As you obviously have to handle uncertain and incomplete information, I'm wondering if probabilistic networks fit your needs? A special kind of probabilistic models (Bayesian networks) is applied in a number of different disciplines. See the proceedings of the Bayesian Modeling Applications Workshop for instance (http://seor.gmu.edu/~klaskey/BMAW2014/agenda.html).

I would be really interested if this approach also fits the needs of your domain.

One common strategy in mixed methods is to begin with qualitative interviews. You can set those up to help you devise you own measures, or you can get responses to existing measures as "stimulus material."

I like to think of the basic idea here as "Why don't you ask them?"

I think it would depend a bit on the theory driving the intervention and where the anticipated changes reside. Using your example of advocacy, if changes reside in individuals within the community, you might take one approach. If they are at a system or societal level, you might use another.

A few references might include:

http://www.hfrp.org/evaluation/the-evaluation-exchange/issue-archive/advocacy-and-policy-change/a-guide-to-measuring-advocacy-and-policy

http://www.aecf.org/resources/a-guide-to-measuring-advocacy-and-policy/

https://www.ncrp.org/files/media/foundationreview-lranghelli-measuringtheimpactsofadvocacyandcommunityorganizing-lowres.pdf

You might also use approaches from other domains (e.g. knowledge translation or policy) depending on what your desired outcomes are. Although it's public health specific, the National Collaborating Centre for Methods and Tools might be helpful: http://www.nccmt.ca/ or the National Collaborating Centre for Healthy Public Policy http://www.ncchpp.ca/en/

-cam