In 2010, three Italian researchers won the Ig Nobel Prize in Management for a curious result: in their computer model, organizations could work better by promoting employees at random. Nine years later, a study of real companies made the joke seem a lot less absurd.

Alessandro Pluchino, Andrea Rapisarda, and Cesare Garofalo revisited the Peter Principle: people get promoted for being good at one job, even though the next calls for different skills. In their model, random promotions could work better than promoting top performers if performance at one level was unrelated to performance at the next. The result depended on that assumption, which may not hold in real workplaces. [14]

In 2019, economists Alan Benson, Danielle Li, and Kelly Shue published a study of nearly 39,000 salespeople at 131 US companies. Over seven years, 1,553 were promoted to management. Companies mostly picked their top sellers. Sales were easy to measure, and the decision was easy to justify.

But the data revealed an uncomfortable pattern: the better people had sold before their promotion, the less their subordinates tended to sell afterward. Among those whose teams could be assessed, managers who had sold twice as much before promotion tended to have subordinates who sold about 6.1% less per person, on average. [1]

The companies could measure sales performance, but had much less evidence about managerial ability. The same problem can arise when several candidates qualify and the remaining differences in their scores say little about how they will perform.

A German research foundation tried something different: it combined expert selection with a draw from a real lottery drum.

85 proposals in the lottery drum

In fall 2017, the Volkswagen Foundation considered 594 proposals for unconventional research projects. An initial screening narrowed the field to 119. An expert committee then judged 85 eligible for funding and selected 17 projects of its own.

The numbers of all 85 eligible proposals, including the 17 the experts had already selected, went into the lottery drum. Seventeen numbers were drawn. The drum and the committee managed to agree on five projects. That left 29 distinct projects to fund. [9]

Volkswagen Foundation: how the grants were awarded, and what followed
594applications received

85passed two rounds of review

17selected by the committee
17drawn from the lottery drum
Five applications appeared on both lists

29distinct projects received funding
Across the full program, who came out ahead?
A tie on publications.Recipients selected by experts and by lottery published research at the same rate.
Selected by experts60%published research
Selected by lottery60%published research
After mixed selection was introduced≈17% → ≈25%the share of women among grantees increased
Program evaluation: a lottery can work alongside expert review. Neither route had an apparent advantage in the share of recipients who published, and the grantee pool became more diverse.
Figures cover the whole Experiment! program, not just the 2017 drawing. Volkswagen Foundation, 2024 evaluation [15].

The idea was not confined to Germany. New Zealand’s Health Research Council also used a lottery after screening proposals, an approach one recipient later defended. [7] [8]

The Volkswagen Foundation kept one detail from the winners: they were never told whether the committee or the lottery had chosen them. It wanted to avoid splitting researchers into those who had “earned” their grants and those who had “just been lucky.” [10]

How choosing a leader changes a team

Inside a team, appointing a manager changes who gets heard. The choice itself can signal to colleagues that this person’s judgment deserves more weight.

In a 2020 experiment, 176 people in 44 teams worked on numerical problems. Leaders were appointed at random, chosen for the accuracy of their earlier answers, or chosen for their confidence. Teams led by the most confident participants made worse use of what their members knew than teams with randomly appointed leaders. Choosing leaders for accurate earlier answers showed no convincing advantage either. [2]

At a project meeting, the manager gives an estimate first. One colleague knows the equipment costs; another has just finished a similar project. Their information is useful, but the boss’s number is already on the table.

In the experiment, participants tended to revise their initial estimates too little. When a leader was presented as especially accurate or confident, that person’s opinion received too much weight, leaving other team members’ knowledge underused. The experiment showed this weighting effect more clearly than any change in the accuracy of teams’ final answers. [2]

A small 1998 laboratory experiment, far removed from real-world hiring, pointed in a similar direction: groups whose leaders were appointed alphabetically performed better on a survival task than groups with systematically selected leaders. [3]

When careful selection matters

For a safety-critical job, you start by checking who has the necessary training. If a skill matters and you can test it, use the test.

For managers, structured interviews and exercises resembling the actual job can help predict performance. [11]

A 2026 study involved 555 participants. Those most eager to lead showed no convincing advantage over randomly appointed leaders. Tests of particular skills, including economic decision-making, were more useful for predicting management quality. [4]

The researchers also studied 225 working store managers. They used the observed links between skills, management quality, and sales to model a different way of selecting managers. The projected gain was roughly 16% in annual sales over existing practice. They did not implement that policy or compare it with a lottery.

A separate field experiment with student teams found that appointing capable, hardworking leaders increased effort and knowledge-sharing. Giving leaders concrete instructions on coordinating the team had an even greater effect. [13]

What we found when we checked the data

At first glance, two studies looked like wins for random selection. We wanted to see how well those conclusions held up in the underlying data.

In an Indonesian experiment, participants played officials deciding how to divide a group’s money. Those selected by exam gave a smaller share to others than those selected by lottery. In the narrower sample, the gap was −9.4 percentage points. We expected the broader sample to sharpen the answer. Instead, when we included all usable money-allocation decisions, the gap shrank to −4.5 points and its confidence interval crossed zero. The evidence for a difference had become much less convincing. So much for the dramatic headline. [5]

Fischer and colleagues’ study gave us a different problem. We reproduced the published effect, then looked at how teams had been assigned: the two formation rules were applied across just four classes. That leaves very little ground for a confident causal claim. [6]

The calculations, including the alternative sample and a test of the four-class design, are below the interactive tool.

When to stop ranking candidates

First, screen candidates against a minimum standard: who has the knowledge, experience, and skills the job requires? Next, test the prediction: do the remaining differences among qualified candidates reliably tell us who will perform better? If so, use that evidence. If not, draw lots among the qualified rather than keep adding interviews and scores.

Passing the minimum standard is only the first test. A lottery makes sense when extra ranking adds little predictive value. You still have to weigh the cost of a mistake, incentives to perform, and the team’s trust in its leader. Promotions encourage effort; making every promotion random could weaken that incentive. [12]

A lottery cannot replace knowledge. It can stop us from pretending to know more than we do.

Try drawing lots

If you have checked your options and all of them would work, but you have no reliable reason to favor one, let chance decide.

Your options

Options entered: 0 of 2

How we checked the studies

Here is what we recalculated and how. We used the original data for both studies. Reproducing their analyses, however, is not the same as independently rerunning the experiments.

Kuipers (2026): selection by exam versus lottery

We compared the share of money experimental officials gave to others when they were selected by exam or by lottery. First we restricted the analysis to groups that completed the full experiment; then we included every group with usable results from the money-allocation task.

Our sample Exam selection compared with lottery selection
Groups that completed the entire experiment −9.4 percentage points; 95% CI: −16.5 to −2.4
All usable money-allocation decisions −4.5 percentage points; 95% CI: −10.5 to +1.4

A negative value means that exam-selected participants gave a smaller share of the money to others. These are two analyses we ran on the same outcome, not two effects reported by the original author. About 55% of lottery-appointed groups and 48% of exam-selected groups completed the full experiment. This high dropout rate limits confidence in the findings. The money-allocation task was only one of several outcomes, so the risk of multiple testing matters too. The change between our estimates shows how sensitive the result is to the sample used; it does not establish that the original study was wrong.

Fischer et al. (2023): randomly formed student teams

Using the original OSF data, we reproduced the published coefficient: self-selected teams scored −0.4151 standard deviations relative to randomly formed teams on written assignments. But the team-formation rules were applied to only four university classes, one class per rule in each of two cohorts. The study describes students being randomly assigned to classes; it does not establish that the rules themselves were randomly assigned to those classes.

To test sensitivity, we hypothetically swapped the two rules between the classes within each cohort. This leaves only four possible assignments, so the smallest attainable one-sided p is 0.25; the conditional two-sided p for the observed difference is 0.50. These are not corrected p-values for the study. The permutation exercise shows the limits of a hypothetical class-level randomization scheme, not the properties of the actual study design.

In both cases, reproducing a number, establishing a causal effect, and applying that result to real workplaces are different questions. The original data are linked in references [5] and [6].

References

  1. Benson, A., Li, D., & Shue, K. (2019). Promotions and the Peter Principle. Quarterly Journal of Economics. DOI.
  2. Büchel, B., Klößner, S., Lochmüller, M., & Rauhut, H. (2020). The Strength of Weak Leaders. Experimental Economics. DOI.
  3. Haslam, S. A., et al. (1998). Inspecting the Emperor’s Clothes. Group Dynamics. DOI.
  4. Weidmann, B., et al. (2026). How Do You Identify a Good Manager? Quarterly Journal of Economics. DOI.
  5. Kuipers, N. (2026). Bureaucratic Selection and Weberian Values. Journal of Politics. DOI; Harvard Dataverse data.
  6. Fischer, M., Rilke, R. M., & Yurtoglu, B. B. (2023). When, and Why, Do Teams Benefit from Self-Selection? Experimental Economics. DOI; OSF data.
  7. Adam, D. (2019). Science Funders Gamble on Grant Lotteries. Nature. DOI.
  8. Ackerley, D. (2020). Grant Lottery Systems: A Winner Responds. Nature. DOI.
  9. Volkswagen Foundation. Give Chance a Chance.
  10. Research on Research Institute. The Experimental Research Funder’s Handbook.
  11. Sackett, P. R., et al. (2022). Revisiting Meta-Analytic Estimates of Validity in Personnel Selection. Journal of Applied Psychology. DOI.
  12. Lazear, E. P., & Rosen, S. (1981). Rank-Order Tournaments as Optimum Labor Contracts. Journal of Political Economy. DOI.
  13. Chemin, M. (2021). Does Appointing Team Leaders and Shaping Leadership Styles Increase Effort? Journal of Economic Behavior & Organization. DOI.
  14. Pluchino, A., Rapisarda, A., & Garofalo, C. (2010). The Peter Principle Revisited: A Computational Study. Physica A, 389(3), 467–472. DOI; confirmation of the 2010 Ig Nobel Prize in Management.
  15. Volkswagen Foundation (30 May 2024). Everyone Is Equal in the Lottery Drum. Evaluation of the Experiment! program: grantee publications and the composition of funded researchers. Foundation report.