The missing player in the standard whistleblowing story
Most whistleblowing reform starts with a two-party story. A worker reports a concern. Management either investigates properly or punishes the worker. If reporting still fails, the usual fixes are stronger employment protection for the reporter and sharper penalties for managers who bury the case.
Ari Ercole's paper argues that this story leaves out the group that often decides the practical outcome: colleagues. In a hospital, a report about unsafe practice rarely stands on one person's evidence alone. It needs corroboration from people who saw the same pattern, worked the same shifts, signed the same notes, or sat in the same meetings.
That is the paper's useful move. It treats whistleblowing as a three-player game between the whistleblower, Trust management, and the colleague group. The model is formal, using a repeated game and global games techniques to obtain a unique equilibrium. But the intuition is plain enough. A report can be true, and management can be formally open to hearing it, while the case still collapses because colleagues do not support it.
This is not just cowardice. It can be individually rational. If a ward has a systemic problem, corroborating the report may expose the witness's own omissions. Silence is safer. The colleague may believe the whistleblower and still choose not to help.
How colleague support becomes the binding constraint
The key constraint in the model is not whether the whistleblower speaks. It is whether enough colleagues support the report for management to validate it.
Management observes the report, the aggregate support rate, and any counter-signal created by colleagues who attack the report's credibility. It then decides whether to validate or decline. Colleagues receive noisy private signals about the severity of the violation. In the model, they may support, stay silent, or, in the active-elimination part of the analysis, help create a counter-narrative.
The global games setup matters because each colleague is not only asking, "Is the report true?" They are also asking, "Will enough other people support it?" If support is likely to be thin, joining the reporter is costly and may not change the outcome. If support is likely to be broad, corroboration is safer and more useful. Small differences in private beliefs can therefore produce a sharp coordination threshold.
Ercole's stronger claim is that collective implication shifts that threshold in the wrong direction. The witness's problem is not only retaliation by management. It is self-incrimination. Supporting the report can invite scrutiny of what the witness knew, did, failed to escalate, or normalized. That cost applies even when management is honest. It also applies even when the original reporter is protected.
That is why the model can produce a grim equilibrium without needing a cartoonishly corrupt manager. Colleagues stay quiet because support exposes them. Management sees little support and infers weak credibility. The report is declined. Nobody needed to coordinate explicitly.
Three mechanisms policy usually misses
The paper's three mechanisms are concrete enough to matter for policy design.
- Collective implication. Corroboration can implicate the witness. This is the central mechanism. If a safety failure was widely visible, the people best placed to validate it are also the people most exposed by validation.
- Active elimination. Colleagues may go beyond silence and build a documented counter-narrative that shifts management's belief about the report's credibility. The point is not general hostility. It is the creation of a record that can be used against validation.
- The turnover ratchet. Short-tenure managers inherit files, not memories. A successor sees that prior reports were "unsubstantiated" but not the reasoning, pressure, missing testimony, or internal politics behind that label. In the model, the recorded prior degrades geometrically through succession.
The distinction between silence and active elimination is one of the sharper parts of the paper. Silence leaves little trace. It is hard to punish and hard to deter. Active elimination, by contrast, creates documents, messages, and attributable acts. If the exposure environment is credible enough, independent elimination can become too risky.
Collusion is different. A coordinated group can trade uncertainty for a higher chance of defeating the report. The paper argues that the policy tool that deters a lone colleague from fabricating a counter-story does not necessarily deter a group acting together. Once suspected, collusion needs a different response: investigate it, confirm it, and punish it promptly, before the record is permanently distorted.
Why protection for reporters may raise reports without raising validation
This is the paper's most practitioner-relevant warning. Employment protection can increase the number of reports while leaving the validation rate unchanged.
That sounds counterintuitive only if the reporter is the bottleneck. In Ercole's model, the reporter's participation constraint is separate from the colleague group's support constraint. Better protection can make the whistleblower more willing to file. It does not reduce the self-incrimination cost borne by colleagues who would need to back the report.
So the Trust can receive more concerns, while the same bottleneck kills them after filing. The resulting pattern can also be misread. Management may see a rising volume of reports and a flat validation rate and infer that the extra reports are lower quality. The model allows another interpretation: the same coordination failure is filtering the cases after they are raised.
This also explains why legal liability for management has limits. Liability bites when a wrongful decline is likely to be exposed. But if colleagues stay silent, exposure is less likely. The mechanism that hides the concern also weakens the penalty meant to discipline the decision-maker.
Why we could not backtest this
We could not backtest this paper on our data because it does not define a tradable signal or a repeated market rule. There is no asset, portfolio, entry condition, holding period, or payoff series to test.
A proper empirical test would need case-level NHS data: who reported, who could have corroborated, who did, whether witnesses faced self-implication, whether counter-narratives were created, manager tenure, investigation timing, and final validation. Those variables are not in market data, and many are not likely to be cleanly recorded in public NHS sources.
That does not weaken the theory by itself. It means the right test is institutional and forensic, not a price backtest.
The proposed fixes and where the theory needs evidence
The proposed fixes follow directly from the model.
Colleague amnesty is aimed at collective implication. The idea is borrowed from cartel leniency programmes. If early supporting witnesses receive protection for related omissions, the support threshold can move. This is a serious proposal, but it needs careful design. Amnesty that is too narrow will not change behaviour. Amnesty that is too broad may look like a pass for unsafe practice.
Mandatory named Trust outcome publication is aimed at the degraded institutional record. The paper's logic is that public reporting can repair the file that future managers inherit. It can also reproduce part of the effect of a public inquiry at administrative cost. The open question is what must be published. Counts alone will not do much. Practitioners would need categories that distinguish withdrawn reports, unsupported reports, validated reports, and cases where corroboration was later found to have been suppressed.
Prompt investigation of suspected coalitions is aimed at collusive elimination. The paper makes a strong sequencing claim: suspected collusion should be investigated before resolving the underlying report, because delay lets the false record harden. That is plausible. It is also demanding. Trusts would need triggers, independence, document preservation rules, and consequences that arrive fast enough to matter.
The theory now needs evidence on magnitudes. How often do potential witnesses stay silent because of self-implication rather than fear of management retaliation? How often do counter-narratives arise independently versus through coordination? Does manager turnover measurably lower the chance that later reports are believed? These are answerable questions if investigators code cases with the colleague layer in mind.
The paper's main contribution is to point to the missing data field: not just "was a concern raised," but "who else knew, and why did they not support it?"