The Bureau of Labor Statistics’ (BLS) 10th Consumer Expenditure (CE) Microdata Users’ Workshop took place in Washington, D.C., July 15–17 and was preceded on July 14 by the CE Survey Methods Symposium, which highlighted the Gemini Project (for redesigning the CE survey). Both the workshop and the symposium consisted of presentations by BLS staffers and academics from around the country who have used CE survey data for economic research, as well as hands-on practical training sessions with the data. There was ample time for networking among the 60 or so participants; one-on-one time with BLS staff was easy to arrange for those with particular projects in mind. There was also plenty of time to talk about the data not collected, but that is interesting to know, such as expenditures on mass transit, alternative fuel, and the shadow economy.
Detailed data from the CE survey on how Americans spend money is used as the basis for all sorts of policy decisions made by the government and businesses. For example:
- The Census Bureau uses it to set thresholds for the Supplemental Poverty Measure.
- Policies concerning foster care and child support are based on the amounts parents spend on healthcare, education, transportation, and other expenses for their children.
- The Centers for Medicare & Medicaid Services uses it to estimate spending on healthcare.
- The IRS (Internal Revenue Service) uses it in calculating sales tax.
- The Department of Defense uses it to update cost-of-living adjustments for military families.
Also, the CE survey is one of two major data feeds for computing the Consumer Price Indexes (CPI; the other is the Telephone Point of Purchase Survey). With so much riding on the data, the BLS has to be sure that the data it gathers is accurate and reflects the demographics of the population.
What Is the CE Survey?
The federal government conducted its first CE survey during 1888–1891. In 1980, it transitioned from a decennial survey (conducted every 10 years) to a continuous survey of the population and how it spends its money. Use of computer-assisted personal interviewing (CAPI) began in 2003, and imputation for missing income variables was implemented the following year (2004). Model-based estimation for federal and state income tax liabilities (based on the National Bureau of Economic Research’s TAXSIM model) was introduced in 2013 to replace collected and missing values. (It turns out that estimates based on input from readily available sources are more reliable than collected data.) Data is released annually, although there is a 2-year lag between data collection and release. The most recent data available is for the period July 2013 through June 2014; the 2013 public-use microdata (PUMD) was released in September 2014.
The CE survey program consists of two separate forms: an interview survey (with 22 sections!) and a diary survey. The interviews require that the interviewee (representing the household) recall large (“big ticket”) or recurring expenditures made over the previous 3 months. The first interview gathers demographic information, expenditures, and income data; the second and third interviews focus on expenditures; and the fourth covers expenditures, income, and assets and liabilities.
The two consecutive 1-week diaries (filled out on paper) are designed to capture small, frequently purchased items, such as details of grocery purchases—not simply “fruit,” but the type of fruit, e.g., “bananas.” Households maintaining diaries (one instrument for the household) receive two or three interviewer visits to assess the recall of expenditures, receipt entry, and information on the buying habits of the household. Income and “consumer unit” (i.e., families and single consumers) characteristics are also captured (using 16 demographic characteristics). Sample size for the quarterly interviews is about 28,000, and the size of the sample maintaining the diaries is about half of that (14,000). Both information-gathering response rates have been declining for the past 40 years.
The Gemini Project
The Gemini Project launched in 2009 to research and develop redesigns for CE surveys—including exploring additional questions to ask, alternative data-gathering mechanisms, and incentives to accurately complete the surveys. Headed by Laura Erhard, the Gemini team is looking at the CE survey with an eye toward improving data quality “through a verifiable reduction in measurement error” (particularly errors caused by underreporting), as well as addressing issues of measurement error and respondent burden.
It is also important that any changes recommended (and ultimately made) have a neutral impact on the budget. This could be challenging to illustrate to policymakers because budgets for data programs are handled discretely and separated even within the same agency. Some changes made to the CE survey might increase its cost but result in cost reductions for other programs.
Who is underrepresented among CE respondents? It turns out that Millennials are reluctant to enter data into a paper diary, preferring to use their smartphones. Also, how likely are teenagers to tell the parent who is reporting household expenditures how much money they spend on video games? Moving away from an individual (known as the “reference person”) reporting on behalf of the household—allowing all members of a household to capture their own expenditures—may prove beneficial by increasing the accuracy of the data collected.
The Gemini proposal suggests sampling consumer units in two waves of data collection, 12 months apart, and with monetary incentives for survey participation. Each wave is composed of two visits with one consumer unit respondent (a recall interview for large expenditures that are easily recallable and a records interview for expenditures likely to be found in receipts). Between these two interviews, there would be a diary period, with each consumer unit member age 15 and older collecting all expenditures in electronic diaries for 1 week.
A comparison of the current and redesigned CE survey data-gathering effort is available at bls.gov/cex/geminiproject.htm#news; details concerning Proof-of-Concept testing are available at bls.gov/cex/gemini_poc_test.htm. The team also will test various incentives for completing each interview and is eager to see if additional money for using receipts at the time of data entry (as opposed to recalling purchases) affects data quality and response rates. If the Proof-of-Concept testing indicates that the BLS should move forward with these changes, a Large-Scale Feasibility Test will ensue.