Case Study 05

Reading Between the Lines: How Workplace Grudges Shape the Interpretation of Coworker Messages

LSE · MSc Behavioural Science

A three-arm randomised controlled experiment with 329 UK employees, testing whether interpersonal history quietly biases how we read neutral, ambiguous instant messages.

MANAGER When can you get the presentation deck done by? TWO READINGS Just asking about timing. Why isn't it done already?

Timeline

Q4 2025

MSc dissertation, LSE

Sample

329

UK employees via Prolific, across 3 randomised conditions

Methods

Randomised controlled experiment · ANOVA · Regression · Mediation analysis

Tools

Qualtrics · Prolific · Stata

The Question

Does a grudge change how we read a coworker's messages, even when the words are neutral?

A colleague once showed me a message from her manager: "When can you get the presentation deck done by?" She read impatience. I read a plain question about timing.

I kept noticing the pattern: neutral messages read through the lens of old frustration. And because instant messaging strips away tone and cues, short messages are wide open to it. No study had tested this directly, so I designed an experiment to find out.

A person at their desk reading an ambiguous instant message from their manager that says: When can you get the presentation deck done by?
The same six words can read as impatience or as a neutral ask, depending on who you picture sending them.
What I Tested

Three questions: does a grudge bias the reading, what drives it, and does it compound?

Relationship grudge · resolved · neutral Rumination Tone reading H1 · direct effect H2 via rumination H3
H1 tests the direct effect, H2 whether rumination carries it, H3 whether a negative reading feeds back into rumination.
H1 · The effect

People holding a grudge read ambiguous messages more negatively than those who resolved a grudge, or feel neutral.

H2 · The cause

Is that negativity driven by rumination, the act of stewing on the coworker?

H3 · The loop

Does a negative reading then feed more rumination, becoming self-reinforcing?

Study Design

Everyone rated the same ten messages. Only the relationship behind them changed.

I ran this as an online experiment inside a Qualtrics survey. 329 UK employees, recruited via Prolific, were randomly split into three groups, and every group then completed the same session.

329 employees randomly assigned Grudge n = 109 still resents them Resolved n = 111 believes it's settled Neutral n = 109 feels neutral
329 employees, randomly split into three groups. Each recalled a different relationship, then all rated the same ten messages (below).
Then every group ran the same session

From there, every participant followed the same 15-minute survey, moving through six steps in order.

1

Baseline

Emotion and trait rumination measured before any conflict recall, so the starting point wasn't coloured by memories.

2

Induction

A five-minute writing task about the assigned coworker evoked the intended emotional state.

3

Rumination T1

A timed one-minute pause, then participants reported the share of time spent thinking negatively. The pause captured rumination indirectly, without asking for it head-on.

4

Tone task

Participants rated the ten ambiguous messages, each shown as if sent by the recalled coworker, on a 7-point scale.

5

Rumination T2

A second pause and the same measure, to test whether the negative reading fed later rumination.

6

Debrief

Demographics, then a positive-affect video so no one left the study in a negative state.

The ten messages rated in the tone task (step 4)

  1. "Do you have a minute?"
  2. "Let's discuss this after the meeting."
  3. "I don't understand what you mean. Please clarify."
  4. "Has this been reviewed?"
  5. "Don't forget we have a meeting later at 3 pm."
  6. "When was the decision to go ahead made?"
  7. "Has this task been completed?"
  8. "I will need this done by tomorrow."
  9. "Is this file the latest version?"
  10. "Let me handle this task."

Rated one at a time on a 7-point scale (very negative to very positive), each shown as if sent by the recalled coworker.

How I Analysed It

Power analysis, randomisation and manipulation checks, ANOVA, regression, and mediation, run in Stata

Each stage below shows its purpose and its key statistical result.

Stage 1
Power analysis
Size the study before running it.
No prior benchmark, so I anchored the effect estimate to a related anger study (d ≈ 0.40). At 80% power, α = .05, that required 99 per group; I targeted 109 to absorb attrition.
Stage 2
Randomisation checks
Confirm the three groups were comparable.
Pearson χ² across gender, age, ethnicity, and working hours showed balance. Recency of contact and supervisor/peer status differed by design and were carried forward as covariates.
Stage 3
Manipulation check
Verify the writing task shifted emotion.
One-way ANOVA on the change in emotion: anger F(2,326) = 70.0, p < .001, with the same direction for positive and negative emotion, Bonferroni-confirmed.
Stage 4 · H1
Primary test · ANOVA
Did the relationship change the reading?
One-way ANOVA with Bonferroni contrasts. Condition explained 27% of variance, F(2,326) = 60.9, p < .001; all three groups separated, every 95% CI excluded zero.
Stage 5
Robustness · regression
Does the effect hold after adjusting for covariates?
Re-ran as pairwise linear regressions adding trait rumination, gender, and recency. The condition effect held essentially unchanged; gender added only ΔR² ≈ 0.01.
Stage 6 · H2
Mediation
Was rumination the mechanism?
Causal mediation per contrast, ACME via quasi-Bayesian Monte Carlo. Rumination did not mediate the grudge effect; a small partial mediation appeared only for the resolved group.
Stage 7 · H3
Hierarchical regression
Does a negative reading feed later rumination?
Two-step model: T1 rumination first (adj R² ≈ .70), then adding tone reading → β = −2.53, p = .01 (a small but significant ΔR² ≈ .005).
What I Found

A grudge made identical, neutral text read as negative. And "resolved" didn't fully reset it.

H1Supported

Grudge 2.87, resolved 3.36, neutral 4.11 on a 7-point scale (lower = more negative). Condition explained 27% of the variance (F(2,326)=60.9, p<.001). The grudge-to-neutral gap was 1.25 points, from text that was neutral.

1 very negative 4 neutral 7 very positive Grudge 2.87 Resolved 3.36 Neutral 4.11
Mean perceived tone by condition. The grudge group sits well into negative territory reading text that was designed to be neutral.
The surprise: the resolved group still read messages more negatively than neutral (difference 0.75, 95% CI [0.48, 1.02]). Subjective resolution left a residue.
H2Not supported

Grudge-holders ruminated more, but rumination didn't predict their negative readings. The judgement looked fast and automatic, an on-guard response. It did partially mediate for the resolved group, where the delay appears to have reactivated dormant memories.

Grudge vs neutral Rumination Tone reading a = 30.3*** b = n.s. c′ = −1.17*** Total effect c = −1.25***. Rumination didn't carry it.
For the grudge contrast, the path through rumination (a → b) was broken by a non-significant b. The direct effect (c′) stayed strong.
H3Supported

A negative reading predicted higher rumination minutes later (β=−2.53, p=.01), after controlling for prior rumination. Small per message, but across the hundreds of exchanges a strained relationship generates, it compounds.

Negative reading Rumination ↑ Primed for the next message × hundreds of messages
A small effect per message, but the loop tightens the more a strained relationship plays out over time.
Why It Matters

The bias starts before a word is read, and it accumulates unnoticed.

A grudge tilts an ordinary message toward a negative reading, and even a grudge believed resolved can tilt it too. Clarifying tone over IM feels awkward, so people absorb the negative read and move on. Over hundreds of messages, that is how minor friction hardens into lasting tension, eroding cooperation, wellbeing, and performance.

Two levers follow: early conflict resolution, because subjective resolution doesn't fully reset perception, and psychological safety, because the fix depends on people feeling able to check intent before assumptions calcify.

Even a grudge believed to be resolved can quietly tilt how we read the next message.