Every year, development organizations publish glossy reports filled with success stories. A microfinance program that lifted a village out of poverty. A health intervention that slashed child mortality. A school-building initiative that transformed educational outcomes. These narratives raise funds, win awards, and shape how the entire sector thinks about what works.

But here's the puzzle: if we have so many proven successes, why does progress remain so uneven? Why do programs that worked brilliantly in one country collapse when replicated elsewhere? The answer lies not in the interventions themselves, but in how we select, narrate, and learn from our experiences.

When we subject celebrated development wins to rigorous scrutiny—controlling for external factors, examining long-term outcomes, accounting for what we don't see—many prove far less impressive than their reputations suggest. This isn't cynicism. It's the foundation of honest learning. And until the development sector gets better at it, we'll keep recycling approaches that look good on paper while missing interventions that actually move the needle.

Imagine a donor funds twenty maternal health programs across different regions. Seventeen produce modest or negligible results. Two fail outright. One achieves dramatic improvements in outcomes. Which program gets featured at the annual conference? Which one becomes a case study taught in development courses? The single standout success becomes the story—and the other nineteen effectively vanish.

This is classic survivorship bias, and it distorts development learning in profound ways. We study winners and reverse-engineer their strategies, assuming we've found a formula. But without examining the full distribution of outcomes—including the programs that used identical strategies and failed—we can't distinguish genuine effectiveness from luck, favorable conditions, or statistical noise.

The problem compounds over time. Success stories attract imitation. Donors fund replications of the celebrated model. When those replications underperform, they're quietly shelved rather than published. The original success story remains untarnished, and the sector loses another opportunity to understand what actually drives outcomes versus what merely correlates with them in favorable circumstances.

Development economist Lant Pritchett has called this the problem of "small n, big conclusions." We draw sweeping lessons from tiny, unrepresentative samples. Rigorous evidence requires examining the full population of attempts—successes, failures, and everything in between. Until reporting norms change to incentivize transparency about the complete picture, our collective knowledge base will remain dangerously skewed toward optimism.

Takeaway

A success story only becomes evidence when you can also see everything that didn't succeed using the same approach. Without the full distribution of outcomes, you're not learning—you're cherry-picking.

Bangladesh is often cited as a development miracle. Between 1990 and 2020, life expectancy rose dramatically, fertility rates fell, and poverty declined faster than almost anywhere on Earth. Dozens of organizations claim credit: BRAC for its community health workers, the Grameen Bank for microfinance, various NGOs for girls' education programs. But Bangladesh also experienced massive remittance inflows, a booming garment industry, favorable demographic shifts, and significant government investment in infrastructure.

Untangling which intervention caused which outcome is extraordinarily difficult—and success narratives rarely even try. They present a program timeline alongside an outcome trend and let the reader assume causation. This is the development sector's most persistent analytical sin. Correlation is not just confused with causation; it's actively marketed as causation to justify continued funding.

Randomized controlled trials have helped in some domains, but they have limits. They work well for discrete interventions—deworming tablets, bed nets, cash transfers—but struggle with complex, systemic changes like governance reform or economic transformation. For these bigger questions, we're often left with before-and-after comparisons that can't account for everything else that changed simultaneously.

Honest attribution requires counterfactual thinking: what would have happened without the intervention? This is uncomfortable because the answer is often "something similar, just slower." Many celebrated improvements in health, education, and income were part of broader secular trends driven by economic growth, urbanization, and technology diffusion. That doesn't mean specific programs were worthless—but it means their marginal contribution was likely smaller than their marketing materials suggest.

Takeaway

Before crediting any program with a success, ask the counterfactual: what would have happened anyway? The honest answer is usually that the intervention's contribution was real but far more modest than claimed.

In medicine, negative trial results get published. A drug that doesn't work is a finding worth sharing because it prevents others from wasting resources on the same dead end. In development, negative results are buried. Failed programs are rebranded, restructured, or simply not discussed. The institutional incentives are clear: donors don't want to fund organizations that admit failure, and staff don't want careers associated with unsuccessful projects.

This creates an enormous knowledge destruction problem. Consider the thousands of agricultural extension programs, community-driven development projects, and vocational training initiatives that have been tried worldwide. Most produced disappointing results. But because those results were never systematically documented and shared, the same mistakes get repeated decade after decade, country after country. Each new program designer starts nearly from scratch.

Some organizations have experimented with failure reports. Engineers Without Borders Canada published annual failure reports for several years, documenting what went wrong and why. Admitting Aid, a movement in the UK, pushed for greater transparency about ineffective programs. These efforts generated attention but didn't fundamentally change sector norms. The incentive structures—competitive funding, reputational risk, career consequences—remain stacked against honesty.

Building a genuine learning culture requires structural change, not just cultural exhortation. Funding mechanisms need to reward rigorous evaluation regardless of findings. Publication norms need to value null results. Career incentives need to recognize that identifying what doesn't work is as valuable as identifying what does. Until then, the development sector will continue operating with a systematically distorted evidence base—and wondering why so many proven approaches fail to deliver on their promise.

Takeaway

An industry that only documents its wins is an industry that cannot learn. The most valuable development knowledge may be sitting in the unpublished evaluations of programs that didn't work.

None of this means development doesn't work. Enormous progress has been made in global health, education, and poverty reduction over the past several decades. But our ability to understand that progress—to know precisely which interventions drove which outcomes—is far weaker than the confidence of our success narratives suggests.

Better development practice starts with better development honesty. That means publishing full outcome distributions, demanding rigorous counterfactual analysis, and treating failure documentation as a professional contribution rather than a career risk.

The goal isn't pessimism. It's calibrated optimism—knowing what we actually know, admitting what we don't, and building the evidence base that lets us direct scarce resources toward interventions that genuinely change lives.