Before the model, the question

It is tempting to begin an analysis by choosing a method. A new model, an elegant technique, a familiar workflow. But the most useful part of the work often comes before any of that: deciding what we actually want to know.

A dataset is not a question. It is a record of choices—what someone measured, whom they asked, and what they left out. Understanding those choices changes the kinds of conclusions we can reasonably draw.

A little time spent upstream

Before fitting anything, try writing the question in one plain sentence. Then ask what a useful answer would change. Would it inform a decision? Suggest an experiment? Rule out an explanation?

This does not make the technical work less important. It gives that work somewhere to go. The best model is not necessarily the most sophisticated one; it is the one that helps answer a question worth asking.