Multi-Subject Studies: Observing Complete Teams
Observe whole teams while distinguishing visits, recorded participants and statistical units of analysis.
Updated:
Recording several people on a visit can reduce travel effort. This operational advantage does not make every participation independent. Distinguish visits, people recorded and the population whose proportion is being estimated.
Three different questions
- Aggregate person-hour distribution: estimate how collective time is distributed, weighting people or groups by represented exposure.
- Distribution for each individual: each estimate needs sufficient precision and a justified purpose; the aggregate sample does not guarantee it.
- Joint team state: record whether a cell is blocked or how many people are waiting. The unit may be team-instant rather than person-instant.
A round of ten people creates one visit and ten participations; walking around means each reading has a different actual time. They are not literally observed simultaneously.
Size the sample for the estimand
The simple planning formula n=ceiling[z²p₀(1−p₀)/e²] gives 1,068 records with z=1.96, p₀=0.5 and e=0.03. It is a reference for one proportion with independent records. It does not automatically require 1,068 per person for an aggregate result, or allow division by ten when ten people share a stoppage.
Within-visit correlation and repeated observations of people can increase uncertainty. Cluster-based analysis, stratification or person-instant selection must match the case. A maximum-subjects-per-observer rule does not replace this assessment.
Organise rounds
Define categories and units, shifts and days, subject selection, planned and actual times and treatment of absence. Do not substitute the nearest person for someone missing without a rule. Maintain safety and avoid routes that always observe some stations before a state change.
Train observers using operation, setup, transport, waiting, coordination, break and other examples. Record facts rather than infer “unjustified absence” from not seeing someone. Randomisation and rotating observers do not guarantee removal of reactivity or classification bias.
Simulated twelve-person example
Assume five eight-hour days with twelve random round starts per hour: 480 rounds and up to 5,760 participations if all twelve people are recorded each time. Check route and recording feasibility; twelve starts per hour do not guarantee every round can finish without overlap.
Assume rounded aggregate shares of 58% direct work, 22% waiting/transport, 12% setup and 8% administration. They sum to 100%, but waiting and transport should be separated for logistics decisions.
If one person had 480 independent records and p̂=0.58, their approximate normal 95% margin would be 4.4 percentage points. That margin does not automatically apply to the team, nor should 5,760 correlated participations be treated as independent. The 42% outside direct work is not removable savings.
Metrics, rights and the tool
Sampling describes proportions and guides workload investigations. It does not independently establish a bottleneck or safe staffing. OEE needs compatible time data, ideal cycle time and total and good units; a table of active people is insufficient.
Minimise identifiers and review purpose, legal basis, access and retention. Team reporting does not necessarily anonymise underlying records. Do not assume automatic compliance with legislation or collective agreements.
WorkSamp supports multiple-subject participation recording; interpret calculations using the actual design. Exporting data is not automatic ERP, maintenance-system or dashboard integration.