Collaborative Research: EAGER: Automating HERD Reporting Using Machine Learning and Administrative Data

国家科学与工程统计中心（NCSES）高等教育研究与发展调查（HERD）数据是通过每年向大约900所大学和学院发送的调查仪器收集的。这些机构都以自己的方式收集数据以回应调查。大多数依靠高度劳动密集型的过程来收集和分类有关工作性质、资金来源和科学领域的支出和研究项目的信息。所采用的特别支出分类方法取决于执行任务的个人，因为他们随着时间的推移发展了必要的评估技能。随着时间的推移和机构之间可能缺乏一致性。本研究开发了利用大学管理数据的必要工具，以自动完成按科学领域、目的和赞助商类型对项目进行分类的必要且耗时的步骤，以响应NCSES HERD调查。结果将更好地理解HERD和STAR METRICS®报告数据的异同，并就如何改进每个来源的数据收集提供建议。

The National Center for Science and Engineering Statistics (NCSES) Higher Education Research and Development Survey (HERD) data are collected through a survey instrument sent to approximately 900 universities and colleges annually. Each of these institutions collects data to respond to the survey in their own way. Most rely on highly labor-intensive processes to gather and classify information about expenditures and research projects in terms of the character of the work, funding sources and fields of science. The ad-hoc expenditure classification methods employed are dependent on the individuals carrying out the task as they develop the necessary evaluation skills over time. There is potential for a lack of consistency over time and across institutions. This research develop the tools necessary to leverage university administrative data to automate the essential and time-consuming step of classifying projects by science areas, purpose and sponsor type required to respond to the NCSES HERD Survey. The results will provide a better understanding of the similarities/differences in the data reported for HERD and STAR METRICS® and provide suggestions about how data collection for each source might be improved.