For every new sample for the commodities and services (C&S) component of the U.S. Consumer Price Index (CPI), the Bureau of Labor Statistics attempts to produce a C&S sample design that allocates outlets and quotes in an optimal fashion. This item-outlet optimization C&S sample design requires the estimation of components of variance for the three factors in the design: non-certainty primary sampling units (PSUs), item-strata and outlets. The fourth component of variance is the error term. To produce these components of variance, a Random Effects model is chosen, with the independent random variable for the model an individual price change. Weighted Restricted Maximum Likelihood (REML) estimates are used to calculate the variance components. This paper compares an earlier set of variance components obtained from 1993-96 CPI data with an updated set of variance components obtained from 1997—2000 CPI data. We compare and critique the methodologies used as well as the empirical results obtained.