Documenting & imputing missing values in a longitudinal survey of students’ personal attributes

HA Statistics

Letham, Collette Alexis ; McColl, John H.

University:University of Glasgow

Department:School of Mathematics and Statistics

关键词: Missing Data, Personal Attributes, Multiple Imputation, Longitudinal;   

Others  :  http://theses.gla.ac.uk/4545/1/2012lethamMsc.pdf

来源: University of Glasgow

【 摘 要 】

The University of Glasgow is currently engaged in a programme of action designed to reduce the proportion of students who withdraw from the university during their first year. Student retention is a cause for concern for higher education institutions in terms of reputation and funding. Previously, researchers have suggested that early withdrawal from university is linked to personal attributes. A questionnaire to explore this was designed consisting of 5 standard psychometric scales measuring respectively mindset, self efficacy, self esteem, resilience and hope. All new entrants to the University of Glasgow in September/October 2009 were invited to take part in a study of these personal attributes.1,098 (20%) new undergraduates and 407 (10%) new postgraduates agreed, and filled in the questionnaire while pre-registering on the university’s computerized registration system (WebSURF). At random, half of the students who took part at baseline were invited to complete the same survey again at the end of teaching in Semester 1 and the other half at the end of teaching in Semester 2. The results obtained on the psychometric scales were linked to routinely-collected data about the same students’ background and their continuation and progression at the end of first year. The aim was to investigate the influence of personal attributes, either on their own or in conjunction with demographic variables, on the continuation and progression of students.A common problem encountered in this study is that data were missing. It is important that the reasons why data are missing are taken into account and that missing data is dealt with, as far as possible, in a way that does not lead to biased results and invalid inferences. For this reason, it was decided not to rely on the results of a complete case analysis, but to use multiple imputation to fill in the missing values and then repeat the analysis using the completed datasets as well.Chapter 2 provides a review of the psychometric scales used in this study. The characteristics of missing data and methods to handle missing data are described. Also in Chapter 2, the theory of various statistical methods used in this analysis is illustrated in detail.In Chapter 3 the completeness of the questionnaire dataset is documented by examining the rates of non-response. The completeness of the questionnaire is also examined to establish if any of the demographic variables such as Sex, Age, Domicile, Faculty and Socio-Economic Class are associated with it. A higher proportion of older than younger undergraduate students completed the questionnaire fully, and more students in a professional faculty than students in a non-professional faculty completed it.The complete case analysis to explore the effect of demographic variables and personal attributes on the outcome of first year for undergraduate students is detailed in Chapter 4. For whether or not first year students continued at the University of Glasgow after first year neither the baseline personal attribute scores nor the difference in personal attribute scores were found to be statistically significant. The change in self esteem score in the course of first was seen to be a significant predictor of whether or not first year students progressed at the University of Glasgow after first year.Chapter 5 focuses on various ways in which that imputation was applied to fill in missing values of the baseline personal attribute scores and the difference in personal attribute scores. However, even after imputing the personal attribute data, neither the baseline personal attribute scores nor the difference in personal attribute scores were found to be statistically significant predictors of Continuation or Progression.Chapter 6 includes a summary of the results of this thesis and discusses the limitations and further work that could be implemented.