Abstract
Missing Not at Random (MNAR) data present challenges for the social sciences, especially when combined with Missing Completely at Random (MCAR) data for dichotomous test items. Missing data on a Grade 8 Science test for one school out of seven could not be excluded as the MNAR data were required for tracking learning progression onto the next grade. Multiple imputation (MI) was identified as a solution, and the missingness patterns were modeled with IBM Amos applying recursive structural equation modeling (SEM) for 358 cases. Rasch person measures were utilized as predictors. The final imputations were done in SPSS with logistic regression MI. Diagnostic checks of the imputations showed that the structure of the data had been maintained, and that differences between MNAR and non-MNAR missing data had been accounted for in the imputation process.
Highlights
Perfect data sets do not exist in the real world, and missing data are an authentic challenge facing social science analysts and researchers
To address the problem of Missing Not at Random (MNAR) data for the dichotomous anchor items, this study investigated methods to handle missing data when the mechanism for missingness is known
Anchor Item 4 has the largest percentage of missing data at 30% of values missing, with 14.68% of those being Missing Completely at Random (MCAR) data
Summary
Perfect data sets do not exist in the real world, and missing data are an authentic challenge facing social science analysts and researchers. Missing values can bias analyses, especially when high percentages are missing or there are patterns in the missingness (Allison, 2002; Osborne, 2013; Wang, Bartlett, & Ryan, 2017). The handling of missing data has been a topical issue in social sciences and methods dealing with missing values have grown exponentially (Enders, 2010; Li, Stuart, & Allison, 2015; Little & Rubin, 2002; Rubin, 1987; van Buuren, 2012). Data are MNAR when missingness on a variable is directly related to the outcome variable (e.g., science proficiency; Enders, 2010; Graham, 2012; Kim & Shao, 2014; van Buuren, 2012). Greater percentages of missing data may cause bias in analyses and should be investigated, and other options such as MI should be considered (Mallinckrodt, Lin, & Molenberghs, 2013; McPherson et al, 2015; Roberts, Sullivan, & Winchester, 2017)
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
