Getting your Trinity Audio player ready...

Yes—failed experiments can be valuable public goods. A rigorous experiment that disproves a hypothesis, identifies an ineffective method, or documents the limits of an intervention gives other researchers information they can reuse. It can prevent duplicated work, improve future experimental designs, correct distorted scientific literature, and direct funding toward more promising approaches.

The problem is not that science produces negative results. Negative results are a normal and necessary part of scientific discovery. The problem is that researchers are often rewarded only when experiments produce positive, novel, and publishable findings.

As a result, much of what science learns about what does not work remains inaccessible.

A better research-funding system would treat well-documented negative results as reusable scientific infrastructure—and compensate researchers for producing them.

What Is a Failed Experiment?

A failed experiment is not simply an experiment that produces an unwanted answer.

Several distinct outcomes are commonly described as failures:

• A hypothesis is not supported.
• A predicted effect is not detected.
• A proposed treatment or technique does not outperform an alternative.
• An existing result cannot be replicated.
• An experimental apparatus or protocol proves unreliable.
• A study is inconclusive because its sample, measurements, or statistical power were inadequate.
• An experiment cannot be completed because of technical or logistical problems.

These outcomes do not all have equal scientific value.

A well-designed experiment that produces a clear negative result may be highly informative. By contrast, an experiment with uncontrolled variables, corrupted data, or an undocumented procedure may reveal very little.

The relevant distinction is therefore not between positive and negative results, but between informative and uninformative research.

A negative result is valuable when it reliably reduces uncertainty about a scientific question.

Why Negative Results Often Disappear

Scientific institutions generally reward publication, citation, novelty, and apparent success. Researchers therefore have strong incentives to prioritize positive findings.

Studies with statistically significant or favorable results have historically been more likely to appear in journals—and to appear sooner—than studies with negative findings. A Cochrane review found that clinical trials with positive findings were more likely to be published and were published more quickly than trials with negative findings.

This creates the well-known file drawer problem: unsuccessful replications, null findings, and abandoned approaches remain in private files instead of entering the scientific record.

The decision may seem rational from the perspective of an individual researcher. Preparing a paper requires time, journals may reject the result as insufficiently novel, and the publication may contribute little to career advancement.

Collectively, however, the consequences are damaging.

When negative findings disappear:

1. Other laboratories may repeat the same unsuccessful experiment.
2. Meta-analyses may overestimate the strength of an effect.
3. Funders may continue supporting approaches that already have substantial contrary evidence.
4. Researchers cannot distinguish an unexplored idea from an idea tested unsuccessfully many times.
5. Machine-learning systems trained on the published literature receive a distorted account of scientific evidence.

Cochrane warns that systematic reviews become vulnerable to bias when studies are missing because of the magnitude, direction, or statistical significance of their results.

Negative Results as Public Goods

In economics, a public good is generally characterized by non-rivalry and difficulty of exclusion. One person’s use of the information does not prevent others from using it, and openly published knowledge can benefit researchers far beyond the institution that produced it.

A documented failed experiment can have these characteristics.

Suppose a laboratory spends six months testing whether a particular catalyst works under specified conditions. It discovers that the catalyst degrades rapidly and cannot produce the expected reaction.

If the result remains private, another laboratory may spend another six months discovering the same limitation.

If the complete result is published—including materials, temperatures, calibration procedures, raw data, and failure analysis—many laboratories can avoid the same dead end. The original experiment has created a reusable map of an unproductive region of the research space.

Its value may include:

• eliminating implausible hypotheses;
• identifying boundary conditions;
• exposing defects in standard protocols;
• improving statistical power calculations;
• revealing unsuitable instruments or materials;
• informing systematic reviews;
• preventing unnecessary exposure of participants or animals;
• training scientific AI systems on both successful and unsuccessful approaches.

The result is non-rival: thousands of researchers can use it simultaneously.

Its benefits are also widely distributed. The laboratory that produces the result may capture only a small fraction of the time and money saved elsewhere. That is precisely why the market and conventional academic incentives tend to underfund it.

How Hidden Failures Cause Research Waste

Imagine that ten laboratories independently test approximately the same hypothesis. Nine obtain null or negative results, but only the tenth obtains a statistically significant positive result.

If only the positive study is published, the literature gives the appearance of strong supporting evidence. Researchers, doctors, policymakers, and funders do not see the nine contrary outcomes.

This is more than an abstract statistical problem. Selective publication can inflate estimated effects and make the available evidence unreliable for decision-making.

The hidden studies also represent duplicated expenditure:

[
\text{avoidable waste}

\sum \text{cost of repeated unsuccessful approaches}.
]

The costs are not limited to money. They include researchers’ time, laboratory capacity, computational resources, participant recruitment, animal use, materials, and delayed progress toward better alternatives.

Publishing negative results does not recover all these costs. It converts part of the expenditure into a durable informational asset.

A Negative Result Is Not Proof That Something Is Impossible

Funding negative results requires careful interpretation.

“An experiment did not detect an effect” is not automatically equivalent to “the effect does not exist.”

A null result may arise because:

• the sample was too small;
• measurements were too noisy;
• the effect occurs only under different conditions;
• the statistical test had insufficient power;
• the experimental implementation did not match the underlying theory;
• the hypothesis was underspecified;
• the selected population was inappropriate.

For this reason, negative-result repositories should not be collections of unsupported declarations such as “Method X does not work.”

They should contain structured evidence:

• the precise hypothesis;
• the experimental design;
• preregistered outcomes where applicable;
• equipment and materials;
• sample-selection rules;
• statistical power;
• raw or appropriately protected data;
• code and analysis procedures;
• deviations from the original protocol;
• uncertainty estimates;
• plausible alternative explanations;
• the domain within which the conclusion applies.

A useful record might conclude:

Under the reported conditions, using this sample and measurement procedure, the experiment did not detect the predicted effect above the stated sensitivity threshold.

That statement is narrower than a universal rejection, but far more useful than an undocumented claim of failure.

How Science Could Fund Negative Results

Traditional research grants generally fund proposed work before its value is known. Publication systems then reward the subset of projects that produce attractive final narratives.

An alternative is to combine prospective research support with retroactive payments for verified scientific outputs.

Under such a system, a researcher could receive funding after publishing a useful negative result, replication failure, protocol warning, or technical postmortem.

The reward should depend on the informational value of the work, not on whether the original hypothesis was confirmed.

A Possible Evaluation Model

A funding system could score a negative result across several dimensions:

[
V = Q \times I \times R \times A,
]

where:

• (Q) is methodological quality;
• (I) is expected informational value;
• (R) is reproducibility and documentation quality;
• (A) is applicability to other researchers.

Additional adjustments could reflect:

• the cost of experiments the result may prevent;
• the number of research groups working on related questions;
• whether the finding contradicts an influential published claim;
• the availability of data and code;
• independent verification;
• ethical importance;
• evidence that researchers have reused the result.

This model would not pay every failed attempt equally. It would reward failures that create credible, reusable knowledge.

That principle is consistent with results-based scientific funding: funding should follow demonstrated scientific utility rather than depend entirely on promises made before the work begins.

Small Rewards May Be Enough

Negative results often do not require grants comparable to major discovery prizes.

A system could provide:

• small automatic rewards for properly structured reports;
• larger rewards after independent replication;
• bonuses when a report prevents documented duplication;
• continuing payments when datasets, protocols, or failure analyses are reused;
• bounties for testing uncertain assumptions in important research programs.

This creates a continuum of scientific compensation.

A modest but reliable reward may make it rational for researchers to spend several days cleaning their data, documenting the protocol, and publishing the result instead of abandoning it.

The goal is not to make failure more profitable than discovery. The goal is to prevent useful information from becoming economically worthless merely because a hypothesis was rejected.

Registered Reports Address Part of the Problem

One established mechanism is the Registered Report.

With Registered Reports, journals evaluate the importance of the research question and the quality of the proposed method before the results are known. A study can receive in-principle acceptance before data collection or analysis, reducing the incentive to reject it later because its findings are negative.

The Center for Open Science explains Registered Reports as a model designed to reduce bias against negative findings by separating publication decisions from the eventual outcome.

Registered Reports are valuable, but they do not solve the entire funding problem.

They primarily address publication. Researchers still require resources to conduct the work, and many exploratory studies, technical failures, replication attempts, and abandoned methods will never enter the Registered Report process.

A comprehensive system therefore needs both:

1. results-independent publication mechanisms, and
2. direct financial rewards for useful negative evidence.

Clinical-Trial Transparency Shows Why Complete Reporting Matters

The clearest ethical case appears in clinical research.

When clinical-trial results remain unpublished, physicians and patients may make decisions using an incomplete evidence base. Treatments can appear more effective or safer than the total evidence supports.

The AllTrials campaign advocates registration and full reporting of all clinical trials, including those with negative outcomes.

ClinicalTrials.gov also maintains procedures and structured fields for submitting trial results.

The same principle extends beyond medicine. In chemistry, engineering, mathematics-assisted experimentation, psychology, computer science, and materials research, incomplete reporting distorts collective knowledge even when the immediate consequences are less visible.

AI Could Help Evaluate Failed Experiments

The number of negative and technical results could be enormous. Human committees cannot manually assess every laboratory note.

AI-assisted evaluation could reduce the cost of processing this information by:

• checking whether required methodological information is present;
• comparing a report with related published studies;
• detecting duplicated or nearly identical submissions;
• identifying whether a claimed negative result actually follows from the data;
• estimating how many projects may benefit from the information;
• matching unsuccessful approaches with researchers considering similar work;
• flagging unsupported generalizations;
• recommending independent replication;
• tracking downstream reuse.

An AI evaluator should not simply assign a probability that an experiment “failed.” It should distinguish among methodological failure, null evidence, contradictory evidence, implementation failure, and genuine falsification.

Such evaluations must remain auditable and open to challenge. The system should state why it assigned a score and what evidence influenced the decision. This follows the broader case for transparent, merit-based research evaluation.

Risks of Paying for Negative Results

A negative-results funding system could itself be exploited.

Researchers might divide one project into many small failure reports, run poorly designed experiments solely to collect rewards, or falsely claim that an obvious technical mistake represents useful scientific evidence.

The system would therefore need safeguards.

Methodological thresholds

A report should not receive a substantial reward unless its design was capable of answering the stated question.

Preregistration bonuses

Preregistered hypotheses and analysis plans should receive stronger evidentiary weight, although exploratory failures can still have technical value.

Independent verification

Important or surprising negative results should receive larger rewards only after another evaluator or research group verifies them.

Duplicate detection

Reports that merely repeat already-known failures without adding evidence should receive little or no compensation.

Impact-based follow-up

Initial rewards can be modest. Additional funding can be released when the result is cited, reused, replicated, or shown to prevent wasted work.

Adversarial review

The evaluation model should be deliberately tested for manipulation. Systems distributing scientific funds need adversarial testing before they are trusted with substantial budgets.

Not Every Failed Experiment Should Become a Paper

Requiring a conventional journal article for every negative result would create excessive editorial work and flood the literature with repetitive reports.

The better approach is a layered infrastructure:

1. Structured research record: A concise, machine-readable description of the experiment.
2. Repository deposit: Data, code, protocol, and failure analysis.
3. Automated screening: Checks for completeness, duplication, and obvious defects.
4. Expert review when warranted: Human evaluation for consequential or disputed results.
5. Full publication: Reserved for results with broad conceptual significance.
6. Funding allocation: Proportional to verified quality and usefulness.

This makes negative evidence discoverable without pretending that every report is a major scientific contribution.

From a Literature of Successes to a Map of Reality

Current scientific literature resembles a travel map that records mainly the roads that reached desirable destinations. Dead ends, collapsed bridges, seasonal routes, and inaccessible paths are often omitted.

Such a map may appear optimistic, but it is not efficient.

Science advances through search. A successful search system must preserve information about both promising and unpromising directions. Otherwise, each generation of researchers is forced to rediscover part of the same search space.

Openly documented negative results can form a distributed memory for science.

They tell researchers:

• which assumptions have been tested;
• which methods failed under specified conditions;
• which effects may be weaker than reported;
• which measurements are unreliable;
• which replications did not succeed;
• which questions remain genuinely open.

This knowledge is not a by-product of science. It is part of science.

Conclusion: Fund Information, Not Just Success

Failed experiments can be valuable public goods when they are rigorous, documented, searchable, and reusable.

Their value does not come from celebrating failure for its own sake. It comes from reducing uncertainty and preventing society from paying repeatedly for the same lesson.

A rational scientific-funding system should therefore ask:

How much reliable and reusable knowledge did this work create?

Sometimes the answer will be a successful discovery. Sometimes it will be a dataset, a tool, a replication, a corrected protocol, or a carefully established negative result.

Funding all of these outputs according to their verified utility would produce a more complete scientific record—and a system in which researchers no longer need to disguise every worthwhile project as a success story.