Index Of Local Sensitivity To Nonignorability For Intensive Longitudinal Data

When analyzing intensive longitudinal data, it is often assumed the missingness is ignorable. Since this assumption is unverifiable, it is crucial to perform sensitivity analysis to assess the potential impact of nonignorability. However considering the complex and non-monotone missing patterns and the large volume of data, a sensitivity analysis that directly fits different nonignorable models can be challenging to perform because of the high dimensional integrations in the likelihood functions from the alternative models. Linear index of local sensitivity to nonignorability method has been developed to avoid fitting complicated nonignorable models and simplify the calculation for different data types and statistical models when missingness occurs in outcome only. Also, this method has been extended for cross-sectional data by introducing the nonlinear index to capture the U-shape impact of nonignorability caused by concurrent missingness in outcome and covariates. In this dissertation, we further extend the application of this nonlinear index of local sensitivity to nonignorability (NISNI) method to longitudinal linear mixed effects models. With selection modeling framework and the non-monotone missingness patterns modeled using transitional multinomial models, we develop formulas and closed-form expressions for both linear and nonlinear indexes when outcome missing only, outcome and one covariate missing simultaneously, and outcome and multiple covariates missing simultaneously. We evaluate the performance of this extended method using simulated data and real intensive longitudinal datasets. The results indicate that our method can maintain the computational simplicity and capture the impact of nonignorability accurately under different situations.