Biostatistics

A linear regression's residual variability appears to increase with fitted values. Why might this matter for conventional regression inference?