1. Recent improvements and current challenges in the BLS Consumer Price Index (CPI) Program Presentation to the Council of Professional Associations on Federal Statistics (COPAFS) December 3, 2010 Mike Horrigan Associate Commissioner Office of Prices and Living Conditions 1

2. 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. 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

4. CPI MODERNIZATION EFFORT

5. 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

6. 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 1/6 th of the full rent sample 6

7. 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

8. 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

9. REDESIGN OF THE CPI ESTIMATION SYSTEM

10. New Estimation System Oldest existing CPI production system  Outdated computer environment  Rigid structures and fixed data tables limit alternative item structures and formula New system to take advantage of new computing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

11. REDESIGN OF THE CONSUMER EXPENDITURE (CE) SURVEY

12. 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 expenditures 12

13. The CE redesign process Survey Redesign Panel Data capture forum AAPOR panel on record use Data users’ forum Household survey producers workshop 13

14. The CE redesign process Methodology workshop Two independent contracts to outside survey houses to provide redesign options CNSTAT consensus panel 14

15. 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

16. The CE redesign philosophy Interview of the future Role of technology in data collection Role of administrative data 16

17. 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

18. TELEPHONE POINT-OF- PURCHASE SURVEY (TPOPS)

19. 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

20. TPOPS Outlet sample frame replacement focuses on 3 approaches  Short run – Add Cell phone households to TPOPS – Planned start summer 2011 20

21. 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

22. CPI-W, CPI-E, AND SOCIAL SECURITY

23. Social Security COLAs based on the CPI-W 23

24. CPI relative importances by population, for selected expenditure groups, December 2009 (based on 2007-2008 Consumer Expenditure Survey weights). Expenditure groupCPI-UCPI-WCPI-E All items100.00 Food and beverages14.8016.4312.35 Food at home7.808.907.16 Food away from home5.946.434.37 Alcoholic beverages1.061.090.82 Housing41.9639.7547.08 Shelter33.2930.1736.55 Rent of primary residence5.978.483.77 Owners’ equivalent rent25.2120.9631.52 Fuel oil0.190.160.28

25. CPI relative importances by population, for selected expenditure groups, December 2009 (based on 2007-2008 Consumer Expenditure Survey weights). Apparel3.703.792.65 Transportation16.6918.6514.22 Motor fuel4.535.783.36 Medical care6.515.2611.07 Medical care commodities1.611.302.95 Medical care services4.903.968.12 Recreation6.446.035.53 Education and communication 6.436.183.91 College tuition and fees1.490.960.55 Other goods and services3.483.923.19 Tobacco and smoking products 0.871.400.59

26. Annualized CPI increases using third quarter averages* CPI population 1982- 1993 1993- 2007 2007- 2008 2008- 2009 2009- 2010 1982- 2010 CPI-E 4.02.85.1-1.41.03.1 CPI-U 3.62.65.3-1.61.22.9 CPI-W 3.42.65.8-2.11.52.8 *For 1982 CPI-E, December figure is used rather than the third quarter average

27. Building Block CPI-UCPI-WCPI-E A. ScopeUrban ConsumerUrban 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. GeographyDesigned for the urban consumer 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

28. Building Block CPI-UCPI-WCPI-E C. WeightingBased 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 OutletsDesigned 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

29. Building Block CPI-UCPI-WCPI-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. RepricingItem substitution, quality change and monthly production requirements are of highest quality Same comments appliesSame comment applies.

30. ALTERNATIVE APPROACHES TO ESTIMATING CPI’S FOR DEMOGRAPHIC GROUPS – EXAMPLE: THE ELDERLY

31. 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

32. 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

33. 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

34. RESEARCH ON CORPORATE AND SCANNER DATA

35. Corporate Data 35 Product (Vendor) CollectionDisagg Weight Benchmark Checklist Evaluation CPI Sample Benchmark Biases New Indexes Homescan (Nielsen) xxxx Packaged Goods - Food & Sundries (Nielsen) xxxxx Academic database (IRI Symphony) xxxxx Random Weight – Bakery, produce, etc. (Perishables Group) xxxxx

36. Corporate Data 36 Product (Vendor) CollectionDisagg Weight Benchmark Checklist Evaluation CPI Sample Benchmark Biases New Indexes Electronics & Appliance data (NPD & screen scraping) xxx New & Used Vehicles (JD Powers) xx Women’s apparel (Retail chain) xx MEPS (AHRQ) x

37. 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

38. 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

39. 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

40. 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

41. 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

42. Biases Various authors have identified potential biases in the CPI  Small sample bias  New and disappearing goods bias  Substitution bias  Weekday/weekend bias 42

43. 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

44. MEASURING COST-OF- LIVING DIFFERENTIALS (PLACE-TO-PLACE COMPARISONS)

45. IMPROVING PRELIMINARY ESTIMATES OF THE CHAINED CPI-U

46. 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. 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. 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. 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. 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. 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. 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