Assessment Regressivity

The Simple Version

Before any charts or jargon, here is what is actually happening.

The city decides how much your house is "worth." An assessor sets that number. Whatever number they set is what you pay taxes on.

Elmira's assessor has not really updated those numbers in a long time. But houses keep selling. When one sells, we find out what it is truly worth, because a real buyer paid real money for it. So we can check the assessor's homework. What did the assessor say the house was worth? What did someone actually pay?

Do that for every city sale and the first thing you see is not which way the errors go. It is how big they are. A typical Elmira assessment is off by about half. Two near identical houses on the same street can be taxed as if one were worth nearly three times the other. We show you exactly how that happens further down.

The second thing you see is that the errors are not random. Cheap houses sell for $30,000 or $40,000, and the assessor's number is usually higher than that. Those owners pay tax on value that is not there. Expensive houses sell for far more than the assessor's number. Those owners get a discount.

Here is the result. A family whose house would sell for under $50,000 pays about twice as much tax per dollar of what their home is worth as a family whose house would sell for $150,000. Same tax system. Very different deal.

Why does this happen? Houses in the better neighborhoods gained value over the years, and the assessor never updated to match. Houses in the struggling neighborhoods did not gain value, so their old numbers stayed about right. Every year the roll sits still, the gap grows.

The J-Curve

Sort the city's sales by price, and a shape appears. Cheap homes are assessed too high. Mid-range homes are assessed far too low. The very top comes partway back. The line starts high, dips, then curls up, like a J lying on its side.

Everything here rests on one number, so it is worth naming it plainly. The assessment ratio is a home's assessed value divided by what it sold for. A ratio of 1.0 means the assessor got it right. Above 1.0 means the home is assessed for more than a buyer would pay. Below 1.0 means it is assessed for less.

Elmira's curve starts at 1.30. Homes that sold for under $40,000 were assessed at 130% of what the buyer paid. These are the houses in the hardest-hit neighborhoods. Their owners have the least time, money and standing to challenge an assessment, and they are the ones paying tax on value that is not there.

The low point is 0.42, in the $130,000–$175,000 band. Those homes are assessed at well under half what they sell for. That is what a frozen roll does. The market climbed for thirty years and the roll did not follow, so the homes that gained the most value are the ones now taxed on the smallest share of it.

Researchers find this same curve in cities across the country that have gone decades without reassessing. It is not an Elmira quirk.

The J-Curve — Assessment Ratio vs. Sale Price Gold is the median ratio actually measured in each price band, and the shaded area around it holds the middle half of sales. Blue dashed is the same figures once the sorting tilt is divided out (see below). The flat dotted line at 1.0 is a fair assessment: above it homes are assessed too high, below it too low. Every band holds at least 90 sales. Hover any point for the exact ratio.

An honest caution about this chart. We sorted these homes by what they sold for. But the sale price is also the bottom half of the ratio. A house that happened to sell a bit under its true worth lands in a cheaper group and shows a higher ratio. So the line would tilt downward a little even if every assessment in the city were perfectly fair.

Sorting by assessed value instead does not fix that. It flips it, because assessed value is the top half of the ratio. So the real test is not "is the line flat." It is "is the line steeper than the tilt alone would make it."

We measured that tilt rather than guessing. In this data, 510 properties sold twice within three years. Comparing the two prices shows how far a single sale strays from what a home is really worth. Simulate a city with that much noise and no unfairness at all, and the tilt it produces is small — nothing like the curve we actually see. The bias-corrected line is what is left after that tilt is divided out, and it still runs 1.20 down to 0.47 and back up to 0.66.

Why it happens. When a city stops reassessing, assessments drift away from real values at different speeds in different neighborhoods. Areas that gained value fall further below market every year. Areas that declined stay near their old numbers, or above them. Tax burden slides from the neighborhoods that gained to the ones that lost. In Elmira, that means it slides from wealthier households to poorer ones.

Measured Against the Industry's Own Standard

Assessors have a professional body, the International Association of Assessing Officers, and it publishes hard limits for how far a tax roll may stray. Those limits are not our invention. Elmira is outside every one of them.

47.9 Coefficient of dispersion
how far a typical assessment misses, in %. IAAO limit: 15

1.276 Price-related differential
above 1.03 means regressive. IAAO range: 0.98–1.03

−0.572 Price-related bias
ratio drop per doubling of value. IAAO range: ±0.05

2.0× Tax burden gap — cheap vs. expensive
per dollar of actual market value

What a "coefficient of dispersion" actually is

It is the first number above, it is the one that matters most, and the name tells you nothing. So here it is in plain terms.

Take every sale in the city. For each one, divide the assessor's number by what the house actually sold for. That fraction is the home's assessment ratio. If the assessor said $63,000 and the house sold for $100,000, the ratio is 0.63.

Now line up all 1,689 of those ratios and find the middle one. Then ask: how far from that middle does a typical home sit? That distance, written as a percentage, is the coefficient of dispersion.

It measures consistency, not who is favoured. A city can have a bad COD while treating rich and poor exactly alike. It only means the assessor is missing, in every direction, by a lot.

Elmira's is 47.9. The IAAO asks for 15 or less on single-family homes. Here is what 47.9 looks like on your street.

Two houses. Both sell for $100,000. The typical Elmira home is assessed at $63,100 — about 63% of its sale price. But "typical" error here is 30 points of ratio. So one of these houses is assessed at $32,900 and the other at $93,300.

Neither one is unusual. That is the point. Both are ordinary misses for a roll with this coefficient of dispersion. And the second owner pays 2.8 times the property tax of the first, on a house worth exactly the same.

No judgement about rich or poor has entered yet. This is only about the roll being inconsistent. The next two numbers are where direction comes in.

The other two numbers: which way the errors lean

The price-related differential and the price-related bias both answer one question. When the assessor misses, does he miss in a way that favours expensive homes? A PRD above 1.03 says yes. A PRB below −0.05 says yes. Elmira posts 1.276 and −0.572.

These hold up however we slice the data. The 2023–25 sales on their own give 1.288 and −0.616.

One check that the method tracks reality: the median assessment ratio for 2023–25 city sales is 0.500. The state's published equalization rate for Elmira is 56%. Two different instruments, different data, nearly the same answer.

A second check, and a stronger one: the state runs these same statistics itself, and has published them every year since 2004. ORPTS measures every New York municipality's roll and reports a coefficient of dispersion and a price-related differential for houses specifically. Outside regulator, its own data, its own method. For 2024 it puts Elmira's COD at 46.2 and its PRD at 1.26. We get 47.9 and 1.276 from sales. Two independent measurements, effectively one answer.

The State's Own Uniformity Measure — Elmira Residential COD, 2004–2025 How far a typical assessment misses, in percent. The IAAO asks for 15 or better on single-family homes. Elmira met that standard in 2006, a decade after its last revaluation, and has failed it by a widening margin every year since. The 2022 dip is a measurement artefact: that year alone was scored with a computer model rather than a sales study.

Two things in that chart are worth saying out loud. The first is that Elmira's roll used to be fine. In 2006 its COD was 12.9, inside the professional standard. Nothing was done to break it. This is simply what happens to a roll left alone while a housing market moves underneath it.

The second is the figure the state published for 2025: a residential assessment ratio of 52.08%, against the 56% equalization rate published for the whole roll. Those are different numbers and the gap is the point. Houses in Elmira are assessed at a smaller share of their value than commercial and utility property is. Anyone working out what a home is worth by dividing its assessment by 56% is using the wrong ruler, and will guess low.

Why the city alone, and not the county? An assessment ratio only means something inside a single assessing unit, because each one sets its own level of assessment. Chemung County has eleven. Their state equalization rates run from about 1% in Ashland and Baldwin to 100% in Big Flats and Catlin. Pool them and you are measuring the differences between town rolls, not unfairness within one. An earlier version of this page published a county-wide curve. We pulled it in favour of the city-only study, which is how a ratio study is supposed to be done.

The Gap Is Getting Worse

Split the city's sales into three price bands and follow each one year by year. All three fall. They do not fall at the same speed, and that difference is the whole problem.

Every band drifts downward, because prices climbed while the roll sat still. That part is just arithmetic. What matters is the spacing between the lines.

Sale price band	2018 ratio	2025 ratio	Change
Under $50,000	1.333	0.991	−26%
$50,000–$100,000	0.727	0.567	−22%
$100,000–$150,000	0.589	0.408	−31%

Read the top and bottom rows against each other. In 2018 a cheap home was assessed at 2.3 times the share of value that a $100,000–$150,000 home was. By 2025 that had grown to 2.4 times. The cheapest band is still the only one anywhere near a fair ratio of 1.0, and it got there by having no value to gain.

The Widening Gap — Ratio Trends by Price Band, 2018–2025 Three bands of sale price, tracked year by year. Gold is under $50,000, mid-blue is $50,000–$100,000, dark blue is $100,000–$150,000. Nothing above $150,000: that band holds six to eight city sales in 2018 and 2019, and a median of six sales is not a trend. Every point plotted here rests on at least 20 sales.

This is what happens when a housing market moves and assessments do not. Prices rose after 2020 in Chemung County the way they rose everywhere. The roll did not follow. The result is a growing discount for homes that gained value, paid for by homes that did not.

The only fix is a citywide reassessment. One thing that is not a fix: the state's equalization rate update, which caused the apparent jump in Elmira's "full value" figures in 2023. That changed how the state measures the city's roll. It is bookkeeping. It did not change a single property's assessed value, and nobody's tax bill moved because of it.

Elmira Specifically

Who is actually carrying this — and how concentrated the damage is.

Of the 1,689 city sales in this study, 17.7% were assessed for more than the buyer paid. That number hides how concentrated it is. Below $50,000, 64.4% of sales were assessed above the sale price. Above $50,000, only 1.9% were.

So over-assessment in Elmira is not spread thinly across the city. It is almost entirely a condition of the cheapest neighbourhoods. The people living there are more likely to be renters, whose landlords pass the tax through in the rent, or low-income owners with few options to appeal or move.

Who Pays What — Distribution of Ratios The taller the curve, the more sales landed at that ratio. Cheap homes (gold, under $50K) spread across and above the fair-value line; expensive homes (blue, $100K and up) bunch tightly well below it. The dashed lines mark each group's median.

At the other end, homes in the middle of the market are assessed far too low. Houses that sold between $120,000 and $140,000 in 2024 carried a median assessed value of $52,000, a ratio of 0.40. That owner pays tax on $52,000 while living in a $129,000 house. Per dollar of what the home is actually worth, they pay less than half what a neighbour in one of those over-assessed cheap homes pays.

This is not a story about the rich dodging taxes. Most of these are modest working-class houses. What drives it is not mansions. It is a whole housing market rising away from a frozen roll, unevenly, along lines that happen to track income and neighbourhood.

This is not unique to Elmira. The same pattern is documented in Chicago, Detroit, Philadelphia and dozens of other cities that reassess rarely. It shows up so consistently that researchers treat it as a built-in feature of low-reassessment systems rather than a mistake by any individual assessor. Elmira fits the pattern exactly.

Source for all three charts: NYS ORPTS SalesWeb — 1,689 arm's-length single-family (class 210) sales, City of Elmira 2018–2025, compiled into jcurve.json by scripts/visualize_jcurve.py. Full method on the Data & Sources page.

Method

Written for anyone who wants to check us rather than take our word for it, so this part stays technical.

The data. 1,689 arm's-length single-family (class 210) sales in the City of Elmira (SWIS 070400), 2018–2025, from NYS ORPTS SalesWeb (data as of May 2025). Each sale's assessment-to-sale-price ratio is assessed value ÷ sale price, using the assessment on the roll at the time of that sale, not a later one. Sales below $10,000 and ratios above 5.0 are dropped as likely data errors or non-market transfers. Also removing multi-parcel, part-parcel, new-construction and personal-property sales changes the headline by under 5%, so we keep the simpler filter.

Uniformity and bias follow the IAAO Standard on Ratio Studies. COD is the mean absolute deviation from the median ratio, as a percent of that median. PRD is the mean ratio divided by the sales-weighted mean ratio. PRB regresses proportional deviation from the median ratio on log₂ of the value proxy ½ × (assessed ÷ median ratio + sale price). That proxy is chosen so neither assessed value nor sale price alone drives the slope.

The bias correction. Grouping sales by price puts sale price on the x-axis and in the ratio's denominator at the same time, which tilts the curve downward by itself. Grouping by assessed value flips the problem rather than solving it, since assessed value is the numerator. So the correct null is not a flat line in either direction. We estimate how far an individual sale strays from true market value using 510 repeat sales, the same parcel sold twice within three years. That gives a per-sale dispersion of 0.232 in log terms, an upper bound because some pairs are genuine renovations. Simulating a city with that much sale noise and no regressivity at all yields a null curve that is nearly flat, and a null headline gap of 1.27×. That simulated series is not plotted — it is an intermediate, since the bias-corrected line is the observed one divided by it — but it is published in jcurve.json under curve.null. The observed gap is the median ratio under $50K (1.175) over the median at $150K and above (0.457), or 2.573×. Dividing out the artifact leaves 2.02×.

The trend chart splits sales into three price bands and takes the median ratio for each band in each year. A band-year with fewer than 20 sales is left blank rather than plotted. Nothing above $150,000 is charted, because that band holds six to eight city sales in 2018 and 2019.

Literature. The regressive pattern measured here is well documented in the assessment literature (e.g. Berry 2021 on Chicago; Avenancio-León & Howard 2022). Every figure on this page is regenerated and re-checked by scripts/audit_jcurve_figures.py, which fails if any of them drifts from the source data.

Who Pays Their Fair Share?