全球智库信息集成服务系统

• The USAF has access to rich historical information on many factors that the established research literature links to turnover.
• The most significant gap in turnover-related information available to REWS is the lack of information on member attitudes and perceptions.
• Machine-learning algorithms can increase the accuracy of individual-level predictions, and these improvements could result in more-accurate group-level estimates for separation rates.
• The REWS decision workflow operationalizes these predictions so that various USAF planners can generate customizable warnings, understand potential drivers, and assess the policy response required to preempt emerging problems.
• Simplified data inputs offer a way to refresh predictions with minimal resources, and longer-term efforts will enable improvements in data inputs, model accuracy, and functionality.

RAND Project Air Force was tasked with developing a new capability for planners: a retention early warning system (REWS) that alerts policymakers when a subgroup of U.S. Air Force (USAF) military members is at risk for future shortages. The goal of the research project was to develop a forecasting model for retention, operationalized within a prototype decision-support application, that can alert decisionmakers to emerging problems and thus allow them enough time to consider adjusting accession and retention policies before shortages occur.

The authors' overall approach to designing the system drew on widely used paradigms for solving data science problems. These paradigms emphasize understanding the business problem, drawing on a wide array of data sources and types, testing several flexible prediction approaches to optimize performance, and operationalizing the information for decisionmaking. To gain an understanding of the data sources that would be desirable for this application, the authors performed an extensive review of the turnover literature and identified gaps in existing USAF data collection efforts.

• Chapter One

  Introduction

• Chapter Two

  What Information Is Most Relevant to Predicting Retention?

• Chapter Three

  Available Sources of Information for Predicting Air Force Retention

• Chapter Four

  Modeling Approaches and Performance Levels

• Chapter Five

  How Retention Predictions Can Be Used to Generate Warnings

• Chapter Six

  Next Steps for Further Development and Implementation

• Appendix A

  Creating the Analytic Data File

• Appendix B

  Machine Learning Algorithms

• Appendix C

  Decomposition Methodology

• Appendix D

  Literature Review Methodology

• Appendix E

  Considerations and Challenges in Applying Data Science to Air Force Human Resource Problems

• Appendix F

  Policy Impact Methodology