Research
How many police cars should there be?
Patrol consumes more than half of most police budgets. In 1975 RAND wrote a program to work out how many cars each command actually needed — and it ran in 240 kilobytes.
Everything else on this site is about individual police cars: how fast they went, what they cost, what broke. This one is about the question underneath all of them, which almost nobody asks.
How many should there be, and where?
It is not a small question. As the report puts it, although patrol operations are only part of police work, in most departments the patrol function consumes over half of the annual budget. Every argument about vehicle price, fuel economy and replacement mileage sits inside that number.
What it replaced
Before PCAM, departments allocated patrol cars using what were called hazard formulas or workload formulas — weighted scoring systems that took crime counts, population, calls and a few other factors and produced a share-out between precincts.
RAND’s assessment of them is one line long and quietly brutal. The model was intended to substitute for those formulas, which were still widely popular although their failings have been pointed out repeatedly.
Six separate references are cited for that sentence. The profession knew. It kept using them anyway.
What the program did
The Patrol Car Allocation Model was developed at the New York City-Rand Institute in 1975, and modernised as PCAM85 under a National Institute of Justice grant. Between 1975 and 1984 it was used by more than forty police departments and folded into the NIJ’s own Managing Patrol Operations programme.
It worked in two directions. Descriptively, it would tell you what your current allocation was actually producing. Prescriptively, it would tell you the allocation that best met whatever standard you set.
The questions it was built to answer are recognisably the ones police departments still argue about:
| The question | When it comes up |
|---|---|
| How many patrol officers should we have in total? | Budget preparation; moving officers between patrol and investigation |
| How do we split a fixed number between commands? | Reallocation between precincts |
| What time should tours start? | Matching officers on duty to when calls actually arrive |
| Where should an overlay tour go? | A fourth shift straddling two others — PCAM’s unique capability |
The example the report gives for setting a standard is the one every citizen would recognise: keep response time under four minutes in every precinct. Set that, and PCAM tells you how many additional officers you need to meet it.
The number that should be better known
Buried in the data requirements section is a fact about patrol work that reframes the whole subject.
Patrol cars are typically busy on activities other than answering calls for service somewhere between 35 and 60 percent of their working hours.
Meals, reports, court, vehicle maintenance, administrative tasks, self-initiated work. Between a third and nearly two-thirds of a patrol car’s shift is not spent responding to anything.
And RAND says obtaining accurate data on that time may be the most difficult task facing anyone using the model. The largest single component of what a police car does was the part departments could least reliably measure.
What it honestly could not do
The most impressive part of this document is its list of limitations, which is longer and more specific than its list of capabilities.
PCAM could not model dispatching cars across command boundaries for high-priority calls. It could not model holding low-priority calls in a queue to wait for the local beat car when others were free. It could not model holding cars in reserve for high-priority work. And it could not model preempting a car off a low-priority call to send it to an urgent one.
Those are four of the most common things a real dispatcher does.
Only simulation models could handle them, RAND says — and then explains why it did not build one. Simulations are more accurate, but they are substantially more expensive to run, require much more detailed input data, and most police departments would need outside assistance to use one.
That is a research team choosing the tool a department could actually operate over the one that would model reality best. It is the same instinct that runs through the whole of this archive — Michigan converting performance into a dollar figure a purchasing office can act on, rather than a verdict nobody can use.
240 kilobytes
The resources section is a period piece worth preserving.
PCAM ran on any computer with a FORTRAN compiler and at least 240K bytes of core storage. Installed for batch operation, nobody at the department needed to understand FORTRAN at all.
| Operation | Cost |
|---|---|
| Compiling the program | about $14 |
| A realistic sample session | under $12 |
| Fairly complex calculations | under $20 |
Twelve dollars of mainframe time to answer a question about a budget line consuming more than half of a police department’s money. The report notes drily that compiling costs more than most runs, so it is desirable to save the object code.
And the distribution method: write to Warren Walker or Jan Chaiken, and they would send you the program on cards or tape.
Why it belongs here, just about
This is a deployment document rather than a vehicle document, and it earns its place on one connection.
Every other study in this archive treats the police car as a thing to be specified, tested, priced and replaced. This one treats it as a unit of capacity — something that is either available or busy, and whose number determines how long a citizen waits. The 1973 economics study worked out what one costs. Fleet managers worked out when to replace it. Nobody in either document asks how many there should be in the first place, and that decision governs both.
One last detail, for anyone who assumes this was a purely American conversation. Among the model’s references is a 1974 paper in the Journal of Criminal Justice by McEwen and Larson: Patrol Planning in the Rotterdam Police Department.