# Slides — Phase B Outcome Correlation · Trellison Institute Briefing Deck
**Audience**: policy briefing · academic seminar · journalism brief
**Length**: 10 slides + title + closer = 12 total
**Speaker notes**: ~30 seconds per slide; 8-minute briefing or 20-minute deep-read
---
## SLIDE 0 — Title
**State-Level Youth Mental Health Need-vs-Access Measures Predict Within-Instrument Suicide Ideation but Not Across-Instrument Mortality**
A Phase B follow-up to *Youth Mental Health Access Gap V1*
Trellison Institute · May 2026 (v1.0 draft)
*Methodology rating: pending · Evidence-chain certificate: pending*
Speaker note: Phase B asks whether the framework's gap measures predict real outcomes. The answer is informative for the framework's interpretation.
---
## SLIDE 1 — The question
> Does the Need-vs-Access Framework's state-level gap classification predict state-level outcomes — youth suicide ideation, suicide attempts, all-age suicide deaths, drug overdose deaths?
The Youth V1 paper identified 3 positive outliers (PR, NC, NJ — supply worse than uninsured predicts) and 2 negative outliers (VT, AK — supply better than predicts).
Phase B tests: are the gap measures policy-relevant in the outcome sense?
---
## SLIDE 2 — The data we joined
[Visual: schematic flow diagram, 4 sources → 1 joined dataset]
| Source | Year | Granularity |
|---|---|---|
| Youth V1 gap measures | 2023 | 35 states |
| CDC YRBSS Mental Health Indicators | 2023 | State (grades 9-12) |
| NCHS bi63-dtpu (Leading Causes — Suicide) | 2017 | State (age-adjusted) |
| NCHS VSRR + ACS pop | 2023 | State |
Output: `analysis_outputs.mh_gap_youth_outcomes_v1` — 35 states × 11 fields.
---
## SLIDE 3 — The headline correlation matrix
| Gap × | YRBSS considered | YRBSS made plan | YRBSS attempted | AADR 2017 | Drug OD |
|---|---:|---:|---:|---:|---:|
| need_value | **+0.82** | **+0.80** | **+0.76** | +0.34 | +0.19 |
| covariate (uninsured %) | +0.43 | +0.59 | +0.43 | **+0.46** | -0.21 |
| access (providers/100K) | -0.03 | 0.00 | +0.11 | +0.14 | **+0.21** |
| log_gap_ratio | -0.03 | -0.02 | +0.03 | -0.16 | -0.10 |
| residual_z | -0.14 | -0.16 | -0.08 | -0.30 | -0.05 |
Strong (bold) → within-YRBSS coupling and uninsured × AADR.
Null → access × outcomes.
Counter (positive) → access × drug OD (deaths-of-despair signature).
---
## SLIDE 4 — Strong: within-instrument coupling
The need metric (sad/hopeless 2+ wks) is a **strong cross-sectional proxy** for state-level YRBSS suicide indicators:
- r = +0.82 with considered suicide
- r = +0.80 with made plan
- r = +0.76 with attempted suicide
The Youth V1 paper chose the right need signal. State-level distress prevalence captures the same underlying pattern as the more severe suicide questions.
---
## SLIDE 5 — Null: state-level supply is not protective
| `access_value` × | r |
|---|---:|
| YRBSS considered | -0.030 |
| YRBSS attempted | +0.109 |
| All-age AADR | +0.143 |
| Drug OD | +0.211 |
**State-level provider density does NOT predict reduced mortality.** It modestly *positively* correlates with drug OD.
[Visual: scatter plot — access (x) vs all-age AADR (y), Alaska in upper-right (high supply + high AADR)]
---
## SLIDE 6 — Why null? Three explanations
1. **Capacity ≠ count.** NPI registration measures structure, not delivery, not hours, not accepting-new-patients.
2. **State granularity is too coarse.** 500 providers in one metro + 0 in 60% of land area looks the same as 250 distributed evenly.
3. **Rural-state confounding.** AK, VT, MT, WY, NM have idiosyncratic supply patterns (IHS, university medical centers) AND idiosyncratic suicide patterns (firearms, isolation). Cross-section confounds.
---
## SLIDE 7 — The orthogonal residual signal
The framework's residual_z (gap-vs-uninsured deviation) is essentially **uncorrelated** with mortality:
- r(residual_z × all-age AADR) = -0.30
- r(residual_z × drug OD) = -0.05
[Visual: outlier-state outcome bars]
**NJ** (positive outlier, supply gap): YRBSS attempted **5.2%** (LOWEST in dartboard), AADR **8.3** (3rd lowest in US).
**AK** (negative outlier, well-supplied): YRBSS attempted **19.0%** (HIGHEST), AADR **27.0** (3rd highest).
The residual captures *policy-relevant workforce capacity gap given insurance landscape*. It is NOT a mortality-risk signal.
---
## SLIDE 8 — Different lists for different questions
| Question | Right list |
|---|---|
| Workforce build-out priority | PR · NC · NJ (positive outliers) |
| Highest youth distress | IN · AR · OK · NV · MO (from Youth V1) |
| Suicide-prevention priority | MT · AK · WY · NM · ID (highest AADR) |
| What's working pattern | VT (UVM + Medicaid expansion + Designated Agencies) · AK (IHS) |
**The framework's outputs are the workforce priority list. They are not the suicide-prevention triage list.**
Conflating them mismatches intervention to problem.
---
## SLIDE 9 — Drug overdose is a different question
[Visual: state-level scatter — uninsured rate (x) vs drug OD rate (y) showing weak negative correlation]
**Drug overdose** has a fundamentally different state-level signature:
- Higher coverage states have *higher* drug OD rates (r = -0.21 on uninsured).
- DC: 3.6% uninsured under-19, drug OD = 95.0 per 100K.
- WV: 3.1% uninsured under-19, drug OD = 79.1 per 100K.
This is the "deaths of despair" geography. It requires a separate analytical pipeline (NPPES SUD treatment + SAMHSA OTPs + NSDUH OUD-receiving), not the mental-health-care access framework.
---
## SLIDE 10 — Limitations + what's next
**v1.0 limitations**:
- Cross-section; no causal identification
- AADR is 2017 (6-year lag)
- All-age outcomes, not youth-specific
- No FBI crime; no HCUP ER
**v1.1 queued**:
- FBI Crime Data Explorer (API key pending issuance — HIT open)
- CDC WONDER state-level youth (10-19) mortality (XML POST API)
- Multivariate residual analysis (uninsured + income + rural-share)
**v2.0 queued**:
- HCUP State ED licensed integration
- Longitudinal panel (YRBSS biennial + multi-year AADR)
---
## SLIDE 11 — Q&A / Discussion
The framework's value is preserved by this finding. What is qualified is the interpretation.
Anticipated questions cataloged in `mh_gap_youth_outcomes_v1_press_qa.md` (12 entries).
Contact:
- [email protected]
- [email protected]
- [email protected]
*Thank you.*