HEALTH INSURANCE SUPPLEMENT NOTES Identification of "dummy" records on the 1989 NHIS Health Insurance Public Use Data File The purpose of this memo is to provide instructions for the identification of 3,300 "dummy records" on the 1989 NHIS Health Insurance Public Use Data File. These records were classified as having "no" coverage instead of "unknown" coverage on several public assistance and military coverage variables. Analysts who are interested in tallying these specific variables may wish to reclassify the "dummy" records. Some analysts may wish to exclude the "dummies" from their analysis and reweight the sample. Introduction As you know, the National Health Interview Survey includes both the Basic Health and Demographic Questionnaire (the "core") and Current Health Topic Questionnaires (also called "supplements"). Not everyone who agrees to answer the core questions also agrees to provide answers for the supplements. In those cases, we create "dummy" records for the supplement files. These "dummy" records are usually created by coding variables as "unknown" or whatever would be the appropriate response when the expected data are missing. Prior to 1989, when health insurance questions were included in the NHIS, they were part of the core questionnaire. As a result, there were only a few respondents with missing data (less than one percent). In 1989, health insurance was moved to a separate Current Health Topic Questionnaire and, as a consequence of this and possibly other reasons, the nonresponse rate increased to about three percent and more "dummy" records had to be created (3,300). We do not identify which records are "dummy" records on the 1989 public use data file. Because of some historical idiosyncracies in the edit specifications (which are noted below), these "dummy" records were sometimes classified as "no" when "unknown" would have been more appropriate. As a result, the total number of persons reporting no health care coverage from certain types of programs, e.g., Medicaid, will be inflated by the number of misclassified "dummy" records. In addition to a short history of the problem, this memo will provide you with the information needed to identify all of the "dummy" records on the 1989 NHIS Health Insurance Supplement data file and to correct those fields that were misclassifed as "no" instead of "unknown". - 1 - Identification of "dummy" records on the 1989 NHIS Health Insurance Public Use Data File (Cont'd) Background Since 1978, the same edits, with modifications only for added questions, have been used on the NHIS Health Insurance file. For items not related to private health insurance or Medicare, the edits did not distinguish between a response of "no" and "unknown" because most of these were originally check-items (e.g., the box checked by the interviewer meant "yes" and an unchecked box meant "no" or "unknown"). Therefore, blanks were coded as "no". In 1984 these questions were reformatted to three response codes: "yes", "no", and "unknown". The edits were not changed for two reasons: (1) to retain comparability with earlier surveys; and (2) there were an insignificant number of "unknown"s (i.e., a few hundred at most) and the expected response for these items was "no" for about 95 percent of the sample. Thus, about 95 percent of a relatively insignificant number of cases were being properly classified without changing the edits. As noted above, in 1989 the Health Insurance Supplement was removed from the main questionnaire and placed in a separate questionnaire for supplemental topics, and the nonresponse rate increased, but edits for "dummy" records were not changed. Therefore, you may want to reclassify responses for "dummy" records from "no" to "unknown" for items related to public assistance and military- Veterans Administration health care coverage. The actual effect of this modification on the estimates is relatively small and analysts may choose not to make it. The following section offers an indication of the magnitude of the changes to be expected. Effect of reclassifying "dummy" records The need to change "no" to "unknown" for "dummy" records applies mainly to items that are included in the Health Insurance Supplement to serve as a basis for creating recodes to classify persons with regard to public assistance and military-VA coverage. For instance, the item in data location 453 on Aid To Families with Dependent Children (AFDC) was included not primarily to estimate the number of these beneficiaries but to serve as one possible criterion for determining whether or not a person is eligible for Medicaid benefits. When this item is included in a recode, the effect of changing "no" to "unknown" for "dummy" records may not be noticeable to the analyst because "no"s and "unknown's" are treated identically in the recode. However, the effect of the modification will be perceptible when the item itself is under consideration. Table 1 shows the effect of changing "no" to "unknown" for "dummy" records on a selected individual item (data location 453 on the public use data file) using weighted data (annual weight - data location 233). - 2 - Identification of "dummy" records on the 1989 NHIS Health Insurance Public Use Data File (Cont'd) TABLE I Number in thousands Percent distributions Unknowns included Unknowns excluded AFDC STATUS Not Modified Not Modified Not Modified Modified Modified Modified All AFDC 243,532 243,532 100.0 100.0 100.0 100.0 Has AFDC 7,729 7,729 3.2 3.2 3.2 3.3 Does not have 235,773 228,857 96.8 94.0 96.8 96.7 AFDC Unknown 30 6,946 0.0 2.9 - - Note: Subfields may not add to totals because of rounding error. As may be noted in the last two columns of the table, when the "unknown" cases are excluded in the calculation of the percent distributions, the estimate of persons receiving AFDC increases from 3.2 to 3.3 percent with the modification and the estimate of those not receiving this type of benefit decreases from 96.8 to 96.7 percent. - 3 - Identification of "dummy" records on the 1989 NHIS Health Insurance Public Use Data File (Cont'd) TABLE 2 Number in thousands Percent distributions Unknowns included Unknowns excluded Public Not modified Modified Not Modified Not Modified Assistance modified modified All Coverage 243,532 243,532 100.0 100.0 100.0 100.0 Covered 14,983 14,983 6.2 6.2 6.2 6.4 Not Covered 226,588 219,672 93.0 90.2 93.8 93.6 Unknown 1,960 8,876 0.8 3.6 - - Note: Subfields may not add to totals because of rounding error. Table 2 shows the effect of the change on the recode for receipt of public assistance (mostly Medicaid) health care coverage (data location 464). Again, weighted data (annual weight - data location 233) are presented. This recode is based on the following variables: (1) AFDC (data location 453), 551 (data location 454), Medicaid card (data location 456), and other public health care (data location 457). The last two columns of the table indicate that when the "unknowns" are excluded, the estimate of persons with public assistance coverage increases with the modification from 6.2 to 6.4 percent and the estimate of those not covered decreases from 93.8 to 93.6 percent. The modification should have little or no effect when tabulating the number or percent of persons with or without any type of health care coverage. This is because the "dummy" records improperly classified as "no" that are related to public assistance and military-VA coverage are "unknown" for the other two types of health care coverage (private insurance and Medicare) and thus are classified as "unknown" when all four types of coverage are considered at the same time. Thus, utilizing the criteria usually used by NCHS to tabulate this estimate produces identical results for the modified and unmodified tabulations for persons with and without any type of health care coverage. Identification of "dummy" records and modification of other relevant fields - 4 - Identification of "dummy" records on the 1989 NHIS Health Insurance Public Use Data File (Cont'd) To identify the 3,300 "dummy" records, control on "unknown" for Medicare (code 3, data location 336) and "both hospital and doctor unknown" or "either or both hospital/doctor coverage unknown" for private health insurance coverage (code 14 or IS, data location 418-419). The following modification will properly reclassify the "dummy" records for those fields in which they were misclassifed as "no" instead of "unknown". (1) To tabulate " no" for these items, exclude records with both 336=3 and 418-419 (14 or 15) even if they are coded "no" on the item in question, (2) To tabulate "unknown" for these items, include records with both 336=3 and 418-419 (14 or 15) even if they are coded "no" on the item in question. Thus, the modified tabulation of AFDC (data location 453) would be: Yes: Data location 453 = 1 No: Data location 453 = 2 and not (336=3 and 418-419=14 or 15) Unknown: Data location 453 = 3 or (336=3 and 418-419=14 or 15). The modified unweighted marginals would be: Yes: 4360 No: 109253 Unknown: 3316 The public use data locations most directly affected by the misclassification of "dummy" records as "no" instead of "unknown" are listed below: Data location Variable name 453 AFDC or ADC 454 Supplemental Security Income 455 Received Medicaid - past 12 months 456 Medicaid - type of card 457 Covered by other public assistance program that pays for health care 458 Now receives military retirement benefits 459 Type of military benefits 460 Now covered by CHAMP-VA 461 Now covered by any other program providing health care for military dependents 464 Public assistance health insurance coverage 466 Military health benefit 495 Covered by CHAMPUS If you have problems related to the subject of this memo, call P. Ellen Parsons at (301) 436-7100. - 5 -