A Case for Testing the Missing Data Mechanism: Can We Identify Teacher Cheating?

• Activity Number: 42
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Johnny Lin*+ and Peter M. Bentler
• Companies: UCLA and University of California, Los Angeles
• Keywords: Missing Data ; Testing Missing Data Mechanism ; Teacher Cheating ; Hypothesis Testing ; MCAR ; MAR

Abstract

Given educational policy reforms such as the No Child Left Behind Act (2001) and Race to the Top (2009), teachers are faced with increasing pressure to improve student achievement on high stakes tests. In the minority of classrooms, teachers may resort to cheating as a way to artificially raise average test scores. We will show how identifying a particular type of cheating may be an issue of distinguishing between random and systematic forms of missing data. Little (1988) developed a statistical test to accomplish this task using observed mean proxies, but there may be instances when random and systematic forms of missing data may generate the same observed mean patterns and lead the researcher to conclude that the data is missing randomly when it is actually missing systematically. As an alternative, we introduce a novel framework to test for the missing data mechanism that replaces observed means with probabilities of missingness. The advantages of this framework are that a) it directly aligns with Rubin's (1976) theory of missingness, and b) it has been shown to perform better than Little's test under certain cases.