Hedonic Regression Models Using In-House and Out-of-House Data: Two Data Sources Used by the Bureau of Labor Statistics to Create Hedonic Regression Models Have Their Own Distinct Advantages and Disadvantages; the Bureau Performs Research on Each Data Source in an Effort to Meet Its Current and Ever-Evolving Future Needs

Article excerpt

The research leading up to the publication of this article was conducted under the CPI initiative to expand the scope of developing hedonic regression models for quality adjustment purposes to more items within the CPI market basket. The primary focus of the article is to provide a detailed analysis of the hedonic modeling process and to illustrate the characteristics of two data sources the Bureau of Labor Statistics has chosen to utilize in its ongoing research on hedonic-based quality adjustment methods.

Early research by BLS personnel and a significant portion of the current research done by the CPI staff in this area rely upon the existing sample of CPI data for the creation of hedonic regression Models. (1) When it was recommended that the Bureau expand its use of hedonic models for quality adjustment purposes to more items within the CPI, situations arose in which the existing sample size of the items chosen were deemed insufficient to support the creation of hedonic models. To alleviate this problem, supplemental samples were designed and collected exclusively for hedonic modeling purposes. (2) Despite the Bureau's having full control over this type of sample data, such an "in-house" prescription was not seen as a cure-all, because designing and collecting these data exhausts many BLS resources. Accordingly, the Bureau was led to investigate the use of hedonic models created with market data purchased from private firms that specialize in collecting point-of-sale observational data. (3) Purchased, or out-of-house, data offer many enhancements over in-house data, but are costly and have their own sets of limitations.

With home-based telephones (corded or cordless), the Bureau has an opportunity to compare the process and results of using both in-house and out-of-house data in the creation of hedonic regression models. This article discusses the issues of data quality, the specification of a model, and the application of hedonic quality adjustments to substitutions in the CPI sample. Empirical evidence and quantitative data support the topics addressed. The next section examines the characteristics of the data. Following that section, the results of the models are presented, and a discussion illustrates how they could have been used in quality-adjusting substitutions in the CPI. The final section is a follow-up of what has changed with the data and presents a brief conclusion.

The data

Sources. The price data used in the analysis that follows were collected by BLS data collectors for the CPI and by the NPD Group, a private firm that collects and sells point-of-sale marketing information. The CPI sample consists of price data for home-based telephones from the official CPI sample and a specially designed supplemental research sample that was created and collected in order to increase the robustness of the existing sample to facilitate hedonic modeling research. CPI statisticians drew the supplemental sample on the basis of current CPI sampling procedures. The CPI sample is designed to be representative of consumer spending habits and is distributed across the country and across all types of retail outlets. The NPD Group sample is a collection of point-of-sale data for home-based telephones from various partnered retail outlets from across the country. The Bureau purchased these data from NPD to explore the use of out-of-house data in developing hedonic regression models. NPD offers point-of-sale data for a wide range of consumer goods and services. Most of the company's clients are private firms that use the sales data to make marketing decisions.

Sample design. All of the price data for this study were collected in August and September of 2000. It was during these months that the Bureau was able to collect its supplemental sample. The agency purchased roughly 2 years of telephone data from the NPD Group, but in order to make timely comparisons with the CPI sample, only data from the aforementioned months were used in the upcoming analysis. …