The Invisible Struggle: Self-Assessment Barely Matches Peer Feedback

An annual check-in question asks employees a deceptively simple thing: "How would you describe your feelings towards your job and your ability to perform it well?" The five answers cross two dimensions: how you feel about the work (love, neutral, dislike), and how you are coping with it (performing well, struggling).

Tracking the 885 people who answered in more than one annual wave showed a distinctive pattern. Attachment to the job is durable: 88% of people who love their job still love it a year later, and direct jumps from love to dislike almost never happen (4 in 948 transitions). What moves is the other half of the answer. Each year, roughly a quarter of the people who say "Love & Perform Well" slip to "Love & Struggle," and a slightly larger share of strugglers recover. The feeling holds; the sense of being able to perform flips.

That raises the question this study answers: when a person's self-assessment flips, did their performance actually change? These workplaces offer an unusually direct way to check. Colleagues send each other structured feedback carrying a four-level assessment, from "Needs Improvement" to "Truly Outstanding," and managers write goal reviews on a four-point scale. We linked every self-report to every assessment the person received within 180 days on either side, 16,379 colleague feedback items and 1,553 manager reviews in all, and measured the match.

There barely is one. The half of the answer that moves from year to year is precisely the half no one else can see.

The annual dance: confidence moves, attachment holds

First, how people answer at all. Nearly half of the 4,834 answers are "Love & Perform Well." Almost one in three is "Love & Struggle": attached to the work, doubting their ability to do it well. Outright dislike is rare, 2% of answers in total, which is why this is mostly a story about the two love states.

Year over year, 60% of consecutive answers repeat exactly, and stability rises with the state: "Love & Perform Well" holds at 67%, "Love & Struggle" at 54%, Neutral at 51%. No one who answered again stayed in a dislike state two waves running, though dislikers also drop out of the panel at higher rates (more on that under Limitations).

When an answer does change, the change has a shape. Half of all changes are the performance half flipping while love holds. Direct flips of the feeling half are nearly nonexistent.

The traffic between the two love states is also asymmetric: 144 slips into struggle against 90 recoveries, because the perform-well pool is twice as large. The rates themselves replicate almost exactly across both wave pairs (a 23% slip rate and a 29% recovery rate in each), so this is a stable annual rhythm, not a one-off shock. The question is whether the rhythm reflects anything others can see.

Colleagues rate strugglers and performers alike

Each colleague feedback item carries the giver's assessment of the receiver: Needs Improvement, Meets Expectations, Exceeding Expectations, or Truly Outstanding. If self-reported struggle tracked observed performance, people answering "Love & Struggle" should receive weaker assessments than people answering "Love & Perform Well." They barely do.

The gap between self-assured performers and self-doubting strugglers is 0.045 points on a four-point scale, with a confidence interval that lets us rule out any true gap larger than about 0.1 points. Colleagues flag "Needs Improvement" at the same 1.6% rate regardless of what the person says about their own coping. Z-scoring ratings within each company removes any nice-rating-culture confound; the gap that remains is 0.06 standard deviations, borderline at conventional significance and negligible in size.

The little gradient that does exist runs along the other axis. Ratings soften slightly from love (36.0%, 33.8%) to neutral (30.2%) to dislike (24.8% for "Dislike & Perform Well," though only 41 answers). Colleagues seem to pick up traces of the feeling half of the answer, the half that shows on your face, and none of the coping half.

Agreement is no better than chance

Collapsing both instruments to a binary puts a number on the disconnect. Self-report: perform well vs struggle. External: mean received rating at or above "Exceeding Expectations" vs below it. Among 1,855 answers with at least two feedback items in the window, the two instruments agree 54.9% of the time, against 53.9% expected if they were statistically independent. Cohen's kappa: 0.02.

Read the off-diagonal cells. 64% of self-reported strugglers (436 of 686) carry an average colleague assessment of Exceeding Expectations or better; their struggle is invisible. And 34% of self-reported performers (400 of 1,169) sit below their feedback culture's norm; their confidence is not corroborated either. This is consistent with five decades of self-other rating research: meta-analyses put the self-peer correlation for job performance around r = 0.36 (Harris & Schaubroeck, 1988), and self-evaluations of ability generally around r = 0.29 (Zell & Krizan, 2014). Our kappa is lower still, likely because this self-report captures felt coping rather than a considered performance estimate.

The year the answer slips, ratings stand still

The strongest test uses each person as their own control. Take everyone who answered in consecutive waves, and compare the colleague ratings they received around each answer. If slipping from "perform well" to "struggle" reflected a real performance dip, the received ratings should fall with it.

Two details deserve attention. The people who slipped into struggle were the highest-rated group before the slip (3.18), and their ratings did not move afterward. And the people who stayed in struggle two years running saw their ratings drift upward while they continued to report struggling. Whatever changed for these people between waves, their colleagues' assessments did not register it.

The same holds in aggregate. Across the companies that ran all three waves, the share reporting struggle rose 7.2 points between 2024 and 2025, from 26.4% to 33.6%. The mean colleague rating received by the people answering did not decline over that period: it went from 3.07 to 3.12, a difference within noise and pointing the opposite way. A visible slice of the workforce started reporting they were struggling, and the assessment record for those years looks the same as before.

The one modest leak: manager goal reviews

Manager reviews are the exception, and a faint one. Managers rate goal attainment on a four-point scale, and here self-reported strugglers do score lower: 81.9% are rated as having met their goals, against 87.4% of self-assured performers. Managers mark strugglers below goal 1.4 times as often (18.1% vs 12.6%).

The pattern fits a simple account. What a person calls "struggling" is mostly an internal state: strain, self-doubt, the feeling of running harder to stand still. Colleagues rate output and collaboration, which apparently hold up. Managers track goals, where sustained internal strain might eventually surface. The instruments disagree because they measure different things, a gap psychologists have documented as the illusion of transparency: people systematically overestimate how visible their internal states are to others (Gilovich, Savitsky & Medvec, 1998).

What this means

The practical conclusion is not that one instrument is right and the other wrong. The two carry non-overlapping information, and each fails in the direction the other covers.

Limitations

• Colleague feedback is heavily positive-skewed (73% of items are Exceeding or better) and often requested by the receiver, which selects toward favorable raters. A blunter assessment channel might detect more.
• The "concern" threshold (mean rating below Exceeding) is a norm-relative choice; results are similar with top-box definitions, but any binarization loses information.
• People answering "Dislike" are rare (2% of answers) and drop out of the panel at higher rates (31–33% next-wave continuation vs 50% for the happiest state), so dislike-state estimates are indicative only, and some of the apparent recovery from dislike is exit.
• Repeat responders skew toward engaged panel stayers; the transition analysis describes people who kept answering.
• Manager reviews cover 19% of answers and skew to companies running structured reviews.
• Ratings cluster within raters and companies; the company-demeaned check controls rating norms, not who chooses to rate whom.
• The 2026 wave was still in the field when this analysis ran, so any level reported for that year is provisional. The aggregate comparison above uses only the two complete waves, 2024 and 2025.