Lehohla model forecasts sharp ANC decline ahead of 2026 local elections

REAL NUMBERS

Dr Pali Lehohla

21 SEP 2026, 08:00

8 min read

The Lehohla Ledger’s application of Fisher Bayes Gauss Krig (FBGK) Actuation Method is a fine grained longitudinal mesh of data that I created.

This method moves beyond simple interpolation, integrating historical knowledge to create a predictive surface that evolves over time.

Here is the logic of how the Bayesian framework applies in the infrastructure of politics:

Defining the structural prior

In a Bayesian context, the "prior" is not a guess but a formal probability distribution representing our belief about the political landscape before we consider the new, specific data from a particular election (the "likelihood").

For the Lehohla Ledger, this prior is constructed from a multi decade longitudinal database. This historical model is the foundation of the fine mesh.

It encodes structural factors through 2,752 instruments and, for assessing and predicting voting patterns, the coding schema considers variables such as:

Voter Demographics: The stable socioeconomic profile of a specific area (e.g., census mesh data).
Historical Partisan Loyalty: The "sticky" nature of liberation movement support in certain regions versus the high volatility in others.
Spatial Correlations: The tendency for neighbouring voting districts to behave similarly.

Before any new poll or election result is fed in, the model already has a "prior expectation" for every single point on the longitudinal mesh.

This prior is a probability distribution, meaning it has both an expected mean (e.g., ANC will get 60% here) and an associated uncertainty (e.g., but it could reasonably be between 55% and 65%).

The Resulting Longitudinal Mesh:

By applying this Bayesian updating process to every point on a dense spatial grid, repeated for multiple election cycles, the Lehohla Ledger constructs a fine longitudinal mesh.

The final product is a dynamic, multi layered digital map where every pixel represents not just a single percentage, but a continuously evolving probability distribution. This mesh allows them to:

Track political movement over time: See exactly when and where a prior collapsed (e.g., the 2016 Nkandla shock).
Quantify uncertainty: Determine not just who is winning, but with what level of confidence.
Integrate diverse data sources: Seamlessly combine high quality but sparse national census data with low quality but frequent local opinion polls to create a unified, robust predictive field.

In the context of South African politics, the high confidence prior to 1994, rooted in the moral authority of the Mandela era, serves as an illustrative baseline. As new events unfold, evidence of policy success, the introduction of new leadership, or revelations of corruption, this incoming data acts as a "likelihood."

Through Bayes' Theorem, this evidence is systematically used to update the initial prior, resulting in a refined "posterior" belief. This posterior then becomes the prior for the next electoral cycle, creating a dynamic, longitudinal framework.

This approach is uniquely powerful for analysing political shocks. Discrete events, such as the Polokwane conference, the 2016 Constitutional Court ruling on Nkandla, or the emergence of the MK Party, the Phala Phala Parliament impasse, are not treated as isolated incidents but as severe Bayesian updates.

These shocks force a rapid and often significant revision of the prior distribution, lowering its mean and increasing its variance, thereby reflecting increased uncertainty and profound shifts in voter trust.

Ultimately, the logic of Bayesian priors prevents simplistic, static analysis.

It models why reputations for competence can be "sticky" over time, requiring a cascade of negative evidence to be shattered, and why, once shattered, they are exceptionally difficult to rebuild.

It provides the analytical structure to quantify the erosion of ANC dominance, illustrating how repeated negative updates have shifted the probability distribution of electoral outcomes from one of near certainty to one of profound fragmentation and volatility.

Based on the Bayesian prior analysis provided in "The Lehohla Ledger," the stage is set for a local government election (LGE) in seven weeks that is fundamentally different from any previous cycle. The key dynamic is the vicious feedback loop between collapsing national confidence and accelerating local governance failure.

Here is a summary of the priors leading into this critical contest:

The long national decline (1996–2026)

The national political landscape has undergone a profound Bayesian update, shifting from a high confidence, consolidated liberation movement to a fragile, fragmented coalition environment.

1996–2001 (High Confidence): The ANC began with a near certainty prior, rooted in the "Mandela Moment" and the early success of the democratic transition.
2006–2011 (Moderate Confidence: The Zuma Shocks): The prior mean began to drift downwards, marked by early evidence of corruption, Polokwane factionalism, and the ANC Youth League disruptions.
2016 (Low Confidence: The Post Nkandla Posterior): This was a significant shock. The Constitutional Court ruling on Nkandla and poor local election results acted as severe negative evidence, shattering the ANC's reputation for clean governance and discipline. The mean probability of success dropped significantly.
2021 (Moderate Low Confidence: The Ramaphosa Hope): There was a modest attempt to update the prior in a positive direction with Cyril Ramaphosa's "New Dawn," but this was tempered by persistent state capture evidence and economic stagnation.
2026 (Low Confidence: The Coalition Reality): This is the current national prior. The posterior from 2021 taught voters that the ANC could no longer govern alone in key metros. The reality of unstable, often paralysed, coalition governments is the new baseline.

The local government acceleration (2016–2026)

The local priors show a more acute and rapid collapse, directly feeding into the national volatility.

2016 (Moderate Confidence): While the national prior was dropping, local governance was still seen as moderately viable, though already showing signs of decay.
2021 (Low Confidence: The Coalition Shock): The LGE acted as a brutal update. The widespread loss of outright majorities in metros and the subsequent chaos of hung councils became the new, permanent local prior. Voters learned that local governance failure is now the norm.
2026 (Very Low Confidence: Governance Collapse Expectation) as emerges from the Madlanga Commission: As we stand seven weeks away, the prior distribution is defined by profound cynicism. Voters are experiencing daily, visceral evidence of total municipal collapse: persistent loadshedding, water crises, and non existent service delivery. The negative evidence from 2021 (coalition paralysis) has been confirmed and deepened by the 2024 national shock (loss of the ANC's outright majority).

Setting the stage for seven weeks out

The stage is not just set for a change; it is set for a new era of minoritarian governance. The ANC enters this LGE with its local brand irreparably damaged by the evidence of the past decade. The upcoming election will not be a contest for ANC dominance, but a fierce battle among various opposition and breakaway formations (like the MK Party) to form viable, often anti ANC, coalition blocs that can pick up the pieces of the shattered local governance prior.

The challenge for all parties is not to shift a positive prior upwards, but to manage a voter base whose prior belief is one of expected governance failure and deep instability.

The deployment of the FBGK method is now applied to the forthcoming local government elections that are seven weeks away. To it, I apply Bayesian priors to the national outcome and Jometro, which will by all counts be heavily contested.

The following shocks apply:

Phala Phala, where the president of the country has a question to answer.
ANC fracture, where its senior leadership are contesting whether or not money exchanged hands at its electoral conference.
Coalition instability, where the GNU has exposed massive challenges.

The graphic below shows that, as contestations emerged, the pre election baseline according to the Lehohla Ledger FBGK analysis reflects ANC taking 46.2% of the poll, but with the three Bayesian priors emerging, the ANC Voting Districts from bottom up aggregates to a drop of 13.8 percentage points from 46.2% to a 32.4% poll.

The analysis is taken to Jometro, where ANC in the pre shock period is predicted to poll 38.1%, but as a consequence of the emergence of these priors, it drops by 10.8 percentage points to 29.4%.

What is at play, for instance, are the dynamics playing out in front of society, such as the Mbalula Zuma Dynamic, the MK Party emergence, Ramaphosa’s Phala Phala cliff hanger, (although in the analysis the guilty or innocent verdict is neither going to exacerbate nor ameliorate the drop because it is linked to a court of public opinion that doubts the judiciary) and the Madlanga Commission and the Public Sector Integrity drivers, which are already factored.

The next seven weeks will continue to throw surprises and the Lehohla Ledger will, based on the FBGK application using Bayes Priors, update the total number of VDs political parties will win and the associated voter poll.

*Dr Pali Lehohla is the former Statistician General of South Africa, Director of the Pan African Institute for Evidence (PIE), and the founder of the Lehohla Ledger. He is a Professor of Practice at the University of Johannesburg and a Research Associate at Oxford University.