Survey Weights Help

[Nathanael Lucas]

I’m wondering if someone can help me with a sample weights question. I’m happy with further output but not convinced I’ve got the weight right, as no Population Size/Strata proportion has been given.
A survey studying survival of melanoma, breast, and lung cancer patients was carried out. Assuming all total cancer populations were the same (N__melanoma=N__breast=N__lung) a sample of 120 patients (40 melanoma, 40 breast, and 40 lung) was completed, set up the survey data and the do analysis……..
So my attempt,
I think survey weight is 1/120=0.00833333333 as everypiece of data is proportional and contributory
gen dataweight=0.008333
svyset [pw = dataweight], strata(cancertype)

It seems like I’m missing something?

[Steve Samuels]

svyset and the other svy commands in Stata apply only to random samples of a defined finite population. Random samples are selected with random numbers. Your n= 40 patients of each types do not appear to have been selected by random sampling, If that's the case, no svy analysis is appropriate. But perhaps indeed each group was randomly selected from a larger group of patients of the same type. If so, please provide details, including the total number in the larger group.

If a simple random sample of size \(n_i\) is drawn from a finite stratum population (labeled "i") of known size \(N_i\), the probability of selection is \(n_i/N\_i) and the sampling weight for a stratum would be:
\[
wt_i = \frac{N_i}{n_i}
\]
In your case \(n_i=40)\ would be the denominator and you have to supply the numerator. If the \(N\)s are unequal, then the stratum weights will be unequal. For a brief introduction to random sampling see the introduction to survey commands in the Survey manual. However even without a random sample, you can do a standard (non-survey) survival analysis. How well the results generalize to other patients is then a matter of judgement.

[Steve Samuels]

Sorry, for the botched Mathjax: I did check it in a test post, Now the Mathjax is not formatting at all. Here's a corrected version"

svyset and the other svy commands in Stata apply only to random samples of a defined finite population. Random samples are selected with random numbers. Your n= 40 patients of each types do not appear to have been selected by random sampling, If that's the case, no svy analysis is appropriate. But perhaps each group was randomly selected from a larger group ("population") of patients of the same type. If so, please provide details, including the total numbers of patients in the larger groups.

If N_i is the size of the population in stratum i and n_i is the sample size, the sampling weight is W_i = N_i/n_i.

In your strata, n_i = 40 would be the sample size and you have to supply the numerator N for each stratum. If the N's are unequal, as they will be in practice, than the sampling weights will be unequal.