Financial Health Barometer

Each indicator combines a government data source with a base index, scaling factor, and regional adjustment:

• Financial Anxiety = (120 + (Unemployment Rate - 3.5%) × 18) × Regional Multiplier + Trends Boost
• Food Insecurity = (85 + (Poverty Rate - 10%) × 6) × Regional Multiplier + Trends Boost
• Housing Stress = (100 + Rent Burden Score + FMR Score + HPI Score) × Regional Multiplier + Trends Boost
• Affordability = (Housing Stress × 0.60 + Food Insecurity × 0.40) + Trends Boost

Housing sub-scores: Rent Burden Score = (MedianRentRatio - 25%) × 3; FMR Score = (State FMR / National Avg FMR - 1) × 40; HPI Score = Housing Price Change % × 2.
Primary sources: Census ACS B25071, HUD FMR API, FRED HPI. Fallback: Harvard JCHS 2025.

The base index values (120 for Financial Anxiety, 85 for Food Insecurity, 100 for Housing Stress) are chosen to place typical economic conditions in the "Moderate" range.

Regional multipliers (0.85–1.35×) are the single most consequential judgement call in this tool, and they are not measurements. Every index value you see is multiplied by its state’s figure. They were assigned by hand to reflect structural conditions that annual federal series are slow to capture — cost of living, entrenched poverty, labour-market depth. They are not estimated from data, not fitted to any outcome, and carry no confidence interval. Reasonable analysts would choose different numbers.

Their effect is large enough that we publish it plainly. If the multipliers were removed and the indices left purely data-driven:

• 42 of 51 Financial Anxiety ranks would change.
• Mississippi would move 24 places.
• The multiplier range spans about 60 index points on the 120 base — against roughly 72 points for the entire national spread of actual state unemployment rates.

Put plainly: roughly half the variation you see between states on the map comes from this table rather than from the government data. Treat state-to-state gaps as indicative, not as measured differences. Every state’s multiplier is published in latest.json under meta.regional_multipliers, and per state under metrics.regional_stress_multiplier, so you can divide it back out and rebuild the unadjusted index yourself.

Tier estimates. Two housing components have a hand-assigned fallback band for states where no reading can be sourced: rent burden (three tiers, worth 21, 12 or 6 points, with every other state scoring 0) and fair market rent (a single high-cost band worth 15 points). Like the regional multipliers these are assumptions, not measurements. They sit below carry-forward in the order of preference, so in normal operation they should never fire; the number of states actually scored from a tier in the current reading is published in meta.tier_estimates.states_scored, and any affected state is labelled tier_estimate in its metrics. The full tables are published alongside them.

Index bounds. Each index is clamped to a fixed range (Financial Anxiety, Housing Stress and Affordability to 80–200; Food Insecurity to 55–160). A state sitting exactly at a bound has been cut off there, and is tied with any other state at that bound — the order between them is an artefact of sorting, not a real difference. Clamped values carry a clamped flag in the published data.

Health Trends boost (+0–10 pts): FinMango accesses the restricted Google Health Trends API, which returns absolute search probabilities (P(term) × 10,000,000) rather than the relative 0–100 scale on the public Google Trends website. The raw API value is used directly as the boost (capped at 10 pts). The quota is too small to read all 51 states daily, so the fetch rotates through the states and refreshes the full set roughly weekly. The boost is only added to an indicator once every state has a current reading — a boost reaching some states and not others would rank states by which ones the quota happened to reach that week. Coverage for the current reading is shown at the top of this section.

Indices are relative measures of economic stress, scaled on a 0–200 reference range (extreme periods, such as 2020–2022, can exceed 200):

• < 90 (Green): Low Stress / Stable
• 90–120 (Yellow): Moderate Stress
• 120–150 (Orange): Elevated Stress — Warning Signs
• > 150 (Red): Crisis Level — Immediate Attention Needed

These bands are not comparable across indicators. The four indices are built from different inputs on different base values (120, 85 and 100) and are clamped to different ranges. A Housing Stress reading of 150 and a Food Insecurity reading of 150 do not represent the same severity of hardship, and the gap between two indicators for one state is not meaningful. Compare a single indicator across states, or across time — not across indicators.

Affordability is not an independent measurement. It is defined as 60% of Housing Stress plus 40% of Food Insecurity, so it restates the two indices beside it rather than adding new information — in the current reading it tracks Housing Stress at r = 0.90 by construction. Do not read it as a fourth, corroborating signal.

The trend chart is not a history of the index. The index is computed daily from current data and has never been back-calculated for past months — the federal inputs behind it do not exist as a consistent daily series going back a decade. What the chart shows instead is the shape of Google Health Trends search interest in a single representative search term per indicator, rescaled so its most recent month equals today’s composite index value, then smoothed with a 3-month trailing average.

The term behind each line:

• Financial Anxiety — “debt help”
• Food Insecurity — “food stamps”
• Housing Stress — “eviction help”
• Affordability — “cost of living”

So the Food Insecurity line traces how often people searched “food stamps”, not how the poverty-based index moved. Read a historical point as search behaviour at that time, drawn on the index’s scale — not as what the Barometer would have read that month. Only the final point is an actual index value.

Because search data is inherently volatile, month-to-month fluctuations may be large — focus on longer-term direction rather than individual data points.

The indices utilize a Heuristic Stress Model to quantify economic pressure, standardizing disparate metrics into a unified 0–200 stress index.

Researcher Note: Composite indices are heuristic — designed for relative comparison between states, not as absolute measures. For econometric modeling, use the raw underlying metrics (provided in the full CSV export).

Everything we know to be a weakness in this instrument, in one place. If you find another, open an issue.

• Roughly half the state-to-state variation is an assumption, not a measurement. The hand-assigned regional multipliers reorder 42 of 51 ranks. See “Scaling & Regional Adjustment” above.
• Two of the four indicators have no change data at all. Food Insecurity rests on annual Census SAIPE poverty figures, and Affordability is derived from the other two indices. Neither has a period-over-period change to report, so both display “no change data” rather than a percentage.
• Affordability is not independent. It is a weighted restatement of Housing Stress and Food Insecurity, so it cannot corroborate them.
• The indicators update far more slowly than “daily” suggests. The pipeline runs every day, but its inputs do not: unemployment is monthly, poverty and rent burden are annual, and house prices are quarterly. Most days, most numbers are unchanged. The Barometer summarises official statistics in near-real-time; it cannot see ahead of their release schedule.
• Annual sources carry their own lag. Census SAIPE and ACS releases run one to two years behind the period they describe. A poverty-driven reading today reflects conditions from a prior year.
• The trend chart shows search interest, not index history. Only the most recent point is an actual index value.
• Clamping creates ties. States pinned at an index bound cannot be ordered against each other, though the ranking still prints a number for each.
• Index bands are not comparable between indicators. Compare one indicator across states or over time, never two indicators against each other.
• State-level figures hide the variation that matters most. Housing cost and hardship differ enormously within a state; a state median describes no particular household.
• When an upstream API fails, the affected numbers are held, not recomputed. The last real measurement is carried forward and labelled in “What today’s reading actually used”. Dropping a term from a composite would be worse than holding it: the house-price term alone is worth about 29 index points, so a state that lost it would show a phantom improvement. Carried values expire once a real release has been missed; past that the index is flagged partial and a warning names the affected states.
• Two housing components have hand-assigned fallback bands. Where no rent-burden or fair-market-rent reading can be sourced or carried forward, a state is placed in a tier by hand. These are assumptions like the regional multipliers; the count of affected states is published each run.
• Unemployment depends on a shared API quota. Without a BLS registration key the fetch uses an anonymous quota counted per IP address, which shared CI runners routinely exhaust. This took unemployment offline for 23 consecutive days in August 2026. The reading now reports the real reason, and the freshness badge reflects how old the measurements are rather than how recently the file was written.
• The index is heuristic. Its base values, weights and multipliers were chosen for readability, not derived from a model of household finances. It has no error bars. For econometric work, use the raw underlying metrics in the CSV export rather than the composite scores.