Attachment I: Katharine G. Abraham, "Statistics in the Spotlight: Improving the Consumer Price Index: Statement."
Katharine G. Abraham, Bureau of Labor Statistics
Bureau of Labor Statistics, Room 4040, 2 Massachusetts Ave., NE, Washington, DC 20212
Paper presented at meeting of the American Statistical Association, Chicago, IL, Aug. 6, 1996
Key Words: CPI, Alternative indexes, CPI revision
Given the importance of the Consumer Price Index (CPI), both as an economic indicator that provides timely information on the prices paid by consumers and as a measure used extensively for indexation, not only in a number of large and visible federal programs but also in many private contracts, it's not surprising that measurement issues pertaining to the CPI have garnered substantial attention over the years.
I probably remember more clearly than most of you the specific events that precipitated the recent intensification of interest in the CPI. Back in the early winter of 1995, Federal Reserve Board Chairman Alan Greenspan testified before the Congress that he thought the CPI substantially overstated the rate of growth in the cost of living. His testimony generated a considerable amount of discussion. Soon afterwards, Speaker of the House Newt Gingrich, at a town meeting in Kennesaw, Georgia, was asked about the CPI and responded by saying, "We have a handful of bureaucrats who, all professional economists agree, have an error in their calculations. If they canít get it right in the next 30 days or so, we zero them out, we transfer the responsibility to either the Federal Reserve or the Treasury and tell them to get it right."
I heard about this the next afternoon when I got a call at home from John Berry, a reporter for the Washington Post, who read this comment to me and wanted to know if I had any response that I'd like to make. I said to him then the same thing that I would say to you today. If there were problems with the CPI that Bureau of Labor Statistics (BLS) staff knew about and knew how to fix but were just refusing to address, Iíd agree with the Speaker: he should zero us out. That is not, however, an accurate characterization of the BLS performance. Indeed, as other speakers have indicated in their comments--and I would like to express my appreciation for their kind words--the staff of the BLS have been at the forefront of trying to identify problems with the way that the CPI is put together, figuring out how to fix those problems, and making improvements in the index.
What I'd like to do in my time this afternoon is to talk about some of the things that the BLS has done recently to improve the CPI, about some of the things that we have planned for the near future, and about some of the things that we'd like to do if we could identify the necessary resources and/or could figure out how to employ them. I'm not going to talk about biases in the CPI, other than to say that I'm considerably more agnostic than the other speakers in my assessment of the overall bias, if any, in the index. There are some things related to the formulas used to construct the CPI on which almost everyone agrees. Most importantly, as an index based upon a fixed market basket, the CPI does not allow for substitution in response to relative price changes and thus has a slight tendency to overstate the growth in the cost of living. There is less basis for agreement around the issues of how well we adjust for changes in the quality of goods and services, how we deal with new goods, and how we treat changes in the relative importance of different kinds of shopping outlets. At this point, there is a great deal that we just donít know about any possible upward or downward biases associated with these things.
Let me turn, then, to talking about the Bureauís continuing efforts to improve the CPI. I'm going to talk about three things: first, some very recent changes made to correct the so-called "formula bias" problem; second, our production of a set of alternative measures that answer different questions than does the CPI; and third, some things that we are doing or would like to do in the areas of quality adjustment, the treatment of new goods and changes in outlet mix.
Let me start with the so-called "formula bias" problem that was in the news this spring, a problem that grew out of the limitations of the data that we have available for use in putting the CPI together. The CPI is designed as a measure of the cost of purchasing a fixed market basket of goods and services. The market basket concept refers to the quantities of goods and services purchased, but the data we have available from our household surveys give us information on the amounts of money consumers spend on different sorts of items at particular stores. After this information has been compiled, our field economists visit stores to collect prices for specific items within each item category. Our procedure for constructing quantity weights for the items whose prices weíre tracking used to be, first, to project the initial price collected for each item backwards using information on price trends for similar items and then to divide the appropriate expenditure figure by this backwards-projected price to obtain a base period quantity weight for the item. This may sound pretty straightforward. The problem is that this procedure led us systematically to overweight items that were on sale as of the point in time when we first priced them--expenditure divided by a low price gives you a big quantity weight. The prices of sale items are apt to rise in subsequent months, however, and our procedures thus were imparting an upward bias to the index.
We only began fully to appreciate the existence and nature of this problem with the index during the course of 1994. In January of 1995, we introduced changes to deal with the problem for food-at-home items, and also made some related changes in the way we were putting the housing component of the index together. This summer, we are making further changes that we believe fully correct the problem. Going forward, then, the "formula bias" problem should be a non-problem.
There are, of course, other outstanding issues related to the formulas used to construct the CPI. As David Wilcox emphasized in his remarks, there are a variety of questions that you might use a consumer price measure to answer. The CPI tracks the price of a fixed market basket of goods and services, but, for many purposes, a measure that allowed for substitution among items as their relative prices changed, and thereby more closely approximated a true cost-of-living index, would be more appropriate.
The Bureau has done a fair amount of work oriented towards producing alternative indexes that answer different questions than the official CPI. We are in the process of producing an experimental measure that, within the most detailed cells in the index, uses geometric mean aggregation rather than Laspeyres aggregation. This measure may be more appropriate for tracking the cost of living than the CPI if you believe that it is a more reasonable approximation to assume that consumersí preferences exhibit an elasticity of substitution of one between items within item categories rather than an elasticity of substitution of zero. We've also produced experimental superlative measures of the sort originally proposed by Erwin Diewert that take substitution across item categories into account.
There are some issues related to these alternative measures that need to be considered. The key question about using the geometric mean formula for within-cell aggregation is whether assuming an elasticity of substitution of one across the board really is a more reasonable approximation than assuming an elasticity of substitution of zero. Evidence on this issue will be hard to come by.
The superlative measures are theoretically elegant, but are likely to be more difficult than the CPI for the general public to understand. From an operational perspective, production of the superlative measures requires expenditure share data that are available only with a lag. Our experimental superlative measures currently are not available until the fall of the year following the year to which they refer. It would be difficult to shorten that production cycle by very much even for an experimental index, and if we were to produce a superlative index subject to the same sort of review as the official CPI the necessary lag might well increase. For certain purposes, it is important to have a measure that comes out promptly.
There is also an issue about the precision of our experimental superlative measures. The weights for the superlative measures are constructed using data from the Consumer Expenditure Survey (CEX). For the official CPI, we use three years of CEX data to construct weights that our statisticians have deemed to be of adequate precision. For the superlative measure, only two years of expenditure data are used, because the superlative measure is based on the average of the expenditure shares for a base year and an ending year. To produce superlative measures that were of comparable precision to the official CPI would require a Consumer Expenditure Survey that was about 50 percent larger than we now have--and that would cost money.
On the general topic of alternative measures, David Wilcox alluded in his remarks to interest in the growth in the cost of living for the elderly. We have for some time now produced an experimental CPI for the elderly, which we construct by reweighting price change data that we already have using information based on the consumption pattern of the elderly. This method has shortcomings, but doing a better job would require selecting a separate sample of outlets and items to reflect where elderly consumers shop and what they buy, and thus would be quite an expensive proposition.
Turning to a third topic, I would like to talk about some things we have been doing or would like to do that relate to our treatment of quality change, new goods and different kinds of outlets in the index. With respect to the treatment of quality change, the obvious strategy is to try to make more use than we have in the past of hedonic adjustments or other explicit adjustments for changes in the features of the items that we're pricing. How much of this we do is mainly, though not exclusively, a resource issue. Making explicit adjustments for changes in item features requires that we collect information not only on item prices but also on item characteristics. This would not have to be done every month, but we would need substantially more information than we now collect to be able to look systematically, item category by item category, at the relationship between price and item characteristics.
Something that we have planned for implementation as part of the ongoing CPI Revision is the introduction of a new way of updating the CPI outlet and item samples. The current procedure is to update those samples each year for 20 percent of the areas in which we collect prices. By moving to a telephone survey to compile the underlying sampling frames, we'll be able to change that rotation pattern. Instead of bringing in new outlets and items geographic area by geographic area, we'll be able to bring in new outlets and items for whole item categories in all geographic areas at once. If there are categories of items for which we know that there has been a lot of change in what people are purchasing or where they are shopping, we'll be able to bring in new samples for those item categories on a more frequent basis.
Quite appropriately, there has been a great deal of attention devoted to the way that medical care is treated in the CPI. Weíre in the process of making some changes there as well. Under the procedures currently in place for constructing the hospital components of the CPI, we sample and collect prices for very specific items when we visit a hospital. We might, for example, end up tracking the price of a unit of blood. The problems with this approach have become clear to all of us. Hospital care really isnít sold specific item by specific item. In January of 1997--that is, this next January--we will be shifting over to an approach to tracking hospital care prices that involves visiting a hospital, picking a patient bill, identifying the key services covered by that bill, and then tracking the cost of providing that bundle of services. This is not, of course, a solution to all of the problems we have with tracking medical care prices, but looking at whole treatment bundles puts us in a better position to begin thinking about how to accommodate changes in treatment protocols in constructing the index.
There are a number of important outstanding issues that I would have to say we donít have good ways to handle. From an operational point of view, for example, we simply donít know how to go about comparing the prices of different items that may satisfy similar needs or even the prices of the same item sold at different types of outlets. Similarly, in an operational context, we donít have any good way to deal with the value consumers may attach to increases or decreases in the variety of items available for sale. We're working on some of these things, but I am not optimistic about our ever arriving at implementable solutions to all of the concerns that have been raised regarding the CPI.
As I've indicated, some of the things we could and would like to be able to do would require additional resources. Money is always tight, and it's even tighter today than in times past. In addition to seeking extra resources to do some of the things I've talked about, we also need to be looking at whether we're using the money we already have in the most efficient possible way. Changing how we put together the area sample for the CPI, for example, might allow us to reduce our costs, and weíve begun to look at that. When we update the sample of geographic areas in which weíre collecting prices--something that we do roughly every ten years--it might be possible to have more overlap between the old and the new areas. The largest 30 or so metropolitan areas appear in the sample with certainty, and our area selection procedures already are designed to give some preference to smaller areas from the prior area sample. Most of the smaller geographic areas, however, are replaced during our regular Revisions. It's very expensive to go into a totally new area, hire staff, and begin collecting prices. Having more overlap in the geographic areas across area samples thus could save some money.
It also may not be necessary to collect prices for all item categories in all areas. The CPI sample of price quotations currently is structured so that we have a set of geographic areas and a set of item categories. With the exception of certain special cases like postage and used cars, we collect prices for all of the item categories in all of the geographic areas. The prices of items in certain categories, however, may be set in national markets, so that filling in the whole area/item-category matrix isnít necessary.
Moving away from our current approach to sample design would carry some risks. Further increasing the overlap between old and new geographic area samples, for example, might well make it more likely that we would end up with an area sample that was not truly representative of current reality. Similarly, selecting and pricing items nationally rather than locally, even if only in certain item categories, might increase the risk of not representing in our market basket items that account for a significant part of consumersí purchases or of not accurately reflecting price trends in individual areas. These risks are real, but I nonetheless believe that we ought to be looking at and evaluating the sorts of possible changes in our sampling strategy that Iíve mentioned.
Let me conclude by saying that, as the BLS moves forward, we can use all the help that we can get with continuing to improve the CPI. We are very eager to have ideas from any of you regarding how we could be doing a better job, and Iíve gotten some good ideas from the other panelists today. We are in the process of constructing research data bases suitable for addressing a wide range of price measurement issues and Iíd invite any of you to talk with us if you have a project for which those data bases might be suitable and that might contribute to an improved understanding of price change in our economy.