Each Public Health Budget Question Requires a Distinct Empirical Study Design
We match linked state records, trend splits, county panels, and outreach trials to each budget question.
Public health leaders answer different budget questions at each stage of launching and scaling a program. Sizing a multilingual outreach cohort, forecasting a statewide paid leave law, scoring county health budgets, and testing phone call scripts each require different data structures. Matching each budget question to its study design prevents costly planning gaps.
Matching each California public health budget question to its study design locates 120,239 Medi-Cal members needing renewal outreach,[1] +39,600 Paid Family Leave claims,[2] 2.38 annual to 4.32 cumulative fewer deaths per 100,000 from county health spending,[3] and an 8% to 11% renewal lift from live calls.[4] Using one raw source alone undercounts Medi-Cal adults by -40.2% in home surveys or mixes trend growth with new policy shifts.[1][2] Pairing linked records, trend splits, county panels, and outreach trials gives leaders a full toolkit.[3][4]
Table 1. Matching Four California Health Policy Questions to the Right Study Design
We compare four CAPHE studies across budget question, raw baseline error, matched design, and main result.
| Study Design Factor |
Rung 1: Target Size
Record + Survey Link
|
Rung 2: Law Shift
Trend Split + Model
|
Rung 3: County Panel
Routine-Budget Panel
|
Rung 4: Outreach Step
Outreach Trial + Funnel
|
|---|---|---|---|---|
| Main Budget Question | How many members need language help? | What will SB 951 + 12-wk PFL cost? | Does county health spending save lives? | Which outreach script lifts renewals? |
| Raw Baseline Gap | ACS alone undercounts by -40.2%152.8K vs 255.6K | Pre-post mixes +1.52% claim drift & +3.5% pay | Pandemic surge grants give +2.47plus sign | Total rate leaves out 85% pre-visit drop |
| Matched Study Method | Fix totals to DHCS; use ACS shares | Split 2025-26 base slope from law jump | Separate pre-COVID routine budgets & horizons | 5-arm trial (N=82,695) + funnel check |
| Main Result | 120,239 Exposed23K Sp/Viet LEP | +39,600 Claims$623.7M Gross | 2.38 Yr / 4.32 Cumul Fewer32-59:1 | +8-11% Call Lift15.0% CHW Link |
| Key Check or Group Detail | 15,559 Spanish + 7,467 Viet + 3,489 Chinese | 325,055 (2025) -> 330,000 base vs. 369,600 | Urban (-2.59) & Rural (-2.42); 13-yr IV -5.31 | Text and email arms: 0% lift |
| Direct Budget Action | Hire bilingual staff for 26,515 adults | Set payroll tax at +0.04 pp$60/yr | Back base county LHA$420K yr / $231K cumul | Fund live calls and CHWs over text tweaks |
Table 1 Notes & * Modeled Estimates
Expand notes
Table 1 Notes (Mode B Empirical Estimates; Rung 2 Mode A Simulation): Each column uses a separate study group (Santa Clara adult Medi-Cal members, California PFL claimants, 57 California counties, and Medicaid outreach and CHW trial participants). The Rung 2 PFL SB 951 claim and cost numbers are Mode A simulation outputs. In Rung 3, the 2020-2023 pandemic surge slope is +2.47 deaths per 100,000 per $10 per resident (p = 0.001) and pooled 2003-2023 TWFE is -0.25 (p = 0.710). Rung 3 pre-COVID routine-budget TWFE (2003-2019, N = 943) yields -2.38 (p = 0.0022), 13-year long-difference OLS yields -4.32 (p = 0.0052), and Long-Difference Lewbel 2SLS yields -5.31 (first-stage F = 111.20, Hansen J p = 0.2132).
Column Scope. Record + Survey Link. SCC Medi-Cal HR 1 Scope.. Trend Split + Model. CA Paid Family Leave (SB 951).. Routine-Budget Panel. 57-County LHA Pay (2003-2019).. Outreach Trial + Funnel. Call Scripts & CHW Reach..
Model Crosswalk. Plain table labels map to standard study models. Raw Yearly Average / Simple Avg is an unadjusted OLS model where routine and outbreak budgets mix. County & Year FE is a Two-Way Fixed Effects (TWFE) panel model. Routine-Budget Analysis / 21-Year County Budget Model is a Two-Stage Least Squares (2SLS / Lewbel 2012) model that isolates routine base budget shifts. County Board Budget Rules uses Brown (2014) county board rules.
Two-Ruler Benchmark Summary. Mode B Multi-Design Summary Across Medi-Cal, Paid Family Leave, County LHA & CHW Outreach Studies. Ruler 1: Single-Source & Raw Estimates (Undercounts, Drift & Surge Mix): -40.2% Undercount | -12.3% Drift | +2.47 Surge Mix. Using an unmatched design undercounts Santa Clara adult Medi-Cal members by -40.2% in ACS surveys (152,811 vs. 255,601 DHCS). It also misreads a -12.31% pre-law enrollment drop or flips the sign in pandemic county spending (+2.47 in 2020-2023). Ruler 2: Matched Design Toolkit (Linked Records, Trend Split, County Panel & Trials): 120,239 Target | +39.6K PFL | 2.38-4.32 Fewer Deaths | +8-11% Lift. Matching each question to its design finds 120,239 exposed members and +39,600 PFL claims ($623.7M). It also finds 2.38 annual / 4.32 cumulative fewer deaths/100K (32:1 to 59:1 BCR) and +8% to +11% live-call renewal lift.
Table 2. How Each Study Design Guides Agency Plans and Budgets
We show how linked records, trend splits, county panels, and outreach trials work together.
| Choice Factor | Record + Survey Link | Trend Split + Model | Routine-Budget Panel | Outreach Trial + Funnel |
|---|---|---|---|---|
| When This Design Fits Best | When a team sizes a target group and finds language or age subgroups before launch. | When a state law (like SB 951's jump to 90% pay) starts on one date statewide. | When budgets shift across 57 counties and pandemic surge grants must be separated from base care. | When a team tests outreach scripts across 82,695 members or checks referral steps. |
| What This Design Adds | Gives the exact county headcount and language shares to staff outreach. | Splits yearly wage and claim growth from the new law jump. | Tracks death rates across 57 counties over annual (-2.38) and 13-year cumulative (-4.32) horizons. | Checks each referral step to show that 78.1% of reached patients link to care once reached (27.8%). |
| Risk When Design and Question Mismatch | Using raw ACS survey counts under-budgets renewal staff by 40.2%. | Treating a pre-law -12.31% enrollment drop as a 2027 law effect overstates loss. | Pooling 2020-2023 emergency COVID-19 grants (+2.47) with routine budgets attenuates prevention gains. | Spending months testing SMS text wording (0% lift) delays live phone calls that lift renewals +8% to +11%. |
| Action Plan for Agency Teams | Link State Totals + Survey Shares. Use DHCS rolls for total headcount and ACS microdata for language shares. | Model Base Slopes Before Laws. Use multi-year trend splits to separate wage growth from new benefit jumps. | Use Routine-Budget Panels for ROI. Judge multi-year county budgets with pre-COVID panels and 13-year cumulative horizons. | Test Outreach Calls & Funnels. Use outreach trials and funnel checks to tune call scripts and CHW caseloads. |
Table 2 Notes: Referral Funnel Math & Multi-Study Provenance
Expand notes
• CHW Referral Funnel Math (Mode B Published Trial Split): In the Bronx CHW study (N = 3,522 patients, 5,021 referrals; INTERVENTION_DESIGN.md:L185-232), total referral completion was 15.0%. That equals 27.8% started outreach × 69.2% of started were reached (19.2% of total) × 78.1% of reached linked to care (0.278 × 0.692 × 0.781 = 0.1502 = 15.0%).
• Data Provenance: Exact numbers come from scc_hr1_medicaid/SCC_HR1_MediCal_Risk_Profiles.md (L13-112), scc_hr1_medicaid/scc_dhcs_medical_enrollment.csv (L2-15), INTERVENTION_DESIGN.md (L19-232), public/tools/pfl-simulator/data/pfl_model_data.json (L20-169), and src/frontend/js/lha-calculator.js.
• Extended Cell Context. What This Design Adds (Record + Survey Link). Gives the exact county headcount and language shares (like 7,467 Vietnamese members) to staff outreach. · What This Design Adds (Trend Split + Model). Splits yearly wage and claim growth (+1.52% claims, +6.88% pay) from the new law jump (+39,600 claims). · Risk When Design and Question Mismatch (Record + Survey Link). Using raw ACS survey counts (152,811 vs. 255,601 DHCS adults) under-budgets renewal staff by 40.2%. · Risk When Design and Question Mismatch (Trend Split + Model). Treating a pre-law -12.31% enrollment drop (120,239 to 105,436 in 2025) as a 2027 law effect overstates loss.
Why Health Agencies Use All Four Methods Together
A county health director launching a Medi-Cal renewal or chronic-care plan needs four types of proof at four steps. First, record-survey linkage sizes the group (fixing the -40.2% ACS undercount to find 120,239 exposed adults and 26,515 LEP members). Second, trend models split baseline enrollment drift (-12.31%) from new laws. Third, multi-year county panels measure department returns (2.38 fewer deaths/100K annually and 4.32 over 13-year cumulative horizons, 32:1 to 59:1 BCR). Fourth, outreach trials test which contact channel works.
Checking each step of the delivery funnel turns study numbers into daily fixes. In California's 5-arm renewal outreach trial (N = 82,695, NBER WP 34191), changing the wording of texts and emails gave a 0% difference. Adding a live phone call lifted renewals by +8% to +11% (+1.0 to +1.8 pp). Likewise, splitting a CHW referral funnel shows that 78.1% of reached patients link to care. The 15.0% overall rate points to an early gap in starting referrals (27.8%), not a gap in CHW support.
Data Sources, Model Inputs, Notes & Endnotes
(Numbered endnotes [1]–[5], state data sources & math rules)
Expand details
Case Study Focus: Study Credibility & Measurement · Matching Designs to Questions (Mode B Portfolio)
Numbered Endnotes & Main Sources
- [1]California Department of Health Care Services (DHCS): Medi-Cal Certified Eligibles & Redetermination Dashboards Linked With U.S. Census ACS PUMS.
- [2]California Employment Development Department (EDD): May 2026 Disability Insurance & Paid Family Leave Fund Forecast.
- [3]California State Controller's Office: 57-County Public Health Spending Panel (2003–2023) & Brown (2014).
- [4]National Bureau of Economic Research (NBER WP 34191): California Medicaid Renewal Outreach 5-Arm Trial (N = 82,695).
- [5]Brown (2014), Health Economics
Mode B Studies + Mode A PFL Simulation: Rung 1 (120,239 DHCS / 152,811 ACS vs. 255,601), Rung 3 (-2.38 annual and -4.32 cumulative routine-budget panel), and Rung 4 (NBER WP 34191 N = 82,695 and Bronx CHW 27.8% × 69.2% × 78.1% = 15.0%) are observed Mode B studies. Rung 2 (330,000 vs. 369,600 PFL claims, $623.7M) is a Mode A simulation calibrated to EDD forecasts. The four rungs use four separate groups (Santa Clara adult Medi-Cal members, California PFL claimants, 57 California counties, and Medicaid outreach and CHW trial participants). Their estimates are not pooled on one scale.
Main Sources: We cite NBER Working Paper 34191 (California Medicaid Renewal Outreach 5-Arm Trial, N = 82,695) and California DHCS Medi-Cal Enrollment Dashboards. We also use California EDD May 2026 DI Fund Forecast and Brown (2014), Health Economics.
Common Questions & Method Notes
- Why does the ACS survey undercount Santa Clara County adult Medi-Cal members by 40.2%?
- Home surveys face a known Medicaid undercount. People mix up Medi-Cal plan names (like Santa Clara Family Health Plan or Anthem Blue Cross) with private plans or leave out household members. Fixing totals to DHCS state records (255,601 adults vs. 152,811 in ACS) while using ACS for language and work-hour shares fixes the group base.
- What did the 82,695-person California Medicaid outreach trial (NBER WP 34191) show about text wording vs. live calls?
- Across four written arms testing different SMS and email wording, renewal lift was 0%. Only the fifth arm, which added a live phone call, raised renewals by 8% to 11% (+1.0 to +1.8 pp). Live human help clears paperwork hurdles where text tweaks do not.
- Why does splitting a referral funnel into start (27.8%), reach (69.2%), and link (78.1%) matter for teams?
- Looking only at the 15.0% overall completion rate (0.278 × 0.692 × 0.781) makes CHW outreach look weak. Splitting the funnel shows that CHWs link 78.1% of the patients they reach. The real operational gap is starting the referral (27.8%).