There has been widespread interest in evaluating the adequacy of the Consumer Price Index (CPI) as used for various purposes. Some of that interest reflects a view that the annual inflation rate of the CPI exceeds that of some “true” cost-of-living index and leads to an overstatement of cost-of-living adjustments for social security and other public programs. But there has also been a growing research literature among economists and statisticians, much of it from within the Bureau of Labor Statistics (BLS), critically examining and exploring means for improving the design and estimation techniques underlying the CPI.
Price indexes have often been popularly labeled cost-of-living indexes; indeed, until it was renamed the CPI in 1945, the index long published by the BLS had officially been labeled a cost-of-living index. Several decades ago the CPI was widely used in labor contracts to index wages, with the goal of providing automatic adjustments to keep wages fully or partially current with changes in the cost of living. With the same goal in mind—protection against changes in the cost of living—the Congress determined in 1972 that the CPI should be used to make annual “cost-of-living adjustments” to social security benefits, and the practice was subsequently extended to many other public transfer payments. Since 1985 the CPI has been used to index tax brackets, exemptions, and deductions in the tax code so as to “neutralize” the effects of inflation. The annual change in the index is widely used in and outside government as a broad measure of inflation. And its components are the main source of deflating the current dollar value of
consumer expenditures as part of the measurement of the nation’s gross domestic product (GDP).
The growth rate of traditional price indexes like the CPI, which measure the cost of purchasing a fixed basket of goods and services, tends to outpace cost-of-living indexes, which attempt to calculate the change in expenditure needed to maintain living standards. Concerns have long been expressed that the CPI does not adequately take account of improvements in the quality of consumer goods and services in a technologically dynamic economy and thereby overstates the price increases consumers are paying for goods of constant quality. As a consequence, indexing wages, social security benefits, or other payments scaled to the CPI would usually overstate the amount needed to compensate for increases in the cost of living. Forty years ago, the Stigler Committee outlined the conceptual and measurement characteristics of the CPI that distinguished it from a “true cost-of-living index”—or, under alternative committee labels, a “welfare index,” or a “constant utility” index (National Bureau of Economic Research, 1961). The principal recommendation of the committee was the establishment of a long-run research program designed to make the CPI a better approximation to a cost-of-living index.
In recent years, as the projected long-term financing deficit in the social security system has grown, the question of whether and to what extent the CPI is biased upward, and therefore “overcompensates” social security beneficiaries, has become a concern among some legislators. In 1995 the Senate Finance Committee appointed an Advisory Committee to Study the Consumer Price Index (widely known as the Boskin commission after its chair, Michael Boskin) to review this issue. In its widely publicized final report of December 1996, the Boskin commission concluded that the CPI was currently overstating the rate of increase in consumers’ cost of living by about 1.1 percentage points a year, and it cited estimates from other research pointing to approximately the same result. The commission recommended a number of steps designed to move the CPI away from what was essentially an index of the cost of purchasing a fixed basket of consumer goods toward what would be more nearly a cost-of-living index (COLI).
In 1997 the BLS reported to Congress that it had been using a COLI concept for many years to help make decisions about the CPI and that it accepted a COLI as the measurement objective for the index (Bureau of Labor Statistics, 1997c).1 The report of the Boskin commission, however, undoubtedly spurred BLS to broaden and make more explicit that commitment, and it only recently began taking steps to modify the fixed-weight structure of the CPI so as to bring it closer to a COLI.
A fixed-basket, or fixed-weight, price index is essentially just that: it measures changes in the cost of purchasing a fixed basket of goods (and services). For the CPI, price quotes are collected monthly, selected to be representative of the various categories of consumer goods and services. The observed price changes are assigned weights, representing the importance of each category in aggregate consumer expenditures during some base period, then combined into the major CPI subcomponents, such as food, shelter, appliances, and so forth and, subsequently, into an overall national average.
A COLI is more ambitious and correspondingly more difficult to produce in that its objective is to measure changes in living costs. Viewed from the standpoint of an individual household, a COLI seeks to measure the percentage change in expenditures a household would have to make in order to hold constant some specified standard of living or level of material well-being.2 In an aggregate COLI, price and expenditure data must be combined to produce an estimate that reflects some measure of average change in the cost of living for all (or some subgroup of) households.
In recently reiterating its acceptance of a COLI as the measurement objective for the CPI, the BLS added a number of important cautions: “It [the COLI] is a theoretical concept based on the well-being of the individual consumer, so . . . additional assumptions about how to apply it as a measurement objective for an aggregated set of consumers . . . must be made” (Bureau of Labor Statistics, 1997b: 3). Further: “While the CPI may be described formally in the context of a cost-of-living index, there is no single all-purpose definition of the target.” The concept of the standard of living that is to be held constant in a cost-of-living index is far from unambiguous. Various analysts have offered different definitions of what universe it should cover (e.g., the standard of living obtainable from public and private goods or from private goods only), and embedding the concept in a regularly published statistical index raises thorny problems.
The discussion and controversy about the CPI reflect a large number of conceptual and measurement issues: As a guide for the BLS in making decisions about how the index should be designed and measured, what are the advantages and limitations of the concepts that underlie fixed-weight and cost-of-living indexes? For many years attention centered on the “substitution issue”: To what extent is it possible to incorporate into the index the tendency of households to shift their purchases toward those goods whose prices have risen the least or fallen the most? But there are other important questions of index design whose
See Chapter 2 for a discussion of standard of living in the context of cost-of-living theory. Briefly, consumers think more goods are better than less and can consistently rank alternative bundles of goods in terms of a set of preferences. Constrained by income and prices, each consumer chooses the most preferred bundle of goods. The consumer’s “standard of living” (or “material well-being”) is a measure of the extent to which preferences are satisfied.
resolution depends in part on whether one evaluates them through the prism of a fixed-weight or a cost-of-living viewpoint and on how the cost-of-living approach is interpreted: How aggressively and comprehensively should the BLS pursue efforts to use econometric techniques to adjust observed prices for the effect of quality improvements? How comprehensive should be the universe of goods covered by the CPI—should it cover private goods only or also encompass public goods? Should the BLS take into account, to the extent measurement is feasible, the effects on living standards—and therefore on living costs—of changes in pollution, crime rates, congestion, and other “environmental” developments? How should the index take account of the effect on living standards of the continual introduction of new goods in our technologically innovative society? And, in either a fixed-weight or a cost-of-living index, how should the experiences of the rich and the poor, the old and the young, be combined into a single index, and should indexes for population subgroups also be published?
The difference between the two approaches to index construction is not fully captured by juxtaposing the terms “fixed-weight index” and “cost-of-living index.” The objectives of the two indexes are not the same. The former seeks to measure the effects of price changes on the cost to a household of purchasing a specified basket of goods and services. The latter seeks to measure the effects in terms of the cost of maintaining the household’s standard of living at some specified level. The two effects are not usually the same. And, in a world in which consumer tastes change and the qualities of many goods and services are constantly being altered, measuring either type of index is a difficult task. A more appropriate terminology would contrast a “cost-of-goods index” (COGI, where “goods” includes both goods and services) with a “cost-of-living index.”3 Considered from the standpoint of an individual household, a COGI seeks to provide a measure of the percentage change in expenditures the household would require to purchase a basket of goods, given a change in prices between some initial period (usually called the reference period) and some later (comparison) period.4 As its name implies, it seeks to measure changes in the cost of goods. In principle, for a COGI, specification of the basket of goods may be based on a past period’s consumption patterns or current patterns, or even a point in between. A COLI, as noted above, seeks to measure the percentage change in expenditures needed to maintain a household’s standard of living at some specified level (typically, but not necessarily, the level it had in the reference period). As its name implies, its objective is to measure changes in the cost of living.
PANEL CHARGE AND REPORT ORGANIZATION
In view of the wide range of important issues that have been raised, the BLS asked the Committee on National Statistics to convene the current panel, which was charged with two primary tasks: “(1) investigating conceptual, measurement, statistical, and data issues in the development of cost-of-living indexes and (2) assessing the appropriate use of such indexes for indexing federal programs and other purposes.” The statement of task further notes: “Topics of the assessment would include the required frequency, the technical appropriateness of revisions, and the treatment of quality change and new products. The panel would be asked to make explicit the assumptions and models underlying different approaches and to recommend a program of research and experimental measures.” The remainder of this initial chapter provides a brief introduction to the key issues and problems considered by the panel in its effort to contribute to the understanding of price and cost-of-living indexes and their limitations and complexities.
Chapter 2 considers alternative conceptual foundations for the CPI, specifically COGI and COLI approaches. The chapter steps through the key attributes (many of which are taken up in greater detail later in the report) that define price and cost-of-living indexes and examines the relative strengths, weaknesses, limitations, and implications associated with each of these approaches.
Chapter 3 presents the panel’s assessment of what goods, broadly defined, are appropriate for inclusion in the scope of CPI coverage. An all-encompassing cost-of-living index would attempt to cover—in addition to private market goods —goods provided by government, environmental amenities, and other nonprivate societal conditions (such as public safety).
Chapter 4 discusses the conceptual rationale, methodology, and limitations of adjusting indexes or observed price quotes to account for changing item quality. The chapter reviews in detail current BLS approaches of price adjustment that come into play when items are replaced in sampled outlets. The panel assesses these methods and advances proposals relating to the use and potential of different quality adjustment methods.
Chapter 5 first discusses two issues related to the introduction of new goods: (1) what criteria should determine when and how new goods are introduced into the index and (2) should estimates of “virtual price” decreases associated with their introduction be made and incorporated into the CPI? The second part of the chapter addresses how changes in the patterns of consumer patronage among different types of retail outlets affect living costs and price indexes. The panel specifically considers what, if anything, BLS could do to identify and estimate quality and pure price components of differences in the observed prices of goods across outlets.
Chapter 6 examines conceptual and measurement issues pertaining to the construction of the complicated medical services component of the CPI. Those complications include high variability of prices paid for equivalent services,
defining a medical “good,” involvement of insurers and government in transactions, pricing risk, and how adjusting medical care prices to account for the quality of outcomes can lead to strange results. The chapter also discusses “outcomes” and direct insurance pricing options.
Chapter 7 examines the relationship between each of the major purposes for which the CPI is used and the appropriate design of the index. It considers the extent to which different index designs are required for different purposes and when a single design can serve as an acceptable measurement instrument for many purposes. It also spells out the implications for various public policy purposes of choosing one index design over another.
Chapter 8 describes the issues that are confronted when a single index must be produced to represent the changes in prices or living costs faced by a heterogeneous population. It emphasizes problems that arise because different consumers buy different types and qualities of goods and pay different prices for them.
Chapter 9 provides an overview and assessment of the current data structure that underlies the CPI and considers ways that data and survey advances might be coordinated to improve the accuracy of the CPI. It also describes the extent to which different data structures permit flexibility in constructing alternative or supplemental indexes (such as for subpopulations).
ALTERNATIVE APPROACHES: A COGI VERSUS A COLI
Between a policy of continuing a traditional COGI with modest changes and one of attempting the modifications necessary to produce a cost-of-living index that reflects the most comprehensive definition of the “standard of living,” there is a wide range of intermediate possibilities. Indeed, starting from either basic approach—the fixed-basket price index or the cost-of-living index—many of the same kinds of questions must be faced.
If one thinks of a “simple” fixed-basket index and a comprehensively defined COLI as opposite ends of a spectrum, it is clear that neither alone provides an operational model with which a CPI can be constructed. For example, in a modern innovative economy, even over a relatively short period of time, the characteristics of a wide range of goods and services are constantly changing. When consumers pay more for a new model of a good, how much of that represents a true price increase and how much a payment for higher quality? Moreover, goods with completely new characteristics, like DVDs, come to the market and gradually take over some or all of the market for other goods. In the case of a COLI, the complexities, lack of data, and deviations from the assumptions of the theory that are sometimes encountered in the real world impose limits on the extent to which its implied objective can be achieved. Attempting to push beyond those limits risks introducing an unacceptable amount of subjectivity and the possibility of significant error into index measurement.
The cost-of-living concept has been used to guide index construction by applying the economic theory of consumption to specific problems. The theory assumes that households act rationally to achieve the highest possible standard of living given their income and the prices they face. They, therefore, allocate their incomes so that, at the margin, goods for which they pay more make a larger contribution to their standard of living. As a result, information about the relative values to consumers of different goods can be inferred from their relative prices. However, since individuals have differing (marginal) evaluations of quality, the conceptual framework for deriving overall quality adjustments from observed differences in market prices raises some difficult issues that have not been fully worked through by the BLS or by academic researchers.
In applying the theory to specific cases, it is essential to examine how closely the underlying assumptions match, or depart from, the actual behavior of consumers and markets in the particular case at hand. To take an important example, the allocation of a consumer’s expenditures on medical care is to a major extent determined not by the buyer (the consumer) but by a physician, and those expenditures often come not directly out of the consumer’s income, but rather from private or public insurance payments.
DOMAIN OF THE CPI
What goods and service should be covered by the CPI—what should be its domain? In the traditional fixed-weight CPI, and in its counterparts in other countries, the domain is specified by the very definition of the index: it measures changes over time in the cost of purchasing a fixed market basket of goods and services. Its domain, therefore, is the universe of private goods and services. A relatively narrow range of essentially private goods that the government produces and sells in the marketplace, such as entrance fees to national parks or fares on a publicly owned transit system, are included. But no effort is made to estimate a “price” for truly public goods and services, such as national defense or the administration of justice. The adoption of a cost-of-living approach to index construction, however, raises a number of questions about what the index ought to cover beyond what is currently included in the CPI since, in addition to the purchase of private goods and services, a large number of economic, social, and environmental factors clearly have an effect on the standard of living and therefore on the cost of living. (For ease of exposition we use the term “outside conditions” when referring collectively to all of these factors.)
Our panel examined the issue of the appropriate scope or domain of a COLI from several perspectives. If we assume (perhaps unrealistically) that tools could be developed to measure the effects on the cost of living that arise from changes in outside conditions and government actions, we must then ask: Should one include the estimated effects of those conditions on a cost-of-living index that is used for the major purposes served now by the CPI? What role should the BLS
play, and what priorities should it give to research devoted to developing experimental measures of the contribution to national output and welfare associated with changes in outside conditions and government programs?
With a very broad definition of what should be included in its domain, a COLI would be adjusted up or down to take account of the positive or negative effects on consumer well-being arising from a wide range of sources outside the marketplace that have not traditionally been considered relevant for inclusion in the CPI. These include, among other elements, the quality of the air, water, and other environmental amenities; the presence or absence of congestion on roads and in neighborhoods; changes in perceptions about personal security associated with trends in the crime rate; the effects of significant climate changes; and increases in longevity arising from broad environmental factors (aside from those associated with specific medical procedures).5 As noted above, the analytic techniques and statistical tools to measure most of these kinds of effects do not currently exist or, if they have been tried, they are still controversial and speculative.
Public Goods, Publicly Provided Private Goods, and Taxes
The CPI is now based on prices charged for private goods, i.e., goods that are sold to individual households. A few of these goods (e.g., tuition at public colleges, bus fares on city-owned buses, or entrance fees to public parks) are private goods produced by government and sold on an individual basis. The CPI does not include public goods (e.g., national defense, clean air) that are made available freely rather than through individual sales.6 And yet the increased or decreased availability of those goods does affect living standards. Should the value of some or all types of these public goods be included in the CPI, with the taxes to pay for them treated analogously to prices? When goods made available by government are classified in terms of how similar or different they are from the kinds of private goods currently priced in the CPI, they range across a wide spectrum. At the “nearly private” end are things like the airline ticket taxes charged by the government and used to finance flight control, safety, and other operations of the Federal Aviation Administration. The connection between gasoline taxes and highway construction is somewhat looser, but the taxes do bear some relationship
to the quantity and quality of the goods delivered. As a general rule, an increase in sales taxes is passed on to consumers in the form of higher prices and shows up as an increase in the CPI. Some economists argue that increases in sales tax rates should be adjusted out of any index whose objective is to measure the cost of living, on grounds that the addition to living costs caused by the higher sales tax rates is offset by the benefits from the additional public goods provided thereby (Nordhaus, 1998).
In the examples noted above, one could argue the existence of a rough connection between what individual consumers pay in taxes and the quantity or quality of the services they receive. But what about pure public goods, such as national defense or law enforcement, the benefits of which are not parceled out to individuals? Every individual, willy-nilly, gets the same “quantity” of national defense. Nevertheless, accepting a broad definition of the standard of living would extend the domain of the index to include the value of public goods with net taxes (i.e., taxes minus transfer payments) treated analogously to prices in a cost-of-living index.
In a similar vein, the enactment of environmental, health, and safety regulations requires businesses to incur extra costs that, when passed on in higher prices, are captured by the CPI. But these regulations reduce environmental damages (broadly defined) and increase consumer welfare. Should an estimate of such additional costs be subtracted from the CPI on grounds that they are balanced by the welfare gains?
Employer-Provided Fringe Benefits
In the United States, employers pay, in part or in full, for a wide range of “inkind” benefits for their workers, health insurance being the most prominent example. The CPI now excludes from the weights assigned to medical care the value of the health insurance premiums paid by employers. Is that treatment appropriate? More generally, how should the BLS treat in-kind employer fringe benefits in designing the CPI?
The traditional Laspeyres version of the COGI weights the prices of various items in both the initial (reference) and ending (comparison) periods by the quantities purchased in the reference period. Considered from the standpoint of an individual household, such an index reflects the percentage increase in expenditures the household would have to incur in order to buy the reference period basket of goods at the new, comparison period prices.7 But when faced with
changes in relative prices—pork prices rise a lot and beef prices only a little— consumers try to minimize the effect of the price rise on their living standards: they shift their purchases, buying fewer of the goods whose relative prices have risen and more of those whose relative prices have fallen. This substitution allows them to improve their living standards relative to what would have been the case had they been constrained to maintain their old buying patterns in the face of the price changes. The traditional Laspeyres version of the fixed-basket index takes no account of these possibilities since it simply weights both sets of prices by the reference period quantities. As a consequence, the Laspeyres index tends to overstate the rise in the cost of maintaining the reference period’s standard of living.
An alternative weighting scheme is the Paasche index, which uses as weights the quantities purchased in the ending, or comparison, period. It measures the percentage difference in expenditures between what it would cost the household to buy the comparison period quantities at the old prices and what it costs at the new ones. But because the comparison period quantities already incorporate the household’s substitution in favor of goods whose prices have risen the least, the Paasche index understates the cost of maintaining the comparison period’s standard of living at the old prices. Equivalently, the Paasche index understates the change in the household’s cost of living, as evaluated at the comparison period’s standard of living.
If substitution behavior plays a major role in explaining changes in quantities purchased between the reference and comparison periods, goods that have experienced relatively large price increases will tend to receive higher weights in the Laspeyres than in the Paasche index, and the opposite will be true for goods that have experienced relative price decreases. Thus, the Paasche index typically tends to produce a lower estimate of average price increase than the Laspeyres index.
Notice, however, that the Laspeyres index overstates the cost of maintaining the reference period’s standard of living while the Paasche index understates the cost of maintaining the comparison period’s standard of living. Under conditions in which those two standards of living are significantly different—due perhaps to the size and pattern of the relative price changes or to changes in the incomes and tastes of consumers between the two periods—it is conceptually possible that the change in quantities is not dominated by the effects of substitution behavior. In that case the Laspeyres might produce a lower estimate of price increase than the Paasche. Nevertheless, empirical studies have shown that when the actual inflation measures produced by the two types of indexes are compared over various historical periods, Laspeyres indexes consistently produce a higher measured rate of inflation than Paasche indexes. This finding is widely interpreted as a testament to the importance of substitution behavior by individual households. Similarly, the magnitude of the differences between the two estimates is usually thought to depend on how much relative prices have changed and how much
consumers alter their spending patterns when faced with changes in relative prices.
In the 1920s American economist Irving Fisher proposed what he called an “ideal” index that is formed as the geometric mean (the square root of the product) of Laspeyres and Paasche indexes and thus incorporates information about consumer spending patterns from both the base and comparison periods (see Fisher, 1922). In a 1924 article (not available in English until 1939), Russian economist Alexander Konus formally showed how to construct a cost-of-living index as the ratio of the minimum costs required for a consumer to achieve a given standard of living. He also established the relationships between the Laspeyres index and the cost-of-living index for the reference period’s standard of living and between the Paasche index and the cost-of-living index for the comparison period. In 1976 W. Erwin Diewert demonstrated that a class of indexes could be constructed using only information on actual quantities and prices in the two periods that would closely approximate a Konus cost-of-living index (see Chapter 2) for some standard of living intermediate between those in the base and comparison periods and would do so for any pattern of (stable) consumer tastes. He labeled such measures superlative indexes. The Fisher ideal index is one of many possible formulations of a superlative index, all of which involve some form of averaging base period and comparison period weights.
Two Levels of Index Construction Underlying the CPI
Familiarity with several key aspects of the way BLS gathers and combines individual price data into an overall index is necessary for understanding how the substitution issue affects the CPI.8 The BLS collects roughly 80,000 individual prices each month from over 21,000 retail outlets in various geographic areas around the country. For a few CPI categories, it also collects data from 7,300 housing units. The individual prices are classified into 218 categories (termed strata) that represent the various types of goods that consumers buy. From the individual item prices that have been collected, separate price indexes are then computed for each stratum in each area, with weights based on the importance in consumer spending of each of the items included in the stratum.9 This process is called lower-level aggregation. The resulting 218 strata indexes are in turn com-
bined into an overall CPI with weights derived from the Consumer Expenditure Survey (CEX), reflecting consumption patterns in the base period (currently 1993-1995). This second stage in the process is referred to as upper-level aggregation.
How BLS Deals with Substitution in the CPI
If the goods within a stratum are similar in terms of meeting a particular consumer demand (a characteristic of most but by no means all strata), consumer substitution among individual products is clearly important—for example, Golden Delicious substituted for Gala apples. Beginning in 1999, the BLS replaced arithmetic with geometric averaging (“geomeans”) to combine the individual item prices in 129 strata (about 60 percent of the strata in the CPI). Under some rather specific assumptions about the degree of substitution among goods and other matters, the geomeans approach will give the right answer from a cost-of-living standpoint. However, the assumptions about the extent of substitution are unlikely to hold precisely. Moreover, consumer responses to price differences may reflect something other than substitution behavior: for example, a consumer stocks up on particular items when sales occur but does not change the amount of those items purchased per month or per year. Nevertheless, most observers regard the adoption of geomeans as moving the CPI closer to a COLI.
The BLS has announced it will continue to use a Laspeyres approach—base period weights and arithmetic averaging—to combine the individual strata indexes into an overall CPI. In 2002 it will also publish an alternative index that uses a superlative index technique to combine the strata. However, a superlative index requires knowledge of consumer expenditure patterns in real time, and no country’s statistical system now produces such data. As a consequence, the superlative indexes that BLS will publish will apply to the period 2 years earlier: the index published in 2002 will measure price changes only through 2000. Recent research studies (e.g., Aizcorbe and Jackman, 1993, and Shapiro and Wilcox, 1997), making comparisons over the period of the mid-1980s to the mid-1990s, suggest that a superlative index would rise at about 0.15 percent a year less than the official CPI using fixed weights at the upper level. Of course, future patterns of inflation may differ, possibly producing a different comparison.
Substantial changes in consumer tastes pose problems for the use and interpretation of either a COGI or a COLI. The weights in a COGI have relevance because the index aims to measure the average price change for the things that people buy. If the pattern of purchases changes substantially, either because of substitution behavior or because tastes have changed, the relevance of a COGI diminishes. Since the composition of people’s spending is related to their income, age, and possibly other characteristics, changes in income distribution or demo-
graphic balance can also lead to changes in aggregate spending patterns. These sorts of factors may very well contribute to long-term changes in the composition of aggregate purchases though they are unlikely to have large effects over periods of 1 or 2 years. (Below, we outline the aggregation problems that arise in constructing a national price index across individuals and households with differing economic, demographic, and other characteristics.)
A superlative index that, in effect, averages beginning and ending weights has some commonsense appeal in that it takes both states of the world into account. However, the theoretical work that demonstrates that a superlative can provide a close approximation to a measure of the change in the reference period cost of living assumes that changes in purchase patterns stem solely from substitution behavior by households with stable tastes or preferences. To the extent that changes in tastes rather than substitution behavior causes purchase patterns to shift, a superlative index will lose some of its accuracy as a measure of the cost of maintaining the reference period level of living.
In the long run, both consumer tastes and the economic and demographic distributions of households can alter substantially. Comparisons of changes over lengthy historical periods in the price level, and perhaps even more so in the cost of living, are difficult to interpret. In the short to medium run, the issue is whether changes in consumer tastes or substitution behavior tend to be the dominant explanations of changes in household purchasing patterns. As noted above, this is essentially an empirical question, one on which the historical data can shed light, though only inferentially.
Given the time lags required to produce a superlative index, the monthly real-time CPI must instead be calculated from a set number of strata indexes aggregated with fixed weights. Assuming that, due to data constraints, this will be the case for the foreseeable future, what alternatives are available to Congress for making cost-of-living adjustments to social security and other public benefits and for indexing the tax system? Should it continue with the traditional fixed-weight index, recognizing that it is likely to modestly overstate rises in the cost of maintaining the reference period’s standard of living? Should it make an initial adjustment based on the fixed-weight index (or, perhaps, on the advance estimate of a superlative index based on historical relationships between a fixed-weight and a superlative index) and then incorporate a correction into the cost-of-living adjustments that are made 2 years later, when the superlative becomes available?
In Chapter 2 the panel discusses the conceptual pros and cons of superlative indexes and other methods of accounting for substitution behavior. We also note alternative approaches to producing a lagged superlative index and techniques for making advance estimates of that index. In Chapter 7 we make recommendations about the use of a superlative index, as well as advance estimates of a superlative, in making cost-of-living adjustments (COLAs) for social security benefits and other public transfer programs.
Aggregation Across Consumers
A single price index must somehow represent the average experience of a very heterogeneous population, whose members buy different goods, of different qualities, at different prices, in different kinds of outlets and who exhibit different substitution behavior when relative prices change. If the differences were purely idiosyncratic, so that buying patterns, shopping behavior, and prices paid for the same good did not vary systematically according to whether people are rich or poor, old or young, or by other socioeconomic characteristics, alternative ways of aggregating individuals’ varied experiences into a single index would not, for most purposes, pose issues of any great significance. But buying patterns, shopping behavior, and prices paid do vary among different groups by income, age, and possibly other characteristics. And so, during any period in which the prices for the particular kinds and qualities of goods that are especially important to one group rise significantly faster or slower than average, the change in the CPI will under- or overstate the rise in the cost of living for that group.
This heterogeneity raises several issues in the construction of the CPI, which we discuss in Chapter 8. The first of these is peculiar to a COLI and the other two are common to both a COGI and a COLI. First, from the standpoint of measuring consumer substitution behavior, different groups may be more or less inclined to switch their expenditure patterns in the face of changes in relative prices. To the extent that heterogeneity of substitution behavior is systematically related to income and demographic characteristics, the substitution effect incorporated in the overall index can vary with changes in income distribution and demographic balance. (In Chapter 2 we briefly consider the conceptual issues raised by this phenomenon, and in Chapter 8 we provide a fuller exposition.)
The second issue can be framed in the form of several questions: When is a single national index appropriate for the whole population, especially for such purposes as adjusting taxes and social security payments, and when are different indexes for different groups and geographic areas needed? If the latter, how does one collect the kind of data needed to produce subindexes that accurately reflect differences among population groups and locations?
Finally, even if it would be desirable to produce one or more subindexes, a single overall index would still be needed for many purposes—as a measure of national inflation, for example. For purposes of combining the prices of individual goods into the overall national CPI, weights are currently assigned to each good based on aggregate consumer expenditures for the item. Since the spending of a household is positively related to the level of its income, the consumption patterns and prices paid by the rich play a greater role in determining the rate of change in the overall CPI than do those of the poor. Because of their expenditure-based weights, the CPI and the corresponding indexes of virtually every country have been labeled plutocratic indexes. In an alternative democratic index, the
purchasing pattern and prices paid of each household would be given equal weight.
Aggregation and Data Collection
Various empirical studies have combined the basic CPI strata price indexes with expenditure weights reflecting the consumption patterns of a particular group—most notably, the elderly and the poor. In general, these studies have tended to show that the individual group indexes have sometimes risen faster or slower than the overall CPI, but the differences were usually small. Common observation shows, however, that within any category of expenditures (such as a stratum) high-income households buy different items, of different qualities, and often at different stores than do low-income consumers. And the probability that a consumer will purchase a new good during the early period, after which the price often falls, is almost certainly correlated positively with income. Simply reweighting strata prices at the upper level will not show whether price or cost-of-living indexes for the rich, the poor, the elderly, and other subgroups in the population sometimes move differently than the index as a whole or whether a democratic index would behave differently from a plutocratic one.
Testing these possibilities can only occur if data are collected in a way that allows the prices of individual items to be linked to the demographic and other characteristics of those who buy them. But, as explained above, the BLS collects data on price changes for individual items not from households but from retail stores and other sellers. There is no link between the prices of individual items and the economic and demographic characteristics of the consumers who bought them. As a consequence, the current collection system cannot produce the data needed to answer the questions posed above.
CHANGES IN THE QUALITY OF GOODS
Ideally, both a COGI and a COLI ought to be based on changes in the prices of “constant-quality” goods. When a consumer switches to a higher (or lower) quality good, the difference in the price paid for the two goods should be adjusted to remove that part of the difference attributable to the change in quality. If, as is usually the case, the average quality of goods that people purchase improves over time, an index appropriately corrected for quality change will rise more slowly than one measured by the change in unadjusted (nominal) prices. The most frequent criticism of the CPI in recent years, typified by the Boskin commission report, has been that it significantly underestimates the extent of quality improvement in goods and services and therefore overstates the rate of inflation. For many decades, the BLS had been aware of problems posed by goods and services whose quality changed over time and had cautiously extended its use of explicit quality adjustments. More recently, it has begun to move somewhat more aggressively in that direction.
Estimates of “quality bias” in the CPI, such as those of the Boskin commission, that combine specific case studies with subjective extensions to the universe of consumer goods, can contribute to informed discussion about the problem. But, as producers of official statistics, the BLS must walk a difficult line: It must seek to develop and apply techniques for measuring quality change, but it also has to recognize that there are substantial conceptual, statistical, and data availability problems to be solved before it can produce careful and replicable estimates that will be widely accepted.
The adjustment of observed price changes, to eliminate those that reflect changes in the quality of the goods purchased, raises conceptual and measurement issues. Even if there are no measurement problems, one would still have to decide how comprehensively the BLS should pursue the goal of quality adjustment. A frequently cited example arises from improvements in specific medical procedures that reduce mortality. Intuitively, many people think it would be inappropriate to adjust the CPI for such quality improvements and thereby reduce the benefits paid to social security recipients to reflect the estimated monetary value of additional longevity resulting from improved medical procedures (recipients should not be put in the position of living a longer life and enjoying it less). Is there a theoretical foundation for this view? Is it exceptional to this particular issue? Are there grounds for establishing limits on the use of quality adjustments? Are there general principles that can be invoked, or is this the kind of issue that must be settled on a case-by-case basis? In Chapter 2 we consider these issues in general and in Chapter 6 examine the medical care example in more detail.
Occasions for quality adjustment continually arise when field agents find that sample items no longer appear on outlet shelves. In these cases, using guidelines established by BLS commodity analysts, the field agent selects a replacement item (which may or may not be new to the market) that is as similar to the old item as possible. About 30 percent of the items being priced disappear each year. In about two-thirds of those cases the field agent can identify a comparable good, which is then treated as if it were the old item. In the other cases the agent identifies a similar but not completely comparable product—e.g., a different version of a dining room table, a lawnmower with a more powerful engine, or a different model of a computer—to price. A quality adjustment to the price of the replacement item must then be made.
Adjustment for differences between the old and the replacement items proceeds along one of two paths. Explicit quality adjustments are currently carried out using either a cost-based method or a hedonic regression technique. For instance, for three decades the BLS has estimated quality adjustments for the annual model changeover in motor vehicles using the cost-based method. The
BLS collects manufacturers’ estimates of the cost increases incurred in adding or changing observable features in the new model. On the assumption that those costs are reflected in reported motor vehicle prices, the prices are adjusted downward to reflect the quality improvements associated with the changed characteristics. This approach is applied not merely to features that are immediately evident, e.g., adding cruise control as a standard feature, but also to more subtle changes, such as the introduction of more corrosion-resistant metals on exposed surfaces. Quality adjustments for changes in the attributes of gasoline are also made with a cost-based technique.
Hedonic techniques offer an alternative method of direct quality adjustment. In this approach, statistical regressions are applied to estimate how much consumers pay for combinations of observable characteristics embodied in goods. Hedonic techniques have long been used in the CPI to make quality adjustments for clothing and rent. In the case of computers, differences in market prices have been linked to differences in speed, memory, reliability, and other performance indicators. These statistical estimates are then used to quality adjust the prices of the newer replacement goods as they are substituted for older items that disappear from the index. As would be expected, the hedonically adjusted price of computers has been falling rapidly for many years. The BLS has recently extended this approach to other goods, including TVs, microwave ovens, audio equipment, and several other types of appliances, and it is experimenting with hedonic techniques for still more products. The recent expansion in hedonic applications has thus far not had a large effect on the CPI. In Chapter 4 we discuss this phenomenon in detail and describe how it and other factors shape our assessment of the potential of hedonic methods in the CPI.
When a noncomparable replacement for an item that has disappeared cannot be explicitly adjusted by one of the methods described above—and this constitutes the large majority of cases—an implicit adjustment is made. In linking the old item price to the new item price (to calculate the price relative for the month in which the replacement takes place), a “pure” price change is imputed for the new replacement item on the basis of the average price change for similar items. Any remaining difference in price between the new and the old items is assumed to represent a quality change and is ignored (i.e., the assumed value of the quality change is adjusted away and that part of the price difference has no effect on the CPI). Subsequently, the new item is priced each month as are all the other items in the index. The large volume of items that are implicitly adjusted each year for quality change suggests the potentially high value of research directed toward developing reliable methods for widening the applicability of explicit quality adjustment techniques.
The rapid growth of research in hedonic techniques (coupled with the lack of research on alternatives) suggests that they may be the most promising approach for exploiting data on differences in the observable characteristics of similar goods to generate measures of quality change. But, in practice, their application is
fraught with difficult problems of data availability, concept, and statistical method. In Chapter 4 the panel considers in detail the proper balance between the pursuit of improved and expanded direct quality measurement and a rigorous program of selection, testing, and experimentation before implementation.
Pricing medical care embodies the most difficult quality-related problems associated with constructing a cost-of-living index, and the panel gives it special attention. The health services sector is a highly complex industry characterized by rapid advances in technology and a continuing stream of new techniques, propelled in part by substantial federal support for research. The industry also possesses a number of special attributes that make the quantity and quality of its output difficult to define. Complications that BLS must deal with include the high variation in prices paid by consumers for equivalent services, the definition of the medical “good” being purchased (e.g., is it a procedure or treatment or the medical inputs?), the involvement of insurers and government in transactions, and the pricing of risk.
Some important improvements in accounting for changes in the quality of medical care have recently been introduced by redefining the units of service for which prices are collected in the CPI. For example, to price hospital services, the BLS has begun to collect prices for the treatment of particular diagnoses or illnesses in place of the earlier practice of pricing inputs, such as days in the hospital or operating room charges. This has led to substantial improvement in the ability of the index to capture advances in medical technology that reduce the cost of treating a given illness. Another major improvement has been in the BLS treatment of generic drugs, in effect counting the difference between their price and those of the same-molecule branded drugs as a price reduction.
These improvements, however, do not directly deal with advances in the quality of a given treatment or procedure, that is, changes in ultimate outcomes— lower mortality and morbidity, fewer undesirable side effects, less pain or trauma, a better quality of life, and so forth. (We touched on the conceptual aspects of this issue above.) To pursue the cost-of-living concept in the pricing of medical care to its logical conclusion, one would need to put a monetary value on how the outcomes of specific medical procedures affect consumer well-being. This obviously poses enormous conceptual and measurement issues. In Chapter 6 the panel considers these conceptual and measurement issues and makes a number of recommendations about the treatment of medical care in the CPI.
In the marketplace, there is rarely a sharp dividing line separating a new good from an existing one whose quality has been improved. As described above, monthly BLS price collection procedures continually lead to the pricing of re-
placements for discontinued items. These replacements, chosen from the same CPI item category, exhibit only incremental quality changes compared to the old products. Goods also appear that are not replacements for any of the items being priced in the current sample but that can be assigned to existing CPI categories. These items may enter the index when retail outlets (and the goods they sell) are rotated in the BLS sample every few years. In addition, though, new goods and services appear that are different enough from existing ones that there is no place for them in any of the CPI categories. Hence, without an explicit decision to add the new item to the list of categories of goods, their impact on general price trends can go unmeasured: VCRs and cell phones are examples of such items that did not enter the index for many years after they appeared on the market.
There are two ways in which the CPI, when used as a measure of changes in living costs, might be biased by the appearance of new goods. First, many of those who advocate a cost-of-living approach to index construction argue that the consumption welfare effects that accompany the appearance of a new good are missed. Under traditional procedures, new goods, both those that enter through item rotation or after item reclassification, are linked into the CPI in such a way that their introduction, in itself, has no effect on the level of the index. But, for example, there were some consumers who found wireless phones so attractive that, rather than do without, they would have paid a higher price than that prevailing when the phones first entered the index (or even at the time of market introduction). An increase in the standard of living of these consumers was made possible by the introduction of the good, but such effects on the cost of living are not captured by simply linking in the new good without making a specific adjustment for the improvement.
The magnitude of the improvement could, in theory, be estimated using detailed market data on the prices of new goods and the quantities sold to infer how much consumers would have paid for the new good rather than doing without. This estimate would then be used to calculate the relative welfare gain associated with the introduction and subsequent consumption of the good. That gain, in turn, would be incorporated into the index as a price decrease. The conceptual and measurement issues at stake here are addressed in Chapter 5. On a measurement basis, is it possible with available econometric tools to estimate the welfare gain with sufficient accuracy and transparency to warrant its use in the way that would be required? Conceptually, even if reliable measurement should become feasible, should the new good welfare effect be treated as if it were just a price decrease and entered into the consumer price index as such? In Chapter 5 the panel examines both the feasibility and advisability, in constructing a COLI, of employing econometric techniques to estimate and incorporate into the index the “virtual” price reduction that accompanies a new good’s appearance in the market.
The second problem posed by the introduction of new goods revolves around the timing of their incorporation into the CPI. New goods are often introduced in
the market at a high price, and they have a low volume of sales. Then for some time the price will tend to fall and the market will expand rapidly, until the new good becomes a mature, established product. For some new goods, however, producers follow a different strategy, introducing the new good at a low price in order to promote high initial sales and make the good more widely known to consumers. In this case, as the market expands, the price rises.
If a new good with an initially falling price and rising sales volume—e.g., cell phones—is incorporated into the index only after a long delay, the period of falling prices will be missed and the overall price index will be biased upward. If, in contrast, the new good is incorporated into a fixed-weight index not long after its introduction, the index will reflect most of the decline that occurs early in its price cycle. Early sales are likely to be modest, and if the weights in the index remain fixed for some time, the declining prices will carry only a very small weight, and again the overall index will be biased upward, although to a smaller degree. If the new good is one whose price initially rises rapidly, the opposite results occur: whether the new good is incorporated late or early, the index will be biased downward. With either rising or falling prices, the faster the price change and the more rapid the sales growth, the larger the bias. Working on the assumption that most new goods experience a period of large price declines, the Boskin commission and other observers have attributed an upward bias to the CPI stemming principally from late introduction of many new goods.
The BLS is making changes that will reduce the magnitude of the problem: much more rapid updating of the index weighting system (every 2 years instead of every 10 years or longer) and a faster turnover of the stores from which it collects prices. But the issue of how best to deal with the introduction of new goods will remain an important problem, and handling it will require tradeoffs among competing objectives. Very early introduction of new goods raises the danger of incorporating “duds” into the index—such as Betamax VCRs or 10-inch video disks. And speeding up the rotation of retail stores (see below) into the sample is quite expensive. Chapter 5 examines and makes recommendations about the procedures and criteria that BLS should use for linking the prices of new goods into the index in a timely and nondisruptive way.
To the extent that new goods are disproportionately purchased by the affluent in the early stages of a product’s cycle, distributional consequences will arise as these goods are brought into the index. If the relative prices of new goods fall, the growth of the overall index will, on this account at least, tend to understate the inflation faced by low- and middle-income consumers.
When purchasing a good at a particular store, consumers are buying not only the good itself but also a package that includes the quality of the shopping experience associated with the store—the services provided, its locational conve-
nience (or inconvenience), the variety of products available, its return policy, and so forth.
The BLS gradually rotates the sample of retail outlets from which it collects prices, and over time the new samples capture the changing mix of outlets patronized by the buying public. Under current BLS procedures, when new stores enter the sample, all of the difference between an item’s price at the old outlet and its price at the new outlet is implicitly assumed to reflect differences in the “quality” of the shopping experience; none of it shows up as a pure price change. This practice can produce a bias if price variation across outlets allows consumers, by altering their shopping behavior, to reduce their consumption costs, adjusted for the quality of the shopping experience, in a way not detected by the CPI.
The clearest evidence that consumers are reducing costs is indicated by the increase, for a number of years now, in the market share of high-volume, low-price retail outlets. The prices paid at these outlets are often substantially lower than those in conventional stores. Under current assumptions, lower prices are being fully offset by a lower quality shopping experience—as represented by “goods,” such as convenience or a store’s return and exchange policy, that are omitted from the CPI. The fact that the market share of the low-price discounters has been steadily growing, however, implies that even after “quality adjustment” the prices at those stores are lower than elsewhere. As new outlets open, consumers in the area gradually change their shopping behavior and take advantage of the lower quality-adjusted prices. At the same time, a minority of consumers who would have preferred to continue shopping at traditional stores find them to be driven out of business by the new outlets, and those consumers suffer an increase in their cost of living.
From the standpoint of a cost-of-living index, the current procedure for handling sample rotation among outlets misses some of the decline in living costs associated with this ongoing shift in consumer purchasing patterns. The few empirical studies that have been done, however, suggest the resulting effect on the overall CPI is relatively small. Identifying and quantifying “quality” differences in the shopping experience offered by different types of outlets, in order to make a proper adjustment for what is happening, poses measurement difficulties for which satisfactory answers are not currently available. In Chapter 5 the panel makes recommendations to the BLS about what to do in the current absence of methods for making outlet-quality adjustments and suggests what priority it should give to research and development efforts in this area.
STOCKS AND FLOWS
Many longer-lived goods, such as owner-occupied housing, automobiles, and appliances are durable capital goods that gradually yield a flow of consumer services over a period of years. Even some “nondurable” goods (e.g., men’s suits) often provide services for some years. While there is no consensus among experts
and national statistical agencies, a powerful argument can be made that the CPI in any period should measure the price of consumption in that period. In that case the index ought to include not the price that consumers pay for a durable good at the time of purchase, but the estimated cost of the services rendered during the subsequent periods in which the good is owned. In the case of owner-occupied housing, the BLS has, since 1983, estimated (“imputed”) a monthly price of the services rendered by that housing (i.e., shelter services). In all other cases the price of a consumer capital good is entered in the month when it is purchased.
There are two ways of estimating the price of the flow of services from consumer capital goods. The first is to estimate the “user” cost of the service (i.e., the cost to a consumer of buying a good at the beginning of a period, using its services during the period, and selling it at the end). Among many other difficulties, the user cost concept would deduct a capital gain (or add a loss) realized by the owner in the transaction. Housing prices can be volatile, and this method of estimating the cost of shelter services can sometimes produce the paradoxical result that just as housing prices are rising rapidly the user cost estimate will show a decrease in the shelter component of the CPI. An alternative method of estimating the cost of housing services, and the one currently used by the BLS, is to try to find a sample of rental housing that is equivalent in quality to owner-occupied housing and use the change in rents within that sample as a measure of changes in the cost of owner-occupied shelter services. This method also poses measurement difficulties of various kinds, including the difficulty of finding, in each geographical area for which the BLS collects prices, a representative sample of rental housing that is truly equivalent to owner-occupied housing.
While a strong conceptual case can be made for incorporating into any index of the price of consumption an estimate of changes in the monthly cost of the services provided by owner-occupied housing and other long-lived capital goods (see Chapter 2), there are some countervailing considerations. For potential low-income home buyers, who are cash constrained and at the margin of acceptability for mortgage loans, sharp changes in housing prices can have a greater impact on their cost of living than would be reflected in cost estimates of the flow of services. And it has been argued that, since the CPI is used as a guide for monetary policy, it should reflect volatile changes in consumer asset prices, as is the practice in a number of other countries (Goodhart, 2001). (We point out in Chapter 7, however, that governments and central banks have plentiful staff resources to consider the effect of changes in assets prices as they relate to monetary policy, even if they are not included in the official index of consumer prices.)
The practices of other countries with respect to housing vary widely. Because of the conceptual and measurement difficulties, many countries simply exclude owner-occupied housing from their consumer price index altogether. A few use the net acquisition price of owner-occupied housing; a few others, in addition to the United States, use equivalent rent; some use a cash flow approach—the sum of down payments, mortgage principal, and interest payments,
or some subset of these items; and still others use the sum of depreciation at replacement cost value and mortgage interest payments as a (very) rough measure of user cost. No other CPIs extend the “flow of service” concept to the pricing of other consumer capital goods, although that course has sometimes been proposed to the BLS for the pricing of automobiles (Boskin et al., 1996:recommendation 8).
This brief sketch gives an indication of the wide range of conceptual and measurement issues that arise in determining how to measure changes in the prices of consumer capital goods and their services. However, given both the importance of covering the many other subjects the panel was asked to address and the constraints of time, we did not examine these issues in any depth or formulate recommendations on them.
INDEX PURPOSE AND DESIGN
It is clear that there are many difficult issues to resolve in designing a consumer price index or indexes. How some of the decisions should be made may be dictated by the purpose for which the index is to be used—the index designer needs to keep the index user in mind. Although no restrictions need be placed on the number or variety of research and experimental indexes, a desire to avoid public confusion may constrain the number of separate official indexes that are published. More importantly, there are inherent limits on the extent to which it is possible to match the design of an official index to a particular purpose, limits that are often dictated by what can reliably be measured. As a consequence, public policy makers and private users of indexes need to be aware of the extent to which a particular price index does not measure exactly what they want measured. In fact, considering its range of applications, it is probably rare when the CPI does measure exactly what is needed. The CPI is currently used in many ways, including:
as a compensation measure to calculate how much is needed to reimburse recipients of social security and other public transfer payments for changes in the cost of living, and for formal or informal use in wage setting;
for inflation indexation in private contracts;
as a measure with which to index the income tax system to keep it inflation neutral;
as a measure of inflation in inflation-indexed U.S. Treasury bonds;
as an output deflator for separating changes in GDP and its components into changes in prices and changes in real output; and
as an inflation yardstick for the Federal Reserve Board and other macroeconomic policy makers.
Throughout the report, we explain how alternative choices in index design could affect use of the index for each of these purposes. Where relevant, we spell out the public policy consequences of using alternative index designs for making cost-of-living adjustments in public transfer programs, indexing the tax system, and for other purposes. And while we make no recommendations on the subject, we explore the public policy implications of using a wage instead of a price index for escalating social security and other benefits.
The CPI is constructed from several sample-based sources: the Consumer Expenditure Survey (CEX), the Point of Purchase Survey, the Commodities and Services Survey, and the CPI Housing Survey. There are two distinct approaches that could be taken to change the data collection apparatus: the first would be to improve each survey component, assuming that the basic structure will remain in place; the second would involve redesigning, from the ground up, the entire data collection apparatus. The panel considered options under both approaches.
The panel addresses questions about both the accuracy and precision of the CEX, which is the primary tool for establishing the CPI upper-level weights (at the basic, 218-item level). The panel’s foremost concern is with the extent of bias in the CEX which, in turn, affects the accuracy of CPI expenditure category weights. In this context, it is worth evaluating the pros and cons of using alternative data sources—such as those used to produce per-capita personal consumption expenditures for the national accounts—for deriving the national CPI upper-level weights.
Assuming that the CEX is the appropriate source for generating CPI weights, there is the question of optimal survey sample size. The report addresses these questions and provides some calculations that indicate the relationship between sample size and the precision (variance) of derived item strata weights. Of course, precision requirements set for the national index will yield very different answers than similar ones for component indexes or if population subindexes are desired.
In addition to questions about sample size and accuracy, there are a number of issues that involve assessing the information content of questionnaires and the general structure of the CEX. There are also questions about how the mode of data collection might be modified to take advantage of new computer-based data collection methods, whether all expenditures for all item categories should be collected from all households surveyed (or just some from each), and what processing system is required for the CEX in order to expedite development of a superlative index. Answers to all these questions hinge on the types of indexes that BLS will be called on to produce.
A second major survey input to the CPI is the Point of Purchase Survey (POPS), which is used to determine which outlets BLS data collectors visit to record price changes of index items. The POPS produces outlet-specific expendi-
ture information for item categories so that a sample of those outlets can be selected with a probability proportional to consumer use. The POPS is needed because the CEX does not ask consumers where they purchased goods. Given that there is some functional overlap between the CEX and POPS, the panel considers, among other things, the possibility of merging or better coordinating these two surveys. The report also reviews the CPI’s Commodities and Services Survey, a longitudinal survey that tracks changes in price quotes for most CPI sampled consumer items over time.
Since most options for improving CPI support data are expensive, particularly those involving the household surveys, and because there is methodological inflexibility under the current system, it is also worth considering entirely new data production alternatives. Therefore, in Chapter 9 the panel considers (1) the tradeoffs associated with changing to PCE-based expenditure weighting; (2) the possibility of combining POPS and CEX into an integrated survey that contains expenditure and outlet-use data at detailed product levels, along with household demographic information needed for subgroup indexes; and (3) what might be gained from moving toward scanner-based collection systems, which could be used to improve the existing surveys or as a component of an alternative.