Article

April 2016

The Consumer Expenditure Survey redesign initiative

To help mark the Monthly Labor Review’s centennial, the editors invited several producers and users of BLS data to take a look back at the last 100 years. This article highlights the past, present, and future of the Consumer Expenditure Survey (CE), an ongoing federal survey program designed in the 1970s to collect information on spending, income, and household characteristics. This paper also presents an overview of the Gemini Project, which was launched in 2009. The project’s objectives are to reduce measurement error, improve overall data quality, enhance the analytic value of the CE data to users, and implement a new design that supports greater operational flexibility to respond to changes in the interviewing environment. This paper discusses the motivation, accomplishments, challenges, and expected benefits of the CE redesign.

The BLS mission statement states that BLS executes its mission by “. . . providing products and services that are accurate, objective, relevant, timely, and accessible.” The Consumer Expenditure Survey (CE) program supports this mission in its research and development projects. In particular, the CE program maintains the relevance of the surveys through research projects and regular biennial updates to the Consumer Expenditure Quarterly Interview (CEQ) questionnaire and improves data quality through methodological and data collection revisions. A team of economists, survey methodologists, and statisticians work together to identify and propose solutions for measurement error problems, develop new approaches to maintaining response rates and other data collection issues, streamline data processing procedures, and investigate and implement estimation methods that are more efficient.

The CE measures spending by consumers for the total U.S. noninstitutional population. The principal purpose of the survey is to provide the Consumer Price Index (CPI) program with the expenditure weights that the CPI program uses to generate an all-items index. The CPI is the primary measure of household inflation and is one of the most important economic measures of the U.S. economy. Accurate information on consumer spending habits—used to determine expenditure weights—is vital to the CPI. An expenditure weight is an estimate of consumer expenditure needed to weight the market basket of goods. In addition to collecting expenditure data, the CE collects information on family income, assets, liabilities, housing characteristics, and detailed demographic information, making the survey a unique data source for policy analysis and research.

The article comprises (1) an introduction to the CE program, (2) a history of the CE’s methodological improvements, (3) an overview of the Gemini Project, and (4) highlights of recent research conducted in support of the redesign. The article concludes with a discussion of next steps for the CE as the Gemini Project transitions from development to implementation.

CE background

The CE program currently consists of two independent surveys, the Consumer Expenditure Quarterly Interview (CEQ) and the Consumer Expenditure Diary (CED). The geographic samples for the surveys are updated every 10 years using the decennial census to create the sampling frame, and new addresses are selected each year from a master address file that is updated with new addresses from a Postal Service file. The unit of measure is a consumer unit (CU), which consists of all members of a household who are related or who share expenditure decisions. The CE data are collected by the U.S. Census Bureau.

In the CED, respondents complete two 1-week expenditure diaries that capture all expenditures for everyone in the CU. The diaries span 2 consecutive weeks. The CED is designed primarily to collect expenditure data on small or frequently purchased items such as food, meals away from home, apparel, and personal care items. Interviewers visit the selected sample units to collect demographic and income information, leave the diaries, and ask the diary keeper to record expenditures daily. About 7,000 CUs participate in the CED per year and complete about 14,000 diaries.

The purpose of the CEQ is to capture expenditures for larger and less frequently purchased items and for items, such as rent or utilities, for which payments are made on a regular basis, such as monthly. The CEQ is a quarterly survey with a rotating panel design. Each CU is interviewed four times over the course of 10 months (although respondents are only in the sample for 10 months, the survey’s recall period still covers a 12-month period). In the first interview, the interviewer collects demographic data for all CU members and establishes an inventory of items such as properties, vehicles, and insurance policies, which are updated in subsequent interviews. The interviewer collects data on all expenditures for the previous 3-month period and updates demographic and inventory data. Income data are collected in interviews 1 and 4 only; assets and liabilities are asked in the fourth interview. The CEQ produces usable interviews for about 28,000 CUs per year.

While there is some overlap in spending categories covered in the CED and CEQ, some items, such as personal care items, are not captured in the CEQ while others, such as food expenditures, are captured only at an aggregate level (i.e., not at finer levels of detail) in the CEQ.

The revision of the CPI biennial expenditure weights remains the primary reason for undertaking the CE. However, CE data are used by a wide variety of researchers, policy analysts, and government agencies that are interested in studying the economic well-being of America’s families. The CE allows outside researchers to relate the expenditures and income of consumers to the characteristics of those consumers. Data from the CE are used by economic policymakers and others to evaluate the effects of tax or other policy changes on levels of well-being among diverse socioeconomic groups. Econometricians use the CE data in various ways. For example, some use it to construct economic models, such as structural models of optimal life-cycle consumption expenditures. Others use it to test whether people’s spending is based on their current income or on what they expect to earn over a longer period of time, or to investigate the distributional effect of consumption taxes. Market researchers find CE data valuable in analyzing the demand for groups of goods and services. These are some specific uses of CE data:

On an exploratory research basis, CE data are used to determine poverty thresholds for the U.S. government’s Supplemental Poverty Measure.13

Recall interview question development. The questionnaire for the visit 1 recall interview of the redesigned CE is in development. Through cognitive interviews, contracted Westat researchers are testing various aspects of the questionnaire, including question wording and the level of aggregation at which each question should be asked. For example, the category for educational expenses may be asked as a single question “What did your household spend on education expenses in the past X months”or as multiple questions—“In the past X months, what did your household spend on

• college tuition?

• tutoring and test preparation?

• educational books and supplies?”

The level of aggregation required for some expenditure categories may result in data provided by some lower level questions in the recall interview and by responses in the records-based interview, so that the category is spread across both interviews.

Recall interview protocol. Although the current CEQ has a protocol in place, this protocol has never been fully evaluated to identify possible improvements. To ensure that the redesigned recall interview uses the best method for collecting recall information, the CE tested two alternative protocols: (1) a respondent-driven versus interviewer-driven interview structure, and (2) a grouped versus interleafed question order.

The respondent-driven versus interviewer-driven dimension concerned interview order and flow. In the respondent-driven protocol, respondents were presented with a list of broad expenditure categories (e.g., trips and vacations, housing repairs and maintenance, education expenses) and controlled the order of the interview. The interviewer-driven protocol progressed with a set order and is the method used in the current CEQ.

The grouped versus interleafed dimension focused on how questions flowed within each section. For grouped sections, the respondents were presented with a list of items in the section and were asked to indicate all items for which they had an expense. Once a list of expenses was complete, followup questions were asked about each of those items. For the interleafed protocol, all followup questions were asked directly after an affirmative item expensethat is, before continuing on to the next item. The interleafed protocol is the procedure used in the current CEQ.

Through a series of cognitive interviews, contracted Census researchers found that the original structure—interviewer-driven with an interleafed question order—worked best for respondents and interviewers alike.

Records interview protocol. For the visit 2 records interview protocol test completed in 2015, CE program staff tested a track in which respondents handle their records and determine the order of the interview. The staff then compared the results with those from a track in which the interviewer organizes the records and follows a scripted interview order.

With the respondent track, the respondent has a great deal of control over the interview. The interviewer asks the respondent to report expenses from the records gathered, in an order chosen by the respondent. An expected advantage of this design is that respondents can complete the interview in a manner that aligns with how they organize and think about their own records. A possible limitation of this design is that respondents may require more time to locate required information from their own records than would experienced interviewers who know where to look on receipts and other expense documents. Additionally, if a respondent’s record collection is not well organized, the interviewer may be required to revisit categories multiple times to collect all the needed data.

Using the interviewer track, the field representative leads the participant through the interview in a defined order. The interviewer follows a predetermined question order and handles the records (with respondent permission). Expected strengths of this design are reductions to both interview length and respondent burden.

The project resulted in findings that both interviewer-driven and respondent-driven protocols were feasible for a records interview. However, the frequency of use of electronic records was significantly greater in the respondent-driven group, and some participants expressed reluctance to hand over records to the interviewer. The team recommended moving forward with a hybrid approach whereby interviewers control the order of the questions and respondents control their own records. The study also showed the importance of a checklist in managing respondent expectations of interview content.

Large-scale feasibility test. The CE program plans to field a large-scale feasibility test (LSF) in 2019. The purpose of the LSF test is to evaluate the effectiveness of the redesigned survey in terms of response rate, data quality, and expenditure reporting differences. The test will benefit from the statistical power of a larger sample size. The LSF test will be designed to reflect all of the redesign components in an environment closer to what the production survey will operate in than the current survey. For example, the most up-to-date questionnaire revisions will be included, and all materials used will be designed specifically for the test. The LSF test will employ new collection instruments and revised protocols based on inputs from all previous field tests and other research activities.

CE research highlights

This section offers a summary of recent research conducted in support of the redesign and is organized by which of the redesign objectives—reducing error, reducing burden, maintaining cost neutrality, or monitoring results—the study was developed to address. It concludes with a discussion of ongoing research challenges.

Reduce error

Reducing measurement error, particularly error generally associated with underreporting, is the primary objective of the redesign. To this end, the CE program is researching the linking of administrative data to respondent records. Recommended by the CNSTAT expert panel, supplementing respondent-provided data with relevant auxiliary data has the potential to both increase the accuracy of CE estimates and reduce data collection and processing costs. This approach is currently being tested with CE housing data, for which there is some evidence of underreporting.

Additional measurement-error research focuses on proxy reporting. CE staff have long been concerned about underreporting due to proxy reporting. Research and development of individual-level diaries to replace household-level diaries offers the potential to address this issue in the CED. The path is less clear for the CEQ. On the basis of questions asked in a 2013 research section, respondents who reported being “very knowledgeable” of other CU member expenditures showed a 5-percent increase in total reported expenditures compared with those reporting less proxy knowledge.19 CE researchers have proposed implementing this threshold.

Similarly, because of evidence that the bounding interview may not be effective in minimizing telescoping errors, and also because its expenditure data were not used in the production of official estimates nor released as part of the microdata files, CE researchers suggested eliminating the bounding interview as a way to reduce data collection costs without adversely affecting data quality. Starting with the 2015 production cycle, the bounding interview has been eliminated.

Monitor results

Along with undertaking projects in support of the redesign effort, CE researchers are developing ways to monitor the results of the redesign and verify its impact on measurement error. A set of data quality metrics is in development, which will be used to establish baselines for monitoring trends in the quality of routine survey production activities.21

However, through innovative design discussions, adherence to the CE’s planned redesign roadmap, iterative updates based on ongoing research and development, regular and interactive communication with stakeholders, and the support of a dedicated research and development staff, the CE program continues toward the successful implementation of a comprehensive survey redesign.

Suggested citation:

Adam Safir, Jay Ryan, Laura Erhard, Lindsay Jilk, and Lucilla Tan, "The Consumer Expenditure Survey redesign initiative," Monthly Labor Review, U.S. Bureau of Labor Statistics, April 2016, https://doi.org/10.21916/mlr.2016.15.

Notes


1 “Research experimental poverty threshold,” U.S. Bureau of Labor Statistics, at https://www.bls.gov/pir/spmhome.htm.

2 “Expenditures on children by family,” U.S. Department of Agriculture, at http://www.cnpp.usda.gov/ExpendituresonChildrenbyFamilies.

3 Karen Goldenberg and Jay Ryan, “Evolution and change in the Consumer Expenditure Surveys: adapting methodologies to meet changing needs” (U.S. Bureau of Labor Statistics, 2009), https://www.bls.gov/cex/nber2009ryan1.pdf.

4 More information about TAXSIM is available at http://users.nber.org/~taxsim/.

5 Geoffrey D. Paulin and William Hawk, “Improving data quality in Consumer Expenditure Survey with TAXSIM,” Monthly Labor Review, March 2015, https://www.bls.gov/opub/mlr/2015/article/pdf/improving-data-quality-in-ce-with-taxsim.pdf.

6 John Sabelhaus, David Johnson, Stephen Ash, David Swanson, Thesia Garner, John Greenlees, and Steve Henderson, “Is the Consumer Expenditure Survey representative by income?” (Cambridge, MA: National Bureau of Economic Research working paper, October 2013), http://www.nber.org/papers/w19589.pdf.

7 Ian Elkin, “Recommendation regarding the use of a CE bounding interview,” bounding interview project paper (U.S. Bureau of Labor Statistics, May 2013).

8 “Gemini Project vision document,” Gemini Project Steering Team, U.S. Bureau of Labor Statistics, April 16, 2010, last revised February 25, 2015, https://www.bls.gov/cex/gemini_vision_2015.pdf.

9 “Gemini Project to redesign the Consumer Expenditure Survey,” U.S. Bureau of Labor Statistics, revised November 5, 2015, https://www.bls.gov/cex/geminiproject.htm.

10 Don A. Dillman and Carol C. House, eds., Measuring what we spend: toward a new Consumer Expenditure Survey, National Research Council (Washington, DC: The National Academies Press, 2012), http://nap.edu/catalog.php?record_id=13520.

11 Jennifer Edgar, Dawn V. Nelson, Laura Paszkiewicz, and Adam Safir, “The Gemini Project to redesign the Consumer Expenditure Survey: redesign proposal,” CE Gemini Project materials (U.S. Bureau of Labor Statistics, June 2013), https://www.bls.gov/cex/ce_gemini_redesign.pdf.

12 “Gemini redesign project high level timeline,” March 7, 2014, at https://www.bls.gov/cex/geminitimeline.pdf. For the project website, see “Gemini Project to redesign the Consumer Expenditure Survey” at https://www.bls.gov/cex/geminiproject.htm

13 Jennifer Edgar, Brandon Kopp, Erica Yu, Laura Erhard, and Adam Safir, “Gemini content team: final report,” Gemini content project paper (U.S. Bureau of Labor Statistics, February 2014).

14 Brett McBride, Erica Yu, Mik Slomka, Laura Erhard, and Jeanette Davis, “2012 research section analysis,” unpublished paper, 2012 CE research section analysis project (U.S. Bureau of Labor Statistics, December 2013); Brett McBride and Lucilla Tan, “Quantifying CHI doorstep concerns as risk factors of wave 1 nonresponse for the CE Interview Survey,” unpublished paper, CHI panel data analysis of survey burden project (U.S. Bureau of Labor Statistics, April 2014).

15 Erica Yu, “Asking questions about household member activities to improve expenditure reporting,” unpublished paper, proxy reporting lab study project (U.S. Bureau of Labor Statistics, September 2013).

16 Brandon Kopp, Brett McBride, and Lucilla Tan, “An exploratory study on the association of doorstep concerns with three survey quality measures for the CE Interview Survey” (U.S. Bureau of Labor Statistics, November 2013) and McBride and Tan, “Quantifying CHI doorstep concerns.”

17 Wendy Martinez and Lucilla Tan, “An exploratory text analysis project for the Consumer Expenditure Interview Survey,” unpublished paper, CE CHI text analysis project (U.S. Bureau of Labor Statistics, June 2015).

18 Daniel K. Yang, “Compiling respondent burden items: a composite index approach,” paper presented at the annual Consumer Expenditure Survey Methods Symposium, June 14, 2015, https://www.bls.gov/cex/respondent-burden-index.pdf.

19 Adam Safir and Lucilla Tan, “Towards Determining an Optimal Contact Attempt Threshold for a Large-Scale Personal Visit Survey,” paper presented at the American Association for Public Opinion Research annual conference, May 16, 2015.

20 Scott Fricker and Lucilla Tan, “A proposal for a preliminary framework for monitoring and reporting on data quality for the Consumer Expenditure Survey,” CE research library papers (U.S. Bureau of Labor Statistics, May 2012), https://www.bls.gov/cex/research_papers/pdf/cesrvmeth_quality.pdf.

21 Brandon Kopp, “The evolution of password requirements in the Consumer Expenditure Diary Survey,” paper presented at the FESAC annual conference, June 12, 2015.


article image
About the Author

Adam Safir
safir.adam@bls.gov

Adam Safir is a program manager in the Office of Prices and Living Conditions, U.S. Bureau of Labor Statistics.

Jay Ryan

Jay Ryan is formerly a program manager in the Office of Prices and Living Conditions, U.S. Bureau of Labor Statistics.

Laura Erhard
erhard.laura@bls.gov

Laura Erhard is an economist in the Office of Prices and Living Conditions, U.S. Bureau of Labor Statistics.

Lindsay Jilk

Lindsay Jilk is an economist formerly in the Office of Prices and Living Conditions, U.S. Bureau of Labor Statistics.

Lucilla Tan
tan.lucilla@bls.gov

Lucilla Tan is an economist in the Office of Prices and Living Conditions, U.S. Bureau of Labor Statistics.

close or Esc Key