A facility count is not a pollution count

There is a number in our environment dataset that would make a very good chart and a very bad argument. It is the count of facilities in each state that report to the EPA Toxics Release Inventory. California has 5,110, the most of any state. The District of Columbia has 22, the fewest. Across the 51 jurisdictions the counts sum to 64,527. Put those on a map, shade the high states dark, and you have produced something that every reader will understand instantly and almost every reader will understand wrongly.

The file itself says so, in its header, before a single number appears: a facility count is industrial density, not environmental quality. I want to take that warning seriously rather than print it under a chart and carry on. So this piece has no chart. The decision is the argument.

What the number actually counts

It counts sites that report. Each facility contributes one, whatever it makes, however much it handles, whatever it released last year. A site reporting a trivial quantity and a site reporting an enormous one are indistinguishable in this dataset, because the dataset carries no quantity at all. There is no chemical, no volume, no sector code, no compliance record. There is a state and an integer.

Two things follow that a shaded map cannot express. The first is that the number has no upper bound in meaning: a state with twice the count has not necessarily released twice, or half, or any particular amount of anything. The second is that reporting is a legal category. A facility is in the inventory because it crosses a statutory trigger, so the count tracks the reach of the programme alongside the activity it observes. That is not a flaw in the data. It is what the data is.

What it is good for

Read as density rather than damage, it earns its place. Divide by the Vintage 2024 population and the ranking rearranges itself into something recognisable. Indiana runs at 37.1 facilities per 100,000 residents, Wisconsin at 36.4, Iowa at 34.2, Ohio at 31.9, Mississippi at 31.3, Arkansas at 29.8. At the other end sit the District of Columbia at 3.1, Hawaii at 6.7, Florida at 8.9, Maryland at 9.6, New York at 10.7.

California, which led the raw count with 5,110 facilities, comes in at 13.0 per 100,000, well below Indiana's 37.1. The state with the most industrial sites in absolute terms is not a dense industrial state; it is a very large one. That single inversion is the whole case for never printing the raw ranking on its own.

The bottom of the density table is not one phenomenon either. The District of Columbia's 3.1 per 100,000 is a city with federal offices in it, and the state metrics file does not even carry a manufacturing employment value for the District, recording the absence rather than imputing one. Hawaii's 6.7 is an island economy. Florida's 8.9 is a very large population with a manufacturing share of 4.3 percent. Three low numbers, three different reasons, and a shaded map would give all three the same pale colour and the same implied compliment.

The density reading is corroborated by a completely separate file. In July 2026, manufacturing employment was 15.7 percent of Indiana's total nonfarm employment and 15.1 percent of Wisconsin's, against 6.6 percent in California, 4.3 percent in Florida and 4.0 percent in New York. Two agencies, two collection methods, one picture: the reporting facilities are where the manufacturing is. That is a genuinely useful thing to know when you are assessing supplier depth or a labour pool, which is why the number is worth keeping.

The objection, and where it partly lands

The serious objection is that I am being precious. More industrial sites do, on average, mean more industrial emissions; a proxy does not have to be exact to be informative; the public understands perfectly well that a map of plants is a map of plants. If a policymaker uses the count to decide where to send an inspector rather than where to assign blame, the proxy has done honest work.

Part of that lands. As a screening device, a count is defensible, and I would rather a regulator started from industrial density than from nothing. But the objection assumes the count is used carefully, and the failure mode of a state ranking is that it is quoted without its denominator and without its caveat. California ranks first of 51 on the raw count and thirty-ninth on facilities per 100,000 residents. That gap is precisely the one a careless citation falls into.

There is a second complication I would rather raise myself. Even as a density measure the number is imperfect. Rhode Island sits at 37.1 facilities per 100,000 residents, the highest of the 51 jurisdictions and level with Indiana to one decimal place, and yet manufacturing was 7.8 percent of Rhode Island's nonfarm employment in July 2026 against Indiana's 15.7. A small population is a small denominator, and small states climb per-head rankings for reasons that have nothing to do with the thing being measured. The density cut is better than the raw count. It is not clean.

So the number stays in the dataset, labelled as what it is, and it does not get a chart in this piece. A reader who wants to know what a state releases needs release quantities, and we do not hold them. Publishing a striking graphic built on a count and hoping the caption does the work of the data is how a brief loses the only thing it has, which is that its readers can believe the pictures.