Eleven dashboards on the data-outcome side: what is the data showing across descriptive, comparative, regression, longitudinal, and equity views? Two surfaced (Description, Open-Ended); the rest live in the More analyses dropdown so the surfaced row stays focused on credibility-first reads.
What do people's answers show?
Per-item mean, SD, distribution shape, response-over-time chart, and distribution histogram. Per-group rollups when a grouping variable is available.
What themes appear in written responses?
Aggregate open-text responses across every open question. Theme overview, theme explorer, response review table. AI work gated behind a Generate button so you control the spend.
How do groups differ?
Welch's t-test, Mann-Whitney U, one-way ANOVA, Kruskal-Wallis. Cohen's d with 95% CI and eta-squared. AI verdict card translates the test into HR language.
How did people change?
Paired t-test or Wilcoxon signed-rank, Cohen's d_z, Jacobson-Truax Reliable Change Index at 90, 95, 99% confidence with per-respondent improved / unchanged / declined counts.
How do outcomes split across groups?
| Subgroup | n | Mean | d vs all |
|---|---|---|---|
| Engineering | 84 | 4.21 | +0.18 |
| Sales | 66 | 4.05 | +0.02 |
| Support | 58 | 3.95 | -0.13 |
| Operations | 104 | 3.78 | -0.34 |
| Hidden for k-anonymity (n < 5): 1 group | |||
Per-subgroup mean, SD, 95% CI, and Cohen's d versus everyone else, ranked by absolute gap. K-anonymity hides rows below the privacy threshold so a department of three never appears.
Where are the equity gaps across the org?
| Axis | Max |d| | Verdict |
|---|---|---|
| Tenure band | 0.82 | Large |
| Race / ethnicity | 0.55 | Meaningful |
| Role / level | 0.28 | Small |
| Age band | 0.14 | Parity |
Per-grouping-axis means, omnibus ANOVA, pairwise Cohen's d, and a 0-100 Equity Gap Score where 100 = parity. K-anonymity floor of 5 hides small subgroups entirely.
What predicts this outcome?
Three modes: Simple (OLS or logistic regression), Mediation (X to M to Y; bootstrap percentile or BCa CI on the indirect effect), Moderation (Johnson-Neyman plot). VIF flagged for multicollinearity.
Which factors matter most?
Pearson r per driver, standardized regression betas, and Johnson's Relative Weights for continuous outcomes. Plus the Action Priority Map 2x2 with Fix First / Protect / Do Not Overinvest / Monitor labels.
How do items behave across trait levels?
Four models from one dropdown: Graded Response (default for Likert), 2PL and 3PL for dichotomous items, two-dimensional MIRT with closed-form Varimax rotation. Test information function plus item characteristic curves.
Do nested groups change the effect?
Two-level linear mixed-effects (REML / EM), two-level logistic GLMM (PQL), three-level linear. ICC at every level, Satterthwaite or Kenward-Roger df corrections, LR test against the simpler model.
What is changing over time?
Wave detection from channel tags or quarterly bins. Trend Score 0-100, per-wave composite mean line, per-construct sparklines, Welch's t-test current vs prior wave.