Why Buildings Fall Down
The structural causes, the failure mechanisms, and what the evidence says actually works
Roshan Kishore describes what building collapse looks like from inside the informal city. The same dynamic, in slightly different institutional clothing, killed 50,000 people in Turkey in 2023 and 1,134 in Bangladesh in 2013. The causes are consistent enough across countries that a checklist is now possible.
Buildings collapse for a finite set of reasons. The engineering literature is consistent across geographies. Analysis across multiple settings finds poor construction materials account for roughly 33 percent of failures, human error for about 21 percent, and natural disasters for around 12 percent, with a further 34 percent attributable to combinations of all three. That last figure matters most: catastrophic collapses are typically multi-factor events, which is why single-point interventions underperform.
The specific mechanisms are well-documented. In many low- and middle-income contexts, collapse has been attributed to weak foundations, substandard materials, poor material mixing, excessive loading relative to structural capacity, and absent strength testing. Beyond materials, structural failures from poor supervision and workmanship, faulty design, and disregard for approved drawings are documented alongside the absence of soil testing, weak coordination between professional bodies and planning authorities, and inadequate enforcement of development laws.
Natural hazards add a multiplier. Earthquakes, ground subsidence, sinkholes, fire, and flooding that undermines foundations each generate structural failure pathways, with damages typically worse for old or poorly maintained buildings. The hazard itself rarely determines the death toll. What determines it is whether the building stock was designed and built to withstand the hazard it was exposed to.
Two cases make this concrete.
The 2013 Rana Plaza collapse in Bangladesh killed more than 1,100 people. The government’s own investigating committee found defects extending from the swampy ground the building was constructed on, to extremely poor quality materials, to vibration from heavy rooftop generators running at the moment of collapse. The owner had obtained a permit for a five-story structure but added three floors without authorization, and the building was constructed on a filled-in pond. Workers reported severe cracks the day before the collapse and were told the building was safe. This was a multi-factor failure, at every stage, with the information to prevent it available and ignored.
The 2023 Kahramanmaraş earthquake in Turkey operates at a different scale but the same logic. Around 50,000 people died and approximately 100,000 were injured, with damage costs estimated at $84 billion for Turkey alone. The earthquake was large. The death toll reflected a building stock eroded by decades of code non-compliance. At least 75,000 buildings in the affected zone had received construction amnesties, legal exemptions allowing developers to bypass safety certification by paying a fee, with estimates of non-compliant structures across Turkey running into the tens of millions. A structural shift in inspection arrangements had also placed private inspection companies, hired and paid by the same contractors they were inspecting, in charge of approvals, effectively removing independent oversight from the process.
The contrast with Japan is instructive precisely because the seismic hazard is comparable. After major earthquakes, Japan progressively tightened its building standards and tracked compliance. Surveys after the 1995 Great Hanshin-Awaji earthquake and the 2016 Kumamoto quakes found a collapse rate of roughly 28 percent for wooden buildings under the pre-1981 standards, compared with about 9 percent under the 1981 standard, and only around 2 percent for those built after the 2000 revision. By 2023, Japan’s national housing earthquake-resistance rate had reached roughly 90 percent, up from 79 percent in 2008. That trajectory took four decades. The mechanism was mandatory certification at point of construction, tracked by a public registry, with liability consequences for non-compliance.
What the permit system actually does
In most cities, the building approval system involves submitting drawings, securing no-objection certificates from multiple authorities, surviving in-person inspections across departments, and navigating queues that, regardless of official timelines, can take months. Each checkpoint is an opportunity for discretion. Each inspector has leverage. The rent extracted at these points is not incidental to system function. It fills gaps that formal municipal finance doesn’t cover. Urban local bodies in many countries are structurally underfunded, and the discretionary approval process is partly how they operate.
This has a predictable distributional consequence. The formal permit system prices out exactly the buildings it’s supposed to regulate. People build anyway, in the only configurations they can afford, without approvals, without inspections, and without records. The buildings that collapse are overwhelmingly these. The permit system, as currently designed in most cities, doesn’t regulate the informal stock. It creates it.
Two Indian states that redesigned the system
Two Indian states have run a credible natural experiment on what happens when you strip the discretion out of the approval process.
Telangana passed the TS-bPASS Act in 2020. For plots up to 75 square yards with ground or ground-plus-one-floor residential construction, no building permission or occupancy certificate is required at all. For plots up to 500 square meters with height up to 10 meters, approval is instant, granted online on submission of self-certification. Larger applications must clear a 21-day window, with deemed approval on deadline breach. The critical design choice was removing the official’s gate entirely. No meeting, no relationship, no file to move. Over 25,000 applications were filed in the first seven months of the system’s operation, with a disposal rate of around 80 to 85 percent. Officials attributed the throughput specifically to post-verification officers not holding up files, because the system didn’t give them a hold-up mechanism.
Andhra Pradesh built a parallel system across 110 urban local bodies. The government and the World Bank designed an AI-based digital platform for building permit provision that used drones, machine learning, and GIS mapping to update property databases and sanction permissions entirely online without human intervention. Applications for a trade license, water connection, or property title transfer are processed within 15 days. A shop registration application post-2019 is processed in 15 minutes. Revenue from property taxes and water charges more than doubled from 2015 to 2019, and property tax revenues increased 30 percent due to previously unassessed properties being brought onto the register. The fiscal gain from formalizing the informal was larger than the rent extracted through the approval queue.
Both systems work. Neither is a pilot. The technology is not the constraint.
Why these two states and not most others
The deeper question is why two Telugu states moved and most places haven’t. The answer sits in political economy, not governance culture or administrative capacity.
From the late 1990s, Chandrababu Naidu built Hyderabad’s economic identity around attracting IT investment. That bet required regulatory credibility as a signal to companies making location decisions between Hyderabad, Bengaluru, and Pune. When Telangana separated in 2014, the KCR government inherited both the asset and the logic. TS-bPASS was introduced precisely as Hyderabad’s real estate market was drawing national and international capital, and the government needed to sustain investor confidence. The building safety benefit was a byproduct. The primary driver was that a large, economically organized constituency, real estate developers, institutional investors, and IT employers making location decisions, found transparent approvals more valuable than negotiated ones.
The Andhra Pradesh case adds one more variable: the fiscal argument. Once the AP system brought unassessed properties onto the tax register and doubled revenue collection, the government had a financial incentive to sustain and expand the reform. Permit simplification and property tax digitization turned out to reinforce each other. Jurisdictions with severe municipal finance constraints have a stronger economic case for this combination than is usually recognized.
What the evidence says actually works
The empirical record across countries points to a consistent set of interventions. Their effectiveness depends heavily on sequencing and the institutional conditions in which they operate.
Codes must reflect actual hazard exposure. A code that doesn’t incorporate current seismic, flood, or wind hazard maps for the zones where it applies is not functional. Japan’s iterative code revisions after each major earthquake are the reference. Turkey’s 2018 Turkish Building Earthquake Code introduced rigorous design criteria for new buildings and retrofitting, but the 2023 earthquake exposed the gap between written standards and actual compliance in the existing stock. The code and the stock are two different problems.
Amnesty programs are a slow-motion catastrophe. Construction amnesties, which allowed non-compliant buildings to be legalized by fee payment, eroded the margin of safety needed to avoid building collapse in Turkey, with the death toll of roughly 50,000 reflecting that erosion directly. Amnesties resolve a political problem, large volumes of informal stock that voters live in, by converting a safety compliance problem into a revenue transaction. The fiscal logic is obvious. The structural outcome is also obvious.
The existing stock problem is separate from new construction. Most cities with large informal populations already have most of their at-risk building stock standing. New construction rules apply only to what’s built after they take effect. Reducing risk in the existing stock requires either retrofit programs with sufficient subsidy to make compliance financially feasible for low-income owners, or managed redevelopment where the economics of the replacement structure fund the removal of the unsafe one. Japan’s retrofit trajectory took 40 years and required sustained public subsidy. There is no fast version of this.
Materials quality control operates upstream of the permit system. A building designed correctly and approved correctly but built with substandard concrete or under-dimensioned rebar still fails. Materials standards enforcement, including random on-site testing, is a complement to permit reform. After the Rana Plaza collapse, 97,000 of 132,000 hazards identified at factories in Bangladesh were eliminated through the Accord on Fire and Building Safety, specifically targeting poor construction and construction without permits. The mechanism that made this work was external audit with purchasing-power consequences, which is an important qualification about what functions when domestic regulatory capacity is the binding constraint.
The enforcement gap: why adopting a better code often changes nothing
The standard post-disaster policy response is to upgrade the code. After a major earthquake, the political cycle is predictable: commission a technical review, announce adoption of more demanding standards, often modelled on systems from high-income countries, and declare the problem addressed. Turkey had, on paper, construction codes that met current earthquake engineering standards before the 2023 disaster. They were too rarely enforced, which is why thousands of buildings crumbled. As David Alexander of University College London put it at the time: “This is a disaster caused by shoddy construction, not by an earthquake.”
The code was fine. The problem was somewhere else entirely.
This pattern repeats. Building codes are in place in many disaster-prone countries across the globe. Large-scale deaths and property losses in recent natural disasters have nonetheless been largely attributed to non-compliance with those codes. Enforcement has remained the major challenge, particularly in developing countries. Turkey’s own Ministry of Environment and Urbanization had acknowledged before the 2023 earthquake that more than half of all buildings in the country, accounting for around 13 million apartments, were not in compliance with current standards. The ministry knew. The code existed. Neither fact prevented the collapse.
Why enforcement fails structurally
The enforcement gap has three distinct causes, and they require different responses. Conflating them produces interventions that address one while leaving the others intact.
The first is capacity. Most urban local bodies in low- and middle-income countries do not have enough trained inspectors to review a construction boom of the scale their cities are experiencing. The inspector-to-working population ratio in developing countries is often three to ten times lower than in European countries, and that gap doesn’t account for the quality of enforcement when inspectors do show up. In South Africa, one recent analysis found that in Limpopo province, just 121 health inspectors serve a province that requires 657, and construction site inspectors face comparable shortfalls, with essential functions such as enforcing approved plans and occupational safety standards left uncovered. When the inspection workforce is this thin relative to the construction volume, discretion about which sites to visit creates its own rent-extraction opportunity, and the riskiest buildings, the ones without organized constituencies, tend to get visited last.
The second cause is incentive misalignment at the point of inspection. An inspector who visits a site, identifies violations, and issues a stop-work order faces immediate friction: the developer objects, delays accumulate, local officials receive complaints, and the inspector bears the cost of the confrontation while the systemic benefit is diffuse and delayed. Where this dynamic is not countered by strong institutional support, clear liability, and protection from retaliation, inspectors learn to look the other way. Turkey’s arrangement, where private inspection companies hired and paid by contractors assumed full responsibility for approvals, eliminating the previous need for municipality involvement, is the extreme version of this: the inspector’s income is directly contingent on not finding problems.
The third cause is unmeasured stock. You can’t enforce compliance you can’t see. In most rapidly urbanizing cities, a large share of the building stock has no permit record, no as-built drawing, and no inspection history. An inspector arriving at such a site has no baseline against which to assess violation. Non-compliance with building codes has been significantly attributed to disasters in Bangladesh and Nepal, and in both countries, gaps in the coverage of formal records mean that violation rates are likely substantially underreported.
What actually works
Three enforcement approaches have evidence behind them. They aren’t alternatives to each other; the strongest systems combine all three.
Risk-based prioritization. Inspecting everything with equal frequency is neither feasible nor optimal. The AP system’s use of GIS mapping and drone-based property databases offers a model. Drones were used to update property databases and undertake GIS mapping to update existing town plans, with the AI platform then sanctioning building permissions and flagging irregularities without human intervention. Applied to enforcement rather than permitting, the same geospatial record allows an authority to triage its inspection workforce toward structures meeting defined risk criteria: age of construction, height, occupancy type, proximity to a fault line, history of previous violations. This is how scarce inspection capacity gets concentrated where it matters most, rather than spread uniformly across a construction universe no inspection force could cover.
Third-party audit with real liability. The self-certification model in Telangana and Andhra Pradesh delegates inspection responsibility to licensed architects and engineers, but attaches personal financial and professional liability to that delegation. Under TS-bPASS, the onus to ensure authenticity rests with the applicant, who is held personally accountable and liable for false declarations. This works if, and only if, professional licensing bodies function and sanctions are real. Where they don’t, the reform outsources rent-seeking to a different intermediary. The question to ask before adopting this model is concrete: in the last five years, how many licensed architects or engineers in this jurisdiction have lost their license, faced fines, or been prosecuted for certifying non-compliant construction? If the answer is zero, the accountability infrastructure doesn’t exist yet.
Occupancy certificates tied to consequences. Much of the enforcement problem lies at completion, not at permit issuance. A building can be approved on a compliant design and then built to a different specification. The occupancy certificate inspection is the point where as-built reality is supposed to be verified against permitted design, but in most jurisdictions it’s either skipped, cursory, or issued as a matter of form. Linking utility connections to a valid occupancy certificate changes the demand side of this equation. An owner who cannot legally connect water or electricity without a certificate has a financial reason to obtain one, which means the completion inspection actually happens. The AP digitization project built exactly this link: new applications for piped water supply were linked to the AI-based platform, so that service delivery was connected to the permit and compliance record.
The post-disaster policy trap
The reason upgrading the code is the default post-disaster response is worth naming directly. Code revision is technically sophisticated, internationally legible, and politically cost-free. It signals seriousness without threatening anyone’s current income. It produces a document that can be announced, published, and cited. Enforcement reform, by contrast, requires confronting the inspection workforce, the developer community, the municipal finance structure, and often the political network that operates through discretionary approvals. It’s harder, slower, and generates opposition. So governments reach for the code revision and call it reform.
A World Bank analysis of building energy codes across 88 countries found that while 71 countries had adopted mandatory codes in at least one major city, only 52 consistently enforced them, with the enforcement gap resulting in actual energy savings reaching only 30 to 40 percent of their potential impact. The same logic applies to structural safety: the gap between adoption and enforcement is where most of the risk lives.
A code that matches the hazard but can’t be enforced is a policy document. An enforceable inspection regime built around existing adequate standards, risk-based prioritization, third-party liability, and occupancy certificates tied to services, is the functional system. Most places that have had disasters already had an adequate code. What they need is the second thing.
The political economy must be activated, not assumed. The Telangana lesson, read carefully, identifies where leverage exists. The reform moved when a powerful economic constituency found transparent approvals more valuable than negotiated ones. In cities competing seriously for investment, that constituency exists and can be activated. The argument is financial: a city with predictable approvals and clean occupancy records attracts capital that a city with opaque approvals does not. The AP lesson adds the fiscal dimension: formalizing the informal building stock generates more tax revenue than the rents the current system extracts.
Neither mechanism is moral. Both are measurable.
Checklist for reform design
In rough order of dependency, since sequencing matters.
Map actual hazard exposure (seismic, flood, wind) at fine spatial resolution. A code calibrated to the wrong hazard level produces no safety gain.
Audit the existing building stock against current standards. The gap between code and stock is the actual risk profile, and it’s usually larger than official estimates assume.
Simplify the permit process to remove queue-based discretion, replacing it with defined timelines, online submission, and automatic deemed-approval on deadline breach. The goal is raising the share of construction that enters the formal system.
Assign self-certification to licensed professionals with personal financial liability. This requires functioning professional licensing bodies with real sanctions. Build those before delegating certification.
Integrate permit data with property tax records. The AP experience shows that formalizing construction generates fiscal gains that justify and sustain the reform. Jurisdictions running on thin municipal finances have a stronger case for this integration than is usually acknowledged.
Link occupancy certificates to utility connections. This converts a paper requirement into something with demand.
Ban construction amnesties. The Turkish data is conclusive.
Fund retrofit at the scale the stock requires. Subsidize upgrading for low-income owners who cannot finance compliance independently, or the reform excludes exactly the stock at highest risk.
Test materials on-site, randomly and often. Approval of a design does not guarantee approval of the inputs used to build it.
Make liability stick after collapse. Where prosecution for code violation following a collapse is consistently pursued, deterrence improves. Where it is abandoned due to political pressure, it does not.
The buildings will keep falling in places where the political calculation, investor confidence, fiscal incentive, or sustained public pressure, hasn’t yet made inaction more expensive than reform. When it does, the technology and the playbook are both ready.
(Opinions are personal)