Hello Stata Users,
I am looking for some help in trying to set the svy design to accurately reflect my sample. I have individuals who are nested in worship communities for a total N of 828.
The data come from an original face-to-face survey of individuals in which the PSU is a worship community. There are 8 PSU in the survey, covering three religious denominations, which I can treat as the strata. The sample sizes across the PSU range from N=59 to N=129. The individuals from each PSU were selected for interview using some version of random sampling without replacement. I have an interest in making comparisons across the PSUs and not just considering the pooled sample of all observations. In all cases I have data about the real population of each PSU, and in several instances the sample is greater than 10% of the PSU's population. For this reason, I want to use a finite population corrector to adjust standard errors.
The fpc command in Stata requires that the fpc is the same across all observations and therefore does not provide the correction that I am hoping for. I created an fpc that summed the finite populations of all PSU, but it is too large to have any difference on the standard errors as some of the PSUs have populations that are much larger than others.
I can manually calculate the fpc for each PSU using the standard formula ( √(N-n) / (N-1)) but I do not know if there is a way to apply that to the standard errors produced in svy to estimate new confidence intervals, etc.
I would value any suggestions about a path forward.
Danielle
I am looking for some help in trying to set the svy design to accurately reflect my sample. I have individuals who are nested in worship communities for a total N of 828.
The data come from an original face-to-face survey of individuals in which the PSU is a worship community. There are 8 PSU in the survey, covering three religious denominations, which I can treat as the strata. The sample sizes across the PSU range from N=59 to N=129. The individuals from each PSU were selected for interview using some version of random sampling without replacement. I have an interest in making comparisons across the PSUs and not just considering the pooled sample of all observations. In all cases I have data about the real population of each PSU, and in several instances the sample is greater than 10% of the PSU's population. For this reason, I want to use a finite population corrector to adjust standard errors.
The fpc command in Stata requires that the fpc is the same across all observations and therefore does not provide the correction that I am hoping for. I created an fpc that summed the finite populations of all PSU, but it is too large to have any difference on the standard errors as some of the PSUs have populations that are much larger than others.
I can manually calculate the fpc for each PSU using the standard formula ( √(N-n) / (N-1)) but I do not know if there is a way to apply that to the standard errors produced in svy to estimate new confidence intervals, etc.
I would value any suggestions about a path forward.
Danielle
