Correlation coefficients fluctuate across samples just as means and percentages do

Evidence

Hyman and Sheatsley take issue with the notion that sampling does not matter in a study of relationships among variables: “Correlation coefficients, just like means or percentages, fluctuate from sample to sample and may well vary in different populations.” — John T. Jost and Jim Sidanius, Political Psychology, 577

Categorization

• XREF: Challenges the standard defense that non-representative samples are fine when only testing variable relationships — relevant to any quantitative social science where sampling frames restrict variance.