When measuring risk attitudes for use in policy analysis it is important to recognize that the context may matter. We propose a method that goes beyond simple context framing using text and pictures and instead relies on Virtual Reality simulations. Such simulations have been shown to improve the accuracy of perceptions in experimental context. We ask whether elicited risk preferences differ when using simulations or stylized lottery tasks, and compare students (a low cost convenient subject pool) to field subjects. While we see initial differences across our 2x2 treatment design, we find that they converge with repetition.