The New York Review of Books

To the Editors:

As the two Commission members most responsible for the treatment of quality change in the report of the Advisory Commission on the Consumer Price Index (hereafter Advisory Commission on the CPI Report, or ACCR), we are writing to defend the report against the recent attack by Jeff Madrick [“The Cost of Living: A New Myth,”NYR, March 6]. Madrick damns the ACCR for three major sins—that its treatment of quality change and new products is “subjective” and “speculative”; that it “largely ignores” the adjustments for quality change already made in the CPI; and that quality change may not be much valued by the poorer consumer. The tone of Madrick’s review is suspicious of any effort that tries to improve our knowledge of price measurement, which is a central ingredient in every national statistic about productivity, real wages, and economic progress in general. His review looks for ulterior motives and focuses on imperfections, without giving credit for the honest attempt to provide an answer, imperfect as it may be, to these important economic questions.

Madrick supports his diagnosis of subjectivity by quoting one of us as calling the ACCR’s numbers “squishy.” Had Madrick used the whole quotation by Griliches, it might have set a more neutral tone for his review of the Commission’s work. The complete quotation said, “Such values are not chiseled in stone. They are squishy numbers but they are better than a firm but wrong zero” (The New York Times, December 16, 1996). The reference to “zero” refers to past evaluations, which simply assume (subjectively) that the CPI bias is zero in all categories except those which have been subject to careful review. In contrast, the ACCR extrapolates research on bias from one category to another when the categories seem related. To assume a zero bias in those categories not subject to careful research represents an extreme one-sided answer to the question as to whether those components of the CPI are biased. “They may be as likely to be subject to the average rate of bias of those components which have been subject to careful research as to no bias at all.”

Madrick chides the Commission for doing “little original research and much of the evidence it cites is old or covers only limited periods of time.” The Commission was not given a budget or a time frame suitable for doing major original research. We agree with everyone else that more research is desirable. However, we did cite all the previous research that we could find, and we reject an implication that our citation of previous research was selective. Indeed, some of the research is “old,” based on Gordon’s book which ends its coverage in 1983, but that research covers thirty-seven years, not a “limited period of time.” Other research, particularly that on apparel, personal computers, cataract and heart surgery, and pharmaceuticals, has been completed within the last few years.

Much of Madrick’s skepticism about our valuation of quality change and new products is based on recent research by Brent Moulton, director of research at the Bureau of Labor Statistics (hereafter BLS), partly in collaboration with Karin Smedley. A paper by Moulton and Smedley (M-S) and a discussant’s comment by one of us (Gordon) is forthcoming shortly in the Brookings Papers on Economic Activity and helps to clarify this controversy. To quantify what is at stake, the ACCR estimate of quality change and new product bias is 0.61 percent annually for the 1995-1996 period. Of this, 0.31 percent comes from appliances, radio, TV, personal computers, drugs, and medical care, all of which M-S accept at face value. M-S argue persuasively that our estimates for fruits, vegetables, shelter, and motor fuel are overstated, but replacing our estimates by theirs for these categories together only eliminates 0.11 points of our 0.61. In two other categories, we are not convinced that our estimates are too high. For apparel our alternative index is based on prices from the Sears catalog for a period when Sears was losing market share to other outlets offering superior prices and quality, and for automobiles any overstatement of the value of increased durability may be offset by the well-documented decline in defects of new automobiles and the reduced frequency of repairs, as documented by the J.D. Power survey, Consumer Reports, and other sources.

Even though, as Madrick and M-S argue, it may turn out that some of our estimates of quality change in particular items may be too high, others are likely to be too low. There are at least three sources of understatement in our estimates of quality change and new product bias. First, we did not take into account numerous but undeniable aspects of quality change—including (as stated above) reduced initial defects and repair frequency on new automobiles; improved sound quality, picture quality, and reliability of audio and video equipment; improved materials that reduce weight and corrosion in a host of consumer durable goods; improved safety of home power tools, lawn mowers, and other products; and improved energy efficiency of a wide variety of consumer goods.

Second, we did not take into account the value of the consumer surplus from the invention of new products. Among the evident benefits of inventions which we ignored are the value to consumers of being able to watch videos at home without spending money on babysitters and parking; the value of microwave ovens in making possible quick and convenient hot meals at home; the value of central air conditioning in opening up the possibility of living in inhospitable climates; the value of the ATM in making cash available at all hours in many new locations; not to mention the value of modern diagnostic and operative techniques in reducing the pain of medical procedures, and in converting previous procedures requiring inpatient hospital stays to simple procedures done on an outpatient basis. While some of these benefits of new products are hard to quantify, and we did not try, Hausman’s imaginative recent treatment of the value of cellular phones shows what can be done; he concludes that the CPI for telephone services is upward biased by 2.3 percent per year, compared to a bias estimate of 1.0 percent in the ACCR.

Third, we did not take into account evidence emerging from recent research that finds rates of upward bias in the CPI substantially greater than that incorporated into the ACCR. For instance, Gordon finds recently (see in the above-referenced exchange with M-S) that the CPI for television sets is upward biased by 8 percent per year, as contrasted to the 3.3 percent per year assumed by our report. Even more important, the Commerce Department publishes an alternative estimate of consumer price inflation (the deflator for personal consumption expenditures) that rose fully 0.7 percent slower than the CPI during 1995 and 1996; it is based on alternative and most think superior price indexes for computers, medical care, and airline fares, and some of these alternative indexes are compiled by the BLS itself.

In addition to questioning our estimates of the CPI’s quality-change bias, Madrick’s second main point is that the BLS already does extensive qualitative adjustments and that therefore our adjustments must be double-counting some of it. His argument leans heavily on the unpublished work of Moulton and Smedley, who estimated, at the time, that the BLS made adjustments that reduced the inflation rate in the CPI by about 2.6 percent in 1995. In the meantime, in their revised version (March 1997), the number has shrunk to 1.76. Excluding “outliers,” commodity pairs where the implicit price-quality differential exceeds 100 percent (which are simply noncomparable pairs and are not likely to reflect what either the BLS or we had in mind by the concept of “quality change”), the number shrinks to a mere 0.3 percent. At this point, the argument loses any quantitative significance.

We could leave it at that, but it is still instructive to discuss it, since it illustrates the substantive and communication difficulties in this field. Most of the reported “quality adjustment” by the BLS, 1.65 out of the 1.76 percent that includes outliers, comes from “linking” procedures, where a missing item is replaced by another. No judgment at all is made about the quality differential between the new and old item. The price change during the link period is imputed, by using either the inflation rate in the overall CPI or of other commodities in the particular class. These adjustments are the consequences of the BLS sampling procedures, which focus on pricing a very specific item in a particular store and city. There are thousands upon thousands of such commodities in the market but only a small fraction of them are in a particular store at any time. The pricing agent has to deal with rapid turnover and high probability of stockout. Roughly one out of two items disappear sometime during the year and have to be replaced by a different item in the same general class: a larger versus a smaller package of yogurt, a blue raincoat versus black, a refrigerator with its freezer at the bottom rather than at the top. But this churning is not what we had in mind by “quality change,” which rather involves the appearance of new and improved goods, such as the increased variety and freshness of vegetables and fish due to the improving transport facilities and the globalization of trade, the substitution of laporascopic procedures for gallstone operations, and more.

The fundamental point, which Madrick ignores, is that for most categories the extent of current BLS quality adjustments is irrelevant, because most of our estimates of quality-change bias are valid independently of how the BLS arrives at its estimates of price change, and the extent to which its own quality-change adjustments are large or small. Most of our bias estimates are based on the collection of price data from independent sources and the careful quality adjustment of those independent data. The difference between these quality-adjusted independent price indexes and the corresponding CPI indexes (however they are quality adjusted, in a major or minor way) forms the basis of by far the majority of our estimates of bias.

Madrick’s third objection is that our recommendations are unfair to the poor and the elderly. Life is unfair to the disadvantaged members of society, but this has little to do with the CPI, which is a measure of prices paid by the average, unaging household, in an unchanging socioeconomic environment. Some of the poor (whether elderly or not) should be provided with a better safety net, but the CPI is only intended to measure changes in the average price level, not the special circumstances of particular groups.

The real issues involved with the special circumstances of the elderly involve health costs, which rise with age and with some improvements in medical technology. These are real drains on the budgets of the elderly, and indeed on all of us. A true cost-of-living index measures the change in income required to maintain a given standard of living for a person or household of a given age. The opportunity provided to live longer by improved albeit more expensive medical technology represents an increase in the standard of living, not an increase in the cost of living.