Post-Data-Fitting Theorizing

If you see an effect in physical science, and it is surprising, you try to propose an explanation, one that fits in with other data. BUT then you do experiments and theoretical calcualations to be sure that the explanation and its mechanism are actual.

Often, in social science, people will do a fine statistical analysis of their data, often big data, and then try to explain various regression coefficients, etc. All these explanations are likely to be made up, often not linked to any well articulated theory. This need not be a problem. But if you propose an explanation, you then have to go out and test that explanation in a different context or with different data or with a finer theory that incorporates the explanation. You can't just tell a story and expect anyone to believe it, even if they find it cogent. For just telling a story does not exclude other good explanations.

So gather data, develop some theoretical explanation (one that at first might be merely a linear or some sort of regression), do your statistical analysis. If all fits, you might be surprised by some of the coefficients, and then you will need to explain them or plea fluctuation (no sin!). And then you need to theorize and test.

1. Data
2. Theory that leads to testable propositions about the data, or at least testable propositions, which might well be a set of reasonable regressors.
3. Statistics
4. Surprise?-- Yes -->2
                        No-->Develop a theory that justifies your #2 theory.-->Write the article.