Abstract
The question this briefing answers is: how much of a performance rating reflects the person giving it rather than the person receiving it, what psychological processes produce that, how far personality on either side adds to it, and what the person rated, the person rating and the organisation can do differently?
It sets out the evidence on how much of a rating belongs to the rater, including the recent argument that the older figures overstate it, and then describes what raters bring to a rating: their temperament, their purpose, their investment in the person, their first impression and their beliefs about whether people change, their sense of who is like them, their own security, and the story the organisation tells about merit. Each is labelled for whether it rests on tested evidence or on theory.
Three chapters weigh personality on both sides of the form, including how personality and networks decide who is seen, with the limits stated. A late chapter covers the people most often mislabelled, including the reader. The practical chapters end with what organisations can design, and say plainly where no trial has been run.
Keywords: performance review; performance appraisal; performance ratings; rater bias; idiosyncratic rater effect; halo effect; rating leniency; escalation of commitment; implicit person theory; rater personality; Big Five; perceived similarity; homophily; social networks; self-monitoring; narcissism; self-enhancement; meritocracy; feedback.
PART I. THE VERDICT
Chapter 1: Two Managers, One Year
1.1 The Reorganisation
The scenario below is a composite. It draws on recurring patterns in research and advisory practice, and it describes no real, identifiable person or organisation.
For two years the pricing analyst at a mid-sized insurer had been rated exceeds expectations by the same manager, a former actuary who valued accuracy, read every model line by line and said little in meetings. The analyst was quiet too, and as careful as he was, and the two of them had settled into a working rhythm in which she brought the numbers early, he checked them, and nobody outside the team heard much about either of them. Her ratings had been the highest in the team in both years, and nobody had questioned them.
In January the pricing and underwriting teams were merged, and the analyst found herself reporting to the head of underwriting, a fluent, energetic presenter who had argued during the merger for a different person to take her role. Her work did not change. She built the same models to the same deadlines, the error log showed nothing new, and the two pricing reviews she led that spring went through without a query.
1.2 The New Verdict
Her year-end rating was meets expectations. The written comments said that her technical work was strong and that she needed to show more strategic presence and visibility with senior stakeholders. Nothing in the review referred to a missed deadline, a mistake or a complaint, because there had been none.
She asked her former manager, now in another part of the business, whether she had misjudged her own year. He said he would have rated it the same as the previous two and that he could not see what had changed. When she asked, her new manager said the rating reflected the whole picture and that the bar was higher on the combined team.
Both managers were honest, competent and sincere, and if asked, each would have said that his rating described her work. That is the reason to start here.
The analyst did what most people do with a rating like that: she took it as information about herself and spent the following months trying to become more visible, presenting more often and volunteering for a cross-functional project she had no particular interest in. Her models were checked less carefully that year because the new manager didn't read them line by line, and two small errors made it into a client file. Her next rating was lower again.
1.3 The Question Worth Asking
The usual questions about a case like this are whether the new manager is biased, or whether the analyst simply needs to be more visible. Neither gets to what happened. The more useful question is psychological: when two capable people watch the same year of work and reach different verdicts, how much of each verdict describes the work, and how much describes the person who wrote it?
Part of the answer, as later chapters show, lies in a difference the analyst never thought about: her first manager worked the way she did, and her second did not.
Stated plainly: the analyst’s rating changed when her rater changed and her work did not, and neither rater would describe what they did as anything other than an honest assessment.
The previous briefing in this series was about what happens to the people who raise a concern, and it framed the cooler appraisal as one of the quiet forms of retaliation. This one is about every appraisal, and about the person holding the pen.
Chapter 2: How Much of a Rating Is the Rater
2.1 The Share That Belongs to the Rater
Most organisations treat a rating as a measure of the person rated, with error that cancels out over time. The clearest test of that assumption separated the parts of a rating statistically, in two large groups of managers, 2,350 in one and 2,142 in the other, each rated on three areas of performance by two bosses, two peers, two of their own staff and themselves (Scullen, Mount & Goff, 2000), and what the authors called the idiosyncratic rater effect, meaning the part of a rating that comes from the particular rater’s own way of seeing and scoring people rather than from the person rated, was the largest single part: 62 per cent of the differences between ratings in the first group and 53 per cent in the second.
The performance of the person being rated, overall and on each area, accounted for 21 and 25 per cent. Differences between bosses, peers, and staff as groups were small, and random error accounted for most of the rest. Read plainly, a rating in those data carried more than twice as much of the person giving it as of the person receiving it. That is the finding this briefing is built on.
How solid is this: evidenced, from two large samples analysed with the same method. The ratings were collected for development rather than for pay, and the next section explains why the size of the share is still argued over.
2.2 Two People Watching the Same Work
A second way into the same question asks how far two supervisors agree when they rate the same person. A meta-analysis, which combines the results of many earlier studies rather than running a new one, found that agreement between supervisors on overall performance was moderate at best, clearly lower than the agreement of one supervisor with themselves over time (Viswesvaran, Ones & Schmidt, 1996). For a quarter of a century, that figure was quoted as evidence that ratings are mostly noise.
The figure has since been revised upwards. Pooling 132 samples with a method that stops a few very large studies from dominating, a 2024 meta-analysis found interrater agreement, the extent to which two raters put the same people in the same order, noticeably higher than the older estimate, and lower for managerial jobs than for others (Zhou, Sackett, Shen & Beatty, 2024). A second analysis restricted to pairs of raters who had both directly supervised the person, across 22 samples, reached a similar figure, and found agreement lower when the ratings were made for real in organisations than when they were made for research (Speer, Delacruz, Wegmeyer & Perrotta, 2024).
That correction matters, and it does not rescue the comfortable assumption. A reanalysis of supervisor ratings in two samples, 392 and 342 people, still found the rater-related part of the ratings larger than the part shared across raters, at more than 49 per cent against more than 27 per cent (Jackson et al., 2022). Earlier work using a different model had argued that the vantage point of the rater, boss or peer or member of staff, explains much more of a rating than the Scullen figures implied, and the person’s general performance much less (Hoffman, Lance, Bynum & Gentry, 2010).
How solid is this: contested as to the exact share, and the contest is live in the specialist literature as this is written. Every estimate located for this briefing leaves the rater’s contribution at least as large as the rated person’s.
2.3 Different Is Not the Same as Wrong
The dispute changes one thing a reader should think, which is why it gets more than a sentence here. Part of what looks like a rater’s peculiarity may be real: a manager who sees somebody handle the board and a peer who sees them handle a crisis are watching different performances, and each may be right about what they saw. Economists studying a Scandinavian service firm found that supervisors varied widely in how they rated people of similar quality, and that part of that variation reflected real differences in how good the supervisors were at managing rather than only how lenient they were (Frederiksen, Kahn & Lange, 2020).
So the rater’s share is not all error. Some of it is a different view, some of it is a manager who genuinely brings out better or worse work in people, and some of it is the psychology the rest of this briefing describes. What it never is, on any estimate, is a neutral reading of the person. The distinction for the analyst in Chapter 1 is between a new manager who sees something the old one missed and a new manager who sees something the old one did not bring, and the evidence says both are common.
Stated plainly: two honest raters can disagree because they saw different things, because they manage differently, or because of what each brings to the judgement, and a single rating cannot tell you which.
The rest of this briefing is about the third of those. Part II describes what raters bring to a rating, one tendency at a time, and each chapter opens on what happens to the person rated before it turns to the evidence, so that the psychology is met where it does its work rather than as a list of errors to be memorised.
PART II. WHAT THE RATER BRINGS
Chapter 3: The Rater’s Temperament
3.1 The Insight
Some managers rate generously and some rate hard, and the difference follows them from team to team. The same piece of work lands differently depending on whose desk it lands on, and part of the reason is the person's personality.
3.2 What the Evidence Says
Psychologists call the tendency to rate everybody high leniency, and the question is whether it belongs to the situation or to the rater, and personality research usually describes people on five broad traits, known as the Big Five: openness to experience, conscientiousness, extraversion, agreeableness, meaning warmth and a wish to keep relationships smooth, and emotional stability. Among 111 students rating their peers, the more agreeable gave higher ratings and the more conscientious gave lower ones, and raters who were both agreeable and low in conscientiousness were the most generous of all (Bernardin, Cooke & Villanova, 2000).
The pattern holds across studies. A meta-analysis of 28 studies found that raters’ agreeableness, extraversion and emotional stability were related to the ratings they gave, and that the Big Five together accounted for between 6 and 22 per cent of the differences in ratings, depending on the setting (Harari, Rudolph & Laginess, 2015). The relationships changed with the context, including the purpose of the rating and whether raters had to account for their ratings to somebody.
How solid is this: evidenced by a meta-analysis of 28 studies and a smaller study of peer ratings. Some studies were laboratory or classroom work, and the 6 to 22 per cent range reflects how much the setting matters.
3.3 What It Means for You
The common belief is that a generous rater is a soft judge and a tough rater an accurate one. The evidence supports only half of that. Agreeable raters do rate higher, and that is a distortion, since it says more about their wish to keep relationships comfortable than about the work. But the harsh rater may be no more accurate, because low emotional stability also links to the ratings a person gives, and anxiety about one’s own position isn't calibration.
The useful habit is to know the rater before reading the rating. A meets expectations from someone who gives few higher marks is not the same verdict as a meets expectations from someone who gives most of their team higher marks. Organisations hold that information, and few of the people being rated ever see it, which is part of why a single number is read as a verdict.
3.4 What It Looks Like in Practice
In my experience with people like him, the analyst's first manager was a careful, conscientious rater who rarely gave top marks and gave them to her twice. Her second was warm and expressive with people he knew well, which is what makes his lower rating harder to read as a matter of temperament alone: his generosity was going somewhere else. Judgment from practice.
Chapter 4: What the Rater Is Trying to Achieve
4.1 The Insight
A rating is often written to do something, and what it is written to do shapes what it says. Raters are rarely only describing. They also reward, motivate, protect a team, avoid a difficult conversation, or build a case, and the number carries all of it. The form records only the number.
4.2 What the Evidence Says
Raters pursue different goals and their goals show up in their ratings. Students who were asked at the start of a semester what they wanted to achieve with their ratings of a teacher, among 19 possible goals, gave end-of-term ratings that differed with those goals, over and above differences between the teachers, and goals measured after they had watched much of the teaching added further (Murphy, Cleveland, Skattebo & Kinney, 2004). The setting was a classroom, and the point it makes is about raters in general.
In interviews with 60 executives from seven large organisations, the rating was routinely described as a tool rather than a record: political considerations were said to shape it nearly always, some executives inflated the ratings of problem employees to move them out of the department, and others deflated ratings to push somebody to leave or to build a case for dismissal (Longenecker, Sims & Gioia, 1987). These are executives’ own accounts, and they cannot say how often it happens, but they describe the goals frankly. The largest test concerns stakes. A meta-analysis of 22 studies covering 57,775 people found that ratings used for administrative decisions such as pay and promotion were more lenient than ratings given for research or development, and that the gap was larger in real organisations, among practising managers and for real staff than in the laboratory (Jawahar & Williams, 1997).
How solid is this: evidenced as to the effect of purpose, from a meta-analysis; evidenced in a classroom as to raters’ goals; and drawn from executives’ own accounts as to the political uses of ratings.
4.3 What It Means for You
Most people assume that ratings with real consequences are made more carefully, and therefore more accurately, than ratings that do not matter. The evidence suggests the reverse. When a rating decides somebody’s pay, raters have more reasons to soften it, and a system that ties pay to ratings asks raters for candour at exactly the moment candour costs them most.
The same logic runs in the other direction. A rating that falls sharply in a year when a rater needs a case, or needs a budget for somebody else, may be doing a job other than describing. In both cases the number on the form looks exactly like one written only to describe.
None of this requires cynicism. Most of the executives in the interview study described their adjustments as fair to the person, the team or the organisation, and a manager who nudges a struggling but loyal colleague up a grade usually experiences it as kindness. The cost falls on whoever is rated against that colleague, and on the organisation that later reads the record as if it were a measurement.
4.4 What It Looks Like in Practice
In my experience, the clearest sign of a rating doing a job is the comment that does not match the number: strong work, no concerns raised, and a mark below what the comments describe. The analyst’s review had exactly that shape: praise for the substance and a mark explained by something never measured. Judgement from practice.
Chapter 5: The Person the Rater Chose
5.1 The Insight
Managers tend to rate the people they chose more kindly than the people they inherited. A rating of somebody you hired is partly a rating of your own decision, and people protect their decisions. Few of them notice doing it.
5.2 What the Evidence Says
Psychologists call the tendency to keep backing a choice because one made it escalation of commitment. In a classroom experiment with 298 business students, participants who had earlier decided to promote a candidate later rated that person’s poor performance more favourably, and gave larger rewards, than participants who had not made the decision (Bazerman, Beekun & Schoorman, 1982); and in a large public-sector organisation, supervisors rating their real employees rated more favourably those they had helped to hire, and, more weakly, rated less favourably those they had argued against hiring who were appointed anyway (Schoorman, 1988). Later experiments refined it. Among 210 participants who first chose one of two sales candidates and later rated a salesperson’s performance, ratings were inflated for the person they had chosen, while the candidate they had rejected was not marked down (Slaughter & Greguras, 2008), and a laboratory experiment by economists found the same favouring of hires and penalising of explicit rejections, and found that co-workers tended to follow the manager’s lead (Angelovski, Brandts & Solà, 2016).
How solid is this: evidenced, from a field study and several experiments. Whether rejected candidates are marked down is less consistent than whether chosen ones are marked up.
5.3 What It Means for You
A rater’s history with you is part of your rating. If your manager fought to hire you, part of what they are rating is their own judgement, and they will tend to see your work in its best light. If they wanted somebody else, the evidence on whether you will be marked down is mixed, but the protective lift you might have had from a manager who chose you is missing.
For the rater, the uncomfortable question is whether you would score this work the same if somebody else had made the hire. Most people believe they would, and the experiments suggest otherwise.
5.4 What It Looks Like in Practice
The head of underwriting had argued for a different person in the analyst’s role, which in Schoorman’s terms makes her the employee he did not choose. That does not show that he marked her down, and the evidence on downward bias is the weaker half. What it does show is that the rating he gave was the first one in which the rater had something of his own invested in the person not being the right choice. Judgement from practice, consistent with the escalation findings.
Chapter 6: The First Impression That Will Not Move
6.1 The Insight
Once a rater has decided what kind of person somebody is, later evidence tends to be read through that decision. A strong first impression lifts every rating that follows it, and a weak one weighs on every rating, sometimes for years, because the rater reads each new piece of evidence as a further instance of what they already know.
6.2 What the Evidence Says
A century ago, Edward Thorndike noticed that when army superiors rated officers and cadets on separate qualities such as intelligence, physique, leadership and character, the ratings were far too closely related to each other to be describing separate qualities, and he named the pattern the halo effect, a general impression of the person colouring judgements of their specific qualities (Thorndike, 1920); and in a much later experiment, students who watched a lecturer behave warmly rated his appearance, mannerisms and accent as more appealing than students who watched the same man behave coldly, and they denied that their liking had influenced those judgements (Nisbett & Wilson, 1977). Whether halo is always an error is disputed. A critical review argued that much of what is counted as halo may be real, since good performers often are good at several things, and that true and illusory halo are hard to separate (Murphy, Jako & Anhalt, 1993).
What fixes a first impression in place is partly the rater’s beliefs about people. Implicit person theory is the name for a person’s working belief about whether ability and personality are fixed or can change. Across four studies, managers who believed people can change were more likely to notice when an employee’s performance improved, and also when it declined, and their judgements were less anchored by their first impression; managers who believed people are fixed shifted towards the other view after a short self-persuasion exercise, kept the shift six weeks later, and became more willing to recognise improvement (Heslin, Latham & VandeWalle, 2005).
How solid is this: evidenced as to halo existing and as to the effect of managers’ beliefs, from experiments and field studies; contested as to how much halo is error.
6.3 What It Means for You
Most people believe that a year of good work will be seen as a year of good work. The evidence says that depends on what the rater decided about you before the year began, and on whether the rater believes people change at all. A manager who thinks ability is fixed may simply not register that you have improved, and will not experience that as bias.
For anybody inheriting a team, the first impression is made quickly and revised slowly, so the first three months of a new reporting line weigh more than they should. A manager who formed a poor view of somebody in their first weeks will often still be reading that person through it at the end of the year.
The effect compounds. A rater who expects little looks for less, gives less demanding work and reads an ambiguous result as confirmation, and the person rated, sensing the expectation, often plays safer in return. Each of those steps is small.
Over a year they can produce exactly the performance the first impression predicted, which then appears to vindicate it. That sequence is an inference from the anchoring evidence and from practice, not a tested finding about ratings.
6.4 What It Looks Like in Practice
The analyst met her new manager during a merger, and in his eyes she was the person he hadn't wanted in the role; he formed his view of her in meetings where she said little. Strategic presence is a judgement about the whole person, which is where halo works best. Judgement from practice.
Chapter 7: The Person Who Seems Like You
7.1 The Insight
Raters favour the people who seem to think the way they do, and the seeming matters more than the fact. The sort of similarity most people worry about, sharing a manager’s age, gender or background, turns out to do little to a rating on its own. What moves the rating is the rater’s sense that this person shares their values, their standards and their way of working. That sense is formed quickly and is often wrong.
7.2 What the Evidence Says
People are drawn to others like themselves is one of the oldest findings in social psychology, and the largest synthesis adds a twist. Across 313 studies and 460 results, both actual and perceived similarity predicted liking, but actual similarity mattered mainly between strangers; in established relationships, it stopped predicting attraction, while the sense of being alike continued to predict it (Montoya, Horton & Kirchner, 2008). Once people know each other, what counts is what they believe they share.
The same split appears in ratings at work. Among salespeople in a national appliance company, whose personalities were measured and whose supervisors and peers rated them, the measured similarity between a salesperson’s personality and a rater’s was largely unrelated to the rating given, while the rater’s perception of similarity in extraversion, conscientiousness and emotional stability was strongly related to it, for supervisors and peers alike, and liking carried part of the effect (Strauss, Barrick & Connerley, 2001). The same raters reported both the similarity and the rating, which may have inflated the link.
A 2026 meta-analysis of 82 samples of managers and their staff tested the pattern across the field. Similarity in age, gender, race, and education showed no relationship with performance, while similarity in values, personality, and attitudes showed a moderate one; within that, the similarity people perceived went with performance nearly twice as strongly as the similarity that was measured (Condrea & Iliescu, 2026). The link was stronger in more collectivist cultures. Most studies measured everything at one point in time, used self-reports rather than records, and asked whether a good working relationship produces similarity rather than the other way round.
Interviews with 120 evaluators at elite banks, law firms and consultancies show what the judgement looks like from inside. Similarity was the most common basis for assessing candidates: shared pastimes, experiences, and ways of presenting oneself. More than half ranked fit as the single most important criterion at interview, and evaluators used their own biographies as the template of merit, so that former athletes read athletic candidates as driven (Rivera, 2012). That study concerns hiring rather than appraisal, but the evaluators are the same kind of people who later write the reviews.
Liking is the route by which much of this reaches the form. A review of the research found that supervisors’ liking for a subordinate is regularly linked to higher ratings, more halo, less accurate ratings and a reluctance to penalise poor performance, while cautioning that the direction of cause is unclear, since good performers may simply be easier to like (Lefkowitz, 2000).
How solid is this? Evidence includes two meta-analyses, a field study, and an interview study. The direction of cause is unresolved, and the meta-analysis does not separate supervisors’ ratings from other reports of performance.
7.3 What It Means for You
The common belief is that favouritism in ratings follows visible lines, such as who looks like the manager or shares their background. On the evidence here, those lines on their own do almost nothing to the rating. The pull comes from something less visible and harder to argue with: whether your manager has decided that you see the world as they do. You cannot inspect that decision, and your manager may not know they have made it.
Two things follow. First, perceived similarity is built from contact and from what people show of themselves, so a quiet colleague whose standards match the manager’s exactly may never be credited with them, while a colleague who mirrors the manager’s style may be credited with standards they do not hold. Second, some of the effect may be real. A manager and a staff member who genuinely share standards may simply work better together, and the evidence cannot yet separate good fit from a flattering impression.
For the rater, the test is plain. Before you rate somebody, ask whether you would rate the same work the same way if the person did not remind you of yourself.
7.4 What It Looks Like in Practice
The analyst and her first manager were, in my reading of the composite, alike in the way that matters here: both quiet, both more at home with the model than the meeting, both trusting numbers over presentation. Her second manager worked through talk and presence. Strategic presence describes how he does the job, and in my experience, a manager’s development advice often reflects who they are. Judgement from practice, consistent with the perceived-similarity findings.
Chapter 8: The Rater’s Own Security
8.1 The Insight
Raters tend to rate down the people who make them feel less capable. It is not their intention, and it rarely looks like that. It looks like a judgement about style.
8.2 What the Evidence Says
When managers’ confidence in their own ability was low, whether measured among 41 managers and 148 of their staff in an energy company or lowered deliberately in an experiment with 131 adults, they sought less input, rated the people who spoke up less favourably and acted on fewer of their ideas, and the effect ran through an urge to protect their view of themselves (Fast, Burris & Bartel, 2014). That study concerned employees who offered ideas and challenges, not ratings in general, and the step to every appraisal is an extension.
How solid is this: evidence that insecure managers rate challengers down, from a field study and an experiment.
8.3 What It Means for You
The awkward implication is that the strongest member of a team can be rated down by a manager who finds them threatening, and the rating will be written as a concern about their style. If the comments on your review describe how you make people feel rather than what you delivered, the rater’s security is worth considering alongside your own conduct. For the rater, the test is whether the person you are about to mark down is the one who most often tells you something you did not want to hear. If it is, look again.
8.4 What It Looks Like in Practice
In my experience, a new head of a merged function is least secure in the first year, inheriting specialists whose work they cannot easily check. A quietly expert analyst who brings precise numbers to meetings can feel, to that manager, like a standing reminder of what he does not know. Judgement from practice, consistent with the Fast finding.
Chapter 9: When Merit Is the Story
9.1 The Insight
An organisation that tells its managers it rewards merit can make their decisions less fair, not more. The belief that the system is objective gives raters permission to stop checking themselves. That is the paradox.
9.2 What the Evidence Says
The first evidence came from one large American service employer, where personnel records for 8,898 support staff over eight years showed that women, ethnic minority and non-American-born employees received smaller pay rises than white men with the same performance rating, even within the same job, unit and supervisor (Castilla, 2008). The author called this performance-reward bias the gap between a rating and what it buys, and argued that systems with little transparency and accountability allow it. The annual gaps were small, at around half of one per cent a year, but they were paid every year and they compound, so that two people rated identically for a decade end it on noticeably different salaries.
Experiments then tested the organisational story directly. Across three studies with 445 participants, most of whom had managed others, those told their company’s values emphasised merit gave an equally rated man a larger bonus than a woman in each study (Castilla & Benard, 2010). In the third study, the gap in the meritocratic condition was about $46, and in the comparison condition women received slightly more than men, a difference that could have been chance.
How solid is this: evidence from a large field study in one organisation and three experiments with business school participants, most of whom had management experience. The experiments concern bonus decisions made from written profiles rather than real staff.
9.3 What It Means for You
The common belief is that the more an organisation emphasises merit, the fairer its ratings and rewards become. The evidence suggests that emphasis on its own does the reverse, because it reassures raters that their judgement is objective. What the field study points to as the protection is accountability and transparency: somebody checking how ratings turn into rewards, and the results being visible.
This is a finding about raters, not about any group’s performance. It says that two people with the same rating were treated differently by the people deciding what the rating was worth. Neither study suggests that those given lower ratings performed worse, since the rating said they did not.
9.4 What It Looks Like in Practice
A combined team with a higher bar is, in my experience, an organisation telling itself a merit story at the moment its raters are least calibrated against each other. Nobody in the analyst’s case set out to treat her differently, and the evidence says that is precisely the condition in which it happens. Judgement from practice.
PART III. PERSONALITY ON BOTH SIDES OF THE FORM
Chapter 10: The Person Being Rated
10.1 The Insight
Almost everybody rates their own year more generously than their manager does, and the people highest in narcissism do it most on the qualities to do with getting ahead. The gap between your view of your work and your rater’s is normal, and part of it is yours.
10.2 What the Evidence Says
A meta-analysis of 128 samples found that self-ratings and supervisor ratings of the same performance agree only modestly, and that in Western samples covering 35,417 people self-ratings ran clearly more generous than supervisors’ ratings, with the gap depending on the context and the rating form (Heidemeier & Moser, 2009). How far the two agreed depended most on the kind of job and on whether the ratings used concrete, non-judgemental indicators of performance. Across all samples, the comparison covered 37,752 people.
Personality shapes the self-rating. Self-enhancement, rating oneself more favourably than observers or objective measures do, rises with narcissism: a meta-analysis of 36 studies and 6,423 people found the link clear for qualities to do with getting ahead and close to nothing for qualities to do with getting along, across self-ratings that included leadership, intelligence and task performance (Grijalva & Zhang, 2016). Others’ ratings can be swayed in the same direction. In an experiment in which each group member held part of the information needed to reach the right answer, leaders higher in narcissism were rated as more effective, through their displays of authority, while their narcissism held back the sharing of information and the group did worse (Nevicka, Ten Velden, De Hoogh & Van Vianen, 2011); and a large synthesis found that narcissism predicts who is picked as a leader more than who turns out to be good at leading, and that the link to being picked halved once raters had known the person for longer (Grijalva, Harms, Newman, Gaddis & Fraley, 2015).
How solid is this? Three meta-analyses and an experiment. None of this describes any particular person, and narcissism here means a trait measured on a scale in the general population, not a diagnosis.
10.3 What It Means for You
If your self-rating and your manager’s rating disagree, the evidence says that the default explanation is ordinary self-enhancement on your side, and you should start there before concluding that the rater is wrong. That is uncomfortable, and it is useful, because it gives you something to check: the concrete indicators, where your view and your manager’s should meet. Ask which of the specific things on the form you and your manager would both count in the same way.
The second finding cuts the other way. Raters can be impressed by confidence and authority in the short run, so the colleague who rates themselves highly and says so with assurance may also be rated highly by others, at least until those others have watched them long enough. That longer-run correction is the one the synthesis found, since people who had known somebody for longer were less swayed.
10.4 What It Looks Like in Practice
The analyst’s self-assessment in her new team was, by her own later account in this composite, modest and detailed. In my experience, people with her profile underclaim, the opposite of the average, and a rater who reads self-assessments as a starting point will take an underclaimer at their word. Judgment from practice.
Chapter 11: Who Gets Seen
11.1 The Insight
A rater can only rate what reaches them, and what reaches them depends on networks, personality and proximity as well as on the work. Social network research treats an organisation as a web of who goes to whom for advice and company. Where you sit in that web shapes what your rater sees of you and how alike you seem. It also shapes how much benefit of the doubt you get.
11.2 What the Evidence Says
Ties at work form disproportionately between similar people, a pattern known as homophily, one route by which similarity in Chapter 7 is built and sustained. Personality decides part of who ends up at the centre of the web. People high in self-monitoring, meaning the habit of reading a situation and adjusting how one presents oneself to fit it, were more likely than others to hold central positions in the networks of a high-technology firm of 116 people, and self-monitoring and network position each predicted workplace performance (Mehra, Kilduff & Brass, 2001).
A meta-analysis of 138 samples confirmed the pattern and set its limits. Self-monitoring predicted centrality, being sought out by many colleagues, in both friendship and advice networks, but personality as a whole explained only between 3 and 5 per cent of network position; being sought out predicted job performance more strongly than bridging separate groups did, and personality predicted performance beyond what network position explained (Fang et al., 2015). Part of that link will be real, since people who are good at their jobs get asked for help.
What raters see is partly the network they imagine. In a study of an organisation and of MBA teams, colleagues credited men more often than women with bridging separate groups even when their actual positions were the same, and a woman wrongly seen as a bridge was rated competent but not warm (Brands & Kilduff, 2014). The judgement was made about a network that existed mainly in the perceiver’s head.
Contact is where similarity turns into advancement. In a commercial bank with 14,638 staff, managers rotated in a way the employees did not choose, and men moved from a female to a male manager rose about 0.6 pay grades further over the following ten quarters, roughly 15 per cent of salary, while women moved to a male manager did not; smokers moved to a smoking manager gained a similar amount (Cullen & Perez-Truglia, 2023). Neither group worked longer or sold more. What changed was how often they took breaks with the manager, and the effect was concentrated where staff sat near their manager.
The outcome was promotion rather than a rating, and the authors estimate that the mechanism could explain about a third of that firm’s gender gap in promotions. A 2025 economic model adds that this compounds: when ties form between similar people and one group is a minority, members are referred less often and thin out toward the top (Neugart & Zaharieva, 2025).
The counterweight comes from a randomised trial. At a technology firm, 1,612 graduate employees were assigned by birthday either to work from home two days a week or to come in every day, and over the next two years of reviews the hybrid group’s performance grades and promotions were no different, while their quit rate fell by a third (Bloom, Han & Liang, 2024). The firm’s 395 managers had expected hybrid working to cost productivity before the trial and expected a small gain after it. Reviews at that firm drew on co-workers, direct reports and customers as well as the manager.
How solid is this: evidence from a meta-analysis, field studies, a quasi-natural experiment in one bank, and a randomised trial in one firm. The bank study concerns promotion rather than ratings, and network position is partly a result of good performance rather than only a cause of good ratings.
11.3 What It Means for You
The common belief is that being out of sight costs you your rating. The trial suggests that is not automatic, at least where several people feed into the review. What the bank study shows is narrower and more uncomfortable: equal work, unequal advancement, separated by who took breaks with the boss.
The personality point matters for anybody quiet. People who adjust easily to their audience build wider networks and are seen more, which is partly real contribution and partly visibility. People who keep their heads down are rated on less information, and that information usually comes from fewer people. That’s why you let more than one person see your work, not to imitate a style that isn't yours.
11.4 What It Looks Like in Practice
The analyst sat outside her new manager’s circle in every sense: different team history, different floor, little informal contact. Her first manager knew her through her models. Her second knew her through meetings, which is to say through the one channel where she was least visible, and in my experience visibility in a review is often a description of the reviewer’s network rather than of the work. Judgement from practice, consistent with the network and contact findings.
Chapter 12: The Rater’s Darker Traits
12.1 The Insight
It would be convenient if unfair ratings came from unpleasant managers. The evidence that the darker personality traits shape the ratings a manager gives is thin, and most of the unfairness described in this briefing needs no dark trait at all, which is a less satisfying conclusion and a more useful one, because it means the remedies in Part V apply to every rater rather than to a few.
12.2 What the Evidence Says
The darker traits are usually studied as the Dark Triad, three overlapping tendencies: narcissism, a sense of grandiosity and entitlement; Machiavellianism, a cold, strategic willingness to manipulate; and psychopathy, low empathy combined with impulsiveness. One study of 87 supervisors and their teams, measured at one point in time, found that supervisors higher on a combined score of the three had poorer working relationships with their teams, and that this went with how team members rated the team’s performance and innovation, while the supervisors’ own ratings of their teams went instead with an inflated sense of the team’s specialness (Fodor, Curşeu & Meslec, 2021). It is a single study using supervisors’ self-descriptions, and it did not compare anyone’s ratings with an objective measure of performance.
No study located for this briefing has tested whether managers higher in the darker traits rate their staff less accurately, more harshly or more strategically than other managers. The closest tested evidence is indirect: in a study of 60 teams, leaders higher in narcissism asked for input less (Zhou, Yin & Sun, 2021), and managers who feel threatened rate challengers down (Fast, Burris & Bartel, 2014), which describes a reaction to threat that anybody can have.
How solid is this: one study, measured at one point in time, on the darker traits and ratings. Everything below is commentary on the general pattern, labelled as inference.
12.3 What It Means for You
What follows is inference from what the darker traits are known to do in general, not a finding about ratings. A rater high in Machiavellianism might be expected to treat a rating as a tool more readily than most, which is the political use described in Chapter 4. A rater high in narcissism might be expected to react most strongly to a subordinate who outshines them, which is the threat response in Chapter 8 at its most intense. Both are plausible and neither has been tested.
The more important point runs the other way. The patterns in Part II appear across ordinary raters, and a reader who goes looking for a dark personality behind an unfair rating will usually be looking in the wrong place. Nothing here can tell you what any manager is like, and a rating is not evidence of a trait.
12.4 What It Looks Like in Practice
Nothing in the analyst’s case requires a dark trait, and in my experience, most cases like hers don't involve one. A fluent manager who worked nothing like her, an investment in a different candidate, a first impression formed in a merger and a merit story were enough. Judgment from practice.
PART IV. BEFORE YOU ACT
Chapter 13: The People You Will Mislabel, Including Yourself
Read this before acting on anything above. Every pattern in this briefing describes what tends to happen across many raters, and each of the people below can be misread by somebody who has just learned it.
13.1 The Rater Who Marked You Down and Was Right
A lower rating from a new manager can mean that the new manager saw something real. Part of the rater’s share in Chapter 2 is a different view rather than an error, and part is a better manager, so the first question about a disappointing rating is what the rater saw, not what the rater is. Asking for the specific behaviour behind the rating is how to find out.
13.2 The Generous Rater
A manager whose team all rate highly may be lenient, which is a distortion, or may run a genuinely strong team. Agreeable raters do rate higher, but a high average is not proof of leniency, and treating it as one punishes the people in that team. The way to tell is the evidence behind the ratings, which a reviewer can ask for.
13.3 The Manager Who Hired You
A rater who chose you may lift your ratings without meaning to. That is a reason for them to check their judgement, not a reason for anybody else to discount your record, which may be exactly as good as it looks. A strong record from a manager who chose you deserves a second reader, not a discount.
13.4 The Colleague Who Seems Like the Boss
A colleague who shares the manager’s outlook and is rated highly may be benefiting from perceived similarity, or may simply be good and well suited to working with that manager. Shared standards can make two people genuinely more effective together, and the evidence in Chapter 7 cannot yet separate good fit from a flattering impression. Resenting the rating is not the correction; asking whether a second rater who does not share that outlook would see the same work is.
13.5 The Halo That Is Real
Somebody rated well on everything may simply be good at most things. Halo can be real, and assuming that consistently high ratings must be inflated is a bias of its own. The test is whether the specific evidence varies, not whether the marks do.
13.6 The Confident Colleague
The person who rates themselves highly and is rated highly by others may be self-enhancing, or may be very good. The research describes average tendencies across groups and offers no way to tell which from the outside. Watching the work over time is the correction the evidence supports, and it is slow.
13.7 You
The evidence says your own view of your year is probably more generous than an outsider’s. The habit worth keeping from this briefing is to ask what each person, including yourself, brought to a judgement, and to ask it of your own ratings of others too, because if you rate people, everything in Part II applies to you. That includes anybody reading the analyst’s case and deciding they now know what her new manager is like.
PART V. WHAT CHANGES IT
Chapter 14: If You Are Being Rated
14.1 Read the Rating as a Sentence About Two People
Before deciding what a rating says about you, ask what you know about the rater: how they rate everybody else, what the rating is being used for, whether they chose you, and how long they have known your work. Judgement from practice, following from Chapters 3 to 7. It does not make a poor rating wrong. It tells you how much weight one person’s view can carry.
14.2 Look for the Second View
Because two raters agree only moderately, a single rating is a weak basis for any conclusion, and the view of a second person who has seen your work, a former manager, a client or a peer, is the cheapest correction available. It is also the correction a reasonable rater is least likely to resent, because it adds information rather than disputing theirs. Judgement from practice, following from the agreement evidence at 2.2. Organisations that collect ratings from several people do this formally, and Chapter 16 returns to it.
14.3 Keep Your Own Record, in Concrete Terms
How far self-ratings and supervisor ratings agree depends on whether they rest on concrete indicators of performance (Heidemeier & Moser, 2009), and Thorndike’s own recommendation in 1920 was to record the evidence for each quality rather than rate the person as a whole. Keep dated, specific notes on what you delivered, so that a conversation about presence can be brought back to work. Evidenced as to concrete indicators mattering to agreement; the record-keeping is judgement from practice.
14.4 Ask What the Rating Is For
A rating that decides pay is under different pressures from one written for development, and the evidence says it tends to be softer (Jawahar & Williams, 1997). Knowing which you are looking at tells you how to read it. A middling rating in a system where most people are rated generously is a stronger signal than the same rating in a system used only for development, and a sharp fall in a year of budget pressure may be the budget talking. Evidenced as to purpose; the reading is judgement from practice.
14.5 Treat a Change of Rater as an Event
A change of manager is when the rater’s share of your rating changes most, and when a first impression forms that will be slow to move. In the first months, make your work visible to the new rater in concrete terms, and ask early what they will be looking for. Judgement from practice, following from Chapters 5 and 6.
14.6 Let More Than One Person See the Work
A rating rests on what reached the rater, and quieter people are rated on less of it. Rather than imitating a style that is not yours, make sure your work is known to more than one person who could speak to it: a peer who uses your output, a client, a manager in a neighbouring team. In the one randomised trial on the question, working out of sight two days a week did not lower ratings where several people fed into the review (Bloom, Han & Liang, 2024). Evidence as to the trial; the advice is judgement from practice, following from Chapter 11.
Chapter 15: If You Do the Rating
15.1 Write Down Why, for Somebody Else to Read
Raters who had to justify their ratings in writing were more accurate in an experiment with 247 students rating simulated subordinates (Mero & Motowidlo, 1995), and accountability weakened the influence of raters’ personalities in the meta-analysis of 28 studies (Harari, Rudolph & Laginess, 2015). The same experiment found a warning: raters told that earlier ratings had been too low simply gave higher ones, so accountability works when it asks for reasons and fails when it signals the answer. This is supported by an experiment and a meta-analysis.
15.2 Train Raters on What Good Looks Like
Frame-of-reference training, which teaches raters a shared picture of what each level of performance looks like in practice and has them practise against it, improves the accuracy of ratings, and an updated meta-analysis found the largest gains in raters’ accuracy about the pattern of each person’s performance across its different parts (Roch, Woehr, Mishra & Kieszczynska, 2012). Evidenced as to accuracy in the studies reviewed. Whether it changes the political uses of ratings in Chapter 4 has not been shown.
15.3 Check Your Theory of People
Managers who believe people can change see improvements others miss, and a short exercise lasting at least six weeks can shift that belief in one study (Heslin, Latham & VandeWalle, 2005). Before rating somebody, ask when you formed your view of them and what would have to happen for you to change it. Evidence as to the belief and its training; the question is judgement from practice.
15.4 Ask the Four Uncomfortable Questions
Before a rating goes in, ask yourself four things: whether you would rate the work the same if you had chosen this person, or had not; whether you would rate it the same if the person did not remind you of yourself; whether the person you are about to mark down is the one who most often challenges you; and whether you would give the same rating to a stranger who had done the same work. Judgment from practice, following from Chapters 5, 7 and 8. None of the four has been tested as an intervention.
15.5 Keep the Hire Out of the Rating
If you chose the person, or argued against them, say so to whoever reviews your ratings, and ask them to look at that rating with particular care. The escalation evidence shows the lift for people you chose more reliably than the penalty for people you did not, but both are invisible to the rater, which is why a second reader helps. Judgement from practice, following from the escalation evidence at 5.2.
15.6 Remember That Feedback Can Make Things Worse
Across 607 results covering 23,663 observations, feedback improved performance on average, but more than a third of the feedback interventions made performance worse (Kluger & DeNisi, 1996). A rating delivered as a verdict on the person rather than on the work is the kind most likely to fall into that third, on the review’s own account of why feedback fails. Evidenced as to the size of the problem; the account of why feedback fails is the authors’ theory.
Chapter 16: If You Run the Organisation
16.1 The Insight
An organisation cannot train the psychology out of its raters, but it can stop any single rater’s psychology from deciding somebody’s career. The levers are structural: more than one view, reasons that somebody reads, and a check on what ratings turn into.
16.2 What the Evidence Supports
Four findings bear on design. Because the rater’s share of a rating is large, averaging several raters dilutes any one rater’s tendencies, which is the statistical logic behind collecting more than one view (Scullen, Mount & Goff, 2000). Because pay-deciding ratings are more lenient (Jawahar & Williams, 1997), an organisation that uses the same rating for development and reward asks it to do two jobs that pull in different directions. And because a merit story without accountability went with more bias (Castilla, 2008; Castilla & Benard, 2010), the protection lies in somebody checking how ratings translate into pay, not in the language of the values statement.
And because the similarity that goes with higher ratings is the perception of a shared outlook (Strauss, Barrick & Connerley, 2001; Condrea & Iliescu, 2026), a criterion such as fit or presence invites exactly that perception onto the form, which the authors of the 2026 meta-analysis themselves give as a reason to use several raters.
How solid is this: each finding is evidenced. Their combination into a design is judgement from practice, and no trial located for this briefing has tested a whole appraisal design against these outcomes.
16.3 What Follows, in Practice
Five design choices follow, and each is judgement from practice rather than a tested intervention. Collect ratings from more than one person who has seen the work, including at least one from outside the person’s immediate circle, and weigh a single manager’s rating accordingly. Define what good performance looks like in terms of the work, so that a judgement of fit cannot stand in for it.
Separate the development conversation from the pay decision where the organisation can, so neither has to carry the other’s pressure. Have a second person review ratings against the reasons given, which is what accountability means in practice, and review what ratings turn into by group, which is the check the field evidence points to. And watch the moments when the rater’s share is largest: a new manager, a manager rating somebody they argued against hiring, and a merger.
The merit evidence adds one caution about language. In the experiments, describing the company as meritocratic went with the gap, and in the third study, a comparison company described without that emphasis, and with less stress on managers’ own discretion, produced none (Castilla & Benard, 2010). The words an organisation uses about its own fairness are not neutral. They can tell raters that the checking has already been done.
Calibration meetings, in which managers discuss and adjust ratings together, are widely used for exactly this purpose. Whether they reduce the rater’s share of a rating has not been established in the evidence located for this briefing. No evidence base located.
Chapter 17: What This Changes
17.1 The Short Version
If you read nothing else, read this. A rating carries more of the rater than of the rated: in two samples of more than 2,000 managers, raters’ own tendencies accounted for 62 and 53 per cent of the differences in ratings, against 21 and 25 per cent for performance. Ratings that decide pay are softer, not harder: across 22 studies and 57,775 people. The rater’s personality is in the rating: agreeable raters rate higher, across 28 studies.
The similarity that counts is the one people perceive: across 82 samples, similarity in age, gender, race and education showed no link with performance, while perceived similarity in values and personality did. Who takes breaks with the boss can move a career: in a bank of 14,638 staff, smokers moved to a smoking manager advanced further without working harder.
Calling a system meritocratic can make it less fair: 445 participants with management experience gave equally rated men larger bonuses when told the company was meritocratic. Your own rating of yourself is probably the generous one: across 37,752 people. And a rating used as feedback can backfire: more than a third of feedback interventions made performance worse.
17.2 What I Still Do Not Know
Five things remain open. The exact share of a rating that belongs to the rater is in active dispute, and the 2024 work puts agreement between supervisors higher than the figures most often quoted; how much of the rater’s share is a valid different view rather than bias is not settled either.
Nor is the direction of the similarity finding: a good working relationship may create the sense of being alike as much as the sense creates the relationship. No study I could locate has tested whether managers higher in the darker traits rate differently, so Chapter 12 is commentary. And no trial has tested a whole appraisal design, or calibration meetings, against the outcomes this briefing describes.
The temptation is to tell this as a story about biased managers, but the evidence supports a less dramatic and more useful account. Ordinary raters bring their temperament, purpose, investment, sense of who is like them, and own security to a judgement they experience as objective, producing ratings that say a great deal about themselves, and the organisation decides whether one person’s view is allowed to stand for the truth.
What I would most like to make harder is reading a single rating as fact about a person. What I'd like to make easier is asking what the rater brought to it.
