It was becoming increasingly clear to me that terrible things happen when our leaders fail to think about data that are outside their typical focus.
Max Bazerman
The Interval - recurrence of conflict
South Africans, and several other states globally, are days away from elections, and again the stakes are high, as are the hopes and expectations. We are easily lulled into forgetting that these were the emotions, the expectations, the saviours only a few months or years ago.
Most wars that end are followed by more violence. Ceasefires and negotiated settlements hang by threads, until the predictable, seemingly inevitable escalation sets us back to square one, or worse. South Africa, days away from a local election already marked by killings, violence and troubling repeat patterns of escalation and conflict rigidity, shows how the same violence changes form rather than stopping. In conflict studies, this is the increasingly clear problem of recurrence.
If peace is an interval, then both our theory and our practice rest on the wrong assumption. Our politicians will answer my questions here in a politically sensible manner: vote, it’s the way we change things. They may even throw in a cheery “This time will be different, we promise”. They may even be right, this time. Philosophically, even theologically, recurring conflict, even violence is the way of the world, the way things are built.
In my own consulting and written work I advocate for not less conflict, but better conflict. But, as this is conflict analysis and not a political speech, we need to notice an important distinction, especially in our South African political landscape. Conflict itself can be an asset.
If ever we get our coalition conflict skillsets correct, it will be the rocket fuel that gets us to where we need to be. Constructive, skilled conflict management is a sine qua non for the modern leader, whether of the political or any other stripe. Wise, skilled conflicts are the bridges to prosperity, progress and stability. But poorly, unskilled conflicts, leadership failures and partisan selfishness leads to another pathology in our conflicts.
As we can see writ large all over our political landscape, mismanaged conflict causes harmful, repetitive cycles of destructive conflict, a hardening of the willingness and ability to work together, a poisonous scepticism as to the motives and abilities of the Other. We start flailing about, blaming, scapegoating, and soon the migrant, the other political party, the government, our neighbour becomes the reason for it all, the solution to it all. Radical alternatives, sold to us by the ignorant and the malicious, become increasingly reasonable and acceptable.
How do we notice these patterns, how do we escape them? How do we change our corrosive conflicts to building blocks?
1. A hardly-noticed problem without a denominator
On 13 October, four members of the ANC Youth League were shot dead in Katlehong, in Ekurhuleni. Two days later, two ANC volunteers were killed while canvassing door to door in Umgababa, south of Durban, and a third was wounded. Six dead in three days, with local government elections due on 4 November. Long lists of other deaths tell the same story.
The response followed a tired, uninspired script, one that we can easily miss because it has become so pervasive: condemnation, an assurance that police are pursuing the perpetrators, a call for political tolerance. The script is old because the problem is old. KwaZulu-Natal’s Premier has said that eighteen councillors in the province have been murdered since the 2021 local elections. Researchers at the Global Initiative Against Transnational Organized Crime have documented close to two hundred political assassination cases nationally since 2000, with a clear association between election periods and spikes in killing. Their count for 2026 stood, as of late August, at twenty-three incidents and twenty-six deaths across five provinces. Other monitors speak of more than forty suspected assassinations this year. A phenomenon that cannot be counted cannot be managed, as the management aphorism tells us, and that is the first lesson of this essay.
The second is that this is not a South African peculiarity, as we tend to accept. It is a local instance of the most under-appreciated fact in the study of organised violence and related conflict: violence that a society has formally put behind it keeps coming back, in the same places, among the same kinds of actors, around the same calendar.
As specialised conflict analysis, my argument has three parts. First, persistence is the central empirical fact of conflict research, and the field is built in ways that look past it.
Second, recurrence is best explained not by the purity or depth of identities, nor by anything intrinsic to peoples, but by the survival of organizations, the structure of incentives, and the decay of institutions: it is a problem for economics, history and the epidemiology of violence, not for the psychology of in-groups. Third, if most peace is an interval, with a hazard that can be estimated and a calendar of events that we can anticipate and track, then practice should be organised around maintenance rather than resolution.
A word on what this is not. It is not an argument for fatalism, and the phrase “perpetual unpeace” is a wonderfully descriptive conflict studies concept, not a finding. The evidence shows risk that is high, front-loaded, and declining. The task is to take that curve seriously.
2. Persistence is the fact
The best-known statistic in the field is that roughly four in ten post-conflict countries relapse within a decade. It comes from the work of Paul Collier and colleagues on the “conflict trap,” and it has acquired the status of folklore. It is also a good example of how fragile the field’s own facts are. Suhrke and Samset showed that figures of this kind depend heavily on how a war is deemed to have ended, on which dataset is used, and on what counts as a new war rather than a continuation of the old one. Relapse estimates move materially with coding decisions. For our purposes, some of the settled wisdoms of how peace is arrived at, and how that peace is then maintained, are under serious revision. The definitional problem runs deeper than any one study.
The major datasets define conflict by thresholds: the Uppsala data count twenty-five battle-related deaths in a year, the Correlates of War a thousand. A conflict “ends” statistically when it falls beneath the line and “recurs” when it crosses it again. Violence that persists below the line, which describes a great deal of what ordinary people experience, is invisible to the headline statistic. In South Africa, our crime statistics read like a badly scripted horror movie, and even then we doubt whether the tape tells the story in its full extent. A country that settles into a steady state of twenty killings a year is, to the dataset, at peace. Still, the underlying pattern survives every coding choice. The World Development Report 2011, drawing on Barbara Walter’s work, found a reversal in the geography of civil war: in the 1960s most new civil wars broke out in countries with no recent history of one, whereas by the 2000s close to nine in ten occurred in countries that had experienced civil war within the previous thirty years.
Walter’s earlier finding, that prior war is a weaker predictor than the political and economic conditions a country is in, is a reminder that the story is not purely mechanical. But the direction of travel is clear. The world’s wars are increasingly the same wars. Two qualifications keep the claim honest. First, the hazard declines. Work by Quinn, Mason and Gurses on civil war recurrence shows risk concentrated in the first years after a war and falling with time; peace, once it has lasted, tends to become more durable. Second, outcome matters.
Negotiated settlements have often recurred more frequently than decisive victories, though the comparison is confounded by selection, since the wars that end in negotiated settlement are those in which neither side could win. Nobody should draw from this that victory is preferable, a conclusion with obvious moral and practical costs. The right picture, then, is a risk curve, not a verdict that simply ties us into further harmful cycles of conflict avoidance or escalations. It is steepest immediately after the guns fall quiet and flattens with the passage of years, provided the conditions that sustain violence are being addressed. “Perpetual” is the wrong word. “Interval” is the right one.
It also helps to distinguish three phenomena that the single word “recurrence” blurs.
Relapse is the return of the same war between the same parties over the same issue.
Transformation is the mutation of one kind of violence into another: a political war becoming a criminal economy, as in the long afterlives of Mozambique’s Renamo conflict, Colombia’s demobilizations, or Ethiopia’s Pretoria agreement.
Chronic violence is the failure ever to fall below a tolerable threshold at all. Most of South Africa’s experience falls into the second and third categories. Describing it as a risk of civil war is lazy; describing it as the persistence of violent institutions is more accurate and more useful.
3. Four mechanisms, ranked by evidence
Why does the interval end? The literature offers many explanations. Four carry most of the weight for the South African experience, they are observable from the sidelines, and they differ in how well they stand up.
The commitment problem
The strongest theoretical account remains Barbara Walter’s, building on James Fearon’s bargaining models. A peace settlement asks armed parties to disarm and trust in promises that no one can enforce. Whoever disarms first becomes vulnerable; whoever keeps their arms signals that they do not intend to honour the deal. Settlements therefore tend to hold when a credible third party guarantees them. Page Fortna’s work on peacekeeping is the empirical counterpart, and her findings are among the most robust in the field: peacekeeping substantially reduces the risk that war resumes, in some estimates by more than half. The lesson is that settlements are not self-enforcing contracts but promises that require enforcement infrastructure.
The survival of organizations
The second mechanism is the one most neglected in public discussion. Wars create organisations: armies, militias, hit squads, smuggling networks, protection rackets. Peace agreements settle the grievances in whose name those organisations fought, but not the organisations themselves. Stephen Stedman’s classic work on “spoilers” made the point for those who reject settlements; the broader literature on war economies (David Keen on Sierra Leone is the foundational case) shows how violence becomes a livelihood, with its own revenue streams and its own interest in continuation.
Alex de Waal’s notion of the “political marketplace” generalizes this: in many fragile states, loyalty is purchased, and violence is a bargaining instrument rather than a breakdown of bargaining. Demobilising individuals is not the same as dismantling organisations. Humphreys and Weinstein’s study of Sierra Leone found that formal reintegration programmes had little measurable effect on how ex-combatants fared, and that what mattered was what they had done and endured during the war and the communities to which they returned. This suggests that the reintegration industry has been aimed at the wrong unit.
The economic and human-capital legacy
Here the intuition in the conventional story, that war plants the next war through poverty, broken schooling and lost trust, needs more care than it usually gets. Low income and slow growth are among the most robust correlates of civil war risk. But Blattman and Miguel’s review of the field is a salutary caution: the causal channels are disputed, the opportunity-cost story that poor young men fight because fighting pays better is more plausible in some cases than others, and the physical economic damage of war is often recovered faster than expected.
Studies of heavy bombing in Japan and Vietnam found limited long-run effects on local economic development. What does persist is damage to people and institutions: the health legacy that Ghobarah, Huth and Russett documented, years of lost schooling, and the long-run erosion of trust in the state. The lasting wound is less to capital than to the capacity of a society to coordinate peacefully.
Institutions and trust
The fourth mechanism we find at the level of political order. Our comparative conflict study now comes home. Douglass North, John Wallis and Barry Weingast argue that most states are “limited access orders,” in which elites keep violence in check by guaranteeing each other privileged access to various privileges and benefits. This produces peace of a kind, but a fragile one, since any shock to the distribution of benefits becomes a threat to the bargain itself. Where the state cannot reliably punish violence, rivals update their beliefs about what is permitted, and the cost of the next killing falls.
We have been tracking that in South Africa for decades now. Stathis Kalyvas’s work on the logic of violence adds that killing is rarely random: it is a tool of local actors seeking to control territory, resources and information. Impunity is therefore not a side effect of recurrence but one of its drivers.
These are not rival theories so much as layers. The commitment problem explains why settlements fail at the moment of signing; organizational survival explains why they erode afterwards; human-capital damage and institutional decay explain why the soil remains fertile.
4. The epidemiology of violence
It is no accident that public health language has become common in this literature. It forces disciplines of measurement that political science and even conflict studies have often avoided: incidence and prevalence, denominators, hazard rates, exposure, and the distinction between a conflict source and a trigger. In those terms, the reservoir of violence, the source, is the stock of armed organisations and weapons that survive the end of a war. The vectors are the networks, whether kin, patronage, criminal or partisan, through which violence is organised and recruits are drawn. Susceptibility is concentrated among young men with poor economic prospects and weak state protection.
Triggers are events that make the use of violence newly rewarding: contested elections, disputed appointments, the award of a large contract. The empirical record supports parts of this analogy and complicates others. Violence clusters in space and time; Weidmann and Ward’s spatial models of Bosnia show that conflict risk at one location is raised by violence nearby. But Buhaug and Gleditsch’s well-known paper “Contagion or Confusion?” cautions that much of what looks like diffusion is really shared structure: neighbouring places experience violence together because they share the same poverty, terrain, ethnic geography and weak institutions, not necessarily because violence passes from one to the other.
Both processes are probably operating, and the relative weights matter for policy. The intervention evidence is similarly mixed, and it should temper the enthusiasm of the metaphor. Programmes that treat violence as a transmissible condition, such as the Cure Violence model of using credible community “interrupters,” have produced contested results across evaluations. By contrast, a Liberian trial by Blattman and colleagues, combining behavioural therapy with a modest cash grant for high-risk young men, found measurable reductions in crime and violence.
The more modest conclusion is that targeted, individual-level interventions can work, but they treat the host rather than the reservoir. The analogy also has limits, as they all do. Violence is not a pathogen. It is chosen, organised, financed and defended. The value of the epidemiological frame is not that it excuses anyone, but that it imposes the discipline of counting, of risk windows and of attention to what sustains transmission. The violence we see in South Africa is inevitable, even at the existing levels, once we understand these underlying mechanisms.
5. South Africa: transformation, not relapse
South Africa is not in a civil war, and in important respects its transition remains among the great political achievements of the late twentieth century. That is precisely why it is an instructive case. But the “rainbow nation” has not avoided its most important conflicts, it has deferred it, it has covered it in promises and platitudes, and it will need its leaders to lift it from conflict management to resolution, or even transformation. The violence that accompanied the transition was concentrated in KwaZulu-Natal and what is now Gauteng.
Over the final decade of apartheid, by common estimates well over ten thousand people were killed in the province; the so-called Seven Days War in the Natal Midlands in March 1990 alone claimed roughly two hundred lives. The Goldstone Commission, which sat from 1991 to 1994, documented the involvement of elements of the security establishment in fomenting the violence, and the Truth and Reconciliation Commission later added to the record. The 1994 elections were, in part, a mechanism for demobilizing that war, and in the national political sense they succeeded.
But the settlement operated at the level of elites and parties. It did not, and in its nature could not, dismantle the local armed networks, the weapons stocks, or the informal command structures that had grown up in the preceding decade. A plausible reading, and one consistent with how the Moerane Commission, appointed in 2016 under Judge Marumo Moerane to examine killings in the province, understood the problem, is that the organisational capital of violence found new employment. Competition for positions in local government, with the jobs, tenders, land and gatekeeping power attached to them, became the principal arena. Analysts of the current wave routinely stress that political competition overlaps with organised crime, the taxi industry, and fights over contracts and public funds.
The taxi industry’s violent origins in the deregulation of the 1980s, the rise of extortion and “construction mafia” networks, and the growth of a market for contract killings supplied both the vectors and the demand. Our Commissions seem to run on these as central, necessary narratives. The calendar matters, as I indicated earlier. Local elections are, nationally, lower-stakes affairs, but locally they are high-stakes: a ward seat is a source of income and a gate through which resources pass.
Understand this, and much of the Battle for Joburg makes a deeper, more manageable sense. The list-nomination period, in particular, concentrates the risk, because the contest is often not between parties but within them. The association the Global Initiative’s researchers report between election periods and rising assassinations is what the hazard framework would predict. It is here that the institutional mechanisms bite. Acting Police Minister Firoz Cachalia admitted publicly, in June, to being angry that suspects keep escaping justice.
A political killings task team built in the aftermath of the Moerane Commission was dissolved in early 2025, months before a known risk window, and a task team is again at work. Whatever the explanation, the episode illustrates a general point: the institutional capacity that lowers the hazard is built slowly and can be dismantled quickly, and the cost of removal is borne after a lag that makes it hard to attribute. Researchers also worry about desensitization. When killings of councillors, candidates and activists become routine and attract limited attention, the political cost of ordering the next one falls.
Two cautions are important, not out of politeness but in order to be effective if ever we are going to deal with these scourges. First, this is not the pathology of any one party. The victims in the current wave include members of the ANC, the IFP, the EFF and civic organisations, and the perpetrators, where they are identified at all, are frequently entangled in motives that are personal, criminal and commercial as well as political. The case is structural and cuts across partisan lines. Second, the broader repertoire of South African collective violence, which includes the July 2021 unrest with more than three hundred deaths, the waves of xenophobic attacks in 2008, 2015 and 2019, and the policing tragedy at Marikana, should not be collapsed into a single story.
Each has its own causes, and the temptation to read them as manifestations of one underlying disease should be resisted. Conflict management, at its best, sees complex conflicts like these as needing a multi-disciplinary approach to properly understand and address. What they share is not a cause but a condition: a society in which the capacity to mobilise violence is distributed, the state’s monopoly is partial, and the price of using force is, too often, low. The risk of violent solutions are deemed negligible against the risk of capture and / or effective punishment.
The lesson for theory is uncomfortable. By the standard of the transition literature, South Africa’s settlement is a success. Judged at the level of the headline conflict, it was. Judged at the level of the organisations and incentives beneath that conflict, the interval is still running, and not running on an endless loop, but running in escalating levels that show, again, very predictable levels of harmful escalation that should wave red flags at any responsible and competent political leader.
6. Why theory and forecasting misfire
If recurrence is this central, why has the field been slow to build around it, if not always at institutional level, at least in a range of South African categories, such as political and security leadership? Part of the answer is in the structure of the research agenda. The dominant debates of the past three decades, greed versus grievance, ethnic fractionalization versus horizontal inequality, have been debates about onset: why does war begin?
These are important questions, but they treat the end of a war as the end of the story. The empirical regularity that most wars occur in places that have recently had one is treated as a control variable, a nuisance to be netted out, rather than the main finding. The global experience and data on regional or intra-regional wars translate well to the South African internal conflict and violence experienced. Part of it is vocabulary. “Resolution,” “settlement,” “termination” and “post-conflict” all imply a closing of the problem. Peace agreement datasets code agreements as ends, and the policy world (not to mention the election and news cycles) seeks and rewards closure.
And part of it is the character of forecasting. Cederman and Weidmann, writing in Science, urged the field to adjust its expectations: armed conflict prediction is dominated by the persistence of the existing state, and models that look impressive often owe most of their performance to the simple fact that places with violence today tend to have it tomorrow. Early-warning systems such as ViEWS are strong on persistence and much weaker on first onsets in quiet places. There are signs of progress. Mueller and Rauh have shown that newspaper text can improve onset prediction in peaceful countries. But the broad picture stands.
The best predictor of violence is violence, which is simultaneously the field’s most reliable result and its most embarrassing. There is a constructive reading of this. If risk is autocorrelated, then the right default is a model in which the hazard is the object of study, with organisations and localities as the units, and time since the last episode as the central variable. Subnational data, long panels, and measurement of violence below the conventional thresholds would become core instruments rather than supplements. A major work that took recurrence, rather than onset, as its organising question, and asked what the economics, history and epidemiology of repeated violence imply for method, would discipline the field in a useful way. This is where conflict studies should get up from behind their desks and join security and policing on the streets.
7. Peace as maintenance
If most peace is an interval, what should practice look like? I have written on this previously, for other political events, and just reaching this stage again in itself is instructive to the harmful patterns I want to highlight. Again, as conflict analysis, and not political commentary, I would suggest the following seven practical strategies to assist in making our world a bit easier to understand and manage:
(i) Measure time at risk, not signatures: An agreement or an election result is a point of departure, not a goal in itself. We need to stop creating and then perpetuating these cycles of hope and despair, and have a more graduated approach to what works and when. The hard yards, the measuring, the rewards and the punishment must happen on a spectrum, not on these election cycles. The relevant metrics are the hazard rate, the time since the last significant episode, and the trajectory of sub-threshold violence. Success should be reported as a survival curve, throughout, not at election times. Political parties should be held to manifestos and promises, by civil society, activists and various organizations. Society needs to tell politicians where they are going, not the other way around.
(ii) Dismantle organizations and their finance, not only their fighters: The central question for any demobilization and ongoing violence or public conflict cause and effect must be what the organization’s next best employer is. If the answer is the extortion economy, the protection racket or the political patron, then the organization has not ended; it has changed its business model.
(iii) Treat the calendar as a forecast this is available, predictable information: As we can see, conflict studies make these events rather predictable. Elections, nominations, budget cycles, large tender awards and leadership contests are predictable risk windows. Protection, investigative capacity and monitoring should be positioned before them, not in response to them.
(iv) Make counting a public good, a habit: Consistent, independent, publicly available data with stable definitions is infrastructure for peace, as essential as a court or a police station. A state that cannot say how many of its citizens have been killed for political reasons is in no position to say whether things are improving. Measure it, manage it. And statistics should stop being a tweet or a few seconds on a news bulletin. It is a starting point, not a nail in the coffin of hope or action.
(v) Fund years three to ten: Again, statistics and comparative studies can help us here. Donor and political attention decays faster than the hazard. The period in which risk remains meaningfully elevated is also the period in which the agreement has long dropped out of the news. We are quick on the promise and slow on the monitoring. We also forget, or give up on, the third phase: consequence.
(vi) Raise the price of violence for those who commission it: In the South African case, the arrest of assassins and other direct actors is necessary but insufficient. What matters is the prosecution of those who order and pay, those who causally benefit, because that is what changes the expected cost to the next person considering it. Witness protection, specialised investigative units and credible convictions are what make the interval longer.
(vii) Change the language: Words like “managed,” “risk-adjusted,” and “interval” are not defeatist. They are accurate, and accuracy is a precondition for allocating resources sensibly. That is the focus of this essay. How do we understand and prolong that interval? Our politicians and law enforcement leaders are often clearly pressurized to tell us the tales of eternal solutions, of home runs and the once-and-for-all fix. This becomes its own conflict cause, and it leads to increased scepticism and radicalization.
There is an obvious objection. To tell policymakers, or I suppose even the public, that peace is only an interval may seem to license cynicism, or to provide an alibi for failure. The response is that the argument runs the other way. The data show that the interval can be lengthened, that hazards decline with time, and that specific interventions, enforcement guarantees, targeted programs and credible punishment, measurably reduce risk.
A view of peace as maintenance is the opposite of fatalism. It is the claim that peace is built from continuing effort and that effort can be measured. It should also, when properly understood, affect our political rhetoric and strategizing in meaningful, more honest ways.
Conclusion
Together with, or even earlier than the return of geopolitical realism, conflict studies have been dragged out of their own rules-based ivory tower into the streets where they so confidently advised on. A state’s internal conflict, which includes violence and the narratives that drive and sustain it, can benefit from that maturity and renewed focus. And it would end, as this essay does, with practice.
The four men shot in Katlehong and the two volunteers killed in Umgababa were doing the ordinary work of a democracy: asking strangers for their vote. Their deaths are not an aberration within a peaceful society. They are a point on a hazard curve that has been running, with variations, for nearly forty years. We put those events in the wrong boxes, and we end up making the wrong promises, working on the wrong strategies.
A peace that has to be maintained is not a lesser peace. It is the only kind that has ever existed, and the first condition for keeping it is to stop pretending otherwise.
Summary of main sources, references and suggested reading Our blog at www.conflict-conversations.co.za has a rich variety of articles expanding on specific aspects of the article, and conflict in general.
Books and other sources:
(Andre Vlok can be contacted at andre@conflict1.co.za for any further information.) (c) Andre Vlok October 2026 * Author’s note on the use of artificial intelligence in writing this article I learned to draft, argue and write in the hard school of litigation. I enjoy and value the very human process of creating ideas, of testing my own knowledge and thoughts. It is a process that I need, for answering some of my professional and even personal questions, it is cathartic and inspiring. Other than the most basic research assistance I do not use any AI in the creation of my written work, this article included. It is a matter of pride, of preference, and of mental health. Whether that is a wise choice or not, I will leave to the reader to decide.