Sample Design

NEHIS was a sample survey of business establishments (specific locations), governments, and self-employed individuals with no employees and no other locations. The sampling unit was the establishment, defined as "an economic unit, generally at a single physical location, where business is conducted or where services or industrial operations are performed." A major reason that establishments rather than firms (that is, a business organization or entity consisting of one domestic establishment or more under common ownership or control) were sampled in NEHIS is that establishments are confined within State borders, enabling State estimates.

Three distinct sampling frames were used for this survey. Business establishments were sampled from the Dun’s Market Identifiers file, a commercially available file maintained by the Dun and Bradstreet Corporation. This file is updated frequently. The version used to select this sample was the one available as of late November 1993. Governments were sampled from the 1992 Census of Governments. The Census of Governments is conducted every 5 years by the Bureau of the Census. The sample of self-employed individuals with no other employees or locations was identified through a screening of respondents in the second half of the 1993 National Health Interview Survey, who indicated that they were primarily self-employed. The National Health Interview Survey (NHIS) is conducted by the Bureau of the Census for the National Center for Health Statistics.

The sample stratification and allocation were designed to support reliable State-level estimates of characteristics related to employers and employees. The stratification in the private and public sectors was similar. Both sectors used State (the District of Columbia was treated as a "State equivalent") as a major stratifier. Within State, sampling units were classified into strata defined by a two-way cross-classification.

In the private sector, the two-way cross-classification of establishments was by firm size and establishment size, where "size" was the number of employees reported on the Dun and Bradstreet Dun’s Market Identifiers file. "Firm" was defined as the entire parent "company" or "enterprise" for most companies. For large companies with subsidiaries, first- and second-level subsidiaries of large companies were split off from the parent company and treated as separate firms for assignment of firm size. The construction of firm size used number of employees and corporate linkage information off the Dun and Bradstreet Dun’s Market Identifiers file.

Health Insurance Plan Subsampling
In addition to a sample designed to provide State estimates of establishments and governments that offer health insurance, another unique feature of the NEHIS is a sample of plans representative of all health insurance plans offered to employees. The NEHIS design subsampled health insurance plans to reduce response burden. When five plans or fewer were offered, data were collected on all plans. However, for businesses offering six or more plans, five plans were subsampled using a methodology that obtained a representative mix of plans in the sample. A separate plan subsampling methodology was developed for firms with large numbers of establishments in the NEHIS sample, where no more than 13 plans were subsampled within the firm. No plan subsampling was performed for the Federal Government and State governments; these government units (with the exception of four States) reported on all of their plans. Subsampling of plans was implemented for about 11 percent of private sector establishments with insurance and for about 9 percent of local governments (and purchasing units) with insurance.

Further details about the NEHIS sample design are found in:
Plan and Operation of the National Employer Health Insurance Survey (PHS) 99-1021.
View/download PDF 3 MB

Data Source and Collection
NEHIS surveyed a probability sample of all U.S. employers in each State in the private and public sectors. Data were collected by Westat, Inc., from April to December 1994. Interviews were conducted with respondents identified during prescreening as the most knowledgeable about health benefits for the establishment. Approximately 39,000 employers completed interviews, producing a final unit response rate of about 72 percent. Response rate, however, varied substantially by type of sample case, State, and firm size. Among private sector establishments, 34,604 interviews were completed for a response rate of 71 percent.

The primary mode of data collection was computer-assisted telephone interviews (CATI).The CATI methodology was implemented for several reasons: complexity of the question sequence, expected use of multiple respondents, the large number of sampled cases, limited subject-matter expertise of interviewers, and rapid data turnaround. However, SENE data were collected by telephone using a paper questionnaire that was modeled after those items collected in the CATI questionnaire. SENE did not use a CATI version because some questions were not relevant for self-employed persons and because there was only one respondent per completed interview.

The NEHIS questionnaire obtained information about the availability and characteristics of employer-sponsored health insurance coverage, plan benefits, and costs. Information on plan characteristics was collected for both plans with enrollees and plans with zero enrollees (offered to employees but not used).

Further details about the NEHIS survey methodology are found in:
Plan and Operation of the National Employer Health Insurance Survey (PHS) 99-1021.
View/download PDF 3 MB

Estimation Procedures
The probability design of the NEHIS establishment sample allows the data to be weighted to produce unbiased State and national estimates. The multistage estimation procedure used for the sample includes four components: 1) Inflation by reciprocals of the probabilities of selection, 2) adjustment for nonresponse, 3)trimming of excessive weights, and 4) post-stratification to independent universe counts.

1. Every establishment and government on the NEHIS sampling frames had a known, nonzero probability of selection. The "base weights" were equal to the reciprocal of the stratum sampling rate, except for non-located establishments (about 9,700 establishments). Because of their uncertainty of eligibility, those cases were adjusted by 0.03. This adjustment factor was derived from results of investigating the eligibility status for a small sample of non-located cases in Maryland. The largest base weights were 714.8 in the private sector and 177.2 in the public sector.

2. A nonresponse adjustment was carried out on the adjusted base weight in several stages. In the first stage, eligible cases were adjusted to account for establishments whose eligibility status was unknown. The adjustment factor was the ratio of the sum of the weights of all sample cases to the sum of the weights of all sample cases except nonrespondents with unknown eligibility. In the second stage, adjustments were made to account for nonresponding establishments whose insurance status was unknown. In the third stage, the adjusted weights were further adjusted to account for the nonresponding establishments whose insurance status was known. For both the second and third stages, the adjustment factor was the ratio of the sum of the weights of eligible cases to the sum of the weights of the respondents. In the fourth stage, used primarily from the public sector, weights were adjusted to account for nonresponse among public sector entities selected with certainty. The adjustment factor was calculated according to the number of employees at the establishment. The largest overall nonresponse adjustment factors were 2.02 in the private sector and 4.7 in the public sector.

3. Weight trimming was used sparingly in the NEHIS. Private sector establishment weights were trimmed in NEHIS when the weighted difference in the establishment frame size accounted for at least 8 percent of the estimated number of employees in the same State. For the public sector weights were trimmed if the weighted difference in the establishment size accounted for at least 5 percent of the estimated number of employees in the State. In addition, the reported establishment size had to be at least 10 times larger than the frame size. Adjustment factors for trimmed establishments ranged from 0.03 to 0.53 in the private sector and from 0.08 to 0.15 in the public sector.

4. A "post-stratification adjustment" was performed on private sector establishment weights so that NEHIS employee counts agreed with independent employment estimates provided by the U.S. Bureau of Labor Statistics. As no exact control totals were available because of differences in reference periods, frame coverage, and definitions, universe counts obtained from the Employment and Earnings Survey were increased to account for almost 8 million additional people included as employees in the NEHIS but not in the U.S. Bureau of Labor Statistics figures. The cell post-stratification factors ranged from 0.4 to 2.1; and the national post-stratification factor was 0.99.

Reliability of Survey Estimates and Standard Errors
As in other surveys, the NEHIS results are subject to sampling and nonsampling errors. Sampling error consists of the error in a survey estimate that is attributed to the fact that a sample, rather than a complete census, was used to compute the survey estimates. The probability design of NEHIS enables calculation of sampling errors.

The standard error of an estimate is primarily a measure of the sampling variability that occurs by chance because only a sample is surveyed. NCHS considers an estimate to be reliable if it has a relative standard error of 30 percent or less (that is, the standard error is no more than 30 percent of the estimate). It should be noted also that estimates based on

fewer than 30 records are also considered unreliable regardless of the magnitude of the relative standard error.

The standard errors generated for most NEHIS published estimates are computed using SUDAAN software. SUDAAN computes standard errors by using a first-order Taylor series approximation of the derivation of estimates from their expected values.

Because computing a direct estimate of sampling error for every statistic is not always feasible, several generalized variance functions were also derived for approximating standard errors of totals and percents for NEHIS establishments and employee estimates for the private sector. Details about computing these standard error approximations can be found in the following NEHIS publication:
Plan and Operation of the National Employer Health Insurance. (PHS) 99-1021.
View/download PDF 3 MB

This page last reviewed October 15, 2008