Report available to the public as a courtesy of HousingEconomics.com
NAHB has developed a model that estimates the price of a home—and the impact various physical and neighborhood features have on the price—using data from the American Housing Survey (AHS). The AHS is a nationally representative survey of housing units conducted by the U.S. Census Bureau and funded the Department of Housing and Urban Development (HUD). NAHB recently updated the model as part of a research paper presented at HUD’s inaugural AHS Users Conference.
The updated model is now available online in a form that readers with compatible web browsers and internet service providers can use interactively. In addition to announcing the new house price estimator, this article provides background on the model and shows some of the results that can be produced from it, such as:
A standard new home will cost more if built in metropolitan California than elsewhere in the west, or in any other Census region.
An extra full bathroom adds an average of about $43,000 to the value of a standard new home built in a Southern suburb, more than any other house feature considered.
Locating the same home on a body of water adds an average of nearly $90,000 to the value of the standard new home, more than any other neighborhood feature considered.
Abandoned buildings in the neighborhood, on the other hand, reduce the value of the standard new home by about $28,000.
Due to the model’s somewhat limited geographic detail, these results are averages across a broad region— i.e. suburbs of metropolitan areas across the South—rather than estimates for a particular house in a specific location. The model is also based on values of existing homes and therefore reflects demand for particular features, as well as the cost of providing them; so the estimated price effect of, for example, a bathroom may be different from the actual cost of constructing a new bathroom.
Background on the Model
The NAHB model uses a conventional statistical technique to control for the impact various characteristics have on the price of a home. The amount of geographic detail in the AHS is somewhat limited, so the model estimates an average home price across a broad Census region—not the price of a particular home in a specific neighborhood.
In this case, price is based on the owner’s estimate of the home’s value, rather than an independent appraisal or recorded transaction price. Accuracy of owners’ estimates of the value of their homes has been systematically investigated by Kiel and Zabel. They found that owners had a tendency to overvalue their homes by about 5 percent, but that (except for length of tenure in the current residence) the estimates were not related to characteristics of the owner, housing unit, or neighborhood. This suggests that the relative impacts of various features on house prices may be estimated in a reasonable way, although the estimated impacts overall could may be inflated slightly.
The AHS contains considerable detail on the physical features of the homes and the neighborhoods in which they're located. The updated NAHB model is based on the 2009 public use AHS file, which contains information on 2,776 different variables.
Even so, no conceivable data set or statistical model can capture every feature that has an impact on the price of a commodity as complex as a house so there's always a chance that a particular feature in the model is acting partly as a proxy for others. In the statistical literature, this problem is called omitted variable bias.
NAHB’s strategy in developing a model from AHS data is to experiment with a large number of variables, including interactions, and for the most part retaining those that have a statistically significant and theoretically plausible impact on house price—taking care not to delete variables if this materially alters the estimated of other variables in the model, in order to avoid increasing omitted variable bias unnecessarily.
A description of all variables used in the current version of the NAHB house price estimator is contained in an appendix, available in the “Additional Resources Box” at the top of the article.
This article illustrates some of the results the estimator can produce by analyzing a benchmark or standard new (built after 2007) single-family detached home, defined by the following “average” features:
2,150 square feet of living space. A rounded number between the median square footage reported by the Census Bureau’s Manufacturing and Construction Division for single-family homes completed in 2009 (2,135) and 2010 (2,169), based on its Survey of Construction (SOC).
2 full bathrooms, one half bath. The SOC shows that the median for new single-family homes completed has been 2 1/2 baths for roughly the past decade.
3 bedrooms. The median for new single-family homes completed every year reported by the Census Bureau from the SOC.
Built on a slab foundation. Since 2002, the SOC shows that over half new single-family homes completed have neither a basement nor crawl space.
A garage. Since the mid-1980s, the SOC shows that over 70 percent of single-family homes completed have been built with garages.
Central Air Conditioning. The SOC shows that nearly 90 percent of single-family homes completed have central air conditioning.
A fireplace. For 2010, the SOC shows slightly under half of new single-family homes completed had a fireplace, but that is the first time the share was under 50 percent since 1974.
A dining room. This is not a characteristic captured by the SOC. In its new construction column for owner-occupied housing the 2009 AHS printed report shows that 2.2 million out of 3.8 million units have a separate dining room (http://www.census.gov/prod/2011pubs/h150-09.pdf, p47).
3 miscellaneous rooms. The AHS printed report (p41) shows a median of 7 rooms in new owner-occupied homes. This excludes baths. With 3 bedrooms and a dining room already in the specification, that leaves 3 other rooms to account for, which could include a kitchen, living room, den, home office, etc.
Satisfactory shopping (grocery or drug stores) within 15 minutes, but none of the specific community or neighborhood characteristics that the estimator can analyze, as shown in the appendix. Satisfactory shopping is treated as the default as it is reported by nearly 90 percent (3.4 out of the 3.8 million) new homeowners in the AHS (printed report, p 50).
As previously noted, the geographic detail available in the AHS is somewhat limited. The four principal census regions and the area’s urban status (central city, suburb, or and non-metro) are identified, but not specific state or local jurisdictions. Information is available for some metro areas (generally aggregations of contiguous counties based on inter-county commuting patterns), but there are generally too few observations in any one metro to treat separately. Based on the number metropolitan areas identified in California, however, it was possible to carve these out and treat them separately from the rest of the West, as a “region" that captures almost all of metropolitan California.
Figure 1 shows how the estimated price of the standard new home varies across these general locations.
The general tendency is for the house price to be higher in the Northeast and West Census regions, as well as in central cities and suburbs. The price tends to be lowest if the home is built outside of a metro area, although some regional variation exists regardless of urban status. The model estimates that the standard new home will cost over $500,000 if it’s built in a suburb of one of the large California metro areas, but only about $155,000 if built outside of a metropolitan area in the Midwest Census region. In general, suburbs show higher prices than their companion central cities (which include the entire political boundaries of a city, not just a central business district or downtown area). This could reflect some generally pervasive difference in the types of neighborhoods that tend to exist in suburbs and central cities—although, as described elsewhere in this article, the house price estimator controls for a substantial number of neighborhood characteristics.
It’s important to remember that a result from the house price estimator is an average price across a broad Census region--not the price of a particular home in a specific neighborhood. House prices within a region may vary substantially. The price of a home in a suburb of Washington, D.C, for example, may be different from the price of an otherwise similar home in the suburb of a different Southern city.
Physical Features of the Home
NAHB’s house price estimator can be used to study what happens when certain features of the home change. Figure 2 shows some examples for a standard new home built in a southern suburb.
With no modification, the estimated average price of the standard new home in a southern suburb is $203,874. Adding 500 square feet of living space with no other changes increases the estimated price by roughly $13,000.
Holding the square footage (and other features) constant, adding another bed or miscellaneous room increases the estimated price by less than $10,000. The impact of adding other types of rooms is greater. The greatest impact comes from adding a third full bathroom, which increases the estimated price by about $43,000. Eliminating the fireplace reduces the estimated price by about $24,000.
When considering estimates such as these, it’s important to remember that unknown characteristics are always a potential problem. The AHS, for example, doesn’t collect information on the size of bathrooms or bathroom features such as whirlpool baths or multiple sinks, so a statistical model based on the AHS can’t control for these features. If homes with more bathrooms also tended to have bathrooms that were larger and were more likely to contain whirlpool baths and more than one sink, the estimated value of an additional bathroom could be in part picking up the value of these extra amenities.
With this caveat in mind, stakeholders may find results from the house price estimator useful in a variety of settings. Home builders might use the estimator to help determine if the cost of providing a particular amenity will be valued by consumers. Households contemplating a home purchase could use it as a preliminary search tool, to get a rough idea of likely price differences for different size and amenity packages. Owners of existing homes could use it to get a rough idea of how much it would cost to trade up to a home that’s newer, larger, or has more amenities in order to make a decision on whether or not to move. Homeowners could also use the estimator to approximate how particular remodeling jobs would influence the value of their homes. Remodelers might find it useful to provide this type of information to prospective customers.
The house price estimator also allows users to see what happens to the price of a home when the neighborhood changes. Figure 3 illustrates this, again from the starting point of a standard new home built in a suburb in the South census region.
The figure shows that the neighborhood amenity with the largest positive impact on the price is a location on a body of water (such as a lake, river or ocean). Moving the standard Southern suburban new home to an otherwise similar neighborhood on the waterfront increases its estimated price by nearly $90,000. Nearness to adequate public transportation comes in a distant second, raising the estimated price by about $26,000.
Neighborhood features may also be “disamenities,” in that their presence tends to reduce the price of the home. Among the disamenities included in NAHB’s house price estimator, the presence of abandoned buildings within a half block (roughly 300 feet) of the home turned out to have the greatest negative impact, reducing the estimated price of the standard new home in a Southern suburb by about $28,000. Bad roads, odors, lack of adequate shopping, buildings with metal bars on their windows, and litter each reduced the estimate price by more than $6,000 also had very significant negative impacts.
That so many neighborhood disamenities turned out to have a statistically significant impact on house price is perhaps surprising. It’s likely that some of these disamenities—abandoned buildings and litter, for instance—tend to be found in the many of the same neighborhoods, and we might expect these correlations to make it difficult to tease out individual impacts; but the statistical relationships between these characteristics and house prices have proven to be very strong.
The effect of many neighborhood characteristics depends on whether the homes are located inside a metropolitan area or not. For example, it turned out to be possible to estimate a significant impact for nearby shopping or public transportation only for homes located inside metro areas. And although the presence of litter tended to depress house values in all locations, the effect was considerably stronger inside metro areas.
Because the house price estimator helps price neighborhood characteristics, developers may use it to help evaluate the desirability of potential building sites. Home owners, community associations, and local governments could also use it to estimate how certain public policies (such as providing public transportation, or finding a use for abandoned buildings) are likely to impact home values in their neighborhoods.
Accessing the Model Online
The graphs in this article illustrate only a few examples of changes in home and neighborhood features that can be investigated with the NAHB house price estimator. For example, the examples shown here were based on a new home, but the estimator can provide results for homes of different vintages as well.
Running the house price estimator requires Microsoft Excel. Excel’s security setting must be adjusted to allow macros to run. In the past, NAHB has encountered compatibility problems with the estimator and Mozilla Firefox, as well as particular versions of some other web browsers. In addition, some internet service providers may apply security or other procedures that prevent the estimator from running properly. If you have trouble getting the estimator to run initially, NAHB suggests you try accessing it from a different computer with a different browser.
Technical details of the statistical model described in this article and used to generate the house price estimator are can be found in the paper presented at the AHS Users Conference. Readers may also contact Paul Emrath, NAHB’s Vice President of Survey and Housing Policy Research, directly with questions: 1-800-368-5242 x8449; email@example.com.
The technique is usually called hedonic regression, a term used loosely to refer to a method that estimates the price of a good based on its characteristics. Hedonic price estimation dates back at least to Waugh (1928), although Griliches (1961) and Rosen (1974) are usually credited for establishing it as a widely used technique. Theoretic work on the technique continues, including a recent paper by a Nobel laureate in economics (Heckman et al, 2003). A relatively recent example from the housing literature is Coulson et al (2002).
References: Edward Coulson, Seok-Joon Hwang, and Susumu Imai (2002) “The Value of Owner-Occupation in Neighborhoods,” Journal of Housing Research, pp. 153-174.
Zvi Griliches (1961) “Hedonic Price Indexes for Automobiles: An Econometric Analysis of Quality Changes,” National Bureau of Economic Research, The Price Statistics of the Federal Government, pp.173-196.
James J. Heckman, Rosa Matzkin, and Lars Nesheim (2003) “Simulation and Estimation of Nonadditive Hedonic Models,” National Bureau of Economic Research Working Paper 9895.
Sherwin Rosen (1974) “Hedonic Prices and Implicit Market Product Differentiation in Pure Competition,” Journal of Political Economy, pp. 34-55.
The metro areas included in the California region are Bakersfield, Fresno, Los Angeles- Long Beach, Modesto, Oakland, Orange County, Riverside-San Bernardino, Sacramento, San Diego, San Francisco, San Jose, Santa Barbara-Santa Maria, Santa Rosa, Stockton-Lodi, Vallejo-Fairfield-Napa, and Ventura.
For more information about this item, please contact Paul Emrath at 800-368-5242 x8449 or via email at firstname.lastname@example.org.