Understanding and Correcting Inaccurate Calorie Estimations on Amazon Mechanical Turk

Lillio Mok, Brenna Li, Stephen Gou, Joseph Jay Williams

Abstract: Current research on technology for fitness is often focused on tracking and encouraging healthy lifestyles. In contrast, we adopt an approach based on improving consumer knowledge of food energy. An interactive survey was distributed on Amazon Mechanical Turk to assess how well crowdworkers can judge the calories in a series of foods. Our subjects yielded results comparable to traditional participants, exhibiting well-known phenomena such as underestimating the energy contained in foods perceived to be healthy. Several techniques from the online education literature, such as prompts for reflection, were also investigated for their efficacy at increasing estimation accuracy. Although calories were more accurately judged after applying these methods on aggregate, the effects of individual techniques on our participants were inconclusive. A more thorough investigation is thus needed into effective educational methods for correcting calorie estimations on the Web.