How the County Distress Index Works

The Five Dimensions

The CDI organizes its 21 indicators into five dimensions. The dimensions were not chosen editorially. They emerged from principal component analysis run on the full indicator set across all 3,144 counties — the statistical procedure identifies independent factors of variation, and the weights are proportional to how much of the cross-county variance each factor explains.

How the Score Is Calculated

The pipeline is thirteen steps. Each county starts with 21 raw indicator values and ends with a single 0–100 composite, a zone assignment, and per-factor breakdowns.

1. Data collection. 21 indicators pulled from 14 federal sources. All data is publicly available. No proprietary datasets or paywalled inputs.
2. Coverage check. Indicators must cover at least 75% of counties (≥2,358 of 3,144) to be included. Three mortgage indicators were evaluated and excluded — see Coverage Gap below.
3. Imputation. Missing values are filled with the state-level median for that indicator. If no state median is available, the national median is used. Total: 1,222 imputations out of 66,024 cells (1.9%).
4. Direction alignment. All indicators are oriented so that higher values mean more distress. Five indicators are inverted: MHI ratio to state median, homeownership rate, wage-to-rent ratio, business applications per 1,000, and FHFA house price YoY change.
5. Winsorization. Extreme values are clipped at the 2.5th and 97.5th percentiles of each indicator before standardization. This prevents a single outlier county from distorting the scale.
6. Z-score standardization. Each indicator is converted to Z-scores (mean=0, std=1) across all 3,144 counties, so all indicators are on the same scale regardless of units.
7. Principal component analysis. PCA is run on the correlation matrix of the 21 Z-scored indicators. Pre-PCA diagnostics confirm the data is suitable for factor analysis: Kaiser-Meyer-Olkin sampling adequacy is 0.85 (meritorious), and Bartlett's test of sphericity rejects the null at p < 0.001 (χ² = 49,636, df = 210). Component count is determined by parallel analysis (Monte Carlo simulation with 1,000 iterations) — five components is also what the Kaiser criterion and the 70% cumulative variance threshold recommend. The five components explain 70.6% of total variance.
8. Varimax rotation. Components are rotated using varimax rotation to maximize interpretability, so each indicator loads strongly on one factor rather than weakly on several. Rotation does not change total variance explained.
9. Weight derivation. Each indicator is assigned to its primary factor (highest absolute loading). Factor weights are proportional to variance explained: Factor 1 carries 47.5% because it explains 47.5% of the retained variance. Within each factor, indicator weights are proportional to squared loadings. Final indicator weight = factor weight × within-factor weight. All weights sum to 100%.
10. Percentile scoring. Each Z-scored indicator is converted to a percentile rank (0–100) across all 3,144 counties. Percentile ranks are more interpretable than Z-scores for county-level rankings — a percentile of 72 means "more distressed on this indicator than 72% of U.S. counties."
11. Small-county adjustment. For any county with population at or below 20,000, each individual indicator percentile is clipped to the range [5, 95]. See the next section for the rationale and an example.
12. Weighted average. The composite is the weighted sum of clipped indicator percentiles: CDI = Σ (indicator_percentile × indicator_weight).
13. Zone assignment. The composite is mapped to one of five zones — Healthy, Normal, Elevated, Serious, or Crisis — using fixed thresholds.

Why These Weights

The question behind any composite index is how to decide which parts matter more. One choice is to weight everything equally. Another is to weight editorially — the designer decides what matters based on theory. The CDI uses neither. It uses principal component analysis to let the data reveal which dimensions of distress vary most across counties, and then sets the weights proportional to variance explained.

The approach is only defensible if the data is actually suitable for factor analysis. Two diagnostics confirm that it is. The Kaiser-Meyer-Olkin statistic measures how much of the variance in each indicator can be explained by the other indicators. For the CDI, KMO is 0.85 — on the scale Kaiser published in 1974, that's "meritorious." The minimum threshold for factor analysis is 0.60. Bartlett's test of sphericity tests whether the indicator correlation matrix is distinguishable from pure noise. For the CDI, the test statistic is χ² = 49,636 with 210 degrees of freedom, which rejects the null at p < 0.001. The indicators are correlated enough to support meaningful factor extraction.

Component count was chosen by parallel analysis — a Monte Carlo procedure that generates random datasets with the same dimensions and retains only the components whose eigenvalues exceed the 95th percentile of what random data produces. The method is considered the gold standard because it does not depend on arbitrary thresholds. For the CDI, parallel analysis recommended five components. So did the Kaiser criterion (eigenvalue > 1) and the 70% cumulative variance rule. All three agreed. The five retained components explain 70.6% of the total variance across 21 indicators — a high share for a cross-sectional factor model with this many variables.

After varimax rotation, each indicator loads most strongly on a single factor, and the interpretation of each factor becomes straightforward. The weights are then derived mechanically: factor weight proportional to variance explained, indicator weight within a factor proportional to its squared loading. At no point does editorial judgment enter the weighting.

Small-County Adjustment

Before the indicator percentiles are combined into the composite, one more adjustment runs. For any county with a population at or below 20,000, each individual indicator percentile is clipped to the range [5, 95].

The reason is statistical volatility. A county of 5,000 people can trip to the 99th percentile on a single indicator because of small-sample noise alone. A handful of extra bankruptcies in a single year, or one unusual ACS housing estimate, can push a tiny county to an extreme reading that would not appear in a larger county with the same underlying conditions. The CDI is meant to measure relative household distress, not to reward or punish counties for the statistical fragility of their data.

The clip prevents any single indicator from dominating the composite in a small-population county. The county can still score at the very top or bottom of the national rankings — but only if many indicators move together, which is precisely what the index is designed to detect. Larger counties are not adjusted.

The most visible effect of the adjustment is on the extremes. Tunica County, Mississippi — population 9,234 — would score 90.8 without the clip and would be the most distressed county in the country under raw percentile scoring. With the adjustment applied, Tunica scores 88.54 and ranks third nationally, and Richmond County, Georgia — population 205,414, no clip applied — becomes the most distressed county at 89.15. At the opposite end, Los Alamos County, New Mexico (population 19,444) would score 9.3 without the clip. With the clip, Los Alamos scores 11.37 and ranks second-to-last, with Hamlin County, South Dakota holding the least-distressed position at 10.85.

Zone Thresholds and Current Distribution

The 0–100 composite is mapped to five zones with fixed thresholds. The percentages shown here reflect the live distribution across all 3,144 counties.

The full interactive distribution, map, and sortable tables are on the county distress hub. Individual county profiles sit at /counties/[state]/[county]/.

Score Distribution

By construction of percentile-rank scoring, the national median composite is approximately 50. Half of counties score above, half below. This is a feature of the methodology, not a coincidence.

Data Sources

All CDI data comes from public sources. No proprietary datasets, no paywalled inputs, no survey-based estimates that cannot be independently replicated.

Coverage Gap: Mortgage Delinquency

The most significant limitation of the current CDI is the absence of a direct mortgage delinquency indicator. Three mortgage-related indicators were evaluated during construction and excluded because none could cover the 75% threshold required for inclusion in the composite. The exclusions are structural features of the source data, not data staleness.

Mortgage distress is captured indirectly in the composite through the subprime credit population indicator (99.7% county coverage), debt in collections, and credit card delinquency — all of which the national American Distress Index validates as leading indicators of mortgage default. When county-level mortgage data coverage improves, direct mortgage delinquency will be incorporated.

Missing Data and Imputation

Not every county has data for every indicator. Small counties and independent cities sometimes lack coverage in one or more sources. When a county is missing an indicator value, the state median for that indicator is substituted. This is a conservative imputation method — it pulls the missing county toward the center of its own state rather than the national center, preserving geographic patterns without fabricating precision.

Across 3,144 counties and 21 indicators, 1,222 values are imputed via state medians — roughly 1.9% of the 66,024-cell matrix. The most affected indicator is the FHFA house price change (445 counties, 14.2% imputed, mostly small counties without enough repeat sales for the index), and it carries the lowest weight in the composite at 0.8%. No county requires imputation for more than half its indicators.

Validation

Face validation. Twenty-eight benchmark counties were tested before the index was locked — sixteen known-distressed (Mississippi Delta, Appalachian coal, Black Belt, border counties, tribal reservations) and twelve known-healthy (wealthy suburbs, tech corridors, college towns, resort areas). Fifteen of sixteen known-distressed counties correctly ranked in the top half (94%). All twelve known-healthy counties ranked in the bottom half (100%). The only misclassification was Towns County, Georgia, which ranked 1,710 — just outside the top half.

Sensitivity analysis. Four candidate formulas were tested against the PCA-proportional specification: equal indicator weights, equal domain weights, and a PCA-plus-geometric-mean variant. Spearman rank correlations between all four pairs range from 0.93 to 0.97. The top 50 most distressed counties overlap 38 to 50 across specifications. Rankings are highly correlated across weighting choices, and the core distressed counties (Tunica, Bolivar, Sunflower, Richmond) appear in the top 50 across all four formulas. Borderline zone assignments are more sensitive to weighting than the core rankings, which is expected.

Limitations

Cross-sectional snapshot. The CDI ranks counties relative to each other at a point in time. It does not tell you whether a specific county is getting better or worse over time. Annual refreshes will enable year-over-year comparisons starting in Year 2.

Data lag varies by indicator. BLS unemployment is roughly two months old; Census ACS housing data is from 2019–2023 (a five-year rolling average). The composite blends indicators of different vintages. All vintages are documented in the factor tables above.

Imputation. 1.9% of indicator-county values are imputed from state medians. Counties with imputed values are flagged in the underlying data but are not downweighted in the composite.

No direct mortgage delinquency signal. The most significant data gap, documented in the Coverage Gap section above. Mortgage distress is captured through proxies, not directly.

PCA captures variance, not welfare. The 47.5% weight on Consumer Credit Distress reflects the fact that debt indicators vary most across counties. This is a statistical observation, not a normative claim that credit distress matters more than structural poverty. Counties with deep poverty but limited credit access — Appalachian coal country, tribal reservations — may score lower than their lived experience warrants. Factor-level scores are reported alongside the composite specifically so researchers can examine each dimension separately.

Source concentration in Factor 1. Five of the six highest-weighted indicators in Consumer Credit Distress originate from the Equifax panel via Urban Institute. PCA correctly identifies them as one factor, but shared source methodology means shared measurement artifacts. Cross-validation against the independently sourced subprime credit indicator (FRED/Equifax, separate pipeline) and bankruptcy filing rate (U.S. Courts, no credit bureau involvement) confirms the factor structure is not an artifact of single-source measurement.

How the CDI Compares to Other County Indices

Several county-level indices exist. Each answers a different question, and the differences are not cosmetic. The CDI is the only one focused specifically on whether households are falling behind on their bills.

The Economic Innovation Group's Distressed Communities Index measures whether a local economy is producing jobs and income. The CDC's Social Vulnerability Index measures whether a population is exposed to environmental and social hazards. County Health Rankings measures health outcomes. None of them ask the question the CDI asks. A county can have strong job growth (low EIG distress) while its households drown in credit card debt (high CDI distress). The CDI captures that divergence by combining credit bureau panel data with federal housing, labor, and bankruptcy sources — a combination no other public index uses.

PCA also separates acute consumer credit distress (Factor 1) from chronic structural poverty (Factor 3). No competing index draws this distinction. Counties in Appalachian coal country score high on structural poverty but low on credit distress — not because they are less distressed, but because the credit system never reached deep enough to create delinquency data. Mississippi Delta counties score high on both. The CDI can tell these two types of distress apart; other indices blend them into a single number.

How the CDI Differs from the ADI

The CDI and the American Distress Index are two separate indexes that answer different questions. Both are produced by American Default Research, and both use the same five zone names and color system for consistency, but the methods and the thresholds are calibrated differently.

The ADI tells you how bad things are compared to how bad they have been. The CDI tells you where things are worst compared to everywhere else. The numbers are not interchangeable — a CDI score of 60 and an ADI score of 60 measure different things using different methods.

Frequently Asked Questions

How often is the CDI updated?

Annually, aligned to Census ACS and Urban Institute Debt in America release windows, which are the slowest-moving source inputs. Between annual refreshes, faster-cadence indicators (BLS unemployment, FRED subprime, BLS QCEW) may drive light mid-year updates if a specific component moves meaningfully.

Why are the weights unequal?

Because the data, run through principal component analysis, said so. Equal weights across five domains was one of the candidate formulas tested in the sensitivity analysis; PCA-proportional weights were chosen because they let the statistical structure of the 21 indicators determine which dimensions of distress vary most across counties. The 47.5% weight on Consumer Credit Distress reflects the fact that debt indicators vary more across counties than any other dimension — not a normative claim that credit distress is the most important form of hardship.

What does a score of 50 mean?

The national median. By construction of percentile-rank scoring, the median county scores approximately 50. Scores above 50 indicate more distress than the typical American county. Scores below 50 indicate less.

Can I compare CDI scores to ADI scores?

Not directly. A CDI score of 60 and an ADI score of 60 measure different things. The CDI's 60 means a county is more distressed than roughly 60% of counties right now. The ADI's 60 means national household distress is moderately above the 2015–2024 baseline. The zone names overlap for consistency, but the numbers are not interchangeable.

Is the raw data available?

Yes. Per-county scores and per-factor breakdowns are exposed on each county detail page under /counties/[state]/[county]/. A downloadable CSV of all 3,144 counties and full methodology PDF are linked from this page and from the county distress hub. All data is released under Creative Commons Attribution 4.0 — free to reuse with attribution to American Default Research.

How to Cite

The County Distress Index is an open data product released under Creative Commons Attribution 4.0. If you use it in research, policy analysis, or journalism, please cite American Default Research as the source. The named methodology author is listed inside the downloadable PDF for researchers who need to attach an individual byline.

APA

Chicago (author-date)

BibTeX

@techreport{cdi_methodology_2026,
  title       = {{County Distress Index: Methodology and Scoring}},
  author      = {{American Default Research}},
  year        = {2026},
  institution = {{American Default Research}},
  url         = {https://americandefault.org/methodology/cdi/},
  note        = {Principal component analysis across 21 indicators
                  in 5 statistically derived dimensions, covering
                  all 3,144 U.S. counties},
}

Journalist short form

Reproducibility

The scoring pipeline is implemented in Python and runs against publicly available source data. The production scoring engine is at scripts/county/compute_county_dci.py. The full factor analysis pipeline — imputation, winsorization, PCA, rotation, weight derivation, face validation, and sensitivity analysis — is in scripts/county/cdi_rebuild_analysis.py, with its output artifacts in docs/cdi_rebuild/ (correlation matrix, rotated loadings, weight derivation, sensitivity tests, face-validation results, all-county scores under four candidate formulas). Running the rebuild script reproduces every number on this page.

Questions about methodology or requests for the raw data extract should go to press@americandefault.org.