M.L.
R^2 measures the amount of variance in the observational data described by the model. Thus, it describes a correlative relationship in past data.
If your model captures a causal relationship, then your R^2 will give you a sense of how well you might predict future outcomes. However, many folks commit statistical malpractice because they 1) don't understand what makes a model a good model and 2) confuse correlation with causation, and so they only interpret the model through the R^2.
As you talk to, in the end, human behavior is subject irrationality and involves a complex interdependence, and so R^2 values tend to be low in the social sciences (this may not be problematic for your model depending on what you are trying to do, but that's a tangent not needed for the thread).
Bookmarks