It is widely accepted that the prediction of building energy performance is strongly related to the occupancy parameters. Currently, existing buildings and laboratories are the main sources for collecting occupancy related data. However, using such data for predicting the energy consumption of future buildings can create a considerable amount of uncertainties. Recent studies show that Immersive Virtual Environments (IVEs) have the potential to generate design and context sensitive occupant-related data. However, extended observations (longitudinal data covering relevant spatial and temporal events) which are necessary for developing quantitative predictive models are impractical using conventional IVEs. To that end, the authors propose a Spatial-Temporal Event-Driven (STED) modeling approach to enable IVEs for longitudinal studies. Using a single occupant office as case study, two sets of occupancy and lighting data, from IVEs and a comparable physical environment (in-situ), were collected. The occupancy/lighting data was organized in form of state transitions at six events (i.e., arrival in the morning, leaving for and returning from a short leave, leaving for and returning from a long leave, and leaving at the end of a day). It was hypothesized that the probabilities of the occupancy/lighting state transitions in a given event across the two experimental environments (i.e. IVE vs. in-situ) are not statistically different. Results revealed similar patterns at four of the six events (α = 0.05), except at the short leave events. Thereby, STED modeling enabled the potential viability of IVEs for extended observations and generating data to support predictive models. Clearly, more basic research is needed to make data collection using IVEs more effective including a better understanding of virtual cue design and participant's physiological and psychological conditions at the time of experiments.