Why the estimates are not counts
When a page says that 31.8% of adults in a ZIP Code have obesity, it is easy to read as "someone counted". Nobody did. The figure is the share the CDC model expects given who lives there. This guide lists what that changes.
Nobody in the area was necessarily asked
The model is trained on the national survey and knows a respondent's county at most. The estimate for a ZIP Code, place or tract is built from the Census profile of the people living there. The same is true of a county estimate, though for large counties the survey usually has respondents from that county and the county effect in the model reflects them.
Self-report, not diagnosis
Every measure is defined by a survey answer: "Have you ever been told by a doctor ... that you have diabetes?", height and weight as reported, days of poor mental health in the past month. CDC's measure notes record the consequences: self-reported height and weight understate obesity; undiagnosed high blood pressure is not captured; arthritis is not confirmed by a clinician. The measure pages quote those notes.
A percentage, not a number of people
Multiplying a prevalence by the adult population gives a rough count, but the interval multiplies too. 31.8% of 14,000 adults with an interval of 29.0–34.7% is somewhere between about 4,060 and 4,860 people, if the model is right for that area. This site does not print such counts.
Estimates for the whole population only
CDC: "There is one estimate per measure for the entire population of each county, place, census tract or ZCTA. Stratified estimates by age, sex, race/ethnicity, or poverty are not available." And: "We do not produce estimates for individuals". A resident's own probability of having any condition is not what the number says.
Fixed population base
ZIP Code, place and tract estimates are post-stratified on the 2020 Census (2010 for releases before 2024). Population change since then does not enter the model, which is one reason CDC says releases cannot be used to track local change.