This reminded me of a related thought from epi methods class I wanted to share --
In terms of evaluating and interpreting results/drawing conclusions of an evaluation of this policy, trying to frame it or understand it as an effect of housing, it reminds me of the question of "consistency", which is one of the criteria for identifying causal effects. It has to do with having different "versions" of treatment, in other words, the details of *how* a policy houses people matters, and is something you'd have to think about when trying to understand/explain the effect of housing (or argue for policies to improve housing in different settings). I can also imagine how policies might interact with local factors and would not be expected to work similarly in diverse settings, because the intervention isn't just "housing", it is the whole policy.
This difficulty also isn't necessarily unique to these kind of research questions/treatments/outcomes -- for example, if we're trying to see if high cholesterol causes heart disease, it matters *how* a study manipulated cholesterol (statin? exercise? diet? not clear how to disentangle the "high cholesterol causes heart disease", "lowering cholesterol prevents heart disease", or by extension, "lowering cholesterol is an acceptable surrogate outcome").
In terms of evaluating and interpreting results/drawing conclusions of an evaluation of this policy, trying to frame it or understand it as an effect of housing, it reminds me of the question of "consistency", which is one of the criteria for identifying causal effects. It has to do with having different "versions" of treatment, in other words, the details of *how* a policy houses people matters, and is something you'd have to think about when trying to understand/explain the effect of housing (or argue for policies to improve housing in different settings). I can also imagine how policies might interact with local factors and would not be expected to work similarly in diverse settings, because the intervention isn't just "housing", it is the whole policy.
This difficulty also isn't necessarily unique to these kind of research questions/treatments/outcomes -- for example, if we're trying to see if high cholesterol causes heart disease, it matters *how* a study manipulated cholesterol (statin? exercise? diet? not clear how to disentangle the "high cholesterol causes heart disease", "lowering cholesterol prevents heart disease", or by extension, "lowering cholesterol is an acceptable surrogate outcome").