A county color summarizes a population. It does not tell you which residents have a condition, explain its cause, or give an individual’s risk. Start with the measure and its uncertainty before writing a headline.
A real example: the same counties, two different summaries
The maps below show diagnosed diabetes among adults in all 14 Massachusetts counties. They use the CDC PLACES 2025 release, whose diabetes estimates refer to 2023. This is a fixed release used for an example, not a claim to be the newest available data. The first map uses crude prevalence; the second uses age-adjusted prevalence. Their boundaries, extent and color scale are identical.
Watch Barnstable and Suffolk. Barnstable’s crude estimate is 10.7%, compared with Suffolk’s 9.3%. With age adjustment, the displayed order reverses: 7.1% for Barnstable and 10.1% for Suffolk. Changing the summary changes the comparison. This does not establish a cause, measure a treatment effect, or identify which residents have diabetes.
The crude intervals are 8.9–12.8% and 8.1–10.5%, respectively. Those ranges overlap, so the point estimates alone do not justify a confident difference claim. An overlap check is not a formal test of the difference; that requires an appropriate method and the dependence between estimates. A darker shade by itself is insufficient evidence.
Know what the estimate measures
CDC’s diabetes definition concerns adults who report having been told by a health professional that they have diabetes, excluding diabetes only during pregnancy. It is not a map of undiagnosed diabetes, newly diagnosed cases during one year, or a resident’s probability of developing diabetes next year.
PLACES produces local estimates using multilevel regression and poststratification, combining BRFSS responses with population and contextual information. These county values are model-based estimates, not counts of every resident’s medical records. The CDC methodology describes the inputs and simulation-based intervals.
Read the map with its data table
Values are percentages; intervals are CDC’s published 95% confidence intervals. Alphabetical order avoids presenting these individual measures as a ranking of overall county health. FIPS codes retain a reproducible join between estimates and boundaries.
| County FIPS | County | Crude % | Crude 95% CI | Age-adjusted % | Adjusted 95% CI |
|---|---|---|---|---|---|
| 25001 | Barnstable | 10.7 | ( 8.9, 12.8) | 7.1 | ( 6.0, 8.3) |
| 25003 | Berkshire | 10.9 | ( 9.2, 12.8) | 8.2 | ( 6.9, 9.5) |
| 25005 | Bristol | 10.8 | ( 9.2, 12.4) | 9.1 | ( 7.8, 10.5) |
| 25007 | Dukes | 10.6 | ( 8.8, 12.4) | 7.3 | ( 6.2, 8.6) |
| 25009 | Essex | 10.3 | ( 8.8, 11.8) | 8.7 | ( 7.5, 9.9) |
| 25011 | Franklin | 10.8 | ( 9.1, 12.7) | 8.1 | ( 6.7, 9.5) |
| 25013 | Hampden | 12.1 | (10.5, 13.8) | 10.5 | ( 9.0, 12.0) |
| 25015 | Hampshire | 8.2 | ( 7.0, 9.6) | 7.8 | ( 6.6, 9.1) |
| 25017 | Middlesex | 8.5 | ( 7.3, 9.6) | 7.7 | ( 6.6, 8.7) |
| 25019 | Nantucket | 8.6 | ( 7.2, 10.1) | 7.2 | ( 6.1, 8.5) |
| 25021 | Norfolk | 9.0 | ( 7.7, 10.5) | 7.7 | ( 6.6, 8.9) |
| 25023 | Plymouth | 10.3 | ( 8.9, 11.9) | 8.2 | ( 7.1, 9.5) |
| 25025 | Suffolk | 9.3 | ( 8.1, 10.5) | 10.1 | ( 8.8, 11.4) |
| 25027 | Worcester | 10.4 | ( 9.0, 11.9) | 9.0 | ( 7.7, 10.2) |
Why two county totals cannot describe individual exposure
Consider a separate, fictional example. County A has 200 recorded cases among 100,000 people; County B has 100 among 100,000. Suppose A also has a higher area-level exposure indicator. These totals do not say whether the people with the exposure are the people with the condition. Many different individual-level arrangements can produce those same county totals.
The ecological inference error occurs when the headline crosses that gap without evidence. “The area-level measures differ” describes the comparison. “Exposed residents are twice as likely to become ill” requires information linking exposure and outcomes for individuals, together with an appropriate study design. The fictional example is not a finding about Massachusetts or diabetes.
Five checks before publishing a county map
- Match the question to the measure. State geography, adult population, reference year, release and crude or age-adjusted basis. Preserve the denominator when comparing places.
- Keep the legend honest. Use the same scale for side-by-side maps. Document breakpoints if using bins. Keep missing estimates separate from zero, and provide a readable table.
- Read uncertainty beside the estimate. Include intervals, rather than letting a small color difference imply a precise ordering. Check population composition and the source’s comparison guidance.
- Keep inference at the supported level. A descriptive county pattern can guide investigation. It does not establish individual exposure, individual risk, or a causal mechanism.
- Respect source-specific limits. CDC says PLACES should not rank overall area health, track local trends under its current modeling procedure, or evaluate local policy/intervention effects. Individual measures can be compared with care. These restrictions belong to PLACES; other datasets require their own methods review.
For those limits and comparison cautions, consult the PLACES FAQs. For navigating geographic layers and legends, see CDC’s interactive-map guide.
A caption that keeps the claim clear
These maps show model-based diagnosed-diabetes prevalence among adults in Massachusetts counties, using 2023 estimates from the CDC PLACES 2025 release. Crude and age-adjusted values answer different comparison questions. The maps display point estimates; read the accompanying 95% intervals. County patterns do not determine individual risk or establish causation.
Sources and reproduction
- CDC county GIS-friendly dataset, 2025 release. Filter: StateAbbr = MA. Fields: CountyFIPS, CountyName, DIABETES_CrudePrev, DIABETES_Crude95CI, DIABETES_AdjPrev, DIABETES_Adj95CI. All 14 counties retained; no values recomputed.
- U.S. Census Bureau 2023 cartographic boundary files: cb_2023_us_county_500k.zip, filtered to STATEFP 25 and joined by GEOID/CountyFIPS. All 14 joins matched. Boundaries are generalized, not a survey or navigation product.
- CDC PLACES measure definitions, methodology, and FAQs.
Data and source pages retrieved 28 October 2026. Maps use an equirectangular display with longitude scaled by cos(42.2°) for this state-sized illustration. County colors reflect the published values; no interpolation or causal model was added. This worked example explains interpretation and is not a clinical assessment.
Related: Age-adjusted versus crude health rates and Uninsured rates by state.
For related comparison questions, see absolute versus relative risk and how boundaries and weighting change area comparisons.
Revision note, 29 October 2026: Added real county maps and a complete data table, replaced general background references with measure-specific sources, and shortened repeated explanations. WorldLensia uses AI assistance in publishing; no independent clinical review is claimed.