Two kinds of data
| Type | What it is | HR examples | What it answers |
|---|
| Quantitative | Numbers you can count or measure. | Headcount, exits, days to fill a vacancy, survey ratings, absence days. | How much? How many? Is it rising? |
| Qualitative | Words and observations. | Open survey comments, exit interview notes, focus group notes. | Why? What is it like? |
The strongest HR stories use both: numbers show the size of an issue, and words explain it.
Core HR metrics and how to calculate them
| Metric | Formula | Notes |
|---|
| Turnover rate | Separations in the period ÷ average headcount in the period × 100 | Often split into voluntary (resignations) and involuntary (dismissals, redundancies). |
| Retention rate | Employees present at the start who are still present at the end ÷ employees at the start × 100 | Not simply 100 minus turnover, because turnover also counts people hired and lost within the period. |
| New-hire turnover | New hires who leave within a set time (often 90 days or 12 months) ÷ new hires × 100 | A direct measure of hiring and onboarding quality. |
| Time to fill | Days from approval of the opening to acceptance of an offer | Definitions vary; state yours. |
| Metric | Formula | Notes |
|---|
| Cost per hire | (Internal + external recruiting costs) ÷ hires | Include advertising, agency fees, recruiter time. |
| Absence rate | Days absent ÷ available working days × 100 | |
| Favourable score | Respondents choosing the top two options on a five- point agreement scale ÷ all respondents × 100 | The usual way to report engagement survey items. |
| eNPS | % promoters (9 to 10) − % detractors (0 to 6) on “How likely are you to recommend us as a place to work?” | Ranges from −100 to +100. Passives (7 to 8) count in the total but not in either group. |
| Survey response rate | Responses ÷ people invited × 100 | A low rate limits what you can conclude. |
Seven ideas that prevent most mistakes
| Idea | Why it matters |
|---|
| Rates need denominators. | “30 people left” means little without knowing how many people there were. Always ask “out of how many?” |
| Mean and median differ. | The mean (average) is pulled by extreme values. The median (middle value) is often fairer for pay and time-to-fill. |
| Do not average averages. | Departments differ in size. A company-wide figure must be calculated from totals, or weighted by group size. The worked example below shows the error. |
| Small groups are unreliable and identifiable. | One person changes a team of five by 20 points, and results for small groups can reveal who said what. Set a minimum group size (commonly 5 to 10) below which you do not report. |
| Correlation is not causation. | If low engagement and high turnover appear together, one may cause the other, or both may be caused by something else. Say “is associated with” unless you have stronger evidence. |
| One point is not a trend. | Compare with previous periods before calling something a rise or fall. |
| Who did not answer matters. | If only 50% of a department responded, the least engaged may be the missing half. |
The practice dataset (fictional) Brightwater Logistics, financial year 2026. Use this for the exercises.
| Department | Average headcount | Survey responses | Favourable engagement | Exits in year | Average time to fill (days) |
|---|
| Warehouse Operations | 200 | 120 | 58% | 76 | 41 |
| Customer Support | 80 | 64 | 71% | 18 | 33 |
| Department | Average headcount | Survey responses | Favourable engagement | Exits in year | Average time to fill (days) |
|---|
| Transportation | 60 | 30 | 62% | 16 | 47 |
| Corporate | 60 | 54 | 80% | 6 | 52 |
| Total | 400 | 268 | | 116 | |
Worked example: the averaging trap. Company turnover is total exits divided by total headcount: 116 ÷ 400 = 29.0%. If you instead average the four departmental rates (38.0%, 22.5%, 26.7% and 10.0%), you get 24.3%, which understates the problem by nearly five points because it gives the 60-person Corporate team the same weight as the 200-person warehouse. The same applies to engagement: weighting by responses gives 66.0% favourable; a simple average gives 67.8%.