The Turnover Number That Wasn't
A city's HR analyst sends the city manager a quarterly dashboard covering 2,400 budgeted FTE. Reported annualized turnover jumped from 9.8 percent to 14.2 percent. The analyst computed it as 79 recorded separation events divided by the quarter-end headcount of 2,220 (the hiring freeze left 180 positions vacant), multiplied by four. On review, the 79 events include 12 internal transfers between departments, 25 seasonal lifeguards whose appointments ended in September as scheduled, and 5 retirements. Exit-reason data comes from a free-text field that is blank for 38 percent of separations. Two departments merged in month two, so cost-center codes changed mid-quarter. The city manager wants to announce a "retention crisis" at the next council meeting and asks HR to "just tell me who is quitting and why."
1. Recompute the measure. What is the defensible turnover figure, and what did the analyst get wrong?
Read the worked answer
Three errors: wrong numerator, wrong denominator, and an unexamined comparison.
Numerator. Internal transfers are not separations from the organization; count them separately as a transfer rate. Scheduled seasonal ends are planned and should be reported on their own line, not blended into turnover. Retirements are separations but are usually reported apart from voluntary quits because they are largely predictable.
79 − 12 transfers − 25 seasonal ends − 5 retirements = 37 voluntary separations.
Denominator. A quarter-end headcount frozen by vacancies understates the base and inflates the rate. Use the average of beginning and ending filled positions: (2,400 + 2,220) / 2 = 2,310.
Rate. 37 / 2,310 = 1.60% for the quarter; annualized ≈ 6.4 percent.
Comparison. The prior 9.8 percent must be recomputed on the same definitions before anyone compares the two. If it was built the old way, the "jump" may be entirely an artifact of method.
State the figure as "voluntary separations, annualized, excluding internal transfers and scheduled seasonal ends, on an average filled-position base." A number without its definition invites exactly this kind of misreading.
2. What must be fixed in the HRIS and reporting process so this does not recur?
Read the worked answer
- Codify separation type. Replace the free-text exit-reason field with a required coded list (voluntary resignation, retirement, involuntary, seasonal end, transfer, death, disability) plus an optional comment field. Free text cannot be aggregated and produces the 38 percent blank problem.
- Make org codes effective-dated. When departments merge, historical records must retain the structure in effect on the transaction date so trend comparisons are not broken.
- Publish a data dictionary with one approved definition of turnover, the numerator and denominator rules, and the treatment of transfers, seasonals, and retirements. Any dashboard number should trace to it.
- Reconcile against payroll each quarter before publishing; HRIS and payroll rarely disagree by accident.
- Build exception reports for separations missing a type, position records missing an org code, and vacancies open beyond a threshold.
- Document the change in method and, where practical, restate prior periods so the series remains comparable. A method change announced in a footnote is defensible; one discovered by a council member is not.
3. What can and cannot be concluded from this data, and what should HR tell the city manager?
Read the worked answer
Can be said: voluntary separations are running at roughly 6.4 percent annualized on the corrected basis; retirements, transfers, and seasonal ends are separate, largely planned categories; the prior figure must be restated before any trend claim is made.
Cannot be said: that there is a retention crisis, or why anyone is leaving. With 38 percent of exit reasons missing, the exit data is not a sample of leavers - it is a sample of leavers who filled in a box. And at 37 separations across 2,310 employees, splits by department or job class become very small numbers; a rate built on three exits is noise.
What to tell the manager: the corrected number, the definition behind it, the fact that the comparison to last quarter is not yet valid, and a plan - run the analysis by department, tenure band, and job class using 12 months of data rather than one quarter; restart structured exit interviews with a standard instrument; check compensation and supervisory indicators in the classes with the most exits. Then give the council a factual statement with a date for the deeper analysis instead of a crisis narrative the data cannot support.