Task 1: Key Concepts about weighting linked data

The properties of the NHANES interview and examination weights for linked data files with incomplete linkage due to ineligibility for linkage and non-matches are unknown. In addition, methods for using survey weights for some longitudinal analyses, where survey data are obtained at one time point and CMS data occur in later years, have not been developed. Because this is an important and complex methodological topic, ongoing work at NCHS and elsewhere is examining the use of survey weights for linked data in multiple ways.

Until specific recommendations are available, preliminary guidance is for you to create adjusted survey weights that account for linkage-ineligibility and non-matches. One approach for adjusting the NHANES sample weights follows:  first, using the whole sample (linkage-eligible, not linkage-eligible, non-matches), obtain estimated population counts by summing the NHANES sample weights within subgroups defined by combinations of gender, age group, and race/ethnicity; second, for the respondents who are linkage-eligible (the criteria by which survey participants can be potentially linked to CMS data as defined in Course 1, Module 3, Task 1), adjust the NHANES sample weight so that the total sum of the adjusted weights in each combination of gender, age group, and race/ethnicity is the same as that for the whole sample; the adjusted weights for the respondents who are not linkage eligible are 0. This approach uses a ratio adjustment of the weights for non-responders (Lohr, 1999).

Using methods similar to those described above you can produce adjusted sample weights using these two steps by programming directly in standard statistical software, such as SAS and Stata.  When deciding on the demographic combinations, particularly age categories, there should be sufficient number of respondents in each combination (about 30 is often suggested but there are no definitive guidelines).  Age categories that are consistent with the NHANES sample design can be found in the NHANES analytic guidelines (see Resources for link). It is recommended that race/ethnicity groups be consistent with NHANES sampling (Curtin, 2012).   Another option for adjusting the weights is to use SUDAAN’s PROC WTADJUST procedure in SUDAAN, version 10. The WTADJUST procedure in SUDAAN computes non-response weight adjustments using a model based calibration approach similar to a logistic regression and offers more flexibility by permitting the use of continuous variables, more main effects, and lower order interactions in the model. The WTADJUST procedure can also post-stratify to population totals.

Other approaches for handling linkage-eligibility for the linked CMS data, including alternative methods for adjusting sample weights, may be used.  Researchers should seek assistance from a statistician for guidance on their particular project. Additional information will be posted on the NCHS Data Linkage webpage as it becomes available. Please contact the NCHS Data Linkage Team (datalinkage@cdc.gov) for more information.

Resources

• Lohr S: Sampling Design and Analysis. Albany, NY, Duxbury Press, 1999.
• NHANES analytic guidelines
• Curtin LR, Mohadjer L, Dohrmann S, et al. The National Health and Nutrition Examination Survey: Sample design, 1999–2006. National Center for Health Statistics. Vital Health Stat 2(155). 2012.
• Research Triangle Institute (2008). SUDAAN Language Manual, Release 10.0 Research Triangle Park, NC: Research Triangle Institute.
• Linkage Eligibility Definition, Course 1 Module 3 Task 1