Skip to main content
v2.0-us · 2025

UQLI US Edition — Methodology

The UQLI™ US Edition composite score is constructed from 20 city-level indicators across ten dimensions, normalized using winsorized min-max scaling and aggregated via a weighted geometric mean. Dimension weights are calibrated via PCA on the normalized indicator matrix and blended with structured expert judgment. All indicators vary meaningfully across US cities — see Data Sources below.

Normalization

Winsorized Min-Max Normalization

Each indicator is winsorized at the 5th and 95th percentiles to limit the influence of extreme values. Values are then scaled to a 0–100 range using min-max normalization. A floor of 2.0 prevents undefined behavior in the geometric mean.

Composite Method

Weighted Geometric Mean

Indicator scores within each dimension are averaged (arithmetic mean). The ten dimension scores are then combined using a weighted geometric mean. This penalizes extreme imbalance: a city with one very low dimension score cannot fully compensate by excelling elsewhere.

Confidence Bands

Data Completeness Confidence

Confidence ratings reflect the share of indicators with available data: High (≥80% data available, ±2.5 pts), Medium (≥60%, ±5.0 pts), Low (<60%, ±10.0 pts). Wider bands reflect greater uncertainty from missing inputs — not statistical sampling error.

Theoretical Framework

The UQLI is grounded in three complementary frameworks for measuring human welfare:

Capability Approach (Sen & Nussbaum)

Quality of life is measured by the real freedoms people have to live well — not merely income or utility. The UQLI captures capabilities across health, education, safety, and civic participation.

OECD Better Life Framework

The OECD's multi-dimensional well-being framework distinguishes material conditions (income, housing, employment) from quality-of-life factors (health, education, environment, civic engagement). UQLI adopts a comparable dimensional structure.

Stiglitz-Sen-Fitoussi Commission

The 2009 Commission on the Measurement of Economic Performance and Social Progress established that GDP is insufficient as a welfare measure. UQLI incorporates sustainability, inequality, and subjective well-being dimensions alongside economic output.

10 Dimensions

20 US city-level indicators distributed across ten dimensions

1

Health & Longevity

Primary Factor

All-cause mortality burden — age-adjusted rate per 100k, 5-year window (CDC WONDER D77, annual); obesity, diabetes, mental health, physical activity, and chronic disease prevalence (CDC PLACES); health insurance coverage (Census ACS). All indicators are city-level.

2

Economic Security & Opportunity

Key Factor
3

Education & Human Capital

Key Factor

Bachelor's degree attainment rate and high school graduation rate among adults 25+ — city-level from Census ACS S1501.

4

Environmental Quality

Key Factor

Annual mean PM2.5 air quality (EPA AQS), EPA walkability index, % population within 10-min walk of a park (TPL ParkServe), and food desert exposure (USDA FARA) — genuine city-level measurements.

5

Safety & Security

Key Factor

Violent crime rate and property crime rate per 100k population — city-level from FBI UCR Table 8.

6

Infrastructure & Mobility

Key Factor

Mean commute time, housing overcrowding rate, gross rent burden rate, and price-to-income ratio — all city-level from Census ACS.

7

Housing & Shelter

Key Factor
8

Civic Governance & Freedom

Key Factor

Governance indicators (rule of law, press freedom, government effectiveness) apply uniformly to all US cities by design and are excluded from scoring. This dimension is suspended in the US Edition — its weight is redistributed proportionally across the remaining nine dimensions via the scoring engine.

9

Social Cohesion & Inclusion

Key Factor
10

Digital Access & Innovation

Key Factor

US City-Level Data Sources

1,852 US Cities · 20 City-Level Indicators · 2025 Data Cycle

CDC PLACESCity-level health measures: obesity, diabetes, mental health, physical activity
CDC WONDER (D77, annual)Age-adjusted all-cause mortality rate — county-level, 5-year window, aggregated to city
US Census Bureau ACS 5-YearIncome, commute, education attainment, Gini, health insurance
FBI Crime Data ExplorerViolent and property crime rates per 100k — city-level UCR Table 8
EPA Smart Location DatabaseBlock-group walkability index (NatWalkInd), aggregated to city
Trust for Public Land ParkServe% population within 10-min walk of a park — city-level
USDA Food Access Research AtlasLow-income & low-access food desert tracts — aggregated to city
OpenAQ (EPA AQS network monitors)Annual mean PM2.5 — aggregated from EPA AQS monitor readings, nearest station within 25km of city center
US Census Bureau ACS 5-Year (Housing)Price-to-income ratio, overcrowding rate, rent burden rate — city-level from ACS B25077/B25014/B25070
Stanford Education Data Archive (SEDA 4.1)District-level learning outcomes — pooled test score composite (2009–2018), matched to city via district name

Missing indicator values are imputed using the US peer-group median — the median normalized score of cities sharing the same population tier. Imputed values contribute to dimension score arithmetic means and are flagged in the indicator-level API response. They are excluded from the count of genuinely available indicators used in confidence interval calculations.

Methodology Improvements (v2.0-us)

The following issues from earlier versions have been resolved in the US Edition:

Federal-level uniform indicators removed

Rule of law, press freedom, and governance effectiveness — which do not vary across US cities — have been fully deactivated and excluded from scoring. The governance dimension is suspended in the US Edition; its weight is redistributed proportionally across the remaining nine dimensions.

Learning outcomes added alongside attainment

Stanford SEDA district-level test scores (pooled 2009–2018) are now ingested and matched to cities, adding a genuine learning outcomes signal to the Education dimension alongside attainment proxies (bachelor's degree rate, HS graduation rate).

Dimension weights calibrated via PCA

Dimension weights are derived as a 70/30 blend of PCA-derived empirical importances (from PC1 of the normalized indicator matrix) and structured expert judgment. This reduces arbitrary subjectivity and partially corrects for inter-dimension correlation.

Dimension intercorrelation partially addressed

The PCA-blended weighting naturally down-weights dimensions that are highly collinear with others (health, income, education), as their indicators contribute less unique variance to PC1. A full Cholesky orthogonalization is planned for v3.0.

Known Limitations

The UQLI research team is committed to methodological transparency. The following inherent constraints of US city-level public data are acknowledged:

Mortality data is county-aggregated, not city-direct

Age-adjusted all-cause mortality rates come from CDC WONDER D77, which publishes at the county level. City-level rates are derived via area-weighted crosswalk from the Census 2020 Place-County relationship file. For multi-county cities, this introduces approximation error. Single-county cities (>85% of UQLI cities) have exact matches. This is the finest geographic resolution publicly available for cause-of-death data in the US.

Temporal heterogeneity in data vintages

Indicators draw from different reference years: CDC PLACES (2025), Census ACS (2023), FBI UCR (2022), SEDA learning outcomes (2009–2018 pooled). Composite scores represent a snapshot of best-available data across sources, not a single calendar year. The data vintage for each indicator is shown on individual city profile pages.

SEDA learning outcomes matched by district name, not FIPS crosswalk

Stanford SEDA test score data is matched to UQLI cities via school district name fuzzy-matching rather than a formal NCES-to-Census-place geographic crosswalk. Cities without a SEDA match receive peer-group median imputation for this indicator. A formal LEAID-to-FIPS crosswalk is planned for the next data cycle.

Composite Score Formula

The UQLI score is computed through a multi-stage normalization and aggregation pipeline. Outliers are statistically bounded before normalization. Dimension scores are aggregated using a non-linear weighted method that penalizes extreme imbalance — a city cannot compensate for very low performance in one dimension by excelling elsewhere.

Step-by-step

1. Winsorize each indicator at p5 / p95 across all cities

2. score_i = (value − min) / (max − min) × 98 + 2   (floor 2, ceiling 100)

3. DS_d = mean(score_i) for all indicators i in dimension d

4. UQLI = ∏d DS_dw_d   (weighted geometric mean; Σw_d = 1.0)

Exact dimension weights are proprietary. The weight tiers shown above (Primary / Key / Contributing / Supporting) reflect the relative magnitude of each dimension's contribution to the composite score.