A unified finite population theory is developed for composite estimators of both the current level and change in level between consecutive occasions when a rotation sample design is employed. Explicit variance functions are given under the assumption that exponential and arithmetic correlation patterns hold over time for the characteristics of interest. Optimum values of the weight coefficients in the composite estimators, the optimum number of consecutive visits by a sampling unit, and the gain in efficiency relative to the use of simple estimators are all determined numerically. Moderate gains for level and appreciable gains for change are achieved when the correlation is high.