Recent improvements and current challenges in the BLS Consumer Price

Index (CPI) Program

Presentation to theCouncil of Professional Associations

on Federal Statistics (COPAFS)

December 3, 2010

Mike HorriganAssociate Commissioner

Office of Prices and Living Conditions1

2

Outline CPI Modernization Effort

Housing Geography

Redesign of the CPI estimation system Redesign of the Consumer Expenditure

(CE) Survey The Telephone Point of Purchase Survey

(TPOPS)

3

Outline CPI-W and Social Security CPIs for demographic groups (elderly) Measuring cost of living differentials by

geography (place-to-place comparisons) Research on corporate and scanner data Improving preliminary estimates of the

Chained CPI-U

CPI MODERNIZATION EFFORT

CPI Modernization Goal is to move the remaining

components of periodic CPI Revision (Housing and Geographic Sampling) to a continuous process

Housing (Rent Sample) will be converted first – beginning in 2010

Geographic sampling will follow after incorporating 2010 Decennial results

5

Housing Three Phases

Augmentation (2010 – 2011)– Increase sample of renters from 35,000 to 50,000

Replacement (2012 – 2015)– Replace original 35,000 unit sample

Continuous Updating (2016 -- )– Annually replace 1/6th of the full rent sample

6

Geographic Two phases

2010 Decennial sample– Use traditional approach where a new sample is

drawn based on the decennial Census– Expectation that 67% are overlap with existing

design– Remaining 33% replace existing areas that were

not reselected – New areas are initiated into CPI over 4-6 years• TPOPS; CE; new Rent samples needed for new

areas

7

Geographic

Post 2010 Decennial– Select new geographic areas from American

Community Survey and Decennial Census– Rotate new areas into CPI design on an ongoing

basis• TPOPS; CE and Rent sample required for each

new area

8

REDESIGN OF THE CPI ESTIMATION SYSTEM

New Estimation System Oldest existing CPI production system

Outdated computer environment Rigid structures and fixed data tables limit

alternative item structures and formula New system to take advantage of new

computing environment Flexibility with respect to Item structures and

index estimation formula Support researchers and index experimentation

10

REDESIGN OF THE CONSUMER EXPENDITURE

(CE) SURVEY

Why redesign the CE? Original design of the surveys - 1980s Improvements in data collection such as

CAPI, improved diary forms Response rates a concern Surveys are burdensome Nature of how members of consuming

units spend money has changed Concerns over underreporting of

expenditures12

The CE redesign process Survey Redesign Panel

Data capture forum

AAPOR panel on record use

Data users’ forum

Household survey producers workshop13

The CE redesign process Methodology workshop

Two independent contracts to outside survey houses to provide redesign options

CNSTAT consensus panel

14

The CE redesign philosophy FY 2011 CE budget initiative

Marginal changes and thinking outside the box

Needs of the CPI Cost weights Item selection TPOPS

15

The CE redesign philosophy Interview of the future

Role of technology in data collection

Role of administrative data

16

The CE redesign philosophy Cognitive methods to reduce burden

and improve the quality of the estimates Role of incentives Proxy reporting Use of global questions and matrix methods Questionnaire order effects Structured vs conversational interviewing Recall period

17

TELEPHONE POINT-OF-PURCHASE SURVEY

(TPOPS)

TPOPS Outlet Sample drawn from Telephone

Point of Purchase Survey Response rate is declining as households are

better at screening unwanted calls Potential biased results because “Cell Phone

only” households are excluded– Latest estimates are 17-25% of all households– Younger households are even more likely to be

cell phone only

19

TPOPS Outlet sample frame replacement

focuses on 3 approaches

Short run– Add Cell phone households to TPOPS– Planned start summer 2011

20

TPOPS Long run

– Examine use of administrative and corporate data as an alternative, examples:• Census of Retail Trade• Medical Expenditure Panel Survey

– Examine alternative survey approaches to replace TPOPS:• Incorporate into Consumer Expenditure

Survey• Mixed mode surveys for targeted

demographics using mail, personal visit, and internet collection

21

CPI-W, CPI-E, AND SOCIAL SECURITY

Social Security COLAs based on the CPI-W

23

Year  Jul   Aug   Sep   Average

Percent change from 

previous peak

2007  203.700     203.199     203.889     203.5962008  216.304     215.247     214.935     215.495 5.8%2009  210.526     211.156     211.322     211.001 -2.1%2010  213.898     214.205     214.306     214.136 -0.6%

CPI relative importances by population, for selected expenditure groups, December 2009 (based on 2007-2008

Consumer Expenditure Survey weights).

Expenditure group CPI-U CPI-W CPI-EAll items 100.00 100.00 100.00 Food and beverages 14.80 16.43 12.35 Food at home 7.80 8.90 7.16 Food away from home 5.94 6.43 4.37 Alcoholic beverages 1.06 1.09 0.82 Housing 41.96 39.75 47.08 Shelter 33.29 30.17 36.55 Rent of primary residence 5.97 8.48 3.77 Owners’ equivalent rent 25.21 20.96 31.52 Fuel oil 0.19 0.16 0.28

CPI relative importances by population, for selected expenditure groups, December 2009 (based on 2007-2008

Consumer Expenditure Survey weights).

Apparel 3.70 3.79 2.65 Transportation 16.69 18.65 14.22 Motor fuel 4.53 5.78 3.36 Medical care 6.51 5.26 11.07 Medical care commodities 1.61 1.30 2.95 Medical care services 4.90 3.96 8.12 Recreation 6.44 6.03 5.53 Education and communication

6.43 6.18 3.91

College tuition and fees 1.49 0.96 0.55 Other goods and services 3.48 3.92 3.19 Tobacco and smoking products

0.87 1.40 0.59

Annualized CPI increases using third quarter averages*

CPI population

1982-1993

1993-2007

2007-2008

2008-2009

2009-2010

1982-2010

CPI-E 4.0 2.8 5.1 -1.4 1.0 3.1

CPI-U 3.6 2.6 5.3 -1.6 1.2 2.9

CPI-W 3.4 2.6 5.8 -2.1 1.5 2.8

*For 1982 CPI-E, December figure is used rather than the third quarter average

Building Block

CPI-U CPI-W CPI-E

A. Scope Urban Consumer Urban Wage Earner and Clerical Worker

Reference person or spouse age 62 plus

  Includes Older Americans / retirees / social security recipients

Does not include retirees - Includes non-social security recipients

      - Does not include surviving spouses less than 62 or their minor children

      - Does not include elderly living with families (children) where the head and spouse are less than 62

B. Geography Designed for the urban consumer

Uses geographic sample for the CPI-U

Uses geographic sample for the CPI-U

    Not designed specifically for the urban wage earner and clerical worker

Not designed specifically to reflect where Older Americans live

Building Block

CPI-U CPI-W CPI-E

C. Weighting Based on 76,000 interviews

Based on 22,500 interviews - results to higher variances

Based on 19,400 interviews - results in higher variances

      Weights on medical and shelter significantly higher; education and food and beverages lower than CPI-U population

D. Retail Outlets Designed for the urban consumer

Uses the retail outlet frame for the CPI-U; The sample is not designed to reflect where the urban wage earner and clerical worker shops

Uses the retail outlet frame for the CPI-U; The sample is not designed to reflect where Older Americans shop

Building Block

CPI-U CPI-W CPI-E

E. Market Basket

Designed to represent the market basket of purchases of the urban consumers

Designed to represent the market basket of purchases of the urban wage and clerical consumers

Designed to represent the market basket of purchases of the urban older consumers

F. Collecting the right price

Designed to collect the out of pocket expenses paid by the urban consumer, including taxes and inclusive of discounts.

Uses the prices collected for items selected to represent the purchasing patterns of the CPI-U. The items and prices may not be representative of the purchasing patterns of the urban wage and clerical worker.

For any item category (apples), the retail stores cannot report reliably the revenue they get for different types of apples they sell to urban wage earners and clerical workers.

Uses the prices collected for items selected to represent the purchasing patterns of the CPI-U. The items and prices may not be representative of the purchasing patterns of the urban older consumers

For any item category (apples), the retail stores cannot report reliably the revenue they get for different types of apples they sell to Older Americans.

Senior citizen discount rates may be far more prevalent for the Older American population than the urban population as a whole.

G. Repricing Item substitution, quality change and monthly production requirements are of highest quality

Same comments applies Same comment applies.

ALTERNATIVE APPROACHES TO

ESTIMATING CPI’S FOR DEMOGRAPHIC GROUPS – EXAMPLE: THE ELDERLY

Estimating Price Indexes for demographic groups – the case of the

elderly

Price indexes for a demographic group, for example, the elderly, require information on: Where they make consumer purchases -

TPOPS What they purchase – CE and TPOPS The ability to select items for pricing at each

outlet – disaggregation

The disaggregation process is a key step Selecting apples for pricing requires getting

information on relative sales revenue by type of apple – from the store manager

31

Estimating Price Indexes for demographic groups – the case of the

elderly

Asking a store manager to estimate revenue by type of product across all customers is difficult enough; Asking them to estimate revenue by type of

product for sales to a particular demographic group (say the elderly) is far more problematic.

Need to know which items you are selecting in advance of walking into the store (based on probability principles) with a minimum of further dissaggregation need in the outlet. 32

Estimating Price Indexes for demographic groups – the case of the

elderly

Possible role of the redesigned CE survey CE has demographics Ask for outlet information Ask for item purchase information

33

RESEARCH ON CORPORATE AND SCANNER DATA

Corporate Data

35

Product (Vendor)Collection Disagg Weight 

BenchmarkChecklist Evaluation

CPI Sample Benchmark

Biases New Indexes

Homescan (Nielsen)

x x x xPackaged Goods -Food & Sundries (Nielsen)

x x x x x

Academic database (IRI Symphony) 

x x x x x

Random Weight – Bakery, produce, etc. (Perishables Group)

x x x x x

Corporate Data

36

Product (Vendor)Collection Disagg Weight 

BenchmarkChecklist Evaluation

CPI Sample Benchmark

Biases New Indexes

Electronics & Appliance data (NPD & screen scraping) 

x x x

New & Used Vehicles (JD Powers)

x x

Women’s apparel(Retail chain) 

x xMEPS (AHRQ) x

Data Collection Vendors may be a source of alternative

data, eliminating the burden on respondents and freeing economic assistants to focus on collecting hard-to-gather prices.

37

Checklist Evaluation Disaggregation uses a checklist

designed to identify the price determining characteristics of each item. Compare the characteristics on the checklists

with the characteristics found in other data sources.

38

Disaggregation Research Disaggregation is the process of

randomly selecting the specific item whose price will be collected over time. Test the efficacy of disaggregation by examining

the distribution of items in the CPI against other the distributions found in other data sources

Test the effect of “volume seller” selection as an alternative to disaggregation

39

Cost Weight Benchmark The cost weights used in the CPI are

derived from the interview and diary sections of the Consumer Expenditure Survey. Compare the distribution of expenditures in the

CE against the distribution found in other data sources

40

CPI Sample Benchmark The CPI sample is the result of

disaggregation in the outlets using checklists designed by the Commodity Analysts. Compare the distribution of items selected for

pricing using the CPI methodology to the distribution of items reported in the other data sources.

41

Biases Various authors have identified potential

biases in the CPI Small sample bias New and disappearing goods bias Substitution bias Weekday/weekend bias

42

New Indexes New data offers the potential to explore

new formulas or create new indexes: Demographic Indexes (for food, at least) New quality adjustment methods Experimental disease based indexes

43

MEASURING COST-OF-LIVING DIFFERENTIALS

(PLACE-TO-PLACE COMPARISONS)

IMPROVING PRELIMINARY ESTIMATES OF THE

CHAINED CPI-U

46

The CPI and Substitution Bias

The CPI-U (Consumer Price Index for All Urban Consumers) is the “headline” CPI

It is a Lowe or “Modified Laspeyres” index, and does not reflect consumer substitution across item categories

BLS considers the C-CPI-U (Chained CPI-U) to be a closer approximation to a cost-of-living index

47

The Chained CPI-U Uses a superlative Törnqvist formula

requiring current spending as well as current price data

Preliminary C-CPI-U values are published monthly in “real time,” but only become final with a 1-2 year lag

In most years the preliminary indexes have underestimated the final C-CPI-U Final indexes have been closer to the

headline CPI-U than the preliminary indexes were

48

BLS Preliminary C-CPI-Us Based on a geometric mean formula BLS is evaluating three different

methods for potential use in generating the preliminary indexes

Objective is to enhance the usefulness of the C-CPI-U by reducing revisions

49

Method 1. CES Model Changes the formula used in the

preliminary indexes Headline CPI-U implicitly allows for no

substitution; elasticity =0 Overstates final C-CPI-U

Preliminary C-CPI-U indexes assume =1 Understate final C-CPI-U

Constant-elasticity-of-substitution (CES) model can be used to estimate a substitution parameter 0 < < 1

50

Method 2. Predicted Expenditures

Changes the expenditure weights used in the preliminary indexes

Headline CPI-U implicitly assumes quantities purchased in a given item category are constant over time

Preliminary C-CPI-U indexes assume that spending shares remain constant

Seasonal adjustment (X-12 ARIMA) models can be used to predict the expenditure weights

51

Method 3. Time Series Models

Uses historical data on final and first preliminary estimates

Regression and Vector Autoregressive Moving Average (VARMA) models can be used to estimate the relationships

Final index estimates can be projected given the initial preliminary starting points, using Kalman Filter methods

52

Upcoming C-CPI-U Revisions

In February 2011 BLS will release final Chained CPI indexes for 2009

The first revision of the preliminary indexes for 2010 will also be released

BLS plans to use the results of the three methods under evaluation in the determination of the preliminary indexes for 2010 and 2011