The Hamilton method for estimating CPI bias is simple, intuitive, and has been widely adopted. We show that the method conflates CPI bias with variation in cost of-living growth across income levels. Assuming a single price index across the income distribution is not consistent with the downward sloping Engel curves that are necessary to implement the method. We suggest an approach that disentangles genuine CPI bias from differences caused by comparing changes in the cost of living across income levels– non-homotheticity. For the period Hamilton studies, this yields substantially lower estimates of CPI bias and therefore implies lower income growth.