Mitigating Correlated Errors

Mitigating correlated errors means actively preventing a group from making the same mistake for the same reason. For example, if a team has only read a single positive report about a project, a good strategy is assigning someone to actively find flaws, as this breaks the shared bias. While the 'wisdom of crowds' can cancel out random individual errors, it fails when everyone is influenced by the same flawed information, causing their errors to align and amplify.

Ask this: Before our team agrees, what common information might be biasing all of us in the same direction?

The effectiveness of group judgment, often called the 'wisdom of the crowd,' relies on a critical assumption: that individual errors are independent and will cancel each other out when averaged. Correlated errors violate this assumption. They occur when a group's members are influenced by the same misleading data, persuasive leader, or shared cognitive bias, causing their judgments to err in the same direction.

Think of it as the difference between weighing an object on ten slightly different scales versus ten scales that were all calibrated incorrectly in the exact same way. In the first case, the average weight will be highly accurate. In the second, the average will be precisely and confidently wrong.

To mitigate this, we must introduce cognitive diversity and structural independence into the decision-making process. Key strategies include:

• Seek diverse information: A hiring committee where members conduct separate, structured interviews and review different references is less likely to fall for a candidate's superficial charm than one where everyone sits in the same room for one unstructured chat.
• Judge alone, then debate: Before a group discussion, have every member commit to their own opinion or forecast in writing. This prevents an influential early speaker from anchoring the entire group's thinking.
• Assign a 'devil's advocate': In a project planning meeting, formally assign one person the role of critiquing the plan. This legitimizes dissent and forces the team to confront potential flaws they might have collectively ignored, a technique often used in high-stakes environments like intelligence analysis.

The study of correlated errors is central to improving collective intelligence and avoiding the pitfalls of groupthink. Research distinguishes between 'noise' (random, independent error) and 'bias' (systematic, correlated error). While averaging judgments is effective at canceling noise, it is useless or even harmful against bias, as it simply reinforces the shared error. Effective 'decision hygiene' involves processes that intentionally decorrelate judgments before aggregating them.

• Kahneman, Sibony, & Sunstein (2021) argue that organizations must practice 'decision hygiene,' such as using independent assessments before group meetings, to reduce both noise (random error) and bias (correlated error) in judgments. [1]
• Tetlock & Gardner (2015) found that elite 'superforecasters' excel not just individually but by using aggregation methods that leverage diverse, independently formed perspectives to cancel out errors. [2]
• Sunstein & Hastie (2014) identify four key problems that cause groups to fail: amplifying individual errors, cascade effects, polarization, and focusing only on shared information, all of which create or worsen correlated errors. [3]

Implementing strategies to mitigate correlated errors can be time-consuming and may feel inefficient for routine decisions. Forcing dissent with a devil's advocate can sometimes create artificial conflict or be dismissed as performative if not managed skillfully. Furthermore, in a hyper-connected world where news and narratives spread rapidly, achieving true informational independence among team members can be exceptionally difficult.