Realistic use of two‐error regression treatments as applied to rubidium‐strontium data

Four two‐error linear regression treatments have been reported in the literature for use in Rb‐Sr geochronology. We have made a general comparison of these treatments and have examined some of the assumptions and concepts implied in their use. For certain cases we show that analytical errors involved in Rb‐Sr measurement are normally distributed and uncorrelated and may thus be regressed with any of the four published treatments. Furthermore, by the use of simulated test data, we show that all treatments provide essentially identical regression parameters (slope, intercept, and associated error estimates), provided that the observed scatter can be accounted for solely in terms of the experimental error. Such data are said to define an isochron. Data that possess scatter in excess of experimental error (i.e., geological error) are said to define an errorchron , and for such data the existing treatments may not furnish reliable regression parameters. One of the most important provisions of several of the treatments is a statistical index for distinguishing between isochrons and errorchrons. Use of this index requires an allowance for small‐sample statistics based both on the number of samples regressed and on the number of replicates used in calculating the experimental error. As the value chosen for the experimental error also directly affects the final isochron error estimate, a Student's t multiplier (derived from the number of replicates used in calculating the experimental errors) should be applied to the final isochron error estimates.