Extrapolaters gonna extrapolate: early vote-counting and its problems

17 November 2015

The fun in by-election analysis  – and it’s not a rich seam of fun – does not come from the Excel sheets and the tables and the maps. The fun comes from people’s reactions to the results.

I am a bit of a nerd for numbers (as many famous sociologists have observed). My day job involves data analysis, data cleaning, data everything. I like to unwind by writing by-election reviews.

I get crapped on from time to time for the reviews, and I’ve written about that here. If the ANC has a good day at the polls and I say so, I am liable to catch flak from opposition parties. If I say the DA has done well then I might as well be cheerleading the return of apartheid as far as ANC fans are concerned.

People love to confirm their biases with election results. I was reminded of this when I read this BizNews.com article about last week’s by-elections. (My own review is here.)

The DA did very well in those elections, winning a ward, defending three others, and increasing its share of the vote in every ward it contested bar one. What does this mean for next year’s municipal elections?

If you are me, it doesn’t necessarily mean much. By-elections are significantly different from general elections (I will explain why in a moment). I’m also not in the business of projecting and predicting too much: I’m not a polling company and I don’t feel the pressure to prognosticate. Mostly, I don’t like to be wrong, and if you don’t make any predictions then you can generally avoid being wrong.

The BizNews.com article took last week’s by-election results and extrapolated mightily from them, suggesting that the DA looked good for absolute majorities in both the City of Johannesburg and Tshwane. Much of this analysis was based on the DA’s gains in the Johannesburg and Tshwane wards that were contested: the DA increased its share of the vote in both wards by over 10 per cent.

By-elections are not general elections. Here’s five reasons why they are different:

The wards that are contested in by-elections are not chosen at random: I wrote about this in more detail in this research note, but the quick summary is that, with the exception of councillor deaths, by-elections are often a managed process. Sometimes councillors are expelled or they resign. Sometimes, following a national election, they are promoted to the National Assembly or a provincial council. Sometimes – well, increasingly often – a whole clump of ANC councillors are expelled or ANC-led councils are dissolved.

The only people who don’t get a say in where and when a by-election takes place are the constituents themselves. Bad councillors are not expelled because people call for their ouster. Good councillors are not retained because of the people’s will.

The ANC generally has more to lose in by-elections: The party holds the majority of council seats. The law of averages suggests that the party will defend more wards. This does contradict the previous point a bit, and the shenanigans in many Western Cape municipalities has seen the DA defend more wards than it should on average. Outside of the Western Cape this rule holds though.

By-elections aren’t public holidays: People have to take time off from what they normally do to vote in by-elections. It’s fair to say that richer voters with private transport are more likely to vote.

The DA is good at mobilising and getting its voters out: It’s a well-observed trend that the ANC’s successes are directly proportional to the voter turnout. Put another way, opposition supporters are more ‘sticky’ than ANC supporters. The DA has fewer voters than the ANC but these voters are more likely to vote in municipal elections.

Voter turnout, as a percentage of registered voters, is generally quite low in by-elections. If you want to confirm real growth in a party’s support it’s better to compare the absolute number of voters for a party than the percentage of the total vote accruing to each party.

There are a few wards where the DA has increased its absolute numbers. I’d be more comfortable extrapolating from these wards. In the City of Joburg ward where the DA went from 73 per cent to 93 per cent the party actually went from 5 601 votes to just 2 985 votes. That’s because voter turnout fell from 57 per cent in the 2011 elections  to just 22 per cent in the by-elections.

In the Tshwane ward, the DA did go from 577 to 646 votes. That’s a better story. An increase of 69 votes is less impressive to report than an increase from 9 per cent of the vote to 23 per cent, however.

By-elections are often a two-horse race: In municipal general elections everyone tends to have a bit of a punt and a flutter. That’s why you’ll find the IFP and the NFP putting up candidates in Gauteng. There’s a bit of strategy in the scattergun approach – if you win 3 per cent in every ward you’ll end up with about 3 per cent of the total vote and you stand a chance of getting one or two PR councillors elected.

In by-elections there’s often no payoff for small parties to put up candidates. Case in point: five parties contested the Ngwathe (Parys) ward in 2011 and just the ANC and DA contested the ward last Wednesday. In 2011 the DA won with 87 per cent of the vote; last Wednesday the party won with 99 per cent.

According to BizNews, this 12-percentage-point increase in the DA’s share of the vote is a portent of things to come in next year’s elections. But in 2011 the ward was also contested by the FF+, which took 9 per cent of the vote. The FF+ didn’t contest the ward on Wednesday. It’s a fair bet that those FF+ voters were never going to vote for the ANC.

The DA, by the way, went from 2 156 votes in the ward to 1 540 as the voter turnout fell from 66 per cent to 43 per cent.


I think the DA will increase its share of the vote in many municipalities come next year, and I think the ANC will be run pretty hard in Johannesburg and Tshwane. But the DA’s share of the vote is unlikely to increase by ten percentage points or more, unless the ANC suffers a dramatic drop in its voters turnout.


Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>