Evidence for Evolution or Bias in Host Extinctions of Type 1a Supernovae at High Redshift

Type 1a supernova magnitudes conventionally include an additive parametercalled the extinction coefficient. We find that the extinction coefficients ofa popular ``gold'' set are well correlated with the deviation of magnitudesfrom Hubble diagrams. If the effect is due to bias, extinctions have beenoverestimated, which makes supernovas appear more dim. The statisticalsignificance of the extinction-acceleration correlation has a random chanceprobability of less than one in a million. The hypothesis that extinctioncoefficients should be corrected empirically provides greatly improved fits toboth accelerating and non-accelerating models, with the independent feature ofeliminating any significant correlation of residuals.Comment: 20 pages, 5 figures, to appear in Ap