MEASUREMENT ISSUES IN THE CONSUMER PRICE INDEX*


Bureau of Labor Statistics

U.S. Department of Labor

June 1997

*Prepared in response to a letter from Jim Saxton, Chairman of the Joint Economic Committee, to Katharine Abraham, Commissioner of the Bureau Labor Statistics, dated January 28, 1997.

I. Introduction

This paper on the Consumer Price Index (CPI) has been prepared in response to a letter from Jim Saxton, Chairman of the Joint Economic Committee, to Katharine Abraham, Commissioner of the Bureau Labor Statistics, dated January 28, 1997. The letter requested "a serious, detailed response by the professional career staff of the Bureau of Labor Statistics (BLS)…to fully inform Congress, the media, and the public of the central issues raised by the Boskin Commission report, and the BLS response to them."

The following pages address the definition and measurement objective of the CPI, together with the BLS response to the estimates of bias put forward in the final report of the Advisory Commission to Study the Consumer Price Index and to the specific recommendations made to the Bureau by the commission. Decisions concerning whether and how the CPI should be used in escalation, however, lie outside the purview of a statistical agency such as the BLS, so the budgetary implications of any bias in the CPI are not discussed.

The Advisory Commission to Study the Consumer Price Index (CPI), established by the Senate Finance Committee and chaired by Michael Boskin, delivered its final report on December 4, 19961. The present paper summarizes the response of the Bureau of Labor Statistics to the findings of the commission.

The advisory commission compares the U.S. CPI to a hypothetical ideal measure of the change in the cost of living and concludes that in several respects the CPI is biased relative to this standard. The categories of bias discussed by the commission include: substitution bias (due in large part to the fixed-weight nature of the index), outlet bias (which may occur if the benefits to consumers from switching to discount outlets are not accounted for in the index), quality change bias (which results when the quality differences between the goods priced in two consecutive periods cannot be accurately measured and deducted from the accompanying price difference between the goods), and new product bias (due to the failure to reflect adequately the value to consumers of new products that are introduced into the market). The commission, using empirical evidence and the members' own judgments about the magnitude of these biases, concludes that the CPI overstates the true cost-of-living change by 1.1 percentage points per year. The commission also discusses the fiscal impact of CPI bias through its use as an adjustment factor in several areas of the federal budget, including Social Security, military and civil service retirement, and the income tax.

The advisory commission emphasizes that the U.S. economy is exceedingly complex and dynamic, with the available offerings of goods and services constantly changing. It also acknowledges that index number construction is a complex and difficult task. It recommends that the BLS make several changes in the methods used in constructing the CPI, including more frequent updates of the market basket and expenditure information required by the index and the use of formulas more consistent with the theoretical cost-of-living concept. Most prominently and fundamentally, the commission recommends explicitly adopting the cost-of-living index as the measurement objective of the CPI, replacing the current index by two indexes-a monthly index that takes account of the changing market basket and a second annual index calculated using a "superlative" formula and subject to revision-and using geometric means for aggregating elementary price quotes. The commission also makes several intermediate and longer run methodological and research recommendations.

The objectives of the present paper are: first, to discuss the relationship of the CPI to the conceptual cost-of-living index; second, to review and critique the advisory commission's estimates of bias; and third, to respond to the detailed recommendations made by the commission. The advisory commission's report also raises a number of issues that will not be addressed here. These include: first, the various uses of the CPI; second, the revenue impacts of changes in the CPI; third, recommendations made to Congress and to the economics profession; fourth, separate indexes for demographic subgroups of the population; and fifth, the potential impact of including social and environmental factors (such as crime, AIDS, and pollution) in an official index. The first three of these topics generally involve the formulation of policy and so are outside the purview of the Bureau of Labor Statistics. The latter two are topics on which relatively little research has been conducted2.

II. The CPI in a Cost-of-Living Framework

The CPI is a measure of the average change in the prices paid by urban consumers for a fixed market basket of goods and services. Measuring price change through the use of a fixed market basket has a long history in economics, going back to the early 1700's in England3. Over time the state of the art for specification of the market basket has evolved from a judgmental selection of representative items to the modern survey-based approach of defining a comprehensive categorization of goods and services, selecting a representative sample of items to track, and weighting them according to the consumption of the average consumer during a base period.

The CPI is computed using an index number formula, known as the Laspeyres formula, that measures the change in the cost of a fixed market basket4. In this formula the quantities of the goods and services purchased by urban consumers during a base period serve as the weights for the prices, so that the value of the market basket represents the cost of purchasing the same items as were purchased during the base period. The CPI measures the current cost of the market basket relative to its cost during a reference period. In other words, the Laspeyres price index answers the question: "What is the value of the base-period market basket in today's prices?" An important underlying assumption in the comparison of market basket values is that the price changes are measured net of any changes in the quality of the goods and services that may have occurred. Indeed, adjusting for changes in product quality is one of the main problems facing index number practitioners and is a problem to which the BLS devotes considerable effort.

The computation of the CPI is an undertaking that involves the collection of prices from approximately 7,300 housing units and 22,500 retail/service establishments each month. The CPI is constructed in two stages. In the first stage, often referred to as the "lower" level, the elementary indexes are constructed. These indexes are the 206 item category indexes constructed for each of the 44 urban areas from which prices are collected for specific items in specific outlets5. In the second stage, the "upper" level, the BLS combines the 206 item indexes formulated for the 44 index areas. Thus the overall U.S. CPI is an aggregation of 9,064 indexes.

To construct the market basket of goods and services, the BLS uses information from the Consumer Expenditure Survey (CEX). This is a household survey that collects comprehensive data on consumer spending. Currently the expenditure base period of the CPI is 1982-84, but with the revision scheduled for 1998 the base period will change to 1993-95. To measure price changes, a sample of outlets is selected from locations identified by consumers from the Point of Purchase Survey (POPS). Specific sample items are then selected from each sample outlet, to ensure that the market basket is representative of what households purchase and where they shop. To keep up with changing shopping patterns, the Bureau replaces about 20 percent of the outlet sample in every year, thus turning over the sample every five years.

The CPI is used for many purposes, but measurement of changes in the cost of living is one of the most important of these. The BLS has for many years used the concept of the cost-of-living index as a framework for making decisions about the CPI and accepts the COLI as the measurement objective for the index6. The cost-of-living index is a theoretical construct, however, not a single or straightforward index formula readily amenable to practical use.

The cost-of-living index compares the cost to the consumer at different points in time of maintaining a constant standard of well-being, without restrictions on the market basket. It is a theoretical concept based on the well-being of the individual consumer, so that additional assumptions about how to apply it as a measurement objective for an aggregated set of consumers (such as the U.S. urban population) must be made. In addition, for an aggregate measure, assumptions must be made about the implications of the distribution of prices paid for the same good across markets. The general cost-of-living theory does not prescribe how any compensation for changes in well-being would or should be administered. Thus, while the CPI may be described formally in the context of a cost-of-living index, there is no single all-purpose definition of this target7.

In the most general sense, the cost-of-living index answers the following question: "What is the minimum change in expenditure that would be required in order to leave a specified consumer unit indifferent (or as well off) between a specified reference period's prices and a comparison period's prices?"8 The consumer's well-being depends on many aspects of life other than market goods and services, e.g., environmental quality and amenities (such as clean air and low crime), goods provided through taxes (such as national defense and fire protection), health status, and future consumption goals (which depend on both current and expected future income, and savings). All of these aspects of life can, and do, change over time along with commodity and service prices. Most of these also are difficult to measure, and it would be even more difficult to translate them into measured increments to well-being. The cost-of-living index approximated by the CPI is a subindex of the all-encompassing cost-of-living concept, specifically a subindex that is conditional on the excluded factors that affect consumer well being, such as health status and the quantity and quality of government-provided goods and services9. The BLS defines the scope of the CPI to include only market goods and services or government-provided goods for which explicit user charges are assessed.

In the case of medical care, for example, the CPI includes direct out-of-pocket expenditures for medical care commodities and services, plus expenditures for the purchase of health insurance. This definition includes the employee-paid share of premiums for employer-provided health insurance coverage, as well as Medicare Part B monthly insurance premiums, but excludes the portion of income and payroll tax payments used to fund the provision of medical care for elderly and low-income beneficiaries.Although the advisory commission states that all medical care spending should be included in the CPI, the BLS believes that the exclusion of Medicaid and Medicare Part A is appropriate and consistent with the treatment of public schools and other tax-funded goods and services10.

Practical price index measures exist that do not hold the market basket of goods and services fixed at its original value. Some of these index measures, known as "superlative" indexes, have been shown theoretically to be closer to the cost-of-living concept than measures that track the cost of a fixed basket.11 The major superlative indexes are the Fisher and Törnqvist measures. Using these formulas, one can construct an index that accounts for the changes that consumers make in the quantities of the goods and services they consume in response to changes in relative prices. By substituting goods that have become relatively cheaper for those that have become relatively more expensive, consumers can achieve the same standard of well-being for less than the cost of purchasing their original market basket. The difference between an index that accurately accounts for this substitution and an index that does not (e.g. the Laspeyres index used in the CPI) is known as substitution bias. Because the CPI holds the market basket fixed at base period quantities, it incurs substitution bias by putting too much weight on the relatively more expensive items from which consumers have shifted away. The superlative indexes, because they adjust for changes in consumer expenditures, tend to avoid this type of bias. The superlative indexes do, however, require estimation of the comparison period market basket. Because it takes time to collect and process consumer expenditure data, a superlative index can be produced only with a time lag.12

III. Review of Advisory Commission Bias Estimates

Substitution Bias

The Commission report produces two estimates of substitution bias in the CPI: one for the lower level of aggregation and one for the upper level of aggregation. At the lower level of aggregation individual price quotes are aggregated to form subindexes for each category of goods, such as apples, watches, or dental services. At the upper level of aggregation these subindexes are collected into an all-items index. The formula currently used to aggregate the individual price quotations to form the subindexes does not account for consumers' ability to substitute across items within item categories when the relative prices of those items change-for example, when the price of Delicious apples increases and the price of Granny Smith apples falls. Similarly, the formula used to aggregate the subindexes to form the overall CPI does not reflect the substitution across item categories that takes place when the relative prices of items in different categories change-for example, when the price of apples falls relative to the price of oranges.

For substitution bias at the upper level the commission's estimate of 0.15 percentage point per year is based on BLS research that compares indexes calculated using superlative formulas to an index calculated using the fixed-weight Laspeyres formula.13 The BLS and the advisory commission essentially agree on the size and nature of the bias at this level.

Substitution bias at the lower level is sometimes confused with the separate problem of formula bias14. To understand what is meant by "formula bias," recall that the CPI measures the change in the cost of purchasing goods and services using a formula that weights each item's price by the quantity that was purchased during a base period. Because the household expenditure surveys give information on dollar expenditures rather than quantities, the CPI quantity weights must be derived indirectly, as expenditures divided by price. Until 1995 quantity weights for the items in the sample were formed by, first, projecting the initial price collected for each item backwards using information on price trends for similar items, and, then, dividing the appropriate expenditure figure by this backwards-projected price. This procedure, however, had an unintended consequence. Items that were on sale as of the point in time when they were first priced were systematically overweighted-expenditure divided by a low price gives a high quantity weight. Because the prices of sale items are apt to rise in subsequent months, this procedure imparted an upward bias, i.e., formula bias, to the index. The BLS introduced procedures (principally what is known as "seasoning") to eliminate this formula bias beginning in January 1995 for food-at-home and shelter, and June and July 1996 for all other items.

To calculate the lower level substitution bias the commission first asserts that the geometric means index is an unbiased estimate of the true cost-of-living index. They cite BLS research showing from June 1992-December 1994 the difference between the growth rate of the geometric means index and the CPI was 0.49 percentage point per year.15 Then the commission makes an adjustment to take account for the changes made by the BLS during 1995 and 1996 to eliminate formula bias, which the BLS has estimated to have reduced the rate of growth of the CPI by 0.24 percentage point per year.16 Their estimate of lower level substitution bias therefore is computed as the difference between 0.49 and 0.24, or 0.25 percentage point per year.

This estimate, however, may be too large. As described in Appendix A, the commission fails to mention several strong assumptions about the distribution of price changes that they implicitly use when claiming that the geometric means index is unbiased, or to note that, under these same assumptions, the Laspeyres formula currently used by the BLS also is unbiased. There is, moreover, reason to believe that the assumptions in question may not hold for many or most of the CPI component strata. If they do not hold, the geometric means index still may be unbiased, but only if the elasticity of substitution is exactly equal to one.17 If, on the other hand, this elasticity is zero, the "seasoned" Laspeyres used by the BLS will correctly show price change with no substitution.

As will be described in section IV, the BLS has made a commitment to evaluate the likely applicability of the geometric mean aggregation formula this year, item category by item category, and to make a decision by the end of the year about whether to adopt the geometric mean approach to calculating some components of the CPI. It is unlikely that the conditions necessary for the geometric mean formula to be unbiased will be found to hold in all cases. Thus, the commission's estimate of lower level substitution bias may be too large.

New Outlet Bias

The commission estimates that the entry of lower-priced outlets causes a bias of 0.1 percent per year. This estimate appears to be based on research conducted at the BLS by Reinsdorf, which compared price levels in newly selected outlet samples with price levels in outlet samples leaving the CPI.18 His estimates imply a price decline of about 0.25 percent a year, which gives a figure of 0.1 percent per year on an assumption that 40 percent of the CPI is affected by new outlet bias.

This estimate is subject to considerable uncertainty for three reasons. First, the effect of outlet entry is likely to vary from year to year, and Reinsdorf's data cover only two years from the late 1980's. Those years may be unrepresentative of long run trends. Second, Reinsdorf's estimates have large enough standard errors so that conservative statistical hypothesis tests would not rule out the hypothesis that the true effect of outlet changes is zero. Third, there is no assurance that the item categories studied by Reinsdorf, food and gasoline, are representative of other categories that may be subject to outlet bias.

Two additional considerations suggest that the estimate of 0.25 percent per year for the items affected by new outlet bias is too high. First, this bias estimate is based on an assumption that the new lower-priced outlets provide service of the same quality as the higher-priced incumbents. In many discount and off-price stores reductions in costly retailer services help make the low prices possible. (Examples of retailer services that might be less available at the lower-priced outlets include knowledgeable sales staff, breadth and depth of product assortment, assurance of item availability and quality, convenient location and hours, liberal return policy and store ambiance.) Furthermore, under some circumstances, entry by low-priced outlets with reduced services also could cause incumbents to reduce their services, thereby creating a downward bias unless adjustments for these quality reductions were made in the CPI. Indeed, because of the likelihood of quality declines, Reinsdorf interprets the 0.25 percent figure as an upper bound estimate of outlet bias in those components of the index where such bias might plausibly exist.

Second, changes other than entry of lower-priced outlets probably contribute to the price declines in Reinsdorf's data. Since 1978, the BLS has updated its sample of brands and product versions at the same time that it updates its outlet samples. Thus, if consumers were shifting over time to cheaper brands or product versions, these choices would be reflected, through the probability sampling methods used by the BLS, in selections of cheaper brands or product versions in the newly sampled outlets, making their price advantage appear larger than it really is. In a more recent study, Reinsdorf compares growth rates of sample average prices for food items and CPI food indexes over periods from 1948 to 1963 and from 1967 to 1976, when the BLS rarely changed the product version in the sample.19 These comparisons imply a price decline from new outlets of just 0.1 percent per year, compared to the 0.25 percent estimate above. This figure reflects price differences between outlets entering and leaving the sample because, for most of those years, the BLS had a policy of allowing price differences between outlets to affect its average price series but not its indexes.

Quality and New Products Bias

The largest share of the bias in the CPI that the commission concludes exists-0.6 percentage point per year, or more than half of the total of 1.1 percentage points per year-arises from an alleged failure to make adequate adjustment for changes in the quality of the goods and services people buy and to account properly for the value to consumers of newly available goods.

Before commenting on the evidence marshaled by the commission in support of its conclusions in the quality/new goods area, we emphasize that the BLS already has procedures in place designed to account for changes in the quality of the items being priced. (It often mistakenly has been assumed, though not by the commission, that the BLS makes few or no such adjustments.) Although these adjustment procedures are not perfect, they do have a very important effect on the rate of price change the BLS reports. The best available information on this point applies to a CPI subindex covering roughly the commodities and services component of the market basket (about 70 percent of the total, with shelter the largest exclusion). During 1995, this subindex would have risen by 3.9 percentage points had these procedures not been applied. Because of their application, however, the subindex actually rose by only 2.2 percentage points over the year. Roughly speaking, these figures imply that the adjustments made by the BLS for changes in the quality of these goods and services amounted to 1.7 percentage points over the course of a single year.20

The BLS also has established procedures for bringing new items into the index. The BLS has updated the expenditure share information used to aggregate the CPI subindexes only once every ten years or so, but the specific stores in which prices are collected and the specific items priced are reselected on a five-year cycle. Although more frequent sample rotations undoubtedly would be desirable, it is a fact that the BLS, by replacing 20 percent of the sample each year through the POPS and the initiation of new samples of outlets and items, already devotes considerable resources to ensuring that the sample of items priced is representative of what consumers actually are purchasing.

The commission does not argue, of course, that the BLS is not making a good effort to address quality/new goods biases, but rather that, in spite of a good effort, residual bias remains. The report's approach to assessing this residual bias is to divide the CPI into 27 categories, and then to make a judgment about the magnitude of the bias in each case. Unfortunately, the evidence applicable to many of these categories is rather sparse.

Of the 27 categories, the commission assigns eight a quality/new goods bias of zero (fuels, housekeeping supplies, housekeeping services, other private transportation, public transportation, health insurance, entertainment services, and tobacco). Each of the remaining 19 categories is assigned an estimated bias, in all cases positive (i.e., they concluded that price change is overstated because quality change is understated or the value of new products ignored). The commission supported its estimates of bias using three types of evidence: first, analysis of published and unpublished studies of quality/new goods bias for particular goods, second, quantitative evidence assembled by the commission from independent sources of data, and third, in the absence of direct evidence, estimates based on the judgment of the members.

For nine of the 19 categories (food at home other than produce, fresh fruits and vegetables, food away from home, alcoholic beverages, other utilities including telephone, other house furnishings, motor fuel, nonprescription drugs and medical supplies, and personal and educational expenses), absent evidence, the commission is forced to fall back on its best judgment. The alleged bias in these categories accounts for 0.11 of the 0.61 percentage point bias the commission attributes to quality/new goods problems. The food and beverages categories are an example; the commission's estimates of upward biases in these categories rest exclusively on judgments regarding the value to consumers of increased variety on grocery and liquor store shelves, together with the value of greater choice in restaurants, as shown in the following quotation from the report:21

"…there is little if any published evidence on the food category, other than [Jerry] Hausman's … attempt to establish the value for the introduction of a new variety of breakfast cereal…How much would a consumer pay to have the privilege of choosing from the variety of items available in today's supermarket instead of being constrained to the much more limited variety available 30 years ago? A conservative estimate of the value of extra variety and convenience might be 10 percent for food consumed at home other than produce, 20 percent for produce where the increased variety in winter (as well as summer farmers' markets) has been so notable, and 5 percent for alcoholic beverages where imported beer, microbreweries, and a greatly improved distribution of imported wines from all over the world have improved the standard of living."

In putting forward these estimates, the commission does not cite any published or unpublished studies, and indeed they comment on the absence of such evidence. Moreover, the commission does not specify how their estimates were developed in the absence of evidence. In several places the report characterizes the commission's specific estimates of bias as "conservative," but it generally is not clear why this is believed to be so. The commission's standard, the cost-of-living index, is defined as a function of consumer preferences, so reasonable questions to ask are, "Whose preferences are being described?" and "How were they assessed?" Although economists have methods for drawing inferences about preferences from market data on observed consumer choices, the report does not indicate that the commission used such methods in these cases. Appendix B presents an analysis of two categories, fresh fruits and vegetables and motor fuel, which attempts to quantify the missed consumer benefit or "surplus" that was described by the commission.22 In both cases this analysis concludes that the commission's estimates overstate the bias.

For four categories (shelter, apparel and upkeep, new vehicles, and used cars) members of the commission have produced evidence that bears on the trend in prices for particular sorts of items. The alleged bias in these categories accounts for 0.16 of the 0.61 percentage point bias the commission attributes to quality/new goods problems. In each of these cases there are significant problems with the inferences drawn by the commission.

An example of these problems is found in the commission's estimate of the quality bias in the index for rent of shelter. The commission's reasoning is essentially as follows. Over the period 1976 to 1993 the median rent increased about 1 percent per year faster than the CPI rent index. This fact might suggest that the quality changes already accounted for in the index are substantial. According to the advisory commission, however, these quality adjustments remain inadequate because of a supposed 20 percent increase in the average size of apartments between 1976 and 1993.23 In addition, they estimate that other improvements including "appliances, central air conditioning, and improved bathroom plumbing, and other amenities" amount to 10 percent over the past 40 years, giving a net upward bias of 0.25 percent per year.

There are two fundamental problems with this analysis. First, rents generally increase less than proportionally to apartment size, which implies that the advisory commission's proportional adjustment for apartment size would overstate the value of the increase. Second, the commission's factual premise-the assertion that average apartment size has increased 20 percent from 1976 to 1993-appears to be wrong. Although data giving an exact measure of the growth in size of rental units since 1976 are not available, a recent study analyzing data from the Residential Energy Consumption Survey, the American Housing Survey, and Current Construction Reports concluded that the increase was probably about 6 percent-i.e., the commission's estimate is too high by roughly a factor of three.24 After correcting this error, the data cited by the commission no longer support an upward bias of the CPI rent index.

Another example is the commission's estimate that the growth in prices of new and used cars has been overstated by 0.6 percentage point per year in the recent past. This estimate is based on a flow of services approach in which the cost of consuming automobile services declines as the useful life of the car increases. The commission presents data showing that the average age of cars on the road has risen, which it takes as a measure of the increase in the useful life of a car. To justify treating the increase in average age of cars as reflective of bias, the commission also assumes that current CPI procedures do not capture any of the increases in automobile durability that may have occurred. This latter assumption, however, is incorrect; Appendix C lists some of the many durability-related model changes for which adjustments have been made in the CPI over the past few years. Like other automobile quality adjustments in the CPI, these are derived from manufacturer cost data, marked up to retail values. The commission provides no evidence that this adjustment procedure would lead to an underestimate of the value of quality improvements that have contributed to enhanced durability.

Finally, the commission's estimate that the CPI has overstated the rate of growth of apparel prices by 1.0 percentage point per year since 1985 rests on a comparison of the official CPI data with price indexes constructed using Sears catalogue prices for items whose characteristics remain unchanged from one year to the next. Clearly one ought to have reservations about drawing any general conclusions based upon the prices charged by a single catalogue merchant. Moreover, BLS research has shown that price changes often are timed to coincide with changes in product characteristics, particularly in the apparel market segment where changing fashion is so important.25 To the extent that this is true, the commission's reliance on the data for unchanging items is likely to result in a downward bias, vitiating its criticism of the CPI apparel index.

For the six remaining categories (appliances including electronic, prescription drugs, professional medical services, hospital and related services, entertainment commodities, and personal care) the advisory commission reviewed existing studies of bias in the price trends for specific items to draw inferences about likely bias in the price trends for unstudied related items within the category. These six categories can be categorized as constituting two major areas of the index: medical care and high-tech consumer goods. More than half (0.34 percentage point) of the quality/new goods bias the commission believes exists in the overall CPI is judged to occur in just these areas of the index. These clearly are components of the index in which the BLS faces particularly difficult measurement problems, though the inferences that the commission has drawn about the magnitude of any bias in these index components involves some degree of speculation and extrapolation.

The advisory commission's estimate of bias in the medical care component of the index appears to have been largely based on just two recent empirical studies, one of cataracts, the other of heart attacks, which both identified large quality improvements that are missed in the calculation of the CPI.26 Although we acknowledge that there have been enormous improvements in medical technology over time, we also note the heterogeneity of the medical services category, which includes services as diverse as dentistry, eyeglasses and eye care, psychological counseling, podiatry, chiropractic, and physical therapy. Thus we are not convinced that the two conditions cited by the commission should be considered representative with respect to the unmeasured quality advances in the treatment of all medical conditions.

In some cases quality bias in the medical care component of the index may have arisen as a result of failure by the BLS to capture improvements in procedures that led to shorter hospital stays and out-patient treatment. The BLS recently has taken steps that, at least in principle, should address medical care quality improvements of this type. For hospital services, beginning in January 1997 the CPI has adopted the practice that previously had been used in the Producer Price Index (PPI) of pricing completed treatments (as represented by the service bundles on selected patient bills) rather than individual medical inputs. This change should permit BLS staff to track changes in treatment over time.27 This change, however, will not resolve all quality adjustment problems in the medical care component. Some kinds of quality change are difficult to evaluate, involving changes in patient outcomes, such as improved mortality or reduction in pain. The BLS is continuing to support and encourage research on this topic, but we are skeptical that it will be possible to develop methods that will permit reliable evaluation of all kinds of quality changes on an on-going basis within the monthly CPI.

The area of high-tech consumer goods (e.g., consumer electronics) is one for which there are a number of published studies documenting systematic quality bias of the CPI. Most of these studies are based on the method of "hedonic" quality adjustment (i.e., adjustments based upon the empirical relationship between the prices of various items and their characteristics), with studies having been conducted of personal computers, television, video equipment, etc. The BLS is currently applying hedonic methods in the PPI for personal computers and peripherals. Projects are underway at the BLS to develop hedonic quality adjustment methods and improved sampling of new products within the appliance category of the CPI.

In addition to these specific comments about the nature of the evidence on quality/new goods biases assembled by the commission, there are several general remarks to be made. The commission's estimates of bias are made case by case using a variety of methods, without any clear statement of what methods are appropriately used in each circumstance. The absence of a well defined methodology for deriving the commission's estimates represents a fundamental reason why the BLS reaction to the quality/new products section of the report has been skeptical. Also, in general, the commission's discussion of quality/new goods biases does not include explicit recommendations regarding the adoption of procedures to correct the problems it believes exist. In part, this appears to reflect a lack of consensus among economists about what is practical and theoretically justified for measuring the benefit to consumers from new products.28 For production of the CPI and other national statistics the BLS must use methods that are objective, reproducible, and verifiable.

The commission also failed to make any systematic effort to explore the possible existence of negative biases in the CPI. Other analysts have hypothesized reduced convenience and comfort of air travel and deteriorating quality of higher education as examples of quality decreases that are ignored in the CPI. More generally, whereas the commission notes some service quality improvements, such as the introduction of automatic credit-card readers at gasoline pumps, the BLS often hears complaints about broad-ranging declines in the quality of customer service, which are equally difficult to incorporate in the CPI.

A more subtle issue is that price increases for many goods occur intermittently and often are timed to coincide with model replacements or other quality improvements. The BLS commonly adjusts for quality differences between successive models by, in effect, treating the difference in price between them as wholly attributable to a difference in quality. There is a risk that this procedure may over-adjust for quality change, imparting a downward bias to the index. Methods have been introduced to try to minimize that possibility, but the commission paid little attention to this potential problem.

From a BLS perspective, the most important question about possible quality/new goods problems is what we might do to improve our procedures and ameliorate those problems. Recognizing the particular difficulties associated with measuring medical care prices and high-tech consumer goods prices, the BLS has devised and announced important improvements in our methods. These include the changes noted above in our hospital price measurement procedures, and prospective changes in our sample rotation procedures that will allow us to update item samples in rapidly changing market segments more frequently than once every five years (at the cost of less frequent updates in more static market segments). In addition, the President's 1998 budget includes funds to improve the accuracy, timeliness, and relevance of the consumer price data available from the BLS. The FY 1998 budget request, if approved, would allow us to make important progress in the quality/new goods area, by supporting greater use of hedonic techniques and implementation of more aggressive procedures for identifying and beginning to price new goods promptly once they appear in the marketplace.

IV. Short Run Recommendations29

Recommendation i. The BLS should establish a cost of living index (COLI) as its objective in measuring consumer prices.

The advisory commission's report begins with one overarching recommendation: "The BLS should establish a cost of living index (COLI) as its objective in measuring consumer prices." The BLS basically concurs with this; indeed, the BLS long has said that it operates within a cost-of-living framework in producing the CPI. That framework has guided, and will continue to guide, operational decisions about the construction of the index.30 Putting things slightly differently, if the BLS staff or other technical experts knew how to produce a true cost-of-living index on a monthly production schedule, that would be what we would produce. While the BLS has no fundamental disagreement with the commission about what the objective of our CPI program ought to be, we disagree to some extent about what changes to the index would be feasible and prudent and about the timetable on which those changes could be implemented.

Because the cost-of-living concept does not imply a single all-purpose cost-of-living index, the BLS will continue to need to make choices about the specific issues of formula, coverage, and index construction. The BLS will continue to describe the scope and theoretical assumptions of its price measures, as well as any necessary caveats with respect to their use.

Recommendation ii. The BLS should develop and publish two indexes: one published monthly and one published and updated annually and revised historically.

Recommendation iii. The timely, monthly index should continue to be called the CPI and should move toward a COLI concept by adopting a "superlative" index formula to account for changing market baskets, abandoning the pretense of sustaining the fixed-weight Laspeyres formula.

Recommendation iv. The new annual COL index would use a compatible "superlative-index" formula and reflect subsequent data, updated weights, and the introduction of new goods (with their history extended backward).

Because these three recommendations address methods for dealing with the upper-level substitution bias problem, we will discuss them together. The commission recommends that the BLS should move to a "trailing Tornquist" formula for the monthly index.31 The Final Report did not explicitly define this formula, but based on subsequent discussions with commission members, we interpret this to mean a geometric mean formula in which the weights are lagged expenditure shares, the weights are regularly updated, and the indexes are chained. The commission also recommends that the BLS develop a new annual index that is calculated using a superlative formula and is subject to revision.

The BLS continues to investigate several experimental indexes that use a superlative formula at the upper level of aggregation. These include formulas which, due to the need for current expenditure data, create indexes that must be produced with a lag, as well as new methods that may approximate the superlative formula and allow the production of indexes in a timely fashion.

While the method of calculating the current CPI could be changed to incorporate a superlative formula, the CPI would then have to be produced with a lag. Moreover, the expenditure data that are required to derive the weights for the superlative index are available with sufficient precision to be used in calculating such an index only at annual intervals, and thus would not support a true monthly CPI.

The timeliness of the CPI might be maintained by using some form of an approximation to a superlative index. The commission's proposed "trailing Tornquist" formula, however, has been shown to produce price changes that systematically understate the increases in the cost of living, as measured by the superlative formulas.32 More recently, other approximation strategies have been proposed, including a method based on the "constant elasticity of substitution" (CES) formula.33 But such an approximation would not track the superlative indexes precisely-during some years an index based on an approximation would rise more than the superlative index, during other years it would rise less. This feature raises the issue of whether such an index subsequently would need to be revised once the data were available to calculate the superlative index. Another issue that needs to be addressed in considering use of approximations is the issue of estimating the subaggregate indexes, i.e., the indexes for intermediate levels of aggregation, such as for "food" or "transportation." Some of these indexes may consist of item categories that are relatively close substitutes-fresh fruits, for example, consists of apples, bananas, oranges, etc.-whereas others may consist of item categories that probably are not close substitutes-medical professional services, for example, includes physicians, dentists, and eyecare. Because the CES function is based on a single elasticity parameter which is assumed to be the same for all items, while consumers' willingness to substitute is likely to vary across categories of items, further research is needed to determine whether a simple approximation such as the CES would produce sensible approximations for all of these subaggregates. Also, the use of an index based on statistical approximation might be difficult to interpret and explain to users of the data. We believe we would gain little, and possibly do much damage to the credibility of our statistical system, if we were to move hastily to adopt untested techniques for producing the official CPI.

The President's 1998 budget includes funds to improve the accuracy and timeliness of the CPI, and an important part of this request will support the production of a superlative index, produced to a greater degree of accuracy than is now possible. The BLS plans to begin publishing this measure in early 2002. In the interim, the superlative measures we currently produce can be used to estimate the magnitude of the upper level substitution bias in the CPI, and indeed are the best measures currently available for this purpose.

Recommendation v. The BLS should change its procedure for combining price quotations by moving to geometric means at the elementary aggregates level.

To address lower-level substitution bias, the commission has suggested adoption of a geometric mean formula for aggregating price quotations, a formula that has been under investigation by the BLS over the past several years. As discussed above, the current CPI formula does not allow for the potential substitution among items within a category, such as between different varieties of apples, when the relative prices of those items change. The proposed geometric mean formula is based on an alternative assumption, namely that consumers substitute among items in such a way as to hold the share of their expenditures devoted to each item constant. Although this assumption is not likely to hold exactly for any particular stratum, the geometric mean formula should provide a close approximation to the exact cost-of-living subindex in cases where the stratum consists of substitutes, such as different varieties of apples, and the price elasticity of demand for each variety is fairly large. If the elasticity of substitution is zero, then the fixed weight Laspeyres formula is the appropriate measure of the cost-of-living subindex. Again this assumption is not likely to hold exactly, but the Laspeyres index should provide a close approximation to the exact cost-of-living subindex in cases where the price elasticity of demand for each variety is quite small. It may be more plausible to assume that consumers substitute freely between, for example, types of apples or between brands of television sets when their relative prices change than to assume similar substitutability between, for example, types of prescription drugs.

The BLS has begun issuing a monthly experimental measure that is constructed using the geometric mean formula in all index components, and will make a decision by the end of this year as to which components of the official CPI should employ the geometric mean formula.34 Scanner data, studies of substitutions between brands, and other information will be used to assess the propensity of consumers to substitute across items within individual item categories as the relative prices of those items change. The likely date for implementation of any changes decided upon for the official CPI is with the release of January 1999 CPI data.

Our best estimate is that the use of the geometric mean formula in all CPI subindexes would lower the growth rate of the index by approximately one-quarter of one percent per year. Partial adoption of the geometric mean formula, which is more likely than a full adoption, would be expected to have a downward impact of between zero and one-quarter of one percent per year, depending on how many, and which, indexes use the new formula.

V. Intermediate Run Recommendations

Recommendation vi. The BLS should study the behavior of the individual components of the index to ascertain which components provide most information on the future longer-term movements in the index and which items have fluctuations which are largely unrelated to the total and emphasize the former in its data collection activities.

Sample resources for the CPI are allocated between the two major price surveys, commodities and services (C&S) and housing, according to the relative importance and variability of the survey estimators for each component, while taking into account the relative costs of each survey. The sample for the C&S component of the CPI was designed to allocate resources systematically among major item groups and sample cities, utilizing models to minimize the sampling variance of estimated price change, as measured by the all-items (less shelter), national CPI, subject to cost and sample coverage constraints. Solution allocations among items, outlets, and cities thus strike a balance with respect to the contributions of components of sampling variability by sample items, their relative importance with respect to the total consumer budget, and the relative cost of data collection and processing, while keeping within the cost and coverage constraints of the program.35

The commission's recommendation suggests that data collection activities should focus on a different objective, namely to provide information on the future longer term movements of individual prices or the index as a whole. Forecasting inflation is a widespread and important use of the CPI, of course, but one that is conceptually distinct from the measurement of cost-of-living changes. If prediction of future inflation, or the measurement of "inflationary pressure," were the measurement objective of the CPI, this might imply different choices with respect to the formulas and weights used in construction of the index, as well as with respect to the allocation of the sample. The commission, however, emphasizes the use of the CPI as a measure of past and contemporaneous changes in the cost of living in choosing the index formulas and weights, on the one hand, while emphasizing the uses of the CPI in forecasting future price movements in determining the sample allocation, on the other. This appears to be an internally inconsistent strategy.

The commission suggests that resources devoted to the sample for bananas, a perishable fresh fruit whose price-change sampling variability has been estimated to be substantial, but whose price fluctuations are "not systematically related to the underlying trend movements of the CPI," would be better allocated to surgical treatments, consumer electronics, and communication services.36 The potential for saving resources by reducing data collection of items like bananas is fairly limited because the marginal cost of collection and processing is quite small-the stores are already being visited to collect other grocery items and very little analysis is required after collection. Because the sample has been allocated to minimize the variance, a reallocation of resources away from any item with a high sampling variance toward other items necessarily would result in an increase in the variability of its index and the reliability of the all-items index would be diminished.

Recommendation vii. The BLS should change the CPI sampling procedures to de-emphasize geography, starting first with sampling the universe of commodities to be priced and then deciding, commodity by commodity, what is the most efficient way to collect a representative sample of prices from which outlets, and only later turn to geographically clustered samples for the economy of data collection.

Because geographical coverage impinges on many aspects of the CPI data collection and index estimation process, the practical meaning of this recommendation is somewhat unclear. By the same token, the importance of the geographic structure underlying the CPI makes it a continuing subject of BLS research.

The statement that the BLS should decide commodity by commodity, what is the most efficient way to collect a sample, has been and will continue to be the standard practice. In several cases, for example, postage and used cars, the BLS currently collects data on a national level. In most cases, however, it is not possible to select samples of specific items at the national level because of the lack of a national list (or frame) of items to sample, together with the sales volume information needed to determine the probabilities of selection. Moreover, if specific items were selected nationally, there would not usually be a feasible way to determine whether a selected item was, in fact, carried by any particular sample retail outlet. These considerations have led the Bureau to do sampling locally, by first selecting the urban area, then the outlet, and finally the specific item within the outlet. This method helps to ensure that the sample of items is timely and representative. The BLS is currently investigating potential uses of point-of-sale (scanner) data which are available from private vendors, and in the future it might be possible in some cases for the BLS to use such data to draw national samples of items.37

Recommendation viii. The BLS should investigate the impact of classification, that is item group definition and structure, on the price indexes to improve the ability of the index to fully capture item substitution.

As part of the 1998 CPI revision activities, the BLS has just completed a process of modifying the item classification structure.38 The ability of the index to capture consumer substitution was one of the prominent factors that was considered in developing the new item classification. In putting together the item classification, the BLS "also tried to see that [the strata] formed natural groups, as consumers would view them...For example, using the consumer view, items within the same stratum should have some affinity, such as substitutes (butter and margarine), or complements (washers and dryers)."39

The commission points to some examples which cross item boundaries, such as "on-line news services which compete with newspapers, automobile purchases with leases, and drugs with surgical procedures they replace" as examples for which direct price comparisons are needed so that the full substitution effect can be measured.40 The BLS is sympathetic to the commission's concern, and will continue to work to improve the CPI item structure.41 It seems to us, however, that no feasible item classification system would completely capture the current and possible future developments in consumer substitution behavior. Nor does it seem to us that the item classification system is necessarily the most significant impediment to measuring the effects of these substitutions. The more fundamental issue is the need to develop systematic methods for identifying the substitution and accounting for differences in quality between the substituted items.

Recommendation ix. There are a number of additional conceptual issues that require attention. The price of durables, such as cars, should be converted to a price of annual services, along the same lines as the current treatment of the price of owner-occupied housing. Also, the treatment of "insurance" should move to an ex-ante consumer price measure rather than the currently used ex-post insurance profits based measure.

When the BLS adopted the rental equivalence approach to pricing housing services in 1983, BLS staff were aware that the same conceptual issues arise in the pricing of other consumer durables.42 In principle the CPI is intended to measure the cost of consuming goods and services, and durable goods provide a flow of services over time rather than immediate consumption. To implement a flow-of-services approach, however, requires information on either rental equivalence or user cost of the durable asset. In the case of housing, the existence of rental markets makes it relatively easy to implement the rental equivalence approach, while the long life of housing assets and the likelihood of price appreciation made the standard asset price approach uniquely problematic. During the mid-1980s, BLS researchers investigated the potential use of automobile leasing data to price automotive services, but at that time concluded that the leasing markets were not sufficiently developed to support a leasing equivalence approach to index construction. Subsequently, automobile leasing has grown to the point that in 1998 an automobile leasing stratum will be added to the CPI market basket. Currently BLS researchers are reexamining the flow-of-services approach for automobiles, possibly using a leasing equivalence methodology. For durables other than automobiles, the lack of widespread rental markets as well as the lack of data needed for direct estimation of user cost suggest that the flow of services approach may not be practicable. As explained in our discussion of the commission's quality bias estimate for automobiles, we do not agree with the commission's premise that failure to price a flow of services necessarily leads to systematic quality bias.

The commission recommends that the BLS move the CPI for insurance to an "ex ante consumer price measure" from the currently used "ex post insurance profit based measure." The current CPI for health insurance does not directly price policies purchased by consumers.43 Instead, an indirect approach to measuring the price of a policy is used; the price is seen as deriving from the services provided by the insurer and the value of benefits paid to providers of health care. The BLS prices these two parts separately, obtaining from insurers information on retained earnings to measure changes in the value of the insurance service component, and using the price indexes in the CPI medical care component to measure changes in the cost of the health benefits paid to providers. It is possible that direct pricing of health insurance policies would have the virtue of automatically reflecting cost-reducing innovations in the treatment of medical problems (such as the substitution of less-costly outpatient procedures). The countervailing difficulty, however, is that health insurance policies can increase or decrease in price due to changes in coverage or in the characteristics of the covered populations, and these changes may be very difficult to observe or adjust for in the index.

The current CPI approach was adopted in 1964. Prior to that the CPI collected the price of the most widely-sold community-rated Blue Cross/Blue Shield policy. That approach was dropped, however, when it became evident that the quality of the policies was changing in ways for which it was difficult to adjust the policy price. In 1984-85 the Bureau experimented with the direct pricing of a sample of health insurance policies but the experiment was terminated because it again proved too difficult to maintain constant quality and coverage of risk over time. The BLS recognizes the importance of the health insurance price movements to consumers as well as to policy makers and will continue to search for ways to overcome the obstacles to accurate adjustment for changes in policy characteristics.44

Recommendation x. The BLS needs a permanent mechanism for bringing outside information, expertise, and research results to it. At the request of the BLS, this group should be organized by an independent public professional entity and would provide BLS an improved channel to access professional and business opinion on statistical, economic, and current market issues.

The BLS already has in place many mechanisms for bringing in outside information, expertise, and research results. Business and labor research advisory committees meet regularly with BLS staff and management and have long been a source of outside information and expertise. A price research division has been a part of the price index programs since 1965, and much of the discussion of CPI bias has been based upon the results of research conducted by BLS staff. BLS economists and statisticians regularly solicit opinions from outside researchers by presenting research papers at conferences and submitting them for publication at peer reviewed journals. Academic researchers are regularly invited to present their research findings to BLS staff in seminars. The Bureau's ASA-NSF-BLS fellowship program brings in scholars for extended on-site research projects. The BLS has funded research by academic economists when research by experts was needed to solve difficult measurement problems.45

The BLS agrees that continued input from outside researchers is useful, and is currently studying the possibility of creating an academic advisory commission. In addition, the BLS is interested in having outside researchers address the important measurement issues that it faces, and will provide researchers with access to research databases to the extent possible, while meeting data confidentiality requirements.

VI. Longer Run Recommendations

Recommendation xi. The BLS should develop a research program to look beyond its current "market basket" framework for the CPI.

This recommendation suggests that the BLS should develop research programs exploring "quality of life" issues such as time-saving benefits of new medical procedures and new communication devices, and changes in the social or natural environment caused by rising crime or new diseases. Because these things clearly affect our standard of living, a complete accounting of U.S. economic progress would include them.

We do, however, have a reservation about this recommendation. Implicit in this recommendation is a suggestion that the BLS should adjust the CPI for these effects. We think that valuing changes in time allocation or in the general social environment may require too many subjective judgments to furnish an acceptable basis for adjusting the CPI. Furthermore, arriving at a comprehensive measure of changes in the quality of life will be quite difficult, yet making such adjustments in only a few selected cases could make the CPI less accurate if these cases are not representative. Finally, it is unclear whether "quality of life" valuations really belong in an index used for the escalation of payments and adjustment of tax parameters. For example, the advisory commission suggests that the CPI rent index should have made a quality adjustment for changes in climate as renters migrated to the south.46 Such a quality-of-life adjustment, however, is properly viewed as out of scope under the current definition of the CPI.47 Most of the uses of the CPI have evolved within the context of an index limited to market goods and services, and presumably the appropriate uses of an index that incorporated changes in crime levels, disease incidence, or income tax rates would be somewhat different from the current uses of the CPI.

Recommendation xii. BLS should investigate the ramifications of the embedded assumption of price equilibrium and the implications of it sometimes not holding.

Any systematic method for distinguishing quality change from price change must be based on some theoretical framework and set of assumptions. In most cases the BLS, like academic economists who do research in this field, relies on one or another assumption about price equilibrium. An equilibrium assumption underlies hedonic methods for quality adjustment, for example, as well as the matched model price comparisons commonly used by the BLS.48 Although virtually all systematic methods for quality adjustment are based to some extent on assumptions about price equilibrium, the nature of the assumptions differs between methods. Of the methods used for quality adjustment by BLS, two (the "overlap method" and the "link method") are based on a particularly strict equilibrium assumption-that quality differences can be inferred from the price differences between individual items.49 The hedonic method, in contrast, allows for random deviations of prices from equilibrium values and may allow for differences in rates of price change between items of different vintages.

The commission recommends that the BLS investigate the assumption of price equilibrium that underlies certain quality adjustment and item substitution procedures. We agree that reducing reliance upon this assumption can sometimes make the CPI more accurate, particularly for long run comparisons. Indeed, the BLS already has made considerable progress in doing this. Recent tabulations indicate that item replacements adjusted for quality using the methods that embody a strong price equilibrium assumption (i.e., the "overlap method" and the "link method") declined from about 2 percent of prices collected in 1983 to 0.62 percent in 1995.50 In addition, the CPI for prescription drugs now reflects consumers' savings from buying therapeutically equivalent generic substitutes for branded products. We plan to continue research on avoiding bias from unwarranted price equilibrium assumptions.

Recommendation xiii. The BLS will require a number of new data collection initiatives to make some progress along these lines. Most important, data on detailed time use from a large sample of consumers must be developed.

The final longer run recommendation is that the BLS should develop new data collection initiatives on time use and "quality of life" issues. These data would support the research programs described in the commission's first longer run recommendation. We agree that time use data would be valuable to researchers, and we concur with the focus on using them for supplementary indicators rather than as part of the main cost-of-living framework.

VII. Conclusion

The advisory commission report has performed a service by calling to the attention of policy makers the many and varied issues that the BLS faces in constructing the CPI. Most public attention has been focused on the commission's estimates of CPI bias, but the central argument of the report is that almost every assumption underlying the procedures used around the world for price index construction is called into question by the pace and form of market developments. The issues are not new to index number experts (many of the issues are discussed, for example, in the articles in the December 1993 Monthly Labor Review), but the quantitative and budgetary importance of price measurement problems and techniques have not always been appreciated by users.

As discussed earlier in this paper, the BLS has a vigorous program of research and development activities aimed at improving the CPI. In one category are the activities related to upper- and lower-level substitution bias. These include:

· The continued monthly publication of the experimental geometric mean index (the CPI­U­XG) and the evaluation of the geometric mean formula for use in the CPI­U and CPI­W, probably beginning in January 1999.

· The continuing annual publication of experimental superlative indexes, and (assuming approval of the Bureau's associated budget requests) introduction of an official superlative index as a supplement to the CPI­U and CPI­W in 2002. With the development of the CPI­U­XG, the experimental superlative indexes can be constructed and compared using individual category indexes based on both arithmetic and geometric mean formulas.

· Introduction of a new CPI market basket in January 1998 based on 1993-95 consumer expenditure patterns, and consideration of a more frequent schedule of market basket updates than the roughly ten-year cycle followed in the past. The BLS FY 1998 budget initiative also calls for development of an enhanced processing system that will enable us to construct expenditure weights that are just two years old when introduced into the index. (By contrast, the 1993-95 market basket will be 3 1/2 years old when it is introduced in January 1998.)

The advisory commission recommends using a geometric mean formula for upper-level aggregation, and annual market basket updates, to approximate a superlative index while avoiding the need for index lags or revisions. Evidence indicates that such an index would be downward-biased relative to a cost-of-living index. As recommended by Shapiro and Wilcox, however, one could develop an index based on the CES formula that provides a close approximation to a superlative index over some historical period. The BLS plans to estimate such an index as part of its experimental superlative index program. Additional research is needed on the approximation properties of the CES formula, especially below the U.S. all-items level, before it could be considered for use in the CPI­U or CPI­W. Moreover, a move away from the arithmetic-mean Laspeyres formula above the category level could make the CPI more difficult to use and explain, and these considerations would have to be weighed against the potential advantages of a closer approximation to a cost-of-living index. Also weighing in would be the potential disadvantages of using a formula based upon an approximation to a superlative index, which might need to be revised once the data were available to calculate the superlative index.

This paper has emphasized that substitution bias, and especially upper level substitution bias, accounts for a relatively small part of the total bias that the advisory commission argues exists in the CPI. Quality change in existing goods and services, the introduction of new products, the establishment of new outlets, and the disappearance of older products and outlets, present extremely important issues for which there are, as yet, no general solutions. The absence of general solutions explains why the commission has no short-run recommendations in these areas. The BLS will continue to study the pertinent intermediate-run recommendations-use of leasing equivalence for automobiles, direct pricing of health insurance, and investigation of improved item classification structures-but these are unlikely to solve the fundamental measurement problems even in specific CPI components. Finally, the absence of systematic, well-accepted ways to deal with these problems also means that there are no rigorous ways to measure the new outlet or quality/new goods biases potentially created in the CPI. The advisory commission, like other observers, was forced to use introspective or extrapolation methods to obtain many of their bias estimates.

The BLS specifically rejects several of the estimated quality or new goods biases, in cases where the commission presented new evidence. Examples of these cases noted in Section III above include the estimates of a 0.25 percentage point annual bias in shelter, a 1.0 percentage point annual bias in apparel and upkeep, and 0.59 percentage point annual biases in new and used cars. Together, these comprise 0.16 of the 0.6 total estimated quality/new goods bias in the overall CPI. In addition, the evidence presented in Appendix B suggests that the commission's estimates of bias for the food and motor fuel components likely are overstated. Most of the remaining estimated bias comes from two areas of the index: medical care and high-tech consumer goods. These clearly are components of the index that present particularly difficult measurement problems, but the quantitative evidence is very fragmentary and the BLS is reluctant to speculate as to what the magnitude of any bias in these index components might be. Finally, some analysts have cited potentially countervailing declines in quality, particularly in services, that are not reflected in the CPI or in the advisory commission's bias estimates.

For the BLS, the primary task is not to evaluate the bias estimates set forward by the advisory commission or other groups, but rather to employ the most accurate methods available for dealing with quality change and with new goods and outlets. Those methods must be rigorous, objective and reproducible, minimizing the role of analyst judgment, although these considerations make it very difficult to incorporate in the CPI the benefits of some types of product innovation.51 Improvements in medical care that enable patients to lead more active lives have undoubted value, for example, but that value cannot now be, and may never be, measured objectively enough to be reflected in official data series. Notwithstanding such limitations, the BLS is taking several steps to improve its methods for dealing with quality change and new products:

· Effective in January 1997, two improvements were made in the hospital and related services component of the CPI. The hospital room, other inpatient, and outpatient subcomponents were consolidated to enable the index to reflect shifts in the mix and importance of treatment. At the same time, there was a shift from pricing individual items (like units of blood) to pricing the collections of services on selected patient bills; among the benefits of this change are a better reflection of alternative reimbursement methods and an enhanced potential for quality adjustment.

· In 1999, the BLS will implement a change in the CPI's sample rotation procedures from a city-based to an item-based sequence. This ultimately will make it possible to update item samples in rapidly changing market segments more frequently than once every five years (at the cost of less frequent updates in more static market segments).

· The BLS FY 1998 budget request calls for data collection to support greater use of hedonic techniques that explicitly account for changes in the characteristics of items being purchased. (Even in the absence of such funding, the use of hedonic regression for quality adjustment likely will expand into product categories such as personal computers and televisions.) The requested resources also would support implementation of more aggressive product initiation procedures for identifying and beginning to price new goods promptly once they appear in the marketplace.

· Other potential intermediate-term changes include the direct pricing of health insurance policies and a leasing equivalence approach to pricing of automobile services, as recommended by the advisory commission. Both approaches have been evaluated by the BLS in the past and rejected as infeasible, but new developments in the leasing and medical care markets argue for their continued consideration.

Unfortunately, the ongoing controversies surrounding cost-of-living measurement and, more generally, appropriate federal indexation policy, have led much of the public to conclude that the CPI is somehow "broken." Although the BLS rejects that notion, it is evident that the expanding number of users of the CPI have objectives and priorities that sometimes can come into conflict. When this happens, the result can be an index that is less than optimal for certain purposes. One example mentioned above reflects the competing objectives of an index that is free of upper-level substitution bias (as might be desired, for example, for benefit indexation), and of one that is not subject to lags or routine revision (for example, for indexing debt instruments). Some also have argued the need for specialized indexes for program beneficiaries or other population subgroups. It is, in fact, commonplace to observe that there is no single best measure of inflation. The BLS response to this situation has been to develop a "family of indexes" approach, including experimental measures designed to answer different questions from those answered by the CPI­U and CPI­W. This "family of indexes" now includes the CPI­U­XG, the CPI­E corresponding to the market basket of elderly consumers, and the experimental superlative measures, and under the BLS FY 1998 budget request will include a production-quality superlative measure beginning in 2002. As mentioned above, an experimental CES index is a likely addition to the group.

The BLS is engaged in numerous CPI program enhancements that have not been mentioned above. Some are part of the six-year CPI revision program now underway: conversion to computer-assisted data collection and a telephone-based POPS survey, improvements to the housing sample and estimator, and enhancements to the CEX survey processing system. In addition, the FY 1998 budget request, if approved, would support an expansion in the CEX sample, permitting more accurate expenditure weights and a more timely CPI market basket. The solutions to many CPI measurement issues, however, must await methodological breakthroughs in economics, or improved availability of data. Unfortunately, the techniques available for measuring the gains in consumer welfare from new products (and the losses from product disappearances) are in their infancy, and may never be adaptable for implementation in a large, ongoing price measurement program like the CPI. The increased use of scanner data in U.S. consumer markets offers broader opportunities, and the BLS has been engaged in a significant research effort to explore the many possible uses of these data, in identification of new products and outlets, sampling of items, and ultimately in the computation of the CPI itself.

In summary, the concluding statements of the BLS report to the House Budget Committee in April 1995 remain applicable today. The BLS is intensely aware of the sensitive nature of the data it produces, and of the critical need for these data to be as accurate as possible. It will continue to investigate the measurement issues that it and others have identified, and will introduce corresponding improvements to the index as quickly as it can.

Appendix A. Technical Issues About Lower Level Substitution Bias

This appendix describes two technical problems with the commission's discussion of lower level substitution bias and formula bias. The first of these problems involves the commission's discussion of "time reversibility."52 The commission describes this property as a "requirement or test for an index number...that the index should remain the same if the underlying prices undergo a reversal." Their example of this property, however, is incorrect. In their example, the quantity of beef is 1.0, and the price of beef starts at 1.0 in period 1, rises to 1.6 in period 2, and then falls back to 1.0 in period 3. The commission claims that in such a case the CPI would add the 60 percent increase between periods 1 and 2 to the 37.5 percent decrease between periods 2 and 3 to show a total increase of 22.5 percent between periods 1 and 3. Adding the percentage changes, however, is contrary to any reasonable procedure and is not an accurate description of current or past BLS methods. The ratio of the price of beef in period 2 to its price in period 1 is 1.6, and the ratio of the price in period 3 to the price in period 2 is 0.625. So in this case the CPI would multiply the relative changes (1.6 0.625 = 1), correctly showing no change in price between periods 1 and 3. Thus it is inaccurate to attribute the bias shown in this example to the CPI.

A second problem is the commission's assertion that the geometric mean formula would eliminate lower level substitution bias. The commission states that the difference between a geometric means index and a Laspeyres index "is an estimate of the bias of the Laspeyres formula, since [Matthew] Shapiro and [David] Wilcox…have shown that the geometric mean provides an unbiased estimate of the underlying cost-of-living index."53 This statement is surprising, because it is well known that the geometric mean index is unbiased only under restrictive conditions. The basis for the commission stated view appears to be as follows:

"Shapiro and Wilcox...have provided an elegant rationale for the geometric approach based on the correlation of relative prices over time. Provided that this correlation is small, a modification of the geometric mean is approximately unbiased for the underlying cost of living index, and this characterization does not require information about the underlying system of consumer's preferences" (U.S. Senate, Committee on Finance, Final Report, p. 19).

This statement mischaracterizes the discussion in Shapiro and Wilcox. That article made several important assumptions that are not mentioned by the commission, including assumptions about consumers' preferences. These assumptions are stated by Shapiro and Wilcox when they describe the results of BLS research:54

"Several recent papers [by BLS authors]...have explored another alternative to the Laspeyres-based formula, namely the modified geometric means estimator...Under the same assumptions as we used above (CES utility, stationary distribution of relative prices, etc.), one can show that the modified geometric means estimator is approximately unbiased for the true cost-of-living index" (Shapiro and Wilcox, "Mismeasurement," p. 111).

The assumptions made by Shapiro and Wilcox are fairly restrictive. For example, the assumption of a stationary distribution of relative prices is an assumption that all of the prices in a stratum follow the same underlying trend. Prices in heterogeneous strata very likely violate this assumption because dissimilar goods may well follow different trends. Prices even in relatively homogeneous strata, such as tomatoes, can violate this assumption if some goods are produced with different technologies, such as hand-picked versus mechanically picked tomatoes.

Moreover, if all of these assumptions hold (stationarity, small correlation of relative prices over time, CES utility), the seasoning method now used to estimate the CPI component indexes also is unbiased. As Shapiro and Wilcox state, under their assumptions:

"If rl-n (the autocorrelation of the relative prices between periods l and n) is small, [the "seasoned" version of the CPI]...should provide quite an accurate estimate of the rate of increase in the true cost-of-living subindex, regardless of the elasticity of substitution" (Shapiro and Wilcox, "Mismeasurement," p. 110).

The fact that the growth rates of geometric means index and the seasoned index actually differ implies that an assumption, probably stationarity, is being violated. This weakens the commission's argument that the geometric means index necessarily approximates a true cost-of-living index and points to the importance of taking account of consumer substitution behavior. If relative prices are not stationary, then the geometric mean formula may still be the exact measure of the stratum cost-of-living subindex, but only if the elasticity of substitution equals one.55 Alternatively, the seasoned Laspeyres formula may still be the exact measure of the stratum cost-of-living subindex, but only if the elasticity of substitution equals zero. The BLS intends to determine which of these assumptions provides the closest approximation, item category by item category.

Appendix B. Critique of Advisory Commission's Bias Estimates for Fresh Fruits and Vegetables and Motor Fuel56

Fresh fruits and vegetables.

The quote cited above [see section III] indicates that the advisory commission attributes a bias of 20 percent over the period 1967-96 due to increased seasonal availability and variety. It is reasonable to think that, to the extent that consumers value the increased seasonal availability of produce, they will consume more of it. Our analytical framework is to consider the "November strawberry" to be a new good, distinct from the "June strawberry," and measure the consumer surplus associated with the new good.57

Among the various methods that have been proposed for incorporating new goods in a cost-of-living index, Jerry Hausman's suggestion of calculating the consumer surplus from a linearized demand curve is particularly easy to apply to back of the envelope calculations.58 Hausman's linearized method implies that the percentage bias of the price index from failure to incorporate the consumer surplus from a new good, n, is approximately

(1)

where is the percentage expenditure share of the new good after introduction and is its price elasticity of demand. Thus the calculation of consumer surplus and bias can be inferred from information on the expenditure share, which is often readily available, and the elasticity of demand, which can be estimated or inferred from elasticity estimates for similar goods.59

New varieties or seasonal availability of fresh fruits and vegetables face many substitutes, not only from other fresh produce, but also from frozen fruits and vegetables. We assume a value of -1.0 for . Under these assumptions, equation 1 implies that the increased consumption of new seasonal items and varieties as a share of current consumption would need to be quite large-about 40 percent of 1996 expenditures-to be consistent with the advisory commission's estimated index bias of 20 percent.60

Table B1 presents U.S. Department of Agriculture data on changes in per capita consumption of fresh fruit from 1975 to 1995. The change in consumption is shown, somewhat unconventionally, as a percentage of 1995 consumption, because the shares in equation 1 refer to current period consumption. As the advisory commission observes, per capita consumption of many fruits has indeed increased substantially over this period: in particular, limes, cranberries, grapes, kiwifruit, mangos, papayas, and strawberries. Despite these large increases, however, most of these items continue to represent a small percentage of overall fruit consumption, so that the total increase in per capita fruit consumption as a share of 1995 consumption is only 14 percent (measured in pounds). The largest absolute increase in consumption of fruit is that for bananas. We are confident there was no important improvement in seasonal availability of bananas and that there were only minor increases in consumption of new varieties of bananas over this period. In addition, consumption of apples did not change significantly and consumption of oranges decreased. We wonder whether the use of apples for baking may have decreased during this period, which might mask a possible increase in the consumption of raw apples.

We do not attempt to calculate the overall bias using equation 1 because doing so would require average price or expenditure data for each of the detailed categories, which we have not been able to assemble. As mentioned earlier, under Hausman's model and our earlier assumptions, to be consistent with the commission's bias estimate consumption of new varieties and seasonal items would need to increase by about 40 percent over thirty years, which annualizes to 25 percent over the twenty years for which we have data. If increased consumption of seasonal varieties was relatively unimportant for apples, bananas, and citrus fruits, which, according to the Consumer Expenditure Survey of the Bureau of Labor Statistics, together represent 61 percent of dollar expenditures on fresh fruit in 1995, it would be difficult for increased seasonal consumption of the other fruits to produce an estimated bias as large as the commission proposes.

Table B2 shows changes in consumption of vegetables from 1972 to 1995. Unlike the data for fruit, the data for vegetables show important increases in consumption for many items and thus appear, at first glance, to be consistent with the advisory commission's estimates of bias. Under the assumptions stated above, our consumer surplus calculations indicate that for the commission's estimate to hold, the growth in consumption over thirty years would need to be about 40 percent of current consumption, which annualizes to 29 percent over the twenty-three years for which we have consumption data. This is, in fact, very close to the overall increase over this period: 27 percent. We are skeptical, however, about concluding that the increase in consumption derives entirely from improved seasonal availability. A BLS food specialist, Bill Cook, has suggested that the increase in seasonal availability of fresh vegetables mostly occurred before 1985, as evidenced by a 1984 internal BLS study showing that 91 percent of the CPI price quotes for the "other fresh vegetables" category were by then available year round.61 Table B2 shows, however, that almost half of the increase in consumption of fresh vegetables occurred after 1985. Part of the increase appears to have been driven by shifts in preferences, perhaps as a response to improved knowledge about the health benefits of fresh vegetables.

Motor fuel

For the motor fuel category, the advisory commission attributes "a small upward bias of 0.25 percent per year to the CPI for ignoring the convenience and time-saving contribution of automatic credit-card readers built into gasoline pumps."62 Because the commission applies this estimate over a ten-year period, the estimate of the cumulative bias from this source amounts to 2.5 percent.63 Our approach to measuring the consumer surplus created by pay-at-the-pump credit card technology is to attempt to value the saving in time. Suppose that paying at the pump saves two minutes per fill-up, and that the customer's time is valued at $18 per hour (average total compensation per hour for all workers in private industry was $17.49 in 1996). Then the value of paying at the pump is 60 cents per fill-up. Assuming that ten gallons are purchased, the quality bias for the customer who pays at the pump is 6 cents per gallon, or roughly 4.5 percent of the cost of a gallon of gasoline.

Since this service is of value only to the customers who use it, one must next determine the approximate percentage of gasoline purchasers who use pay-at-the-pump technology. Although we have not found direct information on this percentage, the September 1996 issue of the trade journal National Petroleum News reports that 28 percent of the retail facilities operated by thirteen oil companies had installed pay-at-the-pump technology as of 1996.64 Since many of the customers at these stations do not use credit cards, we attempt to find the percentage of gasoline customers who do so. We have not found published information, but an industry source has told us that roughly 35 percent of sales are made through credit cards.

A naive estimate of the proportion of sales using pay-at-the-pump technology would thus be 10 percent (28 percent 35 percent). However, there are at least three reasons why this estimate is too low: first, pay-at-the-pump technology was doubtless first targeted at high-volume sites in areas with high credit card usage; second, the availability of the technology induces customers to make more use of credit cards; and third, the technology is spreading rapidly, so that even estimates published in September 1996 will understate current availability. Consequently we take 25 percent as our estimate of the percentage of customer sales made with pay-at-the-pump technology at the end of 1996. Under these assumptions, we calculate the cumulative index bias from neglecting the benefits of this technology as approximately 1.1 percent (4.5 percent 25 percent), which is less than half of the advisory commission's estimate.

Table B1. Per Capita Consumption of Fresh Fruits, by Type, 1975-95

Units as indicated

Change, 1975-95
Pounds per capita As percentage of In
Type of fruit 1975 1985 1995 1995 consumption pounds
Citrus
Oranges and temples 15.9 11.6 12.3 -29.6 -3.6
Tangerines and tangelos 2.6 1.5 2.0 -27.9 -0.6
Lemons 2.0 2.3 2.9 32.1 0.9
Limes 0.2 0.6 1.2 81.7 1.0
Grapefruit 8.4 5.5 6.0 -38.4 -2.3
Total 29.0 21.5 24.4 -18.9 -4.6

Noncitrus
Apples 19.5 17.3 18.9 -3.0 -0.6
Apricots 0.1 0.2 0.1 20.0 0.0
Avocados 1.2 1.8 1.4 10.9 0.2
Bananas 17.6 23.5 27.4 35.6 9.8
Cherries 0.7 0.4 0.2 -187.5 -0.5
Cranberries 0.1 0.1 0.3 53.3 0.2
Grapes 3.6 6.8 7.6 52.7 4.0
Kiwifruit 0.1 0.5
Mangos 0.2 0.4 1.1 85.8 1.0
Peaches and Nectarines 5.0 5.5 5.4 8.5 0.5
Pears 2.7 2.8 3.4 19.4 0.7
Pineapples 1.0 1.5 1.9 46.6 0.9
Papayas 0.2 0.2 0.4 56.8 0.2
Plums and Prunes 1.3 1.4 0.9 -41.5 -0.4
Strawberries 1.8 3.0 3.8 52.1 2.0
Total 55.1 65.1 73.5 25.0 18.4
Total 84.1 86.5 97.9 14.1 13.8

Source: U.S. Department of Agriculture, Economic Research Service, Fruit and Tree Nuts, FTS-278, October 1996) (table F-29).

Table B2. Per Capita Consumption of Fresh Vegetables, by Type, 1972-95

Units as indicated

Change 1972-95
Pounds per capita As a percentage of In
Type of vegetable 1972 1985 1995 1995 consumption pounds
Asparagus 0.4 0.5 0.6 33.3 0.2
Broccoli 0.7 2.6 3.2 78.1 2.5
Carrots 6.5 6.5 10.1 35.6 3.6
Cauliflower 0.8 1.8 1.3 38.5 0.5
Celery 7.1 6.9 6.4 -10.9 -0.7
Sweet Corn 7.8 6.4 7.8 0.0 0.0
Bell Peppers 2.4 3.8 5.8 58.6 3.4
Onions 10.7 13.6 17.7 39.5 7.0
Tomatoes 12.1 14.9 16.6 27.1 4.5
Cabbage 8.5 8.8 9.1 6.6 0.6
Spinach 0.3 0.7 0.6 50.0 0.3
Cucumbers 3.0 4.4 5.6 46.4 2.6
Artichokes 0.4 0.7 0.4 0.0 0.0
Snap Beans 1.5 1.3 1.6 6.3 0.1
Eggplant 0.4 0.5 0.4 0.0 0.0
Escarole or endive 0.6 0.4 0.2 -200.0 -0.4
Garlic 0.4 1.1 2.1 81.0 1.7
Lettuce
Head 22.4 23.7 21.6 -3.7 -0.8
Leaf or Romaine 3.3 6.0
Watermelon 12.3 13.5 15.9 22.6 3.6
Cantaloupe 7.0 8.5 9.9 29.3 2.9
Honeydews 1.0 2.1 2.4 58.3 1.4
All Others 0.8 0.8 0.7 -14.3 -0.1
Total 107.1 126.8 146.0 26.6 38.9

Source: U.S. Department of Agriculture, Economic Research Service, Vegetables and Specialties: Situation and Outlook Yearbook, VGS-269, July 1996 (table 14).


Appendix C. Examples of New Car Reliability/Durability Quality Adjustments in the CPI Since 1992

  • Improved corrosion protection - body, electrical system, fuel tank, pump, shocks, brakes and cables
  • Increased warranties
  • Body side cladding
  • Sealing improvements
  • Stainless steel exhaust
  • Longer life spark plugs - 100,000 mile life
  • Improved steering gears
  • Powertrain improvements
  • Dextron III transmission fluid - 100,000 mile life
  • Water pump front face - 150,000 mile life
  • Battery saver
  • Increased catalyst load - 100,000 mile life
  • Rust resistant fuel injection -100,000 mile life
  • Clearcoat paint
  • sided galvanized steel body panels
  • Serpentine drive belt

 

Footnotes

1 U.S. Senate, Committee on Finance, Final Report of the Advisory Commission to Study the Consumer Price Index. Print 104-72, 104 Cong., 2 sess., (Washington, D.C., Government Printing Office, 1996).

2 For experimental index results for the poor and elderly subgroups, see Thesia I. Garner, David S. Johnson, and Mary F. Kokoski, “An Experimental Consumer Price Index for the Poor”, Monthly Labor Review, vol. 119, no. 9, September, 1996, pp. 32- 42; and Nathan Amble and Ken Stewart, “Experimental Price Index for Elderly Consumers”, Monthly Labor Review, vol. 117, no. 5, May, 1994, pp. 11-16. These experimental indexes simply reweight CPI price measures to reflect the expenditure patterns of the poor and the elderly, so the price measures are not necessarily representative of the outlets at which these groups shop or the specific items that they purchase. The non-market aspects of quality of life present conceptual and theoretical problems that have not been resolved and, thus, have not received a comprehensive empirical treatment to date.

3 See W. E. Diewert, “The Early History of Price Index Research,” in W. Erwin Diewert and Alice O. Nakamura, eds., Essays in Index Number Theory, Volume 1 (Amsterdam, North-Holland, 1993).

4 The formula used by the BLS for the CPI is sometimes referred to as a “modified” Laspeyres formula because the market basket is representative of expenditures during an earlier period than the period in which it is first used for price comparisons.

5 Until January 1997 there were 207 strata of items. The construction of these indexes involves the random sampling of outlets and areas, and the use of an aggregation formula. The 27 largest Metropolitan Statistical Areas (MSA) along with Anchorage and Honolulu are selected as self-representing Primary Sampling Units (PSU) with certainty. To represent the remaining urban areas a random sample of representative PSU's is selected. The sample of areas underlying the CPI will change in January 1998 as part of the CPI Revision process, as discussed by Janet L. Williams, “The Redesign of the CPI Geographic Sample,” Monthly Labor Review, 119, no. 12, December 1996, pp. 10-17.

6 For a discussion of the relationship of the CPI to the cost-of-living index, see Robert Gillingham, “A Conceptual Framework for the Consumer Price Index,” Proceedings of the American Statistical Association 1974 Business and Economic Statistics Section, (Washington, D.C., American Statistical Association, 1974).

7 To some extent, the definition of a price index is motivated by the uses for which it is intended; see Jack E. Triplett, “Escalation Measures: What is the Answer? What is the Question?” in W.E. Diewert and C. Montmarquette, eds., Price Level Measurement: Proceedings from a Conference Sponsored by Statistics Canada (Ottawa, Statistics Canada, 1983), pp. 457-87.

8 An alternative formulation of the cost-of-living index is in terms of required income rather than expenditure. This formulation would imply the inclusion of income- and wage-based taxes. See, for example, Robert A. Pollak, “The Treatment of Taxes in the Consumer Price Index,” in The Theory of the Cost-of-Living Index (New York, Oxford University Press, 1989), pp. 193-199, and Robert Gillingham and John S. Greenlees, “The Impact of Direct Taxes on the Cost of Living,” Journal of Political Economy, 95, no. 4, August 1987, pp. 775-796.

9 See Pollak, Theory of the Cost-of-Living Index, and Gillingham, “A Conceptual Framework.”

10 U.S. Senate, Committee on Finance, Final Report, p. 37. The weight for employer-provided medical care also is excluded from the CPI. The proper treatment of this benefit, as well as other in-kind employee compensation, involves more difficult conceptual issues, and depends in part on the uses made of the index. See Ralph Turvey et al., Consumer Price Indices: An ILO Manual (Geneva, International Labour Office, 1989).

11 See W. E. Diewert, “Exact and Superlative Index Numbers,” in Diewert and Nakamura, eds., Essays in Index Number Theory, Volume 1, pp. 223-252.

12 Whereas the BLS collects and processes CPI data monthly, most CPI expenditure data are drawn from the CEX household interview survey, which is conducted quarterly. Fully edited expenditure data for a given year are not available until late in the following year. As will be described in section VII below, the BLS plans to take steps to expedite the processing of the CEX data, but updating of expenditure weights on a monthly basis would be prohibitively expensive.

13 The original research was published in Ana M. Aizcorbe and Patrick C. Jackman, “The Commodity Substitution Effect in CPI Data, 1982-1991,” Monthly Labor Review, 116, no. 12, December 1993, pp. 25-33. These estimates subsequently have been updated by BLS staff. Until the introduction of the updated market basket in January 1998, the CPI may exceed the superlative indexes by somewhat more than this amount, as indicated by John S. Greenlees, “Expenditure Weight Updates and Measured Inflation,” paper prepared for Third Meeting of the International Working Group on Price Indices, Voorburg, Netherlands, April 16-18, 1997 (Washington, D.C., Bureau of Labor Statistics); and Matthew D. Shapiro and David W. Wilcox, “Alternative Strategies for Aggregating Prices in the CPI,” paper presented at Federal Reserve Bank of St. Louis Fall Policy Conference on Measuring Inflation and Real Growth, St. Louis, October 16-17, 1996 (University of Michigan).

14 For example, the commission is unclear about whether, in their usage, substitution bias equals, includes, or is distinct from formula bias. At one point they state “…what we called `formula bias' [we] now refer to as Lower Level Substitution Bias” (see U.S. Senate, Committee on Finance, Final Report, p. 19). Subsequently, they state “BLS has reduced so-called formula bias, the part of Lower Level Substitution Bias resulting in substantial measure from the introduction of sample rotation procedures” (p. 44). Then they state “Changing to geometric means will not only solve the `formula bias' problem…but will also alleviate the below-stratum-level substitution bias” (p. 51). The last of these statements, indicating that formula bias and substitution bias are distinct phenomena, most closely agrees with definition of formula bias that was given when it was originally identified by BLS research; see Marshall B. Reinsdorf, “Price Dispersion, Seller Substitution, and the U.S. CPI,” BLS working paper 252 (Washington, D.C., Bureau of Labor Statistics, 1994). Appendix A of the present report describes an additional confusion with the commission's example of the related property of “time reversibility.”

15 The research paper cited by the commission is Brent R. Moulton and Karin E. Smedley, “A Comparison of Estimators for Elementary Aggregates of the CPI,” paper presented at Western Economic Association International conference, San Diego, July 7, 1995 (Washington, D.C., Bureau of Labor Statistics).

16 See Robert McClelland, “Evaluating Formula Bias in Various Indexes Using Simulations,” BLS working paper 289, 1996; and Brent R. Moulton, “Estimation of Elementary Indexes of the Consumer Price Index,” paper presented at American Statistical Association conference, Chicago, August 5, 1996 (Washington, D.C., Bureau of Labor Statistics).

17 The elasticity of substitution is a measure of consumer willingness to substitute between commodities and is defined by economists as the proportionate change of relative quantities demanded divided by the proportionate change of relative prices.

18 See Marshall Reinsdorf, “The Effect of Outlet Price Differentials on the U.S. Consumer Price Index,” in Murray F. Foss, Marilyn E. Manser and Allan H. Young, eds., Price Measurements and Their Uses, (Chicago, University of Chicago Press, 1993).

19 Reinsdorf, “Price Dispersion.”

20 See Brent R. Moulton and Karin E. Moses, “Addressing the Quality Change Issue in the Consumer Price Index,” forthcoming in Brookings Papers on Economic Activity 1997:1 (Washington, D.C., Bureau of Labor Statistics, 1997). These figures are somewhat different from those reported in an earlier version of their paper. They refined their prior estimates principally to exclude some “quality adjustments” that are made to account for simple changes in units of measurement or package size.

21 U.S. Senate, Committee on Finance, Final Report, p. 28.

22 The analysis is taken from Moulton and Moses, “Addressing the Quality Change Issue.”

23 U.S. Senate, Committee on Finance, Final Report, p. 30. The commission provides no direct support for this estimate, although reference is made to the changing characteristics of new single-family houses over the same period. They also cite increases in the average number of bathrooms, and in the share of units containing central air conditioning, within the stock of rental units.

24 Brent R. Moulton, “Issues in Measuring Price Changes for Rent of Shelter,” unpublished paper presented at Conference on Service Sector Productivity and the Productivity Paradox, Ottawa Canada, April 11-12, 1997 (Washington, D.C., Bureau of Labor Statistics).

25 See Jack E. Triplett, “Quality Bias in Price Indexes and New Methods of Quality Measurement,” in Zvi Griliches, ed., Price Indexes and Quality Change: Studies in New Methods of Measurement, (Cambridge, MA, Harvard University Press, 1971); Paul A. Armknecht, “Quality Adjustment in the CPI and Methods to Improve It,” in American Statistical Association 1984 Proceedings of the Business and Economic Statistics Section (Washington, D.C., American Statistical Association, 1984); Paul A. Armknecht and Donald Weyback, “Adjustments for Quality Change in the U.S. Consumer Price Index, Journal of Official Statistics 5, 1989, pp. 107-23; Paul R. Liegey, Jr., “Adjusting Apparel Indexes in the Consumer Price Index for Quality Differences,” in Murray F. Foss, Marilyn E. Manser, and Allan H. Young, eds., Price Measurements and Their Uses, (Chicago, University of Chicago Press, 1993); Paul R. Liegey, Jr., “Apparel Price Indexes: Effects of Hedonic Adjustment,” Monthly Labor Review 117, May 1994, pp. 38-45; Marshall B. Reinsdorf, Paul Liegey, and Kenneth Stewart, “New Ways of Handling Quality Change in the U.S. Consumer Price Index,” BLS working paper no. 276 (Washington, D.C., Bureau of Labor Statistics, 1995); and Moulton and Moses, “Addressing the Quality Change Issue.”

26 The studies cited by the commission are Matthew D. Shapiro and David W. Wilcox, “Mismeasurement in the Consumer Price Index: An Evaluation,” in Ben S. Bernanke and Julio J. Rotemberg, eds., NBER Macroeconomics Annual 1996, (MIT Press, 1996); and David M. Cutler, Mark McClellan, Joseph P. Newhouse, and Dahlia Remler, “Are Medical Prices Declining?” NBER working paper no. 5750 (Cambridge, MA, National Bureau of Economic Research, 1996). The latter study, of heart attacks, was supported in part by the BLS.

27 See Elaine M. Cardenas, “Revision of the CPI Hospital Services Component,” Monthly Labor Review, vol. 119, no. 12, December 1996, pp. 40-48.

28 For examples of some of the methods that have been proposed, see the papers in Timothy F. Bresnahan and Robert J. Gordon, eds., The Economics of New Goods, (Chicago, University of Chicago Press, 1997).

29 The advisory commission uses two different methods for numbering their recommendations. See U.S. Senate, Committee on Finance, Final Report, pp. 2-3 and pp. 49-55. Herein we follow the numbers and text from pp. 2-3.

30BLS Handbook of Methods, Bulletin 2490, 1997, p. 170.

31 U.S. Senate, Committee on Finance, Final Report, p. 50.

32 See Ana M. Aizcorbe, Robert A. Cage, and Patrick C. Jackman, “Commodity Substitution Bias in Laspeyres Indexes: Analysis Using CPI Source Data for 1982-1994,” paper presented at the Western Economic Association International Conference in San Francisco, July 1996 (Washington, D.C., Bureau of Labor Statistics); and Shapiro and Wilcox, “Alternative Strategies.”

33 See Shapiro and Wilcox, “Alternative Strategies.” The CES formula that they proposed was originally derived by P.J. Lloyd, “Substitution Effects and Biases in Nontrue Price Indices,” American Economic Review, vol. 65, June 1975, 301-13, and was suggested by BLS staff as a method for approximating a superlative index without current expenditure data.

34 See Bureau of Labor Statistics, “The Experimental CPI using Geometric Means (CPI-U-XG),” (Washington, D.C., Bureau of Labor Statistics, April 10, 1997).

35 See S.G. Leaver, W.H. Johnson, R.M. Baskin, S. Scarlett, and R. Morse, “Commodities and Services Sample Redesign for the 1998 Consumer Price Index Revision,” Proceedings of the Survey Research Methods Section, American Statistical Association, 1996, forthcoming.

36 U.S. Senate, Committee on Finance, Final Report, p. 51.

37 See Ralph Bradley, Bill Cook, Sylvia G. Leaver, and Brent R. Moulton, “An Overview of Research on Potential Uses of Scanner Data in the U.S. CPI,” paper presented at the Third Meeting of the International Working Group on Price Indices, Voorburg, Netherlands, April 16-18, 1997 (Washington, D.C., Bureau of Labor Statistics).

38 Walter Lane, “Changing the Item Structure of the Consumer Price Index,” Monthly Labor Review vol. 119 no. 12, December 1996, 18-25.

39 Lane, “Changing the Item Structure,” p. 22.

40 See U.S. Senate, Committee on Finance, Final Report, p. 52.

41 The January 1997 consolidation of three CPI item strata—hospital room, other inpatient services and outpatient services—into one hospital services stratum was designed in part to capture substitution among those three settings for treatment provision. The inclusion of new cars and new trucks in a single new vehicles stratum is an example of a similar change taking place as part of the January 1998 introduction of the revised CPI market basket.

42 See “Changing the CPI Homeownership Method to Rental Equivalence,” CPI Detailed Report, Bureau of Labor Statistics, January 1983, pp. 3-17.

43 Automobile and tenants insurance policies are priced directly in the CPI.

44 For discussions of past BLS research on the direct pricing of health insurance policies, and on the user-cost and leasing-equivalence approaches to pricing of automobile services, see Paul A. Armknecht and Daniel H. Ginsburg, “Improvements in Measuring Price Changes in Consumer Services: Past, Present, and Future,” in Zvi Grilliches, ed., Output Measurement in the Service Sectors, (Chicago, University of Chicago Press, 1992).

45 See, e.g., Pollak, Theory of the Cost-of-Living Index; and Cutler, et al., “Are Medical Prices Declining?”

46 U.S. Senate, Committee on Finance, Final Report, p. 30.

47 The commission's discussion of the appearance of AIDS, however, suggests agreement with the idea that not all changes in the quality of life ought to be reflected in the CPI (U.S. Senate, Committee on Finance, Final Report, p. 47).

48 See Jack E. Triplett, “Concepts of Quality in Input and Output Price Measures: A Resolution of the User Value–Resource Cost Debate,” in Murray F. Foss, ed., The U.S. National Income and Product Accounts: Selected Topics (Chicago, University of Chicago Press, 1983).

49 For discussion of the quality adjustment methods used by the BLS, see Paul A. Armknecht, Walter F. Lane, and Kenneth J. Stewart, “New Products and the U.S. Consumer Price Index,” in Timothy F. Bresnahan and Robert J. Gordon, eds., The Economics of New Goods (Chicago, University of Chicago Press, 1997); Reinsdorf, Liegey, and Stewart, “New Ways of Handling Quality Change;” and Moulton and Moses, “Addressing the Quality Change Issue.”

50 Moulton and Moses, “Addressing the Quality Change Issue,” Table 4.

51 Martin Feldstein, in testimony before the Senate Finance Committee (February 11, 1997), has agreed that the CPI must be based on tested and reliable statistical methods, even though in his view the resulting estimate will overstate the true increase in the cost of living.

52 U.S. Senate, Committee on Finance, Final Report of the Advisory Commission to Study the Consumer Price Index. Print 104-72, 104 Cong., 2 sess. (Washington, D.C., Government Printing Office, 1996), p. 17.

53 U.S. Senate, Committee on Finance, Final Report, p. 42. The article cited in this quotation is Matthew D. Shapiro and David W. Wilcox, “Mismeasurement in the Consumer Price Index: An Evaluation,” in Ben S. Bernanke and Julio J. Rotemberg, eds., NBER Macroeconomics Annual 1996, (MIT Press, 1996).

54 The BLS authors cited by Shapiro and Wilcox are Brent R. Moulton, “Basic Components of the CPI: Estimation of Price Changes,” Monthly Labor Review, 116, no. 12, December 1993; Marshall B. Reinsdorf and Brent R. Moulton, “The Construction of Basic Components of Cost-of-Living Indexes,” in Timothy F. Bresnahan and Robert J. Gordon, eds., The Economics of New Goods (Chicago, University of Chicago Press, 1997); and Brent R. Moulton and Karin E. Smedley, “A Comparison of Estimators for Elementary Aggregates of the CPI,” paper presented at Western Economic Association International conference, San Diego, CA, July 7, 1995 (Washington, D.C., Bureau of Labor Statistics).

55 Shapiro and Wilcox, “Mismeasurement,” fn. 22, p. 111.

56 This appendix is an excerpt from a study by BLS researchers: Brent R. Moulton and Karin E. Moses, “Addressing the Quality Change Issue in the Consumer Price Index,” forthcoming in Brookings Papers on Economic Activity 1997:1 (Washington, D.C., Bureau of Labor Statistics, 1997).

57 See W. Erwin Diewert, “The Treatment of Seasonality in a Cost-of-living Index,” in W.E. Diewert and C. Montmarquette, eds., Price Level Measurement: Proceedings from a Conference Sponsored by Statistics Canada (Ottawa, Statistics Canada, 1983).

58 See Jerry Hausman, “Cellular Telephone, New Products and the CPI,” unpublished paper (Massachusetts Institute of Technology, 1997). Hausman refers to his linearized method as a “lower bound” on the consumer surplus, but it is unclear to us whether the conditions for the method to be a lower bound—a convex shaped demand curve—necessarily hold in all cases.

59 If a new variety fully replaces an old one, the consumer surplus calculation should deduct the lost surplus of the disappearing variety from the surplus gained from the new variety.

60 Ideally, one would examine monthly consumption data to isolate seasonal changes in consumption, but such data do not appear to be available.

61 Internal memorandum from William L. Weber to Dan Ginsburg, U.S. Bureau of Labor Statistics, May 25, 1984.

62 U.S. Senate, Committee on Finance, Final Report of the Advisory Commission to Study the Consumer Price Index. Print 104-72, 104 Cong., 2 sess. (Washington, D.C., Government Printing Office, 1996), p. 36.

63 We also note that the report does not address possible unmeasured decline in retail services, such as the introduction of fees for providing air for tires at some service stations. In addition, the advisory commission incorrectly assumes that the CPI does not make quality adjustments for air pollution mandates and, agreeing with this supposed BLS practice, makes no bias adjustment for the mandates itself. Since BLS does, in fact, make cost-based adjustments for motor fuel pollution mandates, the commission presumably should have counted these as downward bias (see U.S. Bureau of Labor Statistics, “Quality Adjustment for Gasoline,” CPI Detailed Report, January 1995, p. 8).

64 “Pay-at-the-Pump Shows Solid Growth in '90s,” National Petroleum News, September 1996, p. 22.

 

Last Modified Date: October 16, 2001