Hello all, welcome to the home of SA election data. I am going to treat this account like a live blog almost, talking about the data I have, how things are being developed and of course the outcomes. If you are interested in that kind of stuff, I would love to talk to you about it. Insights, thoughts, requests all welcome. All of the final data will appear on my blog - inside-politics.org - it already has a ton of stuff up on it, and I am always growing and developing it. So, if this kind of thing appeals to you, welcome!

A thought or two on the DA's internal polling, as reported in News24. There is a lot to be taken away from that, but one figure above all others stands out: 51% in CT. That will be of deep, deep concern to the DA, 20 years of progress hovering on a 1% knife edge. It is difficult to explain too: metro performances for the DA usually follow a uniform trend. It is highly unusual for the DA to be up in almost every metro, but down in CT, and significantly down at that. The results suggest no one cause. Rather, the "death by a thousand cuts" syndrome that has plagued the party for years now. A hundred small parties all taking a bite. That is incredibly difficult to counter, so many battle fronts. The DA's national squeeze message is "Stop the Chaos". Although there are variants of it (I particularly like the ETH message: You could wake up to an MK government or a DA government, the choice is yours). I am not sure if the party has a tailored squeeze message in CT, the national one doesn't seem to make sense to run there, as there is no coalition. It needs a very powerful one though. 51% is a crisis. That 20 years of progress vs 1% knife edge seems pretty powerful, really does bring home the "every vote counts" message. There are upsides to close calls in politics, can make your squeeze campaign far more effective. DA needs the mother and father of squeeze campaigns in CT, and those polling numbers could be exactly what delivers the message. Time is short though!

Some thoughts on Herman Mashaba and arrogance Herman Mashaba has now made it clear, on numerous platforms that he will not work with Helen Zille, if a coalition between ASA and the DA is a possibility. Most recently, on the Sharp Sharp podcast, he said, “I have made it clear to the DA, in the city of Johannesburg, in the event - if they are lucky - if they get anywhere close to an opportunity to work with us, they must present someone else.” As things stand, the DA looks set to win the most votes in JHB, with 30%+, ASA, looks set to come in somewhere around 5th, with around 10%. Now, Mashaba has also - many, many times - declared the DA “arrogant”. It is an odd juxtaposition - one man, who would dismiss the will all those voters who voted for the DA and Zille, on the basis that he, Herman Mashaba, has an issue with her. But, simultaneously, label the DA arrogant. It would seem to me his position is the very definition of arrogance. And here it is important to understand what Mashaba is saying. It is perfectly reasonable for a party to write off any arrangement with another party. But you can’t have your cake and eat it. You cannot write off that party’s democratically elected representative, and have the party. The difference between those two things can only ever boil down to some personal grievance, for it does not apply to the voters - for them the two are intricately linked. Indeed, for many the individual - Zille - may well supersede the appeal of the party. And right there is where you find Mashaba, a man who’s vote share in JHB looks set to decline (in other words, he has lost support amongst the people), who seeks to dictate to a party that looks set to growth (in other words, which has grown its trust among the people). Any person who believes their personal experience so important it deserves to be elevated above the democratic will of hundreds of thousands of voters cannot but suffer some form of grandiosity that is out of control. It is an important insight into Mashaba: if he is willing to do that, to what degree, in any coalition arrangement, can he be expected to ever cater to the electorate? If his personal disposition is more important than voting itself, surely there is nothing he will respect other than his own will? One cannot be a good public servant, if the public’s position counts for nothing. If the situation does come down to a possible DA/ASA coalition, you can be sure Mashaba will replay every defamatory card he has shown this campaign. But that really will not be the point. The point, which all parties need to bear in mind, is this is a man who has promoted himself above the electorate. And it doesn’t matter which political party you are, that is the most dangerous type of person in politics. People say politics is the art of the possible. Nowhere is that more apposite than coalitions, which require compromise and maturity. For Mashaba, politics is the art of the personal, and that is the very definition of arrogance. No flag burns a brighter red.

An update on the impossible model: predicting turnout. I have continued work on the comprehensive model I tweeted about earlier, I think it is quite good and we will see how it holds up. However, I wanted to talk about new registrations only, as an indicator for turnout (this is a component of the bigger model, but it worth talking about alone). Repeatedly - more than changes in the voters roll, and re-registrations or even all registration activity - new registrations demonstrates the most powerful relationship with turnout. So, without being too complicated, I have simply used that as a "predictor". its results differ from the bigger model, fairly substantially - it predicts 47.9% (using only LGEs) - the bigger model predicts 37.2% (using only LGEs) - this is helpful for me, as I have two models that quite nicely cover growth and decline, which will be great when evaluating the final result, with a view to getting this all 100% before 2029. Below are what you get if you only use new registrations to predict turnout (these numbers are only available by province). NPEs are included just for reference, but they aren't used for the prediction. Some key points: 1. The column Corr (correlation) is important. This is the measure of the strength of the relationship between new registrations and turnout, within each province, for all LGEs. These are very strong readings, (1 is a perfect relationship, so around 0.8, 0.9 very strong) 2. Each cell contains three figures: turnout, the new registration rate and the "fit" (Near 1 = strong fit; near 0 = outlier). 3. The final column, the one you will be interested in, is the 2026 prediction. New registration rates were higher this election than 2021 (2.86 vs 1.8) If the historical relationship holds, this data set says turnout should increase marginally from 2021, and in every province bar the NC and WC. This is a massively counter-intuitive finding, and frankly difficult for me to accept. All other evidence (turnout trends in particular) say turnout will drop. But it is hard to argue with the historical strength of the relationship and the fact that the new registration rate was substantially up in 2026 (up a 1pt, which is very big). As a result, I can't disprove these outcomes as they stand, and one must take them seriously. I was hesitant to even publish them to be honest, and maybe the actual result will reveal a fundamental flaw in how this was constructed, but, as it stands, well, there it is.

In the first iteration of the JHB 2026 map, there were around 26 (3%) of all VDs "unclassified". This was because of boundary changes. To try and account for those, I classified them as ANC-first, DA-first or Other-first, by using their largest geographic overlap with a 2024 VD. Not 100% perfect but I think leads to a more accurate full accounting of registration trends. Doesn't change much as far as the DA's ratio goes, drops from +2.83 to +2.59, but does boost the ANC from an average of 12 to 22. Sorry, these are consequences of live tweeting your thinking! Changes will be made. Here is revised summary graph.

“The rabbit-hole went straight on like a tunnel for some way, and then dipped suddenly down, so suddenly that Alice had not a moment to think about stopping herself before she found herself falling down a very deep well.” Ok, so I have now plotted the DA’s turnout advantage over the ANC, for the same election set: 2006-2021. It uses the same VDs – those where the ANC finished first and where the DA finished first, as well as the average turnout across the whole metro. Here is where things get really interesting, we can work out a set of ratios for the DA's turnout advantage as well (top boxes), and then we can test the hypothesis: does a registration advantage translate into a turnout advantage, by seeing if the two sets of ratios correlate in anyway. First, the graph below. Turnout in DA-first VDs is continuously higher than ANC-first VDs. That is an encouraging first sign, that is the same pattern as in the registration graph. So there is one big obvious correlation. But how precise can we be? I wont set out all the turnout ratios here, you can read them on the graph if you interested, but I will translate both sets of ratios into percentages. Do that, and you get the following: Election 2006: +375.00% [DA registration advantage]; +18.00% [DA turnout advantage] 2009: +39.00%; +3.00% 2011: +94.00%; +19.00% 2014: +44.00%; +7.00% 2016: +96.00%; +29.00% 2019: +28.00%; +9.00% 2021: +543.00%; +35.00% 2024: +83.00%; +17.00% Now, we have some more practical choices to make. 2021 seems extreme, and the temptation may be to exclude it. But actually, I think it should be left in. The proportions make sense at first glance. It is 2006 that is odd, and which I am inclined to leave out. I have no good statistical reason for doing this, I just feel it complicates things in an unexplained way, 2006 is a random start date anyway (don’t have 2000 or 2004), it is the oldest data set, and it doesn’t interrupt the timeline by excluding it. Let’s just start at 2009 and see what the stats say. I am going to use a Spearman correlation. There are other options, but I like this one, for this. A Spearman correlation measures how strongly two variables move together in terms of their relative rankings. It works by ranking the values of each variable, then measuring how closely those rankings correspond. The result ranges from −1 to +1. +1 = Perfect positive relationship. As one variable increases, the other always increases. 0 = No consistent relationship. −1 = Perfect negative relationship. As one increases, the other always decreases. If we exclude 2006, we get a Spearman correlation of 0.893! That is an amazing result, a very strong positive correlation. Rare to get such a high number in the social sciences. Now, the fun part. We can put in the 2026 registration ratio (+2.83) and, using those past patterns, get an estimate for the DA’s turnout advantage in 2026 in JHB. The number that comes out is 19.49% - essentially a 2011-level advantage. Hmmmm, does that make sense? In 2011 the DA’s turnout advantage was 19%, in 2016 it was 29% and in 2021 it was 35%. Given its great registration ratio - +2.83 – one of its best ever, 19.5% seems a bit low. But ask yourself this: will the DA repeat 2016? Unlikely, and that was 29%. Will we have a covid like turnout collapse in 2021 (a drop of 13pts)? Unlikely, turnout will probably decline but 2024 suggests it has recovered a bit and won’t plummet. So maybe a 19.5% advantage entirely realistic? At any rate, this was a fun experiment. Will be interesting to see what happens on the day and reflect on the correctness or incorrectness of the maths here. But I am right at the bottom of this particular rabbit hole now. Even Alice found the bottom, and I plan to now find my way out and onto other things.

Ok, the rabbit hole has me. What can I say, couldn't resist. Have now mapped the changes in voters roll per VD over registration weekends back to 2006 for Johannesburg, which is as far back as this kind of data goes. Here is the summary graph. It is profoundly depressing, an ode to voter apathy and alienation. I had to fundamentally redo the scale prior to 2016, so dramatic was the change with earlier registration weekends. Here is the ANC vs the DA vs the average in each election. In the top boxes is the DA's ratio - the rate at which the voters roll in JHB increases per VD in the DA's favour (a "+" sign means the ratio is in the DA's favour; "-" would be in the ANC's favour, but that never happens - blue line always above the green line). 2006: +4.75 (DA controlled VDs grew the voters roll over both registration weekends by an average of 4.75:1 compared to what happened in ANC controlled VDs) 2009: +1.39 2011: +1.94 2014: +1.44 2016: +1.96 2019: +1.28 2021: +6.43 2024: +1.83 2026: +2.83 I think one can ignore 2021, as a Covid driven outlier event, for the most part, the DA's advantage hovers between 1.2 and 1.9. Seen in this grander historical context, 2026's figures really are exceptionally good. As I said before, the context will mean this counts for less (an advantage in a depressed environment is less of an advantage - the entire bar drops) but there is nothing the DA can do about that. On the day, even if the pool of actual voters is smaller, the scene would seem set for one of its best turnout differentials yet. That said, just look at those earlier rates, particularly in NPEs, 2009 and 2014. The average growth in the voters roll back then was 449 new voters per VD (!) and 277 respectively. Those days long gone now. 2026's pathetic average growth per VD? Just 31 new voters. And the graph is headed downwards, no doubt about that.

Some picture thoughts on Gauteng and coalitions One of the problems with local elections is they are incredibly myopic, every metro is a government unto itself. Campaigns and thinking is structured that way. Gauteng is an outlier in this regard. It’s three metros - JHB, TSH and EKU - sit side by side. It is also SA’s economic hub. I did propose elsewhere that, for parties like the DA, selling a “big picture” unified vision for these three metros was a necessity. Voters like grand visions. And it speaks to stability as well. Alas, that hasn’t happened. But coalition negotiations still mean it is possible. The lay of the land suggests all three metros will follow a similar pattern when results come in - an opportunity for three similar administrations, that work together rather than compete. The idea of three metro mayors sitting down to discuss economic growth and integration is an idea so alien to current campaigning that it probably seems far-fetched. But it is quite possible. And, I would argue, necessary. If you want to fix any one metro in Gauteng, you need to fix all of them. I would hope parties thinking about this, and not too ad hoc about coalitions. Real opportunity here to craft three similar administrations, with the same agenda, centred on restoring Gauteng to its position as continental economic powerhouse.

At the risk of going down a JHB rabbit hole, at the expense of other metros, I have worked out the average change in the voters roll overall, for ANC-controlled VDs and DA controlled VDs, for the 2016, 2019, 2021, 2024 and 2026 registration weekends. This was important to give some context to the 2026 numbers, which showed the Voters Roll in DA VDs grew by about 3x as much as it did in ANC VDs, after both registration weekends in 2026. The graph below suggests, while the DA still dominates the ANC in terms of a ratio - which is great for the party - the raw number is low, far lower than 2016. In 2016, the average DA VD grew by 142 voters, the average ANC VD by 72, for a ratio of 2:1. In 2026, the average DA VD grew by 34 voters, the average ANC VD by 12, for a ratio of 2.8:1. So the bottom line is, the DA's ratio is very promising for disproportional turnout, but voter enthusiasm in all markets, including the DA's is dropping. And while the DA's turnout differential looks like it will be strong, it's very likely going to be achieving that in a smaller pool of actual voters.

2021 only had one registration weekend, but my goodness was it bad for the ANC. Look at all that red (voters roll decline) in Soweto and surrounds. (This map only shows VDs in which the voters roll shrank in 2021).

Let’s look at the N24/Sabi and ENCA/Markdata polls in JHB and see which “big party” mayoral candidate/party is running the most effective campaign. We will use the 2021 results in JHB as the benchmark against which to measure their performance. • Helen Zille [DA]: 2021: 25.5% (N24: 29.2%, ENCA: 35.0%) - growth of 4pts to 10pts • Tlaleng Mofokeng [EFF]: 2021: 10.1% (N24: 11%, ENCA: 11.9%) – growth of 1pt to 2pts • Bongani Baloyi [MK]: 2021: Did not stand (N24: 12.3%, ENCA: 12.5%) – no baseline. MK got 13.2% in JHB in 2024, against which it would have a decline of around 1pt. • Herman Mashaba [ASA]: 2021: 18.1% (N24: 7.0%, ENCA: 10.3%) - decline of 8pts to 12pts • Dada Morero/ANC [ANC]: 2021: 33.2% (N24: 22.6%, ENCA: 21.8%) - decline of around 11pts There is a graph setting all this out below. The biggest problem here (apart from the ANC, but it seems to have given up and not even had a campaign or mayoral candidate, so there is little to analyse) is ASA. If these trends hold, it will have a disastrous campaign, in its stronghold. Its strategic approach to it seems to have been to go intensely negative on the DA, and Helen Zile (she is Verwoerd, she will bring back apartheid, she is responsible for JHB’s condition) and on immigrants (they are responsible for JHB’s condition). As things stand, it seems to have backfired. ASA continues to appear held hostage by the personal animosity its leaders have towards the DA. It has spent 3 months now, venting its spleen. The results are not encouraging. In this environment, where the ANC’s vote is fracturing left, right and center, it will be a dire result if it holds. Still some small amount of time to improve those numbers, but you feel the party is unlikely to escape the trappings of its own emotional disposition in any meaningful way before election. MK and the EFF are more of less stable, EFF shows some prospect of some small growth - but within margin of error. DA looks set for a very successful election, but that might well be forgotten in the coming battle for coalition control and the mayoralty.

These are the 2026 provincial predictions that underpin that national result in previous tweet (with 2021 turnout in brackets for comparison).

I need to quite clear that this turnout model is an experiment. It has never been done before and, as a result, there are a lot blind judgment calls. The truth is, the actual 2026 turnout number is the final ingredient, to properly understand which modules worked, and which didn't and how to refine it all hopefully with a view to make it really good by 2029. That said, nothing ventured, nothing gained. This is the first estimate of the model, based on 7 modules, 4 elections and combined NPE+LGE results. I am going to keep testing it. Will keep you updated.

An update on the turnout model. I have finished four modules, each of which will feed into the final model. Module 01 [urban/rural] - looks at the relationship between urban/rural VDs, new registrations and turnout Module 02 [new registrations] - looks at relationship between new registrations alone and turnout Module 03 [historic turnout] - looks at historic turnout patterns (particularly change between NPEs and LGEs] Module 04 [political geography] - looks relationship between ANC-first, DA-first and Other-first VDs, new registrations and turnout. Each of these modules have two baselines. One is a six election data set (2011, 2014, 2016, 2019, 2021 and 2024) and one is four election data set (2016, 2019, 2021, 2024). The benefit of the 6 election baseline is more data. The downside is older data sets impact result. The benefit of 4 election baseline is less data but the module far more attuned to recent trends. These are not all the modules, there will be around eight of them in the end but, just for fun - I wouldn't rely on any one of these modules in and other itself, rather we have to see what they all produce together - I asked each completed module to make a 2026 turnout prediction: one using NPE results only, one using LGE results only and one using both NPE and LGE results. These are the outcomes: Module 01 [urban/rural] 6 Year module [2011, 2014, 2016, 2019, 2021, 2024] Historical calibration and predicted 2026 turnout • NPE only: 38.31% • LGE only: 45.20% • NPE and LGE combined: 43.05% 4 Year module [2016, 2019, 2021, 2024] Historical calibration and predicted 2026 turnout • NPE only: 39.44% • LGE only: 34.79% • NPE and LGE combined: 40.18% Module 02 [new registrations] 6 Year module [2011, 2014, 2016, 2019, 2021, 2024] Historical calibration and predicted 2026 turnout • NPE only: 37.84% • LGE only: 45.53% • NPE and LGE combined: 42.49% 4 Year module [2016, 2019, 2021, 2024] Historical calibration and predicted 2026 turnout • NPE only: 38.54% • LGE only: 41.35% • NPE and LGE combined: 40.37% Module 03 [historic turnout] 6 Year module [2011, 2014, 2016, 2019, 2021, 2024] Historical calibration and predicted 2026 turnout • NPE only: 39.43% • LGE only: 45.26% • NPE and LGE combined: 41.96% 4 Year module [2016, 2019, 2021, 2024] Historical calibration and predicted 2026 turnout • NPE only: 37.49% • LGE only: 32.87% • NPE and LGE combined: 36.34% Module 04 [political geography] 6 Year module [2011, 2014, 2016, 2019, 2021, 2024] Historical calibration and predicted 2026 turnout • NPE only: 38.24% • LGE only: 44.59% • NPE and LGE combined: 42.42% 4 Year module [2016, 2019, 2021, 2024] Historical calibration and predicted 2026 turnout • NPE only: 36.64% • LGE only: 39.82% • NPE and LGE combined: 39.82%

Am trying to build a model to predict turnout. I feel I am getting closer, but it is very difficult. On version 58 now. Essentially it was initially designed to use changes in the voters roll over registration weekends. It was solid (out by a few points on most elections). But 2021 is proving a very year to get right, and that means the others - even though they appeared rightish, weren’t. So I have moved onto a more complex model, that sits outside the data. It will work on a range of variables, including urban/rural divide, population group, all registration activity, previous turnout levels and party support. You can in theory then feed in the 2026 data, and see what it says. These are a lot a variables to build and identify meaningful predictive relationships with. And I’m not sure if it will work, but we shall see. It would be such a game changer if you could get an accurate forecast of turnout. I really do feel it is possible. Interestingly, one of the strongest predictive variables so far is actually new registrations. Two weeks to get this right, if possible at all.