Mapping Attention, Performance, and Generalized Event Rates: A Multi-Task Application

Multitasking is an area of great research and offers many opportunities to ascertain how individuals divide attention. Variables such as performance and attention are typically assessed; however these can be modulated by the nature and status of the assigned task. The status of a given task can be identified in many ways and is often directly output in experimental contexts; in real-world situations, however, this is rarely the case. One persistent feature of a task is the time of observable events, which can be used to evaluate performance relative to task events. While different methods of analyzing the timing of task events exist, hazard functions are a well-defined way of doing so. We present here the use of hazard functions, both to understand behavior in context of task events and to predict future performance. Previous work is leveraged to identify focus of attention (FoA) using neural data, alongside measures of multitasking efficiency derived from performance data, to which is added the cumulative hazard value determined from task events. Qualitative and quantitative examinations are performed. Traditional models show statistically significant improvements with the addition of this variable; its use to compare values across different contexts (performance and FoA) shows statistically significant differences between those contexts; finally, machine learning models using hazard values as an input feature show good prediction of future performance.