# Method

New York City's expense budget assigns money to about 140 agencies. Three of
those agencies do not run anything. Agency 095 holds the city's pension
contributions, agency 098 holds health insurance, payroll taxes and legal
payouts, agency 099 holds debt service. Between them they hold about $31
billion of a $118 billion budget.

Every published chart of the city budget inherits that structure, which is why
they all show the Police Department as a $6.3 billion line with no pension
contribution and no health insurance in it. This page moves the pooled money
back to the agencies that incur it, and says for every dollar how confident
that assignment is.

## Sources

| What | Dataset | Notes |
|---|---|---|
| Expense budget by agency, unit, object code and position | [`mwzb-yiwb`](https://data.cityofnewyork.us/City-Government/Expense-Budget/mwzb-yiwb) | Office of Management and Budget |
| Settled claims against the city, by agency | [`ex6k-ym48`](https://data.cityofnewyork.us/City-Government/Claims-Report-Underlying-Settlements-and-Claims-Fi/ex6k-ym48) | Comptroller's claims report |
| Population denominator | Census Bureau vintage 2025 estimate, 8,584,629 | Same denominator as The New York City ledger; held flat across the series, so per-resident figures for earlier years are dollars against today's population |

The dataset carries fiscal 2017 through 2027, and the page builds all eleven.
Each year has to pass every check on its own before any of it is written.

Two features of the dataset complicate a multi-year build.
**Agency numbers lose their leading zero before fiscal 2027** — the pension
agency is `095` in one year and `95` in the next, and Education is `040` or
`40`. A filter that does not pad silently matches nothing, and the year builds
with an empty pension pool rather than failing. Object class numbers do the
same, `01` against `1`. The build pads both, and refuses to write a year that
comes back without all three central agencies.

The line names, by contrast, are stable: every pension fund, every benefits
object code, and the uniformed and pedagogical payroll splits carry identical
strings in all eleven budgets, which is what makes the series comparable.

The budget dataset also keeps every publication snapshot of a fiscal year as its own
set of rows — a year is republished two or three times as it moves from
preliminary to executive to adopted. Grouping on `fiscal_year` alone therefore
sums the snapshots and overstates the year by two or three times. The build
takes the newest `publication_date` for the year and records which one it used
in `data.json`.

## Pensions: named in the budget

The pension agency writes its payment to each retirement system on its own line,
and four of the five systems belong to a single workforce:

| Budget line | Goes to | Fiscal 2027 |
|---|---|---|
| Teachers' Retirement System (two lines) | Department of Education | $3.62B |
| Board of Education Retirement System | Department of Education | $0.24B |
| Police Actuarial Pension Fund | Police Department | $2.54B |
| Fire Actuarial Pension Fund | Fire Department | $1.63B |
| Contingent Reserve Fund | shared — see below | $2.32B |
| Non-city pensions | left unassigned | $0.11B |

$8.0 billion of the $10.5 billion pension bill is assigned exactly as the budget
writes it. Nothing is modelled.

The contingent reserve fund is the city's payment to the New York City
Employees' Retirement System, which covers nearly everyone the other four funds
do not. It is divided across agencies by payroll, after removing the pay that
belongs to the four named funds. That removal uses the budget's own object
codes, not an estimate: `FULL TIME UNIFORMED PERSONNEL` at the police and fire
departments, and all Education payroll, come out of the base. What is left at
the police and fire departments is their civilian and emergency medical staff,
who are in fact employees' retirement system members, so they get a share.

**Known limitation.** Pension contributions are actuarially determined per
fund, not proportional to payroll, and the employer rate differs sharply
between funds — the fire fund's rate runs near 100% of covered payroll while
the employees' system runs near a quarter of it. Sharing *within* the
employees' system by payroll is reasonable because one rate applies across it.
Sharing *between* funds by payroll would not be, and is not done here.

## Health insurance and payroll taxes: modelled, then checked

The miscellaneous agency's fringe benefits unit holds $9.39 billion. It is split
three ways.

**Held out.** The retiree health benefits trust, budget code 3006, is $2.94
billion. It pays for people who no longer work for the city. It is not charged
to any current agency.

**Divided by budgeted positions, $4.74 billion.** Health insurance, welfare fund
contributions, workers' compensation, unemployment and uniform allowances follow
people rather than salaries, so they are divided by each agency's budgeted
positions.

**Divided by payroll, $1.70 billion.** Payroll taxes track wages.

The Department of Education and the City University buy their own health
insurance and pay their own payroll taxes out of their own budget lines. They
are detected by rate rather than by name — an agency spending more than $5,000 a
position on health insurance, or more than 3% of payroll on payroll taxes, is
taken out of the corresponding pool rather than charged twice.

**The check.** This is the only modelled step on the page, and it can be tested
against agencies that buy the same coverage directly. The pooled headcount rate works out to **$29,895 a budgeted
position**. Education, buying the same coverage directly, spends **$28,270** a
position on the same components, and the City University **$32,576**. Three
independent numbers within about a tenth of each other. The build fails if any
self-funding agency’s rate falls outside 0.55 to 1.75 times the pooled rate, in any year.

**Known limitation.** Workers' compensation is not evenly distributed — sanitation,
police and fire generate far more of it per employee than an office agency does,
and dividing it by headcount understates them. It is $616 million of a $9.4
billion pool.

## Judgments and claims: followed to the agency that got sued

The budget provides one line, $823 million in fiscal 2027, for judgments and
claims across the whole city. The comptroller's claims report names the agency
in every settled claim. The line is split on each agency's share of settlement
dollars over fiscal 2021 to 2023, the most recent three years published.

95.5% of settlement dollars name an agency that also appears in the budget. The
rest stays unassigned rather than being spread around. The build fails below
85%.

The police share is 35.8%, transportation 13.6%, sanitation 10.4%, education
10.0%, health and hospitals 9.3%.

**Known limitation.** The judgments line is budgeted well below what the city has
recently paid. Actual settlements ran about $1.4 billion in fiscal 2025. Using
the budget's own figure keeps the page internally consistent, but it means the
payout component here is conservative for every agency.

## What is deliberately not assigned

$13.3 billion, 11% of the budget, stays where the budget puts it.

- **Debt service, $4.87 billion.** It could be approximated by tracing which
  function the capital was built for, but that attribution would not be
  supportable from the budget itself. It falls due against capital budgets of
  past years.
- **Retiree health care, $2.94 billion.** Uniformed workers can retire after 20
  or 25 years of service and are covered for longer than other city employees,
  so holding this aside lowers the police and fire figures relative to the rest.
- **Reserves and contingency, $1.58 billion.** Not yet assigned to anything.
- **Transit subsidies, $1.30 billion.** The city pays the Metropolitan
  Transportation Authority; it does not operate it.
- **State building aid, $1.20 billion.** A pass-through tied to school
  construction debt.
- **Collective bargaining reserve, $673 million.** Raises not yet written into
  any agency's payroll line.
- **Other central charges, $719 million.**

Capital spending is not on this page at all. This is the expense budget, the
same thing the published charts show. The city spent about $15.6 billion on
capital in fiscal 2025 on top of it.

## What the series shows

Across eleven budgets the multiples move very little. The Police Department has
been understated by 77 to 93 percent in every year since fiscal 2017. The Fire
Department has run between 1.85 and 2.06 times its published line. Education has
drifted down, 1.17 to 1.11, as it moved more of its own benefits onto its own
budget line. Health and Hospitals fell from 1.18 to 1.05 as its staff moved off
the city payroll.

The pooled share of the whole budget has been close to flat: 16.2% of the budget
sat in the three central accounts in fiscal 2017 and 14.9% does in fiscal 2027.

One caveat applies to the series specifically. The retiree health benefits trust swings
hard from year to year — $1.57 billion in fiscal 2018, $269 million in fiscal
2022, $2.66 billion in fiscal 2027 — because the city funds and draws on the
trust unevenly. That money sits in the unassigned pile in every year, so it moves
the "belongs to nobody" total around without touching any agency's multiple.

## What reconciles

The build fails rather than writes if any of these do not hold:

1. Agency totals sum to the citywide total.
2. Pension object codes sum to the pension agency's total.
3. Miscellaneous detail sums to the miscellaneous agency's total.
4. Each pool's allocations plus its unassigned remainder equal the pool.
5. Fully loaded agency costs plus unassigned central charges equal the citywide
   total, within the rounding on individual lines. Drift runs $0 to $8 a year
   across roughly 150 rounded rows.
6. Every self-funding agency's own benefits rate lands within 0.55 to 1.75 times
   the pooled rate, in every year.
7. At least 85% of settlement dollars map to a budget agency.
8. Every year returns all three central agencies, which is what catches the
   agency-number padding if the dataset's formatting changes again.

## What the figures do not say

A larger figure here carries no argument about whether an agency should cost
more or less. What the arithmetic establishes is narrower, and it is about the
chart rather than the policy: the number printed next to an agency's name is not
that agency's cost, the size of the gap differs by agency, and so agencies
cannot be compared against one another on published lines.

A low multiple is not a judgment either. Agencies that move least — homeless
services at 1.03, education at 1.11 — mostly spend on contracts, grants and
benefit payments rather than on staff. The multiple measures how much of an
agency's work is done by people on the city payroll.

## Rebuilding

    python3 build/build.py              # every year the dataset carries
    python3 build/build.py 2024 2027    # a range
    python3 build/build.py 2022         # a single year

The script writes `data.json` only if every check passes. Bump the `?v=` stamp
on the asset links in `index.html` when deploying, so a cached copy of one file
is never paired with a fresh copy of another.