Rival estimation theories rest on data yielded by different methods.
Effects of anchors and sampling were tested in 2 × 2 design.
Free and anchored estimates were best predicted by a mixed log-linear model.
Logarithmicity of estimates declined as a function of age and anchors.