Comparing analyses across differently defined populations reveals selection bias.

Evidence

Unlike the usually uninterpretable ρ coefficient in Heckman selection models, we have seen that there is often a clear logic to the selection biases that reveals itself when we look at statistical analyses comparing different populations. Many have stressed how power and deterrence factors can produce selection effects. Our analyses have confirmed in a very clear fashion that these concerns are well founded. In contrast, there is almost no work on variables that should produce no selection effects. For example, we found that rivalry variables were much less susceptible to these kinds of problems. This chapter then confirms some well-founded hypotheses, but at the same time pushes the selection bias agenda forward. — Gary Goertz, Social Science Concepts, 1095

Categorization

• SEED: Angle: ‘Almost nobody studies what doesn’t break’ — the neglected research program of variables immune to selection effects.
• XREF: connects to: Findings may depend on the particular groups researchers select. — Susan T. Fiske et al., A Model of (Often Mixed) Stereotype…, 158