Concept formation should be driven by semantics and theory, not by empirical data.

Evidence

It is worth stressing that this discussion has made very little reference to empirical issues. How concepts relate to the coding of cases and the distribution of those cases will be a key theme in later chapters. My approach to concepts is top down: it is driven by semantics, normative concerns, and causal mechanisms and theories. — Gary Goertz, Social Science Concepts and Measurement, 550

Categorization

• SEED: Angle: a methodological essay arguing that measurement debates invert the proper order by letting available data determine what concepts mean.
• XREF: Connects to Sartori-style concept-stretching debates and to the measurement literature in the reader’s social-science methods shelf.