Bias of polarimetric estimators for binary star inclinations

It is shown that the polarimetric 'modelling' used by previous authors to obtain the least squares fit to polarimetric binary data will tend to yield inferred values of inclination greater than the true value. This statistical bias is most pronounced at high noise levels and low inclinations when the inferred value and the formal linear error will have no bearing on the actual value. As noise levels increase this inferred inclination approaches 900. This complication in the parameter determination using the canonical model and least squares procedure has been obscured by the fact that the binaries observed tend, through selection effects, to have high inclinations. Errors for inclination which are established by formal techniques are seriously over optimistic except at extremely low noise levels. (author)