Statistics for people analytics and HR research: your workflow, one tool
People analytics is where organisational research meets business decisions. The daily work is a compact toolkit — engagement-scale reliability, turnover prediction, promotion-time analysis, pay-equity comparisons — applied on modest samples with tight timelines. This post maps that workflow onto what thericerca does.
What people-analytics statistics actually looks like
A typical people-analytics project answers one of four questions: are our engagement / satisfaction / burnout scales reliable (measurement); who leaves and when (turnover / retention); what predicts an outcome we care about (performance, engagement, promotion); and are there differences between groups (representation, pay, promotion rates). The samples are usually smaller than product-analytics data but need more careful inferential work — the stakes are people's careers.
People analytics rarely stumbles on statistical ambition. It stumbles on an engagement average without a confidence interval, and a "significant" pay-gap finding that never checked for confounding.
Your typical study, mapped
- Engagement / burnout / satisfaction scale reliability — Cronbach's α with an item-drop table; McDonald's ω for a strict take.
- Engagement dimensionality (a big battery of items) — exploratory factor analysis to see if your "engagement" is really one thing or three.
- Engagement across departments / tenure bands / managers — one-way ANOVA / Kruskal-Wallis with post-hocs.
- Turnover prediction (leaves in 12 months, yes/no) — logistic regression, with hazard-ratio-style odds ratios and 95% CIs.
- Time to exit / time to promotion — Kaplan-Meier by group + log-rank + Cox regression for the multivariate story.
- Representation across levels — chi-square with Cramér's V, or Fisher's exact for small counts.
- Pay-equity comparison — multiple regression on salary, adjusting for role, tenure, and location — the "unexplained gap" analysis.
- Pre/post organisational-change survey — paired t-test (matched respondents) or independent-samples t-test (different waves).
- Inter-rater agreement (calibrated performance ratings) — Cohen's κ (two raters) or ICC (multiple raters).
The workflow, cell by cell
- Export your survey or HRIS extract — CSV, XLSX, or SPSS all work.
- Enter the objectives (e.g. "compare engagement by function", "predict 12-month exits from role, tenure, and manager").
- thericerca picks the right test per objective, checks assumptions, and applies non-parametric fallbacks when needed.
- Post-hocs fire only after significant omnibus tests, so pairwise comparisons never over-claim.
- Download the report as DOCX (for the leadership readout), PDF (for the archive), and a Jupyter notebook (for the analytics-team's evidence trail).
For quick one-off checks, the free tools land the most common asks: the reliability tool for a fast Cronbach's α on any Likert scale, the sample-size calculator for engagement-survey planning, and the test-picker when you're not sure whether a comparison should be a t-test or a Mann-Whitney U.
When you'd need something else
Beyond thericerca's scope. Propensity-score matching for policy evaluations (MatchIt / WeightIt). Multilevel models with random slopes for employees-within-teams-within-locations data (nlme / lme4). Structural equation modelling for engagement-driver models that need latent variables (lavaan). Causal-impact estimation for a specific policy change. Network analysis for organisational structures (igraph). Machine-learning models for large-scale attrition prediction (scikit-learn, xgboost). For any of these, thericerca still handles the descriptive and primary-analysis layer of the study — then hands off.
A concrete example
A first-year attrition study on 400 new hires: engagement survey at 30 days, exit status by 12 months, and demographic + role variables from HRIS.
- Cohort snapshot — descriptives + frequencies as Table 1 (age, tenure at 30 days, function, location, manager tenure).
- Engagement scale reliability — Cronbach's α on the 8 items, plus item-drop.
- Engagement across functions — one-way ANOVA with post-hoc.
- Predictors of 12-month exit — logistic regression on function, engagement score, manager tenure, and role level; odds ratios with 95% CIs.
- Time-to-exit by function — Kaplan-Meier + log-rank across functions; then Cox regression to adjust for the covariates.
All of that runs in a single thericerca project. The leadership readout comes out as a DOCX; the archive gets the Jupyter notebook.
The short version. If your people-analytics work is survey- and cohort-shaped — engagement, turnover, promotion, pay equity — thericerca covers the full workflow. Reach for the specialist tools only when your question specifically demands them.