consumer expenditure survey data user needs

Acknowledgement: We thank Jennifer Edgar from the Office of Survey Methods Research for assisting us in the design, implementation, and data collection of the 2010 CE Data User Survey.

FINAL REPORT

Consumer Expenditure Survey Data Users’ Needs

Submitted: August 13, 2010 Revised: September 14, 2010 Steven Henderson, Victoria Brady, Janel Brattland, Brett Creech, Thesia Garner, & Lucilla Tan

CONTENTS I. Introduction ............................................................................................................................................... 1

II. Executive Summary of Findings about Data Concerns ............................................................................. 3

III. Methodology: identification of data applications and concerns ............................................................. 6

IV. Detailed Findings ..................................................................................................................................... 8

a. CE Data applications, concerns and future needs................................................................................. 8

a.1. Government programs and publications ....................................................................................... 9

a.2. Research ....................................................................................................................................... 18

b. Alternative data sources to the CE ..................................................................................................... 26

c. Recommendations from data users .................................................................................................... 27

V. Highlights of concerns about current CE data ........................................................................................ 28

Appendix A. Speakers at Data Users’ Needs Forum ................................................................................... 29

APPENDIX B. Tabulations & write-in responses from the 2010 CE Data User Survey ................................ 31

APPENDIX C. Notes from post-Forum wrap-up meeting ............................................................................ 49

I. Introduction The Consumer Expenditure Survey (CE) Program initiated the Gemini Project in 2009 to provide

oversight and guidance for the development of a detailed research roadmap for the redesign of the CE.

The objectives of the CE redesign are to improve data quality through a verifiable reduction in

measurement error without increasing respondent burden. A key input to this process is an

understanding of the needs of CE data users so that alternative redesign options and trade-offs between

these options may be appropriately assessed. The CE Gemini Project Steering Team chartered the CE

Data User Needs Team (the Team) to conduct a Data Users’ Needs Forum for “CE data users to present

and discuss with program office staff their current and desired data needs” and “to produce a report

detailing the current and desired needs of identified CE data users”.1

The Team charter specified the following guidelines for the CE Data User Needs final report:2

• The focus is to be on summarizing the information gathered from data users;

• The Team should use the method described in the CE Data Quality Definition Report for

developing an approach to summarize data user needs;3

• The Team is not responsible for determining the priority of data users or their data needs, or for

resolving conflicts between data needs.

In this report, we describe what we learned about the seventeen diverse uses/applications of CE data,

the features of data needed to support these applications, and the concerns that data users have about

the data as they affect their specific data application. These findings are based on the CE Data Users’

Needs Forum and a CE Data User Survey conducted in the summer of 2010. Of these seventeen data

applications, eight are used in support of government programs and publications, and the rest support a

variety of research topics undertaken by government and non-government researchers.

1 Gemini Project: User Needs Team Charter (February 19, 2010). 2 See Section V in Gemini Project: User Needs Team Charter (February 19, 2010). 3 Definition of Data Quality for the Consumer Expenditure Survey: A Proposal (2009). http://blsspapp1/OPLC/CE/Research/Gemini/Shared%20Documents/Project%20Documents/DQDProposal_FinalReport_20091022.doc

http://blsspapp1/OPLC/CE/Research/Gemini/Shared%20Documents/Project%20Documents/DQDProposal_FinalReport_20091022.doc

http://blsspapp1/OPLC/CE/Research/Gemini/Shared%20Documents/Project%20Documents/DQDProposal_FinalReport_20091022.doc

The survey source, periodicity of the data, level of geographic detail, demographic variables, and data

files used by each application are reported in detail in Table 1 (page 10) and Table 4 (page 18).

Additional data features and concerns data users have about the current data are reported in Table 2

(page 11) and Table 5 (page 20). Future needs of data users are documented in Table 3 (page 16) and

Table 6 (page 25), alternative data sources utilized by CE data users are listed in Table 7 (page 26), and

recommendations from data users appear in Table 8 (page 27).

This report documents the opinions and perspectives as they have been expressed by the data users at

the CE Data Users’ Needs Forum and the CE Data User Survey.

II. Executive Summary of Findings about Data Concerns Data concerns raised by CE data users are summarized below.4 These concerns are classified according

to the Total Quality Management and Total Survey Error framework for data quality proposed for the CE

(2009):

Relevance: the degree to which the survey products meets the user’s specific needs in terms of both content and coverage.

• Greater detail needed for expenditure items and characteristics of the sample population. The CPI is interested in having the CE collect detailed cluster-level expenditures. Other users requested additional categories, depending on data application. In addition to the currently collected Entry Level Item (ELI) information, the CPI is also interested in obtaining outlet location as a Point of Purchase data source. The US Department of Agriculture is interested in food categories such as organic or low-fat or dark green leafy vegetables. Researchers using CE medical data are interested in the health status of each member, including height and weight. The USDA is interested in learning the proportion of housing expenditures allocated to children.

• Data users indicate they will continue to need total spending by a consumer unit. Total quarterly spending information is used for shares calculations by the Department of Defense and others. Total spending across all four quarters of a calendar year is used by non-CPI researchers looking for seasonal trends. Other researchers use data from four interviews to examine spending behavior by a consumer unit over the year it participates in the survey.

• The different samples for the CE Quarterly Interview survey (CEQ) and the CE Diary Survey (CED) and the differential quality of survey estimates from these two data sources causes problems for data applications that require complete expenditures measured for a specific consumer unit (CU). This affects the Supplemental Poverty Measure (SPM) which is forced to use expenditures for food and apparel from the CEQ, even though the published source for these items is the CED. Non-federal data users have the same problem in determining total spending by consumer units.

• The different time coverage of income and assets compared to spending in the CEQ causes problems for all researchers and data users who use income data. Income for the past twelve months is collected in interview two and five in the CEQ, while expenditures for the past three months are reported in interviews two through five. Except for those CUs that complete the fifth interview, this different time period causes a mis-match in comparing quarterly spending to annual income.

Coherence: the degree to which different sources or methods on the same phenomenon are similar. • Non-federal researchers stated that major spending categories of the CE lack regular

comparisons with external data sources. These researchers seemed unaware that the CE has published comparisons regularly and will continue to do so.

• Federal and non-federal data users were concerned about the lack of comparability to other data sources, particularly to the National Income and Product Account aggregates.

4 This summary is based on Table 2 and Table 5 of this report.

Timeliness: the interval between the time data are made available to users and the event or the phenomena the data describe.

• Quarterly or biennial data – The chained CPI-U would be updated and published with a shorter time lag if the CE were released on a rolling four-quarter basis. The Bureau of Economic Analysis (BEA) could also use CE data sooner for their annual revisions of the NIPA in order to better capture movements during economic turning points in the economy.

• Annual calendar year data release - earlier release of annual data is preferred by the Census Bureau, the Internal Revenue Service (IRS), the Center for Medicaid and Medicare Services (CMMS) and the BEA. Presently the IRS needs the sales tax data in September, one month before the CE tables are currently published. The Census Bureau’s work on the SPM requires CE data by July, as does the CMMS for the National Health Expenditure Accounts.

Accessibility: the ease with which statistical information and appropriate documentation describing that information can be obtained from the statistical organization.

• Researchers want an integrated CED and CEQ microdata file. (The post-survey adjustment quality dimension below contains additional microdata suggestions.)

• Researchers wish for more comprehensive coverage of lower level geographies. They requested more regional data (the published four Census regions are too broad), State tables, and more MSA tables.

• Researchers want less restricitve topcoding for owned vehicles, state identification, and income values.

Interpretability: the availability of adequate information to allow users to properly use and interpret the survey products.

• Researchers who utilize or would like to utilize the panel feature of the CEQ request for more guidance in linking data for a consumer unit across the four interviews and in computing standard errors when data are linked this way.

Accuracy: the degree to which the estimate is similar to the true value of the population parameter. Three subcomponents of the accuracy dimension mentioned as areas of concern were measurement error, sampling error, and post survey adjustments.

Measurement error: the difference in the response value from the true value of the measurement. • The chief concern expressed at the Data Users’ Needs Forum was under-reporting of

expenditures for various categories, depending on data application. • Accurate CPI weights are critical. • Data users in Treasury, the Federal Reserve and non-federal researchers reported that savings

rates calculated using CE after-tax income minus outlays are too high; federal and state income tax data are missing.

• Researchers noted that expenditures greatly exceed income in the lower income quintiles and may be underestimated in the higher income quintiles.

• Researchers noted that CE data on participation in government assistance programs (food, shelter, and other income supplemental programs), and the value of benefits received are too low when compared with other federal data.

Sampling error: the error that results from drawing one sample instead of examining the entire target population. It also refers to the difference between the estimate and the parameter as a result of only taking one sample instead of the entire population. • Population coverage is considered inadequate for various subpopulations, depending on data

application. -The CPI needs expenditure data collected in the same geographic locations priced by the CPI, incorporating the CPI’s continuous sample rotation schedule, and needs more reported expenditures per month for the chained CPI-U. -BEA needs more rural farm households and second homes in the sample to improve the Owners’ Equivalent Rent data. The USDA also needs more farms in the sample for determining farm household well-being. -The Department of Defense, which uses three years of CE data annually for military families living off of military bases, is concerned that the CE sample contains too few military families. -The IRS notes that there are too few CUs in each state for local-level sales tax calculations. -State governments want larger sample in their states to improve the accuracy of state-level estimates. -Researchers studying poverty want a larger sample of low-income households, and researchers looking at total aggregate expenditures want a larger sample of wealthy households.

Post-survey adjustment: the extent to which survey estimates are affected by errors in adjustment procedures (e.g. weighting and imputation) that are initially designed to reduce coverage, sampling and non-response errors. • Microdata users who utilize the panel design of the CEQ need accurate longitudinal weights. • State governments and researchers need accurate state level weights in order to calculate

spending. Presently the CE population weights do not represent state populations, but only Census regions or self-representing MSA populations.

• Missing federal and state income taxes should be imputed, the NBER TaxSim program was recommended for imputing these taxes.

• Some researchers noted that CE top-coding of income is a problem, since both tails of the income distribution are important for some research applications looking at consumer behavior and needs.

III. Methodology: identification of data applications and concerns

The Team first developed a preliminary list of CE data uses (applications) based on findings from past CE

User Surveys.5 In addition, the Team reviewed data user lists maintained by the CE Branch of

Information and Analysis (BIA). Next, the Team consulted staff in BIA and the Division of Price Index

Number Research to refine the list of data applications, and to obtain the names of data users proficient

in each unique CE data application, including the speakers at the 2010 CE Public Use Microdata Users’

Workshop. This list of CE data applications and user names formed the basis for recruiting participants

for the CE Data Users’ Needs Forum, and an on-line CE Data User Survey.

1. CE Data Users’ Needs Forum

The Forum was hosted at the Bureau of Labor Statistics over two days, June 21-22, 2010. The Forum

consisted of brief presentations by a variety of data users on how they use CE data, the data attributes

they needed for specific data applications, and concerns they had about CE data. There was also time for

questions-and-answers between speaker presentations. The Forum agenda appears in Appendix A.

About 70 persons associated with government agencies, academia, and policy research organizations

attended the Forum. The majority of speakers and audience were affiliated with Federal Government

agencies. A breakdown of the affiliations of speakers and audience are provided below:

Affiliation of Participants at the 2010 CE Data Users’ Needs Forum

Participants Federal

Government State

Government Academia

Policy research

organization Total Speakers & Session Chairs 23 1 5 2 31 Audience 37 1 4 0 42 Total 60 2 9 2 73

2. CE Data User Survey

The Team also systematically collected information about CE data applications and data needs through

an on-line data user survey administered through Survey Monkey.6 The field period for the survey was

5 The two previous CE Data User Surveys were [1] Consumer Expenditure Data User Survey Results, U.S. Department of Labor, Bureau of Labor Statistics, March 1993, Report 832; [2] CE Data User’s Survey, Jennifer Edgar, 2005. 6 SurveyMonkey.com, Portland, Oregon USA.

officially from July 1 through July 16, 2010. The survey was sent out to 99 data users, of whom 29 were

affiliated with Federal Government agencies. All data users received the same set of questions, but data

users from Federal Government received a different survey cover because the team wanted to identify

their agency or department in this report. A copy of the survey appears in Appendix B. The overall

survey response rate was 41%.

IV. Detailed Findings

a. CE Data applications, concerns and future needs

In this section, descriptions are presented of the data applications, the data features used to support each

application, and the concerns that data users have about the data for their specific application. Data applications

that support the programs and publications of Federal and State Government agencies (subsection a.1) are

distinguished from data applications in support of a wider variety of research issues undertaken by researchers

from government agencies, non-government organizations, and academia (subsection a.2). The enumeration of

data applications is intended to assist the reader, as information about data applications appear across several

pages; the enumeration does not indicate a data application’s ranking in priority. Summary tabulations and

write-in responses from the Data User Survey appear in Appendix B.

As mentioned in the Background and Executive Summary sections, this report classifies the data concerns

according to the dimensions of data quality described in the data quality definition for the Consumer

Expenditure Survey (2009). The data quality dimensions are:7

Relevance: the degree to which the survey products meets the user’s specific needs in terms of both content and coverage.

Coherence: the degree to which different sources or methods on the same phenomenon are similar.

Timeliness: the interval between the time data are made available to users and the event or the phenomena the data describe.

Accessibility: the ease with which statistical information and appropriate documentation describing that information can be obtained from the statistical organization.

Interpretability: the availability of adequate information to allow users to properly use and interpret the survey products.

Accuracy: the degree to which the estimate is similar to the true value of the population parameter. The error components of accuracy that data users reported on in Tables 1 & 2 are:

Sampling error: the error that results from drawing one sample instead of examining the entire target population. It also refers to the difference between the estimate and the parameter as a result of only taking one sample instead of the entire population.

Measurement error: the difference in the response value from the true value of the measurement.

Post-survey adjustment error: the extent to which survey estimates are affected by errors in adjustment procedures (e.g. weighting and imputation) that are initially designed to reduce coverage, sampling and non-response errors.

7 Data quality dimensions not reported on by any data user were excluded in this report.

a.1. Government programs and publications Table 1. CE Data Applications used in Government Programs and Publications: Data Needs

Data Application & User

Survey Periodicity Demographics Geography Variables/Files

1. Expenditure weights and ELI selection User: CPI

CEQ, CED Monthly, Annual

CPI populations: non-institutional - urban, wage earner, elderly

Index Areas, County

CPI definition of consumption- INCLUDES all market goods & services, government-provided goods for which users pay; EXCLUDES investments (financial, real estate, owned home repairs), life insurance, illegal goods, gambling winnings & losses, fines, cash gifts, child support, alimony payments, interest payments.

2. Consumption thresholds for Supplemental Poverty Measure (SPM) User: BLS & Census Bureau

CEQ Annual, Quarterly

All CUs, Age, CU composition

National, Regional, State

Food, Clothing, Shelter, Utilities(FCSU) out-of-pocket: food and clothing bought and received as gifts, non-vacation shelter, utilities for primary residence. Census tract and county for estimating rent subsidies. Resources: pre-tax income, subsidies, transfers, in-kind benefits, member-level work-related transportation expenses, CU level work-related child care expenditures, member-level medical expenditures including insurance premiums for FCSU Summary level, e.g. food at home, clothing, vehicles

3. Optional State & Local Sales Tax tables on sales tax paid User: IRS

CEQ and CED, Special request tabulations

Annual All CUs, Income, CU size

State UCC level

4. National Health Expenditure Accounts User: CMMS

Integrated published tables

Annual All CUs, Age, Income

National, State Total Out of Pocket spending (OOP), OOP health care, Premiums (private health insurance & Medicare), Pre-tax income

5. Expenditures on Children by Families (annual report) Child Support & Foster Care Guidelines User: USDA

CEQ, Microdata Quarterly data annualized

CU composition, Age, Income

National Region Urban / rural

Child-specific expenditures: clothing, child care, education Household expenditures: housing, food, transportation, health care, miscellaneous items.

Table 1. CE Data Applications used in Government Programs and Publications: Data Needs


Survey Periodicity Demographics Geography Variables/Files

6. National Accounts - Benchmarking in NIPA & Annual growth rates User: BEA

Integrated published tables, Microdata, Special request CE tables

Annual, Quarterly

All CUs, Housing tenure

National Housing: space rent, rental income Shelter expenditures: maintenance and repair, capital improvements, lawn care, swimming pools Household expenditures on private health insurance and its components, including commercial health insurance, HMO, Medigap.

7. To assign weights to market basket for computation of cost of living allowances User: DOD

Integrated published tables, Special request tabulations (Military uniformed member)

Quarterly All CUs, Military (uniformed member)

National 120 individual goods and services across 11 major expenditure categories (e.g. transportation, food at home, clothing, etc.)

Table 2. CE Data Applications used in Government Programs and Publications: Data Concerns Part 1


Relevance Timeliness Coherence Accessibility


Outlet /point of purchase for all or a subset of expenditure categories (a high priority for CE survey improvement to CPI). Lower level of item detail than ELI (for Cluster) to reduce burden at outlets. - Gasoline expenditures by grade of fuel

Monthly data for C-CPI-U lag of 2 years; Annual data needed by September; When CPI changes definitions of Item Strata or ELI, CE to provide expenditure estimates within 1 calendar year of change.


Collecting FCSU expenditures requires: - complete expenditures on a CU (now, CED and CEQ are from 2 different samples) - inclusion of all subsidies in FCSU expenditures (e.g. WIC, SNAP, LIHEAP, Section 8, public housing) Need to match time coverage of resource with FCSU spending Identify proportion of employment-related expenses, e.g. transportation for work commute, child care expense incurred so that adult can work.

By July (Census releases thresholds in September)


By late September/mid-October. Two year lag with tax year is of concern during cyclical turning points.





Need detailed health spending by category: - hospital, physician, dentist, other professionals, durable medical equipment, - prescription drugs, over-the-counter drugs - other non-durable medical products - nursing home, home health & other

Need data in July (for their August deadline)

Percent allocation to Medicare premiums differs from the Medicare Beneficiary Survey (reference: A. Foster’s MLR article).


More child-specific items on health care & miscellaneous Proportion of housing expenditures allocated to children Child expenditures by non-household member

Food and health care expenses are lower than other surveys


Prefer data on actual transactions to imputations. Lack good estimates on average interest rates paid on different types of loans. Trip expense: want foreign/domestic split Vacation trips: want relative amount of mileage driven. Lack data on second homes More data on closing costs

Late for 1st annual revision Lack of consistency with other indicators.

Published tables: for all housing-related expenditures.





Need military-specific data – spending by military members at CU level

Need data by October

Table 2. CE Data Applications used in Government Programs and Publications: Data Concerns Part 2 Data Application & User

Interpretability Sampling error Measurement error Post survey adjustment - Longitudinal weights

Post survey adjustment - Other


CPI area and populations: insufficient sample for 1 year base period estimates; Insufficient sample at ELI-PSU level


Total income and income sources Government assistance programs: -program participation, amount of benefits/subsidies

Needed Integrate CEQ and CED for food and apparel Standard errors of thresholds with housing expenditure substituted


Change in definitions of UCCs over time.


Need larger sample – concerned about impact of small sample on data trend.

Underreporting of premiums (private health insurance)


Guidance on linking quarters of data vs. annualizing quarterly data.

Inadequate sample for State estimates

Food, Health care Needed Too much top-coding

Table 2. CE Data Applications used in Government Programs and Publications: Data Concerns Part 2 Data Application & User




Need larger sample size Difference between actual rent and homeowner’s perceived attainable rent


Sample size too small (in 2009: 252 CUs in CEQ sample)

Table 3: CE Data Applications used in Government Programs and Publications: Future Needs


Future Needs Other Comments


For demographic or household indexes, need: - Complete detailed set of expenditures from one CU (not choose between CEQ or CED) - Point of purchase - Maintain detailed demographics CE expenditure classifications should accommodate future changes to market basket structure and the introduction of new products. Rural area spending (for HICP)


Adjust FCSU thresholds for price differences across geographic areas Develop thresholds by: - housing tenure (renters, owners with mortgage, and owners without mortgage), state level, urban/rural, county level

Expenditure distributions at micro-level, and how they change over time are important.


Uses CPI for IRS indexing project. To the extent that CE weights are inaccurate, all taxpayers are affected, so the impact is nationwide. Two year lag with tax year is of concern during cyclical turning points.


Evaluation of Affordable Care Act - out-of-pocket health spending health insurance coverage & premiums


Table 3: CE Data Applications used in Government Programs and Publications: Future Needs


Future Needs Other Comments


Lag in availability of CE is of special concern during cyclical turning points in the economy


a.2. Research

Table 4. Research Use of CE Data: Data Needs

Data Application & User Survey Periodicity Demographics Geography Variables/Files 8. Assessing well-being of farm households User: USDA – ERS

Integrated published tables, CEQ

12 months for each CU

All CUs, Age, Income, Occupation (Farm households)

National, Regional, MSA

Consumption-income comparison

9. Distribution effects of consumption-based taxes (sales, cap & trade, VAT) User: Minnesota State, Utah State, CBO, Urban Institute

Integrated published tables, Special request tabulations, CEQ & CED

Annual, Multi-year

All CUs, Income, CU composition, CU size, Home ownership status, Age, Marital status, Hispanic status, Student status

National, Regional All income & wealth variables; Summary level expenditures; Detailed expenditures on items subject to special tax policy UCCs: Level of data needed to be able to construct variables that conform to all parts of the state sales tax base

10. Food spending in American households User: USDA – ERS, academic researchers

Integrated published tables, CED & CEQ

Annual, Weekly

All CUs, Income, CU size

National Food at Home & Away expenditures and demographic information

11* Analysis of consumer behavior, Economic-well-being. User: Federal Reserve, Academic researchers, Public policy think tanks

Integrated published tables, Special request tabulations, CEQ & CED

Annualized quarterly data, Biannual

All CUs, CU composition, CU size, Income, Asset ownership, Mortgage status, Age, Education, Race, Occupation and other demographic details.

National, Urban, Rural, MSA, State

Summary level expenditures. Total spending for each consumer unit across all categories. Some detailed family expenditures (e.g. kids’ clothing), luxury items (e.g. boats, swimming pools), energy, change in assets and liabilities, after-tax income, pension contributions

12. Analysis of recreational marine fishing expenditures by anglers User: NOAA

Integrated published tables, Special request tabulations

Annual All CUs National, Regional, State

Spending on vehicles averaged by region & year

Table 4. Research Use of CE Data: Data Needs

Data Application & User Survey Periodicity Demographics Geography Variables/Files 13. Estimating tax rates for National Retail Sales Tax Macro project User: Federal Reserve Board

Integrated published tables, Special request tabulations, CEQ

Annual, Quarterly

All CUs, Age, Income, Race, Hispanic/Latino subgroups

National, Regional Subcategories from summary level (e.g. children’s clothing)

14. Child-care spending, poverty measurement User: Public policy research organization)

Integrated published tables

Quarterly All CUs, Age, Income

National

15. Obesity Research User: Academic researcher

CED Weekly All CUs, Age, Income, Race, CU composition

National, Urban

16. Mortgage debt, housing User: Academic researcher

CEQ Annual, Quarterly

All CUs MSA Income; Detailed level information on debt, mortgage payments, housing expenditures

* Specific research topics mentioned by data users for data application #11 were: Consumption and income inequality; Consumer demand; Engel curves: to predict employment cyclicality by categories of goods; as instrument for quality of goods; Comparing consumption and income measures of inequality, poverty analysis, standard of living; Trends; How savings vary across groups and time; Tax evasion behavior; Does gambling crowd out savings or other expenditure? Develop socio-demographic profiles of households and their expenditures in different categories of goods and services, and examine the effects of seasonality (defined by quarter).

Table 5. Research Use of CE Data: Data Concerns Part 1



8. Assessing well-being of farm households user: USDA – ERS

Sample attrition in CEQ Matching of time period for expenditures with income and assets Need current value of vehicles (as for housing) Report disposable income components

Definition of farm characteristics in CE differs from Agricultural Resource Management Survey (ARMS)


Sample attrition in the CEQ Lack quantity data on goods subject to excise taxes (cigarettes, alcohol) Child care expenses – identify proportion incurred for work, or for other reason. CED and CEQ refer to different samples

Decline in total consumption relative to NIPA CE Consumption-Income ratio difficult to explain: • too high at low income • too low at high income Differences in estimates in some categories between CED and CEQ

Published tables: more MSA-specific, State-specific. Microdata: - link CUs across a panel for annual consumption at the CU level - integrate CEQ and CED


More detailed food expenditures, especially food at home categories (e.g. products with whole wheat grains, vegetables that are dark green & deep yellow).

Published tables: median expenditures


CED and CEQ refer to different samples – prefer all consumption measured on one CU. Match expenditures to income for time period More information on financial asset holdings (what & how much), type of pensions Include more categories related to

Prefer annual data available by September 1

Discrepancies with NIPAs - ratio of CE aggregate to NIPA aggregate steadily fallen over time. Consumption-income ratios across income distribution (especially at both extremes of income distribution) are not credible

Provide multi-year spending tabulations for Blacks and Hispanics by age, income, and education attainment. Need state of residence of CU Need more support for historical CE data. Interest




changes in net wealth. Declining coverage of expenditure items Current income and labor market information (e.g. current work status) lacking - not asked every wave in CEQ. Inadequate coverage of small expenditure categories (e.g. gambling, games of chance)

Income tax payments from TAXSIM very different from reported taxes in CE CE data lack external validation – CE aggregates do not match aggregates from Census Retail Sales (“a well measured external source”) in corresponding categories. - Aside from housing, CE wealth data do not match Survey of Consumer Finances. - Income from benefits programs not comparable to CPS and other surveys

in long term trends – need good documentation on changes in variables definitions (breaks in series) Need integrated CEQ and CED microdata

12. Analysis of recreational marine fishing expenditures by anglers User: NOAA

Vehicles –need type of model Need data by July Published tables: vehicle expenditures by state

13. Estimating tax rates for National Retail Sales Tax Macro project User: Federal Reserve Board

Need more detail on specific insurance products (e.g. funeral, life, disability)

Need earlier data release – quarterly, biannual

Published tables: by gender

14. Child-care spending, poverty measurement User: Public policy research organization

Be able to identify working parents with children under age 13. Child-care expenses: identify if child care was for parent employment/ other reason; distinction needed for “NAS-type” poverty measurement.

Published tables: for family with working parents and children under age 13; - state-level




15. Obesity Research User: Academic researcher

Demographics: need height and weight Quantity of (food) purchase is needed

16. Mortgage debt, housing User: Academic researcher

Need more detail on consumer credit, mortgage debt and repayment expenditures, cost on use of credit

* Specific research topics mentioned by data users for data application #11 were: Consumption and income inequality; Consumer demand; Engel curves: to predict employment cyclicality by categories of goods; as instrument for quality of goods; Comparing consumption and income measures of inequality, poverty analysis, standard of living; Trends; How savings vary across groups and time; Tax evasion behavior; Does gambling crowd out savings or other expenditure? Develop socio-demographic profiles of households and their expenditures in different categories of goods and services, and examine the effects of seasonality (defined by quarter).





8. Assessing well-being of farm households User: USDA – ERS

Guidance on computation of SEs when linking CUs across quarterly data.

Needed Standard errors of annual estimates in published tables assume independence across quarterly observations


Provide cross-walk or adjustment factors between NIPA PCE and CE’s UCC codes

Sample size of low and high income groups too small Sample size at state level is inadequate for state level estimates.

Differential reporting error across income groups is a concern

Needed Top-coding of income is a problem – need tails of income distribution


Under-reporting of Supplemental Nutrition Assistance Program recipients and benefits Volatility of food at home expenditures relative to SEs from year to year Low income CUs spend more than income






Would like to know why a CU leaves the survey before final interview

Differences between CED and CEQ estimates

Sample size too small for poverty analysis

Increasing degree of underreporting in expenditures Income, assets and liabilities; Changes in credit are poorly tracked Mis-measurement of consumption aggregates. Accurate aggregates more important than item detail for macroeconomic analysis. Lack understanding of source of measurement error, how it is changing over time, and if these errors are correlated with respondent characteristics

Needed Need weights for state level analysis

Note: No other comments were made on “Data Concerns Part 2” aside from those reported in Table 5. * Specific research topics mentioned by data users for data application #11 were: Consumption and income inequality; Consumer demand; Engel curves: to predict employment cyclicality by categories of goods; as instrument for quality of goods; Comparing consumption and income measures of inequality, poverty analysis, standard of living; Trends; How savings vary across groups and time; Tax evasion behavior; Does gambling crowd out savings or other expenditure? Develop socio-demographic profiles of households and their expenditures in different categories of goods and services, and examine the effects of seasonality (defined by quarter).

Table 6. Research Use of CE Data: Future Needs and other comments

Data Application & User Future Needs Other Comments 8. Assessing well-being of farm households user: USDA – ERS

Good data on rural household expenditures as input to constructing price index for rural households. -annual measure of expenditures by major categories for farmers. Annual disposable income by income type

Distributions are important


State level estimates Internal inconsistency: the reconciliation of income and total consumption across the income distribution is not plausible, suggesting both income and consumption are not measured well

10. Food spending in American households User: USDA – ERS, academic researchers)

Include more food commodity detail (e.g. price, nutrition)


Internal inconsistency - lack of balance in expenditures- income at CU level. Although this may be difficult to implement, the principle is still important to keep in mind for improving data accuracy Concerned about erratic changes in assets and liabilities As the service sector grows in the U.S. economy, data on service expenditures has become very useful.

Note: No other comments were made on “Future Needs & Other Comments” aside from those reported in Table 6. * Specific research topics mentioned by data users for data application #11 were - Consumption and income inequality; Consumer demand; Engel curves: to predict employment cyclicality by categories of goods; as instrument for quality of goods; Comparing consumption and income measures of inequality, poverty analysis, standard of living; Trends; How savings vary across groups and time; Tax evasion behavior; Does gambling crowd out savings or other expenditure? Develop socio-demographic profiles of households and their expenditures in different categories of goods and services, and examine the effects of seasonality (defined by quarter).

b. Alternative data sources to the CE Participants at the Data Users’ Needs Forum and the CE Data User Survey also mentioned other data sources they used to complement or to substitute for the CE Survey. These data sources are listed in Table 7. Table 7. Alternative data sources to the CE Survey

Data Source Used in addition to CE

Used as an alternative to CE

Aggregate excise tax collection data American Community Survey American Housing Survey Army & Air Force Exchange data British Household Panel Census Retail Sales Child Care Expenses from Other Surveys County Business Patterns Current Population Survey Defense Manpower Data Center data Dept of Defense Commissary & Exchange data Education Dept Surveys Health expenditure databases Health and Retirement Study Internal Revenue Service Management Survey Data Medical Expenditure Panel Survey - Household Component Medical Expenditure Panel Survey - Insurance Component Medicare Current Beneficiary Survey National Association of Home Builders National Association of Realtors National Household Travel Survey National Income and Product Accounts National Marine Manufacturer Association National Health and Nutrition Examination Survey Navy Exchange data Nielson Homescan Consumer Panel National Survey of America's Families Panel Study on Income Dynamics Personal Consumption Expenditures Proprietary Market Research Data Service Annual Survey - Census Bureau Survey of Consumer Finances Survey of Income and Program Participation USDA Food Plans USDA's Agricultural Resource Management Survey

c. Recommendations from data users Participants at the Data User Forum and the CE Data User Survey offered recommendations for addressing some of the data issues that were raised. These recommendations are summarized in Table 8. Table 8. Recommendations from data users

Data Issue Recommendation

External validation Need to undertake regular external validation, and document findings from comparisons for data users

Income tax payments Use the TAXSIM software by NBER

Internal consistency of spending and wealth at CU-level

Try “top-down” approach to structuring interview, asking aggregate expenditures first and compare total of expenditure aggregates to reported income and assets Match Haig-Simons wealth measure to [income – expenditure]

Assessment of whether spending is more or less than income

Add selected attitude and expectation questions (as in Survey of Consumer Finances), e.g., is income higher or lower than usual, retirement, health assessment Have occasional survey supplement, like the CPS

Expenditures on Children by Families annual report - for determining proportion of housing expenditures allocated to children

Ask parents / guardians to provide an estimate of this proportion

Need complete measures of expenditures on one CU, but differential quality in expenditure estimates between CEQ and CED

Create public use file that integrates best data from each survey using imputation

Occupation codes Use Census 3-digit occupation codes

CU composition Follow Census household typology more closely, so that sum of each type of CU adds up to the total number of CUs

Respondent burden Use administrative records to supplement CE survey Explore electronic record-keeping by respondents

V. Highlights of concerns about current CE data The previous section described the data needs, data user concerns about the current CE data, and anticipated

future data needs for seventeen CE data applications. These were based on information learned from the CE

Data Users’ Needs Forum and the CE Data User Survey conducted in 2010. Of these seventeen data

applications, eight are used in support of government programs and publications; the rest support a variety of

research topics undertaken by government and non-government researchers.

While details of data concerns associated with each of the data applications appear in Tables 1 through 6, the

following highlights those concerns common to three or more data applications:

Relevance • Lower level expenditure item needed – various categories, depending on data application • Different samples for the CEQ and CED and the differential quality of survey estimates from these two

data sources are problematic for data applications that require complete expenditures for a CU • The different time coverage of income and assets compared to spending in the CEQ

Coherence • Major spending categories of the CE lack regular comparisons with external data sources • Lack of comparability to NIPA aggregates

Timeliness • Earlier annual release of annual data would be preferred. The SPM requires data by July, which would

meet the timing need of all other data applications.

Sampling error • Population coverage inadequate – various subpopulations, depending on data application

Accessibility • Integrate CED and CEQ for microdata file • Lower level geographic estimates - Census regions too high-level, more State and MSA tables needed

Measurement error • Under-reporting of expenditures – various, depending on data application • Participation in government assistance programs (food, shelter, and other income supplemental

programs), and the value of benefits received underestimated Post-survey adjustment • Longitudinal weights are needed for the CEQ • Top-coding of income is a problem since tails of income distribution are important for some research

applications

Appendix A. Speakers at Data Users’ Needs Forum Monday, June 21 — Meeting Government Data Needs

Welcome & Introductions

Steven Henderson, Chair, Data Users’ Needs Forum Team

Keith Hall, Commissioner, Bureau of Labor Statistics

User Needs in the CE Redesign

Michael Horrigan, Associate Commissioner, Office of Prices and Living Conditions

Jay Ryan, Chief, Division of Consumer Expenditure Surveys

CPI Needs

William Casey, Economist, Index Cost Weights and Special Projects Section, Division of Consumer Prices and Price Indexes8

Session: Aggregate Statistics Chair: Bill Passero, Division of Consumer Expenditure Surveys

Cathy A. Cowan, Economist, Office of the Actuary/National Health Statistics Group, Center for Medicare and Medicaid Services

Arnold J. Katz, Economist, National Economic Accounts Research Group, Bureau of Economic Analysis

Wu-Lang Lee, Economist, Research, Analysis, and Statistics, Internal Revenue Service

Session: Detailed Data Needs I Chair: Laura Paszkiewicz, Division of Consumer Expenditure Surveys

Mark Lino, Economist, Center for Nutrition Policy and Promotion, USDA

Kathleen S. Short, Economist, Bureau of the Census

Carol Adaire Jones, Senior Economist, Economic Research Service, USDA

Session: Detailed Data Needs II Chair: Rob McClelland, Chief, Division of Price and Index Number Research

Ed Harris, Analyst, Tax Analysis Division, Congressional Budget Office

Phillip Anthony, Research Analyst, Division of Tax Research, Minnesota Department of Revenue

Kenneth Hanson, Senior Economist, Economic Research Service, USDA

8 For the paper, CPI Requirements for CE, William Casey (2010), see: \\psbres2\dces-public\Teams\Gemini\5 User Needs\CPI\CPIRequirementsforCE.pdf

Meeting Government Data Needs

Jesse Rothstein, Chief Economist, Department of Labor

Tuesday, June 22 — Macroeconomic Statistics, Economic Well-Being, and Analysis of Consumer Behavior

Welcome to Day 2

Jay Ryan, Chief, Division of Consumer Expenditure Surveys

Session: Macroeconomic Statistics Chair: Barry Bosworth, Senior Fellow, The Brookings Institution

Christopher D. Carroll, John Hopkins University

Thomas Crossley, University of Cambridge [invited, but did not attend ]

John Sabelhaus, University of Maryland

Mark Bils, University of Rochester

Session: Economic Well-being Chair: Thesia Garner, Division of Price and Index Number Research [Mina Kim substituted for Thesia]

Jared Bernstein, Chief Economist and Economic Policy Adviser to the Vice President

Geng Li, Economist, Division of Research and Statistics, Federal Reserve Board

Session: Analysis of Consumer Behavior Chair: Geoffrey Paulin, Division of Consumer Expenditure Surveys

Bruce Meyer, Northwestern University

Janet Wagner, Director, Center for Excellence in Service, University of Maryland

Sherman Hanna, Consumer Sciences Department, Ohio State University

Panel Discussion Moderator: Jay Ryan, Chief, Division of Consumer Expenditure Surveys

Brent Moulton, Associate Director, National Economic Accounts, Bureau of Economic Analysis

Laura Wheaton, Senior Research Associate, The Urban Institute

John Eltinge, Associate Commissioner, Office of Survey Methods Research, Bureau of Labor Statistics

NOTE: Speaker presentations made at the Forum can be found in this folder: \\psbres2\dces-public\Teams\Gemini\5 User Needs\DUFpresentations\

APPENDIX B. Tabulations & write-in responses from the 2010 CE Data User Survey NOTE: The survey responses reported in Appendix B reflect all responses received as of August 2010; write-in responses shown are “as is” reported in the survey. SURVEY COVER PAGE9 The Consumer Expenditure Survey program (CE) has embarked on a major survey redesign to improve the quality of the CE data. Understanding your data needs is a key input to this process. Information collected through this survey will assist the CE in evaluating alternative redesign options. Please contact Steven Henderson at 202-691-5124 or [email protected] if you have any questions. Your participation in this survey is voluntary. We estimate that it will take you 15 minutes to complete this survey. We are collecting this information under OMB Number 1225-0059. Without this currently-approved number, we could not conduct this survey. (Expiration: November 30, 2012)

Please note that this survey is being administered by SurveyMonkey.com and resides on a server outside of the BLS domain. The BLS cannot guarantee the protection of survey responses and advises against the inclusion of sensitive personal information in any response. We will not be using your identifying information in our report. (Wording for non-federal users) Please note that this survey is being administered by SurveyMonkey.com and resides on a server outside of the BLS domain. The BLS cannot guarantee the protection of survey responses and advises against the inclusion of sensitive personal information in any response. We will be using your agency information in our report. (Wording for federal users) Please answer each question: • For questions where more than one option applies to your work, please mark all that apply. • We understand that not all the questions may apply to your circumstances. Please mark “not applicable” for any question that doesn’t apply to you. Please complete your survey by July 16. Summary tabulations and write-in responses to survey questions appear below.

9 While all survey recipients received the same questions about the CE data, Federal and non-Federal government survey recipients received a different survey cover page and options for organization affiliation.

Q1. For what topics do you use the CE data? For example, studying the impact of fiscal changes, inequality of expenditures, poverty analysis, and consumer demand.

Number Non-FED responses 1 Obesity research 2 Consumer demand and saving 3 Analysis of incidence of state and local sales taxes 4 Saving

Consumption Tax Incidence Inequality

5 Lifecycle consumption decisions, consumption behavior at retirement and differences in consumption across racial groups

6 Examining changes in household net wealth, studying overspending, examining household living standards (i.e. types and levels of consumption), studying consumption patterns generally

7 I'm not a user of the CE data ---not sure why this was sent to me 8 Consumption inequality, Consumer demand 9 Mortgage debt, housing consumption

10 Food away from home (expenditure and frequency) and expenditure of prepared meals. 11 Inequality of expenditures, poverty analysis, consumer demand 12 Poverty analysis 13 Inequality in expenditures and its change over time and its correlation with macroeconomic

Variables; the impact of changes in taxes on expenditures 14 State tax analysis, general economic data by household 15 Inequality, consumer demand 16 Studying consumer spending and saving decisions. 17 consumer demand 18 Various uses; most recently, have tried to look at the data on child care spending 19 Consumption and excise tax, poverty analysis 20 Uses the data for segmentation purposes. Develops socioeconomic and demographic

profiles of households (CUs) and their expenditures if different categories of goods and services.

21 For market segmentation purposes. What is the effect of various socioeconomic and demographic variables on consumer expenditures in various categories of goods and services? What is the income elasticity? How does seasonality moderate the effects of CU characteristics on expenditures?

Number FED responses

1 Incidence of excise taxes, distributional analysis of consumption taxes, measures of inequality

2 Estimating tax rates for a National Retail Sales Tax Macro Project; and Consumer Demand 3 Poverty analysis, expenditure analysis 4 Analysis of recreational marine fishing expenditures by anglers 5 consumer expenditures on food at home and away from home 6 Expenditure patterns, and poverty analysis 7 Estimating family expenditures on children, which are used to set child support and foster

care guidelines. 8 national accounts 9 Sales tax amounts paid

10 poverty analysis

11 consumer demand 12 Out-of-pocket health costs, individually purchased premiums 13 household expenditures on private health insurance and its components--including

commercial health insurance, Blue Cross Blue Shield, HMOs, Medigap and other health insurance

14 Distributional effects of consumption-based taxes (excise taxes, energy taxes, sales/value-added taxes

15 Household consumption smoothing liquidity constraints Data quality comparison tax evasion gambling behavior Engel curve

16 studying household wellbeing based on consumption measures estimated from the expenditure data

17 consumption tax analysis microsimulation tax models

18 Assign relative weights (importance) in a market basket of approximately 120 goods and services for cost of living allowances.

Q2 What data do you use? (Mark all that apply)

Response options Non Fed Fed Total Integrated aggregate tables published by the CE 8 12 20 Weekly Diary Survey only 5 4 9 Quarterly Interview Survey only 15 8 23 Special tabulations prepared by the CE 2 7 9 Selected microdata variables 7 6 13 Other (please specify) 5 2 7 answered question 20 17 37 skipped question 0 0 0

Other –specify responses

Number Non-FED 1 EXPN files, FMLY files, MTAB files 2 none of the above 3 Interview and Diary data 4 None. None are good enough to use. 5 in the past, used the microdata -- haven't used it recently so can't answer

those questions Number FED

1 Spending on Vehicles - new and used - averages by year and region 2 Military (Uniformed Member) tabulations of consumer expenditures

Q3. Which microdata variables do you use?

Non-FED FED No. responded 5 6 No. skipped question 17 12

Variable descriptions provided by respondents

Number Description from Non-FED 1 change in assets and liabilities, after-tax income, total expenditures, pension

contributions, and most of the characteristics variables 2 I use several different ucc codes. 3 Most--MTAB, ITAB, FMLY detailed 4 income, debt, mortgage payments, housing payments, etc 5 microdata variables on the quarterly interview files to calculate income under various

definitions and consumption, generally aggregated to 17-20 types Number Description from FED

1 Too many to enumerate 2 Most of the expenditure type ones in order to analyze what share of family budgets go to

different areas. 3 Energy expenditure variables (electricity, natural gas, etc.)

Family size Household income State (would be more useful if this was weighted)

4 Those related to children's clothing (coats, footwear, dresses, etc.), child care (own home, other home, and day care), and children's education (tuition, books, fees, and supplies). Also, we subtract from housing expenses: care for older adults, and from other expenditures: pet, lottery, funeral, dating service, and occupational expenses.

5 space rent 6 All income variables

Disaggregated spending variables Wealth variables Demographics of the consumer unit (age, sex, family relationships, education)

Q4. What level of geographic detail do you use? (Mark all that apply)

Response Options Non-FED FED Total National (urban and rural together) 20 17 37 Urban only 3 2 5 Rural only 2 2 4 Regional 4 8 12 State 4 5 9 Metropolitan Statistical Area 2 1 3 Other (please specify) 2 2 4 answered question 19 17 36 skipped question 1 0 1

Other-specify responses Number Non-FED

1 occasionally, could use regional or state breakdowns if available

2 Use both urban and rural together. Run regression analysis and creates a dummy variable for one category and compare the results to the other

Number FED

1 Would love to be able to use weighted state data! 2 broken out by tenancy

Q5. What level of demographic detail do you use? (Mark all that apply)

Response Options NONFED FED Total National (All Consumer Units) 17 18 35 Subset by age 15 9 24 Subset by income 14 11 25 Subset by race 9 2 11 Other (please specify) 10 7 17 answered question 19 17 36 skipped question 1 0 1

Other-specify responses

Number Non-FED

1 household type 2 family type, earner composition, education level, 3 Subset by wealth holdings, mortgage status, occupation, industry, education 4 household size 5 Educational category 6 Subset by asset ownership.

7 subsets by educational attainment, income quintiles, occupation, race & ethnicity 8 for child care analysis, only makes sense if data can also be subset to just look at families

with children <13; and ideally, just families with working parents and kids ,13 9 marital status, home ownership, hispanic status, student status

10 Run regression analysis, uses detailed information on the demographic features of CUs.

Number FED

1 Special CE tabulations by Hispanic/Latino sub-group 2 family size 3 family composition 4 Family size 5 Subset by family type (married/single, with/without children) 6 Subset by family size, Subset by total consumption percentile 7 Military (Uniformed members)

Q6. What time period of data do you use? (Mark all that apply)

Response options Non Fed Fed Total Annual 14 15 29 Biannual 1 0 1 Quarterly 10 8 18 Weekly 1 1 2 Other (please specify) 4 0 4 answered question 19 17 36 skipped question 1 0 1

Other – specify responses

Number Non-FED 1 combination of several years of micro data 2 monthly 3 No data frequency is reliable. 4 when I used microdata in the past, we combined quarters to get an annual picture, but had

issues with weighting

Q7. If you use the Interview Survey microdata from more than one quarter, do you treat each consumer unit‘s quarterly information as if it is independent from other quarters of data?

Response options Non Fed Fed Total Yes 3 7 10 No 14 5 19 Not applicable 4 5 9 answered question 19 16 35 skipped question 1 1 2

Q8. Do you need a full year’s worth (four quarters) of data for each consumer unit?

Response options Non Fed Fed Total Yes 17 16 33 No 4 2 6 answered question 19 17 36 skipped question 1 0 1

Q9. What level of detail do you need on expenditures?

Response options Non Fed Fed Total Not applicable 0 1 1 Total expenditures 2 0 2 Summary levels such as “food at home,” "clothing," "vehicles" 11 6 17 More detailed levels such as “beef” or “poultry,” "men's shirts," "tires" (please specify) 7 11 18 answered question 18 17 35 skipped question 2 0 2

More detailed descriptions provided by respondents

Number Non-FED 1 individual food categories 2 data is needed to have the ability to construct variables that conform to all parts of

the states sales tax base 3 I require both summary level and particular expenditures. I look at family

expenditures in detail (e.g. exp. for children's clothing) and also "luxury" or "status" purchases (e.g. boats, swimming pools, etc.)

4 as noted above, housing and mortgage related items 5 iuse almost every level of detail that you provide 6 Specifically, for child care expenses -- need to know if the child care expenses are for

purposes of parent employment or education; this affects whether or not the expenses are relevant for "NAS-type" poverty measurement

7 Summary level + detail on items currently subject or likely subject to special tax policy

such as excise tax Number FED

1 sub-categories from summary levels, such as 'children's clothing,' etc 2 new and used vehicles, different vehicle types 3 we would like more detail on food at home categories 4 Summary level data for most categories, but we break up some variables (i.e.

education, pettoy ...) 5 Need child-specific expenditures (clothing, child care, and education). Also, need to

subtract expenses for care for older adults from housing expenses and expenses on pets, funerals, dating services, lotteries, and occupational services from other expenses.

6 specific expenditures like maintenance& repair, capital improvements, lawn care, swimming pools

7 At UCC levels/details. 8 Health premiums, health spending by category - hospital, physician, dentists, other

professionals, dme, prescription drugs, over the counter drugs, other non-durable medical products, nursing home, home health and other

9 commercial health insurance, BCBS, HMOs, Medigap, and toerh health insurance 10 Very detailed, we match the UCCs to PCE categories in BEA Table 2.4.5u 11 Price over 120 individual goods and services across 11 major categories, such as

transportation, food at home, clothing, etc.

Q10. In your work, do you use alternative sources of data on consumer spending, either in addition to or instead of, CE data?

Response options Non-FED FED Total Yes 15 14 29 No 6 4 10 answered question 19 17 36 skipped question 1 0 1

Q11. What other data sources do you use IN ADDITION to CE data? (List data sources) Response options Non-FED FED Total answered question 14 14 28 skipped question 8 4 12

Descriptions of additional data sources provided by respondents

Number Non-FED 1 national accounts, PSID, SCF, and HRS 2 PCE 3 SCF, SIPP etc for wealth change 4 PSID, data from scanner agencies (Nielsen), British Household Panel 5 Survey of Consumer Finances, Health and Retirement Study 6 PSID

7 American Housing Survey BEA data, county business patterns Current Population Survey

8 PSID 9 NIPAs, PSID

10 Nielsen Homescan Consumer Panel, PSID, SCF, IRS data 11 Census Retail Sales

PSID spending measures 12 National aggregates (PCE, retail sales); Panel Study on Income Dynamics; Survey of

Consumer Finances. 13 information on child care expenses from other surveys: SIPP, education dept

surveys, NSAF 14 NIPA PCE

Number FED

1 Information from National Income and Product Accounts along with other surveys for items such as durable sales, health care expenses

2 Survey of Consumer Finances ACS CPS SIPP Proprietary Market Research data

3 CPS, ACS, SCF 4 Data from the national marine manufacturer association, that provides spending

on recreational boats and boat related items. Data on housing prices and sales from the National Association of Realtors

5 Nielson homescan, BEA PCE, NHANES 6 Medical Expenditure Panel Survey

USDA Food Plans National Household Travel Survey

7 some Census, trade group (NAR, NAHB) 8 SIPP

MEPS AHS

9 Service Annual Survey - Census Bureau Medicare Current Beneficiary Survey Medical Expenditure Panel Survey - Insurance Component Medical Expenditure Panel Survey - Household Component

10 NIPA PCE aggregates 11 Panel Study of Income Dynamics

Survey of Consumer Finances 12 for the farm household population, I use USDA's Agricultural Resource

Management Survey data. 13 Aggregate excise tax collection data

Health expenditure databases American Housing Survey

14 Dept of Defense Commissary and Exchange data; Army & Air Force Exchange data, Navy Exchange Data, and Defense Manpower Data Center data

Alternative data sources listed by respondents

Number Non-FED 1 NA 2 none 3 SCF-CAMS 4 American Housing Survey 5 Nielsen Homescan Consumer Panel, PSID, SCF, IRS data 6 Survey of Consumer Finances

PSID SIPP

Number FED

1 We do not use other data INSTEAD of CE but IN ADDITION to the CE, we use: statistics of income data, survey of consumer finances, survey of income and program participation, NIPA data, current population survey, medical expenditure data, federal program data, etc.

2 Nielson homescan, BEA PCE, NHANES 3 SIPP

AHS 4 Service Annual Survey - Census Bureau

Medicare Current Beneficiary Survey Medical Expenditure Panel Survey - Insurance Component Medical Expenditure Panel Survey - Household Component

5 Survey of Consumer Finances to estimate saving/wealth by population subgroup

Q13. Do you have concerns about the accuracy of the CE data?

Response options Non-FED FED Total No 5 4 9 Yes. Please explain your concerns 16 14 30 answered question 19 17 36 skipped question 1 0 1

Concerns about accuracy of CE data listed by respondents

Number Non-FED 1 Quantity is not measured, so expenditures are not very revealing/accurate. 2 There is a growing degree of underreporting in the expenditure data, and the changes

in assets and liabilities appear to be very erratic.

Q12. What other data sources do you use INSTEAD OF CE data? (List data sources) Response options Non-FED FED Total answered question 6 5 11 skipped question 16 13 29

3 There appears to be trouble correlating income to spending at low levels of income. 4 Underreporting, unexplained consumption/income ratios 5 The declining coverage of CE data is very worrisome when trying to think about

measuring consumption inequality. Also, the lack of labor market variables (current work status for example) make the CE ill suited to study important questions pertaining to consumption smoothing during recessions and how that has changed over time.

6 Concern about accuracy of total expenditures for individual households, plus other resources for low income households

7 Because my primary interest is in changes in net wealth, the CEX is somewhat untested in its ability to track wealth. Results (including my own) indicate that it does surprisingly well, but more data on wealth would permit a better comparison to more established wealth tracking surveys like the SCF.

8 Lack of representativeness vis-a-vis NIPA 9 Differences between Diary and Interview figures, differences with NIPA aggregates

10 Discrepancies with NIPAs 11 The detail required of households seems to come at the cost of complete measurement

of total and subgroups. Income information seems poor; asset, saving, measures very limited and poor. The increasing difficulty of getting participation in surveys and the correlation of participation/data quality with the scarcity of time/resources is a serious concern. Suggestions: use the budget constraint to help households get total spending correct, ask only a little detail from each household, and bring in as much administrative data as possible.

12 I'm not confident that all income of lower-income households is reported 13 The aggregates do not match aggregates from the Census retail sales in corresponding

categories. How can anyone have confidence about conclusions if they cannot be validated by a (well measured) external source? If the averages are meaningless, how can anyone believe the subaggregates?

14 There is a high degree of measurement error. Furthermore, we don't know the source of the error (i.e. the degree to which it reflects a non-representative samples vs. biased or income responses of the participants). And, we don't know if it is changing over time or whether it is correlated with the characteristics of the participants.

15 I always have data accuracy concerns, but compensate by being very cautious about drawing any conclusions from the data or about describing any changes from year to year

16 I am concerned by decline in total consumption relative to NIPA, unexpectedly high consumption/income ratios at lower income levels and unexpectedly low consumption/income ratios at higher income levels, and the effects of sample attrition when trying to construct annual estimates.

Number FED

1 It is difficult to reconcile spending by category as compared with NIPA data. If the NIPA is right, some items seem more underreported in the CE than in the PCE. In addition the reconciliation of income and total consumption across the income distribution does not seem plausible suggesting that both income and consumption is not being measured well.

2 It would be great to have data that would be representative at the state level. 3 the year to year change in food at home expenditures tends to jump around more than

the standard error suggesting a concern about accuracy. 4 Data on income from benefits program does not seem comparable to similar data in

CPS and other surveys 5 Food and health care expenses are lower than similar expenses in other surveys. Also,

concerned whether publically released state-level data are an accurate representation of state population.

6 small sample sizes on specific issues; consistency with other indicators 7 The data collected are very hard for respondents to report accurately. 8 We use CPI for our indexing project. To the extent to which CE data/weight over or

under states the true values, it affects our indexed values. Those values are applied to all the taxpayers, so the impact is nationwide.

9 Due to small sample size, concerned about the trend in the data Also concerned about underreporting in certain areas such as premiums

10 Aggregate spending appears too low compared with other benchmarks, such as the PCE. Correspondingly, national savings rates appear too high. Low-income households appear to spend more than can be accounted for by their earnings, transfers and drawdown of assets. High-income households appear to spend too little, with implausibly high savings rates. The survey seems to have too few very high-income individuals.

11 Whether data accuracy and consistency is preserved in imputations Some smaller categories of expenditure is not well covered

12 1. survey burden is high, do respondents have an opportunity to evaluate implied totals for major commodity categories after providing data on all the small categories? 2. procedures for estimating annual expenditures from quarterly data are problematic: weights do not address sample attrition within the panel, to use data only for CU with a full panel; multiplying quarterly data by 4 seems problematic for distributional analysis across CU (means may not be adversely affected) 3. Calculation of standard errors (in published data) assuming independence across quarterly observations lowers standard error estimates

13 Difference between aggregate CE and PCE at a detailed consumption level (BEA Table 2.4.5u) Distribution of consumption by income class, especially concerns about left & right tails Overall small sample size creates problems when data are subdivided into multi-dimension subgroups Lack of balancing between income, consumption and saving on a per record basis Survey drop-outs & related reduction in sample size for those with all 4 quarters of Consumption data Inconsistency between interview and diary aggregates Large proportion of records for which income is based on imputation

14 Our sample in the CES (military members) is probably too small

Q14. Because of internal processing and publication requirements, CE currently publishes tables and produces public-use microdata for a calendar year during the following October. Does this timing meet your data needs?

Response options Non FED FED Total Yes 19 10 29 No 0 4 4 Not Applicable 2 4 6 answered question 19 17 36 skipped question 1 0 1

Q15. Please describe what timing would best meet your needs. Response options Non Fed Fed Total answered question 0 4 4 skipped question 22 14 36

Description of preferred timing of data provided by respondents

No response from non-FED

Number FED 1 Faster release dates; quarterly 2 If the data could be ready by July that would be preferable 3 Our project needs the CE data to be completed as soon as possible. No later

than late September of the year will meet our need. 4 Our numbers need to be finalized by the end of August, so would prefer the

numbers available in July.

Q16. What timing would better meet your needs, if the best timing is not possible? Response options Non Fed Fed Total answered question 0 4 4 skipped question 22 14 36

Description of better timing needs provided by respondents

No response from non-FED

Number FED 1 Biannual 2 If the data could be ready by July that would be

preferable 3 Mid-October of the year. 4 Data available by end of August

Q17. Apart from any issues of accuracy or timeliness, are there other ways in which the CE data do not meet your needs (e.g., topic coverage, demographic subgroups, sample size, data documentation)?

Response options Non Fed Fed Total

No 7 6 13 Yes. Please describe the ways in which the data do not meet your needs

14 11 25

answered question 19 16 35 skipped question 1 1 2

Other data concerns described by respondents

Number Non-FED 1 It would be nice to be able to build state level estimates 2 The income and labor market coverage is not ideal given the current timing.

Collecting information in period 1 and 5 really does not allow researchers to take full advantage of the panel aspect of the CE. I would much prefer if the CE added a short (very short) labor market update in all quarters (current employment status of head and spouse, weeks unemployed - if currently unemployed, was the worker laid off, quit, etc., and whether their job was eliminated or their plant closed).

3 I would encourage the CEX to expand its coverage of categories related to changes in net wealth. In particular changes in "credit" (i.e. UCC 006002 and "CREDITX5") are poorly tracked. A more detailed breakdown of changes in credit (like that in the "MOR" file for mortgages) would be appreciated. Also, I would encourage the Bureau to continue to cover a broad variety of spending and demographic categories, as these are vital for anyone trying to connect specific social groups with their spending patterns.

4 does not release state of residence information for many CU 5 It would be nice if the CE data use Census 3 digit occupational codes. 6 Diary and Interview data refer to DIFFERENT samples 7 Sample size is too small for poverty analysis 8 A greater focus on the broad categories, on asset holding and financial products, and

more intertemporal variation. The occasional special survey supplement (like the CPS) would be a great addition to the survey.

9 Yes, we would like to be able to use MSA-specific or state-specific data, but you only provide that for a few selected areas that aren't really representative (i.e., California heavily skews West region data, so I just use national data, even though we're in the west, because our demographics are very different from California)

10 Aside from housing, the wealth data are very poor (don't match the SCF). 11 It would be great to get some more information on financial asset holdings. Even just

knowing a little more detail about what they own (as opposed to amount) that would be helpful. And, what type of pensions households have --- DC, DB or both.

12 It would be immensely helpful if the tabulations called "composition of consumer unit" were to follow the Census Bureau household typology more closely such that the sum of each type of CU would add to the total number of consumer units.

13 As mentioned earlier, the child care expenses data aren't collected in a way that is useful for purposes of expanded poverty measures. Child care that is for purposes of parent employment needs to be separate from other expenses -- such as preschool for enrichment, or general "babysitting" while parents go out in the evening, etc.

14 The CE doesn't do well enough at capturing the highest income families, to the extent

needed to make good estimates of the distributional impact of consumption taxes. I have not completely reviewed the needs of the CE relative to the supplemental poverty measure, but would reaffirm any changes needed for that. One change for the SPM is to be able to distinguish whether child care expenses are to enable the parent to work, or are for some other reason. It would also be nice if the interview file had a complete measure of consumption (right now a few categories are only reflected in the diary, so are missed when using the quarterly interview microdata).

Number FED

1 More detail on specific insurance products (funeral, life, disability, etc.) 2 Yes, it would be great to have data that would be representative at the state level. 3 if the diary has more detail on food at home items it would be useful to report the

detail. for instance, products with whole wheat grains, vegetables that are dark green and deep yellow.

4 I would love to be able to use weighted state-level data. Census regions are just too big to use for geographical analysis.

5 sample size is big, some definitional issues, breakouts of series 6 Need participation in government programs, transportation for commuting only, child

care while parents work only. 7 Sample size, 8 Would like to know why a CU leaves the survey before the fourth interview, whether

it was due to moving out? 9 1. Need an annual measure of expenditures by major categories.

2. Need annual disposable income, by income type - (current measures are not good estimates.) 3. A measure of current value of each vehicle would be useful.

10 Lack of a total consumption question on the diary survey is a problem 11 Military specific data – how military members expend their money across all goods

and services

Q18. Are there additional tables that you would like published, either in more detail or sorted by additional demographic variables?

Response options Non Fed Fed Total Not applicable 5 3 8 No 11 7 18 Yes, please describe 5 7 12 answered question 19 16 35 skipped question 1 1 2

Descriptions of additional data requested by respondents

Number Non-FED 1 More detail on consumer credit and mortgage debt and repayment expenditures 2 Data for more MSAs, not just huge population centers 3 If the aggregates are not credible, who cares about the demographic breakdowns

which are even less credible. 4 Using multiple years it would be very helpful to have spending tabulations for

Blacks & Hispanics by age, income & educational attainment. 5 Published data on child care expenses are not particularly useful given the current

subgroups. To be most useful, a table would need to be focused on families with working parents and children <13.

Number FED

1 by Gender 2 I'd like to see the expenditure tables by state, for the expenditures on vehicles. 3 tables with median expenditures reported that could be compared to reported

mean values of expenditures 4 Energy spending by state. 5 all housing-related expenditures 6 For reporting by income levels:

- break up the top income category. - use an equivalent-income measure, taking into account CU size.

7 Military specific data – how military members expend their money across all goods and services

Q19. Please list any changes you would like to see in CE data, or other recommendations below. Response options Non Fed Fed Total answered question 12 8 20 skipped question 10 10 20

Number Non-FED 1 Include height and weight information to be able to calculate BMI. Also for food

purchases, record quantity purchased. 2 New data collection methodology focused on internal consistency 3 some means of improving accuracy of aggregate expenditure measure. I am not sure

how, perhaps a top down approach. 4 Adding selected attitude and expectation variables to match Survey of Consumer

Finances -- assessment of whether spend more than income, whether current income is higher or lower than usual, planned retirement age, health assessment

5 Again, more detail on use and cost of credit 6 Integrated micro information from the two surveys 7 The survey needs a complete redesign in light of user needs and modern survey

methods. 8 More support of historical data. 9 Start over from the ground up with a survey of household balance sheets in which Haig-

Simons wealth is matched to income minus expenditure. 10 See my answer to question 17 above. 11 If there were any way the annual data could be published on or about September 1st of

each year that would be wonderful. 12 If the current structure is maintained, then it would be helpful for BLS to construct files

that link the same consumer units across the quarters of the year and provide appropriate weight adjustments. Various approaches to this have evolved, and it would be helpful if BLS could document/investigate the benefits/drawbacks of various

approaches to attempting to construct annual consumption at the individual household level.

Number FED 1 This data is a national treasure and our only reliable public domain source of tracking

consumer expenditures in a timely manner (other than relying on bar-code corporate/market proprietary sources).

2 price data that match with the expenditures 3 Collect child-specific expenditure data with regard to health care. 4 Industry Division also needs specific expenditures (always asking us). 5 Release sooner. 6 Impute spending from the diary to the interview for items where the diary is deemed

superior. 7 see above (17 and prior comments) 8 More detailed analysis for each PCE consumption group regarding the proportion of

the difference between CE and PCE that is attributable to (1) definitional and coverage differences and what proportion is attributable to (2) measurement differences. The MLR article provides helpful qualitative information regarding definitional and coverage differences. In addition, it would be extremely useful to have research yielding quantitative estimates for each PCE group that could help allocate the aggregate differences between the two causes. More detailed analysis by PCE group on how much of the measurement differences (i.e., differences after accounting for definitional and coverage differences) are attributable to under-reporting of amounts on the CE and how much is caused by non-reporting More detailed analysis for the CE/PCE groups on how much of the measurement differences (i.e., differences after accounting for definitional and coverage differences) are attributable to under-reporting of amounts vs. how much is caused by non-reporting More detailed analysis on how much of the measurement differences for a consumption category are attributable to under-reporting of amounts and how much is caused by non-reporting

Q20. What is your position or affiliation?

Response options Non Fed State/local government 2 Academia-faculty 10 Academia-student 1 Policy research/Think-tank 5 Market research 1 Other (please specify) 2 answered question 19 skipped question 1 FED US Department of Defense 2 Bureau of the Census 1 Bureau of Labor Statistics 0 Bureau of Economic Analysis 3 Internal Revenue Service 1 Congressional Research Service 0 Congressional Budget Office 1 National Health Statistics 0 Federal Reserve Board 2 US Department of Agriculture 3 Other (please specify) 5 answered question 16 skipped question 1

Other –specified affiliations by respondents

Number Non-FED 1 academia- researcher 2 most of the above categories

Number FED

1 Public Policy Think Tank * 2 Department of Commerce/NOAA 3 Center on Budget and Policy Priorities 4 CMS 5 US Treasury Department

* This is a non-Federal respondent who received the Federal version of the Data Users’ Survey.

APPENDIX C. Notes from post-Forum wrap-up meeting On July 9, 2010, the Gemini Steering Team led a post Data Users’ Needs Forum wrap up discussion. The summary document follows: July 12, 2009 TO: Data User Needs Team FROM: Gemini Steering Team SUBJECT: Summary of July 9 DUF Wrap-up Meeting The CE program held a wrap-up meeting for BLS attendees of the July 9, 2010, Data Users’ Needs Forum. The purpose of the wrap-up meeting was to solicit information from internal attendees for inclusion in the meeting report and forthcoming statement on CE data priorities. The meeting agenda included (I) dominant user need themes, (II) perceived data flaws, (III) concerns or issues raised in side conversations, and (IV) ideas on the type of work CE should undertake based on the Forum. A brief summary of the meeting discussion follows. I. User Need Themes 1. Question about whether the CE program should focus on improving the quality of aggregate

estimates, or on continuing to collect very (or more) detailed data 2. Suggestions about using administrative records to supplement or improve existing data 3. Interest in obtaining accurate aggregate estimates at the National level 4. Interest in obtaining longitudinal weights 5. Requests for making it easier to use the panel data 6. Interest in improving timely access to the data 7. Interest in obtaining sub-national weights 8. Requests for ongoing comparisons to auxiliary data sources 9. Interest in more accurate information on savings 10. Interest in poverty thresholds 11. Requests for larger sample size 12. Interest in an extended longitudinal period, e.g., one that goes beyond 12 months 13. Interest in better capturing various payment methods 14. Questions about incorporating a balancing edit in the survey (e.g., for income, expenditures,

savings, and asset changes) 15. Statements that expenditure distributions, not just means, are important to some users, such

as for the supplemental poverty measure 16. Users mentioned several times that “Survey X measures Y better than CE”; highlighting a

need to follow-up on methods used by similar surveys (e.g., British, Canadian) for best practices

17. CPI needs/wants: a. point-of-purchase data b. geographic sample to be in continuous rotation c. cluster-level detailed expenditures

d. expenditure weights at the lowest level they can possibly get, for prices e. expenditure estimates for the items they’re pricing; everything else is a compromise

II. Perceived Data Flaws 1. Concern that CE overestimates rental equivalence compared to PCE (although some point

out that PCE underestimates rental equivalence) 2. Concern about evidence of underreporting, leading to low expenditures 3. Concern that some (possibly new) products are more difficult to measure 4. Questions about measurement and data quality issues for the high- and low-income

populations; question about whether we should consider oversampling to address 5. Questions about timeliness of data and lag in publication 6. Concern that CE cost weights are wrong 7. Concern that CE mis-measures the relationship between income and expenditures 8. Concern that CE shows an increase in savings while the National rate of savings has fallen III. Internal Observations 1. All users individually seem to want more detail for their items, but often at the expense of

items that others care about 2. CE can’t address all needs simultaneously and still be able to implement the survey 3. Concern that we didn’t hear from the full spectrum of users, and that the macro-economists

dominated the discussion 4. Conflict between macro- and micro-economists; e.g., some tension between users who

strictly want expenditure data, and those who want data that go beyond expenditures, e.g., income, savings, health care, etc.

5. Will need to make a decision about how high of a priority it is to address deficiencies in the relationship between income, expenditures, savings, and asset changes in the survey

6. Disconnect between concerns at the macro-level and how the interview is conducted IV. Suggested Action Items 1. Need to better understand the impact of design changes on users 2. Need someone to evaluate how good our allocation procedures are in CE 3. Need to better educate users to address misinformation about what data are and are not

available 4. For perceived data flaws, need to:

a. Educate users about those “perceived” flaws that don’t actually exist; e.g., select top three and write position papers about them

b. Conduct analysis on perceived flaws in question c. Address flaws that do exist

5. Need to develop a statement on and/or inventory of identifiable and quantifiable measurement error in CE, in order to operationalize the project’s mission of achieving a “measurable” decrease in measurement error

The full sets of meeting notes from the discussion can be found in this folder: \\psbres2\dces-public\Teams\Gemini\5 User Needs\ForumFeedback\