Improvements to the food at home, shelter, and prescription drug indexes in the cpi
BLS WORKING PAPERS
U.S. DEPARTMENT OF LABORBureau of Labor Statistics
Improvements to the Food at Home, Shelter, andPrescription Drug Indexes in the U.S. ConsumerPrice Index
The authors wish to thank Karin Smedley, who developed most of the non-shelter index simulations presented in this paper.
We also wish to thank Steve Henderson, Walter Lane, Sylvia Leaver, and Marshall Reinsdorf for their input, but any errors or
omissions are our own. The views expressed are those of the authors and do not reflect the policies of the U.S. Bureau of
Labor Statistics or the views of other BLS staff members.
IMPROVEMENTS TO THE FOOD AT HOME, SHELTER AND
PRESCRIPTION DRUG INDEXES IN THE U.S. CONSUMER PRICE
Paul A. Armknecht, Brent R. Moulton and Kenneth J. Stewart
While the U.S. Consumer Price Index (CPI), as a Laspeyres-type index, attempts tomeasure the average change in the prices paid by urban consumers for a fixed marketbasket of goods and services, new samples for most item categories are routinelyintroduced over time to keep the CPI sample representative of consumer spendingpatterns.
This paper will describe methodological changes being implemented in January 1995 tothe food at home, shelter, and prescription drug components of the U.S. Consumer PriceIndex (CPI). These changes reflect improvements designed to make the CPI morerepresentative of price changes experienced by consumers.
Food at Home.
The U.S. Bureau of the Census conducts a Point-of-Purchase Survey(POPS) to provide the sampling frame of outlets for most non-shelter components of theU.S. CPI. To keep the outlet and item samples representative of consumer expenditures,the POPS is conducted annually in approximately 20 percent of the Primary SamplingUnits (urban areas) in the CPI and new outlets are selected to replace the old samples.
The procedures currently used by the U.S. CPI cause it to give high weight to sampleitems that are on sale during the month that a replacement sample is introduced, and lowweight to sample items that are not on sale. These procedures can cause an overstatementof price change in an urban area immediately after its sample is replaced. We discuss asample aging procedure that will be used to alleviate this problem for food-at-homeitems.
. Further investigation of the functional form bias noted for non-shelter items(Reinsdorf, 1994) has uncovered an analogous problem with the estimator used forimputation of implicit rent changes for owner occupied housing. In addition, rent dataare collected less frequently than other prices, at 6-month intervals. The current rentindexes are estimated using a composite of 6-month rents and renters' reports of 1-monthrent changes. We discuss both improvements in our strategy for imputing changes forowners' implicit rents, as well as changes in our treatment of estimating changes for rent.
. Another potential source of bias in the U.S. CPI is the treatment ofbrand name prescription drugs after their patent expires. Griliches and Cockburn (1993)present evidence that there is substantial substitution of generic drugs for brand drugswhen patents expire. The U.S. CPI does not pick up such substitution in a timelymanner. To address this situation, BLS will institute new substitution procedures for theprescription drug component of the CPI.
Improvements to the Food at Home, Shelter and Prescription Drug Indexes
in the U.S. Consumer Price Index
Paul A. Armknecht, Brent R. Moulton and Kenneth J. Stewart
A cost-of-living index (CLI) is used as a conceptual framework for dealing with practical
questions that arise in the construction of the Consumer Price Index (CPI) of the U.S.
Bureau of Labor Statistics (BLS) (BLS Handbook of Methods, 1992
). While the
Laspeyres approach used in the U.S. CPI can provide an approximation to a CLI,
measuring the average change in the prices paid by urban consumers for a fixed market
basket of goods and services has many limitations when interpreted relative to a true
cost-of-living index. For example, consumers shift spending patterns in response to
changes in relative prices, items and outlets available in the original or base period
disappear, and new items and outlets enter the marketplace. To alleviate some of these
problems, the U.S. CPI uses a modified Laspeyres approach, which allows for product
substitution and the introduction of new samples of outlets and items.1
The universe of consumer goods and services in the U.S. CPI is divided into 207
component item strata and 44 geographic areas,2 stratifying the CPI into 9108 basic item-
area components (207 item components x 44 areas). A two-tiered weighting system is
used to calculate the CPI. At the first or lowest level, within each basic item-area
component, price relatives are calculated using a weighted average of price change for
sampled items. At the second or higher level of aggregation, those price relatives are
used to update expenditures across items and geographic areas. At each level, we
periodically update the samples and/or expenditure weights to keep them representative
Within basic item-area components, the CPI has used a Point-of-Purchase Survey
(POPS) since 1978 to provide the sampling frame of outlets for most goods and services.
Item and outlet samples are scientifically selected using probability proportionate to
sales. To keep CPI samples representative of consumer purchases, the POPS survey is
conducted annually in approximately twenty percent of the Primary Sampling Units (or
urban areas) in the CPI, where new outlets are selected to replace the old samples. (BLS
At the second or higher level of weighting, base period expenditure weights for each
basic item-area component of the CPI are derived from the Consumer Expenditure (CE)
Surveys. The index (and thus the expenditure weight) of each component item-area is
effectively updated each month by its basic component price relative, and the updated
expenditure weights are aggregated across items and geographic areas. To keep these
expenditure weights representative of consumer purchases, CE weights are updated every
ten years or so. The present CPI is based on 1982-84 CE weights; in January 1998, we
are planning to introduce 1993-95 CE weights into the CPI.
The effect of consumers shifting spending patterns in response to price change at the
aggregate level in the U.S. CPI has been investigated at length. The "substitution bias"
caused by shifting purchasing patterns has been measured at between 0.1% and 0.22%
per year (Braithwait, 1980; Manser and McDonald, 1988; Aizcorbe and Jackman, 1993).
A portion of these substitution effects could be eliminated with more frequent updating
of the CPI expenditure weights (Schmidt, 1993).
The potential effects of substitutions by consumers at the lowest level of estimation
within item and area has been explored recently by Armknecht (1993), Moulton (1993),
Reinsdorf (1993 and 1994), and Reinsdorf and Moulton (1994). Much of this research
examines the assumptions of Leontief preferences in which consumers choose to
purchase fixed quantities of goods and services over time, versus Cobb-Douglas
preferences in which consumers will substitute goods and services across outlets but keep
expenditure shares constant. Leontief preferences imply no substitutions and that
consumers have demand elasticities equal to zero. With Cobb-Douglas preferences
consumers' demand elasticities are assumed to be unitary. The Leontief or fixed quantity
preferences are consistent with the Laspeyres price index for the CLI, while the Cobb-
Douglas or constant expenditure shares preferences are consistent with the geometric
mean form of the Laspeyres concept for the CLI.
While researching in this area of the appropriate functional form of the estimator for the
basic indexes or elementary aggregate, Reinsdorf (1994) has uncovered a systematic bias
with the estimator used in the U.S. CPI when new samples are introduced. Moulton
(1993) summarizes the type of estimation bias inherent in the U.S. CPI at the first or
micro-level of estimation -- the calculation of basic item-area price relatives -- caused by
the introduction of new outlet and item samples into the index. Further investigation of
the functional form bias has uncovered an analogous problem with the estimator used for
imputation of implicit rent changes for owner occupied housing.
Another potential source of bias in the U.S. CPI is the treatment of brand name
prescription drugs after their patent expires. Griliches and Cockburn (1993) present
evidence that there is substantial substitution of generic drugs for brand drugs when
patents expire. The U.S. CPI does not pick up such substitution in a timely manner. To
address this situation, BLS will institute new substitution procedures for the prescription
This paper summarizes BLS research into the causes of these biases, presents estimates
of their magnitude, and discusses actions to reduce these effects in the U.S. CPI. In
section I we discuss the sample rotation problem and alternative solutions, including the
one we are implementing. In section II we demonstrate the shelter estimation biases and
present the new estimators we plan to introduce. In section III we discuss the new
procedures that are being introduced to improve our ability to capture consumers
switching to generic drugs as patents on brand drugs expire. In section IV we present a
summary and suggestions for future research.
I. Sample rotation in the U.S. Consumer Price Index
Within most basic item-area components, expenditure estimates for each sampled item
are derived from the POPS. Retail outlets are also drawn from the POPS, and are
introduced into the index usually a year or two after the POPS survey is conducted.
Following the methodology presented by Moulton (1993) and others, the price index for
each item-area component is updated each month by a price relative, as in
Ideally, Rt, t − 1 is a Laspeyres-based estimate using the ratio of the current cost of
purchasing the base period quantities in t to the cost in the previous period t-1,
where P is the price of an item and Qb its base period quantity.
In the U.S. CPI, we do not have base period quantities (Qb); we only have estimates of
base period expenditures (EXb ). The Laspeyres-based formula can be rewritten in terms
Expenditure information is then used to draw the sample of specific items. Because the
sample is drawn after the base period, the base-period prices are not known. A proxy
base price (Pb) for each selected item is estimated by dividing the link month price of the
item by the price change of the component item-area relative from the POPS reference
where Ib is the component index value in the POPS base period and Ilink is the index
Again, after the new sample has been linked into the index, price relatives are chained
together to effectively update fixed-quantity base-period expenditures to the current
period. In the first month after the new sample is introduced then, the price relative is
where X = Ib / Ilink , the index change for that item-area from the link month back to the
POPS period. This can be further simplified to :
Reinsdorf (1993) noted that price indexes were rising at significantly faster rates than
average prices in the U.S. CPI, particularly for food items. Most of the divergence
between indexes and average prices can probably be attributed to a bias in the functional
form for estimating price change at the basic component index level, as noted in a
number of recent articles and papers (Armknecht, 1993; Moulton, 1993; Reinsdorf,
1994). Related discussions of functional form bias in basic component indexes also
appear in Dalén (1992), Szulc (1983), and Reinsdorf and Moulton (1994).
This occurs because the current method of setting base period prices in the U.S. CPI
often creates a significant positive correlation between the weight of an item priced and
its price change immediately after introducing the new sample. Forsyth and Fowler
(1981), Szulc (1989) and Dalén (1992) also note that the "bouncing" of prices from the
base expenditure period to their introduction into the sample often results in a correlation
between the weight of an item introduced into the sample and its subsequent price
change. This can occur with seasonal prices (e.g., strawberries), sale prices, as well as
when product substitutions occur (e.g., apparel). Szulc (1983) and Reinsdorf (1994) also
note that the effect is particularly severe when indexes are chained, as they are in the
As seen from the equation above, in the U.S. CPI, newly sampled items that are on sale
or have unusually low prices at sample rotation will tend to have unusually low base
period prices, and thus unusually high implicit quantity weights. These items are likely
to go off sale the next period, so that a high weight is given to items that are increasing in
price. Since the weights are inversely proportional to the starting price Plink , Reinsdorf
(1994) notes that this estimator has properties similar to that of a Sauerbeck index
immediately after the new sample is linked into the index. A Sauerbeck index is derived
from the unweighted average of price relatives.
For example, presume a set of banana prices where the average price in an area is always
$1.00, but prices within a specific outlet can vary between $0.50 to $1.50.
If a new sample of two equally weighted banana quotes in that area is introduced into
then the price relative in link month +1 would be
Even without the correlation between new weights and subsequent price change, this
estimator fails the permutation test suggested by Dalén (1992).
Alternatives for alleviating sample rotation bias in the U.S. CPI
Many proposals have been forwarded to alleviate the bias caused by the rotation of new
item and outlet samples in the U.S. CPI. If we had unlimited resources, we could expand
our surveys to get more timely information on expenditures or even actual quantities
sold. These are options that will be explored in the future for specific commodities as we
examine data available from private market research groups. In the interim, there are
three ways that have been systematically investigated in which the current bias in the CPI
sample rotation process may be alleviated:
1) using geometric means to calculate basic item-area price relatives;
2) setting base period prices using pre-link month "initiation" prices; and
3) pricing both the old and new samples for a period of time before introducing the new
Using geometric means to calculate basic item-area price relatives.
Price relatives constructed using geometric means have many desirable statistical
properties (Dalén, 1992; Turvey, 1989; Szulc, 1989) and, under certain conditions, can
be demonstrated to be superior to Laspeyres when approximating cost-of-living indexes
In Moulton (1993), most item categories within the CPI were recalculated from June
1992 through June 1993 using geometric means to calculate basic item-area price
relatives. Mathematically, the geometric mean price relative is:
where Sb is the base-period expenditure share for that item. Since items are effectively
weighted by expenditure shares, it is not necessary to estimate (or misestimate) base
period quantities. Unlike the true form of the Laspeyres concept, where base period
quantities are fixed, here relative base period expenditure shares are fixed.
For items included in the simulation by Moulton, the simulated Laspeyres arithmetic
mean CPI went up 2.95 percent; the index using geometric means rose 2.48 percent. The
differences between the two indexes were most striking for food and apparel items.3
It is the intention of the Bureau of Labor Statistics to create experimental indexes for the
U.S. CPI using geometric means in calculating basic item-area price relatives starting
sometime in 1996. These test indexes will be used to generate further research. We are
hopeful that a parallel production-quality index using geometric means could be
published on a monthly basis late in the year 2000.
In addition, the U.S. CPI is seeking additional research into the conceptual underpinnings
associated with using geometric means at the basic component level. At a recent
conference sponsored by the National Bureau of Economic Research on the economics of
new products, this approach to estimation was raised and received considerable
encouragement for future research efforts.
Using initiation prices to set base period prices.
Changes have been suggested in our current estimator to create an independence between
the price used to set a base price and the price change following the introduction of a new
sample. One potential way to reduce the correlation between the link month price and
the base-period price would be to set the base price by taking the initiation price,
observed several months before the sample is linked, and rebasing it to the POPS year.
The initiation price for a quote is typically collected about three to five months before
link month, but approximately one-sixth of quotes are initiated during link month (see
Table 1. Number of months between initiation price and link month price, drawn fromquotes in the April 1993 CPI database.
In this approach, the base period price would be defined as
where Xz = Ib / Iinit , and the price relative would become
To test the impact on the CPI of setting base period prices on initiation prices, indexes
were recalculated for most item categories using the new base period prices (see
Table 2. Percent change from May 1993-May 1994, selected item categories, current
If this change had been implemented in the CPI for the items included in the test, the all-
items CPI would have been reduced by around 0.07% per year [(2.13%-
2.02%)x64%]=0.07, where 64% is the approximate weight of the items included in the
test. About half of the effect would occur in the "Food at home" category, which has
approximately 10% of the weight in the CPI-U population [(1.59%-
Redesigning the computer system to set base prices on initiation prices, however, was not
feasible because the required resource commitment was too large. A similar approach,
discussed in the next section, was chosen because it was more cost effective.
Other proposals aimed at creating an independence between the price used to set a base-
period price and the price collected at sample introduction have not been fully
investigated at this point. These proposals include using an average price calculated for a
similar cluster of items, using an average price over the past year supplied by the
respondent, and obtaining an average price from private market research data for a
particular item. Preliminary research into these ideas has suggested they would either be
difficult operationally and, therefore, costly to implement, or may introduce other errors
Pricing both the old and new samples for a period of time before introducing the new
An alternative solution is to price overlap samples for food-at-home categories for three
months. This means that the CPI would continue to price and use in the index
calculation the old samples for food-at-home observations for three months after it had
discontinued the rest of the old sample for other items; at the same time it would price
but not use in the calculation the new sample for food-at-home observations. By doing
this, base-period prices for items in the new sample can be calculated using these earlier
prices, and are much less likely to be correlated with price change after the new sample is
linked into the index. Pricing double samples for food at home for three months is
expected to have the same effect as setting the base price on the initiation price. Although
food at home represents only about 10% of the weight of all items, it would benefit the
most by the institution of overlap samples
Mathematically, the calculation of the price relative would be similar to that for setting
The only difference is that the initiation price can typically be collected as early as Plink - 6
and as late as Plink , as seen in Table 1 above.
While we cannot make a prospective estimate, our research into setting base prices on
initiation prices indicates that the bias in the entire CPI by delaying the linking of new
samples for food at home could be reduced around 0.04% per year. We should have a
retrospective 12-month impact of this change in the middle of 1996.
We will implement this method for food at home items in those cities whose samples will
be linked in January 1995 (St. Louis, Missouri and Nashville, Tennessee). A new sample
for Chicago will be linked in February 1995.
II. Rent and owners' equivalent rent
The shelter components of the U.S. CPI also present some unusual estimation problems.
The most prominent of these is the estimation of changes in homeowners' equivalent rent,
a concept that must be estimated indirectly because the owners' implicit rents are not
directly observable. Another issue is that rent data are collected less frequently than
other prices, at 6-month intervals. This led to adoption in 1978 of a composite estimator
that combines information from the 6-month collection intervals with respondent-recall
information about rent changes during the preceding month.
One outcome of research by BLS staff into the sample rotation and chaining properties of
CPI estimators has been an examination of the chaining properties of the imputation
formula used to calculate owners' implicit rents. BLS adopted the concept of owners'
equivalent rent in 1983 as the economic concept to be measured by the price index for
owner-occupied housing (Gillingham and Lane, 1982). From 1983-1986 the index was
calculated by simply reweighting the rent sample to represent owner-occupied units.
Beginning in January 1987, BLS began estimating owners' equivalent rent using a sample
of owner-occupied units. Each owner unit was assigned an initial implicit rent based on
a rent level estimated by the field representative. Subsequently, every six months each
owner unit was matched to one or more rental units, and the implicit rent was imputed
forward by the average rent change of the matched rental units. The matching procedure
attempts to match the owner to rental units in the same neighborhood with similar
structural attributes. See BLS Handbook of Methods
(1992) for details on the matching
and imputation procedures and Lane and Sommers (1984) for a discussion of the model
underlying the matching and imputation methodology.
The formula currently used to measure average rent change of matched rental units is an
equally weighted average of renter relatives, or Sauerbeck formula. These relatives are
chained together each time an owner unit is matched, every six months. The tendency of
the Sauerbeck formula to exhibit chain drift could therefore lead to substantial
overstatement of the change in owners' implicit rent. Our research, presented below,
suggests that the overstatement relative to other formulas with better index-number
properties has been about 0.5% per year.
To understand how the chaining of imputed owner rent changes causes upward drift,
consider the following example. Suppose that an owner with an initial implicit rent of
$500 is matched to two renters, each of whom also has an initial rent of $500. The first
period, one of the renters experiences a rent increase to $600, and the other rent remains
unchanged. The next period, the first rent remains equal to $600, but the second rent
increases from $500 to $600. An appropriate implicit rent formula should have the
owner's implicit rent increase to $600, or in other words satisfy the "proportionality"
property. The current formula does not satisfy this property, however, and the owner's
implicit rent would increase to $605.
This example illustrates that the current formula causes the implicit rent (and therefore
the index) to overstate the rent change of matched rental units. The current implicit rent
formula was introduced in January 1987—since then the growth rate of the homeowners'
index has exceeded the growth rate of the renters' index by 0.9% per year.
Approximately half of the difference can be attributed to using an inappropriate
The current Sauerbeck imputation formula for the 6-month owner relative is (except for
where Pi, t is the current period pure rent of matched renter i, Pi, t − 6 is the pure rent from
6 months previous, Qj is the set of renters matched to owner j,
and nj is the number of
An alternative formula that satisfies the proportionality criterion is the ratio of sums (A):
The fact that this formula satisfies the proportionality principle is illustrated using the
earlier example, in which the ratio of sums moves the owner's implicit rent to the correct
In the simulations reported below, the age (depreciation) factor was applied to the renter
relatives, as is done in the current index, and then formulas S and A were used to
calculate implicit rents from the matched renters. Similar expressions were applied to
obtain the 1-month changes in implicit rent.
The current composite estimator for shelter was introduced with two objectives: to
improve the timeliness of the index and to reduce the variance. The form of the
composite estimator currently used for rent and homeowners' equivalent rent is
It = 0.65 × Rt, t − 1 × It − 1 + 0.35 × Rt, t − 6 × It − 6 ,
and t − 6 are the current period, previous period, and period t − 6 indexes
for the area, Rt, t − 1 is the 1-month relative, which is calculated using respondent recall of
the previous month's rent, and Rt, t − 6 is the 6-month relative, which is calculated from
comparing current rent to the rent reported during the last interview, six months
The composite estimator was intended to produce an unbiased estimate of the current
index, under the assumption that the 1-month and 6-month relatives are each unbiased
estimates of relative rent change during the corresponding interval. Research by staff of
the Price Statistical Methods Division of BLS, however, has found that the 1-month
relatives tend to underestimate rent change. One of the reasons for this is that rent
changes often occur when the tenant changes, and the new tenant may not be aware that a
rent change has taken place. The CPI uses a vacancy imputation procedure to try to
adjust for units that were vacant during the previous month, but these imputations are
necessarily imperfect (Baskin, 1994). Even among long-term tenants, however, the
evidence indicates that there is significant under reporting of 1-month rent changes
Because the 6-month relative is considered much more reliable, in general, than the 1-
month relative, the following "6-month chained" formula has been studied as an
alternative to the current composite estimator:
This index formula also eliminates the oscillating patterns that occur in the composite
formula because the 6-month relative is multiplied by the 6-months previous index. For
analysis of the properties of the composite index and various alternatives see Kosary,
Branscome, and Sommers (1984) and Jacobson (1994b). Research has shown that the
variance of the 6-month chained formula is much lower than for the composite formula,
and furthermore the validity of the measure is improved to the extent that under reporting
of 1-month rent changes is eliminated (Leaver, 1994). The only potential disadvantage is
a possible reduction in the timeliness of the indexes, since it would take an average of 3-
4 months for rent changes to appear in the index. It should be noted, however, that some
of the improved timeliness of the current composite estimator is illusory, since much of
the 1-month rent change is derived from vacancy imputations which in turn are derived
from 6-month rent changes. Since rents are largely determined by long-term contracts
and tend to move gradually, the disadvantage of reduced timeliness is not so great for
shelter as it would be for other components of the CPI.
To test these alternative formulas using actual production data, we simulated the index
calculations for owners' equivalent rent under alternative implicit rent imputation
formulas and under the composite and 6-month chained index estimation formulas over
the period from March 1992 to June 1994. These simulations use the same owner-renter
matches that were assigned when the published indexes were produced, and thus these
simulations give us accurate estimates of what the indexes would have been had the
alternatives been adopted in April 1992. The alternative implicit rents and indexes were
carried forward for each subsequent month's calculations.
Chart 1 shows the 1-month index changes for rent using the current composite estimator
and the alternative 6-month chained formula. The lower variability of the 6-month
chained formula is easily seen in the chart. Table 3 shows that the composite estimator
produces an annual growth rate for the index of about 2.1%, compared to 2.3% for the 6-
month chained estimator. This difference is most likely attributable to under reporting of
1-month rent changes. Chart 2 shows that the 6-month chained formula results in a
similar reduction in variability for homeowners' equivalent rent. Table 3 shows that
changing the imputation formula (while continuing to use the composite estimator)
reduces the annual growth rate of the owners' equivalent rent index from 3.0% to 2.5%,
which is attributable to the chain drift problem of the Sauerbeck formula. Changing
from the composite to the 6-month chained estimator increases the index growth rate
from 2.5% to 2.6%, which can be attributed to the under reporting of 1-month rent
changes. Overall, the alternative estimator is expected to reduce the growth rate of
owners' equivalent rent by 0.4% per year.
Table 3. Percent change from March 1992-June 1994, rent and homeowners' equivalentrent, current CPI versus alternatives.
estimator with Sauerbeck imputationOwners' equivalent rent: composite
estimator with ratio-of-sums imputationOwners' equivalent rent: 6-month chained
Our research has also examined two other imputation formulas for owners' implicit rents
that also satisfy the proportionality and reciprocity properties (the geometric mean and a
generalized ratio-of-means formula). The indexes derived from these two formulas
moved very similarly to the index based on the ratio of sums. On the basis of these
findings, the simpler ratio-of-sums imputation formula, along with the 6-month chained
estimator is scheduled to be implemented by the CPI shelter indexes effective with the
January 1995 CPI. Research is continuing for further improving the housing components
during the next major revision of the U.S. CPI.
Chart 1. Comparison of Monthly Changes (%) for Rent:
Current, Composite Estimator vs. 6-Month Chained Estimator
Chart 2. Comparison of Monthly Changes (%) for Owners' Equivalent Rent:
Current Composite Estimator with Sauerbeck Imputation vs. 6-Month Chained Estimator with Ratio-of-
Griliches and Cockburn (1993) note that traditional price indexes do not capture price
declines enjoyed by consumers who switch from a branded drug to its generic
"equivalent" when the branded drug loses its patent. Indeed, in the U.S. CPI, when a
branded drug under patent is sampled, we have historically continued to price that brand
even when it loses its patent, as long as it remains for sale in the outlet. Generic drugs
typically enter the CPI only through sample rotation, where any price change from the
previous sample is linked. On the rare occasions when a pharmacy has discontinued
selling the branded drug but offered the generic equivalent, we have substituted to the
generic form and shown a price decline.
Effective with the calculation of the index for January 1995, however, the U.S. CPI will
change its treatment of prescription drugs that lose patent protection. The CPI
prescription drug analyst will keep track of when all branded drugs lose patent
protection. Six months after a drug loses its patent, that particular drug will be resampled
using probability proportional to expenditures for that drug in that specific outlet.
For example, the patented brand Capoten (generically, Captopril), used in the treatment
of hypertension and congestive heart failure, will go off patent August 8, 1995. In March
1996, we will select between Capoten and other, generic versions of Captopril, based on
revenues between these. When a generic version is selected in place of Capoten, we will
reflect in the U.S. CPI the entire price difference (i.e., price decline) between the brand
drug and the "equivalent" generic. This change in CPI procedures will have the effect of
slightly slowing the rate of growth in the CPI prescription drugs component.
This method is in one respect less conservative than that laid out by Griliches and
Cockburn. While some consumers, even those who switch to generics, perceive some
quality difference between the brand and the generic, we will treat use the entire price
difference between the brand and the generic as pure price change. On the other hand, by
basing our substitution procedures for prescription drugs on revenues, there will be
somewhat less substitutions taking place in the CPI sample for prescription drugs than if
The U.S. CPI uses a Laspeyres-type estimator to measure price change in consumer
markets. There are many concerns about fixed market basket indexes being used to
measure changes in the cost of living. Some of these concerns revolve around ways in
which such indexes can incorporate new products as well as reflect changes in consumer
tastes and preferences that occur over time. The U.S. CPI approaches these problems by
holding expenditure patterns fixed at the basic index level, but allows the sample of items
and weights to vary within the elementary index. Thus, new samples of products and
services are introduced through a rotating sample scheme that updates 20 percent of the
sampled items in geographic areas on an annual basis. The base period expenditures are
escalated every month by chaining together weighted averages of monthly price changes
using the products and services that are actually being purchased by consumers.
While this approach helps resolve problems of declining samples with less representative
products and services in the market place, it has been discovered that the procedures used
to introduce the new samples are not fully congruent with the type of Laspeyres estimator
employed at the basic index level. The combination of introducing new weights and the
price oscillations in several components of the CPI contributes to a problem of chain
index drift. Previously, a number of researchers have observed this phenomenon for
indexes computed at higher levels of aggregation. This fact demonstrates the importance
of knowing the relationship between the estimator used at the micro-index level and the
weighting structure used for combining price observations.
A similar problem also occurs in the shelter component of the U.S. CPI. The use of a
Sauerbeck estimation formula (average of relative price change) for imputing the implicit
rents of owner-occupied housing causes these values to be overstated in comparison to a
more traditional Laspeyres estimator (ratio of average prices). This fact highlights the
importance of being cognizant of the statistical properties of index imputation formula as
Another issue that has become of increasing interest in calculating consumer price
indexes is the relationship of the a CPI to a true cost-of-living index (CLI). The
Laspeyres-type indexes that are predominantly used in most countries attempt to keep
quantities fixed and, thus, only allow prices to change in response to changes in relative
prices. The CLI should reflect the substitution effects of consumer responses in such
situations. Where the Laspeyres index assumes no substitution (price elasticity of
demand is zero), the geometric mean index assumes the consumer will substitute with
unitary price elasticity. This has been an area of research to which we have devoted
considerable resources. We will continue this research to determine if the geometric
mean form of the Laspeyres index is a better estimator of the components of a CLI than
In the future we will publish a number of alternative research indexes on an annual basis
so that users can see the effects of different index estimators and understand the different
assumptions about consumer behavior that underlie each. We have published both Fisher
Ideal and Tornqvist indexes (Aizcorbe and Jackman, 1993) and will update these
annually. In 1996, we plan to publish a complete set of test indexes that use the
geometric mean estimator. To date, we have only published indexes for selected
components of the CPI with the shelter component being the most prominent area
missing (Moulton, 1993 and Reinsdorf and Moulton, 1994). We will be very interested
in research conducted by other statistical agencies in this area.
The authors wish to thank Karin Smedley, who developed most of the non-shelter index simulations
presented in this paper. We also wish to thank Steve Henderson, Walter Lane, Sylvia Leaver, and
Marshall Reinsdorf for their input, but any errors or omissions are our own.
1A true cost-of-living index would also take into account the effect on consumers of such things as income taxes, changes in real income, and the consumption of non-market items.
2The 44 geographic areas of the country are represented by 88 primary sampling units (PSUs). Of these 88 PSUs, 32 are self-representing cities such as Washington, DC and St. Louis, MO. The remaining 56 PSUs are probability selected to represent the four U.S. Census regions and 3 city-size classes within each region, for a total of 12 additional geographic areas.
3Only about 70 percent of CPI items were included in Moulton's simulation. The most notable missing component was shelter, which accounts for over 25 percent of the market basket.
Aizcorbe, Ana M., and Jackman, Patrick: "The commodity substitution effect in CPI data, 1982-1991," Monthly Labor Review,
December 1993, pp. 25-33.
Armknecht, Paul A.: "Effects of changing market baskets and product substitution in the U.S. Consumer Price Index," Economic Commission for Europe/International Labour Organization Meeting of Consumer Price Indexes, Geneva, October 1993.
Armknecht, Paul A. and Ginsburg, Daniel H.: "Improvements in measuring price changes in consumer services: past, present, and future," Output Measurement in Services Sectors
, National Bureau of Economic Research, Studies in Income and Wealth Volume 56, University of Chicago Press, 1992, pp. 73-156.
Baskin, Robert: "Imputation of Vacant Units in the Residential Rent Index of the U.S.
Consumer Price Index," 1994 American Statistical Association Proceedings of the Survey Research Methods Section
Braithwait, Steven D.: "The substitution bias of the Laspeyres price index: An analysis using estimated cost-of-living indexes." American Economic Review
, March 1980, pp. 64-77.
Dalén, Jörgen: "Computing elementary aggregates in the Swedish Consumer Price Index," Journal of Official Statistics
8 (1), 1992, pp. 129-147.
Diewert, W. Erwin: "Index numbers," The New Palgrave Dictionary of Economics
, Vol. 2, The Stockton Press, New York, 1987, pp. 767-780.
Forsyth, F.G. and Fowler, R.F.: "The Theory and Practice of Chain Price Index Numbers," Journal of the Royal Statistical Society (A Series) 144(2): 224-46.
Gillingham, Robert: "A Conceptual Framework for the Revised Consumer Price Index," Proceedings of the Business and Economic Statistics Section, American Statistical Association, 1974, pp. 46-52.
Gillingham, Robert, and Lane, Walter: "Changing the treatment of shelter costs for homeowners in the CPI," Monthly Labor Review,
Vol. 105, June 1982, pp. 9-14.
Griliches, Zvi and Cockburn, Iain.: "Generics and New Goods in Pharmaceutical Price Indexes," National Bureau of Economic Research, Working Paper No. 4272, February 1993.
Jacobson, Shawn: Memorandum to CPI Housing Steering Committee, January 11, 1994 (a).
Jacobson, Shawn: "Evaluation of alternative rent estimators for the Consumer Price Index," 1994 American Statistical Association Proceedings of the Government
Statistics Section (to appear), referenced as 1994 (b).
Kosary, Carol, Branscome, James M., and Sommers, John P.: "Evaluating alternatives to the rent estimator," 1984 American Statistical Association Proceedings of the Business and Economic Statistics Section
Lane, Walter F., and Sommers, John P.: "Improved measures of shelter costs," 1984 American Statistical Association Proceedings of the Business and Economic Statistics Section
Leaver, S.G.: Internal communication to Housing Sample Design and Estimation Teams, July 6, 1994.
Manser, Marilyn E. and McDonald, Richard J: "An analysis of the substitution bias in measuring inflation, 1959-85," Econometrica
, Vol. 56, July 1988, pp. 909-930.
Moulton, Brent: "Basic components of the CPI: Estimation of price changes," Monthly Labor Review
, December 1993, pp. 13-24.
Pollak, Robert A.: "Subindexes in the cost-of-living index," International Economic Review
, February 1975, pp. 135-150.
Reinsdorf, Marshall: "The effect of outlet price differentials on the U.S. Consumer Price Index," Price Measurements and their Uses
, Vol. 57, National Bureau of Economic Research, University of Chicago Press, 1993.
Reinsdorf, Marshall: "Price dispersion, seller substitution, and the U.S. CPI," U.S. Bureau of Labor Statistics Working Paper 252, March 1994 (revised version of paper presented at Statistics Canada's Price Measurement Advisory Committee, Ottawa, Ontario, Canada, May 1993).
Reinsdorf, Marshall and Moulton, Brent: "The construction of basic components of cost of living indexes," National Bureau of Economic Research, Conference on Research in Income and Wealth - The Economics of New Products, April 1994.
Schmidt, Mary Lynn: "Effects of updating the CPI market basket," Monthly Labor Review
, December 1993, pp. 59-62.
Szulc, Bohdan J.: "Linking price index numbers," Price Level Measurement: Proceedings from a Conference Sponsored by Statistics Canada
, Ottawa, Ontario, 1983.
Szulc, Bohdan J.: "Price indices below the basic aggregation level," in Consumer Price Indexes: An ILO Manual,
International Labour Office, Geneva, 1989.
Turvey, Ralph: Consumer Price Indices: An ILO Manual
, International Labour Office, Geneva, 1989.
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index Detailed Report
, September 1994.
U.S. Department of Labor, Bureau of Labor Statistics, Handbook of Methods
, Bulletin 2414, 1992.
PATIENT MEDICAL HISTORY 1. Are you currently being treated by your physician for any medical condition? ____________________ ________________________________________________________________________________ 2. Physician’s Name _________________________________ 3 .Please list ALL medications you are currently taking: ______________________________________ ________________________________
Lowcountry Psychiatric Associates Intake Form Name______________________________ DOB____________ SSN_______-_____-_________ Emergency Contact: _________________ relation: ____________ Tel.# ______________________ Referred by _________________________ Primary Physician _____________________ What are the problem(s) you are seeking help for today? ___________________________________