Every accelerator brochure performs the same inference: our graduates raised more, grew faster, survived longer, therefore our program works. The inference has a hole a first-year statistician can name. Programs choose their cohorts, choosing is a skill, and a program that admits the ten best applicants from three hundred would show impressive graduate outcomes if its entire curriculum were a month of silence. The question that matters, for funders allocating scarce ecosystem money and for founders spending scarce months, is the counterfactual one: how much of the graduate’s trajectory is selection, the firm was already good, and how much is treatment, the program made it better? The research that takes this seriously, comparing admitted firms against near-miss applicants just below the cutoff, finds treatment effects that are real but selective: meaningful on financing access and network breadth, weaker and slower on revenue and survival, and heterogeneous across founder types. That pattern is not a scandal. It is a design brief, and programs that read it honestly build differently than programs that keep marketing the selection as if it were the treatment.
Key Takeaways
- Graduate outcomes conflate two forces: selection (picking firms already destined to perform) and treatment (changing their trajectory), and brochures systematically attribute the first to the second.
- Credible designs, near-miss comparisons and randomized admissions, find real but selective treatment effects: strongest on financing and networks, weaker on near-term revenue and survival.
- Selection skill is genuinely valuable, a good screen is information the capital market lacks, but it should be sold as certification, not transformation.
- The Contribution Ledger framework makes a program’s claims honest: publish the funnel, track near-misses, attribute by channel, and price the certification separately from the treatment.
- For founders, the finding is liberating: the program’s stamp opens doors, but the trajectory-bending inputs, peers, brokered customers, and operating discipline, are identifiable and demandable.
Why is the brochure inference broken?
Because admission is not random and the comparison group is fictional. When a program reports that graduates outperform “average startups,” it compares firms that survived a competitive screen against firms that never applied, could not pass it, or were screened out, a comparison that measures the screen. The fix is old and rarely used: compare admitted firms against the marginal rejects, the applicants just below the cutoff, who resemble admits in everything but the treatment. Studies using this design, including the well-known work on top US accelerators and the growing GALI dataset spanning emerging markets, consistently shrink the headline effects: much of the raw graduate advantage survives in the near-miss group too, because it was selection all along. What survives as genuine treatment is instructive: accelerated firms show measurable gains in fundraising probability and speed, network size, and follow-on relationships, while revenue and survival effects are smaller, slower, and concentrated in particular founder segments. The program changes access more than it changes operations, at least on the timescales evaluations can see.
Is selection skill worthless, then?
The opposite: it is valuable information sold under the wrong label. In capital markets as thin and noisy as East Africa’s, a rigorous screen, hundreds of applications, structured diligence, evidence-based filters, produces something investors and buyers cannot cheaply produce themselves: a credible shortlist. That is certification, the same economic function a credit score or audit performs, and it deserves to be priced and marketed as what it is. The dishonesty enters when certification wears transformation’s clothes: when a program’s fundraising pitch claims its curriculum caused outcomes its screen selected, it is reporting ceremony as results, and worse, it misallocates its own budget, overfunding content and stage production while underfunding the brokered access that the evidence says is the actual treatment. Programs confident in their screens should also confront the uncomfortable corollary: if selection is most of the value, the highest-impact move may be certifying more firms more cheaply, not treating a tiny cohort more elaborately, the cohort-model question this industry keeps avoiding.
What is the Contribution Ledger?
The named framework of this essay: four practices that let a program know, and honestly say, what it adds.
- Publish the funnel. Applications, shortlist, admits, with criteria. A program proud of its screen should show it; the funnel is the certification’s audit trail.
- Track the near-misses. Follow the just-below-cutoff applicants with the same annual survey graduates get. This is the cheapest honest counterfactual available, and the delta, not the graduate number, is the impact claim.
- Attribute by channel. When graduates report gains, ask through what: peer cohort, specific introductions, curriculum, capital access, the stamp itself. Channel attribution converts a warm testimonial into a budget allocation signal.
- Price certification separately. Offer the screen as a product, diligence-backed shortlists for investors, verified evidence packs for lenders and buyers, so the information value is monetized honestly rather than laundered through transformation claims.
A program running this ledger can say something rare and fundable: here is our selection value, here is our measured treatment, here is the channel it flows through, and here is what your next dollar buys.
What should founders and the region’s funders do with this?
Founders: use the certification, demand the treatment. The stamp is real, wear it into investor and buyer rooms, but choose programs by the channels the evidence validates: dense peer cohorts, brokered customer access, living alumni networks, and operating-discipline formation, and treat curriculum-heavy, access-light programs as expensive certificates. Funders: pay for deltas, not graduate outcomes, require near-miss tracking in every grant, and consider funding the certification function directly, screens that serve the whole capital market rather than one cohort’s stage. And for the faith-driven programs this corpus keeps in view, the selection-treatment distinction carries a theological edge worth sitting with: a program that merely selects the already-strong and crowns them has run a tournament, not a formation, and formation’s evidence is precisely treatment, the near-miss founder whose trajectory bent because the program taught, disciplined, and companioned. Selection finds the gifted. Treatment serves the called. An honest program knows which it is doing, measures both, and charges for neither what it cannot show.
FAQ
What is the selection vs treatment distinction?
Selection is picking firms already likely to succeed; treatment is changing a firm’s trajectory. Graduate outcomes blend both, and only comparison against similar non-admitted firms separates them.
What does credible evidence find?
Real but selective treatment: gains in fundraising probability, speed, and network breadth, with weaker, slower effects on revenue and survival, and much of the raw graduate advantage explained by selection.
Is a strong screen valuable if it isn’t treatment?
Yes, as certification: a rigorous shortlist is information thin capital markets lack, worth pricing and selling honestly, the way audits and credit scores are.
What is the Contribution Ledger?
Four practices: publish the admission funnel, survey near-miss applicants as a counterfactual, attribute reported gains to specific channels, and monetize certification separately from treatment claims.
How should a founder choose a program?
By validated channels: peer density, brokered customer and capital access, and living alumni infrastructure. Treat curriculum-heavy programs without access as certificates, and price the months accordingly.
Related Reading
- Does Startup Acceleration Work? The Evidence
- After the Demo Day: What Replaces the Pitch Ritual
- Alumni Networks Are the Product; Cohorts Are the Funnel
- Against the Three-Month Cohort
Sources and Evidence
- Global Accelerator Learning Initiative: multi-country application-and-outcome data enabling near-miss comparisons.
- Hallen, Cohen & Bingham research on accelerators: evidence on treatment effects at leading programs.
- World Bank evaluations of entrepreneurship programs: randomized and quasi-experimental findings in developing markets.
- ANDE: sector evidence reviews on entrepreneur support effectiveness.