The National Survey of Family Growth, Cycle IV, evaluation of linked design.

Research was undertaken to quantify the effects of costs of alternative methods for selecting sample women for the National Survey of Family Growth (NSFG) from the National Health Interview Survey (NHIS). This report presents estimates of the effects of alternative design options, obtained by statistical modeling techniques, for linking the NSFG with the NHIS; the cost data and the statistical precision of estimates were based on data from the NSFG, Cycle IV. The estimated survey costs and projected response rates for alternative linked design options and for the unlinked design are compared for fixed precision. The findings confirm that substantial gains in the NSFG design efficiency were obtained by linking the NSFG sample design to that of the NHIS.

PIP: This article describes the research methodology used in Cycle IV of the National Survey of Family Growth (NSFG). The NSFG used a subsample of women living in households that participated in the National Health Interview Survey (NHIS), which is a continuous survey in 198 metropolitan areas and clusters of nonmetropolitan counties in the United States. Chapter 1 is devoted to a description of the sampling design and subsampling design, respectively, of the NHIS and NSFG. An intensive follow-up was conducted among a 50% subsample of nonresponse cases. Chapter 2 covers a discussion of the application of "balanced repeated replications," which was the technique used to estimate within-unit variance and within-segment variance. In general the between-segment variance was small for most estimates. There was a 7% between-primary-sampling-unit (PSU) variance for all races combined and an 11% between-PSU variance for Black women. Chapter 3 includes a presentation of a theoretical model of the variance of a ratio-adjusted estimate. The effects of variation in weights is discussed. A fitted model for design effects was compared to a direct design effects model. The aim of examining design effects was to provide a comparison of NSFG and NHIS linked and unlinked designs. Chapter 4 considers four designs which vary in the number of PSUs, the number of segments, the number of designated households, the number of designated women from those households, and the number of interviewed women. The design effect of oversampling of strata with a high proportion of Blacks was calculated. The number of women that would have to be interviewed was calculated in an unlinked design in order to attain the same precision comparable to a linked Cycle IV design. Chapter 5 gives the direct costs for each of the four design options for 10 activities in data collection.

Sample design: Third National Health and Nutrition Examination Survey.

This report presents a detailed description of the sample design for the Third National Health and Nutrition Examination Survey, 1988-94, including a brief description of research that led to the choice of the final design. The National Health and Nutrition Examination Survey (NHANES) is one of the major surveys of the National Center for Health Statistics, Centers for Disease Control. Information on the health and nutritional status of the noninstitutionalized population of the United States is collected through the NHANES household interviews and standardized physical examinations.

Random digit dialing in selecting a population-based control group.

Results are described from four epidemiologic studies in the United States which used random digit dialing in over 30,000 households to identify controls from the general population for use in case-control studies. Methods and problems in telephone sampling are discussed. It is concluded that if complete population rosters are unavailable and if the population to be sampled has the high rates of telephone ownership typical of much of the United States, telephone-based sampling can yield a nearly random sample of the individuals in a population, often at much less expense than can dwelling-based sampling.

Reliability of Estimates With Alternative Cluster Sizes in the Health InterviewSurvey.

The practice of grouping the elements of the universe of study into clusters and sampling the clusters is a common feature of sample designs. Dealing with clusters of elements ahnost invariably increases the sampling error of the statistics estimated from the sample. However, clustering usually reduces the per-unit costs of the sampling and data collection. As the cluster size is increased, the costs of conducting a survey are usually reduced but the sampling errors of the statistics generally increase for a given sample size. The increase in variance due to clustering may be considered as a "loss" in the reliability of a statistic. The purpose of this study is to obtain measurements of the loss in reliability due to clustering of households using various cluster sizes for data collected by the Health Interview Survey (HIS).