Two tests have been proposed for testing the hypothesis that a linear rather
than a ratio model explains the relation between variables. A parametric
F-test and the nonparametric peak test both rely upon testing residuals for
homoscedasticity as against the alternative that residual variances are monotonic
in the independent variable. It should be noted that the peak test can
also be used as a two-tailed test when one examines whether the residual variances
are monotonically increasing or decreasing in the independent variable.
Finally, it should be mentioned that although we have not explicitly dealt
with the case treated by Theil where the variance of the residuals is proportional
to the square of the mean of the dependent variable,’6 the peak test appears
applicable to this case as well if the residuals are considered ordeped by the
predicted value of the dependent variable. The power of both tests is estimated
from sampling experiments and found satisfactory.