Chapter 8 Errors of observation: Missing data
8.1 How much and what data are there
Sensor data often accumulate missing data that can decrease the inferential value of WAS data. We focus on the extent of missingness across different data types (geolocation, physical activity, smartphone use data, etc.) and discuss patterns of missing data in WAS studies.
8.2 Types of missing data
We discuss mechanisms responsible for missingness for different types of WAS data, including technical errors, device errors and differences between types of sensors, and user behavior.
8.3 How design decisions affect missing data
We focus on study design decisions to help minimize missing data as well as on device characteristics, such as brand and model, that can affect missing data (e.g., app shuts down in the background by an operating system to save battery life). We provide practical advice on how to reduce missing data when conceptualizing a study.
8.4 Imputation of missing data
Imputation of missing data for WAS data can be more complicated than for traditional social science and behavioral data sources due to the dynamic nature of the data. We discuss emerging imputation methodologies and provide practical exercises that the readers can apply in practice.