# Press Q&A — Mental Health Access Gap V1

**Companion to**: press release · working paper
**Audience**: journalists, policy researchers, advocates
**Format**: 12 anticipated questions with sourced answers

---

### Q1. Is this saying the mental health access crisis is overblown?

No. The crisis is real and the data confirms it: 41 million American adults experience frequent mental distress (≥14 days of poor mental health in a 30-day period) and the population-weighted prevalence is 16.80%. The study's contribution is that "the access crisis" is *two distinct problems* — geographic access (6.3M Americans) and capacity access (238M Americans) — with completely different policy answers. Calling it one problem and applying one policy solution to both is what's stalled the conversation.

### Q2. How can 91.3% of Americans be within 30 minutes of a provider and still have an access crisis?

Because *physical proximity to a licensed provider is not the same as actually getting an appointment*. The data flags this directly: in Cabell County, West Virginia (Marshall University), the nearest psychiatrist is 0.7 minutes away — and the frequent-distress prevalence is 36.9%, more than double the national average. Distance is not the constraint; appointment availability is. The eight college towns we identify show the same pattern. For these areas, the policy answer is workforce capacity (BHWET) and payment-model reform (CCBHC) — not telehealth.

### Q3. Aren't drive-time estimates a poor proxy for actual road travel?

We use a haversine-distance × 1.4 road-network multiplier and an urban/rural speed model (35 mph / 55 mph) drawn from NCHS Urban-Rural Classification. We validated the proxy against OSRM (Open Source Routing Machine) on a representative sample of the dartboard tracts and found agreement within ±15% nationally. The methodology supplement lists the sensitivity analyses we ran at ±10% speed assumptions. The headline two-problem finding is robust across all of them.

### Q4. Where does the provider count come from?

The CMS National Provider Identifier (NPPES) registry. We query five mental-health taxonomies: psychiatry, psychology, clinical social work, marriage and family therapy, and psychiatric nurse practitioners. The pull yielded 102,036 providers across 8,609 unique practice ZIP codes (May 2026 snapshot). The full taxonomy list is in the methodology supplement §3.

### Q5. NPPES doesn't tell you who's accepting new patients. How do you handle that?

Honestly: we don't. This is the central limitation we flag. NPPES is a supply registry, not a capacity registry. It doesn't see provider hours, network status, accepting-new-patients status, wait times, or quality. The residual analysis is what compensates for this — when a tract's gap is far worse than its insurance coverage predicts, the framework flags it as an unexplained shortfall whose most likely cause is the unobservable capacity constraint. The college-town finding is exactly this signal.

### Q6. Why use insurance coverage (uninsured rate) as the single covariate? Aren't there other determinants?

Yes, many. We chose uninsured rate (PLACES `access2_crudeprev`) as the single covariate for v1 because (a) it is also a peer-reviewed PLACES small-area estimate at the same geography as the prevalence, (b) it is the most policy-actionable proxy (Medicaid expansion + parity enforcement), and (c) starting single-covariate makes the residual analysis interpretable. Multivariate extensions are planned for v1.1 (uninsured + income). The framework is parameterized for either approach.

### Q7. The college-town finding is striking. Was it pre-specified or post-hoc?

The framework was pre-specified before any tracts were examined. The dartboard (case-study sample) is population-weighted random sampling stratified by residual class with a fixed RNG seed — there is no cherry-picking. The eight college towns in the published outlier table are the top-z-score positive outliers (with population ≥ 1,000 to exclude small-N noise) — they sit at the top of the regression's outlier list, not selected by us. The fact that they all share the college-town structural pattern *is the finding*.

### Q8. What about Medicaid expansion's effect — is the negative-outlier finding causal?

No. The residual analysis identifies places where the gap is *better than the SES proxy predicts*. The fact that they cluster in Medicaid-expansion states with CCBHC payment models is a *correlational signal*, not a causal claim. We label them "replication candidates" because the structural recipe (expansion + CCBHC + parity enforcement) is testable elsewhere. The methodology supplement explicitly disclaims causality.

### Q9. How did Hawaii and Alaska factor in?

Both are included. Alaska contributes 175 tracts (mostly large, sparsely populated census areas); Hawaii contributes 426 (denser, urban). Alaska shows up in the geographic-desert table (Kusilvak Census Area, North Slope — drive times over four hours). Hawaii shows up across all three residual classes, with no single dominant pattern. The animated visualizations include both via inset axes; the dataset CSV includes them as standard rows.

### Q10. What is "Trellison Institute" and why should anyone trust this analysis?

Trellison Institute is the methodology-audit arm of the DaedArch platform. We don't influence research; we are the audience. We rate methodology, not conclusions. Our credibility rests on the methodology being open, the data sources being authoritative (CDC, CMS, Census), the framework being open-source (CC-BY-4.0), and the entire analysis being reproducible from public data. The methodology rating itself is pending — once Trellison rates its own first audited output, the rating goes on the page. LedgerWell will then issue the evidence-chain certificate that the audit was followed faithfully. Both gates are public.

### Q11. Can other researchers run this analysis on a different access domain?

Yes. The framework is the product. It is registered as a reusable DB-native tool (`atlas.need_vs_access_framework_v1`) and the methodology supplement gives the input contract: bind a need metric, an access metric, a population metric, and a covariate metric, and the framework produces the same analytical outputs. We have eleven other domains queued internally (poverty, ELL, jobs, postsecondary, library access, police per capita, maternal care, dental care, broadband, oncology, crisis response).

### Q12. Where is the dataset and how do I cite the work?

Dataset (78,815 tracts × 27 fields, CSV + sha256 manifest):
**https://api.daedarch.ai/api/v4/t/media/serve?bucket=daedarch-public-media&object=mental_health_gap/dataset_v1/mh_gap_tract_v1.csv**

Working paper and content arsenal hub:
**https://trellison.com/research/mental-health-supply-demand-gap**

Recommended citation (DOI pending Zenodo registration):
> Trellison Institute. *The Mental Health Access Gap in the United States Divides into Two Distinct Problems*. Working paper, v1.0 (May 2026). DOI: [pending]. https://trellison.com/research/mental-health-supply-demand-gap

The Need-vs-Access Framework code is documented in the methodology supplement and registered as `atlas.need_vs_access_framework_v1` v1.0.0 in the DaedArch tool registry.