Bounding causal effects with an unknown mixture of informative and non-informative missingness.

Facial expressions represent one of the most fundamental non-verbal communication channels in human interaction, serving as windows into our emotional states. Although the universality of expressions between cultures remains an area of scholarly debates [], [], [], robust evidence indicates that facial expressions function as reliable biomarkers of emotional and psychological states between various age groups []. These natural indicators of affective states offer a uniquely accessible pathway for monitoring mental well-being, potentially enabling early detection of conditions such as depression and anxiety disorder when integrated into appropriate sensing technologies. This capability becomes particularly valuable for vulnerable populations, such as older adults in long-term care facilities, who may experience neglect or inadequate mental health monitoring. The growing global mental health crisis, exacerbated by the widespread social, economic, and psychological impacts of the COVID-19 pandemic [], underscores the urgent need for unobtrusive, continuous affective monitoring solutions. VR and AR systems, typically deployed as head-mounted devices, have emerged as promising platforms for mental health interventions by providing immersive therapeutic environments that can enhance traditional behavioral therapy approaches when used alongside tele-health consulting [], []. These wearable platforms present a unique opportunity for integrating affective sensing technologies that can continuously monitor users’ emotional states through facial expression analysis. Beyond therapeutic applications, they also hold potential for assistive devices that can transform the lives of individuals with severe disabilities, including tetraplegia [], []. By embedding emotion recognition capabilities within wearable VR and AR systems, we can enable both passive monitoring of mental well-being as well as create novel emotion-responsive interfaces that adapt to users’ affective states, offering a more natural and empathetic computing experience []. Computer vision has historically dominated facial expression recognition (FER) approaches, evolving from early heuristic methods to contemporary deep learning systems. Heuristic pattern recognition algorithms for facial expression recognition typically employ geometric feature-based approaches that track displacement of facial landmarks and appearance-based methods like local binary patterns and Gabor wavelets that capture texture information correlated with specific expressions. Many of these approaches operate within the framework of the facial action coding system, which decomposes expressions into action units corresponding to specific muscle movements. More recently, vision systems have largely transitioned to deep neural networks [] and attention-aware transformer models [], leveraging large annotated datasets to achieve state-of-the-art performance. Deep learning approaches offer significant accuracy gains over heuristic methods but also present significant limitations in wearable contexts, particularly for VR/AR applications. These systems require unobstructed views of the face, are sensitive to ambient lighting conditions, and raise substantial privacy concerns that may limit their acceptability across diverse user populations and regulatory environments.