Wellness MetricsAbout the Data
Short reference for sources, definitions, and limits. The full method lives on Data Workflow.
Analysis stays at the population level. We do not impute suppressed cells and we do not treat suicidal thoughts as suicide deaths.
Sources
- Statistics Canada 13-10-0972 — perceived mental health (CCHS)
- Statistics Canada 13-10-0465 — suicidal thoughts and related indicators
- Statistics Canada 13-10-0802 — stress and coping
- CIHI / PHAC Health Infobase — mental-health services
Expected files
| File | Role | Status |
|---|---|---|
| perceived_mh_annual.csv | Perceived mental health (CCHS) | Loaded |
| suicidal_thoughts.csv | Suicidal thoughts and positive mental health | Loaded |
| stress_coping.csv | Stress and coping | Loaded |
| cchs_mh_disorders.csv | CCHS mental health disorders | Loaded |
| perceived_health_quarterly.csv | Perceived general health (context only) | Loaded |
| cihi_mh_services.csv | CIHI mental-health services | Loaded |
Trend Direction Model
The Trend Direction Model analyzes historical StatCan/CIHI observations and estimates whether the next reported percentage in a comparable series is more likely to move up or down. It is an analytical signal based on historical patterns, not a clinical prediction and not evidence of a statistically significant change.
| Model | Training data | Target | Test accuracy | Test macro-F1 | F1 (Down / Up) |
|---|---|---|---|---|---|
| HistGradientBoostingClassifier | Real, cleaned Statistics Canada and CIHI observations (data/processed/mh_long.csv) - no synthetic data | Direction (Up / Down) of the next reported percentage in a comparable series | 75.8% | 0.745 | 0.687 / 0.803 |
Important limitation: This model predicts the direction of the reported point estimate, not statistical significance. Confidence intervals on these estimates are often 7-11 percentage points wide, so a small predicted move may fall well within normal sampling variation. “Up” is not the same as “worse” - it depends on what the indicator measures.
Predictions are precomputed by the training notebook and served as a static file — the dashboard never trains or runs the model itself. Trained on 8,142 real cycle-to-cycle transitions built from the tables above; about 79% of observed transitions in this data are smaller than their own confidence interval.
Limitations
- Most tables have two or three far-apart cycles, not a continuous series.
- Perceived mental health is 18+ only. Age lives on other tables.
- CCHS excludes reserves, institutions, and full-time military.
- Status F and x cells are dropped, never filled.
- MHACS microdata and the synthetic Kaggle set are out of this dashboard.