# Press Q&A — Phase B Youth Mental Health Outcomes Correlation
**Companion to**: press release · working paper
**Audience**: journalists, policy researchers, advocates
**Format**: 12 anticipated questions with sourced answers
---
### Q1. The headline says "supply doesn't predict reduced suicide." Isn't that backwards?
Not backwards — clarifying. The expectation is that states with more mental-health providers per capita would have lower suicide rates. The data does not show that pattern. Both the all-age suicide age-adjusted death rate (NCHS 2017) and the drug overdose rate (2023) are *flat to slightly positive* with respect to state-level provider density. Alaska has the highest youth-serving provider density in our dataset (1,085 per 100K under-18) and the highest all-age suicide AADR (27.0 per 100K). Vermont has high supply and moderate suicide AADR (18.3). New Jersey has low supply and very low suicide AADR (8.3). At state-level cross-section, there is no protective correlation. Why exactly is a separate question — the paper proposes three non-mutually-exclusive explanations (capacity ≠ count, geographic confounding, state granularity is too coarse).
### Q2. Doesn't this mean adding more providers is a waste?
No. The paper does not say supply expansion is ineffective; it says state-level cross-section provider density is not correlated with state-level mortality at this aggregation. Three things this finding does *not* establish:
1. It does not establish that supply expansion at the *local* level (county, ZIP) is ineffective.
2. It does not establish that supply expansion over *time* (longitudinal) is ineffective. A state that increases provider density from 50 to 500 per 100K could see mortality reductions that a cross-state snapshot cannot detect.
3. It does not establish that *no* mental-health intervention works. It says the specific signal "more state-level providers → fewer state-level deaths" is absent in this data.
Supply expansion is necessary but not sufficient. The paper makes that explicit.
### Q3. Why is your residual_z signal uncorrelated with suicide if your framework is supposed to identify access gaps?
Because the framework was designed to identify *workforce-build-out priority states given insurance coverage*, not *suicide-risk priority states*. Those are different questions.
A state can be supply-undersupplied for its uninsured population (the framework's positive_outlier classification) and simultaneously have low overall suicide rates — because the population is small, urban, demographically homogeneous, and the uninsured share is itself low. New Jersey fits this exactly: positive_outlier on supply gap, but the third-lowest all-age suicide AADR (8.3) in the country. The framework's residual measure captures policy-decision-relevant capacity gap. It does not capture mortality risk. We draw this distinction explicitly in §5 of the working paper.
### Q4. Then what is the framework actually useful for?
It identifies which states should build more youth-serving workforce capacity given the insurance landscape they face. Concrete policy actions: BHWET (Behavioral Health Workforce Education and Training) program with youth-specific tracks; CCBHC (Certified Community Behavioral Health Clinic) payment model with explicit youth-serving certification; Title V Maternal and Child Health Block Grant; state-level loan-repayment programs targeting community youth mental-health workforce.
Puerto Rico (workforce density 5.8 per 100K under-18), North Carolina (9.1), and New Jersey (17.9) are the framework's three positive outliers — the workforce-build-out priorities. None of these are also the highest-suicide states; that list is Montana, Alaska, Wyoming, New Mexico, Idaho, and the policy levers for those states are different.
### Q5. Why is the supply-vs-drug-overdose correlation positive? Higher supply, more deaths?
The "deaths of despair" geography (West Virginia, Ohio, New Mexico, parts of Appalachia) is overlapping but not coincident with the under-supply geography. WV has moderately high youth-serving provider density (229 per 100K under-18) and the highest drug overdose rate in our dataset (79.1 per 100K). DC has very low uninsured rate (3.6%) and a drug overdose rate of 95.0 per 100K — among the highest. The positive correlation reflects this overlap, not a causal pathway.
The paper explicitly says drug overdose mortality has a different signature from suicide and requires a separate analytical pipeline (NPPES SUD-treatment taxonomies, SAMHSA OTPs, NSDUH OUD-receiving rates). We report the drug-OD correlation for completeness but do not interpret the supply coefficient on this outcome.
### Q6. The all-age AADR is from 2017. Why aren't you using more recent state-level suicide data?
The most recent state-level all-cause suicide death rate available via Socrata API (data.cdc.gov dataset `bi63-dtpu`) is 2017. CDC publishes more recent data via the WONDER system, but the WONDER public API requires an interactive license-acceptance flow that our automation does not currently handle. We treat the 2017 vintage as a real limitation. State-level suicide AADR is fairly stable year-over-year (the autocorrelation is ~0.95), so the 6-year gap introduces noise but not systematic bias. v1.1 will integrate the WONDER POST API.
### Q7. Why no FBI crime data?
The FBI Crime Data Explorer API requires an `api.data.gov` API key. The key is free and instant from https://api.data.gov/signup/ but it has not yet been provisioned in our credential vault. A HIT is open for issuance. Once the key is in the vault, the framework's crime correlation extension will run in approximately one analysis turn.
### Q8. Why no ER visit data?
The Healthcare Cost and Utilization Project (HCUP) State Emergency Department databases require licensed access via the HCUP Data Distributor. CDC's DOSE Dashboard for drug-overdose ER visits does not currently expose a public API. v2.0 of this study will integrate HCUP licensed data; v1.0 omits it.
### Q9. What's the validity of YRBSS as a "youth-specific outcome"?
YRBSS is the CDC's biennial Youth Risk Behavior Surveillance System, a school-based self-administered survey of high-school students grades 9-12. It is the federal standard for state-level youth health-behavior surveillance and is published with state-by-state response rates. The suicide-related questions (considered, planned, attempted in past 12 months) are validated measures and have been continuously administered since 1991 with reformed wording from 1999 forward.
YRBSS is self-report, school-based (excludes truant + dropout youth), and biennial. The three known limitations: (i) social-desirability response bias, (ii) sampling bias toward enrolled students, (iii) state-level participation is voluntary so coverage is incomplete. Within these limits, YRBSS is the strongest state-level youth suicide-indicator dataset available to public researchers, and our within-instrument correlations (r > +0.75 between sad/hopeless and the suicide questions) are well-calibrated to the underlying state-level patterns.
### Q10. The dartboard says 35 states. Where are the rest?
The Youth V1 analysis required four data sources to join: YRBSS state participation, NPPES youth-serving provider count, ACS under-18 population, and ACS uninsured under-19. The binding constraint is YRBSS — 17 U.S. states did not release state-level data for 2023 (state non-participation or non-releasable response rates). California, New York, Florida, Georgia, and several others are not in the 2023 YRBSS state-level Total-demographic release. They are in the dataset structure but with null `need_value`.
This is the most consequential limitation: our 35-state sample over-represents the smaller-population states (the participating set skews toward states that have continued YRBSS state-level participation) and excludes some of the largest states (CA, NY, FL).
### Q11. Did you adjust for multiple comparisons in your correlation matrix?
We did not. The matrix has 25 cells. Under a strict null and α=0.05, expected false-positive count would be ~1.25. Our headline correlations (r = +0.82, +0.80, +0.76 for the within-YRBSS bloc) have |r| an order of magnitude larger than the false-positive threshold, so multiple-comparison correction does not affect the headline. For mid-magnitude correlations (r ≈ +0.34, +0.21, +0.46), we report them as effect sizes rather than significance tests. The sample size (n = 24-35) is sufficient for descriptive cross-sectional interpretation but not for formal hypothesis testing on individual coefficients.
### Q12. How do I cite this work? Where is the dataset?
Recommended citation (DOI pending Zenodo registration):
> Trellison Institute. *State-Level Youth Mental Health Need-vs-Access Measures Predict Within-Instrument Suicide Ideation but Not Across-Instrument Mortality.* Working paper v1.0 (May 2026). Phase B follow-up to the Youth Mental Health Access Gap V1 study. DOI: [pending]. https://trellison.com/research/youth-mental-health-outcomes-correlation
Dataset (35 states × 11 fields, sha256-stamped):
**https://api.daedarch.ai/api/v4/t/media/serve?bucket=daedarch-public-media&object=youth_mental_health_outcomes/dataset_v1/mh_gap_youth_outcomes_v1_state.csv**
Full content arsenal hub: https://trellison.com/research/youth-mental-health-outcomes-correlation
The framework: `atlas.need_vs_access_framework_v1` v1.1.0 in the DaedArch tool registry. License: CC-BY-4.0.