prof, good afternoon
- 0:28Speaker 1: to you. Thank you so much for your time. I'm
- 0:31Speaker 1: sure you're standing by with your prediction model ready to go.
- 0:34Speaker 1: How soon can you start the engine rolling and get
- 0:38Speaker 1: it wound up?
- 0:41Speaker 2: Yes, good afternoon, Mandy, and welcome. We do the predictions, which,
- 0:47Speaker 2: you know, this is being done in partnership with the SABC.
- 0:53Speaker 2: And we would start the predictions from the very first
- 0:58Speaker 2: voting districts that is being declared. So it'll be from
- 1:02Speaker 2: a little after midnight tonight.
- 1:05Speaker 1: And then what patterns do you look at in terms
- 1:07Speaker 1: of voting behavior?
- 1:08Speaker 2: Yeah. So the way the model actually works is, so firstly,
- 1:18Speaker 2: we use 2024 elections data, so the results for 2024,
- 1:24Speaker 2: where we form voting districts that, or we cluster voting
- 1:30Speaker 2: districts that are most similar to each other And when
- 1:35Speaker 2: the actual results start to come in, we compare those
- 1:39Speaker 2: actual results to these clusters that we have already developed.
- 1:45Speaker 2: So what it's actually doing is it's comparing the way
- 1:49Speaker 2: in which people have voted at a particular voting station
- 1:54Speaker 2: to our clusters. And therefore, if there's changes at the
- 1:59Speaker 2: voting district, that is observed through the model. And therefore,
- 2:06Speaker 2: you know, that is then applied to all voting districts
- 2:08Speaker 2: that have not been declared. And once that is tallied,
- 2:12Speaker 2: you know, you get the percentage for each of the
- 2:14Speaker 2: political parties. Yes.
- 2:16Speaker 1: I remember in 2024, the CSIR's model called a national
- 2:20Speaker 1: result at about lunchtime on the day after the elections.
- 2:23Speaker 1: Do you have an idea of when you might be
- 2:25Speaker 1: able to do the same tomorrow, considering it's a local
- 2:28Speaker 1: government election?
- 2:30Speaker 2: Yes. So actually in 2024, it was actually the latest
- 2:37Speaker 2: that we actually called out any result. But it wasn't
- 2:40Speaker 2: around lunchtime. It was actually at half past seven in
- 2:43Speaker 2: the morning. But that was the latest ever that we
- 2:46Speaker 2: caught up.
- 2:46Speaker 1: I must apologize then because I have a different recollection.
- 2:49Speaker 1: But you obviously know better than me.
- 2:53Speaker 2: So sometimes, you know, if we are lucky and the
- 2:58Speaker 2: data comes in quickly, like in 2019. In 2019, you know,
- 3:04Speaker 2: we received about 5% of the voting district declared by
- 3:07Speaker 2: quarter to three in the morning. And therefore, we released
- 3:11Speaker 2: the results as early as that. But usually, you know,
- 3:15Speaker 2: one can expect results by around about half past five
- 3:21Speaker 2: tomorrow morning.
- 3:22Speaker 1: Look forward to chatting to you tomorrow then. Prof, thank
- 3:25Speaker 1: you so much. Prof Pravesh Deba, researcher and analyst at
- 3:28Speaker 1: the CSIR.