Mark Bils at Rochester has done careful work estimating quality-change bias on manufactured goods, finding the bias is quite significant for that important subset of goods, demonstrating that systematic correction is possible for narrow categories even if hard to generalize.