Swapping the comparison group can flip a coefficient’s sign or significance.

Evidence

There are no easy answers to this heterogeneity problem. It depends a great deal on the causal inference goals of the scholar and the kinds of heterogeneity of the negative cases. The stakes are high: by choosing one versus another comparison group, statistical results can flip sign or significance level. It is worth stressing that this is maintaining exactly the same positive pole cases. — Gary Goertz, Social Science Concepts and Measurement, 456

Categorization

• QUESTION: How often do published findings in political science or sociology hinge entirely on an unexamined choice of negative cases? A replication sweep could quantify this.