QuantDev Brown Bag Seminar Series (In-Person)

Google map

Robustness Checks in Modeling Intensive Longitudinal Data: Sensitivity Analysis for Missing Data.

Abstract: 

In real-world empirical research, handling missing data presents a challenge due to the absence of a known ground truth, as typically available in simulation studies. Sensitivity analysis offers a method to explore the robustness of results across different strategies for addressing this missingness.

Using the TIME study's intensive longitudinal data, which tracked participants' affect and physical activity daily over a year, I employed a multilevel VAR (vector autoregressive) model. This model elucidates the reciprocal influences between daily affect and physical activity while accounting for the nested structure of the data. Given the intermediate missing data in both variables, I adopted various joint models. Each model was added to the primary VAR model based on differing assumptions about the reasons for missingness, and their influence on the primary model's results was subsequently observed.

The findings revealed that the autoregressive parameters for affect and physical activity remained stable across the different missing data assumptions. However, cross-regressions between affect and physical activity were less consistent, with outcomes varying based on the selected joint missing data model. This highlights the importance of sensitivity analysis when dealing with missing data in longitudinal studies.