Skip to main content
v2026.11,986 entries · CC-BY 4.0

Hedonic pricing: formula and worked example

Understand hedonic pricing, interpret regression coefficients and reproduce a published housing example in Excel, R or Python.

Written and maintained by CASRAI Editorial Board

Published

On this page

Hedonic pricing estimates how the characteristics of a good contribute to its observed market price. Researchers use hedonic regression to separate attributes that are traded together, such as a home’s size, location and environmental surroundings. A coefficient describes a conditional price relationship; interpreting it as willingness to pay requires additional economic assumptions.

Definition in one paragraph

A hedonic pricing model treats a differentiated good as a bundle of attributes. The whole property is sold, while features such as an additional room or quieter surroundings have no separately observed transaction price. Regression estimates their implicit contributions by comparing observations with different bundles. The Handbook on Residential Property Prices Indices distinguishes estimating these contributions from constructing a quality-adjusted price index. The distinction determines the research question: an attribute study examines a coefficient, whereas an index study examines price movement after controlling for the composition of transactions.

This is a particular application of regression analysis, rather than a separate estimation algorithm. In environmental economics, it is a revealed-preference approach: observed housing choices provide evidence about tradeoffs involving environmental attributes. EPA’s economic-analysis guidelines emphasize that the market, available choices and information held by participants matter to this interpretation. A regression label alone does not establish that an estimated relationship measures preferences.

Hedonic pricing formula and every symbol explained

A linear specification models price directly:

Pᵢ = β₀ + Σⱼ₌₁ᴷ βⱼxᵢⱼ + εᵢ.

Here Pᵢ is the observed transaction price for observation i; i identifies the property; K is the number of included characteristics; j identifies a characteristic; xᵢⱼ is its measured value; β₀ is the intercept; βⱼ is a population regression coefficient; εᵢ is the unexplained error; and Σ means to add the K characteristic terms. For a continuous characteristic, βⱼ is the model’s price slope per additional unit, holding the other included characteristics constant.

A semi-log specification instead uses the natural logarithm of price:

ln(Pᵢ) = β₀ + Σⱼ₌₁ᴷ βⱼxᵢⱼ + εᵢ.

The symbols have the same meanings, but the coefficient now describes a change in log price. Write bⱼ for the estimated coefficient. For a change Δxⱼ in one untransformed characteristic, the corresponding fitted price ratio is exp(bⱼΔxⱼ). Here Δ means a change and exp is the exponential function with base e. The exact percentage contrast is:

Percentage contrast = 100 × [exp(bⱼΔxⱼ) − 1].

This compares exponentiated fitted log prices, or fitted geometric price levels. It does not by itself estimate the arithmetic mean transaction price. The approximation 100bⱼ applies to a one-unit change when the coefficient is small; it is not the exact finite-change calculation. A binary characteristic compares its coded 1 category with its coded 0 reference category. These readings also depend on the stated functional form: if a predictor is logged or interacted with another variable, the simple untransformed-predictor interpretation changes.

Worked example with published numbers and arithmetic

Table 11.6 of the handbook supplies a published housing example based on 796 observations from a Canadian city’s MLS data for 2008. Its semi-log model includes bedrooms, bathrooms, age and land area. The bedroom coefficient is 0.1156791, with standard error 0.0098159 and a 95% coefficient confidence interval from 0.0964108 to 0.1349473. These are the handbook’s example results, not a new empirical study.

For one additional bedroom, hold the other included variables constant and set Δx = 1. The calculation proceeds as follows:

  1. Change in fitted log price: 0.1156791 × 1 = 0.1156791.
  2. Fitted price ratio: exp(0.1156791) = 1.1226355606.
  3. Proportional contrast: 1.1226355606 − 1 = 0.1226355606.
  4. Percentage contrast: 100 × 0.1226355606 = 12.2636%.

The small-change approximation gives 100 × 0.1156791 = 11.5679%. The exact calculation exceeds that approximation by about 0.696 percentage points. That difference comes from exponentiation, not from fitting another model or adding a control variable.

Transforming the published confidence endpoints through the same increasing function gives 100 × [exp(0.0964108) − 1] = 10.1211% and 100 × [exp(0.1349473) − 1] = 14.4476%. Thus the transformed contrast is approximately 12.26%, with a 95% interval of 10.12% to 14.45%, conditional on this specification and the source’s inferential assumptions.

This comparison does not estimate the return from renovating a house. It compares the fitted price levels for different characteristic bundles in the published model. Changing a building could alter other attributes, and a causal intervention requires its own identification argument.

Interpretation: what the result means

The outcome transformation determines how to read the coefficient. The following guidance follows the handbook’s linear and semi-log equations; the percentage expression is the algebraic transformation shown above.

Model or output Reading guidance
Price in levels; untransformed continuous predictor The coefficient is a price-unit difference per predictor unit, conditional on the included variables.
Natural log price; untransformed predictor Use 100 × [exp(bΔx) − 1] for the finite percentage contrast in fitted geometric price levels.
Period dummy in a pooled log-price model Exponentiating its coefficient gives a quality-adjusted index relative to the omitted base period, under the model’s restrictions.

A positive bedroom coefficient therefore describes a positive conditional association in the example. It does not show that the same percentage applies to every city, every dwelling or every period. The sample and specification define the comparison being estimated.

For environmental valuation, EPA distinguishes a marginal implicit price from a large-change welfare measure. Under the relevant equilibrium assumptions, the slope of the hedonic price function can represent willingness to pay for a small attribute change. A substantial environmental improvement can change sorting, preferences represented in the neighborhood and the market relationship itself. Multiplying a local slope across a large improvement can therefore answer a different question from the intended benefits calculation.

EPA also separates capitalization studies, which examine changes reflected in property values over time, from an unrestricted interpretation as willingness to pay. Researchers should name the output they estimate. The downloadable data table below compares these research outputs, their inputs and the boundary on their interpretation.

Assumptions and common mistakes

Omitting correlated characteristics

A price model needs relevant property and location characteristics. An apparent environmental premium can also reflect school quality, crime or other neighborhood attributes correlated with the environmental measure. EPA identifies this as an identification problem, not simply a low model-fit statistic. A confounding variable can distort the coefficient even when the estimated association is precise. Explain which comparisons identify the attribute of interest and which competing explanations remain plausible.

Ignoring geography and market boundaries

EPA cautions against comparing widely separated markets as though households face the same choices. Nearby transactions may also share unmeasured influences or exhibit spatial dependence. Geographic controls, spatial methods and appropriately structured uncertainty estimates address different aspects of that problem; none removes the need to justify the market definition. For environmental attributes, the measure must also correspond to something participants can perceive or understand if the objective is to infer their preferences.

Confusing prediction with a stable attribute value

Closely related predictors can make individual coefficients unstable. The handbook notes that this matters especially when the coefficients themselves are the object of interest, even when fitted prices remain useful. Review multicollinearity and variance inflation before interpreting a single characteristic’s contribution. Deleting an economically relevant variable solely to improve coefficient stability can introduce omitted-variable bias instead.

Cleaning toward the desired answer

The handbook warns that removing unusual observations merely to stabilize coefficients can bias results. Distinguish entry errors from unusual but valid sales, document missing characteristics and explain exclusions. APA’s quantitative reporting standards likewise call for diagnostics, transformations and missing-data procedures to be described. A reproducible analysis preserves those decisions alongside the model output, rather than presenting only a final coefficient table.

How to report it: an APA example

APA’s JARS–Quant guidance calls for clear descriptions of design, measures, analysis and uncertainty. For a hedonic study, identify the transaction sample, period, geography, outcome transformation, characteristic coding and model specification before interpreting an estimate. The following APA-style results paragraph reformats the handbook’s published example; it does not claim a new analysis:

In the handbook’s 2008 housing example, an ordinary least squares model predicted natural log transaction price from bedrooms, bathrooms, age and land area (N = 796). Bedrooms were positively associated with log price, b = 0.11568, SE = 0.00982, t(791) = 11.78, p below .001, 95% CI [0.09641, 0.13495]. The model had R² = .4410 and adjusted R² = .4382. The bedroom coefficient corresponds to a fitted geometric price contrast of 12.26% per additional bedroom, holding the other included variables constant.

The source prints the test probability rounded to three decimal places; it should not be described as a probability of zero. Keep the coefficient’s confidence interval on its stated log scale, and explicitly label any interval transformed to percentages. An accompanying regression results table should preserve the distinction between estimates, uncertainty and fit. In an original report, also describe the actual standard-error procedure and diagnostics; the example’s conventional output does not establish that a particular procedure suits another dataset.

How to calculate it in Excel, R or Python

These snippets reproduce the exact percentage transformation of the published coefficient and confidence limits. They calculate the interpretation after estimation; they do not refit the hedonic model. Use natural-log coefficients, preserve their precision until the final display and apply the same transformation to both interval endpoints.

Excel

Microsoft defines EXP as the natural exponential. For the coefficient, enter:

=100*(EXP(0.1156791)-1)

The result is 12.2635560573 before display rounding. Replace the coefficient with 0.0964108 and 0.1349473 in two further cells for the lower and upper percentage limits. The formula already returns a percentage number: applying Excel’s percentage display format would multiply its displayed value again. Use an ordinary numeric format and label the result as percent.

R

The base R function expm1 computes exp(x) − 1 and accepts a vector. This expression returns the estimate, lower limit and upper limit, in that order:

round(100 * expm1(c(0.1156791, 0.0964108, 0.1349473)), 4)

The values are 12.2636, 10.1211 and 14.4476. Their order follows the input vector; it is not the order of the original regression table’s columns. No extra package is needed for this arithmetic.

Python

Python’s standard math module supplies the corresponding scalar function:

from math import expm1
values = (0.1156791, 0.0964108, 0.1349473)
print([round(100 * expm1(v), 4) for v in values])

This produces [12.2636, 10.1211, 14.4476]. Both R and Python document expm1 specifically for computing the exponential minus one, including small arguments where direct subtraction can lose precision. Regardless of software, the arithmetic cannot correct omitted characteristics, an inappropriate functional form or an unsupported causal interpretation.

Sources

Data: Hedonic research outputs and interpretation boundaries

An original crosswalk of hedonic research outputs compiled from the international residential property price handbook and EPA economic-analysis guidelines. Rows distinguish the object estimated, required information and interpretation boundary; this is a methodological compilation, not newly observed transaction data.

Hedonic research outputs and interpretation boundaries
Research outputObject estimatedInformation usedInterpretation boundarySource locator
Linear hedonic attribute slopePrice-unit change per characteristic unitIndividual prices and measured characteristicsConditional on the model and included characteristicsHandbook, chapter 5, equation 5.2
Semi-log hedonic attribute contrastChange in fitted log price, convertible to a geometric price ratioNatural log prices and measured characteristicsA fitted price contrast does not by itself establish an intervention effectHandbook, equations 5.3 and Table 11.6; EPA, pp. 7-39–7-40
Time-dummy hedonic price indexQuality-adjusted price movement relative to a base periodPrices, characteristics and period indicatorsPooled specification constrains characteristic coefficients across included periodsHandbook, chapter 5, paragraphs 5.11–5.12
Marginal environmental willingness to payLocal tradeoff implied by the hedonic price-function slopeMarket choices, prices and perceived environmental attributesEquilibrium and information assumptions matter; large changes require additional analysisEPA, chapter 7, pp. 7-36–7-40
Capitalization effect over timeAmenity changes reflected in property valuesProperty values across time and changing amenitiesEquating capitalization with willingness to pay requires additional stability assumptionsEPA, chapter 7, p. 7-40

Sources

Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · free to try

Ask about Hedonic pricing: formula and worked example

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →