necessary to account for stratification when only interested in proportions, not CIs or p-values?

[laura944]

I'm analyzing data from the National Survey of College Graduates (NSF, 2019). I am only interested in reporting on descriptive data, specifically what the proportions of the population are in certain groups e.g. X% of the population has a PHD and is female and Hispanic, not confidence intervals or p-values. The file includes a "survey specific weight" variable, which I would apply to the analysis.

Question: The design of the survey is described as a "stratified sampling design" with stratification cells (exact description below from NSF site). If I am only interested in the proportions, do I need to account for stratification? Or is just applying the weight variable sufficient?

Full description: "Sample design. The NSCG uses a stratified sampling design to select its sample from the eligible sampling frame. Within the sampling strata, the NSCG uses probability proportional to size or ystematic random sampling techniques to select the NSCG sample. The stratification cells are defined by the following variables: • Demographic group • Highest degree type • Occupation field and bachelor’s degree field As has been the case since the 2013 NSCG, the 2019 NSCG includes an oversample of young graduates to improve the precision of estimates for this important population." - from NSF website

[ballardw]

Learning suggestion: Use the data from that source.

The run the analysis with and without strata and sample information. See what it does to proportions.

[SAS_Rob · Accepted solution]

The weighted proportions themselves do not depend on the stratification variable so you can safely report them without including the stratification.  The standard error and confidence intervals that will change however.

Of course, assuming you have the stratification information and given the ease of simply using the STRATA statement in Proc SURVEYFREQ, it is probably a good idea to include it anyway in case you want other measures.

[laura944]

Very helpful, thank you.