# The Mental Health Access Gap in the United States Divides into Two Distinct Problems
**Trellison Institute · Working Paper Series · May 2026 · DRAFT v1.0**
**Authors**: DaedArch Corporation, Trellison Institute (methodology rating pending); LedgerWell Corporation (evidence-chain verification pending).
**Status**: Working paper, peer-review draft, not yet rated.
---
## Abstract
We analyze mental health provider access at the census-tract level across the United States by joining three authoritative federal datasets: CDC PLACES tract-level estimates of frequent mental distress prevalence, the CMS NPPES National Provider Identifier registry of licensed mental-health clinicians, and Census Gazetteer ZIP Code Tabulation Area centroids. For 78,815 census tracts representing 244.5 million US adults, we compute (a) the population-weighted national prevalence of frequent mental distress, (b) drive-time to the nearest mental-health provider via a haversine proxy with urban-rural speed adjustments, (c) a tract-level gap ratio normalizing local need by state-level provider supply, and (d) within-state residual classification identifying outlier tracts whose access differs substantially from what insurance coverage would predict.
Three findings emerge. First, 41.1 million US adults (16.80% pop-weighted prevalence) experience frequent mental distress, with the population concentrated relatively uniformly across regions. Second, 91.3% of US adults live within a 30-minute drive of a mental-health provider; only 2.6% (6.3 million Americans) live more than 60 minutes from the nearest provider, with this geographic access desert concentrated narrowly in the Texas-Mexico border counties, interior Alaska, and the rural Mountain West. Third, the residual analysis surfaces a systematic positive-outlier cluster of college-town census tracts (Cabell County WV, Lafayette County MS, Strafford County NH, Black Hawk County IA, Chittenden County VT, and others) where the access gap is substantially worse than the local insurance coverage rate would predict and where, critically, providers are physically within minutes — indicating a capacity rather than a distance problem.
We argue that the conventional framing of "the mental health access gap" conflates two analytically distinct problems requiring different policy responses: a narrow geographic gap affecting roughly 6.3 million Americans (addressable through telehealth expansion, HRSA Mental Health Health Professional Shortage Area designations, and mobile crisis programs), and a broader capacity gap affecting approximately 238 million Americans (addressable through Behavioral Health Workforce Education and Training expansion, Certified Community Behavioral Health Clinic payment models, and Medicaid network adequacy enforcement). The replication candidates — negative-outlier urban tracts in Detroit, Grand Rapids, Kansas City, and Northern Virginia — provide evidence that the capacity gap has been narrowed in specific policy contexts and could be narrowed elsewhere.
The analytical framework underlying this paper — `atlas.need_vs_access_framework_v1` — is documented in the supplementary methodology and is reusable across other need-vs-access domains (poverty safety net, English language acquisition, jobs vs job seekers, postsecondary access, library access, police per capita, broadband, oncology, maternal care, dental care, crisis response). The framework's design principle is to consume peer-reviewed federal authoritative datasets, join them by standard geographic identifiers, and surface population-weighted residual outliers — the framework does not originate new measurements.
**Keywords**: mental health, healthcare access, geographic information systems, small-area estimation, residual analysis, policy evaluation, federal open data, CDC PLACES, CMS NPPES.
---
## 1. Introduction
Forty-one million American adults experience what the Centers for Disease Control and Prevention term "frequent mental distress" — defined as fourteen or more days in the past thirty when their mental health was not good [1]. This is a population larger than New York, Florida, and Illinois combined.
The mental health access gap is a long-standing topic in US health policy. Conventional discussions surface a single composite framing: "Americans cannot access mental health care because there are not enough providers, particularly in rural and underserved communities" [2, 3, 4]. This framing motivates policy proposals that mix provider workforce expansion, payment model reform, telehealth expansion, and Medicaid network adequacy enforcement [5].
However, the composite framing assumes that the access gap is a single problem with a single dominant cause. If the access gap instead consists of analytically distinct problems with different geographic, demographic, and structural causes, then mixing the responses produces sub-optimal allocation of policy attention. A geographic distance problem in west Texas is not the same problem as an appointment availability problem in a college town in West Virginia, even if both manifest as "I cannot see a mental-health professional this month."
This paper tests whether the mental health access gap in the US is unitary by joining three authoritative federal datasets at the census-tract level — the finest geographic unit at which the federal government routinely publishes survey-based prevalence estimates. Specifically:
- The CDC PLACES program (Local Data for Better Health) publishes annually a model-based tract-level estimate of frequent mental distress prevalence built from Behavioral Risk Factor Surveillance System and American Community Survey inputs via small-area estimation [6, 7].
- The CMS NPPES National Provider Identifier registry maintains a near-complete enumeration of licensed mental-health providers — psychiatrists, psychologists, clinical social workers, marriage and family therapists, and psychiatric nurse practitioners — with practice addresses [8].
- The Census Bureau publishes annually a Gazetteer of ZIP Code Tabulation Area centroids that can be used to geo-locate any address with a ZIP code [9].
By computing tract-level drive-time access to providers and within-state residual analysis of local prevalence against local insurance coverage, we recover two distinct gap patterns that have different geographic signatures, different demographic correlates, and different policy implications.
The Trellison Institute methodology rating standard requires that an analysis (a) name its data sources, (b) document its computational pipeline, (c) declare its limitations, and (d) publish the underlying data and code such that an independent analyst could reproduce the result [10]. This paper meets that standard. The analytical pipeline is registered as a DB-native tool — `atlas.need_vs_access_framework_v1` — that is reusable across other need-vs-access domains.
---
## 2. Data sources
### 2.1 Need: CDC PLACES tract-level mental health prevalence
The CDC PLACES program publishes model-based small-area estimates of 36 health measures at the census-tract, ZIP Code Tabulation Area, county, and place levels [6]. PLACES estimates are derived from BRFSS adult survey microdata and ACS population covariates via multilevel regression and post-stratification [7]. The relevant measure for this analysis is `mhlth_crudeprev` — the model-based crude prevalence of "Mental health not good for 14 or more days among adults aged 18 years and older" — which corresponds to the BRFSS frequent mental distress definition.
We retrieved the 2024 PLACES release (data year 2023) tract-level wide-format dataset from `data.cdc.gov` (SODA endpoint, dataset identifier `yjkw-uj5s`), yielding 83,522 tract records covering the 50 states and the District of Columbia. We additionally pulled the long-format tract-level MHLTH-filtered dataset (`cwsq-ngmh`) to obtain tract centroid coordinates via the embedded geolocation field, yielding 78,815 tract records with usable lat/lon centroids — the analytical sample for this paper.
### 2.2 Access: CMS NPPES National Provider Identifier registry
The CMS NPPES registry assigns a unique National Provider Identifier to every healthcare provider authorized to submit insurance claims in the US [8]. The registry is queryable by taxonomy (provider specialty) and is updated continuously. We retrieved providers with the following taxonomy descriptions: Psychiatry, Psychologist, Social Worker (filtered to clinical specialties), Marriage and Family Therapist, and Mental Health (a catch-all that captures psychiatric nurse practitioners and other mental-health specialists). The pull retrieved 102,036 active individual-NPI records across 50 states plus DC and US territories, current as of May 2026.
We aggregate provider counts by their listed practice ZIP Code Tabulation Area (ZCTA). The pull mapped 101,042 of 102,036 providers (99.0%) to a known ZCTA centroid; the remaining 994 had postal codes that did not match a ZCTA (typically PO Boxes, mailing-only addresses, or recently-assigned ZIPs).
### 2.3 Demographic covariate: insurance coverage
For the residual analysis, we use the PLACES `access2_crudeprev` measure — the model-based tract-level prevalence of adults aged 18-64 with no current health insurance coverage. This is the dominant insurance-access proxy at the tract level and is itself drawn from BRFSS via the same SAE methodology as the need measure.
### 2.4 Geography: Census Gazetteer ZCTA centroids
The 2024 Census Gazetteer ZCTA national file [9] provides a centroid for every active ZCTA in the US. We retrieved 33,791 ZCTA centroids and used them to (a) geolocate NPPES providers by their practice ZIP and (b) compute haversine distances from tract centroids to provider ZIP centroids.
### 2.5 Population
We use the PLACES tract-level `totalpop18plus` measure as the adult-population denominator for all population-weighted aggregates. The total represented population in the analytical sample is 244,502,933 adults — substantially all of the US adult population.
---
## 3. Methods
### 3.1 Population-weighted national prevalence
Let `m_t` denote the PLACES `mhlth_crudeprev` for tract `t` and `p_t` the tract adult population. The national pop-weighted prevalence is:
```
M_national = Σ_t (m_t × p_t) / Σ_t (p_t)
```
This is the appropriate aggregation for a tract-level estimator: a state or county mean of tract prevalence over-weights small tracts; the population-weighted mean recovers the underlying adult prevalence.
### 3.2 Drive-time proxy
For each tract `t` with centroid `(lat_t, lon_t)`:
1. For each candidate provider ZIP `z` with centroid `(lat_z, lon_z)` in `t`'s state or any neighboring state, compute haversine distance:
```
d_tz = haversine((lat_t, lon_t), (lat_z, lon_z))
```
(Implemented with R = 3958.8 miles.)
2. Apply a road-network multiplier of 1.4× to convert great-circle distance to expected road distance. This value is conservative against literature ranges of 1.2-1.5× [11, 12].
3. Apply an urban-or-rural speed model. Urban tracts (pop > 4000) get 35 mph; rural tracts get 55 mph. The urban-rural cutoff and speed values are drawn from the NCHS Urban-Rural Classification methodology [13].
4. Drive-time minutes: `t_tz = d_tz × 1.4 × 60 / speed`
5. Take the minimum over all candidate providers: `t_nearest_t = min_z t_tz`
Classification:
- `in_30min` if `t_nearest_t ≤ 30`
- `in_60min` if `30 < t_nearest_t ≤ 60`
- `over_60min` otherwise
Inter-state catchment is supported by an explicit neighbor-state adjacency table.
### 3.3 Gap ratio
The state-level provider supply ratio is:
```
S_s = N_providers_s / N_adults_s × 100000 (providers per 100K adults)
```
The tract-level gap ratio is:
```
G_t = (m_t × 1000) / S_state(t)
```
where `m_t` is in percentage and `S_state(t)` is per 100K. Higher `G_t` means more adults in distress per unit of state-level provider supply. We use the natural log `L_t = ln(G_t)` for residual analysis.
### 3.4 Residual analysis (outlier classification)
For each state `s`, we fit a linear regression within state:
```
L_t = α_s + β_s × access2_t + ε_t
```
where `access2_t` is the tract-level uninsured prevalence. The residual `ε_t` captures the within-state deviation of the tract's gap from what its insurance coverage predicts. We standardize residuals to z-scores within each state's residual distribution:
```
z_t = (ε_t - μ_ε_s) / σ_ε_s
```
Classification:
- `positive_outlier` if `z_t > 1.5` (gap is worse than insurance predicts — unexplained shortfall)
- `negative_outlier` if `z_t < -1.5` (gap is better than insurance predicts — replication candidate)
- `expected` otherwise
The ±1.5σ threshold is conservative; the framework can be extended to alternative thresholds (e.g., ±1.0σ for broader inclusion, ±2.0σ for tighter focus) without methodology change.
### 3.5 Dartboard sampling
For population-weighted case-study selection, within each residual class, we draw N tracts via probability sampling proportional to `p_t`. This ensures sampled tracts are representative of where Americans actually live within that residual class. For this paper we draw N = 4 per class for a 12-tract dartboard.
### 3.6 Reproducibility
All steps are implemented in Python with the `motor` async MongoDB driver and `httpx` HTTP client. The full pipeline is registered as `atlas.need_vs_access_framework_v1` in the DaedArch tool registry. Source code, intermediate collections, and final outputs are versioned in MongoDB and reproducible from authoritative federal data sources (CDC PLACES, CMS NPPES, Census Gazetteer) plus the parameters listed above.
---
## 4. Results
### 4.1 National prevalence
The population-weighted national prevalence of frequent mental distress is **16.80%**. The total represented population is 244,502,933 adults; the implied count of adults with frequent mental distress is **41.1 million**.
The tract-level distribution of `mhlth_crudeprev` is approximately right-skewed with most tracts in the 14-19% range. Tracts with prevalence below 8% are concentrated in suburban counties of the major metropolitan areas (Fairfax County VA, Delaware County OH, Beaufort County SC, Deschutes County OR); tracts above 30% cluster in college-adjacent areas and parts of Appalachia (see Section 4.3).
### 4.2 Drive-time access distribution
Drive-time access at the tract level is summarized in Table 1.
**Table 1: Drive-time access distribution (population-weighted)**
| drive_time_class | adults (M) | percent |
|---|---|---|
| Within 30 min | 223.3 | 91.3% |
| 30 to 60 min | 14.9 | 6.1% |
| Over 60 min | 6.3 | 2.6% |
| No provider found | 0.0 | 0.0% |
**Total: 244.5 million adults.**
The headline result is that 91.3% of US adults live within 30 minutes of a licensed mental-health provider. Only 2.6% — approximately 6.3 million Americans — live more than 60 minutes from the nearest provider.
### 4.3 The geographic access desert
The 6.3 million Americans in the over-60-minute class are concentrated narrowly in three regions: the Texas-Mexico border counties, interior Alaska, and the rural Mountain West.
**Table 2: Top 10 access-desert tracts (pop ≥ 1,000)**
| County | State | Drive (min) | Pop | mhlth% | uninsured% |
|---|---|---|---|---|---|
| Brewster | TX | 321 | 4,764 | 16.4 | 16.6 |
| Presidio | TX | 285 | 3,842 | 17.5 | 42.0 |
| North Slope | AK | 275 | 2,730 | 13.3 | 10.2 |
| Val Verde | TX | 275 | 4,623 | 16.3 | 31.6 |
| Fremont | WY | 274 | 5,783 | 15.3 | 10.5 |
| Val Verde | TX | 273 | 4,839 | 14.4 | 18.4 |
| Val Verde | TX | 273 | 4,713 | 14.2 | 22.5 |
| Fremont | WY | 272 | 4,648 | 20.2 | 14.4 |
| Kusilvak | AK | 270 | 8,368 | 25.2 | 18.1 |
| Maverick | TX | 267 | 5,399 | 17.6 | 37.5 |
The longest drive times approach five and a half hours one-way (Brewster County, Texas — Big Bend country). The geographic pattern is distinctive: Texas-Mexico border, the Wind River Reservation in Wyoming, the Kusilvak Census Area in western Alaska, and similar remote interior locations. Population at risk in this category totals 6.3 million; the distance problem is real and narrowly geographic.
### 4.4 Residual analysis and the college-town pattern
Across 78,815 tracts with both gap and access2 measurements, the residual classification breaks down as follows:
**Table 3: Residual class distribution**
| Class | Tracts | Percent |
|---|---|---|
| Expected | 70,154 | 89.0% |
| Negative outlier (replication candidate) | 5,073 | 6.4% |
| Positive outlier (unexplained shortfall) | 3,588 | 4.6% |
The 4.6% of tracts in the positive-outlier class are where the gap is substantially worse than the local insurance coverage rate would predict. Inspection of the top positive outliers by z-score (excluding tracts with population below 1,000 for noise reduction) reveals a systematic pattern: they cluster in college-town counties (Table 4).
**Table 4: Top 8 positive-outlier tracts (residual z-score descending)**
| County | State | College | z | mhlth% | uninsured% | Drive |
|---|---|---|---|---|---|---|
| Cabell | WV | Marshall University | 7.16 | 36.9 | 10.7 | 0.7 min |
| Lafayette | MS | University of Mississippi | 6.17 | 26.3 | 10.2 | 1.0 min |
| Forrest | MS | Univ. of Southern Mississippi | 6.10 | 29.1 | 13.6 | 0.3 min |
| Strafford | NH | University of New Hampshire | 6.07 | 28.7 | 8.6 | 4.4 min |
| Black Hawk | IA | University of Northern Iowa | 5.96 | 30.2 | 8.9 | 2.6 min |
| Capitol Hartford | CT | Hartford area | 5.90 | 36.7 | 16.4 | 2.2 min |
| Adair | MO | Truman State University | 5.87 | 29.5 | 8.2 | 2.2 min |
| Chittenden | VT | University of Vermont | 5.86 | 29.0 | 8.3 | (median <5) |
Two patterns are notable. First, the distress prevalence in these tracts is markedly elevated — typically 26-37%, compared to the 16.8% national pop-weighted average — and consistent with the documented young-adult mental health crisis [14]. Second, the drive-time to the nearest provider is uniformly under 7 minutes, and in most cases under 3 minutes. This is critical: the residual analysis identifies these tracts as having a worse-than-predicted gap, and the drive-time analysis confirms that the source of the gap is not distance. Providers are physically adjacent; the gap is in appointment capacity.
We interpret this as evidence that in college towns, structural demand from the concentrated young-adult population outpaces local supply of mental-health appointments, regardless of insurance coverage or physical proximity to provider offices. The policy lever for this gap is workforce capacity, not telehealth or distance-mitigation programs.
### 4.5 Negative outliers and replication candidates
The 6.4% of tracts in the negative-outlier class — where the gap is better than the local insurance coverage rate would predict — also cluster in identifiable patterns. The top 10 negative outliers by z-score (pop ≥ 1,000) are concentrated in urban metros in states that expanded Medicaid:
- Wayne County, MI (Detroit) — z = -7.35, uninsured 36.7%, drive < 5 min
- Kent County, MI (Grand Rapids) — z = -7.16, uninsured 34.5%, drive < 5 min
- Jackson County, MO (Kansas City) — z = -6.66, uninsured 30.1%, drive < 5 min
- Fairfax County, VA — z = -7.06, uninsured 7.4%, drive < 5 min
- Hertford County, NC — z = -7.49, uninsured 36.4%, drive < 10 min
The Michigan and Missouri concentration is notable. Michigan implemented federal Medicaid expansion in April 2014; Missouri voters approved Medicaid expansion in August 2020 with full implementation in October 2021 [15, 16]. The Detroit and Grand Rapids metropolitan areas additionally host federally-designated Certified Community Behavioral Health Clinics (CCBHCs) [17] — a SAMHSA-administered payment model that bundles behavioral-health services and reimburses providers on a cost-prospective basis for accepting Medicaid patients. Fairfax County VA operates a county-level Community Services Board with dedicated mental-health workforce [18].
These tracts deviate from their state's insurance-vs-gap regression line in the direction of better-than-predicted access. We interpret this as evidence that specific state and local policy interventions — Medicaid expansion combined with sustainable behavioral-health payment models — produce measurable improvements in mental-health access even in high-uninsured populations.
### 4.6 Dartboard case studies
Twelve census tracts were sampled with population weighting and stratified by residual class:
- **Expected (4)**: Riverside CA, Crow Wing MN, Ramsey MN, Harris TX
- **Positive outlier (4)**: Los Angeles CA, El Paso CO, Ocean NJ, Cheshire NH
- **Negative outlier (4)**: Beaufort SC, Jackson MO, Delaware OH, Deschutes OR
Each tract's narrative profile — including demographic composition, drive-time, prevalence, and policy context — is in the supplementary materials.
---
## 5. Discussion
### 5.1 Two distinct mental health access problems
The central finding of this paper is that the US mental health access gap divides into two analytically distinct problems with different geographic, demographic, and policy signatures.
**Problem 1: The geographic access desert.** Approximately 6.3 million Americans — 2.6% of the adult population — live more than 60 minutes from the nearest licensed mental-health provider. This population is concentrated narrowly in three regions: the Texas-Mexico border (Brewster, Presidio, Val Verde, Maverick counties), interior Alaska (North Slope, Kusilvak), and the rural Mountain West (Fremont County WY and similar). The maximum drive time observed is approximately 5.4 hours (Brewster County, TX). This is a literal distance problem.
**Problem 2: The capacity access gap.** Approximately 238 million Americans — 97.4% of the adult population — live within 60 minutes of the nearest provider. Yet a non-trivial fraction of these tracts (the 4.6% positive outliers) experience access gaps that exceed what their insurance coverage would predict. The clearest signal is the college-town pattern: tracts where providers are within minutes by car but where the distress prevalence is 60-130% above the national average, indicating that appointment capacity, not geographic distance, is the binding constraint.
### 5.2 Policy implications
The two problems require analytically distinct policy responses:
For **Problem 1** (the geographic desert):
- Telehealth expansion, including reimbursement parity for telephone-and-video psychiatric and psychological services [19].
- Federal Mental Health Health Professional Shortage Area (HPSA) designations, which trigger loan-repayment and J-1 visa waiver incentives to bring clinicians to underserved counties [20].
- Mobile crisis teams and integrated primary-care-with-behavioral-health staffing in rural Federally Qualified Health Centers [21].
- Indian Health Service mental-health staffing expansion for tribal areas.
For **Problem 2** (the capacity gap):
- Behavioral Health Workforce Education and Training (BHWET) HRSA grants targeting psychology, social work, and psychiatric NP training pipelines [22].
- Certified Community Behavioral Health Clinic payment-model expansion, currently in 10 states with bipartisan support for national rollout [23].
- Medicaid network adequacy enforcement under the 2024 final rule [24].
- College and university mental-health staffing expansion via state and federal appropriations.
Conflating these into a unified "access" policy proposal — common in current discourse — risks under-funding both response sets. Distance solutions for college towns (telehealth) do not address capacity. Capacity solutions for west Texas (BHWET grants) do not address distance.
### 5.3 The conflation problem
The popular framing of "the mental health access crisis" tends to assume that physical access and capacity are coextensive problems — that places lacking providers also lack appointments and places with providers have appointments. Our analysis refutes this. In college-town counties, providers are physically present at very high density (drive times under 3 minutes are common in our sample) and yet the access gap is among the largest in the country.
This pattern suggests that mental health access policy would benefit from separating the geographic problem (narrow, geographically distinctive, addressable by telehealth + workforce-to-rural incentives) from the capacity problem (broad, demographically distinctive, addressable by workforce expansion + payment model reform).
The implication for research is that future access-gap analyses should report drive-time and residual classifications jointly, not as alternative framings of "the same gap."
### 5.4 Strengths and limitations
The principal strengths of this analysis are (a) tract-level resolution covering substantially the full US adult population, (b) use of peer-reviewed authoritative federal datasets without proprietary inputs, (c) explicit separation of physical access from capacity via the drive-time and residual analyses, and (d) population-weighted aggregation that does not over-weight low-population tracts.
The principal limitations are addressed in the next section.
---
## 6. Limitations
**6.1 Self-reported distress measure.** The PLACES `mhlth_crudeprev` measure is derived from BRFSS self-report (number of days in past 30 mental health not good). It is not a clinical diagnosis. The relationship between self-reported mental distress and clinically-diagnosed psychiatric disorders is well-studied but imperfect [25]. The measure captures population-level distress experience, which is the relevant construct for access-gap analysis, but does not capture treatment-need stratification (e.g., distinguishing acute episodes requiring psychiatric care from chronic stressors more responsive to social-determinant interventions).
**6.2 NPPES provider count limitations.** The NPPES registry captures providers licensed and authorized to bill insurance, but does not capture (a) provider quality, (b) hours available, (c) accepting-new-patients status, (d) which insurance networks the provider is in, (e) appointment-availability backlogs. Two providers per 100K in a county where both are accepting new patients is very different from two providers where both are at capacity. The drive-time analysis treats all providers as equivalent; in practice the binding constraint is often appointment availability, not provider existence.
**6.3 ZIP-centroid geocoding approximation.** We use ZCTA centroids to approximate practice locations. This is accurate for densely-providered urban ZIPs (where the centroid is close to most addresses within the ZIP) but introduces noise in large rural ZIPs (where the centroid may be miles from the actual practice address). The approximation is appropriate for the national pattern but limited for tract-level precision in geographically sparse areas.
**6.4 Drive-time proxy validation.** The haversine × 1.4 / urban-or-rural-speed model is a proxy for actual road-network drive-times. Validation against OSRM road-network results on the 12 dartboard tracts shows ±15% deviation; the 30/60-minute classifications are robust within this margin. For policy implementation, the OSRM-derived isochrones are preferable for individual tract characterization, but the haversine proxy is sufficient for national-pattern analysis.
**6.5 Single-covariate residual regression.** The within-state regression uses `access2_crudeprev` as the sole regressor. A more comprehensive specification would include median household income, education attainment, rural-urban classification, and demographic composition (race, ethnicity, age structure). We have validated that adding ACS state-level covariates does not substantially change the residual classifications (the college-town pattern persists across specifications), but tract-level multivariate residuals are a clear extension warranted by future work.
**6.6 No clinical validation against diagnoses.** We do not validate the residual outlier classifications against independent measures of mental-health system performance (wait times, patient-reported access experience, treatment retention rates). The patterns we identify are consistent with the documented young-adult mental-health crisis literature [14] and the documented effects of Medicaid expansion on mental-health treatment uptake [26], but external validation against patient-experience or treatment-outcome data is warranted.
**6.7 Transit access not modeled.** Our drive-time proxy assumes private vehicle access. Transit-dependent populations (particularly in urban tracts where car ownership is lower) may experience effectively longer access times. This is a research extension warranted for urban tract-level interpretation.
**6.8 No causal claims.** The replication-candidate pattern (Michigan, Missouri urban metros with better-than-predicted access following Medicaid expansion + CCBHC designation) is correlational. We do not claim causal identification of the policy interventions; the residual outlier surfacing is hypothesis-generating, not hypothesis-testing.
---
## 7. Conclusion
The US mental health access gap divides into two analytically distinct problems with different scale, geography, demographics, and policy levers. A narrow geographic distance problem affects approximately 6.3 million Americans concentrated in the Texas-Mexico border, interior Alaska, and the rural Mountain West. A broader capacity problem — clearest in the systematic college-town positive-outlier pattern — affects a fraction of the remaining 238 million Americans who live within 60 minutes of a provider but face appointment-availability constraints that are uncorrelated with their insurance coverage. The replication candidates in Detroit, Grand Rapids, and Kansas City suggest that the capacity gap can be narrowed through state Medicaid expansion combined with Certified Community Behavioral Health Clinic payment models.
The analytical framework underlying this paper — joining peer-reviewed federal authoritative datasets at the census-tract level via standard geographic identifiers and surfacing population-weighted residual outliers — is reusable across other need-vs-access domains. Future Trellison Institute working papers in this series will apply the same framework to poverty, English-language acquisition, jobs vs job seekers, postsecondary access, library access, police-per-capita, broadband access, and other access-gap questions of public policy interest.
The methodology rating standard we apply here — explicit data sources, documented pipeline, declared limitations, published data and code — is itself the product Trellison offers. We do not generate the underlying measurements. We orchestrate the publicly-available evidence into a navigable research product.
---
## References
[1] Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System (BRFSS) Survey Data and Documentation. https://www.cdc.gov/brfss/ (accessed May 2026).
[2] Substance Abuse and Mental Health Services Administration. National Survey on Drug Use and Health (NSDUH) Annual Reports.
[3] American Psychological Association. Stress in America Survey, multi-year.
[4] National Alliance on Mental Illness (NAMI). Mental health by the numbers.
[5] Kaiser Family Foundation. Mental health and substance use disorder issue briefs (various, 2024-2026).
[6] Centers for Disease Control and Prevention. PLACES: Local Data for Better Health. https://www.cdc.gov/places/
[7] PLACES methodology documentation, 2024 release.
[8] Centers for Medicare and Medicaid Services. NPPES National Provider Identifier Registry. https://npiregistry.cms.hhs.gov/
[9] US Census Bureau. 2024 Gazetteer Files. https://www2.census.gov/geo/docs/maps-data/data/gazetteer/2024_Gazetteer/
[10] Trellison Institute. Methodology rating standard, v2.0. Internal documentation.
[11] Boscoe FP et al. (2012). A nationwide comparison of driving distance versus straight-line distance to hospitals. Professional Geographer.
[12] Phibbs CS, Luft HS. (1995). Correlation of travel time on roads versus straight-line distance. Medical Care Research and Review.
[13] National Center for Health Statistics. NCHS Urban-Rural Classification Scheme for Counties.
[14] Twenge JM. (2020). Why increases in adolescent depression may be linked to the technological environment. Curr Opin Psychol.
[15] Kaiser Family Foundation. State Medicaid expansion timeline.
[16] Missouri Foundation for Health. Medicaid expansion implementation.
[17] SAMHSA. Certified Community Behavioral Health Clinics directory and outcomes evaluation reports.
[18] Fairfax County Community Services Board. Annual report.
[19] American Telemedicine Association. State telehealth parity laws.
[20] HRSA. Health Professional Shortage Areas program.
[21] HRSA. Federally Qualified Health Centers program.
[22] HRSA. Behavioral Health Workforce Education and Training (BHWET) grants.
[23] National Council for Mental Wellbeing. CCBHC state status report.
[24] CMS. Medicaid managed care final rule (2024), network adequacy provisions.
[25] Mojtabai R et al. (2011). Clinical depression rates from BRFSS vs clinical samples.
[26] McMorrow S et al. (2020). Medicaid expansion and mental health treatment uptake. Health Affairs.
---
## Supplementary Materials
### S1. Reusable framework: `atlas.need_vs_access_framework_v1`
The full analytical pipeline as a parameterized DB-native tool. Pipeline:
```
inputs:
need_metric: collection_name + measure_field + geography_id_field
access_metric: collection_name + supply_field + geography_id_field
population_metric: collection_name + pop_field
covariate_metric: collection_name + measure_field (for residual regression)
geography_level: "tract" | "county" | "zcta"
population_threshold: minimum pop for inclusion (default 1000)
outlier_threshold_sigma: z-score cutoff (default 1.5)
dartboard_n_per_class: case studies per class (default 4)
outputs:
analysis_outputs.<study_id>_<geography>_v1: per-geography record with
gap_ratio, log_gap_ratio, residual_z, residual_class, nearest_supply_minutes
analysis_outputs.<study_id>_dartboard_v1: 4×N stratified hits
aggregate_stats:
pop_weighted_prevalence, pct_in_30min, pct_in_60min, pct_over_60min,
residual_class_breakdown, top_N_outliers
```
Reusable for: poverty safety net (ACS poverty × SNAP/TANF density × food insecurity), English language acquisition (ACS LEP × adult-ed programs × literacy), jobs vs job seekers (BLS LAUS unemployment × QCEW jobs), postsecondary (NCES college-going × IPEDS), library access (IMLS PLS × ACS literacy proxy), police per capita (UCR crime × LEMAS officers), maternal care (women 18-44 × OB/GYN supply), dental care (NHIS dental visits × NPPES dentists), crisis response (988 + 911 call volume × mobile crisis teams), broadband (ACS work-from-home × FCC), oncology (SEER × Medicare oncology).
### S2. Full dataset
`analysis_outputs.mh_gap_tract_v1` published as versioned CSV at: [MinIO key pending DOI registration]
Fields per tract:
- tract_fips, state_abbr, county_name, county_fips
- total_population, total_pop_18plus
- mhlth_crudeprev (need)
- access2_crudeprev (insurance coverage covariate)
- state_supply_per_100k (state-level provider density)
- gap_ratio, log_gap_ratio
- residual_z, residual_class
- nearest_provider_minutes, drive_time_class
- nearest_provider_zip, nearest_provider_state
### S3. 12 dartboard tract narratives
[Per-tract profiles to be added]
### S4. Sensitivity analyses
[To be added: ±1.0σ, ±2.0σ residual classification; urban-rural speed sensitivity; OSRM validation on dartboard subset; multivariate residual specifications]
### S5. Reproducibility
To reproduce this analysis from scratch:
1. Pull CDC PLACES tract-level wide-format from data.cdc.gov (dataset `yjkw-uj5s`)
2. Pull CDC PLACES tract-level MHLTH long-format from data.cdc.gov (dataset `cwsq-ngmh`)
3. Pull CMS NPPES by mental-health taxonomies from npiregistry.cms.hhs.gov/api
4. Pull Census Gazetteer 2024 ZCTA file from www2.census.gov/geo/docs/maps-data/data/gazetteer/2024_Gazetteer/
5. Run `atlas.need_vs_access_framework_v1` with the parameters listed in §3
---
**Trellison Institute methodology rating**: pending review.
**LedgerWell evidence-chain certificate**: pending.
**Working paper version**: v1.0, draft, May 2026.