In a hypothetical advertising decision involving 142 US MBA students, stronger shifts in ratings of four information items, pretested to be approximately neutral on average, were associated with larger changes in confidence towards the preferred option. The relationship was statistically significant but modest. Because the task used one fixed information order and did not manipulate distortion, it did not establish that rating shifts caused confidence to change. The campaign decision also had no objectively correct answer.
The paper, Information distortion as a source of overconfidence in managerial decisions, was written by Peter J. Boyle of Central Washington University and Pete Nye of the University of Washington Bothell. Cambridge University Press published it online in the journal Judgment and Decision Making on 9 February 2026 as volume 21, article e3.
Boyle and Nye use “information distortion” for ratings that favour the current leader more than an objective assessment of the information would warrant.
The authors propose that this slanted evaluation can provide additional support for the leading option, which may then increase confidence in that option. Their new study measured whether the two patterns moved together while a business decision developed.
MBA students assessed a fictional advertising campaign
The researchers combined two samples of MBA students from US universities, one with 89 participants and another with 53. Both groups completed the same core task.
In the second sample, 70% reported prior full-time work experience, and the median reported experience was five years. The paper does not provide a corresponding work-experience figure for the first sample.
Participants took the role of a senior marketing executive at a fictional consumer-products company. They had to decide whether to launch a magazine campaign claiming that an oat-bran cereal reduced cholesterol.
The campaign could increase the brand’s market share, but the claim risked misleading consumers. The fictional case included mixed medical and commercial information. For example, a low-fat diet that included the cereal had reduced cholesterol, but the evidence did not establish that oat bran produced the benefit. Consumer testing also suggested that the proposed advertisement could lead people to infer more medical support than existed.
The task had an ethical dimension and no objectively correct launch decision. The researchers were interested in how participants evaluated the information and built confidence, not whether they approved the campaign.
Four information items were pretested for neutrality
After an introductory scenario, participants received four additional information items one at a time. Each paragraph contained points supporting and opposing the campaign.
More than 500 people took part in several rounds of pretesting. Each pretest respondent assessed only one item and did not make the launch decision. An item qualified as approximately neutral when its average rating was within one point of the midpoint on a nine-point scale.
This calibration applied to the four additional items, not to the introductory scenario.
After each new item, the MBA participants rated whether that information favoured launching the campaign. They also identified the option currently leading and reported how certain they were that they would ultimately launch.
The main analysis compared trends in four information ratings with trends in five interim confidence reports, one after the introduction and one after each additional item. A separate confidence rating recorded after the final choice was not included in that analysis.
Evidence ratings and confidence moved in the same direction
The central result was a positive correlation between the trend in participants’ information ratings and the trend in their confidence. The correlation was r = 0.279, with a one-sided p value of 0.0004.
The two trend measures were signed. Positive values represented movement towards launching and negative values represented movement towards not launching. Their positive correlation therefore captured information ratings and confidence moving in the same direction. The paper did not report separate effect estimates for the launch and no-launch groups.
The authors interpret the relationship as evidence that a desire for cognitive consistency can influence how incoming information is evaluated. Once an option begins to lead, later evidence may look more favourable to that option, and that interpretation may provide additional confidence.
The statistical test itself was correlational. Information distortion was not randomly assigned, so the experiment does not isolate a causal effect of distorted ratings on confidence.
No starting preference was assigned
The design did not assign a starting preference. Boyle and Nye argue that the pattern can emerge while a choice is taking shape, without a settled initial view.
The paper calls this predecisional information distortion, distinguishing it from confirmatory actions taken after a choice.
Self-ratings did not identify the largest confidence changes
The second sample of 53 participants completed an additional question about whether they had evaluated the information without bias.
Their self-ratings were almost entirely unrelated to the size of their confidence change during the task. The correlation was r = -0.005, with a one-sided p value of 0.49.
This was a narrow result based on one self-report item. It did not test whether a structured self-review could reduce bias, and the self-rating was compared with the confidence trend rather than directly with the study’s information-distortion measure.
The same 53-person sample completed measures of need for cognition, a preference for effortful thought, and tolerance for ambiguity. Neither score was significantly associated with the size of the confidence change. The correlations were r = -0.020 for need for cognition and r = -0.039 for tolerance for ambiguity, with p values of 0.44 and 0.39 respectively.
Boyle and Nye describe the trait analyses as exploratory and caution that the small sample does not support a firm conclusion about either characteristic.
Business risks remain possible rather than demonstrated
The study raises a practical distinction for companies: increasing confidence does not necessarily mean that the underlying evidence has become more diagnostic.
Boyle and Nye suggest that a manager whose confidence rises quickly might stop seeking information before contradictory evidence appears. The experiment did not test that possibility. Every participant received the same amount of information and could not decide when to end the search.
The order of information may also matter. Earlier controlled research has shown that placing information favourable to an inferior option early in a sequence can establish that option as the leader and influence later evaluations. The new MBA study did not manipulate presentation order, so it cannot estimate an order effect.
Earlier experiments reduced, but did not remove, distortion
The 2026 study did not test a management intervention. In their discussion, the authors point to two approaches examined in previous experiments.
A 2012 group-decision study found that competing preferences among group members kept distortion and confidence lower while the disagreement continued. Once a consensus emerged, both began to rise.
A separate five-study paper found that priming a counterarguing mindset reduced information distortion. Its fifth study also found lower commitment to the preferred option. Boyle and Nye describe the work as evidence of lower attendant distortion-driven confidence.
Neither approach eliminated the pattern. The evidence also does not establish that the same procedures will improve investment, product-launch or strategy decisions inside companies.
Study used MBA students in a fixed-order, hypothetical task
The participants were MBA students, not a representative sample of senior executives. The task was fictional, and no participant faced financial, legal or career consequences.
All participants received the information in the same fixed order. The authors acknowledge that the sequence may have contributed to the changes in confidence. They also note that the ethical theme running through the items may have given one option an early advantage.
The study did not define a point at which confidence became overconfidence, nor could it judge decision accuracy because the campaign choice had no objectively correct answer. The authors’ narrower position is that the high confidence observed was not wholly justified by an objective evaluation of the available information, including four attributes pretested to be approximately neutral on average.
Boyle and Nye call for further research using counterbalanced sequences of unrelated information. They also say larger samples are needed before drawing conclusions about self-awareness or individual traits.
