Engagement survey study: from fieldwork to a funded decision

Palmstone Group (a fictional services company with 1,400 employees in six departments) invited every employee to its annual engagement survey: a census design, open for three weeks. 778 people responded (55.6%).

The executive committee is worried about retention. Item q6 asks “I intend to be working here in 12 months' time” (1 = strongly disagree … 5 = strongly agree). The Chief People Officer has a proposal on the table: a retention and workload programme for the departments with the weakest intent to stay, costing SAR 120,000 in fixed design costs plus SAR 1,500 per employee covered. For this scenario, assume that replacing a leaver costs about SAR 40,000 (recruitment, onboarding and lost productivity). These costs are planning assumptions for a fictional company, not industry figures; real replacement costs vary widely by role.

You are the people-analytics lead. The committee wants a written study, not a dashboard:

1. Can the survey be trusted? Who responded, how well, and what the design implies for next year.
2. What do staff say about staying and about workload? Summarised in a way that respects the 1–5 scale.
3. What is the company-wide picture once each department counts in proportion to its size, and how uncertain is it?
4. Do groups differ (departments; remote vs onsite staff) — without fishing through dozens of tests?
5. Should the programme be funded, for whom, and how will we know whether it worked?

Constraints. Results are reported by department only (no group under 10 people is ever shown), individual answers are confidential, and the programme budget must be approved before the next survey cycle.

The data. engagement_survey has one row per respondent (department, tenure band, work mode, six Likert items stored as codes 1–5 with some items skipped, completion time in seconds, submission date). engagement_population has each department's headcount, responses and response rate. Both datasets are synthetic — generated for teaching, describing no real company or employee — so none of the results is evidence about real workplaces.

Stakeholders. The executive committee (decides the budget), the Chief People Officer (sponsor of the programme), department heads (whose teams will be compared), and employees (who were promised confidentiality and action).