Attention to co-worker contribution and perceived own contribution in construction teams: a weekly field experiment

A short weekly survey asks field employees of an electrical contractor to estimate how much of their team's work over the past seven days they did themselves and how much each of their closest co-workers did. Respondents are randomly assigned, within work site, to one of two orders in which these two sets of questions are presented. The questions, and the rest of the survey, are identical across arms.

2026-10-05

2026-11-13

Claimed own share of the team's work over the past seven days, 0 to 100. The confirmatory contrast is at Survey 1, the first exposure to the manipulation; the pooled contrast over Surveys 1 to 5 is secondary.

Taken directly from the own-contribution screen as entered. Implied responsibility, the claimed share multiplied by the size of the respondent's reference set (the respondent plus each of up to five co-workers not marked N/A), is reported as the first secondary outcome; it is the measure comparable to the prior literature but depends on a reference-set size recorded after randomisation, and it is reported with that caveat.

Implied responsibility; perceived fairness of workload sharing (0 to 10); perceived effort of other team members (0 to 10); interest in working with this team again and enjoyment of working in this team (1 to 7 each); ease of observing how team members are working (0 to 10); likelihood of five responses to an underperforming team member, the focal item being talking directly to them (four-point scales); commitment to the project's success (1 to 5); four common-interest and team-identification items (1 to 5); two items on an employee programme asked in later waves only; weekly hours worked and safety incidents from company records.

All survey items are analysed as entered. Hours worked and safety incidents come from the company's administrative records, linked to survey responses by an employee identifier and de-identified before analysis; they are examined as before-and-after contrasts around the announcement and as order-by-period interactions, with no predicted direction.

A two-arm survey experiment delivered in five weekly waves to field employees at participating work sites of one company. Individuals are randomly assigned to arms in a single draw before the first wave, blocked on work site so that each site contributes equal numbers to both arms, and assignment is held fixed across all five waves. Each participant receives a personal survey link by text message from the research team each week, with one reminder to those who have not completed that week's survey, and the company sets aside working time on site for completion.

Not available

Randomisation done by computer: a seeded script, archived at registration, assigns individuals to arms within each work site using the company's roster.

Individual, blocked on work site.

No

Not applicable; individual randomisation across 184 individuals.

184 individuals; up to five person-wave observations each, so up to 920 person-waves.

92 individuals self-first (control), 92 individuals others-first (treatment), balanced within each site.

Outcome: claimed own share of the team's work, in percentage points (0 to 100). Randomisation is at the individual level, blocked on work site, so there is no clustering; the calculation allows for site fixed effects absorbing an assumed 15 percent of the variance and, for the pooled estimate, an assumed within-person correlation across waves of 0.5. Assuming 90 percent response at Survey 1 falling five percentage points per wave (83 per arm at Survey 1, 64 at Survey 5), the minimum detectable effect for a two-sided 5 percent test with 80 percent power is about 0.40 standard deviations at Survey 1 and about 0.32 standard deviations pooled over the five waves. The standard deviation of claimed share is not known before fielding; if it lies between 15 and 20 percentage points, the Survey 1 figure corresponds to roughly 6 to 8 percentage points of claimed share and the pooled figure to roughly 5 to 6 points, against a control-arm mean likely to be near 20 to 25 percent in teams of four to five. The closest prior study (Schroeder, Caruso and Epley 2016, Experiment 3) found an effect of about 0.27 standard deviations on the same contrast; power against an effect of that size is about 0.47 at Survey 1 and about 0.66 pooled.