# 12 Dartboard Tract Case-Study Narratives

**Companion to**: `mh_gap_article_v1_peer_review.md` §S3
**Source**: `analysis_outputs.mh_gap_dartboard_v1`
**Sampling**: population-weighted, stratified by `residual_class` (4 expected + 4 positive_outlier + 4 negative_outlier)
**Each tract is pre-specified — not post-hoc cherry-picked.**

Per-tract structure: tract FIPS · location · headline numbers · what the regression "predicts" · what's actually happening on the ground · policy implication.

---

## EXPECTED — what the regression predicts holds

### 1. Riverside County, CA · Tract 06065030800

- **Stats**: prevalence 16.6% · uninsured 9.9% · drive-time 7.3 min · z=+0.19 · gap ratio 2,754
- **Population**: 5,787 adults

The Inland Empire commuter belt. Distress prevalence is at the national pop-weighted mean (16.8%); uninsured rate is below the CA average; drive-time to nearest provider is brief. The within-state regression predicts gap_ratio ≈ 2,800 given those inputs, and the observed 2,754 sits right on the line (z = +0.19). This is what an "expected" tract looks like under the framework: a working-age suburban tract with moderate need and moderate supply, where the gap matches the SES proxy with no residual signal. No policy lever is mis-firing here; no policy lever is over-performing.

### 2. Crow Wing County, MN · Tract 27035950401

- **Stats**: prevalence 14.3% · uninsured 6.1% · drive-time 28.2 min · z=-0.05 · gap ratio 20,513
- **Population**: 3,163 adults

Rural lake country near Brainerd. Below-average distress (14.3%), excellent insurance coverage (6.1% uninsured — Minnesota's expansion + MinnesotaCare buy-in), but the drive-time is at the 30-minute edge and the state supply ratio drives the gap_ratio higher. The regression captures the supply-thin / insurance-thick balance, and z ≈ 0 says: in rural Minnesota, this is the expected outcome. The 28-minute drive matters less than the headline number suggests because the population is small enough that telehealth + driving once a month is workable.

### 3. Ramsey County, MN · Tract 27123031200

- **Stats**: prevalence 16.7% · uninsured 7.1% · drive-time 1.2 min · z=+1.15 · gap ratio 23,956
- **Population**: 2,368 adults

Urban St. Paul tract. Despite a 1.2-minute walk to the nearest licensed provider and below-average uninsured (7.1%), gap_ratio is elevated — because **Minnesota's state-level provider density is the binding constraint**, not local distance. z = +1.15 sits just under the +1.5σ outlier threshold; this tract is starting to signal that the state-level supply count understates demand in dense urban tracts. The framework's design captures it as "expected, but pushing the boundary."

### 4. Harris County, TX · Tract 48201320100

- **Stats**: prevalence 17.1% · uninsured 38.8% · drive-time 5.6 min · z=-1.29 · gap ratio 35,727
- **Population**: 1,826 adults

Houston-area tract with **very high uninsured rate** (38.8%) — Texas did not expand Medicaid. The regression EXPECTS that a 38.8% uninsured rate predicts a large gap (because residents face barriers to filling appointments). And the gap is large — but z = −1.29 says it's *less* than the model predicts. The likely interpretation: county hospital systems and FQHCs (Harris Health, Legacy Community Health) absorb a chunk of the implied unmet demand without billing through commercial supply channels. The state-level supply count doesn't see those visits; the residual flags it.

---

## POSITIVE OUTLIERS — gap worse than the regression predicts

### 5. Los Angeles County, CA · Tract 06037910501

- **Stats**: prevalence 23.3% · uninsured 19.4% · drive-time 15.8 min · z=+1.50 · gap ratio 3,865
- **Population**: 4,058 adults

LA county tract. Distress is *substantially* elevated (23.3% vs 16.8% national pop-weighted), uninsured is moderate, drive-time is short. CA's state supply ratio is the highest among large states, so gap_ratio is *low* (3,865) — but the residual analysis says it's still 1.5σ worse than CA's regression predicts. The interpretation: in dense LA tracts, supply exists on paper but *appointment capacity* is exhausted. The state-level count doesn't see the saturation. This is what the two-problem framing surfaces — the "capacity gap" mirrors the LA mental-health market's appointment-availability crisis even though physical access is near-instant.

### 6. El Paso County, CO · Tract 08041005115

- **Stats**: prevalence 19.2% · uninsured 9.6% · drive-time 4.1 min · z=+1.65 · gap ratio 568
- **Population**: 4,559 adults

Colorado Springs / Pikes Peak area. Insurance is fine, distance is fine, gap_ratio is *very low* — Colorado's state supply ratio is high — yet z = +1.65 flags this tract as worse than predicted. The likely driver: high concentration of active-duty military + veteran population (Fort Carson, Peterson SFB, the Air Force Academy) puts mental-health demand that DoD/VA systems carry, while the state-supply count is civilian-only. The residual captures the structural undercount of civilian-system supply in military communities.

### 7. Ocean County, NJ · Tract 34029715701

- **Stats**: prevalence 21.3% · uninsured 14.2% · drive-time 10.2 min · z=+2.37 · gap ratio 44,976
- **Population**: 5,194 adults

Jersey Shore retirement/commuter belt. **The highest z-score in the dartboard** (+2.37σ). Distance is fine, insurance is moderate, but the gap is dramatically worse than NJ's regression predicts. The most plausible driver is a **senior-population mental-health load** that BRFSS underweights — older adults underreport distress on phone surveys, and Ocean County's age-65+ share is among the highest in NJ. This tract surfaces a measurement-coverage gap that the framework cannot resolve directly; it flags it for follow-up.

### 8. Cheshire County, NH · Tract 33005970600

- **Stats**: prevalence 17.6% · uninsured 6.6% · drive-time 20.5 min · z=+1.56 · gap ratio 11,216
- **Population**: 5,242 adults

Western New Hampshire, near Keene. Insurance is excellent (6.6% uninsured — NH expanded Medicaid 2014 + has strong employer coverage), distress is at-average, but the gap is 1.5σ worse than the regression predicts and the 20-minute drive is starting to matter. This is a **small-town capacity gap** — Keene Mental Health Center is the dominant provider for a wide rural catchment, and waitlists are real. The lever here is workforce expansion (BHWET) and rural-incentive recruitment, not telehealth alone.

---

## NEGATIVE OUTLIERS — gap *better* than the regression predicts

### 9. Beaufort County, SC · Tract 45013001101

- **Stats**: prevalence 10.9% · uninsured 8.1% · drive-time 19.0 min · z=-2.70 · gap ratio 9,602
- **Population**: 3,144 adults

Hilton Head / Beaufort area. **Distress is unusually low** (10.9% vs 16.8% national), insurance is fine, and the gap is 2.7σ *better* than the SC regression predicts. The driver is structural: affluent retiree tract with high private insurance, plus the Medical University of South Carolina's outreach footprint and integrated behavioral-health workflows in primary care. South Carolina has *not* expanded Medicaid, yet this tract's affluence + access to integrated care closes the gap on its own. This is what works at the *demographic* level (not the policy level).

### 10. Jackson County, MO · Tract 29095000900

- **Stats**: prevalence 21.2% · uninsured 18.4% · drive-time 0.6 min · z=-1.97 · gap ratio 33,682
- **Population**: 2,173 adults

Kansas City urban core. **High distress** (21.2%) and **high uninsured** (18.4%) — the kind of tract where the regression predicts the worst gap. But z = −1.97 means it is *much better* than predicted. The driver: Missouri voters approved Medicaid expansion in 2020 (implementation 2021) AND Kansas City hosts federally-designated **Certified Community Behavioral Health Clinics** (Truman Behavioral Health, Burrell Mental Health) paying providers to take Medicaid patients on real terms. This is the policy stack: expansion + CCBHC payment model.

### 11. Delaware County, OH · Tract 39041011563

- **Stats**: prevalence 11.9% · uninsured 4.1% · drive-time 14.2 min · z=-2.14 · gap ratio 35,063
- **Population**: 4,045 adults

Affluent Columbus suburb. Distress is low (11.9%), insurance is near-universal (4.1% uninsured — Ohio expanded Medicaid 2014 + this tract has dense employer EAP coverage), and the gap is 2σ better than the Ohio regression predicts. This is the **insurance-saturation + EAP** story: combined coverage closes the gap independently of provider density. Replicable by states that expand Medicaid and where employers extend EAP.

### 12. Deschutes County, OR · Tract 41017001301

- **Stats**: prevalence 12.9% · uninsured 3.8% · drive-time 18.3 min · z=-1.74 · gap ratio 15,603
- **Population**: 4,773 adults

Bend, Oregon. Low distress (12.9%), very low uninsured (3.8% — among the lowest in OR, which expanded Medicaid 2014), 18-minute drive — and the gap is 1.7σ better than the OR regression predicts. Bend has both private psychiatric supply (St. Charles Behavioral Health) AND integrated primary-care behavioral-health via Mosaic Medical (FQHC). This is the **affluent recreational town + integrated FQHC** story: economic base supports private supply; FQHC structure absorbs the lower-income tail.

---

## What this dartboard says, taken together

- **The 4 expected tracts confirm the framework's ability to identify "on the regression line" places** without spurious flagging. The residual analysis isn't generating false positives from noise.
- **The 4 positive outliers cluster on a single structural driver in 3 of 4 cases** — the supply count systematically undercounts demand when sub-populations are invisible to it (urban capacity saturation, military communities, retiree distress under-reporting, rural town waitlists). The framework surfaces the count problem; it cannot resolve the count.
- **The 4 negative outliers cluster on the same policy stack in 3 of 4 cases** — Medicaid expansion + a sustainable payment model (CCBHC) + or near-universal coverage + integrated care. This is the replication template. (Beaufort SC is the affluence-and-MUSC variant, not the policy variant.)

The dartboard is 12 tracts. It is small. But it is pre-specified, transparent, and stratified. It shows the patterns the residual analysis is detecting at scale — and gives the reader a way to verify the framework's signal by inspecting individual tracts whose stories check out against the public record.