Robustness of Multiple Imputation under Missing at Random (MAR) Mechanism: A Simulation Study

<p>Missing data is an unavoidable issue in controlled clinical trials and public health research and practice. Presence of missing data and applying inappropriate methods of analysis generates biased estimates and reduces power of study. It is very important for investigators to use appropriate methods of analysis to deal with missing data in order to maintain internal (power of study) and external (generalization of sample results to larger population) validity of study. The focus of this dissertation is to compare different methods to deal with missing data in controlled clinical trials and public health research and practice. In addition, this dissertation also discusses that current approaches to deal with missing data might not produce valid inferences and may affect internal and external validity of results. Furthermore, emphasis is put on demonstrating how well multiple imputation works to deal with missing data under Missing at Random (MAR) mechanism with monotonic and non-monotonic missing data patterns for a range of percent missing under both normal and non-normal distributions. The results of this dissertation showed that multiple imputation is an efficient technique to obtain valid inferences compared to single imputation methods. In addition estimates obtained from multiple imputation also preserve the internal validity of study.</p>

Name thesis:degree_name

Doctor of Public Health in Biostatistics (Dr.P.H.)

Level thesis:degree_level

Dissertation (open access)

Discipline thesis:degree_discipline

Department of Biostatistics (COPH)

Year dc:date.available

2013

Author dc:creator

• Garg, Priyanka

Contributors dc:contributor

• Hani M. Samawi

• ETD
• Monotone missing data pattern
• Non-monotone missing data pattern
• Sensitivity
• Specificity
• Public Health

Repository record dc:identifier

https://digitalcommons.georgiasouthern.edu/etd/35

OAI identifier oai:identifier

oai:digitalcommons.georgiasouthern.edu:etd-1034