Abstract

States often turn to a data masking procedure called microsuppression in order to reduce the risk of disclosing student records when sharing data with external researchers. This process removes records deemed to have high risk for disclosure should data be released. However, this process can induce differences between the original data and the data that ultimately gets used in education research. This article assesses the extent to which microsuppression can bias key statistics in state education data and finds that while marginal test score means tend to be preserved in the masked data, conditional means for subgroups can exhibit bias as large as 0.3 standard deviations.

摘要

为了减少与外部研究人员共享数据时泄露学生记录的风险，各国经常采用称为微抑制的数据屏蔽程序。如果发布数据，此过程将删除被认为具有较高披露风险的记录。但是，此过程可能会导致原始数据与最终用于教育研究的数据之间出现差异。本文评估了微抑制可以使州教育数据中的关键统计数据产生偏差的程度，并发现尽管边际测试得分均值倾向于保留在被掩盖的数据中，但亚组的条件均值可能会出现高达0.3个标准差的偏差。