Traditional science has a clear idea of how research is to progress, rationally speaking: First you build a theory, for example, by observation, expert interviews, and the like, and then you generate hypotheses to test the theory. Over time, some theories will stand the test of time and will be considered valid.
Sadly, most software engineering research today, even the one published in top journals and conferences, often skips the theory building process and jumps straight to hypothesis testing. Vicky the Viking, from the accordingly named TV series of way back comes to my mind: Out of the blue, some genius idea pops into the researcher’s mind. This idea forms the hypothesis to be tested in research.
This approach to science comes with high costs. What if the hypothesis is wrong? The researcher, convinced of his or her idea, may be consciously or subconsciously tempted to fudge the data to make the hypothesis come true, after all, it makes so much sense.
Theory building first helps avoid this problem. Done right, it guides researchers towards sensible theories and hypotheses so that all the money spent on the hypothesis-testing research has a real chance of coming to both a correct and desired result, that is, true hypotheses and thereby validated theories.