I want to combine three cycles of NHANES data (2013-2014, 2011-2012, and 2009-2010), and I have a question about the weights. I want to exclude participants who did not answer a particular question. However, if I remove these people from the dataset, my weights will not add up to 100. Is this OK to not have the weights add up to 100 or do I have to multiply each individual weight by a number across the board so that the sum adds up to 100?
For example, let's say I have 100 people, and I want to exclude 5 of them, for the remaining 95, would I multiple their weight by 1.053? (95*1.053=100)
Thank you all for your help!
Dear Walter
NHANES is a survey study based on multistage complex sampling from the US general population. You have to use Stata or SAS for statistical analysis using survey analysis method. In the survey analysis you have to introduce strata and cluster in the sampling for the software (based on two or more variables), after that, you can use any other statistical methods such as logistic model and etc. Weighting method is different based on your interested variables. You can find you weighting variable by reading this instruction: http://www.cdc.gov/nchs/nhanes/nhanes_questionnaires.htm
Sample weights consider unequal probabilities of selection due to different sampling designs, non-response or non-coverage than number of participants per se. You may be aware of the previous pooling of NCHS data sets to produce 2000 CDC growth charts. Pl refer to the following article for methods:
http://www.cdc.gov/nchs/data/series/sr_11/sr11_246.pdf
Hi Walter, At first glance I would say yes you could weight each subject as with the formula you gave. But you're still left with the caveat that some subjects didn't answer some questions. Depending on your data and how you analyze this, it sounds like you are then 'filling in' the non-responders with an averaged response from those who did respond. This may be legitimate as long as you explain this in your methods and note any limitations this could introduce, if any. But as Sashi above pointed out, NHANES already over-samples certain demographic subgroups so as to get a more accurate picture of their data and you may be further weighting this downstream, rather like a series of lenses that successively distort the original picture. This can get complicated, and it depends on what you are analyzing and what the purpose of your weights are. If the weights are simply to give equal weight to the sample population size of each NHANES cycle, then weighting them like you describe should be okay. But I recommend you keep searching the literature a bit more for guidance. And the NHANES statisticians themselves might be able to guide you if you make an inquiry.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis