Should Some Research Grants Be Allocated by Lottery?

Getting your Trinity Audio player ready...

Yes, some research grants should be allocated by lottery—but only after proposals pass meaningful eligibility, quality, ethics, and feasibility checks.

A research funding lottery should not treat a rigorous project and an obviously defective proposal as equals. Its proper purpose is narrower: to choose among multiple fundable projects when peer review cannot reliably determine which one deserves the final available grant.

This model is usually called a partial lottery or partially randomized funding allocation. Experts first identify proposals that meet a defined standard. Random selection is then used within that qualified group, particularly near the funding threshold.

The principle is simple:

Peer review should reject proposals that are clearly unfit, but it should not pretend to measure tiny differences between similarly strong proposals with scientific precision.

What Is a Research Grant Lottery?

A research grant lottery is a funding mechanism in which chance plays some role in selecting grant recipients.

There are two fundamentally different models.

A full lottery

Every eligible application has an equal chance of receiving funding. Scientific reviewers may check only basic eligibility or compliance.

This method greatly reduces administrative costs, but it risks funding weak projects while rejecting substantially better ones. It is therefore difficult to justify for large, expensive, safety-sensitive, or mission-critical research programs.

A partial lottery

Peer reviewers first assess the proposals. Applications that fail minimum standards are rejected, while proposals judged sufficiently strong enter a randomized selection process.

Some systems also fund the highest-rated proposals automatically and use a lottery only for the difficult middle group. This is generally the more defensible model.

A partial lottery therefore does not replace scientific judgment. It limits the use of judgment to distinctions that reviewers can plausibly make.

Why Conventional Grant Ranking Is Less Precise Than It Appears

Traditional grant competitions usually produce scores, rankings, panel discussions, and a final funding line. This creates an appearance of exact measurement: proposal 12 is funded, while proposal 13 is not.

But research quality is not measured on a perfectly calibrated scale. Reviewers can disagree because they have different expertise, theoretical commitments, risk tolerances, and interpretations of novelty. Scores may also change depending on which reviewers happen to receive an application.

Near the funding line, the difference between success and rejection may reflect:

• reviewer assignment;
• wording and presentation;
• disciplinary fashion;
• institutional prestige;
• familiarity with the proposed method;
• panel composition;
• small and uncertain score differences.

The Swiss National Science Foundation has used random selection when proposals around the funding boundary are judged to be of comparable quality. The rationale is not that every project is equal, but that the evaluation process cannot confidently rank all good proposals in a strict order.

In this situation, a transparent lottery may be more honest than an overconfident ranking.

The Strongest Case for Grant Lotteries: Uncertainty Near the Funding Line

Suppose a funder receives 500 applications and can support 50.

Reviewers might identify:

• 20 unusually strong proposals;
• 100 additional proposals that are credible and potentially valuable;
• 380 proposals that do not meet the program’s standard.

It may be reasonable to fund the top 20 directly and reject the bottom 380. The harder question is how to select the remaining 30 grants from the 100 qualified proposals.

A conventional panel might rank them from 21 to 120. But does the evidence really establish that proposal 49 is materially better than proposal 73?

Often, it does not.

A partial lottery acknowledges three categories:

1. Clearly fundable
2. Fundable but not reliably rankable
3. Not fundable

Randomization is applied only to the second category.

Grant Lotteries Can Reduce False Precision

Peer-review scores are estimates, not physical measurements. A proposal receiving 4.6 out of 5 is not necessarily objectively better than one receiving 4.4.

Small numerical differences may be smaller than the uncertainty of the evaluation itself. Converting those differences into a strict ranking can therefore produce false precision.

Randomization can make this uncertainty explicit. Instead of claiming that the panel has identified the uniquely correct winner, the funder states:

These proposals have all crossed our quality threshold. Available evidence does not justify a confident ranking among them, so the final selection will be randomized.

This is not a rejection of merit. It is a more careful statement about how accurately merit can be measured before research has actually been performed.

Lotteries May Reduce Prestige and Network Bias

Grant review can be influenced by an applicant’s institution, publication history, collaborators, reputation, country, or disciplinary network. Blinding can reduce some forms of bias, but it cannot remove all of them. Research topics and citations may reveal the identities or intellectual communities of applicants.

A lottery among qualified proposals limits the ability of small subjective preferences to determine the final result.

It may particularly help:

• early-career researchers;
• independent researchers;
• interdisciplinary teams;
• researchers at less prestigious institutions;
• unconventional projects;
• applicants outside dominant professional networks.

A lottery cannot eliminate inequality before the selection stage. Researchers with more institutional support may still produce more polished applications. Nevertheless, randomization can reduce the influence of marginal score differences through which prestige advantages are often amplified.

Lotteries Can Make Space for Risky and Unconventional Research

Expert review is valuable, but experts are usually trained within existing research paradigms. They may be well positioned to identify technical defects, yet less able to predict which unconventional ideas will produce major discoveries.

Novel projects often have less preliminary evidence precisely because they are novel. They may appear less safe than incremental proposals with familiar methods and predictable results.

New Zealand’s Health Research Council uses a modified lottery for its Explorer Grants, which support transformative, innovative, exploratory, or unconventional health research. Applications must first be judged suitable for funding; qualified proposals then enter random selection. Its 2026 Explorer Grants offer up to NZ$150,000 for projects lasting up to 24 months.

The Volkswagen Foundation also introduced partially randomized selection in its completed “Experiment!” initiative for bold research ideas. Across eight calls, the program received 5,051 applications and funded 183 projects; randomization was introduced in 2017.

These programs use lotteries not because scientific quality is irrelevant, but because the future value of genuinely exploratory research is unusually difficult to predict.

Lotteries Can Reduce Wasteful Proposal Competition

Writing a major grant proposal may require weeks or months of work. Applicants collect preliminary results, build consortia, estimate budgets, obtain institutional approvals, and adapt their language to the preferences of a particular funder.

Most of that labor produces no direct scientific output when success rates are low.

Competitive funding can become a costly contest in which researchers devote increasing effort to improving proposals rather than performing research. Economic models have suggested that, under some conditions, the aggregate effort spent competing for grants may consume a substantial share of the scientific value created by the funded projects. Partial lotteries are one proposed method for reducing this waste.

Lotteries do not automatically solve the problem. If applicants must still prepare 100-page submissions to enter the randomized pool, the administrative burden remains. The greatest efficiency gains arise when funders combine partial randomization with:

• shorter initial applications;
• standardized evidence requirements;
• automated compliance checks;
• staged evaluation;
• reusable researcher profiles;
• limited resubmission requirements.

Randomization Can Improve Portfolio Diversity

Scientific impact is highly uncertain. A proposal that looks safe may produce little value, while an unusual project may generate an unexpected breakthrough.

A funder should therefore think like a portfolio manager rather than attempt to identify one certain winner.

A diversified research portfolio can include:

• established work with high feasibility;
• replication and validation;
• enabling infrastructure;
• high-risk exploratory projects;
• neglected research areas;
• early-career investigators;
• independent and interdisciplinary researchers.

Random selection within qualified categories can prevent the entire portfolio from converging on the same fashionable methods, institutions, and assumptions.

The purpose is not diversity for its own sake. It is protection against correlated error: if every panel favors similar projects, the funding system may repeatedly make the same mistake.

What Are the Main Objections to Grant Lotteries?

Research lotteries have genuine limitations. They should not be presented as a universal cure for peer review.

“A lottery ignores scientific merit”

A full lottery largely does. A properly designed partial lottery does not.

Scientific merit determines whether a proposal enters the fundable pool. Chance is used only where the funder lacks enough reliable information to make finer distinctions.

The key policy question is therefore not “merit or lottery?” It is:

At what point does additional ranking stop measuring merit and start measuring noise?

“Researchers deserve reasons for rejection”

Applicants reasonably want to understand why they were unsuccessful. A lottery can feel frustrating because no corrective explanation is available.

However, conventional reviews often provide contradictory or low-value feedback. A proposal may be rejected despite being scientifically acceptable simply because the budget is insufficient.

A partial-lottery funder should clearly distinguish between:

• rejection for identified deficiencies;
• qualification for the lottery followed by non-selection.

Being unsuccessful in a lottery should not be recorded as evidence that the research was scientifically weak.

“Random selection may damage public trust”

A public announcement that “research funding was awarded randomly” can sound irresponsible without context.

Funders must explain that the randomization occurred only among proposals that passed expert assessment. They should publish the procedure, thresholds, number of qualified proposals, lottery method, conflicts policy, and audit records.

Transparency is essential. A hidden or poorly documented lottery would create rather than reduce distrust.

“Researchers may optimize for the minimum threshold”

If all qualified proposals receive equal odds, applicants may aim merely to cross the threshold instead of producing their best possible plan.

Several mechanisms can reduce this risk:

• automatic funding for exceptional proposals;
• weighted probabilities based on broad quality bands;
• separate lotteries by scientific category;
• milestone-based continuation;
• post-award evaluation;
• retroactive rewards for demonstrated results.

Weighted systems require caution. Giving a proposal a probability of 4.7% rather than 4.3% may reintroduce the same false precision that lotteries are meant to avoid.

“Randomness does not eliminate bias in qualification”

Correct. Reviewers still decide which proposals enter the pool. A biased threshold can exclude unconventional researchers before the lottery begins.

Partial randomization should therefore be combined with reviewer audits, conflict-of-interest controls, appeal procedures, structured criteria, and tests for systematic differences across institutions or applicant groups.

Systems distributing public or charitable money should also undergo adversarial testing to identify manipulation, collusion, identity fraud, strategic scoring, and other failure modes.

Which Research Grants Are Best Suited to a Lottery?

Partial lotteries are most defensible when four conditions apply:

• many proposals meet a reasonable quality threshold;
• differences among qualified proposals are highly uncertain;
• the program seeks exploration or portfolio diversity;
• the cost of detailed ranking is high relative to the grant size.

Good candidates include:

Small exploratory grants

Seed grants allow researchers to test whether an unconventional idea deserves larger investment. Because expected outcomes are uncertain and award sizes are limited, randomized selection can be proportionate.

Early-career grants

Reviewers have less evidence with which to assess new researchers. Small differences in publication records may reflect opportunity and institutional support rather than underlying ability.

A qualified-pool lottery can prevent premature career stratification in which receiving one early grant makes every later grant easier to obtain.

High-risk, high-reward programs

Panels may be able to identify whether a speculative project is coherent without being able to predict which speculation will succeed. Portfolio randomization can support multiple approaches rather than one consensus favorite.

Grants near an indistinguishable funding boundary

This is the least radical use. The funder retains conventional review but uses a lottery to break ties or select among proposals with overlapping evaluation uncertainty.

Microgrants

For very small grants, elaborate review may cost almost as much as the research being funded. A lightweight qualification process followed by a lottery can direct more of the budget toward actual work.

Which Grants Should Usually Not Be Randomized?

Lottery allocation is less appropriate where proposals differ substantially in quality or where mistakes impose unusually high costs.

Examples include:

• major research infrastructure;
• projects involving serious safety risks;
• programs with strict operational deliverables;
• emergency research requiring rapid strategic coordination;
• projects where only one team has the necessary capability;
• grants dependent on highly specific facilities or datasets;
• late-stage translational work with measurable execution requirements.

Even in these cases, randomization could sometimes resolve a genuine tie. It should not replace due diligence, technical verification, safety review, or assessment of delivery capacity.

A Better Model: Review, Lottery, Milestones, and Retroactive Funding

The strongest research funding system does not need to choose one mechanism for every stage.

A hybrid process could work as follows:

Eligibility and integrity screening

Check identity, conflicts, research ethics, legal compliance, budget plausibility, and whether the proposal falls within the program’s scope.

Scientific threshold review

Experts determine whether the project is coherent, sufficiently original, technically plausible, and potentially valuable.

Direct funding for exceptional cases

A small number of proposals with unusually strong evidence or strategic importance may receive immediate funding.

Lottery among comparable qualified proposals

Remaining fundable applications enter a transparent lottery. The selection procedure and probability rules are published and auditable.

Milestone-based continuation

Initial grants remain relatively small. Projects demonstrating credible progress become eligible for additional funding.

Retroactive rewards

Research outputs that create verifiable scientific or public value receive further support after results exist. This shifts some funding from prediction toward evidence.

This hybrid model separates different questions:

• Is the project eligible and responsible?
• Is it scientifically fundable?
• Can reviewers reliably distinguish it from other fundable projects?
• Did the resulting work create value?

No single mechanism answers all four questions well.

How AI Could Improve—But Not Control—Grant Lotteries

AI systems could help analyze applications, detect unsupported claims, compare budgets, identify possible conflicts, map related literature, and estimate uncertainty in reviewer disagreement.

AI should not conceal uncertainty behind an even more complicated score. A model producing “87.31 scientific merit points” may create greater false precision than a human panel.

A more defensible role for AI is to identify broad categories:

• clearly below the funding threshold;
• uncertain and requiring human examination;
• qualified for randomized selection;
• unusually strong or strategically relevant.

AI could also audit whether apparently neutral criteria produce systematic disadvantages. However, any such system must be tested for manipulation, model bias, unstable scoring, and dependence on incomplete scientific literature. The AI alignment requirements for research funding should therefore include transparent uncertainty, appeal mechanisms, and limits on autonomous decision-making.

Should Lottery Winners Receive Equal Chances?

Not necessarily. Several designs are possible.

Model	How it works	Main advantage	Main risk
Equal qualified-pool lottery	Every proposal above the threshold receives equal odds	Simple and transparent	Encourages threshold optimization
Tie-break lottery	Randomization applies only to proposals tied near the funding line	Minimal change to peer review	Preserves most ranking costs
Quality-band lottery	Proposals enter broad bands with different probabilities	Recognizes major quality differences	May reintroduce score gaming
Automatic top tier plus lottery	Exceptional proposals are funded; other qualified proposals enter a lottery	Balances excellence and uncertainty	“Exceptional” category may reflect prestige bias
Portfolio lottery	Separate pools exist for fields, risks, career stages, or project types	Produces deliberate diversification	Category design can become political
Staged lottery	Winners receive small grants before competing for continuation	Limits the cost of mistaken selection	Requires later evaluation infrastructure

There is no universally correct design. The mechanism should follow the purpose, grant size, uncertainty, and risk profile of the funding program.

Rules for a Credible Research Funding Lottery

A legitimate system should publish:

• the criteria for entering the lottery;
• the reasons a proposal may be excluded;
• whether exceptional proposals bypass randomization;
• the probability assigned to each qualified application;
• how conflicts of interest are handled;
• the random selection method;
• an auditable record of the draw;
• rules against duplicate or coordinated applications;
• the appeals process;
• post-award reporting requirements;
• aggregate data on applicants and recipients.

The random-number process itself should be verifiable. A blockchain can provide a tamper-evident record and publicly checkable randomness, but blockchain technology does not make the surrounding evaluation fair by itself. Biased eligibility rules recorded immutably remain biased.

Governance, review quality, identity safeguards, and accountability matter more than the choice of database.

The Deeper Principle: Randomness Is Sometimes More Rational Than Pretended Certainty

Scientific institutions often treat randomness as the opposite of rational decision-making. But randomization is already fundamental to rigorous science. Clinical trials randomize participants because uncontrolled selection creates bias. Sampling procedures use randomness to improve representativeness.

Grant allocation presents a different problem, but the underlying lesson is related: when decision-makers cannot reliably distinguish between alternatives, controlled randomness may be less biased than arbitrary discretion.

The alternative to a grant lottery is not always a perfectly informed meritocracy. It may be an opaque mixture of noisy scores, reviewer preferences, institutional reputation, strategic writing, and panel dynamics.

A lottery does not claim to know which qualified project will succeed. That intellectual humility can be a strength.

Conclusion: Use Lotteries Where Prediction Becomes Guesswork

Some research grants should be allocated by lottery, particularly small exploratory grants and proposals clustered near an uncertain funding threshold.

However, the strongest model is not a full lottery. It is a qualified partial lottery embedded within a broader funding portfolio:

• peer review removes clearly unsuitable proposals;
• exceptional projects may receive direct support;
• comparable fundable proposals enter a transparent lottery;
• milestones control later funding;
• retroactive evaluation rewards demonstrated scientific value.

Grant lotteries cannot eliminate bias, guarantee breakthroughs, or replace scientific expertise. They can, however, prevent unreliable distinctions from being presented as objective merit.

The correct goal is not to randomize science funding. It is to use randomness precisely where human and algorithmic prediction no longer justify confident selection.