Teams got the hiring decision right 15 percent of the time when the newcomer held the information the group needed, and 28.7 percent of the time when an original member held it — and the originals had been talking to each other for five minutes

The setup

How long does a group have to exist before it has an inside?

Two people met in a text chatroom and spent five minutes discussing what would make someone a good airline pilot. A third person, the one who would become the newcomer, watched a five-minute nature video instead. Then all three were told they would pick one of four applicants for a pilot job.

Across 370 such teams, when the unshared information the group needed sat with an original member, 28.7 percent of teams chose the right applicant. When it sat with the newcomer, 15 percent did. The design and the numbers come from Sydney Phlegar, Kareena del Rosario, Oana Dumitru and Tessa West, whose study of newcomers in decision-making teams appeared in Communications Psychology in June 2026. The journal has so far published it as an accepted manuscript ahead of final editing. The authors’ own reading is that five minutes is already enough to matter. No team was run without those five minutes, so what the experiment actually varied is who held the information, and the newcomer’s assigned status.

Everyone received the same positive and negative notes on each applicant, and those shared notes made one applicant, Candidate C, look like the weakest of the four. One person in each team also received extra information that made Candidate C, in the paper’s words, “clearly the best suited for the job”. In the researchers’ own pilot testing with 159 people, someone holding the full set of information saw Candidate C as the best option 75 percent of the time.

So the information the group needed was inside every team. It simply sat with one person, and the experiment varied which person that was: either one of the two originals, or the newcomer. The three then had ten minutes to talk.

Behind the two percentages

In the paper’s Bayesian model, the gap between 28.7 percent and 15 percent is a coefficient of -0.77, with a 95 percent credible interval running from -1.26 to -0.26. That interval excludes zero, which is the paper’s own threshold for treating an effect as reliable. The authors also report that a conventional frequentist analysis returned nearly identical results.

The analysed sample is 1,110 participants, recruited through Prolific, all United States citizens with English as a first language, aged 17 to 79 with a mean age of 35, and 56.8 percent women to 40.8 percent men. Seventy-three of the 443 recruited teams were dropped before analysis: someone in the team suspected the status feedback was not real, or someone typed nothing at all in the chat, or a connection dropped.

One thing the paper states plainly: the hypotheses, design and analyses were not preregistered. The predictions came from existing theory rather than from a filed plan, which means the result is a finding to be tested again rather than a confirmation of something already committed to in advance.

A chatroom is not an office

A typed chatroom has no eye contact, no interruptions, no differences in how long anyone speaks, and none of the small physical signals through which rank is usually conveyed. Strip all of that out and the effect of being new can be read apart from other social identity characteristics, which is the point. The authors suggest the missing signals might also explain in part why the status manipulation did not override newcomer status: prestige had no channel to travel down. They add a second partial explanation: a group normally infers a member’s prestige over time rather than being told it, so handing someone an explicit percentile is not how earned prestige usually arrives.

Each team had exactly one newcomer, so being new and being the only one of anything are tangled together. The initial group was a pair, and pairs may be unusually hard to break into, since two-person bonds tend to run on rapport in a way that larger groups do not. The authors push back on their own point here, noting that the newcomer penalty also turns up in research on people joining larger teams, so they do not think it is an artefact of starting with a dyad. And this is one task, one US sample, ten minutes of typing.

The function-word score

The measure the paper reaches for is linguistic coordination, computed with the ConvoKit toolkit over eight categories of function word, including articles, prepositions, conjunctions and pronouns. The paper’s own example: a speaker’s score of .10 on articles could mean they use articles ten percentage points more often when a partner has just used them than they otherwise would. It is a measure of accommodating to someone else’s speech habits, and the authors describe it as subtle and unconscious. They also caution that it is not one psychological process, since accommodation can serve rapport-building or persuasion, and in a decision task those two goals pull against each other. And the paper’s main text reports no analysis linking coordination scores to the decision itself.

Here the scoping matters more than the direction. There was a credible interaction between role and holding the critical information, and inside the newcomer group it pointed one way: newcomers who held the critical information coordinated about two percentage points less than newcomers who did not. Among the original members, holding it made no credible difference.

What the study did not find is a general newcomer effect. Newcomers as a class did not coordinate credibly less than original members: the role effect came out at -0.0002, with an interval running from -0.0060 to 0.0055 and straddling zero. The newcomer-versus-original comparisons inside each information condition were not credibly different either, and both of their intervals include zero. The authors call the effects small, and they are. The one other credible result on this measure was that high-status members coordinated roughly half a percentage point less than low-status ones.

Friction, self-reported

The self-reported side lines up and complicates itself at the same time. Newcomers holding the critical information reported more task conflict than newcomers who did not, and original members showed no credible difference either way. Among the people who held no critical information, though, it was the originals who reported more conflict than the newcomers. That is the opposite of a simple story in which the new person always feels worse. And among the people who did hold it, newcomers and originals were not credibly different from each other.

That last null is worth pinning down, because the paper’s own discussion section is looser than its results section. The discussion says newcomers holding critical information perceived more conflict “than others overall”, which the results do not support: the comparison that reached credibility was against other newcomers, not against original members. This account follows the results.

The authors read the pattern as a split experience. A newcomer struggling to get a point taken up may register friction, while the majority, talking mostly about information they all share, may find the same conversation smooth.

The status boost with no detectable effect on the decision

Every participant sat a fake assessment called the Person Insight Test and was told they had scored in either the 40th or the 80th percentile. Assignment to a percentile was random, with the two originals always split one high and one low, and the ranks were put in front of people again before the three-way task. Manipulation checks confirmed recall tracked the rank people had been given; belief was handled by exclusion rather than measurement, since any team where someone said they suspected the status feedback was not real was dropped.

It did not rescue the decision. Teams where a low-status member held the critical information chose correctly 16.0 percent of the time; where a high-status member held it, 23.5 percent. The coefficient is 0.45 with a credible interval from -0.07 to 0.98. That interval crosses zero, so the study did not detect a status effect on the outcome. The authors also report that no credible interaction with status turned up anywhere in their models, though those models sit in a supplementary note rather than the main text.

The interval’s upper end still allows a real advantage this sample could not resolve either way, so the finding is an absence of detection rather than a demonstrated absence. The authors do not read it as rank being beside the point: they point to the missing signals, and to the artificial way the ranks were conferred, as partial explanations, and they note that the manipulation moved one thing, the coordination scores, so it was not inert.

The 2009 newcomer study this site covered in August

In August, Silicon Canals covered Phillips, Liljenquist and Neale’s 2009 experiment on socially distinct newcomers. Its comparison is narrower than our August headline suggested: every four-person group in the study received a newcomer, and the contrast was between groups whose newcomer came from outside their own social group and groups whose newcomer came from inside it. The out-group version performed better while reporting less confidence and rating its own interaction as less effective. And the abstract is specific about what the gain was not: the performance gains “were not due to newcomers bringing new ideas to the group discussion”.

That last clause is the one that matters here. In 2009 the group with the socially distinct newcomer outperformed the group with the socially similar one, and the gain was not down to the newcomer’s ideas. In 2026, when the unshared information was the newcomer’s to supply, the team reached the right answer 15 percent of the time. Phlegar and colleagues cite the 2009 work themselves, for a narrower point still: newcomers who resemble the existing group often end up quietly adopting its view.

Ten minutes of typing in a chatroom is the whole of the group task that produced these numbers. A newcomer on a real team has a manager, a title, a reason to be there and more time than that, and none of those were present to be measured. Nobody writing this has sat in on the version that matters. Neither paper compares a team that received a newcomer with one that did not. Data and analysis code for the 2026 study are posted on the Open Science Framework.

Five minutes bought two strangers an inside. A newcomer holding the answer spent ten minutes outside it, coordinating a little less than newcomers who held no critical information, reporting a little more friction than they did, and getting the group to the right candidate 15 percent of the time. The identity of newcomer, the authors write, “can form rapidly”. It does not wait for the information to arrive.