Reading the numbers

Reported crime counts what somebody told the police and the police wrote down. Two towns with identical experiences can post very different rates if one has a department that records diligently and the other does not, or if residents in one are likelier to call at all.

The larger problem is participation. Departments submit returns to the FBI voluntarily, and many submit for part of a year or not at all. A rate built on nine months of reporting, annualized, looks like a real number and is not one. That is why nothing appears here unless a department reported all twelve months — a rule that withheld 1,546 department-years.

Watch the denominator too. A city that draws a large daytime workforce is policing far more people than live there, and every incident involving a commuter or a visitor lands in a rate divided by the residential population.

What it does answer: how much crime is being recorded here, relative to places recording theirs the same way.

A death certificate is filed whether or not the local police department submitted anything that year, which makes homicide the most consistently counted violent outcome in the country. When a crime rate and a homicide rate disagree sharply, the crime rate is usually the one bending.

The trade is resolution. These are published by county and nowhere finer, and a county can hold a dense city center and a quiet suburb at once. Fulton County records 13.7 homicide deaths per 100,000 and contains both downtown Atlanta and Johns Creek — a figure that describes neither. That is the whole reason it is shown here and never folded into a score.

Two more limits. These are crude rates, deaths divided by population and not adjusted for age, from provisional death certificate records. And where a county recorded too few deaths for the CDC to publish the count (fewer than ten, or a range up to fifty), the count is suppressed at source and the CDC publishes a modeled rate instead, one that borrows stability from neighboring counties. Those figures are marked “modeled, not measured” on the page, and they are best read as an order of magnitude: a rate built on single figures moves sharply from one release to the next.

Firearm deaths count every intent together, and nationally most are suicides rather than homicides, so the firearm figure and the homicide figure count some of the same deaths. They overlap and should not be added together.

Every state writes its own test and sets its own passing mark. A state that sets a demanding bar produces lower proficiency rates than one that does not, and the difference between them measures the exam rather than the schools. Comparing 62% in Georgia against 62% in Massachusetts is comparing two unrelated things.

So proficiency is ranked within its own state here. That is the only comparison the data supports, and it is also the comparison a family actually faces — they are choosing between schools in one state, not across fifty.

Two smaller traps, both of which changed figures on this site. Fully virtual schools enroll students from across a state while reporting one administrative address; counting them gives a town a student body it does not have. And results for small groups are published as a range rather than a number — a school reported as “50 to 100 percent” arrives as its midpoint, 75, which is not a measurement of anything. One Oregon town read as the best in the state on six tested fourth-graders before that was caught.

A dash in a proficiency column is an absence in the federal file, not a bad result, and it has three causes. Some schools have no tested grades, a kindergarten-to-second-grade school for one. The release publishes a range rather than a rate, and any range wider than twenty percentage points is dropped here rather than guessed at. And a school with fewer than ten tested students is suppressed at source.

Schools are placed by ZIP code, so a profile shows which schools are in a place rather than which one a particular address is zoned for. District boundaries do not follow municipal ones, and enrollment zones change. The school directory is current, from 2024–25; the proficiency figures are from 2018–19, the last full year of state testing before the pandemic, so they describe how students tested some years ago rather than now.

What it does not answer: what a school adds. Proficiency tracks the circumstances students arrive with at least as strongly as it tracks teaching, which is why free-lunch eligibility is shown beside it and never scored.

The census figure is what tenants are currently paying, averaged across everyone in place — including households that signed a lease a decade ago and have never moved. It is an accurate description of what the existing tenancy stock costs.

Fair market rent is HUD’s estimate of the 40th percentile of what recent movers pay. It is an administrative standard, set for housing assistance programs rather than surveyed from listings, and it is the closest published figure to what signing a lease today costs. The two can sit a long way apart: in Burbank, CA a new lease is benchmarked at $3,900 against $2,192 for tenancies already in place, a difference of 78%.

Which is larger is not fixed, and that is the useful part. Where the fair market rent runs above the census figure, the market has moved away from people already in place and a new arrival pays the difference. Where it runs below — which happens, and in some cities the two are within a dollar of each other — sitting tenants are not being protected by an old lease, and the headline rent is closer to what you would actually sign. If you are moving, read the fair market rent; the gap between the two tells you which of those situations you are walking into.

The rent share and the price-to-income ratio both come from the same American Community Survey release as the income they are set against, so the comparison is like for like. They describe housing cost against what people nearby earn. Transport, groceries, healthcare, insurance and taxes are not in them, and a place can look affordable on these measures and still be expensive to live in for reasons they do not capture.

The index compares what a dollar buys against a national average of 100. It is weighted the way the country as a whole consumes, not the way you do, and the spread between places is driven almost entirely by housing — groceries and goods vary far less than the headline suggests.

It also excludes taxes, which frequently run in the opposite direction. A state with a low price level and a high income tax may leave you no better off, and the index will not tell you that.

Used properly it is a conversion rate: divide a salary by it to compare offers in different cities in the same currency. Used improperly it is a claim about your monthly outgoings, which it is not.

The temperature and precipitation figures are thirty-year normals from the nearest weather station. They describe what a typical year looks like, not what the weather is doing today, and the distance to the station is stated because a station across a ridge or a bay is not always the place you mean.

Air quality is measured rather than modeled, by EPA monitors, and published by county only where a monitor sits. 682 of the 1,297 counties published here have one; the rest show nothing rather than an estimate. Air masses are regional, so a county reading describes the air over a place reasonably well, but it is still the county reading.

The median hides the tail. Wildfire smoke and summer ozone show up in the worst tenth of days, not in the middle of the year, which is why the worst-tenth figure is shown beside it.

This is the question that catches people most often, because the answer is rarely stated. A figure printed on a page about a city may have been measured for the county around it, the metropolitan area, or the city itself, and the three can differ enormously.

The rule worth carrying: the wider the geography, the more it averages away exactly the difference you are trying to see. County crime tells you little about which suburb to pick. Metropolitan price levels are genuinely regional, because people shop and commute across a metro, so there the wider figure is the right one. Every measure here states which geography it was measured on, and inherited figures say so explicitly.

Where a figure is missing here, it is missing for a reason that is stated: no department reports for this town, no air monitor sits in this county, too few students sat the test. That is more honest than an estimate, and more useful — a town with no municipal police department is telling you something real about itself.

Be wary of any source that has a number for everywhere. Complete coverage of a patchy federal record means somebody filled the gaps, and filling gaps is where confident-sounding fiction comes from.