Semester of Gradation
Fall 2025
Degree Type
Dissertation
Degree Name
Data Science and Analytics
Department
School of Data Science and Analytics, College of Computing and Software Engineering
Committee Chair/First Advisor
Kevin Gittner
Second Advisor
Lauren Matheny
Third Advisor
Gita Taasoobshirazi
Fourth Advisor
Karen Nielsen
Abstract
Wearable devices have potential for sleep research due to their ability to collect individual level objective data. However, their use in low-resource settings is hindered by high missingness. This dissertation will determine the best method of managing missingness in longitudinal sleep data from wearable devices in low-resource settings.
Analytical methods were used to determine the likely mechanism of missingness. Then, four methods of missing data imputation, fully conditional specification - standard (FCS-S), joint modeling - multivariate normal (JM-MVN), k-nearest neighbor (KNN), and expectation maximization (EM) imputation, and maximum likelihood estimation (MLE) without imputation were compared. The imputed sample was modeled and the results compared using root mean square error, relative bias, and the Frobenius norm value of the V matrices. Finally, new data was imputed, then total and deep sleep duration were modeled with an additional variable incorporated into the model.
There is evidence that the missingness was missing at random. Comparing methods to handle missingness showed that EM imputation performed best. On new data, change in anxiety was incorporated into the models and was significant in predicting deep sleep duration.
When using wearable devices to collect sleep data in low-resource settings, missingness can be mitigated by using EM imputation without adding considerable bias to the model.
Included in
Applied Statistics Commons, Data Science Commons, Longitudinal Data Analysis and Time Series Commons, Other Public Health Commons
Comments
Committee also included Dr. Monica H. Swahn, Dean, Virginia Commonwealth University School of Public Health.