Briefing on the Consumer Price Index

---

December 3, 1996

The CPI Revision

• BLS is now in the middle of a six-year (FY 1995-2000) program of Consumer Price Index (CPI) methodological and technological improvements. The most critical deadlines are in the near future (see also Attachment A):
• January 1998: Update CPI Expenditure Weights
• Effective with the February 1998 release of January 1998 data, 1993-95 consumer spending patterns will replace 1982-84 spending patterns as the basis for weighting the approximately 200 item strata in the CPI. For example, revised structure is likely to have a lower weight for food and beverages (see Attachment B for preliminary evidence on this; final 1995 data is not yet processed).
• New item structure (e.g., restructure food-away-from-home, new major group Education and Communications. See Attachment C for current and 1998 structure) New geographic sample from 1990 Census
• We have estimated that updating weights may reduce measured inflation by 0.1 or 0.2 percent per year.
• Early 1999: Improved Sample Rotation Procedures
• Replace subsets of items in all cities each year, instead of all items in some cities each year, as currently
• Important because it facilitates more rapid introduction of new goods and outlets
• Technological Improvements
• Improved sample, estimator, and data processing system for Housing, introduced in 1999
• Redesigned computer system for Consumer Expenditure Survey, beginning in 1997
• Computer-assisted price data collection using pen-pad computers, beginning in 1997
• POPS survey using computer-assisted telephone interviewing, already underway

Changes to Hospital and Related Services Component

• Shift to pricing of bundles of services on selected patient bills, rather than pricing individual items like room charges or units of blood
• Consolidation of current three hospital strata into one
• The changes are designed to provide a better representation of the hospital services that are actually delivered and to provide a context within which changes in the quality of these services can be evaluated. (See Attachment D for the July 1996 announcement of these changes.)

Experimental Geometric Mean Indexes

• For about two years BLS has been building and testing the computer programs required to compute individual stratum indexes using a geometric mean formula. Are simulating monthly historical indexes starting in December 1989.
• The index will be released beginning in early 1997, and published regularly thereafter.
• The original stimulus was the "formula bias" problem, which created an upward bias in the official CPI-geometric mean indexes are not subject to formula bias.
• Evidence from 1992-94 indicated that a geometric mean index would rise about 0.5 percent per year more slowly than the official CPI.
• Now that the formula bias problem has been corrected, the difference would probably be around 0.25 percent per year.
• The attraction of a geometric mean index now is that, under certain restrictive assumptions, it corrects substitution bias.
• Superlative indexes are designed to account for the fact that consumers shift their purchases in response to relative price changes. An appropriate standard against which to evaluate the geometric mean measure thus would be a properly constructed superlative index that accounted for substitution across items within strata.
• The superiority of the geometric mean index rests on the assumption that the elasticity of substitution is one below the stratum level.
• Research is required to establish the extent to which this assumption is accurate--or more nearly accurate than the CPI's implicit assumption of no substitution--and thus if or how the geometric mean index could be used to provide a closer approximation to a cost-of-living index. (See Attachment E for a discussion of BLS plans for evaluating the geometric mean index.)

Experimental Superlative Indexes

• Superlative indexes that reflect substitution among the 207 CPI strata have been produced by BLS since 1993 on experimental basis.
• These indexes must be produced with a lag, or subject to revision, as current spending data become available
• Published BLS research has indicated that reflecting substitution among item strata would reduce measured inflation by somewhere in the vicinity of 0.2 percent per year (more recent calculations show a slightly smaller effect).
• As noted below, we have developed plans to enhance these measures in the future, subject to resource constraints.

Other Experimental Indexes

CPI for Elderly

• Produced monthly (data going back to December 1982) and available on request
• Several limitations-weights have a high variance due to small sample size, no separate samples of outlets and items, no reflection of special discounts sometimes given to elderly
• Does provide some evidence that inflation experience may be different for the elderly (see Attachment F for a description of the CPI-E).
• Other series, produced at irregular intervals on research basis
• Interarea indexes (the CPI measures only price change over time, not comparisons of the cost of living in different areas at a point in time)
• CPI's for poor Americans

Scanner Data

• The use of scanner data offers tremendous potential long-term opportunities for producing a more accurate CPI, but there are significant obstacles in the nearer term (see Attachment G).
• Using scanners, stores collect vast amounts of data on both prices and quantities of goods sold.
• These much larger samples could mean greater accuracy
• The availability of real-time quantity data would permit real-time indexes free of substitution bias
• Unfortunately, scanner data are now limited to a minority of consumer purchases
• Available mostly in the food-at-home component, which accounts for about 10 percent of the CPI market basket; no data on most services
• Even within food at home, there are significant exclusions of items (e.g., meat and produce) and outlets (e.g., most convenience stores)
• BLS is engaged in a research program comparing scanner-based indexes to the current CPI, and also evaluating how scanner data could be used in the near term. Other CPI Program Initiatives under Development
• Decrease the time required to compile and install new market baskets during future revisions
• Expand the Consumer Expenditure Survey sample, making it possible to construct CPI expenditure weights using two, rather than three, years of data
• Streamline the systems needed to put new market baskets into place
• The combination of these activities would reduce the average age of the market basket by 18 months when first used in the CPI (from 3 1/2 years now, to 2 years).
• The benefits of a more timely market basket accrue regardless of how frequently the market basket is updated.
• Produce alternative superlative measures of changes in living costs
• These would be complete, "production-quality" indexes as opposed to current experimental measures
• Produced closer to "real time"
• Expand pricing of goods and services in the CPI
• Augmented collection of product characteristics and prices to permit increased use of "hedonic" quality adjustment
• Selective reinitiations to enhance capacity to bring "new" goods into the index

Additional Background Information

• We have included two other attachments that provide more detailed statements about the BLS position on measurement issues in the CPI
• "Bias in the Consumer Price Index: What is the Evidence?" by Brent Moulton (Attachment H)
• Presentation by Katharine Abraham at the August 1996 meeting of the American Statistical Association, reviewing BLS efforts to improve the CPI (Attachment I]

---

• Attachments for Briefing on the Consumer Price Index, December 3, 1996