# The Map Doesn't Predict the Outcome — Phase B Narration Script
**Version**: v1.0
**Target**: ~640 words at 78 wpm ≈ ~490s
**Source data**: `analysis_outputs.mh_gap_youth_outcomes_v1` (35 states × 11 fields)
**Methodology rated**: Trellison Institute. Verified by LedgerWell.
**Companion to**: Youth Mental Health Access Gap V1 + Mental Health Access Gap V1 (adults)
---
## Panel 1 — The Question (0–80s, ~105 words)
> Last month we mapped the Youth Mental Health Access Gap at state granularity. The Need-vs-Access Framework identified three states whose youth-serving workforce was substantially undersupplied relative to their insurance landscape — Puerto Rico, North Carolina, New Jersey. It identified two whose workforce was substantially overbuilt — Vermont, Alaska.
>
> The natural follow-up question is whether those identifications *predict* anything. Do the under-supplied states have worse outcomes? Do the well-supplied states have lower suicide rates?
>
> Phase B tests this. We joined the gap measures with CDC YRBSS state-level youth suicide indicators, NCHS state-level age-adjusted suicide death rates, and NCHS drug overdose deaths.
>
> The answer is informative for the framework's interpretation.
---
## Panel 2 — Strong: Within-Instrument Coupling (80–170s, ~130 words)
> The strongest signal in the entire Phase B analysis is the within-YRBSS coupling. State-level prevalence of sad-or-hopeless 2-plus weeks correlates with state-level YRBSS suicide questions at r equals 0.82 for considered suicide, 0.80 for made a plan, 0.76 for actually attempted suicide.
>
> Translation: states where more teens report sustained distress also have more teens reporting suicide ideation, planning, and attempts. The pattern is tight and consistent.
>
> This validates the Youth V1 paper's choice of the sad-or-hopeless prevalence as the principal need metric. We picked the right state-level signal. The same distress-prevalence measure that we use to compute the gap ratio captures the underlying suicide-ideation pattern at the state level.
>
> If we wanted a state-level surveillance proxy for youth suicide indicators, this is it.
---
## Panel 3 — Modest: Across-Instrument Signal (170–260s, ~115 words)
> The signal is weaker for the across-instrument outcomes.
>
> The same need metric correlates only 0.34 with the NCHS all-age suicide age-adjusted death rate, and 0.19 with the drug overdose rate.
>
> Two factors drive the degradation. The demographic mismatch — YRBSS measures grades nine through twelve; NCHS AADR is all-age across the full population. And the rarity — state-level suicide deaths are rare events for any single age cohort, and the signal-to-noise ratio is lower than for the high-prevalence ideation questions.
>
> These are real correlations. State-level youth distress *does* track state-level mortality. But the predictive power is much lower than the within-instrument relationship. State-level surveillance of youth distress is not a substitute for direct mortality surveillance.
---
## Panel 4 — Null: Supply Does Not Predict Reduced Mortality (260–360s, ~125 words)
> The most policy-consequential finding is what *isn't* there.
>
> The framework's supply-side metric — youth-serving providers per one hundred thousand under-eighteen — correlates *zero* with YRBSS considered suicide. Zero point one one with attempted suicide. Zero point one four with all-age suicide AADR. Zero point two one with drug overdose.
>
> The drug overdose number is *positive*. Higher-supply states have higher drug-overdose rates.
>
> The Alaska case anchors this. Alaska has the highest youth-serving provider density in our dataset — one thousand eighty-five per hundred thousand. Alaska also has the highest YRBSS attempted-suicide rate. Nineteen percent. Nearly twice the national average. And the third-highest all-age suicide AADR — twenty-seven point zero.
>
> Counting NPI-registered providers does not predict reduced state-level mortality. The state-level cross-section data does not support that expectation.
---
## Panel 5 — Why Null? Three Explanations (360–440s, ~110 words)
> The paper does not claim to have identified *why* supply is not protective. It proposes three explanations.
>
> First — capacity is not count. NPI registration measures structure, not delivery, not hours, not accepting-new-patients status. A state with one thousand registered child psychologists may have very few of them taking Medicaid.
>
> Second — state granularity is too coarse. Five hundred providers in one metro plus zero in sixty percent of the state's land area looks the same as two hundred fifty distributed evenly.
>
> Third — geographic confounding. The rural-state suicide cluster — Montana, Alaska, Wyoming, New Mexico — has idiosyncratic supply patterns and idiosyncratic suicide patterns that operate on different axes from the framework's supply-vs-uninsured residual.
---
## Panel 6 — What the Framework Is Actually For (440–490s, ~65 words)
> The framework's residual classification is a workforce build-out priority signal. It is not a suicide-prevention triage signal. Those are different lists. The published Phase B paper makes that distinction explicit.
>
> The framework's value is preserved. Its outputs identify which states have under-built capacity given their insurance landscape — exactly what the Youth V1 paper said it does. What is qualified is the interpretation.
>
> Trellison Institute. The map is online.
---
## Stats
- **Word count**: ~640 words
- **Estimated duration at 78 wpm**: ~493 seconds
- **Panels**: 6 × ~80s each
- **Companion length**: similar to Youth V1 narration (~660 words)
## Citations index
| Claim | Source |
|---|---|
| within-YRBSS r=+0.82 | analysis_outputs.mh_gap_youth_outcomes_v1 Pearson |
| Across-instrument r=+0.34 with AADR | same |
| access × AADR r=+0.14, drug OD r=+0.21 | same |
| Alaska 1,085 per 100K | analysis_outputs.mh_gap_youth_v1_state_v1 |
| Alaska AADR 27.0, YRBSS attempted 19.0% | NCHS bi63-dtpu 2017 + YRBSS 2023 |
| NJ AADR 8.3, attempted 5.2% | same |
## TTS production notes
- Voice: warm, documentary, matching the Youth V1 narration's pacing
- Music underscore: instrumental at 12.5%
- Tone: clarifying, not defensive — Phase B is a passing test of the framework's interpretability boundary, not a failure
- Visual cues: scatter plot animations (need × YRBSS attempted strong; access × AADR null), state-level chloropleth of outlier classification overlaid with mortality rate
## Versioning
- v1.0 (current): foundation narration. ~640 words / ~490s.