Abstract

The variance in isotope-dilution measurements of Rb87/Sr86in geological samples is shown to be proportional in general to (Rb87/Sr86)2, so that simple regression methods for estimating Rb-Sr isochrons are not formally valid. A new regression method which allows for known experimental error in both Sr87/Sr86 and Rb87/Sr86 and which weights the data to allow for the nonuniform Rb87/Sr86 variance is given. The method is illustrated with the data for the Heemskirk granite, western Tasmania. Analysis of the data shows that, although the simple method gives quite accurate estimates of the isochron parameters, it tends to underestimate the uncertainty in the age and to overestimate that in the initial Sr87/Sr86. Models with geological variation over and above the experimental variation were also examined. Alternative models considered are geological variation independent of Rb87/Sr86 or proportional to (Rb87/Sr86)2, or a compromise between these.