Three Blind Assessors

Every tax on property rests on a number nobody paid. A house sells, and there's a price. But most houses don't sell in any given year, and a tax on property still has to be charged on all of them. So someone has to say what each one would have sold for. That statement is an assessment, and it's one of the oldest acts of government there is.

It's also one of the least examined, perhaps because it's usually treated as a calculation and not as what it really is: a practice, something a community does over and over, gets better or worse at, and has to keep up. People argue about property tax rates all the time. They argue about exemptions, caps, and who should pay more. Almost nobody argues about how the underlying number is produced, even though the rate can't be fair if the number isn't. For a Georgist tax, which falls on land value alone, the problem is sharper still. The number the tax depends on is one that the market almost never states directly.

This essay starts with assessment in general, then turns to what it takes to assess land well enough to tax it. It asks how much of that work can be handed to machines, and ends with the direction assessment ought to be moving in, to the extent it has one.

What an assessment is

The Domesday Book, compiled for William the Conqueror in 1086, is the famous early case. Royal commissioners went shire by shire recording who held each manor, how many plough-teams it supported, how much meadow and woodland it had, and what it had been worth before the conquest and was worth now. It was a survey of land and what the land could yield, made so the crown knew what it could claim. Egypt's scribes had done something similar along the Nile for millennia, re-measuring fields after each flood. The Roman census did it for the empire.

What these have in common holds for every assessment since. An assessment is three things at once.

It's a price without a sale: an estimate of what something would trade for, made by someone who isn't buying or selling it.

It's an official claim: a number the government stands behind and charges against, which the owner can dispute.

And it's a record over time: the same parcel valued again and again, so that changes can be seen and compared.

The third is what makes assessment a practice and not an event. A single valuation is an estimate. A valuation repeated every year, checked against what sells, corrected, and argued over is a discipline, with standards, habits, and a memory. Most of what goes wrong with property taxes goes wrong because a community stopped practicing: it let the record lapse, froze it, or stopped checking it.

Every assessment also answers, openly or not, three questions: worth to whom, at what moment, and under what rules. Modern practice in most American jurisdictions answers them as "a typical buyer," "a fixed valuation date," and "the property's current legal uses." Each of those answers is a choice, and each could be made differently.

How it's done now

Assessors have three classic methods, all borrowed from single-property appraisal. The sales comparison approach looks at what similar properties sold for and adjusts for the differences. The cost approach adds up what it would cost to rebuild the structure, subtracts depreciation, and adds a figure for the land. The income approach capitalizes the rent a property produces or could produce. Houses are mostly valued by comparison, new or unusual buildings by cost, and commercial property by income.

No county can send an appraiser to every parcel every year, so assessors don't use these methods one parcel at a time. They use mass appraisal: statistical models, fitted to recent sales, that estimate every parcel's value from its recorded characteristics. The software is called CAMA, computer-assisted mass appraisal, and it has been standard for decades.

The profession also has a way of grading itself. A ratio study compares assessed values to the prices of properties that actually sold. The International Association of Assessing Officers publishes standards for how tightly those ratios should cluster. For single-family homes, the coefficient of dispersion, the average percentage by which assessments miss, should generally fall between five and fifteen. The price-related differential tests something subtler: whether cheap and expensive homes are assessed at the same fraction of their value.

That second test is where American assessment most often fails. Christopher Berry's work at the University of Chicago, covering most of the country's counties, found that assessments are widely regressive. Cheaper homes tend to be assessed at a higher share of their sale price than expensive ones, so their owners pay a higher effective rate. The Chicago Tribune and ProPublica documented the same pattern in Cook County in 2017, in detail and at great cost to the owners of modest homes. No one voted for that result. It came out of the models, the appeals process, and the difference in who can afford to contest a bill.

That's the first lesson for anyone designing a Georgist system. The fairness of a property tax is decided less by its rate schedule than by its assessment, and assessment failures are quiet. They don't show up in a statute. They show up in a ratio study most people never read.

Why a land tax needs more from assessment

An ordinary property tax asks for one number per parcel: the value of the whole thing, land and building together. Sales supply that number fairly directly, because what sells is the whole thing. The split between land and building matters little, since both are taxed at the same rate. Many assessors treat the land figure as an afterthought, the leftover once the building has been costed.

A land value tax reverses that. The building's value drops out of the tax, and the land's value is the entire base. The one number the tax depends on is the one the market states least often, since almost every sale is a bundle. Vacant lots do sell, but in most built-up places they're rare, odd, and not much like the developed parcels around them.

Assessors have several ways to pull land value out of bundled sales, and a working system uses all of them.

• Vacant sales, where they exist, are the most direct evidence.
• The residual method takes a sale price, subtracts the depreciated cost of the building, and treats what's left as land.
• Allocation applies a typical land share for a neighborhood to its total values.
• Location models estimate how value changes with position, relative to a corner, a transit stop, a shoreline, a school boundary, and draw a smooth land value surface across a city.

The last of these has a long history in Georgist practice. William Somers, a St. Paul assessor around the turn of the twentieth century, built a unit-foot system that valued each block's frontage from public hearings on relative desirability, then applied standard adjustments for depth and corners. His insight holds up: land value is a property of location more than of the parcel, so it's best estimated as a continuous map and not one lot at a time.

There's a second problem, and it's specific to land taxation. A land value tax is capitalized into land prices. A buyer who knows they'll owe a large annual tax on a parcel will pay less for it. As the tax rises toward the full annual rent, the selling price of land falls toward zero. At that point sale prices stop measuring what the land is worth at all. They only measure what's left over after the tax.

So a mature Georgist system can't assess land by its selling price. It has to assess what the land would rent for each year, before tax. That's the quantity George meant to collect, and it's the quantity sales can no longer show once the tax is doing its job. Britain's old domestic rates were levied on annual rental value, not capital value, so the idea isn't exotic. It does mean that ground rents, leases, and rental markets become the main evidence, and that a land tax at full strength needs a different assessment apparatus from a property tax at one percent.

A third problem runs underneath both. Land value depends on what the land is allowed to be used for, and the community sets that. A parcel zoned for one house and the same parcel zoned for forty apartments are worth very different amounts. The assessment carries the zoning code into the tax bill. It isn't a neutral measurement of nature. It's a measurement of what the community has made of a location, including the rules the community has chosen to place on it. That's exactly why the value belongs to the community. It's also why an assessor valuing land is always partly reading policy, not just market data.

Three things that pull against each other

Assessment is asked to deliver three things that seem to trade off against each other.

Accuracy. The assessed value should track what the land is actually worth, as closely and as currently as possible.

Stability. Owners should be able to predict their bills. A tax that jumps forty percent in one year, however accurate, forces people out of homes they could afford the year before.

Legitimacy. People should understand how their number was produced and be able to contest it when it's wrong.

Push accuracy and bills swing with every market move. Push stability and the number drifts from reality, as it did in Pittsburgh, where deferred reassessment ended an eighty-eight-year split-rate tax in 2001 (covered in Transitioning to LVT). California took stability as far as it can go. Parcel Duality laid out what Proposition 13 did: assessed values are frozen except at sale, and the tax roll describes a state that no longer exists. Estonia, which taxes land alone, went two decades without a general revaluation, from 2001 into the 2020s, and found that the eventual correction was politically painful in exactly the way Pittsburgh's was.

Accuracy and stability were never in conflict. They were being asked of the same number when they belong to different parts of the system. Accuracy belongs to the assessment. Stability belongs to the bill. The assessed value should always be as current and correct as the evidence allows, and it should never be frozen, capped, or smoothed. If the bill needs smoothing, that can be done openly and separately: phase changes in over several years, average the tax over a rolling window, let owners on fixed incomes defer payment as a lien against the land, due at sale or death. Each of these protects the owner without corrupting the record. A frozen assessment protects the owner and destroys the record. Every later decision is then made against a false number, and the eventual correction lands all at once.

Legitimacy is the third force, and it's what lets the other two coexist. An accurate assessment that nobody understands will be resented until it's capped. An open one can be defended.

How much can be automated

Most of it, and less than enthusiasts expect.

Mass appraisal was already automation, and modern tools push it further. Machine-learning models trained on sales, permits, and parcel characteristics routinely beat older regression models on ratio-study measures in dense residential markets. Aerial and street-level imagery can detect new construction, additions, and demolitions without a site visit. Rental listings, lease filings, and land sales can be pulled in continuously. Cook County's assessor's office, after the Tribune series, began publishing its residential valuation model and its data openly so outsiders could inspect and rerun it. That's the right model, and it isn't technically hard.

The limits are real, though, and they're worth being specific about.

Thin markets. Models learn from sales. Where there are few, as with rural land, industrial parcels, large downtown sites, or unusual properties, the model has little to learn from and the error grows. These are often the most valuable parcels in a jurisdiction.

The land share. Models are good at predicting total sale price. Separating land from building is a judgment about a quantity no transaction states, and a model's split is only as good as the assumptions built into it.

The capitalization trap. Once a land tax is substantial, sale prices stop reflecting land value, and a model trained on them learns the wrong thing. Automation that works under a one-percent property tax can quietly fail under a serious land tax.

Private-sector failures. Zillow's home-buying business ran on an automated valuation model and closed in 2021, after writedowns of roughly half a billion dollars, because the model overpaid in a market that turned. That's a warning about using any model's point estimate as if it were a price.

Scaling problems. Denmark suspended its property valuations in 2013 after errors in the old system, and its replacement, built around automated models, took years longer than planned to deliver new values. Automation at national scale is a large public project, and it can fail as one.

So automation should produce the estimate, not the final word. Models should do the routine work of valuing dense, similar parcels, flagging changes, and keeping values current every year. People should handle the parcels where evidence is thin and the land split is contestable. They should run the ratio studies that check the models against reality, especially for regressivity. And they should hear appeals. Anything automated should be public: the model, the data, and the reasons for any individual number, stated well enough that an owner can check it. An owner can argue with an assessor. An owner can't argue with a model nobody can see. If automation takes away that ability, it has made the tax less legitimate, however accurate it is.

Letting owners assess themselves

There's one alternative to the assessor that Georgists should take seriously, even if they end up rejecting it. Sun Yat-sen proposed it for China early in the twentieth century, and Eric Posner and Glen Weyl revived it in Radical Markets in 2018: let owners declare the value of their own land, tax them on the declared figure, and oblige them to sell to anyone willing to pay it. An owner who declares low pays little tax but risks losing the land cheaply. One who declares high is safe from purchase but pays for it. The incentives push declarations toward true value without any assessor at all.

It's an elegant mechanism, and it removes most of the measurement problem. It also makes every holding provisional, open at all times to anyone with the money. That cuts against something this collection of essays has argued for repeatedly: that secure possession, the ability to stay put and build a life on a place, is itself part of liberty. A partial version keeps most of the benefit with less of the cost. The public assessment stays the default, and an owner can declare a different value, with the declaration binding the owner in some limited way, for example as the price at which a public buyer may acquire the land. Owners then become a check on the assessor, and the assessor a check on them.

Assessment as stewardship

A libertarian movement that talks about responsibility has to talk about assessment, even though assessment is the kind of administrative detail such movements usually skip. Responsibility needs an account. A steward can't be held to what's in their care unless someone knows what that is, what it was worth when they took it on, and what it's worth now. Assessment is that account for land.

The Myth of the Public Steward argued that an owner who has to live with what happens next takes better care of land than an agency that doesn't. A Georgist system doesn't take that away. The owner still holds the land, still decides what to do with it, and keeps everything they build on it. What changes is that the owner pays the community for the location, the part of the value the community created. The assessment sets the terms of that arrangement. It's the line between what the owner is responsible for and what the owner merely holds.

Drawn well, that line rewards stewardship in a way the current property tax doesn't. Today an owner who repairs a building, adds a unit, or improves a structure sees the assessment rise, and pays more for having done better. An owner who lets a building decay sees it fall, and pays less for having done worse. A property tax on total value charges people for care and discounts neglect. An assessment of land alone reverses that. What an owner builds and maintains is theirs and untaxed, and the charge for the location is the same whether they tend it or let it go.

There's a harder case, and a practice of assessment has to handle it. The land itself can be improved or degraded. An owner can drain a field, rebuild soil, restore a creek, or plant trees that will outlive them. Another can deplete the soil, contaminate the ground, or strip it bare. A crude land value assessment would tax the first owner more for having made the land better and the second less for having made it worse. That's the property tax's mistake one level down. Valuation practice in New Zealand and Australia has long dealt with a version of this by distinguishing a site's unimproved value from improvements made to the land itself. A Georgist system should go further and treat stewardship as a category of its own:

• What the location is worth belongs to the community, and the assessment should collect it.
• What the owner has added to the land, like restored soil, drainage, or habitat, is the owner's work and should be exempt, just like a building.
• What the owner has taken from the land, like depleted soil, contamination, or lost cover, is a debt, and the assessment's record should show whose.

Only an assessment that keeps a record over time can make those distinctions. It has to know what condition the land was in when the owner took it on, and what condition it's in now. That's the strongest reason the record can't be allowed to lapse. Without it, stewardship and neglect look the same on the tax roll.

Responsibility also runs the other way. A movement that asks owners to answer for what they hold has to ask the same of the community that assesses them. Letting the assessment rot is itself a failure of stewardship. Freezing it, as California did, or deferring it until it breaks, as Pittsburgh did, or quietly overcharging the owners of cheap homes, as Cook County did, are all failures of the same kind. The shared value of a place is also something held in trust. The practice of assessment is how a community takes care of it: by measuring it, keeping the record accurate and public, and correcting it when it's wrong. A libertarian case for owner responsibility that skips this part is only half an argument.

Where assessment should be heading

Assessment doesn't have a final form, which is the strongest reason to think of it as a practice. Land value is a social fact that changes as the community around it changes, so there's no last, correct number to reach. But it does have a direction. The practice gets better the further it moves along a few lines.

From total value to land value. Every assessment should report land and improvements separately, and publish both, even where the tax still falls on both. New Zealand's rating valuations have long reported land value and capital value side by side, and New South Wales values every parcel's land alone each year for its land tax. A jurisdiction that already knows its land values can shift its tax toward them one step at a time. One that doesn't know them can't start.

From capital value to annual rent. As a land tax grows, assessment has to move from asking what the land would sell for to asking what it would rent for. That's the quantity the tax is meant to collect, and the only one that survives the tax.

From periodic to continuous. Annual revaluation should be the floor, not the goal. There's no technical reason values can't be updated as fast as evidence arrives, with stability handled in the bill, not the record.

From price to condition. Alongside what land is worth, assessment should record what state it's in: soil, water, cover, contamination. That record is what lets the tax exempt the owner's improvements to the land and charge for its degradation. Without it, stewardship and neglect can't be told apart.

From opaque to open. The parcel map, the model, the sales data, and the reasoning behind each value should be public and reproducible. Anyone should be able to rerun the county's numbers and show where they're wrong.

From verdict to argument. An assessment should be treated as a claim the government makes and must defend, with appeals cheap and quick enough that the owner of a modest home can bring one as easily as a commercial landlord with a tax attorney. That's the main defense against the regressivity the ratio studies keep finding.

From local agency to shared tools and local judgment. Models, imagery, and data pipelines can be built once and shared across many jurisdictions. The judgments that follow, like how to treat a thin market or a zoning change, belong to the community whose land is being valued.

From a number to a ledger. At its furthest reach, a land assessment becomes something more than a tax base. It becomes a public account of where the community's value comes from. A transit line, a park, a school, or a rezoning each shows up as a change in the land values around it. A community that can see that can charge for what it creates and answer for what it destroys. The same logic reaches past urban land to the natural rents covered in Single Tax for the Living World: water, spectrum, the atmosphere's capacity to absorb waste. Each can be valued the same way, and each is shared in the same sense.

That last direction is the one that matters most to a Georgist. The claim at the center of the tax is that land value is created by everyone and so belongs to everyone. The assessment is where that claim becomes visible, one parcel at a time. Done badly, it hides the value, misattributes it, and charges the wrong people for it. Done well, it gives a community an accurate picture of what it has made. The tax, the dividend, and the argument all rest on that picture.

---

A note on how this piece was written: every argument, source, and structural decision in this essay is mine — I chose the examples, worked out the logic connecting them, and directed what each paragraph needed to say. The sentences themselves were drafted with AI assistance from that outline and then edited by hand. I'd rather say that plainly than have a reader guess at it.