Justification for Change Letter

DEPARTMENT OF HEALTH & HUMAN SERVICES Public Health Service

Centers for Disease Control and Prevention

National Center for Health Statistics

3311 Toledo Road

Hyattsville, Maryland 20782

April 3, 2015

Margo Schwab, Ph.D.

Office of Management and Budget

725 17th Street, N.W.

Washington, DC 20503

Dear Dr. Schwab:

Please accept these proposed plans for the Follow-Back Survey respectfully submitted as a non-substantive change request in compliance with the terms of clearance for the National Health Interview Survey (NHIS) (OMB No. 0920-0214 exp. 12/31/2017).

The National Health Interview Survey (NHIS) Information Collection Request approved on 01/01/2015 included description of a follow-back survey to the 2015 NHIS. In developing the detailed plans for the follow-back, minor changes have been made to the design. This document describes which aspects have changed and what has remained unchanged from the original description.

As initially planned, the survey will be fielded for six months to follow back with 2015 NHIS Sample Adult respondents approximately two months after their NHIS interview. The sampling frame will consist of all Sample Adult respondents who complete the NHIS interview between July and December of 2015 and who are not selected into the sample frame for the Medical Expenditure Panel Survey (MEPS). This will result in a sample size of approximately 8,400 respondents. Originally it was estimated that as many as 12,000 respondents would be recruited, but due to a modified data collection window that estimate has been reduced.  Also as initially described, almost all of the questions in the follow-back instrument pertain to health care reform and are taken from the NHIS questionnaire. Among other research questions (described below), this will still allow the planned examination of respondents’ insurance status, with focus on the uninsured, newly insured, and those who have changed insurance.

However, unlike the initially-described multimode approach (web, mail, phone), the 2015 NHIS follow-back survey will be carried out exclusively by telephone, with mailed advance letters. This decision was driven by two considerations: First, the limited success of the email and web modes in the NHCIS (the 2013 NHIS follow-back survey); and second, the desire to utilize the 2015 follow-back survey to inform the upcoming 2018 NHIS questionnaire redesign. In light of the second consideration, the questionnaire for the 2015 follow-back survey differs from the draft version originally submitted. The new questionnaire includes all the 2015 NHIS family health insurance questions, and omits questions about interview mode and recontact information included in the previous draft.

As now planned, the 2015 NHIS follow-back survey is conceptualized as a randomized experiment in which Sample Adult respondents will be assigned to receive either the complete or a shortened version of the NHIS family health insurance questions.

Assignment to one of the two versions will occur at random. To arrive at the shortened version of the instrument, in which approximately 40 family health insurance questions are eliminated, applicable skip patterns will be programmed into the instrument to switch off the relevant questions. The enclosed draft instrument denotes the questions to be skipped for the short version via grey-shaded background color.

Naturally in the NHIS, approximately 60% of households are multi-adult households, and 40% are single-adult households. Within each multi-adult household, the NHIS instrument randomly selects an adult to complete the Family and the Sample Adult section of the interview from the household roster. In approximately 70% of cases, this results in the Sample Adult being the same person as the Family respondent; in the remaining 30% of cases, the Family respondent and the Sample Adult respondent are two different adults. This design element of the NHIS thus results in three distinct groups of Sample Adult respondents, each of which allows us to ascertain different research questions. They are denoted in summary form in Table 1, below, along with approximate sample sizes in each group.

Table 1. Proposed experimental design and group assignment

	Multi-adult household (HH)		Single-adult HH1
	Family = Sample Adult Respondent (n=3,528) **Random assignment to long or short version**	Family ≠ Sample Adult Respondent (n=1,512) **All cases to long version**	Single Sample Adult Respondent (n=3,360) **Random assignment to long or short version**
Long version	Test-retest reliability Mode effect How much do time and mode bias our estimates?	Construct validity Can we get the same information from the SA as we do from the FR?	Test-retest reliability Mode effect How much do time and mode bias our estimates?
Short version	Concurrent validity How does the shorter version perform compared to the longer, gold-standard version WHEN THERE ARE MULTIPLE ADULTS?	***No analytic benefit of assigning cases to this combination; Since we’re primarily interested in assessing whether the Sample Adult can answer for the Family respondent, it’s in our interest to concentrate the cases in that cell***	Concurrent validity How does the shorter version perform compared to the longer, gold-standard version WHEN THERE IS ONLY A SINGLE ADULT?

¹Including single-adult households allows us to assess the degree of complexity in asking about one vs. multiple adults’ insurance

As shown in the above table, this design allows us to explore two distinct questions pertaining to the redesign, along with providing built-in measures of potential bias. For the redesign, administering the short version to those cases where the sample adult and the family respondent are the same person (either in multi- or single-adult households), we can determine whether the short version of the family health insurance questions provides the same results as the longer, current version. By contrast, administering the long version to those cases where the sample adult and the family respondent are different adults allows us to assess whether asking the same questions of the two different respondents yields the same answers to the health insurance questions (and hence estimates). To ascertain potential biases resulting from interview mode (in-person NHIS interview vs. telephone follow-back interview) or delays between the initial NHIS and the follow-back interview, we can examine responses to the long version of the health insurance questions administered to those cases in which the Family and the Sample Adult respondent are the same person.

These questions are important to the 2018 NHIS redesign for two reasons. First, one of the goals of the redesign is to significantly shorten the length of the NHIS core questionnaire, of which the health insurance questions comprise a large portion. Second, to improve completion rates, under consideration is a revision to the response rules regarding who in the household can answer for the other members. It is conceivable that the NHIS may move to a design in which a single adult respondent completes all modules of the new instrument. This proposed follow-back survey provides an elegant assessment of the impact thereof on a set of estimates that is of great interest to the Department of Health and Human Services.

The proposed changes will not alter the amount of time needed to complete the questionnaire; however, the reduced estimated sample size for the follow-back survey will result in an overall reduction of 1200 burden hours for the currently approved NHIS Information Collection Request.

cc: V. Buie

T. Richardson

2