Revision of Consumer Price Index is now under way Essay

Revision of Consumer Price Index is now under way

The Bureau of Labor Statistics is in the midst of a 5-year program
to update and improve the Consumer Price Index (CPI). The resulting
changes will be introduced in the January 1987 indexes. The 1987
revision will use the Consumer Expenditure Survey data from 1982-84 and
population distributions from the 1980 census to update the CPI market
basket. A greatly enhanced housing survey is being developed that will
improve the rental equivalence measure of homeowner costs recently
introduced in the CPI. Many of the sampling advances introduced in the
1978 revision will be refined for 1987 and other methodological
enhancements will be made.

This article explains why periodic CPI revisions are needed,
briefly reviews previous revisions, and describes the current revision

Why periodic revisions are needed

The CPI is a measure of price change for a fixed market basket of
goods and services of constant quantity and quality purchased for
consumption. It is essential to update that market basket periodically
so that the CPI reflects price changes of items currently purchased by
consumers. Consumers change their purchasing patterns as a result of
changes in a number of factors, including relative prices, real income,
demographic characteristics, and tastes.

Price changes over time may differ among items and these
differences can affect consumer demand. This is illustrated by rapidly
rising prices for energy items over the last decade. In the Consumer
Price Index for All Urban Consumers (CPI-U), energy products (gasoline,
motor oil, electricity, natural gas, fuel oil, bottled gas, and coal)
rose 218 percent from December 1972 to December 1980, more than twice as
fast as the average increase for all items. According to data from the
Consumer Expenditure Surveys, urban consumer units1 reported an average
annual expense for energy items of $743 for the 1972-73 period and an
annual average of $1,783 for the 1980-81 period. This 140-percent
increase is substantially smaller than the change that occurred for
prices and implies a reduction in consumption of energy items as a
result of higher relative prices. This adjustment was also seen in
related consumption such as the increased demand for smaller and more
fuel-efficient automobiles.

Another factor which can influence consumers’ consumption
patterns is changing real income. If prices paid by consumers and their
money income were to increase at the same rate, consumers’ real
income would remain unchanged. Average money income in constant dollars
declined about 7 percent between 1972 and 1981, both for households and
for families. However, per capita average real money income increased
by more than 3 percent during the same period.2 The rise in per capita
income, in contrast to the decline experienced by families and
households, is a direct result of the average size of families and
households becoming smaller.

The impact of rising prices on some families can be offset by
having additional income from another member joining the labor force.
For example, in 1972, 41.5 percent of married women with a husband
present were in the labor force. By 1981, the percentage increased to
51. The labor force participation rate for married women, husband
present, and with children under 6 years of age increased from 30.1
percent in 1972 to a rate of 47.8 in 1981.3 Demographic-related changes
of this kind affect expenditure patterns. For example, expenses for
such items as day care nursery school and babysitting might increase and
there could be additional expenses for eating meals out and

Still other factors which affect the pattern of consumption over
time are product changes and technological changes which can affect
consumers’ demand for various goods and services. The electronics
industry in particular has influenced consumer preferences by its
introduction of such items as the personal computer, video games, and
video recorders. Over time, a number of products are modified,
expanded, or improved, depending on the demand of consumers, and these
changes influence subsequent purchasing decisions of consumers.

Finally, a more subtle phenomenon which contributes to changes in
the relative importance of items in the market basket is that tastes of
consumers change. There are a variety of ways in which lifestyles and
tastes change, such as the increasing number of persons who are active
in some form of physical exercise such as jogging, cycling, or using the
facilities of a physical fitness organization. These preference shifts
also change expenditure patterns for items such as sports clothing and
equipment, and fees paid for recreational facilities.

Population changes. Not only do the consumption patterns of
individual consumer units change over time, but also the geographic
distribution of the population may change. Between 1970 and 1980, the
total population of the United States grew 11.5 percent. The population
of the South grew 20 percent and the population of the West increased 25
percent.4 This means that consumer units in the South and West
represent approximately 52 percent of the population for which the
revised CPI market basket will be based, compared with 48 percent in the
1972-73 market basket currently being used. Thus, consumption patterns
of consumers in the warmer climates of the South and West will have a
greater influence on the CPI than before.

Prior revisions

The first major activity in prior revisions of the CPI has been the
implementation of a Consumer Expenditure Survey as the basis for
selecting and weighting a new market basket of goods and services to be
priced. Until these data are in hand, it is impossible to complete a
revision of the CPI. The periods when expenditures were collected that
were the basis for the last four revisions are as follows:

Reference year(s) ; Release of revised CPI

1934-36 1940

1950 1953

1960-61 1964

1972-73 1978

The time between the reference years of the Consumer Expenditure
Survey and the introduction of the CPI with revised expenditure weights
was typically 3 years, except for the 1978 revision.

The 1940 revision introduced the concept of a sample of cities and
items and the principal of imputation, permitting the CPI to represent
price change in all cities and all items purchased for consumption.
Prior to 1940, the CPI measured the price change in only the 33 cities
being surveyed and for only the items actually priced.

Prior to the 1950 Consumer Expenditure Survey (on which the 1953
revision was based), BLS conducted experimental surveys and test
pricings to improve data collection methods and to establish the basic
procedures for processing these data. The 1953 revision took 3 years to
implement surveys which revised the areas and weights, and updated the
item samples priced. This effort was primarily a clerical operation.

After the 1953 revision, it became apparent that the CPI should be
revised every decade. By the late 1950’s, dramatic changes had
occurred: The composition of the urban population changed, with rapid
growth of suburban areas, increased use of the automobile affected
lifestyles, and new shopping centers catered to the American consumer.
A contributing factor to this growth was the 37-percent increase in
personal disposable income between 1950 and 1956, with more than
two-thirds of the rise being reflected in real income. The BLS
received, in mid-1959, authorization for a revision program, which was
completed in 1964 with the release of an index with revised weights and
outlet samples which included, for the first time, areas outside the
central city of metropolitan areas.5

The first year of the 1964 revision was dedicated to pilot surveys
for testing and debugging procedures to be used nationwide. After
clerical edits and professional reviews of the data, the computer was
used to process estimates of expenditures and indexes.

The 1978 CPI revision took longer than the previous revisions
because it included the introduction of new approaches to the collection
of consumer expenditures and a number of complex improvements and
innovations in pricing for the CPI. A thorough examination of the CPI,
its concepts and operational processes, was made during the revision.
The growth of computer applications during the decade of the 1960’s
made it possible to introduce statistical techniques and monthly
operational processes which were not feasible in earlier efforts of
producing estimates of monthly price change.

Innovations of the 1978 revision

Innovations in collecting expenditure data for the 1978 revision
contributed to a longer time between the Consumer Expenditure Survey
reference data and the introduction of the revised CPI. Prior to the
1972-73 Consumer Expenditure Survey, interviewers visited all sample
households during February through June and asked the respondents questions needed to reconstruct their living expenses for the previous
calendar year. These global estimates of expenditures were used to
obtain annual expenditures for most items. Respondents were asked to
recall weekly expenditures for food store items and small frequently
purchased items.6

Several changes in these procedures were made early in 1972. A
quarterly interview survey for a sample of consurmer units was
introduced. Expenditures for a number of items were collected for
purchases made throughout the preceding 3 months, while other items were
surveyed for varying reference and recall periods. Another separate
sample of consumer units was asked to keep two 1-week diaries in which
each purchase was recorded on the day it was made. Although this change
in methodology was more expensive and took somewhat longer to process,
it resulted in a marked improvement in the data used in the estimation of expenditure weights. It reduced the length of recall in the
collection of data, and, therefore, reduced response errors associated
with either telescoping purchases from an earlier period or forgetting
certain purchases. (Telescoping occurs when the respondent inadvertently
recalls and reports a purchase made prior to the survey period.)

Another significant innovation in the 1978 revision was the
introduction of the Point-of-Purchase Survey.7 In earlier revisions,
the BLS had to rely on secondary data to establish sampling frames used
in selecting outlets in which to price items comprising the market
basket. These secondary data provided only the broadest classification
of the outlet and provided no detail on the merchandise lines actually
purchased. For example, it was not possible to identify all the types
of outlets where motor oil was sold, and it was impossible to tell
whether a particular grocery store sold fresh fish. As a result,
despite substantial efforts, it was impossible to obtain a statistical
sample of outlets for the CPI that represented where people shopped.
The growth of metropolitan areas and the spread of shopping centers
added to the concern about the quality of outlet samples.

In the Point-of-Purchase Survey, consumer units were interviewed in
each local area where prices were to be collected for the CPI.
Respondents specified the amount they actually spent in each outlet in
which they shopped for a category of items. Each category was
structured to be compatible with a major line of goods or services sold,
and so that the category would contain one or more “entry level
items’8 for which a relatively broad class of products or services
could be priced to represent that entry level item. (The current CPI
market basket contains 382 such items.) The Point-of-Purchase Survey
respondents were asked if they purchased an item within a specific
category during a prescribed reference period. If a purchase was made,
the name and address of the outlet was recorded along with the cost of
each transaction.

Prior to the introduction of the Point-of-Purchase Survey, each
outlet was selected and weighted without specific regard for the
relative sales that the outlet had for the priced item. The only
exception to this procedure was in the grocery store food index where
sales data were obtained from food chain organizations so that
differential weights could be used to weight prices in the food index
for large food chains. Since 1978, the BLS has used a probability
procedure with the value of purchases of each outlet as a measure of
size to select outlets for each Point-of-Purchase Survey category. This
ensures that the outlet sample has an unbiased representation of large
and small establishments and also allows for the estimation of variances
and sampling error.

In addition, prior to 1978, there was no systematic statistical
process for replacement of outlets which closed, moved, or changed
merchandise lines. BLS had to rely primarily on its CPI field
representatives to locate a comparable establishment to obtain price
quotes for the specific items to be priced. With the composition of
outlets gradually changing due to the entry of new establishments, it
was difficult to ensure the representativeness of the sample of outlets.
In 1978, BLS introduced a new system, based on data from the continuing
Point-of-Purchase Survey, to update CPI outlet samples in each urban
area on a 5-year cycle. Outlet samples in about 20 percent of the urban
areas priced for the index are updated each year so that the entire
outlet sample is completely updated over a period of 5 years.

When substituting a price quote for an outlet item selected from
the updated sample for a corresponding outlet item quote previously
priced, it is necessary to factor out of the index measure any
difference between the two prices which results from this substitution.
For example, if a man’s 100-percent cotton dress shirt is selected
in the newly selected outlet to replace a cotton blend shirt which was
priced in the outlet to be replaced, the prices of these two items would
not be viewed as comparable for an index measuring price change.
Differences that may exist between the new outlet item quote and the old
one are factored out of the measurement of price change by a method
described as linking with an overlap price. This linking method used in
outlet sample updating requires that both the new and old outlet item
quotes be priced in the same month. The price for the item quote or the
outlet being replaced measures any price change from the previous index
month up to the link month when both outlets were priced. The price of
the item quote from the newly selected outlet is used to measure price
change from the link month forward. This linking method assures that
the process of introducing the new item has no effect on the index.

Prior to 1978, an item designated for pricing in an outlet would
have characteristics specified by commodity analysts in the national
office. The detailed specification was usually the same for all outlets
in the country and would generally limit the number of products that
could be priced to represent the expenditures within the item class. In
the 1978 revision, the BLS introduced new sampling procedures to permit
all products or services within a respective item classification to be
eligible for pricing, thereby increasing the efficiency and
representativeness of the index. However, once an outlet item is
selected, the field representative records the specific narrow
characteristics of the item to identify it for continuous pricing as
long as the item is available in the store.

The process used in the selection of an item within an outlet is
called disaggregation. This process gives an opportunity for every
variety of an item within a store to be selected to represent purchases
for the whole item class. This disaggregation is an objective and
efficient process which results in the selection of a sample of
varieties that covers the full spectrum of purchased items.9

During the 1978 revision, a great amount of time was spent
examining alternative methods of measuring price change in
homeownership. This effort resulted in the definition of a
flow-of-services approach10 which is consistent with the economic
concepts on which the CPI is based. This approach was not introduced
during the 1978 revision because of difficulties in developing a
workable flow-of-services measure and because of the diversity of views
held by various advisory groups.11

Shortly after the revision, concern over the measurement of
homeownership costs increased because of the major changes that were
occurring in the financing of homes and the increasing difficulties of
obtaining adequate house price data. Because of these changes and the
increasing impact and importance of the CPI, BLS changed the
homeownership component of the index between revisions.12 A rental
equivalence measure13 was introduced as the measure of homeowner cost in
the January 1983 CPI-U index and in the CIP-W with the January 1985
index. The rental equivalence measure estimates the change in shelter
costs as the change in rents which would have to be paid for occupancy
of housing units occupied by owners. This new measure replaced the
previous treatment in which homeownership costs were measured by current
house prices, mortgage interest, costs of new mortgages, property taxes,
property insurance, and maintenance and repair costs. Because it
measures the cost of consuming shelter services provided by a house
(that is, the rent that would be paid), rental equivalence is consistent
with the underlying concept of the CPI as a measure of price change for
consumption. The old homeownership measure included investment aspects
of homeownership associated with obtaining and maintaining the house as
an asset.

The 1987 revision

The CPI relates to expenditures of the civilian noninstitutional
urban population of the United States. The urban population is defined
as persons who live in Metropolitan Statistical Areas as defined by the
Office of Management and Budget (including the rural nonfarm within
these areas) and urban areas, including places with 2,500 or more
persons outside of the Metropolitan Statistical Areas.

Since the 1978 revision, the CPI has been calculated for two
populations. The All Urban (CPI-U) index is based on expenditures
reported by all consumer units in urban areas of the United States with
two exceptions: consumer units on farms within urban areas and consumer
units receiving a majority of their income from a member who is in the
military and lives off base with the unit. The CPI-U population
represented 81 percent of the total U.S. civilian noninstitutional
population in 1981. Because a large proportion of the population is
covered, this index is extremely useful in discerning the effect of
changing prices on consumers.

The CPI for Urban Wage Earners and Clerical Workers (CPI-W) is
based on urban consumer units who meet additional requirements related
to their employment: more than one-half of the consumer unit’s
income has to be earned from clerical or wage occupations, and at least
one of the members had to be employed for 37 weeks or more in an
eligible occupation. The CPI-W population was 30 percent of the total
U.S. population in 1981.

Geographic coverage. The first phase of a revision is to make a
new selection of the geographic areas, or primary sampling units, in
which price data collection will be done. The new area sample for the
1987 revision is based on the 1980 Census of Population and uses the new
Consolidated Metropolitan Statistical Area definitions.14 The use of
these definitions resulted in some changes. For instance, the
definition for the New York area now includes Danbury and other parts of
Connecticut; Philadelphia includes Wilmington and Trenton; Los Angeles includes Riverside-San Bernardino; and San Francisco includes San Jose.
The Metropolitan Statistical Areas which are not a part of a
Consolidated Metropolitan Statistical Area were defined as individual
primary sampling units. All nonmetropolitan counties were grouped into
primary sampling units to allow urban places with a population greater
than 2,500 outside metropolitan areas an opportunity to be selected.
The overall primary sampling units design consisted of 278 metropolitan
areas and 810 nonmetropolitan urban areas, which cover all urban
population. Primary sampling units with at least 1.2 million persons
were designated “certainty areas.’ This means that each of
these areas represents itself in the weighting of the estimates to the
total CPI population. The noncertainty selections have a population
weight that represents the population of all cities including their own
population in their stratum–a collection of similarly sized areas in
the same general geographic region. Twenty-nine largest primary
sampling units and two unique areas–Anchorage and Honolulu –were
designated certainty areas.15

The remaining primary sampling units were assigned to three
city-size classes–medium-sized cities, small-sized cities, and
nonmetropolitan areas–within the Northeast, North Central, South, and
West regions.16 The result of the sampling process was the selection of
39 new areas with the retention of 52 primary sampling units from the
old sample, of which 30 were certainty selections in the new sample.
Overall, the number of primary sampling units to be surveyed for the CPI
has increased by six. A comparison of primary sampling units in the old
and new samples by population size and region is shown in table 1.

The South will have eight more primary sampling units than it had
previously. Despite the West’s large population growth between the
1970 and 1980 censuses, it will still have the same number of primary
sampling units; however, it will have two more certainty selections.
Two reasons account for the unchanged overall number of primary sampling
units in the West. First, additional primary sampling units were
allocated disproportionately to the West in previous allocations to
permit publication of a separate nonmetropolitan urban index for the
region. Second, use of Consolidated Metropolitan Statistical Areas
resulted in two certainty selections, Los Angeles and San Francisco,
becoming substantially greater in population. Because of their larger
populations in the new CPI design, the number of items and outlets
priced in each of these two areas will be expanded.17

Allocating samples to produce the most accurate national CPI
possible with the funds available will affect the frequency of
publishing CPI’s for 13 local areas. Beginning with the January
1987 CPI, a monthly index will continue to be published for only the
four largest local areas–New York, Los Angeles, Chicago, and
Philadelphia. The index for Detroit, the smallest of the areas now
published monthly, will be compiled on a bimonthly basis only for even
unmbered months. Bimonthly indexes will be published for each of the
next 10 largest areas. Bimonthly indexes which are now published for
the 12 smaller local areas will be replaced by semiannual average
indexes, and the index for Northeast Pennsylvania (Scranton) will be

Expenditure weights. The relative weight of each entry level item
in the CPI is tabulated from data obtained by the Consumer Expenditure
Survey. This survey is actually composed of two separate surveys–an
interview survey and a diary survey–both conducted by the Bureau of the
Census for the BLS.

As in the 1972-73 interview survey, Bureau representatives collect
data for expenditures which respondents can remember fairly accurately
for periods of approximately 3 months. Each consumer unit designated
for sampling is contacted each quarter for five consecutive quarters.
The initial contact is used to collect socioeconomic characteristics of
the unit–an inventory of properties, vehicles, major durable goods, and
insurance policies. In addition, purchases of goods and services made in
the past month are recorded together with a date of purchase and a
description of each item.19

BLS uses only the second, third, fourth, and fifth interviews in
estimating a 12-month consumption pattern for the consumer units
surveyed. The current interview questionnaire differs from that used in
the 1972-73 survey in that it has a uniform reference period of 3 months
for each expenditure item, whereas the 1972-73 questionnaire allowed for
variable lengths of the period of recall. The major advantage of a
uniform reference period is that it permits each interview to be used in
a quarterly estimate, even when a consumer unit was not interviewed for
the full 12 months of consumption. All data collected from consumer
units are used, in contrast to 1972-73 when data from consumer units who
later moved were not used.

The uniform reference period facilitates rotating the sample. Each
quarter, one-fifth of the consumer units are interviewed for the first
time, an additional one-fifth for the second time, and so on. The
rotation spreads over the calendar year any bias which may result from
either conditioning or fatigue on the part of respondents as they
progress from the first to the fifth interview. Because many expenditure
items are seasonal, it is advantageous to have a mixture of interviews
in each quarterly estimate of consumption patterns.

The purpose of the diary survey is to obtain expenditure
information for small frequently purchased items which consumers tend to
forget. Each selected sample unit is asked to keep 1-week diaries of
expenditures for 2 consecutive weeks. The diary sample is spread among
the 52 weeks of the year. However, the sample size is doubled in the
last 6 weeks of the year to obtain better estimates of items purchased
seasonally. The interviewer, when placing the first-week diary, obtains
the socioeconomic characteristics of the consumer unit and provides
instructions to the respondent. The respondent records purchases made
by any member of the unit during the week. (This eliminates any
questions the respondent might have in determining if an item is within
the scope of the survey.) The diary focuses on the recording of
purchases made in grocery stores and of meals, snacks, and beverages
purchased in restaurants or other eating places. Other purchases are
also recorded; therefore, a number of items reported in the interview
survey can also be recorded in the diary. A major difference in the two
surveys is that the diary does not record expenses made while out of
town on trips. Both surveys have a sample size of approximately 4,800
consumer units per year. However, in the interview survey, each unit
can potentially provide four quarters of data, whereas in the diary only
2 weeks of data can be obtained from the same unit.

The BLS staff has to identify from which survey–interview or
diary–estimates should be used in developing expenditure weights and
selecting item samples. For many items, the design of each survey
predetermines which data should be used. For example, the diary
estimates are used for all individual food and beverage items because
the interview survey only collects a total estimate of expenses for
these items. The diary is also used for a number of small and
frequently purchased items in the categories of personal care, household
supplies, and nonprescription drugs and supplies which are not covered in the interview survey. For other expenses, the interview survey is
the better source as it has an effective sample size of 4,800 units each
quarter and expenses are recalled for a period of 3 months. The diary
panel, in contrast, only has an effective sample of 1,200 units per
quarter for a total of approximately 2,400 diary weeks. There are a few
expenditures that are collected in both surveys for which an evaluation
is necessary to determine which estimate is best. For example, gasoline
purchases are a frequently reported entry in the diary, and the estimate
obtained from the interview is based on an average monthly expense
pattern. Also, small clothing items such as hosiery and accessories
could be overlooked in the 3-month recall which is the heart of the
interview survey, but are likely to be recorded in the diary.

Each expenditure reported in these two surveys is coded to one of
the 382 entry level items which constitute the lowest level of the CPI
classification structure. The highest level of the CPI structure
consists of the seven major groups of expenditures: (1) food and
beverages, (2) housing, (3) apparel and upkeep, (4) transportation, (5)
medical care, (6) entertainment, and (7) other goods and services.
Expenditures within a major group are divided into expenditure classes
which have been established either by categories of commodities or
services and with some regard to similarity in their characteristics.
The CPI structure currently has 68 expenditure classes and a new one
will be established in this revision for electronic products covering
personal computers, computer software, calculators, telephones and other
information processing equipment. (See exhibit 1.)

Most of the expenditure classes are divided into two or more
strata.20 The stratum is the lowest level for which expenditure weights
are calculated, and thus, the level at which the priced market basket is
fixed between revisions. Because the allocation of the sample of quotes
and outlets is also done at the item stratum level, the unmber of strata
within an expenditure class generally has some overall relationship to
the relative importances of expenditures in that class. The variances
of the CPI can be greatly influenced by the way price quotations are
allocated among the item strata. In this revision, a paramount
consideration was to maximize the efficiencies that could be achieved
through sample designs and the allocation of samples. Using data from
the 1980-81 CES and preliminary data on variances, item strata were
restructured so that, given the available resources for pricing, the
variance of the All Items CPI would be a minimum. A very few selected
strata were left unchanged because of their individual uses or interest.
The number of strata for which expenditure weights are calculated will
drop from 265 to 203.(21) (See exhibit 1.)

Each item stratum has at least one entry level item which is
usually structured to facilitate the selection of a unique item to be
priced. If there is much heterogeneity among the goods or services
which comprise an item stratum or in the types of outlets where they are
purchased, the stratum is usually subdivided into two or more entry
level items. Currently, there are 382 entry level items and although
the composition of several will be changed in this revision, the total
number will probably not change by much.

In the past, there have been a few sample entry level items which
have not been priced. Sometimes the item was difficult to price because
its quality changed constantly. An example would be the pricing of
books purchased through book clubs. The book offered varies
substantially over time and various discounts or premiums may be earned.
Also, an entry level item may not have been priced because an
appropriate outlet sample could not be established. This is the case
particularly for services provided by household workers and babysitters.
If an entry level item or a potential one has a small relative
importance, the Bureau does not go to a great disproportionate expense
to price it. In the current revision, the Bureau plans to use the
relative importances of entry level items reported in the interview and
diary surveys to identify those that have become more significant since
the last revision. As a result, unpriced strata are expected to
comprise only 1.5 percent of the CPI, compared with 3.7 percent

Outlet selection. The 1987 revision will rely primarily on the
continuing Point-of-Purchase Survey for the selection of outlet samples.
When this survey was initially designed in 1974, there was some concern
that it would not be useful in selecting outlets for entry level items
which were purchased either infrequently or by a relatively small
percentage of consumers. In updating outlet samples over recent years,
a number of these entry level items have been added to the
Point-of-Purchase Survey. By extending the reference period for such
items, the continuing Point-of-Purchase Survey has proved effective in
securing a sufficient outlet sample.

There are a few entry level items for which outlet samples are
obtained from sources other than the Point-of-Purchase Survey.
Generally, these items are found in a relatively small number of
establishments, and reliable information is readily available for
establishing a measure of size in the sampling frame. Examples of such
entry level items are natural gas, electricity, basic telephone service,
casualty insurance premiums, postage rates, and train fares. The
ongoing Consumer Expenditure Survey is collecting outlet information
(along with the expenditure data) for a small number of these entry
level items. After these data are evaluated, we will determine if it is
possible to use the Consumer Expenditure Survey for selecting outlet
samples for such entry level items as electricity, natural gas, and
tuition. Data collected in the Consumer Expenditure Survey with regard
to consumption quantities on utility bills will be used for selecting
the consumption amounts to be priced for the CPI.

New Strategy: “rolling-in’ samples

In previous CPI revisions, a new area sample (primary sampling
unit) and new item and outlet samples were introduced at the same time.
The 1987 revision will use a concept of rolling-in the new area, item,
and outlet samples. That is, the composition of the area and item
samples will be gradually updated over a period of years, rather than
substituting the full set of new area and outlet samples at a single
time. Two innovations of the 1978 revision facilitate this rolling-in
strategy: the use of the continuing Point-of-Purchase Survey for a
systematic updating of outlet samples, and the broader definition of the
characteristics of items which define strata. The first stage of
rolling-in the new sample is to initiate pricing in new areas which will
be needed in January 1987 for updating the U.S. CPI to reflect changes
in population distributions. A number of the areas which had been
representative of a specific city-size had sufficient population growth
between 1970 and 1980 so that they no longer represented that particular
city-size. There are 19 new areas classified as either small- or
medium-sized or nonmetropolitan, and one large-sized area which have to
be surveyed prior to 1987 in order for the U.S. CPI to reflect the
distribution of the U.S. population as enumerated by the 1980 census.

The second aspect of this phased update pertains to the item
samples in all CPI areas retained in the new design. Any new entry
level item or any entry level item that is substantially modified in
definition will be initiated in all areas prior to the issuance of the
revised CPI for January 1987.

The continuing Point-of-Purchase Survey for 1985 will be conducted
in the 20 new areas so the item and outlet samples for these areas can
be initiated and results introduced in the January 1987 CPI. (An
additional 19 new areas will be initiated and introduced over the
1987-89 period.)

The new item expenditure weights tabulated from the 1982-84
Consumer Expenditure Surveys will replace those tabulated from 1972-73
survey data. To make this substitution of expenditure weights without
causing a discontinuity in the CPI’s measurement of price change,
the index levels using the new expenditure weights will be set equal to
those published for the old series in December 1986. The official CPI
for January 1987, therefore, will reflect the price change between
December and January based on the new expenditure weights. As in the
past, the Bureau will continue to publish overlap indexes using the old
expenditure weights for 6 months after the issuance of the revised CPI,
for the convenience of users.

Outlet samples for entry level items retained from the old primary
sampling unit design will be updated through the use of the continuing
Point-of-Purchase Survey and the existing outlet updating procedures. A
few of the retained areas will have their outlet samples updated in 1987
when 10 more new areas are rolled-in. The remaining areas will have
item and outlet samples rolled-in over the next years.

Advantages. By rolling-in the new areas and using the established
outlet updating process for areas retained in the CPI design, it is
possible to effect significant time and cost savings. One of the most
costly activities of past revisions was the initiation of pricing of the
item and the outlet samples in all areas selected in the primary
sampling unit redesign. Prior to the introduction of the revised CPI,
all of the item and outlet samples had to be initiated and priced in the
same month as the existing samples. Even for the areas retained, a
reselection of item and outlet samples required substantially more new
pricing because the probability of reselecting the same outlet for an
entry level item is very small. Additional field representatives had to
be hird and trained to do this work while pricing was continued to
produce the ongoing CPI.

Because the existing CPI is official until the revised index is
released, the review and processing of data from the new samples must be
done in a framework which does not jeopardize production schedules.
Rolling-in the new areas into the CPI estimate over a 3-year period
allows more time to train field representatives and lessens problems
associated with a rapid expansion and subsequent reduction in staff.
More importantly, using the existing updating procedures for introducing
new outlet samples on a systematic basis precludes the need to maintain
extended dual operations– one for the existing CPI and one for the data
scheduled to supersede it. Over the past 6 years, the Bureau has used
this technique for updating outlet samples. A few modifications to
accommodate new areas and entry level items will increase the amount of
data requiring processing, but by substantially less than the old

Expenditure weights for the 203 strata in the CPI market basket
will be tabulated using 3 years of Consumer Expenditure Survey
data–1982, 1983, and 1984. Because the CPI is a base-weighted index
designed to reflect price change (and not changes in the quantities
purchased), these expenditure weights will remain fixed until the next
revision of the CPI. As in the past, of course, BLS will continue to
update the outlet sample in one-fifth of the CPI areas each year.

Within the CPI fixed-weight constraint, however, BLS intends to
make maximum use of data from the ongoing Consumer Expenditure Survey to
keep the items priced to represent the strata up to date. A number of
CPI strata, for example, are represented by 2 or more entry level items.
The sample of entry level items for these multiple-entry level item
strata have been selected from the Consumer Expenditure Survey.

Beginning in 1987, when the outlet samples are updated for
one-fifth of the urban areas and new detailed items are selected for
pricing, this sample of entry level items will also be updated based on
the two most recent years of Consumer Expenditure Survey data. If
relative shifts of consumption occur among items within a stratum or new
products appear within the stratum, then entry level item reselection
will gradually change the composition of the entry level items being
priced. In other words, the entry level item sample will begin to
reflect the changes consumers are making in the variety of products
purchased which make up an item stratum of he index. The reselection of
the item samples within each fixed-weight category for one-fifth of the
area sample does not alter the fixed-weight nature of the CPI because
the population-expenditure weights will remain fixed, as now, at the
item strata level until the next revision. This reselection will not
affect entry level items which have a very large relative importance or
are the only ones in the particular strata and, therefore, are certain
to be priced in all urban areas.

Although the CPI will continue to have its basic fixed-weight
character, the existence of annual expenditure data will offer a number
of opportunities for developing experimental indexes with different
characteristics. For example, while the expenditure weights for the
official CPI are updated only about once every 10 years, experimental
indexes could be developed with more frequent weight changes.

Improvements of the 1987 revision

Enhanced shelter survey. The adoption in 1983 of rental
equivalence to measure changes in the cost of the shelter component of
owner-occupied homes put the housing component of the CPI on a
flow-of-services conceptual footing, and isolated the consumption
element of owner housing from its investment element.

In addition to updating the housing sample based on the 1980
census, the 1987 CPI revision program will enhance the rental
equivalence method adopted in 1983.22 The selection of a new housing
sample is designed to represent optimally both owners and renters. A
multi-stage sampling procedure was used that stratifies the residential
areas of each primary sampling unit by tenure (percent owner-occupied)
and rent level. Smaller areas are then defined and sampled within each
selected area. The housing units of each selected small area are
screened for tenure and sampled at differential rates according to
tenure. In heavily owner-occupied areas, for example the renters are
selected more frequently than owners in order to find renters who are
like owners, because it is from these rentals that the best estimates
can be made in the implicit rent of owner-occupied dwellings.

Enhancement of statistical techniques. Because the Consumer
Expenditure Survey estimates for each of the individual areas of the
country are based on relatively small samples, BLS has undertaken
research in statistical techniques to reduce the error on local areas
index weights. In the 1978 revision, a compositing technique was used
in which the local area average expenditures were weighted together with
the expenditure estimates for the same item class for the geographic
region to which the local area belongs.

Research done at the BLS during the current revision involved use
of the composite estimation of relative importances rather than of mean
expenditures of the item categories. Relative importances in the CPI
are the mean expenditures for each item as a percentage of all
expenditures. BLS statisticians found compositing of relative
importances to be more effective in reducing the average mean squared
error than compositing of the expenditures themselves.23

Another refinement under consideration is to replace estimates for
each of the four broad geographic regions of the country (Northeast,
North Central, South, West) with two sub-area estimates–one for the
certainty areas within each region and one for all other areas within
each region. The relative importances of each certainty area within
each region would be estimated based on composites which use relative
importances of expenditure patterns from all certainty areas in the
region. The division of the regional estimate between certainty areas
and all other areas has also proved effective in reducing the average
mean squared error.

Publication of quantitative measures of sampling error for selected
indexes is planned for the 1987 revision. Initially, estimates of the
index variance will be available in the All Items CPI and for some of
the major group indexes. Eventually, more indexes will have an estimate
of variance published.

Enhanced quality. During the implementation of the 1987 revision,
the Bureau will add a new dimension to quality assurance and control of
the CPI program. Throughout the years, the staff has devoted
substantial time to the inspection of data collection and processing
activities. The goal of the inspections was to identify and correct
individual error. The goal of the new audit process to be instituted in
this revision will be to achieve long-term quality improvement. This
will be accomplished, in part, by an independent staff which will
systematically evaluate survey processes empirically.

By having independent audit data for comparison purposes, error
profiles can be used to identify the type of errors, diagnose their
sources, and prescribe procedural changes to prevent these errors from
occurring in the first place. The techniques used will include special,
detailed evaluation studies of specific processes, ongoing process
controls and reports, statistical quality control and measurement, and a
system for information feedback and corrective action. The goal is to
develop processes that will result in enhanced estimators of price

Other concepts to be investigated

As part of the revision, BLS will investigate the appropriate
treatment of insurance premiums in the CPI. Currently, premium costs
for health insurance and casualty insurance for vehicles and household
furnishings are priced for the index. The overriding issue in the
pricing is the one of constant quality in the coverage. Quality changes
that affect premium level should be removed before being used in the
CPI. Using health insurance as an illustration, there are four factors
which affect changes in premiums: (1) changes in the costs of medical
procedures, (2) administrative cost and surplus requirements and the
profit needs of commercial carriers, (3) policy benefit changes, and (4)
utilization changes, that is, changes in the frequency of a covered
event occurring. Changes in the first two factors do not affect policy
quality, whereas changes in the latter two will. For the past 20 years,
the Bureau has used an indirect method of pricing health insurance
because it has been unable to develop an effective methodology for
removing the effect of most changes in the coverage or the utilization
rate.24 The indirect method of pricing health insurance measures changes
in medical costs (factor 1) by using the price changes which have
occurred in physicians’ and hospital fees in the CPI to represent
the change in costs that insurance carriers have incurred for their
policy holders. Changes in costs for carriers (factor 2) are measured
by the annual changes in the retained earnings (premium revenue less
benefit payments) of insurance carriers. Thus, the indirect method
measures changes which affect policy premiums while excluding from the
measurement the two factors which affect quality.

Direct pricing of a sample of policies was tried during the 1978
revision, but was dropped due to the unresolved issue of quality
adjustment. BLS was unable to measure satisfactorily the premium value
for changes in the coverage of the policies and for the impact of
changes in the utilization of policies. For the 1987 revision, research
is continuing to determine if a procedure can be adopted that produces
adequate direct adjustment for changes in coverage and utilization.

Casualty insurance on vehicles and household effects is directly
priced in the CPI. Factors for removing quality changes from these
kinds of policies were developed for changes in deductible provisions
and for mandatory “no fault’ automobile insurance. Generally,
other policy coverage changes are treated by not using the policy and
its premium in the index calculation for the month of the change. With
casualty insurance, however, price changes which result from changes in
utilization rates are usually reflected in the index. The difference in
the treatment of utilization changes for health insurance and casualty
insurance is being reviewed as part of the revision.

Evaluating substitute items

One of the most difficult problems for those who compile price
indexes is that of quality change. Products and services change
constantly, and new items replace old ones on the market. There is a
large body of literature on the effect of quality change on Consumer
Price Indexes.25 Most of these studies show mixed results. Although it
is generally agreed that quality adjustment error exists, the extent of
the error, and, indeed, even its direction, are not known.

A series of practical techniques for handling substitution and
quality change issues in an operating environment has been developed.
Briefly, if an item and its substitute are comparable, with no
significant difference in quality, then the prices are directly compared
and used in the index. If the items are judged not comparable, then the
price difference must be broken down into quality change and price
change. This process results in one of three actions: (1) a quality
adjustment is made by using the difference in production costs and
adding a markup to retail or by some other method of valuing the
difference in characteristics, (2) if both the old and new items can be
priced in the current period, the difference in price in this period is
considered the value of quality change (this “overlap pricing’
is the technique used in outlet updating), or (3) if neither a quality
adjustment nor an overlap price is possible, then the price change of
the new item is not used in the current estimate, and a current price
for the old item is imputed using price movements of the quotes with
comparable prices in both periods in the item strata or market basket.
This third action (referred to as linking) not only precludes a quality
change from being reflected in the index, but can also preclude capturing the price change–either positive or negative– which may have
occurred at the time of the substitution in the specific item.

Of the more than 1 million distinct price quotes obtained for items
other than shelter in the index during 1983, only 3.8 percent were
substitutions. But this relatively low frequency of substitution still
had a major impact on the CPI. Price changes associated with the
substitutions accounted for more than one-half of the total price change
in the year, and quality changes equal to about one-third of the total
price change were excluded from the index. More than 40 percent of
these substitutions were comparable, and an additional 8 percent were
adjusted explicitly for quality changes. An additional 45 percent of
the substitutions were judged noncomparable and linked, while an overlap
price was obtained about 6 percent of the time. The highest
substitution rate (17.3 percent) was for apparel and upkeep items.26

In cases where noncomparable substitutes are “linked’ out
of the index, there is a danger that the CPI is missing some real price
change. The converse danger of including some quality change in the
index also occurs when two versions of an item are declared comparable.
Because of the significant impact such substitutions have on the index,
research is under way to identify methods to reduce the risks associated
with missing price change by linking and with reflecting quality change
as price change when declaring substitutes comparable.

As the revision progresses, detailed reports will be prepared on
the results of specific investigations and research.


1 A consumer unit is comprised of either all members of a
particular household who are related by blood, marriage, adoption, or
other legal arrangements such as a foster child; a person living alone
or sharing a household with others or living as a roomer in a private
home or lodging house or in permanent living quarters in a hotel or
motel, but who is financially independent; or two or more persons
living together who pool their income to make joint expenditure

2 See Current Population Reports, Consumer Income, Series P-60,
No. 142 (Washington, Bureau of the Census, February 1984), tables 3,
14, and 38.

3 Elizabeth Waldman, “Labor force statistics from a family
perspective,’ Monthly Labor Review, December 1983, pp. 16-20.

4 Derived from table 52, Persons by Race for Regions: 1980 and
1970, United States Summary, General Population Characteristics,
PC80-1-81 (Washington, Bureau of the Census, May 1983).

5 For a detailed description of the 1964 CPI revision, see The
Consumer Price Index: History and Techniques, Bulletin 1517 (Bureau of
Labor Statistics, 1966).

6 For a detailed description of the 1960-61 survey, see Consumer
Expenditures and Income: Survey Guidelines, Bulletin 1694 (Bureau of
Labor Statistics, 1971).

7 The Point-of-Purchase Survey is a household survey conducted by
the Bureau of the Census each year in one-fifth of the areas sampled by
BLS for the CPI. The survey is designed to periodically update the
outlet sample used for pricing various items. Approximately 4,000
households are contacted each year and asked to provide data on names of
retail, wholesale, or service establishments for purchases of 156
categories of goods and services.

8 An entry level item is the ultimate sampling unit for
expenditure items selected from the Consumer Expenditure Surveys by the
Washington office. Each entry level item establishes the definition to
be used by data collectors in the identification of unique items within
an outlet that can be selected for pricing an entry level item.

9 For further elaborations of the CPI methodologies, see BLS
Handbook of Methods, Volume II, The Consumer Price Index, Bulletin 2134
(Bureau of Labor Statistics, 1984).

10 The flow-of-services approach measures the cost of consuming
shelter services provided by a house. The approach focuses on
consumption and abstracts from the investment aspects of home purchase
decisions. See the following Monthly Labor Review articles: Robert
Gillingham, “Estimating the user cost of owner-occupied
housing,’ February 1980, pp. 31-35; and Robert Gillingham and
Walter Lane, “Changing the treatment of shelter costs for
homeowners in the CPI,’ June 1982, pp. 9-14.

11 For more information, see “Changing the Homeownership
Component of the Consumer Price Index to Rental Equivalency,’ CPI
Detailed Report, January 1983, pp. 7-13.

12 Janet L. Norwood, “Statement Regarding Changes in the
Consumer Price Index,’ USDL News Release, 81-506, Oct. 27, 1981.
This release explains reasons for introducing rental equivalence between
revisions. See also “Changing the treatment of shelter costs’
and “Changing the Homeownership.’

13 The rental equivalence approach as incorporated into the CPI
attempts to answer the following question: How much rental income do
the owners of housing units forego when they choose to occupy the units
themselves instead of renting them out?

14 Consolidated Metropolitan Statistical Area is an area which has
more than 1 million population and is contiguous to one or more primary
metropolitan statistical areas.

15 Anchorage and Honolulu have been designated certainty areas
since 1964, shortly after these territories were legislated to
statehood. They are great distances from the areas comprising the West
region so it is unlikely that a population market basket of other areas
would provide a good representation of them.

16 In order to allocate the remaining primary sampling units to
each size class of the design as proportionally as possible to its share
of the urban population, the population demarcation between medium-sized
cities and small-sized cities varies by region–from 330,000 in the West
to 500,000 in the Northeast. Further, proportional allocation will
preclude the publication of nonmetropolitan urban areas in the Northeast
and West as a minimum of four primary sampling units are required, and
these regions received only two.

When selecting the sample of primary sampling units, major
considerations are the costs of hiring and training field staff in new
areas as well as the requirements related to the linking of CPI region
city-size indexes. Thus, the BLS uses a statistical procedure which
maximizes the probability of retaining primary sampling units from the
old design.

The goal of this procedure is to increase the number of primary
sampling units overlapped between the two designs, compared to an
independent selection of primary sampling units, while at the same time
reflecting the shifts in population of primary sampling units between
the censuses. The BLS also uses a controlled selection to ensure that
the representation of the sample by State is directly proportional to
the population of the State. (See Cathryn S. Dippo and Curtis A.
Jacobs, “Area Sample Design for the Consumer Price Index,’
1983 Proceedings, American Statistical Association.)

17 The budget of the CPI constrains the number of items and
outlets which can be priced. The item sample design developed in the
1978 revision designates the number of price quotes which are required
for each item stratum in the CPI market basket. Some variability in the
number of quotes obtained occurs because of the greater relative
importance of some items and differential allocation based on collection
costs and variances of price change. The basic unit for allocating item
quotes among the primary sampling units selected for pricing is called a
halfsample. Each halfsample has approximately 1,100 quotes and is
called a halfsample because at least two are required to calculate a
price index for a specific CPI market basket area. The proposed budget
for maintaining the CPI after the 1987 revision supports 127

When allocating the 127 halfsamples among the primary sampling
units of the new design, the primary objective was to make the sample as
efficient as possible to minimize the sampling error of the national
index. Each of the 91 primary sampling units was allocated one
halfsample. For the optimization of the design, a primary sampling unit
should only receive an additional halfsample if its population is
greater than 1/127 of the total population. The remaining halfsamples
were allocated among the 15 largest primary sampling units. By doing
this, the efficiency of the national CPI estimate was improved. With
other changes made by establishing population proportionality among the
region according to size of cities, and optimizing the sample allocation
between major groups, the overall efficiency of the national index will
be improved by approximately 35 percent. However, the policy of
optimization of the area design did have an impact on the publication

18 Bimonthly indexes will continue to be published for the local
areas of Boston, Pittsburgh, Cleveland, St. Louis, Baltimore, Dallas,
Houston, Miami, Washington, D. C., and San Francisco. Semiannual
averages will be published for Buffalo, Cincinnati, Kansas City,
Milwaukee, Minneapolis-St. Paul, Atlanta, Anchorage, Denver, Honolulu,
Portland, San Diego, and Seattle.

19 These data are used in a technique described as
“bounding’ the reference period of the subsequent interview.
Bounding minimizes response errors which may result from the respondent
inadvertently duplicating purchases from an earlier period. By
recording dates and descriptions of purchases for the preceding month of
each visit, the technique can be repeated in each subsequent interview.

20 Item strata constitute the level of detail for calculating the
expenditure weights of the CPI market basket, and the qualities and
implicit quantities of this market basket are kept fixed between
revisions. That is, any change in the CPI from one month to another is
the effect of price changes of the item strata comprising the market

21 The reduction in the number of strata will affect the number of
indexes that are currently published. The BLS will, however, produce a
number of substratum (entry level items) indexes for old item strata
that are now published and that have a significant number of price
quotations collected.

22 See Walter F. Lane and John P. Sommers, “Improved Measures
of Shelter Costs,’ 1984 Proceedings, American Statistical

23 Michael P. Cohen and John P. Sommers, “Evaluation of
Methods of Composite Estimation of Cost Weights for the CPI,’ 1984
Proceedings, American Statistical Association.

24 See Daniel H. Ginsburg, “Medical care services in the
Consumer Price Index,’ Monthly Labor Review, August 1978, pp.

25 For a detailed discussion of this and other problems, see Janet
L. Norwood, Problems in Measuring Consumer Prices, Report 697 (Bureau of
Labor Statistics, 1983) and Jack E. Triplett, “Quality Bias in
Price Indexes and New Methods of Quality Measurement,’ in Zvi
Griliches, ed., Price Indexes and Quality Change (Cambridge, MA, Harvard
University Press, 1971).

26 Paul A. Armknecht, “Quality Adjustments in the CPI and
Methods to Improve It,’ 1984 Proceedings, American Statistical

Table: 1. Current and new primary sampling units, by city-size and

Table: Exhibit 1. Strata titles for revised Consumer Price Index


I'm Tamara!

Would you like to get a custom essay? How about receiving a customized one?

Check it out