I am working with the HCUP NIS dataset and have aggregated 5 years of data (2009-2013) for analysis. I am interested in doing trend analysis by the year (percent mortality each year, etc) as well as analysis that spans all 5 years of data (overall percent mortality, etc).
Would it be correct to use the original survey weights for the yearly trend analysis and original survey weights divided by 5 (number of years aggregated) for the analysis that spans all 5 years?
Thanks in advance!
Dividing by 5 won't affect any means (percentages, etc), only totals, so you can equally well use the original weights or the weights divided by 5. Dividing by 5 is a good practice because you might want to compute some totals at some point.
However, if you want to compute overall percent mortality and the population size might be changing, you do need to decide whether to weight individuals equally or years equally. To take an extreme example: suppose the population expanded ten-fold in the last year. Using original survey weights (divided by 5 or not) would give ten times as much weight to the last year, but in some situations you might prefer to give the same weight to each year. To weight years equally, you'd need to scale the weights for each year to the same total.
