The bias of the estimated totals using different ratio type imputation procedures was examined under models and a Monte Carlo study. In addition the mean square error was computed in the Monte Carlo study.

The current imputation procedure for estimating current month sales is p - unbiased under the ratio model with model variance either proportional to the previous month sales or to the square of the sales.

Under the assumption that the data are missing at random, the bias and MSE for estimating total current month sales from the given data by using different ratio and regression imputation procedures were derived. An optimum radio imputation procedure was also derived. The reported data for 9 SIC’s of the 1982 Monthly Retail Trade Survey were used for the Monte Carlo study.

The results of the Monte Carlo study showed that the current ratio imputation procedure is competitive with the optimum ratio procedure and better than other comparable procedures studied. Under the current ratio imputation procedure, a better set of imputation cells can be formed by using sales quantiles as cutoffs within groups as opposed to the current procedure that uses fixed sales cutoffs. For example, the decrease in MSE in 6 of 9 SIC’s ranges from 12% to 59% by using 1/8 quantiles. We recommend that changes in the current imputation cell definitions be considered.