Regression: Helpful Tool or False Precision?
Regression analysis occupies an unusual position in appraisal methodology. Among the analytical tools available for supporting adjustments, it is simultaneously the most statistically sophisticated, the most frequently misapplied, and the most misunderstood — by both the appraisers who use it and the reviewers who evaluate it.
Proponents argue that regression is more objective and rigorous than paired sales or grouped data analysis because it controls for multiple variables simultaneously and produces statistically grounded results. Critics worry that it is being applied in contexts where the data cannot support it, producing the illusion of precision where none actually exists.
Both sides have a point. Regression is a legitimate and valuable analytical tool when the conditions for its use are met. It becomes a problem — a significant one — when those conditions are not met and the appraiser applies it anyway. Understanding the difference is essential for anyone who uses regression to support adjustments or evaluates work that relies on it.
What Regression Analysis Actually Does
At its core, regression analysis is a statistical method for measuring the relationship between variables. In its simplest form, simple linear regression measures the relationship between one independent variable and one dependent variable. Multiple regression — the form most relevant to appraisal work — measures the relationship between multiple independent variables and one dependent variable simultaneously.
In the appraisal context, the dependent variable is typically sale price (or price per square foot, or some other price metric), and the independent variables are property characteristics — size, age, condition, location, lot size, bedroom count, and so on. A well-specified regression model estimates the contribution of each characteristic to sale price, holding all other included characteristics constant.
That "holding all other things constant" feature is regression's key analytical advantage over paired sales. Paired sales can only isolate one variable at a time, and doing so requires finding closely matched transactions. Regression controls for multiple variables simultaneously across the entire data set. When the conditions for its use are met, that is a genuine methodological strength.
When Regression Works
Data Density
Regression works best in markets with abundant, relatively homogeneous sales data. High-volume residential markets — suburban subdivisions, condominium complexes, similar single-family neighborhoods — are environments where regression tends to perform well. The method rewards data density. The more observations available, the more reliably the model can estimate the contribution of individual characteristics.
As a rough general guideline, a meaningful regression model typically requires at least ten observations per variable included in the model — and preferably more. A model with five independent variables therefore needs at least fifty sales observations to produce reasonably reliable results. In practice, larger data sets are almost always better, and the adequacy of any given data set depends on its homogeneity as well as its size.
Statistical Reliability Indicators
A well-executed regression model reports indicators of statistical reliability that appraisers should understand and be prepared to explain. The R-squared value measures how much of the variation in sale prices is explained by the model — a higher value indicates a better-fitting model, though a very high R-squared in a small data set may indicate overfitting rather than genuine explanatory power.
P-values for individual coefficients indicate whether the estimated value contribution of each variable is statistically distinguishable from zero. A p-value above 0.05 or 0.10 (depending on the significance threshold applied) suggests that the coefficient may be the result of random variation in the data rather than a genuine market relationship. Confidence intervals around the coefficients show the range within which the true value contribution likely falls.
These are not academic formalities. They are the tools that tell you whether the regression results are meaningful or whether they may be the product of chance in a limited data set. An appraiser using regression to support adjustments should understand what these indicators mean and be prepared to explain them to a reviewer, a client, or a court.
Statistical output from a regression model is not self-validating. The numbers mean something only if the underlying data is adequate and the model is properly specified. Presenting regression output without understanding what the diagnostic statistics indicate is a significant professional risk.
When Regression Becomes a Problem
Inadequate Data
The most common and most serious misapplication of regression in appraisal work is applying it to data sets that are too small to support statistically significant results. This problem was introduced in Article 4, and it warrants deeper examination here.
When a regression model is applied to a limited data set — ten or fifteen sales in a rural market, a specialty property segment, or a rapidly changing market — the model will still produce coefficient values for each variable. Those values will appear precise. They may be presented in a report with apparent confidence. But they will not be statistically reliable.
What is happening is that the model is fitting the noise in the data as well as the signal. With a small number of observations, random variation in sale prices has an outsized influence on the estimated coefficients. The result is a model that describes the particular sample well but does not reliably represent what the broader market shows. Applying those coefficients as adjustments produces results that look quantitative and sophisticated but are actually no more defensible than an unsupported judgment.
The damage is worse than simply having a weak adjustment. It is that the regression output implies a level of analytical rigor that is false. A reviewer or opposing expert who examines the statistical diagnostics will identify the problem immediately. An appraiser who cannot explain why the model's results are reliable — or who did not examine the diagnostics at all — faces serious credibility consequences.
Model Misspecification
A second significant problem is model misspecification — building a regression model that omits important variables, includes irrelevant ones, or uses the wrong functional form. When important variables are omitted from a regression model, their influence is absorbed into the coefficients of the variables that are included, biasing those coefficients in unpredictable ways.
For example, a regression model that estimates the value contribution of garage space but omits overall condition will produce a garage coefficient that partly reflects condition differences between properties, not just garage differences. Applying that coefficient as a garage adjustment will produce an estimate that is biased to an unknown degree in an unknown direction.
Avoiding model misspecification requires genuine familiarity with the market, careful thought about what variables drive value in the subject property segment, and testing of the model's results against other market evidence. It is not something that can be done mechanically.
Regression as One Input Among Several
The most defensible use of regression in appraisal work is as one analytical tool among several, rather than as a standalone answer. When regression results, paired sales analysis, and grouped data analysis all point toward a similar adjustment range, the convergence of methods provides strong support. When regression results diverge significantly from other market evidence, that divergence is a signal — either the regression model has a problem, or the other evidence does, and the discrepancy deserves investigation before either is applied.
Regression results should also be tested for reasonableness against market participant knowledge. An adjustment derived from regression that no market participant — no buyer, seller, broker, or experienced appraiser familiar with the market — would recognize as reflecting actual buyer behavior is a signal that something in the model needs reexamination. Statistical results that are inconsistent with market reality are almost always the product of a modeling problem, not evidence that the market behaves in surprising ways.
The question to ask about any regression result is not "does the math work?" It always works. The question is "does this result reflect what is actually happening in the market?" Those are very different questions.
Communicating Regression Results
When regression is used to support adjustments, the report needs to explain the methodology in terms that an intended user — including a non-statistician judge or attorney — can follow. This means explaining what the model does, what data it was applied to, what the results showed, and why those results are reliable.
It also means acknowledging limitations. If the data set is smaller than ideal, say so and explain why the results are still meaningful despite that limitation — or explain why a qualitative approach informed by the regression results is more appropriate than applying the coefficients directly. Transparency about limitations is more credible than false confidence.
In Yellow Book and litigation contexts, regression methodology is particularly susceptible to expert challenge. Opposing experts will examine the model specification, the data set, the diagnostic statistics, and the reasonableness of the results. An appraiser who can explain and defend their regression methodology clearly and specifically is in a strong position. One who cannot is not.
The Bottom Line
Regression is a legitimate and valuable tool when the data supports it, the model is properly specified, and the appraiser understands what the results mean. It becomes a problem when it is used to manufacture the appearance of precision that the underlying data cannot justify.
The question to ask is not whether regression produces a number — it always will. The question is whether that number means anything. Answering that question requires understanding the statistical foundations of the method, examining the diagnostic indicators, testing the results against other market evidence, and being honest about what the data can and cannot support.
Have a complex valuation methodology question or need expert witness support? Blue Ridge Valuation Services LLC provides appraisal consulting and litigation support services. Visit
blueridgevaluationservices.com to get expert valuation assistance today.
Next in the series: Article 8 — Supporting Time Adjustments in Changing Markets
