Mass Appraisal Methods

Definition

Mass appraisal is the practice of valuing many properties simultaneously, on a common valuation date, using standardized statistical methods rather than individual expert appraisals. Computer-assisted mass appraisal (CAMA) systems apply hedonic regression — modeling a property's sale price as a function of its location, lot size, structure characteristics, and other measurable attributes — to estimate land and building components separately across an entire jurisdiction. Because location dominates the pricing of otherwise-similar parcels, these systems can isolate the land component even where few or no unimproved lots actually sell.[1]

More recent refinements include cooperative-game approaches. A Shapley-value method for allocating a combined sale price between land and structure offers a principled, replicable way to perform land/building separation, formalizing what has historically been an assessor's judgment call.[2]

Core Methods

Hedonic Regression

Hedonic regression treats a property's sale price as the sum of implicit prices for its individual characteristics. Land-related attributes — location, lot size, frontage, topography, zoning — are modeled alongside structure-related attributes — square footage, age, condition, number of rooms. The regression coefficients on land-related variables yield an estimated land value surface across the jurisdiction, even when most transactions involve improved parcels.[1]

This approach is standard in modern assessment practice. The IAAO's Standard on Mass Appraisal of Real Property (2017) — the assessing profession's consensus standard — notes that "Multiple regression analysis (MRA) and related techniques have been successfully used" in valuation and directs that sale data be maintained for "the development of land values."[8] Production examples include Philadelphia, whose Office of Property Assessment reports that its "CAMA system, iasWorld, went live in February 2020" (iasWorld is a Tyler Technologies product) and calls implementing a CAMA system "an industry best practice";[9] and Cook County (Chicago), whose assessor publishes an open-source machine-learning CAMA model "used to generate initial assessed values for single- and multi-family residential properties in Cook County."[10]

A 2026 case study by Dr. Chester Harvey and colleagues at the National Center for Smart Growth applied a hedonic model directly to non-apartment residential sales across Baltimore (2013–2022 sale data, tested on 2022 properties), incorporating zoning, downtown proximity, job accessibility, crime, and tree-canopy coverage as predictors; the model explained roughly 85% of sale-price variability. The study's three practical lessons illustrate the gap between technique and implementation the rest of this page discusses: incomplete public data on property improvements constrained accuracy despite the method's theoretical soundness; areal-level Census data proved too coarse and produced implausible negative improvement values when substituted in; and adding spatial-lag terms (modeling nearby-parcel similarity) inflated land values unrealistically, suggesting the local housing market was less spatially homogeneous than the technique assumed.[18] Baltimore's hedonic land-value exercise is methodologically distinct from, but politically adjacent to, the city's separate 2026 split-rate hearing bill discussed on the split-rate taxation page — one asks whether land value can be estimated accurately there, the other whether the city should tax it differently once estimated.

Residual and Comparable-Sales Methods

Three traditional methods for isolating land value remain foundational and are formalized within the CAMA framework:

• Comparable vacant-lot sales: where unimproved lots do sell, their prices directly indicate land value for nearby parcels. Even in extremely dense markets this can yield a usable land-value series: Barr, Smith & Kulkarni built a 1950–2014 land-value index for Manhattan entirely from vacant-land sales, valuing the borough's developable land at ~$1.47 trillion in 2014.
• Teardown sales: when a buyer purchases a property and demolishes the structure, the purchase price approximates land value (minus demolition cost), providing a land-value signal even in built-up areas.
• Residual estimation: land value equals sale price minus the depreciated cost of the structure, estimated from construction-cost data and depreciation schedules.[1]

The writer Lars Doucet, whose popular Does Georgism Work? series made the assessment case to a general audience, argues that these methods work because land value is spatially smooth — neighbouring parcels have similar land values — whereas building value varies house-by-house; on his reading, that smoothness lets location-based methods estimate land value with accuracy comparable to or better than whole-property assessment, which must also value the more heterogeneous building component.[1] The comparison is his synthesis of the assessment literature rather than a measured result reported by the profession's own standards bodies.

Cooperative-Game (Shapley-Value) Separation

Özdilek (2020) proposes a cooperative-game approach in which a property's combined sale price is allocated between land and structure using Shapley values — a concept from cooperative game theory that distributes a coalition's total value among its members according to their marginal contributions. Applied to land/building separation, this method treats land and structure as complementary "players" whose combined value exceeds either alone, and assigns each a share based on its incremental contribution to total property value.[2]

This method offers a replicable, axiomatically grounded alternative to ad hoc residual estimation, but no assessment jurisdiction is known to have adopted it in production. The 2020 paper that proposed it validated the approach only as an empirical case study — on 14,715 residential sales across Montreal's 27 districts — rather than in an operational assessment roll,[2] and a 2025 follow-up applying game theory to cadastral valuation states plainly that for the land/building split "there are no established methods" and that its own game-theoretic approach "is new in the theory and practice of valuation."[13]

Why It Matters for LVT

Mass appraisal is the practical answer to the objection that land value can't be assessed: assessors do not need a pure land sale next to every parcel, because the spatial smoothness of land values lets statistical methods interpolate from available data points across a jurisdiction.[1]

The Lincoln Institute of Land Policy's standard policy reference on land value taxation, by Dye & England (2010), treats assessment methodology — including these techniques — as an implementation question to be addressed with better tools and investment, rather than a fundamental barrier to a land value tax.[3] The report surveys experience across more than 30 countries and US municipalities, including the Pennsylvania split-rate cities, and covers assessment methods, transition issues, and the political economy of adoption.[3]

Existing Practice at Scale

Separate land-value assessment is not merely theoretical. Several jurisdictions already assess land values separately as routine practice:

• Estonia has levied a land value tax since 1993, requiring land-only assessments nationwide — national valuations were carried out in 1993, 1996, and 2001, moving to mass-valuation models with value zones.[12]
• Denmark operates a grundskyld (land tax) assessed on land value separately from buildings.[4]
• Several Australian states, including New South Wales (land tax since 1895 on unimproved capital value), assess land value separately as a matter of long-standing practice.[5]
• Most US assessments already publish a land/improvement split, though the quality and methodology of that split varies widely across jurisdictions.[4]

New Zealand's historical use of land-value rating as a primary local government funding mechanism provides another significant case, and the country's gradual shift from land-value to capital-value rating bases has given researchers a quasi-natural experimental setting for studying the effects of the assessment base choice.[6]

Beyond these established systems, a synthesis of six single-country case studies (Croatia, Turkey, Lithuania, Spain, China, Quebec) shows statistical, fuzzy-logic, and multi-criteria land-valuation methods being attempted independently across unrelated legal and academic traditions.[15] Individually most of these studies are thin — several are readable only at abstract or bibliographic level — so they support breadth of attempt, not confirmed accuracy in any one country; one case in the set (Lithuania) is itself a caution, showing three defensible valuation methods diverging by 3.4× on the same parcel.[15]

Assessment Quality and Error

Because land is immobile and visible, gross mis-assessment is relatively easy to challenge through appeal processes. Assessment quality improves with investment in data, methodology, and professional capacity — a focus of the Center for Land Economics, co-founded by Lars Doucet.[4]

The key insight from the assessment literature is that land-value assessment errors are bounded and contestable in ways that many other tax-base errors are not: land cannot be hidden, moved offshore, or recharacterized as another asset class. Under-assessment, not evasion, is the primary practical risk — a concern addressed in the wiki's tax-you-can't-dodge narrative.[4]

How Assessment Quality Is Actually Measured: Ratio Studies

The profession's own instrument for testing whether a mass-appraisal model is any good is the ratio study, standardized by the IAAO's Standard on Ratio Studies — read directly for the wiki rather than at second hand.[14] A ratio study compares appraised values against known sale prices and reports two kinds of error: level (are values close to 100% of market value overall?) and uniformity (are similar properties assessed at similarly close-to-market ratios?). Three metrics carry that testing:

• Coefficient of dispersion (COD) — average percentage deviation of individual ratios from the median; the standard's acceptable range is tighter for improved residential property (5.0–10.0/15.0) than for vacant land (5.0–25.0), a gap plausibly driven by thinner vacant-land sales samples rather than land being intrinsically harder to value.[14]
• Price-related differential (PRD) and price-related bias (PRB) — measures of vertical equity, testing whether low-value and high-value properties are assessed at the same fraction of market value; PRDs should sit between 0.98 and 1.03, and PRBs between −0.05 and 0.05.[14] These are the formal metrics behind Berry (2021)'s finding, above, that US assessments are "typically regressive" — in IAAO's own vocabulary, a PRD/PRB result.[14]

The standard also explicitly names land-only ratio studies as routine practice — "for a ratio study of vacant land, the numerator in the ratio should reflect only the appraised value of the land" — direct professional confirmation that isolating and testing land value alone is standardized methodology, not a Georgist-specific workaround.[14] The companion Standard on Property Tax Policy states IAAO's "core principles" for property tax design but, read in full, says nothing substantive about land value or split-rate taxation as a policy choice beyond a single unglossed bibliography citation — a silence worth noting rather than assuming otherwise.[14]

A direct empirical test of these methods against real sales data comes from Gloudemans's two Lincoln Institute working papers (2000, 2002), which calibrated combined vacant-and-improved-sales CAMA models on three North American residential databases. Adding vacant-land sales to a combined model barely moved predictive accuracy for improved properties (one jurisdiction's COD rose only from 5.87 to 5.97; another's actually fell from 5.75 to 5.55) while producing more plausible land/building splits than improved-sales-only models, which Gloudemans found tend to mechanically overstate land's share by absorbing buildings' fixed costs into the land term. Vacant-land CODs themselves ran higher than improved-property CODs throughout (9.55–22.96 across models and jurisdictions), consistent with the IAAO's own wider tolerance band for vacant land noted above.[16]

The Failure Mode Isn't Bad Models — It's Rolls Nobody Updates

Most of this page concerns whether assessors can estimate values accurately. A September 2026 Times Union investigation of New York State is a reminder that the binding constraint in practice is often whether they estimate them at all, and how recently. Analysing official state data, the paper found that nearly 55% of New York's residential assessment rolls in the previous year failed the state's own uniformity test — spreads between assessments of similar properties too wide for the state to consider equitable — and that nearly half of all municipalities had not conducted a full reassessment in at least a decade, with 61 communities whose last one preceded 1974.[19]

The mechanism is arithmetic rather than methodological. A property tax is zero-sum within a jurisdiction: the levy is set first, and assessments only determine each owner's share of it. So when a roll goes stale, properties whose market values rose fastest become undertaxed and those that rose slowest or fell become overtaxed, with no error in any individual valuation required. New York, unlike most states, mandates no reassessment cycle at all — it only offers aid of up to $5 per parcel to municipalities that reassess at least every four years, against per-reassessment costs that can reach millions in a large suburb.[19]

Three further findings bear on the wiki's assessment-administration coverage:

• Uniformity can be faked. Because municipalities assess at differing fractions of market value, the state assigns each an annual equalisation ratio. Assessors in many communities simply apply that state-provided rate to every residential assessment from the prior year — producing, as the reporting puts it, "the bogus impression that many residences may have gained the same percentage of value." State lawyers consider the method improper; one Capital Region assessor, speaking anonymously, said of it: "Yeah, that's really what I do. I don't make it harder than it needs to be." A roll produced this way can pass a uniformity test while resting on values a generation old — so the ~55% figure is a floor, not a ceiling.[19]
• Selective reassessment persists despite being illegal. Consultants described the "welcome stranger" practice — raising assessments on homes that have just changed hands while leaving neighbours' untouched — which courts have consistently held impermissible. One Syracuse buyer's assessment jumped from $80,000 to $96,000 after a $150,000 purchase, in a city that had not fully reassessed in 30 years.[19]
• Appeals redistribute rather than correct. Every reduction won shifts burden onto owners who did not file, and the contingency-fee firms that dominate the process concentrate where bills are highest: one firm filed just under 30,000 grievances in a single year across twelve downstate counties, charging half of a client's first-year savings. This is the same access asymmetry the wiki documents in Memphis and Dallas, observed here at statewide scale.[19]

The political economy of the fix is the discouraging part, and is the reason the problem is durable rather than accidental. Reassessment changes nobody's aggregate tax bill but reshuffles every individual one, so it reliably produces roughly as many losers as winners: when Albany reassessed after an eight-year gap, its assessor projected 49% of residential owners would see lower taxes and 51% higher, and grievances jumped from a few hundred in a typical year to more than 1,700. A state Senate bill requiring reassessment at least every eight years from 2030 has sat in committee since 2025 and carries no new funding.[19] For LVT specifically the implication cuts two ways: a land tax does not escape this failure mode, since a stale land roll misallocates burden exactly as a stale property roll does — but land values are the component that moves most sharply and most unevenly across a metro area between revaluations, which makes assessment frequency, not just assessment method, a first-order design parameter rather than an administrative detail.

An Alternative: Avoiding the Split Altogether

Rather than improving land/building separation techniques, some designs try to avoid needing the split at all. Lars Doucet's 2026 DALT (Depreciation-Assisted Land Value Tax) proposal keeps a conventional, unified property-tax roll but fully exempts new-construction improvement value for roughly the first half of a building's expected lifetime; because depreciation is front-loaded and present value weights early-year taxes most heavily, Doucet argues this can approximate a pure LVT's incentives in fast-depreciating markets while requiring assessors to estimate only new-construction values, not a full land/building split across the existing parcel roll.[17] It is a legal- and assessment-burden workaround, not a validated substitute for the mass-appraisal methods above, and remains an untested, single-source proposal as of September 2026.

Limits and Open Questions

• Data quality varies by jurisdiction. CAMA systems require transaction data, structure characteristics, and geographic information; jurisdictions with poor records or few arm's-length sales face greater estimation uncertainty. Thin-data settings can substitute structured expert weighting for missing sales: Bencure et al. (2019) built an analytic-hierarchy-process land-valuation model (iLVM) for sub-urban Baybay City, Philippines — the developing-country case Doucet cites — that explained ~67% of land-value variability and beat ordinary multiple regression on RMSE, though its expert-dependent weights transfer rather than eliminate subjective judgment. Berry (2021), regressing sale prices against assessed values in 2,628 US counties, found an average county-level r-squared of just .46 with an interquartile range of .25 to .67, concluding: "Clearly, there is variation in performance across assessors, but there is a great deal of unexplained variation in sale prices in the vast majority of jurisdictions."[11] The same study finds assessments "typically regressive," with bottom-decile homes assessed at roughly twice the ratio to sale price of top-decile homes within a jurisdiction — though this concerns whole-property assessment, the current baseline against which land assessment would be compared.[11]
• The land/building split is hardest in dense, built-up areas. Dye & England note that "traditional methods of assessing land values in developed urban areas have various drawbacks": the common abstraction (residual) method requires "subjective judgments" to extract land value, and the allocation method "makes the arbitrary assumption that land values are the same fixed percentage of total value" for every property.[3] Doucet reports the same difficulty for statistical methods: in the Berlin studies he reviews, nonparametric kernel interpolation from pure land sales was "not able to produce estimates for the city center, only the outlying areas," which is what motivates semiparametric methods that add building characteristics to the model.[1]
• Shapley-value and other advanced methods remain largely academic. Operational adoption of cooperative-game separation methods in assessment practice has not been documented in the sources reviewed here.[2]
• Agricultural and peri-urban land presents special challenges where planning-permission value creates a large gap between use-value and market value — a concern noted in the Mirrlees Review and relevant to the farmer impact objection.[7]

See Also

• New York City's Pied-à-Terre Tax (2026) — a levy whose two-phase design exists because the city's statutory condo and co-op valuations sit far below sale prices
• PLACES Lab — Boston University research group whose machine-learning land-value maps are a modern, nationwide-scale demonstration of the assessment feasibility this page argues for
• McCanless: Constructing Public Rent — Property Tax Appeals and Rentier Capitalism — a caution that assessment accuracy alone doesn't guarantee equity if the appeals process is itself unequally accessed
• Residual Land Valuation — how developers price land in practice (sales value minus costs), explaining why land absorbs planning gain in full
• International Association of Assessing Officers (IAAO) — the professional standards body behind mass appraisal and ratio-study methodology
• IAAO Standards: Ratio Studies & Property Tax Policy — the COD/PRD/PRB definitions and quality bands read first-hand
• Mass Appraisal International Cases — the six-country breadth synthesis, honestly graded by access level
• Gloudemans & Almy, Fundamentals of Mass Appraisal — the IAAO textbook backbone behind the technique inventory (comparable sales, allocation, abstraction, residual estimation) this page describes
• Gloudemans: LVT Assessment Feasibility — the same author's direct empirical test of CAMA land models, both papers read in full
• Kolbe et al.: Berlin land-value appraisal — the accuracy evidence, first-hand · Almy: OECD valuation survey — the cost evidence · Hefferan & Boyd: Australian mass appraisal — revealed taxpayer acceptance
• Hernandi et al. (2026): The SALAD Model, Indonesia — a Land Value Zone / Social Validation Weight model field-tested for community acceptance, not just technical accuracy
• DALT: Depreciation-Assisted Land Value Tax — a 2026 proposal that sidesteps the land/building split entirely rather than improving it
• Objection: Land value can't be assessed accurately
• Land Value Tax
• Site Value
• Split-Rate Taxation
• Research: Dye & England, Assessing the Theory and Practice of Land Value Taxation
• Research: Doucet, Does Georgism Work?
• Center for Land Economics
• Olea, Hoyos & García Enríquez: Bizkaia Land Value Tax — a hedonic spatial model applying this page's methodology to simulate a specific LVT rate for a real jurisdiction
• Yılmaz, Alkan & Teke: A Causal GeoAI Framework for the Unearned Increment — extends ex-post assessment techniques into ex-ante causal prediction, for sizing self-financing infrastructure levies before the investment occurs
• Tubío-Sánchez & Reyes-Bueno: Ecuador LVT Capitalization — direct evidence that systematic underassessment of fast-appreciating areas, not just the tax rate, determines whether capitalization delivers the tax's intended equity outcome

Sources

1. Lars Doucet (2022), "Does Georgism Work? Part 3: Can Unimproved Land Value Be Accurately Assessed?", Astral Codex Ten — wiki summary · original — used for the claim that land's spatial smoothness lets comparable-sales, teardown-sale, and residual methods estimate land value at accuracy comparable to or better than whole-property assessment; and for the hedonic regression framework description.
2. Ünsal Özdilek (2020), "Land and building separation based on Shapley values," Humanities and Social Sciences Communications (Nature), DOI 10.1057/s41599-020-0444-1. Article — used for the cooperative-game (Shapley-value) formalisation of land/building value separation, verified as an empirical case study on 14,715 Montreal residential sales rather than an operational deployment (author confirmed as Ü. Özdilek, sole corresponding author, 2026-07-10).
3. Richard F. Dye & Richard W. England (2010), Assessing the Theory and Practice of Land Value Taxation, Lincoln Institute of Land Policy — wiki summary · Report — used for framing assessment methodology as a practical implementation issue and for the survey of international and US experience.
4. Lars Doucet (2022), "Does Georgism Work? Part 3" — as summarised on the wiki's assessment objection page — used for the Denmark and Australia examples, the claim that most US assessments publish a land/improvement split, and the Center for Land Economics reference. (An earlier draft also attributed the Estonia example to this article; Estonia does not appear in it — the Estonia claim is now sourced to [12].)
5. Robert Andelson, ed. (2001), Land-Value Taxation Around the World — wiki summary — used for the comparative survey of site value rating in Australia and New Zealand, including NSW's land tax on unimproved capital value since 1895.
6. Gemmell, Grimes & Skidmore (2019), "Do Local Property Taxes Affect New Building Development?", Journal of Housing Economics — wiki summary — used for New Zealand's land-value vs capital-value rating distinction and the Auckland quasi-natural experiment.
7. James Mirrlees et al. (2011), Tax by Design: The Mirrlees Review, Institute for Fiscal Studies — wiki summary — used for the agricultural/development value gap concern relevant to assessment difficulty.
8. International Association of Assessing Officers (2017), Standard on Mass Appraisal of Real Property, approved July 2017. PDF — used for the professional-consensus status of mass appraisal, the successful use of multiple regression analysis, and the requirement to maintain sale data for developing land values.
9. City of Philadelphia, Office of Property Assessment (2023), Mass Appraisal Valuation Methodology Overview, Tax Year 2023. PDF — used for the production example of the iasWorld CAMA system (live February 2020) and the characterisation of CAMA as industry best practice.
10. Cook County Assessor's Office, Residential Automated Valuation Model (model-res-avm), public code repository. GitHub — used for the production example of an open-source machine-learning CAMA model generating initial assessed values for Cook County residential property.
11. Christopher Berry (2021), "Reassessing the Property Tax," University of Chicago, Harris School of Public Policy, working paper. PDF — used for systematic evidence on assessment accuracy variation across 2,628 US counties (county-level r-squared averaging .46) and on the regressivity of existing whole-property assessments.
12. Tambet Tiits, "Current Situation and Practice of Land Valuation and Taxation in Estonia" (Lincoln Institute of Land Policy). PDF — used for Estonia's Land Tax Act (levied on the value of land, national collection) and the 1993/1996/2001 national land valuations with value-zone mass-valuation models (verified against the PDF, 2026-07-10).
13. Michael Laskin (2025), "The method for the land plot value appraisal as part of the single real estate object, based on game theory approach," Business Informatics, DOI 10.17323/2587-814x.2025.1.93.107 — used to corroborate that game-theoretic (Shapley) land/building separation remains a research method: the paper states there "are no established methods" for the split and that its own approach "is new in the theory and practice of valuation" (verified via abstract/introduction, 2026-07-10).
14. International Association of Assessing Officers, Standard on Ratio Studies (2013) and Standard on Property Tax Policy (revised 2020). wiki summary · Ratio Studies PDF · Property Tax Policy PDF — both read in full (2026-07-18); used for the COD/PRD/PRB definitions and acceptable-range tables, the explicit land-only ratio-study instruction, and the confirmed absence of land/site-value-taxation discussion in the policy standard's body text.
15. Kilić, Rogulj & Jajac (2019, Croatia); Yalpır & Ünel (2017, Turkey); Raslanas et al. (2010, Lithuania); Aragonés-Beltrán et al. (2008, Spain); Xue et al. (2008, China); Kettani & Khelifi (2001, Quebec). wiki synthesis — used for the six-country breadth evidence and the Lithuanian valuation-divergence caution; see the synthesis page for per-source access-level grading.
16. Robert J. Gloudemans (2000, 2002), Implementing a Land Value Tax in Urban Residential Communities and An Empirical Analysis of the Incidence of Location on Land and Building Values, Lincoln Institute of Land Policy Working Papers WP00RG1/WP02RG1 — wiki summary, both read in full — used for the direct empirical COD/accuracy comparison between improved-only and combined vacant+improved CAMA models.
17. Lars Doucet (2026), "DALT: Depreciation-Assisted Land Value Tax," Progress and Poverty (Substack, Center for Land Economics), July 29, 2026. Article — wiki summary — used for the DALT design as an alternative that avoids the land/building split rather than solving it.
18. Dr. Chester Harvey et al., "Modeling Land Values in Baltimore," Progress and Poverty Institute, August 2026. progressandpovertyinstitute.org — fetched and read 2026-08-21; used for the hedonic model's methodology, its 2013–2022 sales data and 2022 test year, the ~85% explained-variance figure, and the three practical lessons (public data limitations, coarse Census data producing negative improvement values, spatial-lag terms inflating land values) (B-claim).
19. Gabe Levin & Julie Zhu, "How NY's 'broken' property tax system creates winners and losers," Times Union (Hearst), 6 September 2026. timesunion.com — retrieved via curl with a browser user-agent and read in full 2026-09-17, with all figures and quotations verified verbatim against the fetched article text — used for the ~55% non-uniform residential rolls figure, the "nearly half of all municipalities" and "61 communities prior to 1974" reassessment-lag findings, New York's absence of a mandated reassessment cycle and its $5-per-parcel aid, the equalisation-ratio workaround and the anonymous assessor's and state lawyers' characterisations of it, the "welcome stranger" practice and the Syracuse $80,000-to-$96,000 example, the Aventine Properties grievance volume and fee structure, the Albany 49%/51% projection and 1,700 grievances, and the stalled Sanders reassessment bill (A-claim; full text read, quotations verbatim). A newspaper investigation analysing official Office of Real Property Tax Services data, not a peer-reviewed study; the underlying dataset was not independently re-analysed for this wiki.