Working with Missing Data in Higher Education Research: A Primer and Real-World Example

Abstract

within six academic years. Whereas 45 percent of the White students successfully transferred, only 31 percent of the underrepresented minorities did so. Among the White students, a high school grade-point average above 3.5, not delaying enrollment into college, being classified as a dependent student, and academic integration increased the odds of transfer, while attending a community college with higher percentages of minority students and students receiving financial aid decreased the probability of transfer. Among the underrepresented minority students, being a continuing-generation student, taking rigorous math courses in high school, and not enrolling in certificate or vocational courses increased the probability of transfer. These differences suggest the importance of disaggregating the data for different racial/ethnic groups when analyzing the likelihood of transfer. (84 ref)-Educational Leadership and Policy Studies,

Military-Connected Student Academic Success at 4-Year Institutions: A Multi-Institution Study

“…In this analysis, missing data were managed using multiple imputations to maintain the statistical power of the data. Imputing data is a common practice in social science research to utilize complete data methods of analysis and in survey research, analysis of data without imputation requires the use of listwise deletion of cases, which could lead to decreased statistical power, pairwise deletion where parameter estimates may be biased in multiple directions and magnitudes (Cox, McIntosh, Reason, & Terenzini, 2014).…”

Replacing missing values using trustworthy data values from web data sources

“…Due to this, missing values in MNAR cannot be imputed by using other observed values [3]. Therefore, advance statistical knowledge is important in order to analyze data with MNAR [18]. In summary, it is important to make a clear distinction between MCAR, MAR and MNAR occurrence in the dataset as it will determined the best approach to deal with missing values.…”

“…The list-wise deletion is simple. Yet, as the sample size decrease, the statistical power of the analysis will drop as well which then arise difficulties to detect small effect or relationship between variables [18]. It became worst in a case of MCAR data type, as it increase the standard errors and decrease the level of significance due to a smaller sample size [17].…”

“…In this approach, the mean value of the observed variables in dataset is computed and used to replace missing values. This approach is easy to use and more assertive compared to the deletion approach as it estimates the missing values based on the non-missing values without reducing the sample size and disturbing the mean variable of dataset [18]. However, using constant mean value to replace missing values in dataset distort the statistical distribution of data and will further risk the result of data analysis [22].…”

“…Besides, using the constant value of mean to replace missing values altered the randomness of data. Escaping the data randomness led to invalid statistical inference [18,23].…”