MEPS Home Medical Expenditure Panel Survey
Font Size:
Contact MEPS FAQ Site Map  
S
M
L
XL
 

Methodology Report #33:
Sample Designs of the Medical Expenditure Panel Survey Household Component, 1996–2006 and 2007–2016


Sadeq R. Chowdhury, PhD, Steven R. Machlin, MS, and Kilem L. Gwet, PhD, Agency for Healthcare Research and Quality


Table of Contents

1.1 Abstract
1.2 Suggested Citation
1.3 Introduction
1.4 General Features of MEPS-HC Sample Design
1.4.1 Overview
1.4.2 Figure 1. MEPS Household Component Overlapping Panel Design MEPS Panel
1.4.3 MEPS Target Population and Sampling Frame
1.4.4 Analytical Goals, Precision Requirements, and Sample Size Targets
1.4.5 Sampling Unit Definition and Eligibility Criteria
1.4.6 Eligibility of Dwelling Units
1.4.7 Interview Eligibility of Persons
1.4.8 Link of MEPS to the National Health Interview Survey
1.5 Overview of NHIS Sample Design
1.5.1 Overview of 1995–2005 NHIS Sample Design
1.5.2 Overview of 2006–2015 NHIS Sample Design
1.6 MEPS Sample Designs
1.6.1 1996–2006 MEPS Sample Design and Selection (Panels 1–11)
1.6.2 Table 1. Summary of Oversampling Domains in MEPS 1996–2006
1.6.3 Table 2. MEPS Households Selected from the Frame (NHIS) for MEPS 1996–2006
1.6.4 2007–2016 MEPS Sample Design and Sample Selection (Panels 12–21)
1.6.5 Table 3. Summary of Oversampling Domains in MEPS 2007–2016
1.6.6 Table 4. MEPS Households Selected from the Frame (NHIS) for MEPS 2007–2016
1.7 Sample Yields and Response Rates in MEPS
1.7.1 Responding Sample Size
1.7.2 Table 5. Number of responding dwelling units, families, and persons by year and panel in MEPS Household Component Full-Year Files, 1996–2016
1.7.3 Design Effects
1.7.4 Table 6. Design effects of selected variables in MEPS, 1996–2006 and 2007–2016
1.7.5 Response Rates
1.7.6 Table 7. MEPS individual panel and combined annual response rates
1.8 Analysis Weights and Variance Estimation
1.8.1 Development of Analysis Weights
1.8.2 Variance Estimation
1.9 References
Appendix A. MEPS sample sizes by selected characteristics, 1996–2016
Definitions for Table A1
Table A1. Sample selection details of MEPS from NHIS, 1996–2016 (Panels 1–21)
Table A2. Number of completed person-level interviews by age, sex, race/ethnicity, region, and MSA status in MEPS Full-Year Files: MEPS-HC 1996–2006
Table A3. Number of completed person-level interviews by age, sex, race/ethnicity, region, and MSA status in MEPS Full-Year Files: MEPS-HC 2007–2016
Appendix B. Calculation of Response Rates in MEPS
Table B1. Example sample size and final response rates - Full Year (Panel 20, Rounds 1–3 combined with Panel 21, Rounds 3–5): MEPS 2016

1.1 Abstract

Three sample designs have been used for the MEPS Household Component (MEPS-HC) since its inception in 1996. The first design covers the period 1996–2006, the second design covers 2007–2016, and the third design was introduced in 2017 and is scheduled to continue until 2025. This report provides a detailed description of the first two sample designs: 1996–2006 and 2007–2016. It also provides information on target sample sizes, number of sampled units, number of completed interviews, and response rates for the panels of the MEPS from its inception in 1996 through 2016.

Return to Table of Contents


1.2 Suggested Citation

Chowdhury, S.R., Machlin, S.R., Gwet, K.L. Sample Designs of the Medical Expenditure Panel Survey Household Component, 1996–2006 and 2007–2016. Methodology Report #33. January 2019. Agency for Healthcare Research and Quality, Rockville, MD. https://meps.ahrq.gov/data_files/publications/mr33/mr33.shtml

* * *

Additional information on MEPS is available from the MEPS Project Director at MEPSProjectDirector@ahrq.hhs.gov.

Center for Financing, Access, and Cost Trends
Agency for Healthcare Research and Quality
5600 Fishers Lane, Mailstop 07W41A
Rockville, MD 20857
https://meps.ahrq.gov_

Return to Table of Contents


1.3 Introduction

The Medical Expenditure Panel Survey (MEPS) Household Component, a nationally representative sample of the U.S. civilian noninstitutionalized population, has been conducted on an annual basis since 1996 by the Agency for Healthcare Research and Quality (AHRQ), U.S. Department of Health and Human Services (HHS). The MEPS provides national estimates of health care use, expenditures, sources of payment, and health insurance coverage as well as information on respondents' health status, demographic/socioeconomic characteristics, employment status, access to health care, and satisfaction with health care. The survey data can be used to produce estimates for persons and families at the national and regional levels for the full target population as well as for many population subgroups.

Three broad sample designs have been used for the MEPS Household Component (MEPS-HC) since its inception in 1996. The first design covers the period 1996–2006, the second design covers 2007–2016, and the third design was introduced in 2017 and is scheduled to continue until 2025. This report describes the first two sample designs of MEPS-HC: 1996–2006 and 2007–2016. Similar reports describe sample designs of MEPS for different time periods in the past. MEPS Methodology Report #2 details the sample design of the 1996 MEPS-HC, and MEPS Methodology Report #11 describes the 1997 MEPS design (Cohen S, 1997; Cohen S, 2000). MEPS Methodology Report #22 describes the MEPS-HC sample design for 1998–2007 and updates descriptions of earlier designs (Ezzati-Rice, et al. 2008). An additional report provides an overview of the core components of MEPS data collection and the statistical features of the survey (Cohen SB, 2003). In this report, we provide a comprehensive summary of the two designs prior to 2017 when the third design took effect. We also provide information on target sample sizes, number of sampled units, number of completed interviews, and response rates for the panels of the MEPS from its inception in 1996 through 2016. A future methodology report will describe the recently implemented 2017–2025 MEPS design.

Return to Table of Contents


1.4 General Features of MEPS-HC Sample Design

Overview

The MEPS-HC is a complex national probability survey of the U.S. civilian noninstitutionalized population. Each year a new panel of households is selected from among those households that participated in the previous year's National Health Interview Survey (NHIS), another large ongoing federal health survey conducted by the National Center for Health Statistics of the Centers for Disease Control and Prevention in HHS. Each new MEPS annual sample is referred to as a panel. In the MEPS, data are collected for each panel through a series of five rounds of computer-assisted personal interviews (CAPI) over 30 months to yield annual data for 2 full calendar years.

A new MEPS panel of households has been selected and fielded each year since 1996. During each calendar year (with the exception of the first year in 1996), data were collected simultaneously for two MEPS panels as illustrated in Figure 1. One panel is in its first year of interviews (e.g., in the year 2015, Rounds 1 and 2 of Panel 20), while the prior year's panel is in its second year of data collection (e.g., in 2015, Rounds 3, 4, and 5 of Panel 19). The reference period for Round 3 for each MEPS panel overlaps 2 calendar years.

Figure 1. MEPS Household Component Overlapping Panel Design

Panel 19 covers Rounds 1 to 5 during years 2014 and 2015, while Panel 20 covers Round 1 to 5 during years 2015 and 2016

In 1996, the first year of MEPS, only one panel (Panel 1) was fielded and the annual data for 1996 were based on this single panel of data. However, starting in 1997, to increase statistical power of annual estimates produced from MEPS, data are combined across two distinct nationally representative samples, making use of the MEPS overlapping panel design. More specifically, annual estimates are made by combining data from the panel in its first year of data collection and the panel in its second year of data collection. For example, 2015 annual estimates were produced using data collected for the second year of Panel 19 and data collected for the first year of Panel 20.

In addition to annual estimates, the MEPS design structure permits longitudinal estimates over 2 consecutive calendar years, thus allowing examination of person-level changes for select variables over a 2-year period. For example, analysts can assess the persistence of high health-care expenditures by examining whether individuals with low/high expenditures in one year have low/high expenditures in the subsequent year (Cohen SB and Ezzati-Rice TM, 2006).

MEPS Target Population and Sampling Frame

The target population for the MEPS consists of all persons who are members of the civilian noninstitutionalized population (e.g., not in prisons, nursing homes, or the military) at any time during the year and living in the 50 states or the District of Columbia. The NHIS serves as the sampling frame for the MEPS. In most years, the MEPS annual household panel sample is selected from responding households in two of the four NHIS panels (which are different than MEPS panels, see section below on overview of 1995–2005 design for an explanation of NHIS panels) during calendar quarters 1–3 of the previous year. The NHIS quarter 4 is not processed soon enough for use in selecting the following year's MEPS sample as each new MEPS panel must be fielded beginning in mid to late January. Thus, a sample representing about three-eighths (2/4 panels x 3/4 quarters) of the NHIS responding households is generally made available for use in the MEPS.

Analytical Goals, Precision Requirements, and Sample Size Targets

The sample is designed to produce estimates to meet a specified level of precision at the national and regional level and for certain subgroups of the population such as selected racial and ethnic groups.

In general, broad sample design goals for the MEPS include:

  • A sample that will provide unbiased national and regional estimates (four census regions) of health care expenditure estimates and other health parameters with targeted precision, and
  • A sample that will meet targeted precision requirements for policy-relevant subgroups of the population.

Based on varying HHS objectives coupled with the MEPS budget resources available, the sample size and subdomains oversampled (i.e., groups that are sampled at a higher rate) for MEPS can vary from year to year. The MEPS person-level precision requirements are specified for national estimates derived from individuals that are considered full-year respondents (individuals with responses for the entire period(s) during the year they were living in the civilian noninstitutionalized population—see https://meps.ahrq.gov/about_meps/hc_sample.shtml). In the determination of sample sizes necessary to achieve the precision requirements, adjustments are made for household (also referred to as dwelling unit or DU) nonresponse and survey attrition to determine the required number of initial sample units per year. Table 2 (for the 1996–2006 design) and Table 4 (for the 2007–2016 design) present the number of DUs selected in each MEPS panel. Table 5 presents the number of responding DUs, families, and persons in each year for the two MEPS panels combined and separately.

Sampling Unit Definition and Eligibility Criteria

The definition of DUs and group quarters in the MEPS-HC are generally consistent with the definitions employed for NHIS. The definitions used are:

  • Dwelling unit (DU) is a house, apartment, group of rooms, or single room occupied as separate civilian noninstitutional living quarters or vacant but intended for occupancy as separate living quarters. This term is the NHIS definition for households and is the unit sampled for the MEPS.
  • Group quarters consist of a single civilian noninstitutional dwelling or structure in which nine or more unrelated persons reside and where inhabitants are not considered a part of any other DU.

After selection of the NHIS households (occupied DUs and group quarters), "reporting units" are formed based on information collected in the NHIS and for fielding of the MEPS sample. A reporting unit (RU) is a person or group of persons in the sampled DU who are related by blood, marriage, adoption, foster care, or other family association. Each RU was interviewed as a single entity for MEPS. Thus, the RU serves chiefly as a family-based "survey" operations unit rather than an analytic unit per se. Regardless of the legal status of their association, two persons living together as a "family" unit are treated as a single RU if they choose to be so identified. Unmarried college students (less than 24 years of age) who usually live in a sampled household but who live away from home and go to school at the time of the MEPS interview are treated as a student RU separate from that of their parents for the purpose of data collection. However, data for a student RU is usually collected from the respondent of the parent RU. Examples of different types of RUs include:

  • A married daughter and her husband living with her parents in the same DU constitute a single RU.
  • A husband and wife and their unmarried daughter, age 18, who is living away from home while at college constitute two RUs (even though the student daughter is not interviewed).
  • Three unrelated persons living in the same DU would each constitute a distinct RU, i.e., a total of three RUs.

Eligibility of Dwelling Units

In the two designs covered in this report, the only major difference in eligibility status for housing units between NHIS and MEPS is that college dorms represent ineligible DUs for MEPS. College-aged students living away from home during the school year were interviewed at their place of residence for the NHIS, but, in contrast, such students are identified by and linked to their parents' household for MEPS. In other words, for the NHIS, college students living in student housing are sampled independently from their families. But for MEPS, such students are identified through the sample selection of their parents' RU. Once the MEPS sample is selected from among the NHIS households characterized as NHIS respondents, RUs representing students living in student housing or consisting entirely of military personnel are deleted from the sample. In MEPS, removing college students found in college housing sampled for the NHIS eliminates the opportunity of multiple chances of selection for MEPS for these students. Military personnel not living in the same RU as civilians are ineligible for MEPS. After such exclusions, all RUs associated with households selected from among those identified as NHIS responding households are then fielded in the first round of MEPS. The number of sampled DUs (households) for each MEPS panel is shown in Table A1 (Appendix A). However, new RUs are created when members of the household leave the primary RU between the NHIS and MEPS and are then followed according to the MEPS rules.

Interview Eligibility of Persons

Three critical factors define a person's interview status for each round of data collection in the MEPS. These factors are: "in-scope" status, "keyness" status, and "eligibility" status.

In-scope

A person is considered as in-scope for a MEPS interview round if he or she was a member of the U.S. civilian noninstitutionalized population at some time during the reference period covered by that round.

Keyness for MEPS—"Key and non-Key Persons"

Since the MEPS sample is a subsample of NHIS respondents, the chance of selection for MEPS is directly tied to the chance of selection to the NHIS. The term "Keyness" is related to whether an individual had a chance of being included in MEPS. A person is "Key" if he or she is linked for sampling purposes to the set of NHIS sampled households designated for inclusion in MEPS. Specifically, a "Key" person was a member of an NHIS household at the time of the NHIS interview or a person who became a member of such a household after being out-of-scope at the time the NHIS was fielded for their household (examples of the latter situation include newborns and persons returning from military service, an institution, or living outside the United States).

A "non-Key" person is one whose chance of selection for the NHIS (and MEPS) was associated with a household that was eligible but not sampled for the NHIS, and who later became a member of a MEPS RU. MEPS data are collected for the period of time a "non-Key" person is part of the sampled unit to provide information for family-level analyses. However, "non-Key" persons who leave a sample household unaccompanied by a "Key" in-scope member are not followed for subsequent interviews.

In summary, "Keyness" status is set at the time the person enters MEPS, and a person's "Keyness" status never changes. It should be noted that a person might be "Key" even though they are not part of the civilian noninstitutionalized portion of the U.S. population. For example, a person in the military may have been living with his or her civilian spouse and children in a household sampled for NHIS. The person in the military would be considered a "Key" person for MEPS; however, such a person would not be eligible to receive a person-level sample weight if he or she was never in-scope during a defined survey reference period.

Eligibility

The eligibility of a person for MEPS pertains to whether or not data are to be collected for that person. All of the "Key" in-scope persons of a sampled RU are eligible for data collection. The only "non-Key" persons eligible for data collection are those who live in an RU with at least one "Key" in-scope person and eligibility continues only for the time they are living with at least one such person. The only out-of-scope persons who are eligible for data collection are those living with "Key" in-scope persons, again only for the time they live with such a person (only persons in the military can meet this description, e.g., a person on full-time active military duty, living with a spouse who is Key).

Link of MEPS to the National Health Interview Survey

The set of households selected for each panel of the MEPS-HC is a subsample of households that responded to the previous year's NHIS. The NHIS sampling frame for MEPS is a nationally representative sample of the U.S. civilian noninstitutionalized population. The use of a subsample of NHIS households provides budgetary savings by eliminating the need to independently list and screen households to locate selected policy-relevant subgroups of the population. Moreover, the MEPS to NHIS linkage provides valuable data on characteristics of MEPS nonrespondents that facilitate construction of analytic weights in a manner that reduces MEPS nonresponse bias. The linkage also provides an additional data point for enhanced longitudinal analyses (Cohen SB, 2003; Cohen SB, Makuc DM, Ezzati-Rice TM, 2007).

The records in each MEPS panel can be linked back to the corresponding records in the previous year's NHIS public use data file. For information on obtaining MEPS/NHIS link files, please see https://meps.ahrq.gov/mepsweb/data_stats/download_data_files.jsp. A discussion of weighting and estimation approaches for a MEPS/NHIS link file can be found in Chowdhury et al. (2012), which is available online at https://meps.ahrq.gov/data_files/publications/workingpapers/wp_12005.pdf.

Return to Table of Contents


1.5 Overview of NHIS Sample Design

Since the MEPS sample of households is subsampled from the NHIS, some knowledge of the NHIS sample design is needed to understand the MEPS sample design. The NHIS has been in continuous operation since 1957. Every 10 years, following each Decennial Census, the NHIS sample is redesigned to accommodate the survey requirements and to reflect changes to the U.S. population and its distribution. Increased coverage of the target populations with an updated sample design based on data from the latest census improves the accuracy of the sample estimates. The subsample of households selected for each MEPS panel from 1996–2006 was based on the 1995–2005 NHIS sample design and the subsample of households selected for MEPS panels from 2007–2016 was based on the 2006–2015 NHIS sample design. Detailed information about the NHIS sample designs is available from the NCHS Web site. For example, the NCHS Series 2, Number 130 report describes the sample design of the 1995–2004 NHIS. The design for 1995–2004 was extended another year to cover the 2005 NHIS. The NCHS Series 2, Number 165 report describes the sample design of the 2006–2015 NHIS. A brief overview of the 1995–2005 and 1996–2015 NHIS designs are provided below.

Overview of 1995–2005 NHIS Sample Design

The 1995–2005 NHIS was based on a complex multistage sample design of the U.S. civilian noninstitutionalized population with stratification, clustering, and oversampling of selected population subgroups. The first stage of sample selection was an area sample of Primary Sampling Units (PSUs), with each PSU consisting of a single county or a group of contiguous counties or a metropolitan statistical area (MSA). Those PSUs defining the largest MSAs were selected with certainty and were designated as self-representing (SR) PSUs. The remaining PSUs in the universe were designated as non self-representing (NSR) or non-certainty PSUs and a sample of these PSUs was selected. The NSR PSUs were stratified by state and sampled into the NHIS using a probability proportional to population size according to the 1990 Census. A total of 358 PSUs was selected for the 1995–2005 NHIS.

At the second stage of sampling, the entire area within each selected PSU was partitioned into segments consisting of single or contiguous blocks or block equivalents. These segments served as the Second Stage Units (SSUs). In parts of the country where local governments issue building permits, the area sample was supplemented with a sample of permits for residential housing units built after the Decennial Census in order to produce as current a sample of households as possible. So within each sample PSU, the area segments and permit frame listings defined the SSUs. The area segments were assigned to density strata defined by the concentration of blacks and Hispanics from the 1990 Decennial Census. The building permit frame listing within a PSU was included as its own substratum. The area segments or permit frame listings within each density stratum were grouped into "super-segments" or "super-SSUs" consisting of clusters of housing units. These super-segments were subsequently sampled for use over the life of the planned NHIS design (i.e., a 10-year data collection period that was extended for an additional year to 2005). Reserve samples for 2 additional years were also selected. Households within super-segments were subsequently assigned to each calendar year, quarter, and week of NHIS data collection. Households containing Hispanics and blacks were oversampled at rates of approximately 2 and 1.5 times that of the remaining households, respectively. The oversampling was done at two different levels. First, households in census blocks or "combined blocks" with higher densities of blacks or Hispanics were selected at a higher rate; then, households with one or more black or Hispanic persons were screened in at a higher rate.

The annual NHIS sample of households was partitioned into four sub-designs, referred to as "panels," each with approximately the same number of households (NHIS, Series 2 report). (The word "panel" used for the NHIS in this context should not be confused with the term's more common meaning of a follow-up longitudinal survey like the MEPS.) The two main objectives of the NHIS panel sub-design structure are: 1) to provide nationally representative sub-designs with "similar" features but with smaller sample sizes to deal with any potential NCHS budget exigencies for the NHIS; and 2) to provide a subsample for use as a sampling frame for a smaller "follow-on" survey (i.e., a survey whose sample design is then said to be linked with that of the NHIS) such as MEPS. Panels can be further subdivided by sample assignment weeks (e.g., calendar quarters) to provide even smaller surveys. This sub-design structure also enables NCHS to produce early release estimates (e.g., insurance coverage) prior to the availability of data for all calendar quarters. Each NSR PSU was assigned to one of the fours panels, while each SSU within an SR PSU was allocated to one of the four panels. The panels were identified by panel labels 1, 2, 3, or 4. Typically, a PSU or an SSU is assigned a panel label that remains fixed for the life of the survey. Starting with the 1995–2005 design, two of the four NHIS panels (labelled 1 and 3) were reserved for use by AHRQ for the MEPS.

The NCHS Series 2, Number 130 report provides more details of the NHIS sample design for 1996–2005, and is available at https://www.cdc.gov/nchs/data/series/sr_02/sr02_130.pdf.

Overview of 2006–2015 NHIS Sample Design

As for earlier NHIS sample designs, the 2006–2015 NHIS design was based on a stratified multi-stage sample design. The 2006 NHIS sample of households was sampled independently from that which was selected under the 1995–2005 design. However, the fundamental design structure of the 2006 NHIS sample design was very similar to the previous sample design, which was in place from 1995–2005. The target universe for the 2006–2015 NHIS was also all DUs in the U.S. that contain members of the civilian noninstitutionalized population. As in the previous design, the target universe was first partitioned into PSUs consisting of a single county or a group of contiguous counties or a metropolitan statistical area (MSA). The PSUs containing the largest MSAs were selected with certainty and were designated as SR PSUs. The remaining PSUs were designated as NSR or non-certainty PSUs which were sampled. Most of the SR PSUs of the previous design were still SR in the 2006–2015 design. The differences in the location of the PSUs between the two designs occurred mostly in the NSR areas. The 2006–2015 design had 428 PSUs compared to 358 in the previous design. The difference in the number of PSUs was largely due to differences in how the PSUs were defined. In the 2006–2015 design, most of the SR PSUs in the NHIS were partitioned into mini-PSUs, i.e., smaller geographic areas (one or more counties) known as SPSUs or stratification PSUs, resulting in a larger number of PSUs.

Oversampling of black and Hispanic households was retained in the new 2006 NHIS design to facilitate estimation of health-related statistics for these two minority groups. An oversample of the Asian population was also incorporated. As in the previous design, each selected PSU was partitioned into a substrata consisting of single or contiguous blocks or block equivalents. These substrata were assigned to density strata defined by the concentration of blacks, Hispanics, and Asians from the 2000 Decennial Census while new construction housing (or permit listings) within a PSU was included as its own substratum similar to the previous design. Also, the definitions that constituted low, medium, and high concentrations of each minority group were allowed to vary slightly from PSU to PSU in contrast to the previous design where the definitions were consistent across all PSUs. Finally, as with the previous design, the area segments within each density stratum were partitioned into super-segments or clusters of housing units which were sampled for use over the 10-year period of the 2006–2015 NHIS design. Also, as in the previous design, the annual sample of households was partitioned into four sub-designs, referred to as "panels." Each of these four panels were nationally representative with approximately the same number of households and two (labelled 1 and 4) were reserved for the MEPS.

The NCHS Series 2, Number 165 report provides a more detailed description of the 2006–2015 NHIS sample design and is available at https://www.cdc.gov/nchs/data/series/sr_02/sr02_165.pdf.

Return to Table of Contents


1.6 MEPS Sample Designs

1996–2006 MEPS Sample Design and Selection (Panels 1–11)

For the 1996–2006 MEPS design (i.e., MEPS Panels 1–11), households were selected from 195 of the 358 NHIS sampled PSUs (except in 1999 and 2000 when only 100 PSUs were used due to budgetary constraints). In 2002, some design enhancements were made with an increased sample size and the number of PSUs used was brought back up to 195 as in the earlier MEPS panels. The sample was generally selected from the first three calendar quarters of households from the two NHIS panels allocated for MEPS (NHIS Panels 1 and 3) as was done for the 1995–2005 NHIS design.

The MEPS sampling frame includes NHIS households that meet the following criteria:

  • Responding Household: NHIS household with ACTION code 10 (complete interview) or ACTION code 4 (partial interview, no follow-up) and at least one person with HHSTAT (household status) not equal to D (deleted).
  • Eligible Household: All responding households in the NHIS quarters and panels set aside for MEPS. (In most years, the households eligible for MEPS are from calendar quarters 1–3 in 2 of 4 NHIS panels.)

Prior to sample selection, the NHIS-occupied DUs within pre-specified sampling domains are hierarchically sorted by the following measures:

  • Calendar year quarter
  • Interview week within each respective calendar quarter
  • Census division
  • State
  • MSA classification
  • NHIS PSU
  • NHIS segment within PSU

The available NHIS households on the MEPS frame were stratified hierarchically into mutually exclusive sampling domains that varied slightly from year to year (see Tables 1 and A1). For example, in the years when Asian, Hispanic, black, and Other domains were used, the domains were created hierarchically as follows:

  1. If a household contained any Asian member, the entire household is classified as an Asian household.
  2. Among the remaining households, if a household contained any Hispanic member, the household is classified as a Hispanic household.
  3. Then, among the remaining households, if a household contained any member classified as black, the household is classified as a black household.
  4. Finally, if a household is not in any of the three previous strata it is classified as an Other household.

The initial sample size for each panel of the MEPS was determined on the basis of the budget resources available at the time of sample selection and on the eligible sample available from NCHS. As a result, sample sizes have varied from year to year. In the next paragraph, a review of the 1996 and 1997 MEPS is provided, followed by more specific details for the 1998–2006 MEPS.

The 1995 NHIS subsample selected for the 1996 MEPS (Panel 1) consisted of 195 PSUs. An initial subsample of 10,597 households was selected from NHIS Panels 1 and 3 in two targeted quarters (2 and 3) of the NHIS (Table 2). The 1997 MEPS panel (Panel 2) sample of 6,300 households was selected in the same 195 PSUs (as in MEPS 1996) and was selected as a subsample of households responding to the 1996 NHIS Panels 1 and 3 in three NHIS quarters (1, 2, and 3). Both the 1996 and 1997 MEPS reflected an inherited oversample of Hispanics and blacks at the same ratios as in the NHIS (Hispanics, 2.0:1; blacks 1.5:1). In addition, the 1997 MEPS oversampled several policy-relevant domains (see Table 1).

Table 1. Summary of Oversampling Domains in MEPS 1996–2006
Year Panel Oversampling Domains1
1996 1 - - - - -
1997 2 18+ yrs with limitations 18-64 yrs with predicted high medical expenditures <18 yrs with limitations Predicted poor Persons with other limitations
1998 3 - - - - -
1999 4 - - - - -
2000 5 - - - - -
2001 6 - - - - -
2002 7 Asian/Predicted poor - - - -
2003 8 Asian/Predicted poor - - - -
2004 9 Asian/Predicted poor - Black - -
2005 10 Asian/Predicted poor - Black - -
2006 11 Asian/Predicted poor - Black - -

1 See Appendix Table A1 for detailed definitions of domains.

The sample of households for MEPS panels 3–6 (1998–2001) ranged from 5,166 to 10,704 households (Table 2). The sample of households for MEPS panels 7–11 (2002–2006) from the 2001–2005 NHIS, ranged from 8,132 to 9,464 households. As in the earlier years, the oversample of Hispanics and blacks in the NHIS carried over to MEPS. In addition, the NHIS responding households eligible for MEPS that contained either Asian Americans or families predicted (based on a statistical model) to be "poor" (i.e., income <200 percent of the federal poverty level) were selected with certainty. For MEPS panels 9–11 (2004–2006), in addition to the certainty selection of Asians and predicted poor families, households containing blacks that were not among those households selected with certainty were further oversampled with certainty, in addition to the earlier oversampling of these households in NHIS. The sampling rates by subdomains, the number of selected NHIS households, number of PSUs, and number of initial MEPS RUs by MEPS panel and year are shown in Table A1. (Note: The sample sizes presented in Table 2 are confined to the new panel introduced each year. The number of responding families and persons used in producing annual estimates from two consecutive panels are shown in Table 5.)

Table 2. MEPS Households Selected from the Frame (NHIS)1 for MEPS 1996–2006
Year Panel Frame/Sample Total Sampling Domain
Asian/Poor Black Other
1996 P1 Frame 10,597 NA NA NA
Sample 10,597 NA NA NA
% Selected 100% NA NA NA
1997 P2 Frame 14,706 footnote 2 footnote 2 footnote 2
Sample 6,300 footnote 2 footnote 2 footnote 2
% Selected 42.8% footnote 2 footnote 2 footnote 2
1998 P3 Frame 5,166 NA NA NA
Sample 5,166 NA NA NA
% Selected 100% NA NA NA
1999 P4 Frame 7,301 NA NA NA
Sample 6,900 NA NA NA
% Selected 94.5% NA NA NA
2000 P5 Frame 7,263 NA NA NA
Sample 5,380 NA NA NA
% Selected 74.1% NA NA NA
2001 P6 Frame 14,508 NA NA NA
Sample 10,704 NA NA NA
% Selected 73.8% NA NA NA
2002 P7 Frame 14,510 2,671 NA 11,839
Sample 8,132 2,671 NA 5,461
% Selected 56.0% 100% NA 46.1%
2003 P8 Frame 13,628 2,448 NA 11,180
Sample 8,400 2,448 NA 5,952
% Selected 61.6% 100% NA 46.1%
2004 P9 Frame 13,618 2,566 1,394 9,658
Sample 8,640 2,566 911 5,163
% Selected 61.6% 100% 65.4% 53.5%
2005 P10 Frame 13,218 2,575 1,387 9,256
Sample 8,546 2,575 1,040 4,931
% Selected 64.6% 100% 75.0% 53.3%
2006 P11 Frame 14,224 2,685 1,474 10,065
Sample 9,464 2,685 1,106 5,673
% Selected 66.5% 100% 75.0% 56.4%

1 MEPS frame includes 3/8 of the NHIS full responding sample.
2 See Appendix Table A.1 for domains used in 1997.
NA: not applicable

2007–2016 MEPS Sample Design and Sample Selection (Panels 12–21)

For the 2007–2016 MEPS design (Panels 12–21), 183 PSUs were used in selecting the MEPS sample from the 428 total NHIS PSUs. NHIS Panels 1 and 4 were set aside for use in the 2007–2016 MEPS design to maximize the number of overlapping PSUs utilized from the previous 1996–2006 design and thereby mitigate increased MEPS data collection costs.

Except for Panel 12, which was sampled from only 2 quarters, each new MEPS panel was sampled from the first 3 calendar quarters of the prior year's responding NHIS annual sample among the two NHIS panels (1 and 4). To reduce operational issues associated with fielding a new sample design in the same year as implementation of a new Windows-based CAPI instrument, the MEPS sample was limited to eligible responding housing units from the first two calendar quarters of the NHIS Panels 1 and 4 of the 2006 NHIS.

The frame for selecting the MEPS sample was created the same way it was created for panels in the MEPS 1996–2006 sampling. All responding households in NHIS quarters 1–3 within NHIS panels 1 and 4 were used as the frame for the 2007–2016 MEPS design. All households with complete interview (ACTION code 10), or partial interview (ACTION code 4) and at least one person with HHSTAT (household status) not equal to D (deleted) were included on the frame.

The household-level sampling domain variables varied only slightly from year to year (see Tables 3 and A1). Also, to increase operational efficiency, since Panel 16, the "Other" domain has been separated into two domains: households with a complete response in NHIS (Other-complete) and households with a partial response in NHIS (Other-partial). Since the households in the Other-partial domain are generally less cooperative and require extra follow up, these households are selected at a lower rate than the complete households to make the design more responsive (Mirel LB and Chowdhury SR, 2017). The sample was selected independently within each domain. Similar to the previous design, the available NHIS households on the MEPS frame were stratified hierarchically into mutually exclusive sampling domains.

Up to 2009 (MEPS Panel 14), the sample of DUs was selected the same way it was selected in the previous design. DUs were selected systematically with equal weight within each sampling domain after sorting hierarchically by the same set of variables used for the panels in the previous design.

Starting with 2010 (MEPS Panel 15) a probability proportional to size (PPS) sampling was introduced to select households from NHIS to MEPS in non-certainty domains (i.e., non-minority Other households). The purpose of the PPS sampling is to reduce the variation in MEPS base weights and thereby increase the precision of MEPS estimates (Chowdhury SR and Baskin RM, 2014). However, since the sampling rate is 100 percent in minority domains, the PPS sampling is applied to reduce the variation in base weights only in the Other-complete and Other-partial domains. For the PPS selection, since the NHIS final household weight is not available when the MEPS sample needs to be selected, either the interim or a predicted version of the NHIS household weight is used as the measure of size. The interim NHIS household weight is the NHIS final household base weight without the nonresponse adjustment. The sort variables used in the PPS sampling were: calendar year quarter, census region, state, MSA classification, NHIS PSU, within PSU stratum code for NHIS, and NHIS segment.

During the 2006–2016 period, the MEPS sample size ranged from 7,319 to 10,610 households (Table 4). As in the earlier years, the Hispanic and black households were oversampled. Households that contained either Asians or families predicted as poor (i.e., income <200 percent of the federal poverty level) were selected with certainty in 2007 and 2008 Panels 12–13) but households predicted as poor were no longer oversampled in subsequent panels, e.g., Panels 14–21. Only Asian, Hispanic, and black households were oversampled in 2009–2016, except in 2011 when the households with one or more cancer patients (based on prior year NHIS data) were also oversampled (Table 3). A household with one or more members with cancer in the household was defined as a Cancer household. A self-administered questionnaire supplement on cancer (termed the CSAQ) was implemented on cancer survivors. All oversampled domains mentioned above were sampled for MEPS with certainty from the frame. Only the households in the Other domain (which was separated into Other-complete and Other-partial starting in 2011; see Table 4) were not sampled with certainty. The sampling rates by subdomains, number of NHIS households selected, number of PSUs, and number of initial MEPS RUs by MEPS panel and year are shown in Table A1. (Note: The sample sizes presented in Tables 4 and A1 are confined to the new panel introduced each year.)

Table 3. Summary of Oversampling Domains in MEPS 2007–2016
Year Panel Oversampling Domains1
2007 12 Asian/Predicted Poor Hispanic Black - -
2008 13 Asian/Predicted Poor Hispanic Black - -
2009 14 Asian Hispanic Black - -
2010 15 Asian Hispanic Black - -
2011 16 Asian Hispanic Black Other-Complete/Partial2 Cancer
2012 17 Asian Hispanic Black Other-Complete/Partial -
2013 18 Asian Hispanic Black Other-Complete/Partial -
2014 19 Asian Hispanic Black Other-Complete/Partial -
2015 20 Asian Hispanic Black Other-Complete/Partial -
2016 21 Asian Hispanic Black Other-Complete/Partial -

1 See Appendix Table A1 for detailed definitions of domains.
2 Since 2011, the Other household domain was separated into Other-complete and Other-partial domains, and the Other-complete domain was sampled at a higher rate than the Other-partial domain to reduce data collection effort.

Table 4. MEPS Households Selected from the Frame (NHIS)1 for MEPS 2007–2016
Year Panel Frame/Sample Total Asian/Poor/Cancer2 Sampling Domain
Hispanic Black Other-Total Other-Complete Other-Partial
2007 P12 Frame 8,055 1,723 970 884 4,478 NA NA
Sample 7,319 1,723 970 884 3,742 NA NA
% Selected 90.9% 100% 100% 100% 83.6% NA NA
2008 P13 Frame 12,452 2,632 1,855 1,498 6,467 NA NA
Sample 9,703 2,632 1,855 1,498 3,718 NA NA
% Selected 77.9% 100% 100% 100% 57.5% NA NA
2009 P14 Frame 12,181 834 2,386 2,019 6,942 NA NA
Sample 9,700 834 2,066 1,816 4,984 NA NA
% Selected 79.6% 100% 86.6% 89.9% 71.8% NA NA
2010 P15 Frame 12,390 860 2,579 1,994 6,957 NA NA
Sample 8,750 860 1,961 1,705 4,224 NA NA
% Selected 70.6% 100% 76.0% 85.5% 60.7% NA NA
2011 P16 Frame 12,067 1,708 2,386 1,894 NA 4,272 1,807
Sample 10,180 1,708 2,386 1,894 NA 3,354 838
% Selected 84.4% 100% 100% 100% NA 79% 46%
2012 P17 Frame 13,701 1,075 2,762 2,053 NA 6,426 1,385
Sample 9,700 1,075 2,762 2,053 NA 3,256 554
% Selected 70.8% 100% 100% 100% NA 51% 40%
2013 P18 Frame 12,565 1,018 2,641 1,990 NA 5,342 1,574
Sample 9,700 1,018 2,641 1,990 NA 3,368 683
% Selected 77.2% 100% 100% 100% NA 63.1% 43.4%
2014 P19 Frame 12,313 976 2,628 2,002 NA 5,360 1,347
Sample 9,700 976 2,628 2,002 NA 3,531 563
% Selected 78.8% 100% 100% 100% NA 65.9% 41.8%
2015 P20 Frame 12,109 954 2,706 1,881 NA 5,117 1,451
Sample 10,610 954 2,706 1,881 NA 4,304 765
% Selected 87.6% 100% 100% 100% NA 84.1% 52.7%
2016 P21 Frame 11,336 963 2,518 1,801 NA 4,551 1,503
Sample 9,700 963 2,518 1,801 NA 3,685 733
% Selected 85.6% 100% 100% 100% NA 81.0% 48.8%

1 MEPS frame includes 3/8 of the NHIS full responding sample.
2 Poor households in Panel 13 only and Cancer households in Panel 16 only
NA: not applicable

Return to Table of Contents


1.7 Sample Yields and Response Rates in MEPS

Responding Sample Size

As mentioned earlier, to produce MEPS annual calendar year (full-year) estimates, data are combined across two overlapping panels. An early data file is also produced in MEPS, which allows selected "point-in-time" estimates for the early part of the year to be produced. Table 5 provides a summary of the number of completed interviews (DUs, families, and persons) by year and panel based on the annual Full-Year Consolidated Files. Tables A2 and A3 in Appendix A provide the number of completed person-level interviews by age group, gender, race/ethnicity, region, and MSA status for annual full-year files.

Table 5. Number of responding dwelling units, families, and persons by year and panel in MEPS Household Component Full-Year Files, 1996–2016
Year Panel Dwelling Units Families1 Persons
1996 Panel 1 8,095 8,588 21,571
1997 Combined 12,043 12,986 32,636
Panel 1 7,366 7,925 19,622
Panel 2 4,677 5,061 13,014
1998 Combined 8,318 8,920 22,953
Panel 2 4,408 4,756 12,260
Panel 3 3,910 4,164 10,693
1999 Combined 8,671 9,278 23,565
Panel 3 3,639 3,925 9,979
Panel 4 5,032 5,353 13,586
2000 Combined 8,849 9,437 23,839
Panel 4 4,850 5,195 13,170
Panel 5 3,999 4,242 10,669
2001 Combined 11,864 12,732 32,122
Panel 5 3,836 4,114 10,298
Panel 6 8,028 8,618 21,824
2002 Combined 13,689 14,712 37,418
Panel 6 7,677 8,326 20,890
Panel 7 6,012 6,386 16,528
2003 Combined 11,929 12,742 32,681
Panel 7 5,771 6,147 16,000
Panel 8 6,158 6,595 16,681
2004 Combined 12,043 12,917 32,737
Panel 8 5,910 6,358 16,058
Panel 9 6,133 6,559 16,679
2005 Combined 11,918 12,680 32,320
Panel 9 5,832 6,278 15,904
Panel 10 6,086 6,402 16,416
2006 Combined 12,127 12,729 32,577
Panel 10 5,725 6,056 15,458
Panel 11 6,402 6,673 17,119
2007 Combined 11,043 11,516 29,370
Panel 11 6,118 6,391 16,355
Panel 12 4,925 5,125 13,015
2008 Combined 11,516 12,228 31,262
Panel 12 4,655 4,890 12,314
Panel 13 6,861 7,338 18,948
2009 Combined 12,901 13,780 34,920
Panel 13 6,542 7,076 18,075
Panel 14 6,359 6,704 16,845
2010 Combined 11,734 12,374 31,228
Panel 14 6,069 6,416 16,055
Panel 15 5,665 5,958 15,173
2011 Combined 12,551 13,380 33,622
Panel 15 5,390 5,717 14,370
Panel 16 7,161 7,663 19,252
2012 Combined 13,694 14,678 37,182
Panel 16 6,833 7,377 18,313
Panel 17 6,861 7,301 18,869
2013 Combined 12,885 13,885 35,068
Panel 17 6,504 7,018 17,745
Panel 18 6,381 6,867 17,323
2014 Combined 12,346 13,334 33,162
Panel 18 6,134 6,699 16,579
Panel 19 6,212 6,635 16,583
2015 Combined 12,886 13,713 33,893
Panel 19 5,929 6,402 15,730
Panel 20 6,957 7,311 18,163
2016 Combined 12,884 13,492 33,259
Panel 20 6,535 6,905 16,823
Panel 21 6,349 6,587 16,436

1 Families defined by variable FAMID[yy] where FMRS1231 = 1

Design Effects

While sample size is often used as an indicator of the precision of estimates obtained from a survey, the sample size of one survey is not directly comparable to that of another survey due to complexities of corresponding survey designs. The effectiveness of a sample size for precision of estimates in complex sample surveys such as the MEPS or the NHIS depends on many other factors that relate to design specifications including the stratification, clustering, and multi-stage structure of the sample. The NHIS sample is clustered and has multiple stages of sampling, including oversampling of selected minority groups. These NHIS design specifications are an inherited and integral part of the MEPS sample design. Moreover, MEPS households are often selected from the NHIS frame with unequal probabilities, and there is nonresponse to the NHIS and then to the first MEPS interview, and attrition in subsequent rounds. MEPS survey weights account for both NHIS and MEPS differential sampling probabilities and nonresponse. Moreover, the construction of survey weights involves post-stratification to external control totals. All these adjustments make the MEPS weights highly unequal, which also affects the precision of estimates. Estimates based on a complex sample design like MEPS usually have a higher variance than they would if based on a simple random sample design due to clustering, multistage sample selection, and unequal weighting. For a fixed sample size, the ratio of the variance of an estimate under a complex design to the variance of the same estimate under a simple random sample design is called the design effect. To compare sample sizes under different designs, the sample size in a complex design can be divided by the design effect to get the statistically equivalent sample size (i.e., effective sample size) under a simple random sample design. Table 6 provides design effects for two main MEPS estimates (percent insured and mean healthcare expense per person) in the 1996–2006 and 2007–2016 MEPS designs. Design effects appear to be slightly higher in the second design than in the first design. For percent insured, the average design effect is 3.65 in the first design and 4.85 in the second design, and, for mean expense, the average design effect is 2.07 in the first design and 2.38 in the second design. Even under the same design, design effects differ across variables due to variations in distributions and clustering of the variables. Design effects can also vary from year to year for the same variable because of the randomness of the sample across years.

Table 6. Design effects of selected variables in MEPS, 1996–2006 and 2007–2016
Year Insured (<65 yrs of age) Total Expenses
Percent Design Effect Mean Per Person ($) Design Effect
First design
1996 87.8 3.7 2,038 1.6
1997 87.7 3.9 2,039 2.2
1998 88.1 3.7 2,049 1.9
1999 89.1 3.1 2,156 2.1
2000 88.3 4.5 2,255 2.0
2001 88.3 3.6 2,555 1.9
2002 88.2 4.0 2,813 2.1
2003 87.9 3.2 3,082 2.4
2004 87.7 3.8 3,284 2.4
2005 87.6 3.2 3,457 2.5
2006 87.4 3.4 3,452 1.7
Average1 88.01 3.65 2,652 2.07
Second design
2007 86.7 3.4 3,737 1.8
2008 86.6 4.4 3,773 2.2
2009 86.5 5.9 4,107 2.8
2010 86.9 4.6 4,094 2.1
2011 87.5 4.3 4,277 1.8
2012 87.3 5.7 4,309 3.0
2013 87.2 6.0 4,436 2.5
2014 90.2 5.2 4,708 2.6
2015 91.9 4.1 4,978 2.8
2016 92.4 4.9 5,006 2.2
Average1 88.32 4.85 4,342 2.38

1 Simple unweighted average across years

Response Rates

Because of the linkage between the NHIS and the MEPS, the overall response rate for MEPS is a combination of the response rate in the NHIS and round-specific response rates in MEPS. Also, since the sample size in MEPS full-year files include samples from two overlapping panels, the calculation of the final annual response rate involves applying composite factors for each panel to panel-specific response rates (the same compositing factors are used to derive the annual estimation weight for the combined panels). Appendix B (which includes Table B1) illustrate the procedure used in calculating the 2016 final MEPS response rate.

The NHIS, MEPS panel-specific, and MEPS overall (i.e., combined panel) response rates for each estimation year from 1996–2006 are shown in Table 7. These response rates are unweighted and reflect response to both the NHIS and the multiplicative MEPS round-specific response rates (see Appendix B). The overall MEPS response rate has decreased steadily over the 20-year period due to the combined impact of declines in the NHIS response rate (from about 94 to 71 percent) and the MEPS conditional response rate (from about 75 to 63 percent). It should be noted that the standardized response rates shown in Table 7 for 1996–2000 are slightly different from those in the 1996 and 1997 MEPS Sample Design Reports (See Cohen S, 1997, and Cohen S, 2000) and in the public use file documentation due to a slight difference in the calculation methods.

Table 7. MEPS individual panel and combined annual response rates
Calendar Year Response Rate (%)
Year 1 Panel Year 2 Panel Combined Overall
NHIS MEPS1 Compositing Factor2 NHIS MEPS1 Compositing Factor2 NHIS MEPS1 MEPS Final
1996 93.9 74.8 NA NA NA NA 93.9 74.8 70.2
1997 93.8 73.8 0.50 93.9 67.6 0.50 93.9 70.7 66.4
1998 93.3 75.9 0.50 93.8 69.3 0.50 93.6 72.6 67.9
1999 92.2 71.0 0.50 93.3 67.6 0.50 92.8 69.3 64.3
2000 92.2 74.1 0.55 92.2 69.1 0.45 92.2 71.3 65.8
2001 89.9 74.3 0.33 92.2 70.9 0.67 90.7 73.2 66.3
2002 89.7 73.1 0.55 89.9 71.2 0.45 89.8 72.1 64.7
2003 90.6 72.7 0.49 89.7 70.1 0.51 90.2 71.5 64.4
2004 90.3 70.3 0.49 90.3 69.2 0.51 90.4 69.8 63.1
2005 87.9 70.8 0.50 90.3 66.8 0.50 89.1 68.8 61.3
2006 87.3 66.7 0.47 87.9 66.6 0.53 87.6 66.6 58.3
2007 88.1 63.3 0.56 87.3 63.5 0.44 87.7 63.4 55.6
2008 87.4 70.7 0.39 88.1 62.9 0.61 87.7 67.7 59.3
2009 85.2 65.1 0.52 87.4 67.2 0.48 86.3 66.2 57.2
2010 84.0 64.6 0.51 85.2 62.0 0.49 84.6 63.3 53.5
2011 80.6 71.1 0.43 84.0 61.4 0.57 82.1 66.9 54.8
2012 82.9 70.1 0.49 80.6 67.5 0.51 81.8 68.8 56.3
2013 78.0 65.1 0.51 82.9 66.0 0.49 80.5 65.6 52.8
2014 76.2 63.6 0.50 78.0 62.3 0.50 77.1 63.0 48.6
2015 75.1 65.2 0.46 76.2 60.5 0.54 75.6 63.1 47.7
2016 71.2 65.0 0.51 75.1 60.9 0.49 73.2 62.9 46.0

1 Response rate in MEPS only, i.e., conditional on NHIS
2 Factor used for the panel to derive composite weight when two panels are combined to produce annual estimates
NA: not applicable

Return to Table of Contents


1.8 Analysis Weights and Variance Estimation

Development of Analysis Weights

Weights are developed for use in the derivation of nationally representative population estimates based on complex national sample surveys like MEPS. The analytic weights typically account for any disproportionate probabilities of selection, unit nonresponse, and an adjustment to make sure that the weighted sample distributions agree with known population totals. In general, the development of MEPS weights for a new panel involves a series of derivations and adjustments starting with the household sample selection base weight derived from the NHIS final household weight followed by an adjustment for DU nonresponse at MEPS Round 1, an adjustment for nonresponse at the person level to account for survey attrition across the multiple rounds of data collection, and a final step of post stratification/raking adjustments to known population totals for the U.S. civilian noninstitutionalized population. Similar procedures are applied to both panels except that the starting weight for the second year panel is the final weight from the first year of MEPS. Then, a combined-panel weight is derived by compositing the weights of both panels. More detailed information on the weight construction procedures used in MEPS can be found in Machlin, Chowdhury, et al. (2010), Wun et al. (2007), and on the MEPS Web site at https://meps.ahrq.gov.

Variance Estimation

Because MEPS is based on a complex-probability sample design, analytic approaches developed for data from a simple random sample are not appropriate for MEPS. Standard statistical estimation software procedures that assume simple random sampling do not properly account for the variability in the MEPS estimates attributable to the complexity in the MEPS design and estimation procedures. Estimates of variability for MEPS estimates (such as the standard error or corresponding confidence interval) should be calculated by using appropriate statistical software procedures. Several methodologies have been developed for estimating standard errors for surveys with a complex sample design, including the Taylor-series linearization method, balanced repeated replication, and jackknife replication. A variety of software packages provide analysts with the capability of implementing these methodologies. Software packages that permit the use of the Taylor-series linearization method are SUDAAN, Stata svy commands, SAS Survey Procedures (SAS version 8.2 and higher), the R survey package, and SPSS (version 12.0 and higher). Users of these software packages should refer to the corresponding software user documentation for complete information on the capabilities of each package. The three variables on MEPS public use files needed to calculate appropriate Taylor-series standard errors for estimates are variance strata (VARSTR), variance PSU (VARPSU), and the survey weight (e.g., PERWTyyF). More information and examples of statistical software programming code can be found at https://meps.ahrq.gov/mepsweb/survey_comp/clustering_faqs.jsp.

Return to Table of Contents


1.9 References

Chowdhury, S., Machlin, S., Wun, L. Linking Medical Expenditure Panel Survey to the National Health Interview Survey: Weighting and Estimation. MEPS Working Paper No. 12005. August 2012. Agency for Healthcare Research and Quality, Rockville, MD. https://gold.ahrq.gov

Chowdhury, SR and Baskin RM. PPS Subsampling from NHIS to MEPS - Effect on Precision of MEPS Estimates. Proceedings of the Section on Survey Research Methods, Alexandria, VA: American Statistical Association (CD-ROM), 2014.

Cohen, S. Sample Design of the 1996 Medical Expenditure Panel Survey Household Component. MEPS Methodology Report No. 2. AHCPR Pub. No. 97-0027. Rockville, MD. Agency for Health Care Policy and Research, 1997. https://meps.ahrq.gov/data_files/publications/mr2/mr2.shtml

Cohen, S. Sample Design of the 1997 Medical Expenditure Panel Survey Household Component. Methodology Report No. 11. November 2000. Agency for Healthcare Research and Quality, Rockville, MD. https://meps.ahrq.gov/data_files/publications/mr11/mr11.shtml

Cohen SB. Design Strategies and Innovations in the Medical Expenditure Panel Survey. Medical Care, July 2003: 41(7) Supplement: III-5–III-12.

Cohen SB, Makuc DM, and Ezzati-Rice TM. Health insurance coverage during a 24 month period: a comparison of estimates from two national health surveys. Journal of Health Services and Outcomes Research Methodology (2007) 7:125–144.

Cohen SB and Ezzati-Rice TM. (2006). Designing national health care surveys to inform health policy. In Statistics: A Guide to the Unknown, Fourth Edition, 89–101. Belmont (CA): Duxberry of Thomson Brooks/Cole.

Ezzati-Rice, T.M., Rohde, F., Greenblatt, J. Sample Design of the Medical Expenditure Panel Survey Household Component, 1998–2007. Methodology Report No. 22. March 2008. Agency for Healthcare Research and Quality, Rockville, MD. https://meps.ahrq.gov/data_files/publications/mr22/mr22.shtml

Machlin, S.R., Chowdhury, S.R., Ezzati-Rice, T., DiGaetano, R., Goksel, H., Wun, L.-M., Yu, W., Kashihara, D. Estimation Procedures for the Medical Expenditure Panel Survey Household Component. Methodology Report No. 24. September 2010. Agency for Healthcare Research and Quality, Rockville, MD. https://meps.ahrq.gov/data_files/publications/mr24/mr24.shtml

Mirel LB and Chowdhury SR. Using linked survey paradata to improve sampling strategies in the Medical Expenditure Panel Survey. Journal of Official Statistics, 2017: Vol. 33, No. 2, 367–384. http://dx.doi.org/10.1515/JOS-2017-0018

Wun LM, Ezzati-Rice TM, Diaz-Tena N, and Greenblatt J. On modeling response propensity for dwelling unit (DU) level non-response adjustment in the Medical Expenditure Panel Survey (MEPS). Statistics in Medicine, 2007: 26:1875–1884.

Return to Table of Contents


Appendix A. MEPS sample sizes by selected characteristics, 1996–2016

Definitions for Table A1:

NHIS Responding Household. A household (HH) with ACTION = 10 (complete interview) or 4 (partial interview, no follow-up) and at least one person with HHSTAT ne D (deleted) is defined as responding for creating the MEPS sampling frame.

Eligible Household. NHIS responding HH in quarters and panels assigned for MEPS. In the 2007 MEPS sample, responding HHs are also eligible HHs, because the NHIS data comprised only the "MEPS" panels (for 2007, panels 1 and 4).

Sampling Domain. An HH classification to describe persons or reporting units (RUs) in the HH based on information collected in the NHIS. There were no domains used for sampling from 1996–2001 except in 1997 when the sample was selected from seven domains. Beginning with the 2002 MEPS, samples were drawn from three or more domains.

The 1997 MEPS was sampled from the following seven hierarchical domains, i.e., an HH in a specified domain would not have been sampled in a previously listed domain:

Domain Definition
Adult ADL At least one adult in HH has an ADL or IADL ((AGE = 70+ & ADLIADL1 = 1) or (AGE = 18-69 & ADLIADL2 = 1)).
Child At least one child in HH has a limitation ((AGE = 0-4 & LIMCHILD = Limit 0,1) or (AGE = 5-17 & SCHOOLAT = 1-4) or (AGE = 0-17 & ANYLIMOT = 1)).
High Expense At least one adult in HH (age 18-64) has probability >.40 of high expenditures.
Predicted Poor At least one person in HH has probability >.30 of low family income.
Other Limit At least one adult in HH has an "other limitation"
(AGE = 18-69 & ADLIAD2X = 2 & WORKLIMT = 1) or
(AGE = 70+ & ADLIAD1X = 2).
Age 65 At least one person in HH is age 65 or older.
Other HH is not in another domain.

Beginning with the 2002 MEPS sample, the following domains have been defined:

Domain Definition
Asian At least one person in HH is Asian (RACE_12 = 5-15 in 2002-2004 MEPS. Beginning with 2005 MEPS: NEWRACE = 3 [or RACE_1 = 9-15]).
Predicted Poor At least one RU in HH has probability of poverty >.30; probability was constructed from the regression model that was developed for the 1997 MEPS sampling.
Hispanic HH is not Asian, not poor, has at least one Hispanic person (NATOR = 1). First used for 2007 MEPS sample.
Black HH is not Asian, not poor, not Hispanic, has at least one black person.

First used for 2004 MEPS (defined as RACE_12 = 2). Beginning with 2005 MEPS: NEWRACE = 2 [or RACE_1 = 2]).
Other HH is not in another domain.

Beginning with the 2009 MEPS sample, the "Predicted Poor" domain was no longer used.

Beginning with the 2011 MEPS sample, the Other category was subdivided into Other-complete and Other-partial:

Domain Definition
Other-complete For Other, distinguish those HH with all complete interviews (defined as ACTION = 10 (OUTCOME = 201)).
Other-partial For Other, distinguish those HHs with at least one partial interview (defined as ACTION = 4 (OUTCOME = 203)).

In 2011 only, a Cancer domain was also used:

Domain Definition
Cancer At least one adult in HH has reported cancer (defined as AGE GE 18 AND (CANEV = 1 OR LAHCA12 = 1)).

Return to Table of Contents


Table A1. Sample selection details of MEPS from NHIS, 1996–2016 (Panels 1–21)
MEPS Year/ Panel NHIS Quarters/ Panels Sampling Domain Household Eligible for Sampling Sampling Rate Household Sampled
1996/P1 Q2,3/P1,3 TOTAL N/A - 10,597
1997/P2 Q1,2,3/P1,3 TOTAL 14,706 0.428 6,300
Adult ADL 478 1.000 478
Child Limitation 601 1.000 601
High Expenses 596 1.000 596
Predicted Poor 2,064 0.600 1,238
Other Limitation 324 0.600 194
Age 65 2,157 0.300 647
Other 8,486 0.300 2,546
1998/P3 Q2/P1,3 TOTAL 5,166 1.000 5,166
1999/P4 Q1,2,3/P1 TOTAL 7,301 0.945 6,900
2000/P5 Q1,2,3/P1 TOTAL 7,263 0.741 5,380
2001/P6 Q1,2,3/P1,3 TOTAL 14,508 0.738 10,704
Q1,2 - 9,338 0.750 7,004
Q3 - 5,170 0.716 3,700
2002/P7 Q1,2,3/P1,3 TOTAL 14,510 0.560 8,132
Q1,2 Asian/Predicted Poor 1,718 1.000 1,718
Other 7,563 0.500 3,781
Total 9,281 0.592 5,499
Q3 Asian/Predicted Poor 953 1.000 953
Other 4,276 0.393 1,680
Total 5,229 0.504 2,633
Total Asian/Predicted Poor 2,671 1.000 2,671
Other 11,839 0.461 5,461
2003/P8 Q1,2,3/P1,3 TOTAL 13,628 0.616 8,400
Q1,2 Asian/Predicted Poor 1,623 1.000 1,623
Other 7,395 0.500 3,698
Total 9,018 0.590 5,321
Q3 Asian/Predicted Poor 825 1.000 825
Other 3,785 0.595 2,254
Total 4,610 0.668 3,079
Total Asian/Predicted Poor 2,448 1.000 2,448
Other 11,180 0.532 5,952
2004/P9 Q1,2,3/P1,3 TOTAL 13,618 0.634 8,640
Q1,2 Asian/Predicted Poor 1,516 1.000 1,516
Black 853 0.750 640
Other 5,923 0.600 3,554
Total 8,292 0.689 5,710
Q3 Asian/Predicted Poor 1,050 1.000 1,050
Black 541 0.500 271
Other 3,735 0.431 1,609
Total 5,326 0.550 2,930
Total Asian/Predicted Poor 2,566 1.000 2,566
Black 1,394 0.653 911
Other 9,658 0.535 5,163
2005/P10 Q1,2,3/P1,3 TOTAL 13,218 0.646 8,546
Q1,2 Asian/Predicted Poor 1,619 1.000 1,619
Black 823 0.750 617
Other 5,658 0.600 3,395
Total 8,100 0.695 5,631
Q3 Asian/Predicted Poor 956 1.000 956
Black 564 0.750 423
Other 3,598 0.427 1,536
Total 5,118 0.569 2,915
Total Asian/Predicted Poor 2,575 1.000 2,575
Black 1,387 0.750 1,040
Other 9,256 0.533 4,931
2006/P11 Q1,2,3/P1,3 TOTAL 14,224 .665 9,464
Q1,2 Asian/Predicted Poor 1,726 1.000 1,726
Black 930 .750 698
Other 6,404 .600 3,842
Total 9,060 .692 6,266
Q3 Asian/Poor 959 1.000 959
Black 544 .750 408
Other 3,661 .500 1,831
Total 5,164 .619 3,198
Total Asian/Predicted Poor 2,685 1.000 2,685
Black 1,474 .750 1,106
Other 10,065 .564 5,673
2007/P12 Q1,2/P1,4 TOTAL 8,055 .909 7,319
Q1 Asian/Predicted Poor 780 1.000 780
Hispanic 505 .750 378
Black 432 .750 324
Other 2,002 .750 1,501
Total 3,719 .802 2,983
Q2 Asian/Predicted Poor 943 1.000 943
Hispanic 592 1.000 592
Black 560 1.000 560
Other 2,241 1.000 2,241
Total 4,336 1.000 4,336
Total Asian/Predicted Poor 1,723 1.000 1,723
Hispanic 1,097 .884 970
Black 992 .891 884
Other 4,243 .882 3,742
2008/P13 Q1,2,3/P1,4 TOTAL 12,452 .779 9,703
Q1,2 Asian/Predicted Poor 1678 1.000 1678
Hispanic 1194 1.000 1194
Black 966 1.000 966
Other 4,157 .600 2,494
Total 7,995 .792 6,332
Q3 Asian/Predicted Poor 954 1.000 954
Hispanic 661 1.000 661
Black 532 1.000 532
Other 2,310 .530 1,224
Total 4,457 .756 3,371
Total Asian/Predicted Poor 2,632 1.000 2,632
Hispanic 1,855 1.000 1,855
Black 1,498 1.000 1,498
Other 6,467 .5749 3,718
2009/P14 Q1,2,3/P1,4 TOTAL 12,181 .796 9,700
Q1,2 Asian 525 1.000 525
Hispanic 1,571 .900 1,414
Black 1,307 .900 1,176
Other 4.561 .750 3,420
Total 7,964 .821 6,535
Q3 Asian 309 1.000 309
Hispanic 815 .800 652
Black 712 .899 640
Other 2,381 .657 1,564
Total 4,217 .751 3,165
Total Asian 834 1.000 834
Hispanic 2,386 .866 2,066
Black 2,019 .899 1,816
Other 6,942 .718 4,984
2010/P15 Q1,2,3/P1,4 TOTAL 12,390 .707 8,750
Q1,2 Asian 534 1.000 534
Hispanic 1,662 .800 1,330
Black 1,279 .900 1,151
Other 4,574 .638 2,919
Total 8,049 .737 5,934
Q3 Asian 326 1.000 326
Hispanic 917 .688 631
Black 715 .775 554
Other 2,383 .548 1,305
Total 4,341 .649 2,816
Total Asian 860 1.000 860
Hispanic 2,579 .760 1,961
Black 1,994 .855 1,705
Other 6,957 .609 4,224
2011/P16 Q1,2,3/P1,4 TOTAL 12,067 .844 10.180
Q1,2 Cancer 595 1.000 595
Asian 485 1.000 485
Hispanic 1,516 1.000 1,516
Black 1,209 1.000 1,209
Other-complete 2,810 .850 2,389
Other-partial 1,177 .504 593
Total 7,792 .871 6,787
Q3 Cancer 287 1.000 287
Asian 341 1.000 341
Hispanic 870 1.000 870
Black 685 1.000 685
Other-complete 1462 .660 965
Other-partial 630 .389 245
Total 4,275 .794 3,393
Total Cancer 882 1.000 882
Asian 826 1.000 826
Hispanic 2386 1.000 2386
Black 1894 1.000 1894
Other-complete 4272 .785 3354
Other-partial 1807 .464 838
2012/P17 Q1,2,3/P1,4 TOTAL 13,701 0.708 9,700
Q1,2 Asian 695 1.000 695
Hispanic 1,783 1.000 1,783
Black 1,348 1.000 1,348
Other-complete 941 0.400 376
Other-partial 4,317 0.525 2,265
Total 9,084 0.712 6,467
Q3 Asian 380 1.000 380
Hispanic 979 1.000 979
Black 705 1.000 705
Other-complete 444 0.400 178
Other-partial 2,109 0.470 991
Total 4,617 0.700 3,233
Total Asian 1,075 1.000 1,075
Hispanic 2,762 1.000 2,762
Black 2,053 1.000 2,053
Other-complete 1,385 0.400 554
Other-partial 6,426 0.507 3,256
2013/P18 Q1,2,3/P1,4 TOTAL 12,565 0.772 9,700
Q1,2 Asian 656 1.000 656
Hispanic 1,709 1.000 1,709
Black 1,339 1.000 1,339
Other-complete 3,669 0.508 1,864
Other-partial 1,086 0.350 380
Total 8,459 0.703 5,948
Q3 Asian 362 1.000 362
Hispanic 932 1.000 932
Black 651 1.000 651
Other-complete 1,673 0.899 1,504
Other-partial 488 0.620 303
Total 4,106 0.914 3,752
Total Asian 1,018 1.000 1,018
Hispanic 2,641 1.000 2,641
Black 1,990 1.000 1,990
Other-complete 5,342 0.631 3,368
Other-partial 1,574 0.434 683
2014/P19 Q1,2,3/P1,4 TOTAL 12,313 0.788 9,700
Q1,2 Asian 616 1.000 616
Hispanic 1,665 1.000 1,665
Black 1,287 1.000 1,287
Other-complete 3,649 0.690 2,519
Other-partial 864 0.440 380
Total 8,081 1.000 6,467
Q3 Asian 360 1.000 360
Hispanic 963 1.000 963
Black 715 1.000 715
Other-complete 1,711 0.592 1,012
Other-partial 483 0.379 183
Total 4,232 0.764 3,233
Total Asian 976 1.000 976
Hispanic 2,628 1.000 2,628
Black 2,002 1.000 2,002
Other-complete 5,360 0.659 3,531
Other-partial 1,347 0.418 563
2015/P20 Q1,2,3/P1,4 TOTAL 12,109 0.876 10,610
Q1,2 Asian 629 1.000 629
Hispanic 1,726 1.000 1,726
Black 1,215 1.000 1,215
Other-complete 3,419 0.868 2,969
Other-partial 989 0.540 534
Total 7,978 0.887 7,073
Q3 Asian 325 1.000 325
Hispanic 980 1.000 980
Black 666 1.000 666
Other-complete 1,698 0.786 1,335
Other-partial 462 0.500 231
Total 4,131 0.856 3,537
Total Asian 954 1.000 954
Hispanic 2,706 1.000 2,706
Black 1,881 1.000 1,881
Other-complete 5,117 0.841 4,304
Other-partial 1,451 0.527 765
2016/P21 Q1,2,3/P1,4 TOTAL 11,336 0.856 9,700
Q1,2 Asian 604 1.000 604
Hispanic 1,639 1.000 1,639
Black 1,202 1.000 1,202
Other-complete 3,228 0.787 2,540
Other-partial 1,004 0.480 482
Total 7,677 0.842 6,467
Q3 Asian 359 1.000 359
Hispanic 879 1.000 879
Black 599 1.000 599
Other-complete 1,323 0.866 1,145
Other-partial 499 0.503 251
Total 3,659 0.884 3,233
Total Asian 963 1.000 963
Hispanic 2,518 1.000 2,518
Black 1,801 1.000 1,801
Other-complete 4,551 0.810 3,685
Other-partial 1,503 0.488 733

Table A2. Number of completed person-level interviews by age, sex, race/ethnicity, region, and MSA status in MEPS Full-Year Files: MEPS-HC 1996–2006
Year 1996 1997 1998 19991 2000 2001 2002 2003 2004 2005 2006
Total 21,571 32,636 22,953 23,565 23,839 32,122 37,418 32,681 32,737 32,320 32,577
Age
<1 321 406 270 266 291 383 455 431 426 424 417
1-17 5,965 9,330 6,569 6,551 6,595 8,774 10,599 9,512 9,353 9,217 9,161
18-24 1,884 2,919 2,154 2,095 2,119 3,010 3,462 3,080 3,072 3,041 2,994
25-44 6,478 9,332 6,412 6,727 6,680 8,869 10,273 8,877 8,819 8,559 8,317
45-64 4,374 6,661 4,753 5,107 5,244 7,228 8,296 7,053 7,303 7,362 7,749
65+ 2,549 3,988 2,795 2,819 2,910 3,858 4,333 3,728 3,764 3,717 3,939
Sex
Male 10,289 15,443 10,859 11,287 11,445 15,369 17,796 15,413 15,439 15,251 15,370
Female 11,282 17,193 12,094 12,278 12,394 16,753 19,622 17,268 17,298 17,069 17,207
Race/ethnicity
Hispanic 4,638 7,542 5,585 5,852 5,936 7,637 9,427 8,866 9,022 8,990 8,906
Non-Hispanic 16,933 25,094 17,368 17,713 17,903 24,485 27,991 23,815 23,715 23,330 23,671
  Black 2,907 4,815 3,430 3,239 3,471 4,699 5,570 5,094 4,991 5,260 5,608
  Asian 582 829 597 635 602 987 1,304 1,349 1,311 1,227 1,243
  Other 13,444 19,450 13,341 13,839 13,830 18,799 21,117 17,372 17,413 16,843 16,820
Region
Northeast 4,275 6,278 4,159 4,031 3,746 5,063 5,840 4,843 4,912 4,734 4,673
Midwest 4,668 6,834 4,537 4,657 4,951 6,679 7,377 6,365 6,224 6,154 6,370
South 7,494 11,446 8,340 8,764 8,901 12,003 14,212 12,704 13,130 12,656 12,341
West 5,134 8,078 5,917 6,113 6,241 8,377 9,989 8,769 8,471 8,776 9,193

Table A3. Number of completed person-level interviews by age, sex, race/ethnicity, region, and MSA status in MEPS Full-Year Files: MEPS-HC 2007–2016
Year 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
Total 29,370 31,262 34,920 31,228 33,622 37,182 35,068 33,162 33,893 33,259
Age
<1 375 492 490 420 425 485 453 405 371 392
1-17 8,079 8,642 9,560 8,290 8,961 9,935 9,447 8,870 8,831 8,458
18-24 2,692 3,073 3,433 3,010 3,265 3,679 3,460 3,197 3,119 3,005
25-44 7,535 8,278 9,154 8,093 8,835 9,827 9,291 8,791 8,925 8,525
45-64 7,119 7,374 8,399 7,695 8,080 8,937 8,430 7,897 8,275 8,233
65+ 3,570 3,403 3,884 3,720 4,056 4,319 3,987 4,002 4,372 4,646
Sex
Male 14,003 14,946 16,634 14,881 16,068 17,822 16,638 15,790 16,268 15,835
Female 15,367 16,316 18,286 16,347 17,554 19,360 18,430 17,372 17,625 17,424
Race/ethnicity
Hispanic 7,659 8,863 10,112 8,511 9,705 11,862 11,539 10,649 11,068 10,843
Non-Hispanic 21,711 22,399 24,808 22,717 23,917 25,320 23,529 22,513 22,825 22,416
  Black 4,953 6,056 7,004 6,106 6,758 7,589 7,312 6,968 6,551 6,082
  Asian 1,400 1,929 2,295 2,206 2,301 2,653 2,630 2,369 2,464 2,452
  Other 15,358 14,414 15,509 14,405 14,858 15,078 13,587 13,176 13,810 13,882
Region
Northeast 4,384 4,810 5,182 4,757 5,319 6,052 5,838 5,133 5,101 5,262
Midwest 5,956 6,174 6,916 6,402 6,655 6,817 6,206 5,942 6,239 6,290
South 11,127 11,809 13,227 11,688 12,764 14,004 12,955 12,519 12,994 12,664
West 7,903 8,469 9,595 8,381 8,884 10,309 10,069 9,568 9,559 9,043

Return to Table of Contents


Appendix B. Calculation of Response Rates in MEPS

The general approach for the calculation of MEPS response rates, along with an illustrative example, is provided in this section. In particular, response rates for annual 2016 calendar year data are discussed. Because of the linkage of the NHIS and the MEPS, the response rate for MEPS is a combination of the response rate for the NHIS and the MEPS round-specific response rates. Due to the overlapping panel design for the production of annual estimates, the calculation of the annual response rates for MEPS likewise comprises the two overlapping panel-specific response rates. Further, the panel-specific response rates get weighted by their respective sample sizes.

To understand the calculation of MEPS response rates, some key features related to MEPS data collection are first discussed. When an RU is visited for a round of data collection, any changes in RU membership are identified. Such changes include RU members who have moved to another location in the U.S., thus creating a new RU to be interviewed for MEPS, as well as student RUs. Thus, the number of RUs eligible for MEPS interviewing in a given round can only be determined after data collection is fully completed. The ratio of the number of RUs completing the MEPS interview in a given round to the number of RUs characterized as eligible to complete the interview for that round represents the "conditional" round-specific response rate expressed as a proportion. It is "conditional" in that it pertains to the set of RUs characterized as eligible for MEPS for that round, and thus is "conditioned" on prior participation rather than representing the overall response rate through that round. For example, in Table B1, for Panel 20, Round 2, the ratio of 7,991 (Row G) to 8,554 (Row F) multiplied by 100 represents the percentage response rate for Round 2 (93.42 percent when computed), conditioned on the set of RUs characterized as eligible for MEPS for Round 2. Taking the product of the response rate of the NHIS sample designated for use in MEPS (Row A) and the product of the response rates (ratio of the number of completed interviews to the number eligible) for each consecutive set of MEPS rounds, beginning with Round 1, produces the overall response rate through the last MEPS round specified.

An example of a response rate calculation for a full-year MEPS is as follows: To produce annual health care and expenditure estimates for calendar year 2016, data from Panel 20 and Panel 21 are combined. More specifically, data collected covering calendar year 2016 for Rounds 3 through 5 of Panel 20 are combined with corresponding 2016 data from the first three rounds of Panel 21 to produce calendar year 2016 estimates. The overall response rate for the combined sample in Panels 20 and 21 for 2016 is obtained by computing the product of the relative responding sample sizes and the corresponding overall panel response rates and then summing the two products. Details of the calculations as carried out for the 2016 MEPS annual response rate are provided below.

Table B1. Example sample size and final response rates - Full Year (Panel 20, Rounds 1–3 combined with Panel 21, Rounds 3–5): MEPS 2016
Calculation Steps Panel 20 Panel 21
A. Percentage of NHIS households designated for use in MEPS (those initially characterized as responding) 75.1% 71.2%
B. Number of households sampled from the NHIS 10,610 9,700
C. Number of households sampled from the NHIS and eligible and fielded for MEPS 10,571 9,658
D. Round 1 - Number of RUs1 eligible for interviewing 11,283 10,280
E. Round 1 - Number of RUs with completed interviews 8,287 7,643
F. Round 2 - Number of RUs eligible for interviewing 8,554 7,870
G. Round 2 - Number of RUs with completed interviews 7,991 7,319
H. Round 3 - Number of RUs eligible for interviewing 7,743 7,035
I. Round 3 - Number of RUs with completed interviews 7,743 7,035
J. Round 4 - Number of RUs eligible for interviewing 7,877 NA
K. Round 4 - Number of RUs with completed interviews 7,621 NA
L. Round 5 - Number of RUs eligible for interviewing 7,698 NA
M. Round 5 - Number of RUs with completed interviews 7,421 NA
Individual panel response rates:
    P20: A x (E/D) x (G/F) x (I/H) x (K/J) x (M/L)
    P21: A x (E/D) x (G/F) x (I/H)
45.7%
(through Round 5)
46.3%
(through Round 3)
Overall combined response rate:
    0.51 x P20 response rate +
    0.49 x P21 response rate
46.0%

1 RU: reporting unit

Return to Table of Contents