Cost of Living Calculator: Why Two Disagree by 40%

Second quarter of 2025, C2ER's Cost of Living Index put San Francisco at 160.1 against a national average of 100 — sixty percent above average. The Bureau of Economic Analysis, measuring 2024, put the San Francisco metro at 115.6. Same city, same year give or take, two federal-grade methodologies, and one says the premium is four times larger than the other does.

Manhattan is wilder. COLI: 232.0, more than double the national average. BEA's regional price parity for the New York metro: 112.6.

Neither number is wrong. I want to be clear about that up front, because the tempting conclusion — one of these is garbage — lets you keep whichever answer you liked better. Both are defensible measurements of different questions. When two calculators disagree by 40 points on your move, the useful response is not to average them. It is to find the input doing the work, and it is almost always housing. Here is how to run that diagnosis on your own pair of cities.

Four instruments wearing the same web form

Most calculator pages look identical: two dropdowns, a salary box, a percentage. Underneath there are only about four data sources in circulation, and the page rarely leads with which one it licensed.

C2ER's Cost of Living Index (COLI). Sixty-one items priced three times a year — the second Thursday through Saturday of January, April, and July — by chambers of commerce and economic development groups in each participating area. 245 urban areas participated in Q2 2025. NerdWallet's calculator, among others, runs on this data. The COLI methodology manual is public and worth twenty minutes; everything I quote from it below was read on 2 September 2026.

BEA regional price parities (RPPs). Federal estimates of price levels for every state and metro, built largely from CPI price samples and American Community Survey rents. Annual, free, published at bea.gov; the current release, 19 February 2026, covers 2024. The metro figures above came from BEA series on FRED, read 2 September 2026.

Crowdsourced sites. Numbeo is the big one. Its methodology page says prices come from user submissions plus some manually collected data weighted three times heavier, filtered for outliers, with entries in popular cities aged out after three months. Useful for a fast international sanity check. Not something to move on, because you cannot see who typed the rent figures or which neighborhoods they live in.

The CPI — which is not on this list, and gets mistaken for it. Metro CPIs are each pegged to 100 in their own base period. They measure how fast prices rose in one place, not whether the place is expensive. Two cities can be compared on CPI arithmetic and give you a confidently wrong answer; the Congressional Research Service's brief on geographic cost-of-living differences (updated 23 February 2026) explains why in three pages. If a calculator cites CPI for a place-to-place comparison, close the tab.

The index has a 2,400-square-foot new house inside it

Now the actual mechanism of the 40-point gap.

COLI's housing component carries a weight of .2836 — about 28% of the composite. And the manual is specific about what gets priced. The house is new construction, 2,400 square feet of living area on an 8,000-square-foot lot, in a subdivision "typical for professional and managerial households in top income quintile." That single item, the monthly principal-and-interest payment on that house at current mortgage rates, is 19.20% of the entire composite index — the heaviest-weighted line in the survey. The rental item is a 950-square-foot two-bedroom in a complex no more than ten years old, again suitable for a top-quintile couple. Read on 2 September 2026 from the 2026 manual.

BEA's rents input is close to the opposite: rents actually paid across the whole housing stock, drawn from ACS data — every tenant, every building age, including people nine years into a lease at a rate no new arrival will ever be offered.

So one index asks what does a newly built executive house cost here, and the other asks what is everyone currently paying. In a market with tight zoning and old stock, those two numbers diverge enormously — and then the divergence gets multiplied by the biggest weight in the formula. BEA's own 2024 state estimates show why housing is where indexes go to disagree: the all-items range across states runs from 86.9 (Arkansas) to 110.7 (California), while the housing-rents component alone runs from 54.2 in West Virginia to 155.0 in the District of Columbia. Groceries barely move between metros. Rents move by a factor of three. Whatever a calculator decides about its housing input, that decision is most of its answer.

The instinct is to hunt for the gap in groceries and gas. The component spreads above say not to bother: set two outputs side by side line by line and the non-housing categories net out to a few points each, in a table where housing alone can span a hundred. The disagreement sits in one row.

Manhattan is a borough; "New York" is a 20-county metro

The second lever is geography, and it explains the most theatrical gaps.

COLI's 232.0 is for Manhattan specifically — the manual has participants price their "urban area," and the New York boroughs report separately (Brooklyn posted 159.4 in the same quarter, Queens 150.6). BEA's 112.6 is the New York–Newark–Jersey City metro: tens of millions of people, most of whom do not pay Manhattan rent. Averaging Scarsdale with Newark is not an error. It is just an answer to a regional question being read as an answer to a neighborhood question.

This cuts the other way too. If you are pricing a modest suburb of an expensive metro, a metro-level calculator overstates your costs; the COLI figure for the central city overstates them more.

The check takes one minute: find the exact place label on the calculator's result page. If it says a metro name and you are moving to one town in it, the housing line is a regional average wearing your town's name.

Whose grocery cart, whose tax bill

Two more inputs move the answer less than housing but are worth knowing before you trust a number.

The household being priced. COLI's reference household has two college-educated spouses, at least one in an established salaried professional career, income in the top quintile for the area. Participants are instructed to sample only stores and neighborhoods such households actually use. If that is not your income band, the index is pricing a life adjacent to yours. RPPs, by contrast, average across everyone.

What is excluded. COLI measures consumer goods and services excluding taxes. State income tax, property tax, sales tax — none of it is in that 160.1 or that 78.8 (Tupelo, Mississippi, the cheapest area in Q2 2025). For a move between, say, a no-income-tax state and a high-property-tax one, the calculator is silent on exactly the lines that swing hardest, which is why the actual bill for a specific parcel is its own lookup and not a calculator output.

And underneath both: participation is voluntary. A place appears in COLI because a local chamber committed staff to pricing 61 items three times a year. If your candidate town is not among the 245, whatever the calculator shows you is a neighbor or an interpolation.

The five-question diagnostic sheet

Run this against any pair of calculator outputs before believing either. Each row is one lookup, and the sheet only works if you write the answers down — the third column is the cell to fill.

When two outputs disagree, the answer to question 3 differs between them nearly every time, occasionally question 2. In the San Francisco case above, both differ at once: a top-quintile new-construction basket in the urban core against an all-tenant average across the metro. A 45-point gap stops being mysterious the moment both definitions are on paper.

Swap in your own housing line and re-run the pair

The endpoint of this diagnosis is not picking a winner. It is replacing the input that caused the fight.

Pull the median gross rent, table B25064, for both candidate places at data.census.gov — city level, not metro, and note it is a 2020–2024 five-year average with the lags and error margins covered in the Census tables walkthrough. Then collect ten live listings in each place matching the unit you would actually rent and take the median asking rent. The first number tells you the market's floor; the second tells you your price of entry. Write both, with today's date, into the shelter line of the seven-line move sheet, and let the calculators keep arguing about a house you were never going to buy.