Academic journal article
By Liegey, Paul R., Jr.
Monthly Labor Review , Vol. 117, No. 5
The Bureau has adopted hedonic techniques to adjust price changes for apparel, which is subject to frequent variations in characteristics; a comparison of the new indexes with those produced using traditional methodology yields some surprising findings
If a poll were taken of professional economists and statisticians, in all probability they would designate (and by a wide margin) the failure of the price indexes to take full account of quality changes as the most important defect in these indexes. And by almost as large a majority, they would believe that this failure introduces a systematic upward bias in the price indexes--that quality changes have on average been quality improvements.
...Even the concept of quality change is not free of difficulty. Changes in buyers' tastes will lead to the appearance of new goods--an uncontroversial example would be fashionable apparel--which are not improvements judged by either previous or subsequent tastes, and the line separating taste changes from quality improvements will depend on the time span invoked.
Since January 1991, economists working on the Consumer Price Index (CPI) have been making direct adjustments for quality differences in the price observations used to calculate apparel commodity indexes.(1) These ajustments compensate for the effect of quality changes on apparel prices and are based on dollar-value estimates for product features developed through regression analysis. Previously, quality adjustments were handled implicitly when estimates of price changes were made through a procedure known as imputation.
The Bureau of Labor Statistics computes apparel indexes and other consumer price indexes from the average change in prices for a sample of consumer items in a sample of retail outlets. The price change for each sample item is the ratio of its price in the current period to its price in the previous period. A price change is valid for index use only if the item is the same or at least "comparable" between the two periods. However, measuring apparel price changes is particularly difficult; these "fashion-driven" goods are subject to frequent product changes and, in some categories of apparel, items rarely remain in the outlets from one fashion season to the next.
When product changes occur, BLS determines whether price comparisons are appropriate for use in the index by carefully reviewing the product specifications (or "characteristics") for the changing item. Because consumer price indexes are designed to measure only price change, the quality of the items in the samples must be held constant. When an item in a sample is no longer available for consumer purchase, it is replaced--ideally, by the most similar item in the same outlet--to provide a continuous measure of price change. BLS refers to the replacement of discontinued items with currently available items as "substitution."
BLS economists called commodity analysts determine whether substitute items are equivalent in quality by comparing data on the characteristics (such as fabric, lining, and so forth) of each item. Price change for substitute items of the same or similar quality is then measured by comparing the price of the replacement item in the current period with the price of the discontinued item in the previous period. Substitutions of this type are known as "comparable" substitutions.
Before January 1991, if the commodity analyst found that a replacement item was not of the same or similar quality, a price change was imputed (estimated) for the substitution by taking the average of all (nonimputed) price changes within the same geographic pricing area(2) that had quality characteristics similar to the replacement item.(3) Substitutions of this type are termed "noncomparable" substitutions. When the CPI uses imputed price changes for noncomparable substitutions, the effective sample size is reduced. This may increase index variance and may also reduce index accuracy. …