Journal of Official Statistics, Vol.19, No.1, 2003. pp. 31–43
Sequential Weight Adjustments for Location and Cooperation Propensity for the 1995 National Survey of Family Growth
Vincent G. Iannacchione
Abstract:Distinct patterns of location and cooperation were found among women selected for Cycle 5 of the National Survey of Family Growth (NSFG-5). For example, minority women were harder to locate than were other sample women. Once found, however, these same women were more likely to participate in the survey than other sample women. Two logistic regression models were developed sequentially so that predictors related to the locating process could be distinguished from those related to the cooperation process. For the location propensity model, the design-weighted logistic regression algorithm was modified to compute adjustment factors for the sampling weights that preserved the full-sample weighted means for specified analysis domains among women who were located for the NSFG-5. For the cooperation propensity model, a similar process was used to preserve the location-adjusted weighted means among women who participated in the survey. Receiver Operating Characteristics (ROC) curves were used to assess the overall predictive ability of the combined models. The linkage of the NSFG-5 to the 1993 National Health Interview Survey (NHIS) enabled a large number of candidate predictors obtained from the NHIS interview to be considered for each model. As expected, predictors indicating the presence or absence of NHIS contact data (e.g., a telephone number) were significant factors affecting location propensity. However, the refusal by some NHIS participants to provide certain types of contact data also adversely affected cooperation propensity and was interpreted as a predictor of resistance to participation in the NSFG-5.
Keywords:Weight adjustments for nonresponse; linked sample design; logistic regression; ROC curves.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group