Because Kaiser/HRET selects firms randomly, it is possible through the use of statistical weights to extrapolate the results to national (as well as regional, industry, and firm size) averages. These weights allow Kaiser/HRET to present findings based on the number of workers covered by health plans, the number of total workers, and the number of firms.
The calculation of the weights followed a common approach as in previous years, but with several notable changes in 2003. First, the basic weight is determined, followed by a non-response adjustment added this year to reflect the fact that small firms that do not participate in the full survey are also less likely to answer the single offer rate question. To make this adjustment, Kaiser/HRET conducted a follow-up survey of all firms with 3-49 workers that refused to participate in the full survey. Each of these 1,744 firms was asked the single question, "Does your company offer or contribute to a health insurance program as a benefit to its employees?" The main difference between this follow-up survey and the original survey is that in the follow-up survey we asked the question of the first person who answered the telephone, whereas in the original survey we asked the question of the person who is most knowledgeable about the firm's health benefits.
Conducting the follow-up survey accomplished two objectives. First, the chi-squared analysis proved that the change in method--speaking with the person answering the phone rather than a benefits manager-- did not bias the results of the follow-up survey. To show this we analyzed firms who responded to the offer question twice (in both the original and follow-up survey) and proved that there was no difference in the likelihood that a firm offers coverage based on which employee answered the question about whether a firm offers health benefits.
Second, we proved that very small firms not offering health benefits to their workers are less likely to answer the one survey question about coverage. We analyzed the group of firms that only responded to the follow up survey and performed a chi-squared analysis between the firms who had responded to the initial survey and those who only responded to the follow up. Tests confirmed our hypothesis that the firms that did not answer the single offer rate question in the original survey were less likely to offer health benefits. To adjust the offer rate data for this finding we applied an additional non-response adjustment to increase the weight of firms in our sample that do not offer coverage.
The second change to the weighting method in 2003 was to trim the weights in order to reduce the influence of weight outliers. On occasion one or two firms will, through the weighting process, represent a highly disproportionate number of firms or covered workers. Rather than exclude observations by trimming them from the sample, our goal was to determine a set cut point that would minimize the variances of several key variables (such as premium change and offer rate) and assure that the weights apportioned to each firm in the sample do not represent any more than X% of the entire weighted sample.3 The additional weight represented by outliers is then spread among the other firms in the same sampling cell to assure broader representation.
Finally, we applied a post-stratification adjustment. In the past, Kaiser/HRET have always post-stratified back to the Dun & Bradstreet frequency counts. However, concerns were raised this year over the accuracy and volatility of these counts. As a result, we used the recently released Statistics of U.S. Businesses conducted by the U.S. Census as the basis for the post-stratification adjustment. Census data indicate the percentage of the nation's firms with 3-9 workers is 60% rather than the higher percentages (e.g. 74% in 2002) derived from Dun and Bradstreet's national database.4 This change has little impact on worker-based estimates, since firms with 3-9 workers accounted for less than 10% of the nation's workforce. The impact on estimates expressed as a percentage of employers (e.g. the percent of firms offering coverage), however, may be sufficient.
Due to these changes, Kaiser/HRET has recalculated the weights for survey years 1999-2002 and modified estimates published in the survey where appropriate. The vast majority of these estimates are not statistically different. However, please note that the survey data published in this book may vary slightly from previously published reports.
The data are analyzed with SUDAAN, which computes appropriate standard error estimates by controlling for the complex design of the survey. Most statistical tests are performed at the 0.05 level. Two types of significance tests performed are the t-Test and the Chi-square test.
Historical Data
Data in this report focus primarily on findings from surveys jointly authored by the Kaiser Family Foundation and the Health Research and Educational Trust, which were conducted after 1999. Prior to 1999, the survey was conducted by HIAA and KPMG using the same survey instrument, but data is not available for all intervening years. Following the survey’s introduction in 1988, HIAA conducted the survey in 1990 and 1991, but most of these data are not available to us with the exception of a few key indicators in 1988 and 1990. KPMG also conducted the survey in 1992, 1994, and 1997; however, only larger firms were sampled in these years and are not comparable to recent estimates. In 1993, 1995, 1996, and 1998, KPMG interviewed both large and small firms.
To further analyze changes in employer-sponsored health plans during the past few years, this report uses data from the 1993, 1996, and 1998 KPMG Surveys of Employer-Sponsored Health Benefits and the 1999- 2002 Kaiser/HRET Survey of Employer-Sponsored Health Benefits. For a longer term perspective, the 1988 survey of the nation’s employers conducted by the Health Insurance Association of America (HIAA), on which the KPMG and Kaiser/HRET surveys are based was used. Many of the questions in the HIAA, the KPMG, and Kaiser/HRET surveys are identical. The survey designs among the three surveys are similar.
[3] Firms sometimes have disproportionate weights when a firm grows rapidly, or when a firm in the sample acquires another firm.
[4] The initial approach designed to minimize variances among key variables required trimming more than 5% of the observations. As an alternative, a rule was created that the trimming point should only trim the largest 5% of observations.
[5] One possible explanation for this discrepancy is that Dun & Bradstreet is slow in purging firms from their database that have gone out of business, or have been acquired by other firms.
