761 Advances in Consumer Research Volume 36, © 2009 1 This article is based on the first author’s dissertation research and is funded by a grant provided by the Social Sciences and Humani- ties Research Council of Canada. Authors are grateful for the comments of Joshua Klayman, Jackie Gnepp, Ashesh Mukherjee, Ulf Bockenholt, and Renaud Legoux on earlier versions of this article. When Will People Tell You Something You Do Not Know? Lei Huang, Dalhousie University, Canada Sema Barlas, McGill University, Canada 1 EXTENDED ABSTRACT Word-of-mouth (WOM) is considered to be an important source of information for individual decisions based on a widely held assumption that other people can bring diverse perspectives to the individual’s decision process (Heath and Gonzalez 1995). For any given decision problem, individuals may hold unique knowl- edge that may be acquired through personal experiences as well as common knowledge that may be obtained from mass media. According to Grecian norms of conversations (Grice 1975) and information theory (Shannon 1964), people should exchange unique information in order to maximize the informational content of available information for decisions. Also, at the deliberate level, people agree to these prescriptions and ascertain that it is the unique information that should be exchanged in interpersonal communica- tions. However, extant empirical research demonstrated that “what people talk about” in actual conversations is not always what is the most informative for decisions. For instance, overwhelming evi- dence has been accumulated in the literature demonstrating that conversations are dominated by the information that members hold in common before discussions, the information that is consistent with members’ pre-existing preferences (Stasser and Titus 1985, 1987), and the information that is extreme or negative. In addition, Huang and Barlas (2006) demonstrated that people exchange emotional and recreational information more often than cognitive and utilitarian information in WOM in order to establish social connections. This paper investigates the exchange of unique information in WOM depending on the nature of this information (e.g., whether it is emotional or cognitive). Note that bias for emotional information in WOM reported by Huang and Barlas (2006) might be observed because emotional information (e.g., consumption experiences) is often uniquely held by participants. If this is the case, one would expect that the bias for emotional information should diminish when emotional information is commonly held by WOM partici- pants. However, causal observation suggests that bias for emotional information is amplified when it is shared knowledge among individuals. For instance, most political conversations revolve around emotional topics (e.g., the gay marriage issue in the 2004 U.S. elections). Candidates’ positions on issues that spur emotions among the voters are widely communicated by the mass media, and therefore, are likely to be common information among voters, whereas cognitive information (e.g., foreign aid) is likely to be the unique knowledge held by individual voters since it is often not publicized. Also, much literature suggests that people try to validate the messengers as well as the messages by talking about the commonly shared information (Stasser and Titus 1985) in order to validate that their partners are similar to them with respect to interests. Indeed, talking about common topics establishes social connections among participants in the same way sharing of emo- tional information. Thus, this paper aims to tease out the effects of information types (emotional versus cognitive) and the distribution of information among communicators (common versus unique) on the sampling of information for conversations and choices. We hypothesize that common information is exchanged more often than unique information in conversations; however, people demon- strate strong preference for common emotional information, but exchange unique information socially when it is emotional. We used a political decision-making task where participants chose between two candidates for the Student President position after discussing candidates’ profiles. Participants exchanged infor- mation with two other individuals before deciding to cast their vote. 120 participants (64 females) were recruited from undergraduate business courses as groups working on a class project together. Some groups were self-selected friends and others were formed by the course instructor choosing randomly among acquaintances. The 40 three-person groups were equally divided into two WOM conditions: common-cognitive-unique-emotional (briefed as com- mon-cognitive) or common-emotional-unique-cognitive (briefed as common-emotional) condition. A control condition of 99 volun- teered participants performed the same tasks as did the WOM participants except for exchanging information with others. In the common-cognitive condition, the profile descriptions of electoral candidates included the same cognitive information across members, but emotional information was unique to the profile each member received. In contrast, identical emotional information was included in the profiles of all members in the common-emotional condition while each member received unique cognitive informa- tion. All participants were explicitly informed that their informa- tion was not complete before the discussion. All conversations were audio-taped for further content analysis. Our results support the widely accepted conclusion in the literature that common information (mean=0.58) is exchanged more often than unique information (mean=0.42) in conversations. However, our results also suggest that unique emotional informa- tion (mean=0.61) that would facilitate interpersonal connections is, indeed, shared with others more often than common cognitive information (mean=0.39), but still less than common emotional information (mean=0.77). Thus one major contribution of this paper is the finding that emotional information enjoys a sampling advantage in WOM; this bias is inflated when emotional informa- tion is also commonly held by the members prior to conversations (e.g., as in online user groups). That is, the nature of the information plays a more significant role in conversations than the distribution of information among individuals. Additionally, these biases in information sampling lead to similar biases in the influence of information on decisions. There was a strong preference for the candidate who was favorable on the basis of common emotional information. By contrast, preference for the candidate who was favorable on the basis of cognitive information was not strong because participants exchanged the unique emotional information in conversations. The biases reported above were more pronounced when members in the group were similar with respect to the importance of emotional and cognitive attributes in their decisions. However, neither gender nor social ties (friends versus acquain- tance) had significant impact. Most existing research in the political marketing focuses on the impact of information valence on voting behavior. Differing from previous research, results from our study suggest that the nature of the information has considerable impact on conversations about candidates as well as on the votes and the size of this impact