Data and Methods
The Medical Expenditure Panel Survey (MEPS)
This chapter uses data from the Medical Expenditure Panel Survey, a longitudinal survey that produces annual public use files on the use and costs of health care, health insurance coverage, and a host of other data on U.S. households (https://meps.ahrq.gov/mepsweb/). The sample is drawn from households that completed the National Health Interview Survey (NHIS) the previous year. Five in-person data collection visits using a computer- assisted personal interviewing (CAPI) system occur with each panel over a 2.5-year period in an overlapping panel design. Data collection is divided into two phases. The first half of each year includes the first-time-in-MEPS interviews with the newest sample panel; third interviews with the previous year's panel; and final interviews with the panel that is exiting MEPS. The second half of the year includes the second and fourth interviews with the remaining panels. Data for this chapter are from work in fall 2018. About 400 field interviewers conduct more than 25,000 interviews with a household informant each year. Answers to most questions are combined with others to create about 4,000 constructed variables for annual estimates. Key survey statistics include the number of hospital stays, medical provider visits for ambulatory care, and prescribed medicines dispensed each year. MEPS is sponsored by the Agency for Healthcare Research and Quality (AHRQ) in the U.S. Department of Health and Human Services (HHS).
To understand the magnitude and rich detail of these data, consider prescribed medicines for older people. MEPS reports that just over 90% of people aged 65 and older obtained prescribed medicines in 2016. The mean annual expenditure for these people was $3,289. This key statistic is quite sensitive to data from outliers and thus puts a premium on high quality interviewing. (See Table A6A.1 in Appendix 6A in the online supplementary materials for more detailed MEPS data on prescribed medicines.)
In 2018 MEPS implemented a major upgrade of the CAPI system, simplifying some questionnaire sections and entry mechanisms. Interviewers were trained on the new system in January 2018. Researchers were keen to observe how the new system was operating in the field, and to determine whether any corrections might be needed.
Two Critical Question Series: The Calendar Series and Provider Probes Series
Two question series were of particular interest because they were asked in all interviews, they were always recorded in CARI (almost all respondents gave consent to record), and they were critical building blocks for producing data on the use and cost of health care services; key MEPS statistics were: (1) use of calendar aids and other records of medical care, and (2) provider probes (filter questions that prompt the respondent to recall services from various types of medical providers). A striking characteristic of health care surveys is the large degree of underreporting: respondents forget events, or remember events that occurred before the interview reference period (telescoping) (see Miller and Mathiowetz, Chapter 2, this volume).
Increased levels of event reporting are associated with the presence of records (Kashihara and Wobus 2006). The calendar series asks if various sorts of records are available in the interview (e.g., a calendar with entries for medical visits, insurance statements, a patient portal, prescription records or bottles, etc.), and who in the household was associated with each record type. The CAPI screen presents the entry area for these items as a grid, with each household member listed on a row and each record type as a column header. Interviewers can enter answers in any order, by person or by record type (see Online Appendix 6B). The design objective is to encourage respondents to bring records for all family members into the interview and to structure the questioning so that the records can be incorporated into the interview in any order. The grid allows the interviewer to follow the respondent's lead (a respondent-driven path). Grids that allow movement from any cell to any other cell are one of the most difficult item types for interviewers to navigate in a standardized manner because they require the interviewer (rather than the CAPI instrument) to decide where to go next (see Edwards, Schneider, and Brick 2008; Sperry et al. 1998). As part of the technical upgrade, interview skip patterns were revised to take advantage of record types mentioned in this grid, streamlining the interview for respondents with records.
The provider probes are a series of 15 questions about various kinds of health care providers. They were re-ordered in the technical upgrade to begin with three that accounted for the highest expenditures. For example, the first provider probe in the 2018 Round 1 interview read: "since January 1,2018, were you [...] admitted to the hospital?" (see Online Appendix 6B). If the answer was yes, the interviewer determined who was admitted, collected details about the event, and administered probes about emergency room visits, outpatient visits, and other visits and events.