LEADERSHIPMonths to result85% confidence

Psychological Safety and the Mistake-Reporting Paradox

Teams with better cultures report more errors — they're not making more, they're hiding fewer.

Problem it solves

Confusing low error reports with high team performance, when in fact they signal suppressed reporting

Best for

Leaders building high-reliability teams in medicine, finance, aviation, or any domain where hidden errors compound over time.

Not ideal for

Individual contributors working alone — the framework is specifically about team error dynamics.

Overview

Why this framework exists

Harvard Business School's Amy Edmondson was studying surgical teams expecting to find that better-functioning teams made fewer mistakes. She found the opposite: teams with stronger cultures, better communication, and higher trust reported significantly more errors. The naive interpretation would be that good culture causes mistakes. The correct interpretation is that good-culture teams surface and report errors that bad-culture teams bury.

The mechanism is straightforward: in low-trust teams, admitting a mistake carries career and social risk. Errors get hidden, attributed elsewhere, or simply not spoken aloud. The team's official mistake count is low; the actual error rate is the same or higher, but the correction mechanism is broken. In high-trust teams, errors are surfaced quickly, corrected, and learned from. The official count looks worse; the actual performance is better.

Harford connects this directly to the Fisher vs Keynes framework: the same ego-protection that made Fisher unable to acknowledge his forecasting errors at a systemic level operates in teams. He also connects it to the cult doubling-down story — the psychological cost of public admission, when high, produces suppression rather than correction. The leadership implication is counterintuitive: measuring error frequency as a KPI of team performance is backwards. The KPI to manage is error surfacing speed and fidelity.

Core principles

5 total
  1. Low error counts in a team often indicate suppressed reporting, not low error rates.
  2. Psychological safety is the condition under which errors are surfaced rather than hidden.
  3. The goal is not to eliminate errors but to ensure that every error that happens is seen and corrected.
  4. Public admission costs create private suppression — team culture determines whether the personal cost of reporting is bearable.
  5. A team that talks about its failures is a team that learns from them; a team that hides its failures repeats them.

Steps

4 steps
  1. Audit error reporting culture, not error frequency
    Ask whether your team's error count is a measure of actual errors or of errors that reached the surface. Conduct anonymous retrospectives on recent projects and compare reported errors to what individuals privately acknowledge. The gap between private and reported error counts is the measure of psychological safety.
    Pro tipA useful question for this audit: 'Has anyone in the last month discovered a problem and not mentioned it? What made you not mention it?'
  2. Model error admission at the leadership level
    Psychological safety is set from the top. Leaders who admit their own mistakes publicly — not performatively, but genuinely — change the calculus for the team. If the person with the most to lose can say 'I got that wrong', the cost of admission for everyone else drops.
    Pro tipTiming matters: admitting the mistake quickly (before the consequences compound) is more credible than admitting it once it's unavoidable. Fisher's inability to admit early is what made the eventual admission catastrophic.
    WarningDo not confuse vulnerability performance with genuine accountability. Teams can see the difference.
  3. Separate error reporting from blame assignment
    When errors are reported and immediately followed by blame attribution, reporting drops to near-zero. Build a period between error surfacing and root cause analysis — the question 'what happened?' must be answered before 'whose fault was it?', and in many cases the answer to the second question is 'the system.'
    Pro tipBlameless post-mortems — common in software engineering — are a formalised version of this principle. The format creates structural protection for the person reporting.
  4. Track error surfacing as a leading performance indicator
    If error reports go up after a culture initiative, treat it as a positive signal — the team is now surfacing what was always there. If error reports go down without any clear operational improvement, investigate whether reporting has been suppressed. An increase in reported errors in a high-trust environment is usually a sign of improvement, not deterioration.
    Pro tipCompare error rates to outcomes: a team with high reported errors and strong patient / customer outcomes is doing better than a team with low reported errors and mediocre outcomes.
    WarningDo not use error report rate as a performance metric in isolation — it must be paired with outcome data.

Checklist

Saved in your browser

Examples

2 cases
Edmondson's surgical team study

Edmondson measured surgical team culture quality alongside error rates. Teams with better communication, trust, and collaboration reported significantly more errors than teams with poor culture. The naive interpretation was that good culture caused errors; the correct one was that good culture revealed them.

OutcomeThe research established 'psychological safety' as a measurable construct and an operational management lever — now widely cited in organisational behaviour and replicated across multiple industries.
The watch sewn into the kidney

Harford uses dark humour: a bad surgical team's response to a mistake was 'nothing happened, no one speaks of this.' Someone sews a watch into a patient's kidney and the official record shows a clean operation. The error is real; only the reporting is absent.

OutcomeThe patient carries an undetected error that may compound. The team never learns. The official record looks clean.

Common mistakes

4 traps
Using low error count as the primary team performance metric
Edmondson's finding is counterintuitive: the best-performing teams had the highest error counts — because they reported them. Using reported errors as a performance metric without understanding whether reporting is suppressed is worse than useless.
Combining error reporting and blame in the same conversation
When a team member who surfaces an error immediately faces 'why did you let this happen?' in the same conversation, they learn that surfacing errors is personally costly. The next error will not be surfaced. Separating these conversations is structural, not optional.
Assuming culture change produces an immediate drop in errors
The first effect of improved psychological safety is more reported errors, not fewer actual ones. Leaders who interpret the increase in error reporting as evidence that the culture change is making things worse will reverse course and destroy the trust they built.
Treating psychological safety as a permanent achievement rather than an ongoing maintenance task
A single high-profile punishment for an honestly-reported error will suppress reporting for months. Psychological safety is fragile — it requires consistent reinforcement because it only takes one counterexample to re-establish the high personal cost of admission.

Origin story

How this framework came to be

Harford references Edmondson's research through his broader interest in mistakes as documented in his Cautionary Tales podcast. He uses it here to illustrate a counter-intuitive finding — the kind of result that C3 (curiosity) leads you to when you resist the obvious interpretation. He frames it as an example of going deeper rather than simply confirming the expected hypothesis: better culture equals fewer errors.

Source

Traced to primary
Source · PODCAST
Do We Need To Change Our Minds About Index Funds?
Tim Harford · 2025
Open source →

Related frameworks

Browse all Leadership →