The trait hiring wrote off.

The write-off of openness to experience is on the record twice: .04 in the meta-analysis that first made the Big Five usable for selection, and .05 in the 2022 re-estimates that reordered the field's validity table, beneath every other predictor listed. Yet openness to experience at work is also one of the two best Big Five predictors of training success, and among the larger personality differences separating creative scientists from their less creative colleagues. It is the trait that predicted who kept making good decisions when an experimenter changed a task's rules without warning. The weakest predictor on the general table holds the strongest claim to a specific niche — and how you resolve that inversion decides whether the trait belongs on your scorecard at all.

The resolution is not that the meta-analyses are wrong. It is that openness is a narrow trait, not a weak one. Overall job performance ratings are dominated by the routine core of work: task completion, reliability, meeting the standard that was set on day one. The outcomes openness predicts (generating ideas, absorbing new material, coping when the ground shifts) are real, consequential, and barely sampled by that criterion. Average a strong signal for a rare outcome into a rating built from common ones and you get .04.

Both misreadings squander signal. An organization that scores openness on every requisition as a general merit indicator is adding noise to most of its decisions. An organization that strikes the trait from every scorecard because of the .04 is discarding a personality signal that speaks directly to innovation, learning speed, and adaptation, in exactly the roles where those outcomes decide whether the hire worked.

When the rules changed, the trait rankings inverted

The sharpest evidence about openness comes from a laboratory task that was never designed as a selection study. LePine, Colquitt, and Erez (2000) had participants work a 75-problem decision task and changed the task's rules partway through, without announcement. Before the change, performance was predicted by general cognitive ability and by nothing else; openness was null, exactly as its reputation would predict. After the change, the ordering moved. Ability's relationship grew, and openness became a significant positive predictor of decision quality for the first time. The people high in openness were not better at the game. They were better at noticing the game had changed.

The unannounced change is what makes the design unusual. It separates performing a task from re-learning one, two capacities that ordinary performance ratings blend into a single number and then average across a year. Field data almost never pulls them apart. A supervisor rating covers the months when the work was stable and the weeks when it was not, weighted by how much of the year each occupied, which means the unstable weeks weigh close to nothing.

The mirror finding is the sharpest sentence in the study: after the change, conscientiousness became a significant negative predictor, an effect the authors traced to its dependability facets, order, dutifulness, and deliberation. The very dispositions that make a person reliable under stable rules kept them executing the old procedure after it stopped working. That is one experiment, on one task, and it does not overturn the broadest record in personality-based selection; the general case for conscientiousness stands and is made in conscientiousness at work. As a bounded result about change, though, it is exact: when the environment moved, the trait rankings inverted. Figure 1 sketches the design and the directions.

Most jobs eventually run the same experiment without a consent form. A market shifts, a toolchain gets replaced, a regulation rewrites the workflow, and the process that defined competence last year quietly stops working. The LePine result says the personality signal for that moment is not the one that predicts steady-state output, and that a scorecard built entirely for the steady state can select, with real fidelity, the people most exposed to the change. For change-dense work, that error has a direction: the scorecard systematically favors the people least equipped for the moment that decides the hire.

That experiment stays in view for the rest of this article, because it explains the shape of the meta-analytic record. Openness looked worthless right up to the moment the criterion changed to something the trait governs. What follows is that pattern in correlational form, and it begins with the numbers that closed the case.

The near-zero validity is a fair reading of the wrong criterion

Barrick and Mount's 1991 meta-analysis in Personnel Psychology pooled the accumulated studies relating the five personality dimensions to work criteria, and it is the source of the number that defined the trait's reputation: .04, averaged across criteria. An averaged validity is a blunt instrument, and the authors knew it. It pools whatever criteria the underlying studies happened to use, lets the common ones dominate, and allows a strong showing on one criterion to disappear into the mean of the others. They also cautioned that averaging correlations instead of compositing them makes the resulting figures underestimates across the board, so the fair statement of their openness result is that it sits close to zero; calling it harmful overstates what the table shows. Close to zero was enough. Conscientiousness became the general-purpose personality predictor, and openness became a footnote with a reputation problem.

The modern re-analysis did not rescue it. Sackett, Zhang, Berry, and Lievens (2022) put openness at .05 for overall job performance, the lowest of any predictor in their table, against .19 for conscientiousness. Their re-estimation applied less aggressive range-restriction corrections than earlier syntheses had used, and what that shift did to the predictor table as a whole is worked through in the general mental ability debate. The openness estimate also carries a lower 80% credibility value of −.03. That figure describes the spread of true relationships across settings, and what it says is that organizations where openness runs slightly negative on performance are entirely plausible. Where the full revised ranking lands is mapped in what actually predicts job performance; openness sits beneath every interview format, test type, and biographical signal the paper considered.

Two things inside those tables complicate the verdict. Figure 2 carries the first, which is that the criterion where openness performs is not overall job performance at all. The second is how the questions are framed: Sackett and colleagues also report .12 for contextualized measures, items that ask about behavior at work instead of behavior in general, more than double the generic estimate. The case for work-framed measurement is argued in the extraversion briefing, which meets that effect from the other side of the Big Five.

The layout explains why the near-zero reading keeps getting rediscovered. Every new reader meets the overall-performance rows first, in summary tables organized around exactly that criterion, while the training row sits deeper in the paper, where reputations rarely reach.

The low overall estimate is not a measurement failure waiting to be corrected away, and this article is not an attempt to rehabilitate a number. At .05 the trait cannot move a general shortlist, and a vendor implying otherwise is selling something the literature will not support. The claim is narrower and testable: change the criterion, and a measure that looked inert starts to do work.

One label, two appetites

Openness to experience describes a taste for novelty in three registers: ideas, aesthetics, and experience itself. High scorers gravitate to abstraction, imagery, and the unfamiliar; low scorers prefer the concrete, the proven, and the routine, a preference that is itself an asset in work built on consistency. Neither pole is pathology, and neither is a work ethic. The domain holds its seat among the five factors for the reason the other four hold theirs, decades of lexical and factor-analytic replication (the trait-versus-type foundations are laid out in the Big Five versus personality types).

The complication is that the label covers two distinguishable appetites. DeYoung, Quilty, and Peterson (2007), working in the layer between the five broad domains and their narrow facets, showed that each Big Five domain resolves into two correlated aspects, and that this one splits into Intellect, the pull toward ideas and abstraction, and Openness proper, the pull toward aesthetics and imagination. At work the halves read differently. Intellect shows up as appetite for theory, models, and problems nobody has framed yet; aesthetic Openness shows up as imagery, design sensibility, and comfort inside ambiguity that has no analytical handle. A research engineer and a brand designer can post the same domain score from opposite halves of it, and everything this article says about matching the trait to the role sharpens when the score is read at that resolution.

The split also explains some of the noise in the validity record. A domain score mixes a component that plausibly bears on analytical, learning-dense work with a component that bears on expressive and imaginative work. Whenever a study's criterion loads on one half only, the other half dilutes the correlation. The near-zero estimate is, in part, an averaging artifact stacked on an averaging artifact. Unbundling the halves is a measurement decision with practical consequences, and it is a decision a domain-level report has already made for you, in the wrong direction.

Openness to experience at work shows up on criteria the ratings barely sample

The trait's oldest and steadiest correlate is divergent thinking, the capacity to generate many varied ideas for an open-ended problem, and the linkage was established with unusual care. McCrae (1987) found the association held whether openness was measured by self-report or by observer rating, so it was not simply people who call themselves creative also calling themselves open; someone else's description of the person carried the signal too. And it survived controls for age, education, and vocabulary, so it was not intelligence or schooling in disguise. Two vantage points, the obvious confounds removed, and the result still standing: that is what a real trait-outcome connection looks like before anyone takes it to a workplace.

Feist (1998), in a meta-analysis of personality in scientific and artistic creativity, drew the more useful boundary. Comparing creative with less creative scientists, openness showed a median difference of .31 standard deviations, among the larger personality gaps in the analysis. But set scientists beside nonscientists and openness nearly disappears from the comparison; the trait that dominates that entry contrast is conscientiousness, with the largest difference in the whole analysis. Figure 3 holds both halves of the pattern. Openness does not mark who enters a field. It marks who does the creative work inside one.

The workplace criterion tells that story at correlation scale. Hammond, Neff, Farr, Schwall, and Zhao (2011) meta-analyzed predictors of individual innovation at work, the generation and implementation of new ideas inside a role, and put the openness relationship, meta-analytically, at about .24, several times the trait's validity for overall performance. The comparison is the point. Personality and innovation are genuinely connected, the connection runs through precisely the domain the general validity table ranks last, and both facts hold because innovation and overall performance are different criteria scored on one set of employees.

One more result tempers the enthusiasm. George and Zhou (2001) found that openness predicted supervisor-rated creative behavior only under conditions that let the trait express: when feedback was positive and the task was heuristic rather than algorithmic, open-ended in method instead of fixed in procedure. Where either condition failed, the open employees were not visibly more creative than anyone else. A trait is a potential, and this one appears to need situational oxygen. That has a hiring implication all by itself: selecting for openness while running procedural, feedback-starved work buys a mismatch.

The training column disagrees with the performance column

In Barrick and Mount's tables, openness correlated .25 with training proficiency, and the fair reading requires the neighboring number: extraversion came in at .26, essentially tied. There is no single training trait, and this article will not pretend openness is it. What the tie establishes is enough: the domain written off at .04 for performance is, for learning outcomes, in the top pair, and any account of personality and trainability that leaves it out is reading the wrong criterion column. The tie makes sense on its own terms, too, since training environments reward both the appetite for new material and the willingness to engage out loud, and the two traits supply one each.

The gap between the two criteria is not mysterious. A training program is a stretch of work in which nearly everything is novel by design: new material, new methods, no accumulated routine to lean on. It is, in effect, the openness band of a job compressed into weeks, which makes training proficiency the closest thing the classic literature has to a purified criterion for how people meet the unfamiliar. Steady-state performance ratings then measure the years afterward, when the novelty has drained out and the trait goes back to sleep.

Where openness belongs on a scorecard, and where it is noise

Everything above resolves into a placement rule for openness to experience at work. The trait belongs on a role's scorecard when the role sits in its band: an explicit innovation mandate, where the Hammond and Feist results apply; a learning-heavy early-career trajectory in which most of the job's knowledge is still ahead of the hire, where the training result applies (this is much of what a campus hiring program is selecting for, whether it says so or not); or an environment where the operating rules change often enough that adaptation is a first-order requirement, where LePine's result applies. Outside that band, in stable, procedural, well-specified work, the .04 is the truth about the trait, and scoring it is measurement theater.

Two roles can carry one job title and still sit on opposite sides of that line. A support engineer keeping a mature product running and a support engineer standing up a new one are indistinguishable on an org chart and unalike in this evidence, and only one of them has a reason to ask the question at all. The unit that determines whether openness belongs in the scoring is the role as it will actually be worked, down to the mandate the hire inherits on the first day.

Inside the band, the question needs asking at the right resolution. An Intellect-loaded role, one that runs on abstraction, analysis, and appetite for ideas, is not an aesthetics-loaded role, and a domain score cannot tell you which half of the trait a candidate brings; the general case for measuring below the domain is made in facet-level personality measurement. A role profile that names the aspect it needs will read a candidate file differently, and more accurately, than one that stops at the domain.

This conditional structure is not unique to openness, and reading it as one trait's oddity misses the pattern. Agreeableness shows the same signature, helpful in some collaborative structures and costly in others, and the agreeableness collaboration paradox maps that terrain the way this article maps openness. The Big Five do not divide into strong traits and weak ones. They divide into broad signals and narrow ones, and the narrow ones repay as much thought as the organization puts into naming the criterion. A hiring team that can say which outcomes a role actually needs, in writing, before the requisition opens, has already done most of the work of deciding whether openness deserves a column.

Weight openness where the work is unsettled

For organizations, the evidence supports one scoring stance: openness as a role-conditional weight, set by the job's innovation and learning demands, never as a general merit signal. In practice the sorting is less subtle than it sounds. The trait gets real weight when the role description reads build, learn, redesign, ambiguous, and near-zero weight when it reads execute, comply, maintain. A platform that lets a role profile carry that weight explicitly, the way modern assessment platforms structure role-referenced scoring, makes the stance operational instead of aspirational, and it removes the temptation to hand-adjust a candidate's standing because an interviewer liked their curiosity.

Pair the trait with ability whenever the read is trainability. In the adaptation experiment, cognitive ability predicted before the change and even more strongly after it; openness added its signal on top of ability, not instead of it. The two answer different questions: ability speaks to how fast the learning happens, openness to how willingly the learner walks into material that resembles nothing they already know. A trainability screen built on openness alone ignores the strongest predictor in the study, and the broader case that ability anchors any learning-heavy selection decision is the flagship argument of this series.

Situation moderation belongs in the hiring plan itself. The George and Zhou result means an open hire delivers creative behavior when the work is heuristic and the feedback constructive, and delivers little visible difference when it is neither. If the role's daily reality is procedural and its feedback culture punitive, the mismatch is predictable in advance, and the failure belongs to the organization's design. The assessment did its job. The interactionist reading runs both directions: hire for the trait and build the conditions, or do neither.

The write-off itself still protects against the opposite error, and it should stay in view. A candidate's high score on openness to experience at work is not evidence of general excellence, work ethic, or judgment; the table says as much, twice. It is evidence about ideas, learning, and response to change, a narrow claim with unusually good backing (the psychometric foundations behind trait measurement are documented in the science of structured assessment). The bottom row of the validity table turned out to be a mislabeled row, not an empty one: the field averaged a specialist trait across generalist criteria and mistook the average for the trait.