Methodology Report #4:  Sample Design of the 1996 MEPS Nursing Home Component

by James Bethel, Ph.D., and Pamela Broene, M.S., Westat, Inc., and John Paul Sommers, Ph.D.
Agency for Health Care Policy and Research
methods report

 

Contents
	Overview of Sample Design			Sampling of Persons Within Facilities
	Sampling Frame			References
	Facility Selection

 

Tables:
	1. Person-level population estimates and relative standard errors for nursing and personal care homes: 1987 National Medical Expenditure Survey Institutional Population Component
	2. Nursing and personal care facilities-number sampled and expected number responding by round: 1996 MEPS Nursing Home Component
	3. Optimum allocation to cost strata based on the 1996 MEPS Nursing Home Component sample
	4. Minimum acceptable response rates for the 1996 MEPS Nursing Home Component
	5. Rules for facility sampling: 1996 MEPS Nursing Home Component
	6. Response rates at end of Round 1: 1996 MEPS Nursing Home Component

 

Figures showing:
	1. First admissions eligibility determination using a single respondent: 1996 MEPS Nursing Home Component
	2. Final determination of eligibility for sampled nursing home residents: 1996 MEPS Nursing Home Component

Overview of Sample Design

The target population consists of freestanding nursing homes with at least three beds staffed and set up for nursing care, as well as nursing care units consisting of a distinguishable group of three or more nursing home beds within a larger facility. Either type of facility must be:

• Medicare certified as a skilled nursing facility and/or Medicaid certified as a nursing facility, or
• Licensed as a nursing home with a registered nurse or licensed practical nurse onsite 24 hours a day, 7 days a week.

The sample of nursing home residents was stratified by whether residents resided at the nursing home at the beginning of 1996 (current-residents sample) or were admitted during the calendar year (first-admissions sample). The target population of the current-residents sample consisted of persons who resided in nursing homes as of January 1, 1996. The target population of the first-admissions sample consisted of persons who resided in a nursing home during 1996 but were not current residents as defined above.

The sample was designed with the goal of estimating a population proportion of 0.20 with a coefficient of variation (CV) of 9.8 percent or less for facilities, 5.5 percent or less for current residents, and 6.5 percent or less for first admissions. Table 1 shows the relative standard errors for selected characteristics of current residents and first admissions obtained in the 1987 National Medical Expenditure Survey Institutional Population Component, a similar previous survey conducted by the Agency for Health Care Research and Quality (AHRQ). These relative standard errors, or CVs, are based on a sample size of approximately 800 responding nursing homes.

The sampling frame for the selection of facilities in the MEPS NHC was an updated version of the 1991 National Health Provider Inventory (NHPI). The 1991 NHPI is a census of approximately 16,000 nursing homes in the United States, collected by the Bureau of the Census for the National Center for Health Statistics and AHRQ. The 1991 NHPI served as the base, and approximately 2,000 new facilities and 275 hospital-based facilities were added to this original list to create the sampling frame.

Facilities were selected as a double, or two-phase, sample. For the first phase, 1,651 facilities were sampled within strata, with probabilities proportional to size. The measure of size was the number of beds in the facility reserved for nursing home use. The first-phase sample was assigned to four strata based on expected survey travel costs, and a second-phase subsample of 1,430 facilities was selected with equal probabilities within these four cost strata.

The second-phase sample was divided into a main sample of 1,150 facilities and a reserve sample of 280 facilities, the latter being divided into four "release groups" of 70 facilities each. The release groups were intended to be sent to the field to supplement the main sample if response and eligibility rates were lower than expected. On the other hand, the main sample was randomly split into 18 recall groups of approximately 64 facilities each. If response and eligibility rates were higher than expected, sampled facilities could be randomly withdrawn from the field by canceling data collection in selected recall groups. In fact, the MEPS NHC Round 1 response and eligibility rates were higher than anticipated. Therefore, at the conclusion of Round 1, the facilities in two randomly selected recall groups were withdrawn from Rounds 2 and 3 of data collection.

In most facilities, a fixed sample of four current residents and four new admissions was selected using simple random sampling within the facility. In facilities with measures of size that were poorly correlated with the number of admissions, the first admissions sample size could be increased from two to three per round (to a total of six). The within-facility sample sizes are intended to yield approximately 3,043 eligible current-resident respondents and 2,218 eligible first-admission respondents, all with complete use and expenditure data. The target sample sizes are summarized in Table 2.

Sampling Frame

Description of the National Health Provider Inventory

A nursing home, according to the NCHS definition, is a facility having at least three beds and identifying itself on the NHPI questionnaire as one of the following:

• A licensed nursing home;
• A skilled nursing long-term care unit of a hospital;
• A nursing care unit of a retirement center;
• A nursing facility certified under Medicare or Medicaid; or
• Some other type of nursing home.

Among the 17,502 facilities on the basic frame, 205 appeared to be board-and-care homes, which do not meet this definition and were excluded. Another 275 facilities that had been excluded by NCHS-mostly Department of Veterans' Affairs (VA) facilities-were added to the frame. The final updated NHPI contained 17,572 (17,502-205 + 275) facilities (Pancholi, 1995).

To be eligible for the MEPS NHC, facilities must have at least three beds and be either Medicare- and/or Medicaid-certified or licensed as nursing homes. Final eligibility for the MEPS NHC was determined during Round 1 of facility data collection; however, the initial sampling frame included all facilities on the updated NHPI that were likely to meet these criteria.

Editing the MEPS NHC Frame

Missing values for variables on the NHPI that were needed for sampling were obtained from external sources when possible. Information on license/certification status, type of ownership, and the number of beds was obtained for all but a small percentage of the new facilities. Certification status was not available for the 275 VA facilities and was imputed for an additional 69 facilities. The facility ownership type (profit, nonprofit, government) was unknown for 216 facilities but was not imputed. Missing telephone numbers were supplied for over 1,700 facilities, but 132 still remained missing on the frame at the time of sampling. A Beale code (also known as the Human Resource Profile Code) was placed on the file to indicate the metropolitan statistical area (MSA) status of each facility. These codes were collapsed for use in sampling. The collapsed values were 0 = large metropolitan core, 1 = large metropolitan fringe, 2 = medium metropolitan area, 3 = lesser metropolitan area, 4 = adjacent to an MSA, and 5 = not adjacent to an MSA.

Measure of Size

Facility Selection

Summary

Cost stratification was then performed on the 1,651 facilities in the first-phase sample, with the actual strata being defined in terms of distance from the nearest of the 50 largest U.S. cities and the expected effect on survey travel costs. Next, the optimal sampling rates were determined for these four cost strata. Using the sampling rates, a cost-stratified subsample of 1,430 facilities was selected from the 1,651 facilities in the main sample. Within each cost stratum, the second-phase sample of facilities not drawn with certainty (noncertainty facilities) was subsampled at a rate of .803, yielding a reserve sample of 280 facilities and a remaining main sample of 1,150 facilities. The four release groups were assigned by sorting the reserve sample by order of selection and consecutively numbering from 1 to 4, repeating until all 280 facilities were assigned. This resulted in four release groups of 70 facilities each. The main sample was randomly divided into 18 subsamples of approximately 64 facilities each by sorting the noncertainty sample facilities in the order of selection and consecutively numbering from 1 to 18, repeating until all 1,139 noncertainty facilities were assigned.

Initial Stratification

The first-phase strata were formed by grouping facilities according to three types of Medicaid/Medicare reimbursement and whether the facility was hospital based or not. The 20 largest facilities were placed in a separate stratum. Eleven of these had been selected previously for the NNHS, conducted by the NCHS. The remaining nine were designated for the MEPS NHC.

Determining Selection Possibilities

However, before the sample was selected, these selection probabilities were modified to minimize overlap with the 1995 NNHS, which was conducted by NCHS and fielded in late 1995. Because the NNHS and MEPS NHC used similar sampling frames, it was important to prevent (if possible) any nursing homes from being included in both surveys.

A Keyfitz procedure was used to adjust the probabilities of selection to minimize this overlap. This procedure provides the desired unconditional probabilities of selection for the MEPS NHC sample while at the same time minimizing the overlap. To compute conditional probabilities of selection for the MEPS NHC, the probabilities of selection for a facility in both the MEPS NHC and the 1995 NNHS frame, as well as which nursing homes were selected in the 1995 NNHS, must be known. The following notation describes the procedure.

The unconditional probability of selection for a facility in the MEPS NHC can be written as:

Given the outcome of the 1995 NNHS sampling, it is shown below that selecting the MEPS NHC sample with these redefined probabilities preserves the original MEPS NHC probabilities of selection.

After these rules were applied in the six noncertainty strata, the facilities were selected for the MEPS NHC using the redefined selection probabilities. Selection was done using a systematic selection process. For this selection the file was sorted by Beale code and ownership to create implicit sampling strata within each of seven explicit strata. In the certainty stratum, the nine facilities not selected for the NNHS sample were sampled with certainty. The remaining 11 facilities were assigned a zero probability of selection. The outcome of using these probabilities of selection was that none of the MEPS NHC sample facilities overlapped with the NNHS sample.

Two additional certainty facilities were selected in two of the noncertainty strata. These two facilities were Case 1 situations and were not selected in the NNHS sample, so their Keyfitz probabilities were set equal to one. The remainder of the sample in the six noncertainty strata were Case 2 facilities.

Cost Stratification

• Stratum 1: Full workload in a single geographic area, such as a city.
• Stratum 2: Partial workload only in a single area, requiring considerable travel.
• Stratum 3: Single facility requiring considerable travel but within the range of other facilities.
• Stratum 4: Single facility at a distance requiring air travel.

The cost stratification process consisted of several steps. The first-phase sample of 1,651 facilities was mapped using computer mapping software. Next, each facility was mapped into the appropriate ZIP Code center point. Then, to approximate the cost strata, a map of the 50 largest U.S. cities and concentric zones around them was overlaid on the facility map. Facilities located within 100 kilometers of a city were assigned to Stratum 1; facilities 100-200 kilometers, to Stratum 2; facilities 200-300 kilometers, to Stratum 3; and facilities beyond 300 kilometers of a city, to Stratum 4.

Minimizing Overlap With the Medicare Current Beneficiary Survey

The Medicare Current Beneficiary Survey (MCBS) is an ongoing survey of Medicare beneficiaries conducted by Westat for the Health Care Financing Administration. As part of this survey, Westat field interviewers visit many nursing homes throughout the United States. As with the NNHS, it was important to minimize the number of nursing homes involved in both surveys. However, an alternative to the Keyfitz procedure was necessary because of the virtual impossibility of calculating the probabilities of selection for the MCBS facilities.

The procedure used was to flag any nursing home that MCBS respondents reported as their current residence as of September 28, 1995, and that was also in the first-phase MEPS NHC sample. There were 71 such facilities. Of these overlapping facilities, one MEPS NHC noncertainty facility was removed from the first-phase sample prior to sampling in the second phase, thus giving a zero chance of selection. An adjustment factor was applied to the weights within each cost stratum to prevent an undercoverage bias. The nine facilities that were included with certainty in the first phase of MEPS NHC sampling were designated to be selected with certainty in the second phase, regardless of whether they overlapped with the MCBS. Based on opinions of health care analysts at AHRQ, facilities excluded in this way were unlikely to differ in any systematic way from other facilities in the first-phase sample. Thus, this procedure was not expected to cause any sampling bias.

Selecting the Second-Phase Sample

where W_h and r_h represent the population proportion and sampling rate for the hth stratum. This formula neglects the variance for the analysis variables, since it is expected that they would vary little between cost strata. The MEPS NHC facilities overlapping with MCBS were not removed prior to determining the optimal allocation, since these facilities will be treated in sample weighting as nonrespondents.

The optimum allocation based on the MEPS NHC first-phase sample is shown in Table 3. The optimal subsampling rates ranged from .78 to .89. The proportions (W_h) shown in Table 3 are those obtained in the MEPS NHC sample of 1,651 facilities. The data collection cost estimates include travel costs, interviewer per diem and salary, and data processing costs.

Sampling Algorithms

First-Phase Sample

Thus the sampling algorithm for the first phase consisted of this step:

Step 1. Within each stratum, sort the facilities by Beale code, type of ownership, and ZIP Code. Calculate the conditional (Keyfitz) probabilities of selection. Select n_h0 facilities with PPS, with the Keyfitz probability of selection as the measure of size. There will be c_h0 certainties, i.e., facilities that will have . For the other facilities, the original unconditional selection probabilities will be

where M_hi is the measure of size for the ith facility in the hth stratum.

Second-Phase Sample

Thus the sampling algorithm for the second phase consisted of these steps:

Step 2. Map the sample of n₀ facilities into the four cost strata using facility ZIP Code and mapping software.

Step 3. Remove noncertainty facilities identified as overlapping with MCBS and certainty facilities from the first-phase sample.

Step 4. To select the second-phase sample of n₁ facilities from the first-phase sample of n₀, sort the facilities in each cost stratum in the original order of selection. Within each cost stratum, draw an equal probability systematic sample of facilities, where the sample size is determined by optimum allocation. (See Table 3.) Subtract the number of first-phase certainty facilities in each cost stratum from the designated sample size in Table 3 prior to sampling.

Step 5. To select the reserve sample of r facilities from the n₁ second-phase facilities, first sort the noncertainty facilities in the second-phase sample by order of selection in each cost stratum. Within each cost stratum, select an equal probability systematic sample of facilities using the sample sizes in Table 3. Create four release groups by sorting the reserve sample in the order of selection, then consecutively numbering the reserve sample from 1 to 4, repeating until the entire reserve sample has been assigned. There will be m = n₁r facilities in the main sample and r facilities in the reserve sample. The reserve sample will consist of four release groups of r/4 facilities each.

Step 6. To create the 18 recall groups from the main sample, sort the noncertainty facilities in the main sample in the order of selection, then consecutively number facilities from 1 to 18, repeating until all noncertainty main sample facilities have been assigned. Each recall group thus will represent a random subsample of the main sample.

For a two-phase sampling process like this, the sampling probabilities for the ith facility in the hth stratum can be written as:

For the "initial certainty/final certainty" facilities--facilities that were selected with certainty in both the first and second phases of sampling--the overall selection probability is 1.00. For the "initial noncertainty/final noncertainty" facilities, the final selection probability would be:

where m_h' is the main sample size in cost stratum h' and n_h'0₀ is the number of first-phase sample facilities in cost stratum h'. If release groups are used, the numerator in the first factor is increased by the extra number of facilities released. If no release groups are used but some recall groups are withdrawn, the numerator is decreased by the number of facilities being withdrawn in cost stratum h'.

Initial Screening of Facilities

• Facilities must have three or more beds that are staffed and set up for residents (or a distinguishable group of three or more beds within a facility).
• Facilities must either be:
• Medicare certified as a skilled nursing facility and/or Medicaid certified as a nursing facility, or
• Licensed as a nursing home with a registered nurse or licensed practical nurse onsite 24 hours a day, 7 days a week.

As a result of the screening, 14 facilities were identified as being out of business and 1 facility was determined to be ineligible.

Round 1 Facility Response Rates

Field Problems

Nursing Homes Associated With Other Facilities

Facilities That Moved or Combined With Other Facilities

A sampled facility that combined with another facility was retained in the sample as long as the other facility was not listed in the NHPI. If both of the original facilities were listed separately in the NHPI, the combined facility had an increased chance of selection because it could have been selected through either one of the original facilities. Either this increased chance of selection had to be accounted for in the facility weight or, alternatively, one of the listings had to be considered out of scope. When the combined facilities could be treated as multiple units of one nursing home, the latter approach was used. Otherwise, weighting adjustments were made.

Facilities With Multiple Units

Survey Database

• Facility name, address, and telephone number;
• Numbers of residents and eligible beds from the NHPI;
• Final measure of size;
• The random numbers used for sampling current residents and first admissions.

Sampling of Persons Within Facilities

Checking Facility Data Against Frame Data

If the nursing home facility existed within a long-term care facility, the interviewer verified that the number of residents listed corresponded to the eligible portion of the facility. The interviewer also verified that no eligible portions of the facility were overlooked.

Current-Residents Sample

The interviewer entered the size of the list of current residents in the CAPI system, which then determined the random start, the skip interval, and the sample of line numbers. The selected line numbers were displayed on the computer screen and stored in memory for later validation. The order of selection for the sampled current residents was stored for inclusion in the final database. At the end of Round 1, the response rates shown in Table 6 were obtained for current residents.

The overall response rate for the current-residents sample is 98.8 percent. Sampled residents needed to have 75 percent of their baseline health status items complete and age, sex, and race reported in order to be considered respondents. Forty-four eligible current residents did not meet this requirement. Of these, four met the baseline health status criteria but were missing at least one of the demographic variables. In addition, 17 sampled persons were ineligible.

First-Admissions Sample

First-Admissions Sample Size

Thus, the sample size for the first-admissions sample at a given facility may be revised based on the relationship between the current residents and the number of first admissions listed. The revised sample size was based on the selection probability:

Thus, the first-admissions sample sizes were adjusted upward or downward according to whether more or fewer were listed, based on the measure of size adjusted by the factor (rho) to reflect the average number of first admissions to residents. However, the within-facility first-admissions sample size was not permitted to exceed three per round, and was less than two only when there were fewer than two first admissions in the facility for the round. Although (rho) is unknown, it can be approximated using 1987 NMES data on the ratio of nursing home admissions to residents. The value of (rho) using 1987 NMES data turns out to be 718,670/1,523,540 = .472 (Agency for Health Care Policy and Research, 1990).

Eligibility Determination

An eligible first admission is defined as a person with no admissions or stays on or after January 1, 1996, in MEPS NHC-eligible facilities prior to the admission for which the person was sampled at the primary sampled facility. Information about where the person lived from January 1, 1996, until the date of admission to the sampled facility, referred to as the pre-stay period, was collected from facility respondents. Using CAPI, data were collected on the beginning and ending dates for each separate period of residence during the pre-stay period, the name and type of each place where the sampled person stayed, and whether the person stayed at that place the whole time between the beginning and ending dates. Place types include the sampled facility, community residence, acute care or long-term care hospitals, and other long-term facilities. All places of residence provisionally identified as long-term care facilities were searched on the NHPI file for a determination of nursing home eligibility status. Since this would include hospitals with long-term care skilled nursing units, the AHA file also was searched during residence history data collection to determine first-admission eligibility.

As an aid in determining eligibility, the NHPI and AHA list (American Hospital Assocation, 1993) were loaded into the interviewers' laptop computers and incorporated into their CAPI software. A search software program allowed the field interviewers to search for an identified long-term care facility on the NHPI or AHA files in different ways, including name, address, State, and telephone number. Interviewers were able to conduct searches based on portions of the information to maximize the likelihood of finding matches. At the conclusion of the pre-stay residence history data collection, the CAPI system automatically brought the interviewer to the NHPI and AHA search functions to search for matches to reported long-term care facilities. Interviewers were trained to search for the facility name and, if that failed, to use the facility address and telephone number. Statisticians could verify NHPI and AHA searches at any time in the home office.

Based on information collected from the facility about prior admissions to other facilities, the sampled admissions were classified as eligible (with no prior stay in an eligible facility during the reference period), ineligible (with one or more prior stays identified), or indeterminate (with some period of time within the reference period for which the facility could not report whether the resident was in an eligible facility). Figure 1 shows the data collection process and the flowchart for determining eligibility. There were four possible outcomes, each having a different protocol for data collection:

• Eligible first admission: No admissions prior to sampling.
-All data collection continues.
• Ineligible first admission: One or more admissions prior to sampling.
-All data collection stops.
• Provisionally eligible first admission: Eligibility cannot be determined because either the facility has a gap in the pre-stay data or there was an admission to a facility but the name of the facility is unknown or it did not match the NHPI listing.
-All data collection continues.
• Sampling error: The sampled admission is listed twice and the entry sampled is the second one.
-All data collection stops.

Note: EF=eligible first admission. IF=ineligible first admission. NHPI=National Health Provider Inventory. PF=provisionally eligible first admission.

Source: Agency for Health Care Policy and Research: 1996 Medical Expenditure Panel Survey Nursing Home Component.

For persons who were eligible or indeterminate, interviewers attempted to complete a community residence history for the pre-stay period by telephoning a knowledgeable community respondent (usually a relative). Information from the community residence history questionnaires was consulted to make an eligibility determination for persons in the indeterminate group. Persons found to be ineligible on the basis of either the facility data or the community data were dropped from further data collection.

The process to determine eligibility was the same for both the facility and the community pre-stay data. Figure 2 shows the rules used to make a final eligibility determination. This process was implemented if determinations of eligibility were made using data from both the facility and community respondent. Table 2 shows the number of sampled first admissions expected to be eligible and the expected final first-admissions sample sizes.

Source: Agency for Health Care Policy and Research: 1996 Medical Expenditure Panel Survey Nursing Home Component.

Resolution of Sampling Errors

• Person sampled as a first admission was a resident on January 1, 1996.
  Resolution:

-	First admission was not listed for current-residents sample: Drop from the first-admissions sample and code as a sampling error; adjust current-residents sample weights.
-	First admission was listed for both current-residents and first-admissions samples: Drop from the first-admissions sample and code as a sampling error.

• Person sampled as a current resident was not a resident on January 1, 1996, but was admitted later.
  Resolution:

-	Current resident was not listed for first-admissions sample: Drop from the current-residents sample and code as a sampling error; add to first-admissions list before first-admissions sampling to ensure a chance of selection.
-	Current resident was listed for first-admissions sample: Drop from the current-residents sample.

• Person sampled as a first admission was admitted and listed twice or more.
  Resolution:

-	First admission was sampled on first admission: Retain first admission in the sample.
-	First admission was sampled on later admission: Drop first admission from the sample.

• Eligible persons were omitted from listing.
  Resolution: Call home office; adjust sampling weights.
• Ineligible persons (e.g., residents of an assisted living wing) were listed or sampled.
  Resolution: Call home office; clean list and resample, if possible. May require CAPI intervention from home office to allow resampling. If resampling is not possible, CAPI software detects ineligible sampled persons in the residence history questionnaire and they are dropped from the sample and coded as out of scope.

These resolutions are not perfect. While they were intended to preserve the rule of a single chance of selection, they do not preserve the clear stratification of the current-residents versus first-admissions samples. In each case, the sampled person being dropped could instead be retained if proper adjustments were made to the sampling weights. It should be noted, however, that this latter resolution would not preserve the stratification of the two samples either.

References

American Hospital Association. American Hospital Association guide to the health field. Chicago; 1993.

National Center for Health Statistics. Sampling frame for the 1995 National Nursing Home Survey (facility file) [microdata tape documentation]. Hyattsville, MD; 1995.

Pancholi, M. Message to P. Broene in electronic mail system, 1995 Oct 18; Pancholi, M. Unpublished memo to P. Broene, "The Updated Health Provider Inventory," 1995 July 17.

Source: Agency for Health Care Policy and Research: 1996 Medical Expenditure Panel Survey Nursing Home Component.

1 At least 1/3 of data completed.

2 All data provided.

Note: FQ =Facility Questionnaire. IUED = institutional use and expenditure data. RH =  residence history.

Source: Agency for Health Care Policy and Research: 1996 Medical Expenditure Panel Survey Nursing Home Component.

Note: The stratum proportions are based on the MEPS Nursing Home Component sample. Costs are based on the data collection budget for the MEPS Nursing Home Component and include travel costs, interviewer per diem and salary, and data processing costs. Based on proximity to large U.S. cities, the strata were defined as follows: Stratum 1—Full workload in a single geographic area, such as a city. Stratum 2–Partial workload only in a single area, requiring considerable travel. Stratum 3–Single facility requiring considerable travel but within the range of other facilities. Stratum 4–Single facility at a distance requiring air travel.

Source: Agency for Health Care Policy and Research: 1996 Medical Expenditure Panel Survey Nursing Home Component.

Source: Agency for Health Care Policy and Research: 1996 Medical Expenditure Panel Survey Nursing Home Component.

1 Either revise computer-assisted personal interviewing (CAPI) to subsample or review in home office for weighting corrections.

Source: Agency for Health Care Policy and Research: 1996 Medical Expenditure Panel Survey Nursing Home Component.

Source: Agency for Health Care Policy and Research: 1996 Medical Expenditure Panel Survey Nursing Home Component.

email MEPS at mepspd@ahrq.gov