Context Dependent Utility: Modeling Decision Behavior Across Contexts

Context Dependent Utility: Modeling Decision Behavior Across Contexts Jonathan Ito (ito@ict.usc.edu) Stacy Marsella (marsella@ict.usc.edu) Institute for Creative Technologies, University of Southern California 12015 Waterfront Drive Playa Vista, CA 90094-2536 USA Abstract One significant challenge in creating accurate models of hu- man decision behavior is accounting for the effect of con- text. Research shows that seemingly minor changes in the presentation of a decision can lead to drastic shifts in be- havior; phenomena collectively referred to as framing ef- fects. Previous work has developed Context Dependent Util- ity (CDU), a framework integrating Appraisal Theory with decision-theoretic principles. This work extends existing re- search by presenting a study exploring the behavioral predic- tions offered by CDU regarding the multidimensional effect of context on decision behavior. The present study finds support for the predictions of CDU re- garding the impact of context on decisions: 1) as perceptions of pleasantness increase, decision behavior tends towards risk- aversion; 2) as perceptions of goal-congruence increase, deci- sion behavior tends towards risk-aversion; 3) as perceptions of controllability increase, i.e., perceptions that outcomes would have been primarily caused by the decision maker, behavior tends towards risk-seeking. Keywords: Decision; Appraisal; Context; Framing; Utility; Introduction Descriptive models of human decision behavior seek to ac- curately describe and predict the decisions people actually make. Creating these models is vital for advancing a more complete understanding of the human decision process and requires addressing the factors that systematically bias the perception and evaluation of decisions. One significant challenge in creating accurate models of human behavior is accounting for the effect of context on de- cision behavior. Research has shown that seemingly minor changes in the presentation, or framing, of a decision prob- lem can lead to drastic shifts in behavior; phenomena collec- tively referred to as framing effects. In a seminal study, now referred to as the Asian Disease Study, Tversky and Kahne- man (1981) showed that when outcomes were described, or framed, as gains participants tended to be risk-averse; how- ever, when the same outcomes were framed as losses par- ticipants tended to be risk-seeking. Subsequent studies in- volving domains as diverse as financial planning (Schoorman, Mayer, Douglas, & Hetrick, 1994), Acquired Immune Defi- ciency Syndrome (AIDS) (Levin & Chapman, 1990), Breast Self Examinations (Meyerowitz & Chaiken, 1987), taxpayer compliance (Liu, Xia, & Xu, 2011), and judgments of website quality (Hartmann, De Angeli, & Sutcliffe, 2008) have also demonstrated framing effects to varying degrees. In addi- tion to gain-loss framing, framing can also involve the role of the decision maker (Wagenaar, Keren, & Lichtenstein, 1988), the salience of outcomes (Van Schie & Van Der Pligt, 1995), decision domain (Vartanian, Mandel, & Duncan, 2011), and perceived need (Mishra & Fiddick, 2012). Despite the highly multidimensional nature of context, the prevalence of framing effects in numerous domains, and the profound impact they can have on the decision process, very few decision models explicitly address the multidimensional impact of context on decisions. Existing decision-theoretic approaches which do address framing and context are gen- erally limited by a narrow, one-dimensional view of con- text. For instance, Prospect Theory (Kahneman & Tversky, 1979) and Cumulative Prospect Theory (Tversky & Kahne- man, 1992) model the effect of context only to the extent that it applies to outcomes perceived as either gains or losses. Therefore, to address the multidimensional effect of context on decision behavior, previous work has developed Context Dependent Utility (CDU), a framework which seeks to ex- plicitly model the multidimensional impact of context on de- cision behavior through the integration of Appraisal Theory and decision-theoretic models (Ito & Marsella, 2011). This work extends previous research by presenting an experimen- tal study exploring the behavioral predictions offered by CDU regarding the multidimensional effect of context on decision behavior. In particular, the results support the behavioral pre- dictions of CDU and suggest that it can dramatically improve the modeling of human decision behavior across distinct con- texts. Context Dependent Utility In previous work, Context Dependent Utility (CDU) was de- veloped to explicitly model the multidimensional impact of context on decision behavior (Ito & Marsella, 2011). The CDU process consists of two primary components: the com- putational appraisal of the decision situation and an evalu- ation function aggregating the appraisal information into a real-valued utility. Appraisal Theory (Lazarus, 1991) is a psychological the- ory which addresses the process by which emotions arise given the subjective evaluation and interpretation of a situa- tion. Because appraisal theory provides a well-defined frame- work for the interpretation of features of a situation in terms of their significance, we argue that it provides the means to identify, encode, and integrate contextual information into the decision process. Appraisal as implemented by CDU con- sists of three distinct evaluations: pleasantness, goal congru- ence, and control. Each appraisal is defined over individual outcomes as a function of diminishing sensitivity evaluated with respect to some reference point. This follows from the