Complex Survey Design - Analysis Conundrum <- Not homework

#1
Hi Stats forum,

So, I am the analyst for a stratified cluster sample survey. We surveyed facilities to evaluate a safety program. The survey had 5 parts. Interviewers went to the facility and staff were randomly chosen to be interviewed. Different staff were interviewed for different parts of the survey, since the topic of each part would be in regards to different types of staff.

The facilities all have weights calculated (weights are split over the number of staff interviewed to sum up to the facility weight because all questions are more facility level.). And I've completed basic descriptive analysis.

Here's where I am entering uncertain analysis territory.

The PI for this study would like me to examine the association of questions from different parts. In this case the outcomes are both counts as well as categorical. For example, section 2 has questions with regard to the number of certain safety devices in stock while section 3 has questions on the number of times a certain injury occurred. Section 2 involved interviewing inventory staff while section 3 involved interviewing floor workers.

On the surface this seems easy - facilities throughout the parts have the same weights, same sampling scheme.

But, the number of staff interviewed is not the same. I am not sure how to combine the data from the two.

Do, I somehow aggregate the questions of interests down to one row per facility?

It's obvious this isn't a simple merge and then plunk it into survey software (either R or SAS).

Any help from survey/sampling vets is welcome.