THE WORK OF THE NATIONAL ACADEMY OF SCIENCES PANEL ON COST OF LIVING INDEXES

Charles L. Schultze, Brookings Institution
Chair, National Academy of Science Panel on Conceptual, Measurement and Other Statistical Issues in Developing Cost of Living Indexes

A year ago the BLS asked NAS to establish a Panel: "To investigate the conceptual, measurement and other statistical issues in the development of cost-of-living indexes." The Panel members were appointed early this year, and we have had two meetings since then. A draft report is due in the fall of 2000. The Panel was not intended to be an alternative Boskin Commission -- we are not charged with making estimates of the possible bias in the CPI. Nor are we primarily concerned with evaluating the detailed procedures now used by BLS to produce the CPI, except as they flow from choices about fundamental concepts and measurement issues.

The Panel includes members from a number of professional disciplines. There are two statisticians, Albert Madansky (Chicago) and Kurt Wolter (National Opinion Research Center); one cognitive psychologist, Norbert Schwartz(Michigan); a sociologist, Christopher Jencks (Harvard); and seven economists in addition to myself: Ernst Berndt (MIT), Angus Deaton (Princeton), Claudia Goldin (Harvard), Zvi Griliches (Harvard), Van Doorn Ooms (Committee for Economic Development), Robert Pollak (Washington University), and Richard Schmalensee (MIT). [Note: W. Erwin Diewert (British Colombia) has been named to replace Zvi Griliches.]

I am reasonably sure that an important part of our recommendations will take the form of suggesting subjects, priorities and timing for future BLS research. But it's much too early for us to have reached the point of formulating recommendations. So, unlike the other presenters at this session, I am not here to suggest a research agenda. However, I thought it might be useful to take a few minutes to outline the major questions the Panel intends to examine.

We have tentatively identified five broad sets of issues on which we plan to make recommendations, although we are far from reaching any conclusions yet on any of them.

First, in recent years the United States, unlike many European countries, has explicitly accepted the cost-of-living concept as a basic theoretical construct that should, to the extent feasible, be used as a guide in pointing the way towards making improvements to the CPI. Is this an appropriate choice? If not, what is the alternative?

Second, granted that the cost of living concept is appropriate, how comprehensive should be the definition of the "standard of living" or "consumer well-being" underlying the cost-of-living index? What is the universe of well-being that is to be held constant? Decisions about this matter involve not only conceptual issues but questions of feasibility, objectivity, and public acceptance.

Third, what is the appropriate way to aggregate across consumers with heterogenous tastes to arrive at a single index? This question immediately raises the issue of plutocratic vs. democratic indexes. And, whatever the aggregation scheme for the overall index, what ought to be the criteria for determining whether or not to publish subgroup indexes?

Fourth, the Panel is explicitly charged by NAS to examine the relationship between alternative designs of an index and its appropriateness for various uses. For example, what kind of an index is most appropriate as a measure of inflation; what index (or indexes) might a central banker want; what are the implications of alternative conceptual choices for producing measures used to index social security benefits, or the tax code, or inflation indexed bonds?

Fifth, we were also asked to consider the problem of balancing competing objectives in index construction. There is, for example, a tradeoff among the objectives of conceptual rigor, measurement feasibility, public credibility, and budget costs. (Budget costs should not be underrated as a criterion: some components that have been suggested for inclusion in a COLI would be exceedingly costly to measure with reasonable accuracy across a broad spectrum of goods and services, assuming they could be measured at all). The Panel will examine the costs and benefits of periodically revising already published indexes.

Finally, we were explicitly charged with spelling out the kind of research program that is implied in our recommendations, looking both to reasonable objectives for short-term improvement and long-run feasibility studies.

While the Panel is not nearly far enough along for me to offer you even a set of tentative conclusions, I can provide somewhat more detail about several of the questions that we have agreed to address. Logically, the first issue that arises concerns the appropriate boundaries, or to use Jack Triplett's term, the domain of the standard of living. What is the universe of consumer well-being that should underpin the index? In discussing its longer run recommendations, the Boskin Commission suggested a very comprehensive definition of the standard of living, including, for example the positive or negative effects on consumer well-being -- and therefore on the COLI -- of changes in the surrounding economic, social, and environmental climate. Other components of the standard of living that have been proposed by the Boskin Commission and broadly in the literature include the creation and destruction of consumer surplus from the introduction and disappearance of new goods, from the increased variety of existing goods, and from the growth of large volume low-price outlets such as Walmarts and Home Depot. The Boskin Commission recognized that many of the economic and statistical tools and the data available to measure these kinds of effects do not now exist, but urged a long-term research effort so that such tools might ultimately be developed.

Between a policy of continuing a traditional "fixed basket" price index with modest changes and one of attempting the modifications necessary to produce a cost of living index that captures the most comprehensive definition of "the standard of living", there is a wide range of intermediate possibilities. We might think of this in terms of what is the appropriate Pollak sub-index of the universe of consumer well-being that ought to be produced in the COLI.

The most comprehensive definition of what ought to be included would almost surely be practically and, in the broad sense of the term, politically infeasible. On the other hand we shouldn't judge what might realistically be feasible over the longer term on the basis of the current state of economic knowledge, and data collection technology. BLS doesn't have to decide many of these conceptual issues in the abstract. It is not even remotely in the position in the immediately foreseeable future of determining whether many of the suggested components of the cost of living can be reliably measured, even if it were desirable to do so. What is actually at stake on many of these issues is an informed judgment about the likelihood of solving the feasibility problem, as guides to establishing the scope, priorities, and timing of a long run research program for the BLS and suggesting an agenda for academic research.

It is possible, nevertheless, to make a few comments about some of the substantive issues involved in establishing the universe or the boundaries of what goes into the COLI. Start with the broadest or most abstract issue - conceptually, should the index incorporate the effects of broad environmental changes on the cost of living: crime, pollution, congestion, etc? One possible approach would be to adopt a definition of COLI based on a Pollak sub-index that would measure the change in expenditures (or income) required to prevent changes in market prices from altering the standard of living. However, there is the view that the COLI should not attempt to measure changes in consumer utility levels that do not operate through the purchase of goods and services. After all, the COLI is still meant as a price index. Loopholes and wiggle-room would still be needed under this definition. If it were decided to include some goods furnished by government, then taxes would have to be treated as prices. There would also have to be some limited exceptions for fictitious prices, e.g. owner occupied equivalent rent. Should this definition be adopted, the effects of broad changes in environmental conditions would not be included in the index.

If the definition of the COLI is not narrowed along the lines I suggested, then the operational question for the BLS is where, in the long list of potential list of issues that need research, should the BLS place the environmental issue? It's for the Panel to decide, but I suspect it would be pretty far down the list of our priorities.

A second set of conceptual issues involves the question of publicly provided goods and the taxes to pay for them. Should the index be expanded in the direction of Nordhaus's augmented COLI? Governments levy earmarked taxes to pay for highways, aviation facilities and the like. They are shifted forward, and show up as increases in prices, but no account is now taken of the benefits of the goods provided. Nordhaus would exclude these taxes, presumably on the assumption that, roughly, the marginal benefits of the goods provided exceed the marginal tax costs. The same line of reasoning applies to regulatory costs imposed by government for pollution abatement, health, and safety. These costs get passed on into prices, but no account is taken of the effect of the benefits on raising the standard of living.

These suggestions raise several major issues. Studies of environmental regulations show that for many of them, marginal benefits are substantially less than marginal costs (even though aggregate benefits often exceed aggregate costs). Similarly, a recognition of the institutions and political procedures that are used to determine various governmental outlays paid for by earmarked taxes leaves a strong doubt that we can assume that the prices paid are a good measure of marginal benefits. We cannot, therefore, simply subtract the estimate of regulatory costs or earmarked taxes from the index. We would presumably have to go outside the price system and make synthetic estimates of benefits with which to adjust the COLI. This is analogous to the case of medical care where, for various reasons, prices usually can't serve as measures of benefits provided, and if we want to make such adjustments we have to impute the value of increased longevity, reduced morbidity, and decreased trauma. The Panel will have to consider and make judgments about the trade-offs involved in extending the domain of the index to at least some publicly provided goods.

The next class of issues involves the question of whether to try to measure the consumer surpluses created and destroyed by the introduction and disappearance of goods, of different varieties of existing goods a la Hausman, and by the changing composition of retail outlets. Whatever the theoretical pros and cons of defining the COLI to include the effect of these consumer surpluses, the fact is that at the moment neither the BLS nor anyone else is in the position of being able to estimate with any confidence the magnitude of the net changes in consumer surplus across a broad range of consumer expenditures. For example there is no agreement on the appropriate econometric specifications for measuring demand curves and virtual prices near the origin -- look at Bresnahan's comments on Hausman. And extending such estimates across the multiplicity of product categories in an economy characterized by imperfect competition and market strategies driven by product differentiation boggles the mind.

On the other hand the Panel has to make a judgment about whether a 10 to 20 year continued research program might bring the techniques and the data more nearly to what is needed. Speculatively, the Nordhaus sampling strategy might conceivably be used cut down the size of the problem. Moreover, the choice need not be made between incorporating such measurements into the official index or nothing. Conceivably, if research dealing with some aspects of this problem began to pay off, the BLS could periodically publish experimental indexes showing what the consequences of incorporating such measures would be. In turn, any gains from such an approach would have to be weighed against the possible cost of sowing confusion among politicians and the public.

I said earlier that we will consider the problem of aggregating across consumers with heterogeneous tastes and the related question of whether the BLS should regularly publish indexes for sub-groups of consumers. Let me just cover one point here. Empirical studies that have compared the path of sub-indexes with the overall CPI have typically found only modest divergence. Essentially, however, these have involved simply re-weighting a common set of component indexes. But some members of the Panel have argued that the central issue is not one of differing weights and common prices, but the fact that the poor and the rich, the young and the old, often buy different qualities or brands of the same goods, face different prices, and tend to shop in different outlets. We will explore this issue, including the question of the most efficient way to acquire the kind of data needed to find out how important these differences are.

A fourth major set of issues I identified was how the appropriate design of an index is related to the uses for which it is intended. Looked at another way, what are the implications of making various conceptual choices about index design for each of the major uses of an index? For example, what kind of an index would a central banker want? Would the monetary authority want to distinguish between the index it sought to stabilize, and a different index that would provide the most informative guide about the current state of inflationary (or deflationary) pressures in the economy? Which index would be most appropriate for escalating social security benefits, and would that be the same as the measure used for indexing the tax code or inflation-indexed government bonds?

Again, for public policy purposes, it might be desirable to have a comprehensive index of medical service prices in which imputed non-market values for reductions in mortality, morbidity, and trauma were used to estimate quality change. But would it then be appropriate to used such imputed values to adjust the medical component of the index used for social security COLA's? Do we really want to reduce the cash benefits of the elderly to offset the improvements in their longevity coming from innovations in medical care?

It is not within the charge of our Panel to make recommendations to the Congress or to anyone else about what kind of an index they ought to use for various purposes. But we are explicitly charged to articulate the implications of alternative conceptual choices about index design on its uses for those purposes.

I have not covered all of the questions with which we shall be dealing. In particular I have not discussed the issue of quality change and the difficult tradeoff between the objective of quantifying such changes in the index and retaining a high degree of objectivity and replicabilty in the "official" (i.e., non-experimental) measures that BLS publishes. But this issue will obviously constitute an important aspect of our work.