Police and fire take about twenty-eight cents of every dollar a large American city spends. That number is smaller than the one in the budget speech, larger than it was in 1967, and almost unrelated to how much crime the city has.
In the fiscal year ending in 2022, the median large American city spent 27.8% of its general expenditure on police and fire. Across the 327 cities with a complete record, covering 97 million residents, that came to $42.7 billion of policing and $23.3 billion of fire protection.
The spread is wide. A tenth of cities spend under 12%, a tenth spend over 41%, and the extremes run from 3.2% to 55.4%. Police outspend fire by about 1.7 to one almost everywhere.
If you have heard a different number, you have probably heard the general fund share. City budget messages quote police and fire against the general fund, which excludes airports, water systems, grants and capital projects. For the same cities in the same year, that framing gives a median of 48% where the Census basis gives 23%.
| City | Share of its general fund | Share of all general spending |
|---|---|---|
| Columbus | 71% | 30.8% |
| Houston | 70% | 28.1% |
| Phoenix | 70% | 30.1% |
| San Antonio | 67% | 28.3% |
Neither number is wrong. The first answers what competes at the council table; the second answers what the city government actually spends. Only the second means the same thing in every city, because each city decides for itself what lives in its general fund.
Take the 131 cities with complete records in every decade since 1972 and follow them. Their median share of spending rose from 21.8% to 25.5%, a change of 3.7 points in fifty years. Over the same span, what they spent per resident, in constant dollars, went from $390 to $778.
How much that last figure means depends on which prices you deflate by, and the choice is not innocent. Against consumer prices it is a 1.99-fold rise. Against the cost of what governments actually buy, which is mostly labour and has risen faster, it is 1.62-fold — from $466 to $754. The second is the more conservative reading, and the better answer to whether cities bought more policing and fire protection. Both are real; neither is flat.
These 131 cities are also not a cross-section of the country. Surviving fifty years of continuous reporting selects for the older and the larger: 48% of them are Sunbelt against 74% of the cities left out, and their median population is 263,094 against 125,587. The cities that drop out spend 29.0% of their budgets today against 25.5% for the panel. This is a trend for big established cities, which is the only group anyone can follow this far back.
Staffing shows the same thing from the other side. The median city employed 1.57 sworn officers per 1,000 residents in 2022, down from 1.70 a decade earlier, while pay per officer rose from $107,788 to $110,977 in constant dollars. Firefighters run about 1.24 per 1,000, and sixteen of these cities have no fire department at all: an independent district serves them.
Read that pay figure narrowly. It is March cash payroll annualised, so it excludes employer pension contributions, health premiums and retiree benefits — which is where public safety compensation has actually been growing. Flat cash pay is not flat compensation, and this dataset cannot see the difference.
Group the plain municipalities into four equal bands by violent crime rate and the staffing pattern is clear and monotonic. The safest quarter of cities employ 1.25 officers per 1,000; the most dangerous quarter employ 2.02, sixty per cent more.
Spending per resident follows more weakly, and the budget share does not follow at all.
| Relationship, across plain municipalities in 2022 | Correlation |
|---|---|
| Officers per 1,000 residents and the violent crime rate | +0.51 |
| Officer pay and the metro price level | +0.74 |
| Spending per resident and the violent crime rate | +0.28 |
| Spending per resident and the metro price level | +0.36 |
| Budget share and the violent crime rate | -0.03 |
| Budget share and city population | -0.10 |
That first relationship is worth pressing on, because Census models a third of the staffing rather than counting it. Splitting the cities by how their staffing figure was obtained: the relationship survives in every group, and it is weakest in exactly the group where Census estimated the number outright.
| Staffing figure | Cities | Correlation with violent crime | 95% interval |
|---|---|---|---|
| All cities, as published | 224 | +0.51 | +0.40 to +0.60 |
| Reported by the department | 138 | +0.57 | +0.44 to +0.67 |
| Imputed by Census | 59 | +0.32 | +0.07 to +0.54 |
| Prorated from a reported total | 20 | +0.72 | +0.41 to +0.88 |
That is the pattern measurement error produces: an estimated value is smoothed toward a model, which blurs the relationship rather than inventing one. So the modelled cities are diluting this finding, not creating it. The reported-only figure is not the better number either — its interval overlaps the published one almost entirely, and cities that answer surveys are not a random sample of cities. Prorated is kept separate from imputed because it is a different failure — the department's total was reported and only the split between officers and civilians was estimated — though at twenty cities its interval spans almost the whole range and it settles nothing on its own.
The strongest relationship in the dataset is not about crime at all: officer pay tracks the local price level at +0.74. Cities pay what their housing markets force them to pay. Crime buys headcount; geography sets the price of each head; and the share of the budget is decided by something else entirely, which is mostly what else the city government happens to run.
The same city produces very different numbers depending on what its books contain. New York spends 6.6% of general expenditure on police and fire, or 13.1% once schools, welfare, hospitals and the airports are taken out. San Francisco moves from 8.7% to 16.4% on the same adjustment.
| City | Of general spending | Of core municipal spending | Of all spending |
|---|---|---|---|
| New York | 6.6% | 13.1% | 6.0% |
| Philadelphia | 11.7% | 17.4% | 10.7% |
| San Francisco | 8.7% | 16.4% | 7.3% |
| Houston | 28.1% | 33.3% | 25.5% |
| Phoenix | 30.1% | 38.7% | 22.3% |
| Plano | 28.1% | 28.3% | 23.6% |
Nothing about policing changed between those columns. What changed is whether a school system, a public hospital, a county jail or an electric utility sits inside the same government. Any ranking of cities that does not say which denominator it used is comparing organizational charts.
Every figure comes from the Census Bureau's individual unit finance files, parsed directly rather than taken from a published table: police is item code 62, fire is item code 24. The parse reproduces the Bureau's own published state totals exactly in the 2012 census year and to within a hundredth of a per cent in 2017. Staffing comes from the Bureau's employment survey and agrees with the FBI's independent count within 10% for 82% of the city-years the Bureau marks as reported. Constant dollars use CPI-U; regional prices use the Bureau of Economic Analysis parities.
The caveats above are not boilerplate. They are the residue of a validation that compared this dataset against five other sources and 29 cities' audited financial reports, and they are documented city by city, with the cities where the numbers do not reconcile named.