A Substantial Bias in Nonparametric Tests for Periodicity in Geophysical Data

See allHide authors and affiliations

Science  13 Nov 1987:
Vol. 238, Issue 4829, pp. 940-945
DOI: 10.1126/science.238.4829.940


A nonparametric test that has been used to conclude that extinction rates are periodic with a period of 26 million years is shown to be substantially biased toward this conclusion, regardless of whether or not the data are periodic in origin (and, indeed, regardless of the actual period if they are in fact periodic). The test is shown to be sensitive to measurement error of a type expected with these data (early recording of extinctions due to missing fossil specimens, the "Signor-Lipps effect"), and it is shown that because of the unequal spacing in time, such models may be expected to produce statistically significant but artifactual periods of (in this case) exactly 26 million years over the span of time actually used.