Time Series Analysis of U.S. Army Enlisted Force Loss Rates

摘要

The analysis and prediction of personnel loss behavior is critical to effectivemanpower planning and to the U.S. Army's Enlisted Personnel Strength Management System (EPSMS). In support of efforts to modernize the EPSMS, this thesis examines the method by which the Enlisted Loss Inventory Model (ELIM) analyzes loss rates and forecasts them into the future. Time series analysis techniques seek to identify patterns in data and forecast them into the future via time based extrapolations. Four such methods were used to construct loss rate forecasts from data. These methods were the arithmetic mean, exponential smoothing (the current ELIM method), seasonal exponential smoothing and an autoregressive moving average model. Forecasted rates were used to project force strengths which were in fact known. The resulting errors in forecasted strength were analyzed, compared and contrasted with respect to the methods. Error analysis revealed no significant performance differences between the methods. Hence, the simplest methods (mean and exponential smoothing) may be viewed as more economical and preferred.