In most surveys we are dealing with characteristics which change over time. In these surveys we may be interested in estimating population parameters in regular time intervals and studying their changes from one period to the next. Due to expanding the use of rotation sampling in statistical centers, a comprehensive study about the variance estimation on rotation sampling schemes seems necessary. Rotation sampling is one of the methods used in repeated surveys. In this paper, we emphasis on two-way balanced one-level rotation designs and by use of U.S. consumer expenditure survey data we compare estimated variances for generalized composite estimator and simple mean estimator of consumer expenditure. Results of this study indicate the generalized composite estimate is more efficient than the simple mean estimate.