"How to solve it"  was helpful for me in this regard, but it isn't really condensed like this blog post.
After you finish "how to solve it", you can read the more advanced "Mathematics and Plausible Reasoning" 
The value of empirical work hinges on the assumption that the economic environment you're modeling is ergodic, or at the very least stationary AND that you have enough data of good enough quality.
If those conditions are not given, then you need theory. To establish relationships, not just time-series properties, this needs to hold for all variables separately as well as their copula, else you need theory. If one of the ingredients in your model is not observable, you need theory.
I am really tired of all this post-adolescent posturing by "hard scientists" who haven't thought things through. Maybe the relevant distinction is not between "real world" and "imaginary world" but between mirage and best-you-can-do-under-the-circumstances. (Then again, I am a theorist so perhaps I am overly harsh.)
I suggest Polya's Plausible Reasoning  as a better guide to "economic epistemology". Maybe look into Rational Belief theory (key reference: M. Kurz's 1994 papers [2,3]) for a "compromise" in modeling. Also, Fischer Black: Estimating Expected Return .
Fresh book recommendations delivered straight to your inbox every Thursday.