Sample or Person Weights, and NIU Issues

Hello, I am analuyzing HIV testing trends using the NHIS.

Since my interest is on African born persons versus those born in Latin America, I have decided to pool the data over the years 2000 to 2015 to increase the sample size.

  1. When I apply my weights, should i divide it by 15? that is PERWEIGHT/15 or SAMWEIGHT / 15?

  2. Secondly, my dataset include socio-demographic variables (that were asked of all respondents), and other health behavior related variables (including those on HIV testing, mental health) that were asked of only adults. Given that my one of my main outcome variables is HIV testing, should I be using the Sample weights (SAMWEIGHT) instead of PERWEIGHT?

  3. After extraction, I realized that I have a number of variables with NIU (not in universe codes)? The question on HIV testing for example, has close to 67% of the sample not in universe, because they were asked of only adults.

– Should I delete these NIU cases from the data before doing any analysis?

– Does restricting my analysis to only those aged 18+ achieve the same result ?

Thanks in advance for all your assistance.

cheers, E

  1. Dividing the weights by 15 as you propose should be sufficient for generating estimates that are representative of the 2000-2015 population.

  2. Based on your description, it sounds like you should use the SAMPWEIGHT variable to weight your analysis. This means that the population you are studying is restricted to people over the age of 17. Though there will be people in the dataset who are over 17 and still NIU for the HIV variables (see 3 below), when SAMPWEIGHT is used your estimates will be representative of the national 18+ population.

  3. A “sample adult” is actually a special designation for respondents who were asked further questions. Sample adults can be identified using the ASTATFLG variable. So, restricting your analysis to people 18+ will still leave a significant number of NIU cases, but restricting to only sample adults using ASTATFLG will drop all NIU cases from your analysis.

I hope this helps