Crossing these two dimensions, four categories of behaviors can be derived (Table 22.1). Compliance versus defiance characterizes the bifurcation of behavior responses resulting from normative considerations (also see Tal-or et al., 2009). Compliance behaviors refer to instances where individuals bring their behaviors closer to the perceived expectations of the referent group. Such compliance can be in the form of initiation or reinforcement of attitudes/behaviors perceived to be prevalent in and/or sanctified by the referent group, for example, teenagers starting to smoke to be part of the peer group (Gunther et al., 2006). Compliance can also be changes from the other end of the behavioral spectrum (i.e., giving up behaviors or positions perceived to be “unacceptable” or “undesirable” to the referent group). Peripheral developmental town residents’ consideration of relocating away from a town that they believed was disliked by the majority others is a case that illustrates such behaviors (Tsfati & Cohen, 2003). Or the heightened sense of alienation and political inefficacy (Tsfati, 2007; Tsfati & Cohen, 2005) as a result of perceived normative expectations is also a subtle form of being co-opted by the “mainstream view” (Gunther et al., 2008, refer to such responses as “withdrawal” behaviors).
Table 22.1 Typology of Behavioral Outcomes Along Two Dimensions
Defiance behaviors, in contrast, refer to those that run counter to the perceived norms. These behaviors are not researched as much in extant IPI research. One example is from the Tsfati and Cohen’s (2005) study, which showed that Gaza settlers who perceived greater negative media influence on the public opinion about their group felt more resistant against evacuation and more justified to resort to violence. Chia and Wen’s (2010) finding that the more influence college male students perceived of media portrayals of ideal bodies on other friends, the less likely they were to indicate an intention of going to a gym regularly may also be suggestive of resistance against or “defiance” of perceived norms.
Under ecological considerations, coordination and rectification are two possible types of behavioral responses. Coordination reactions (see also Cohen & Tsfati, 2009) refer to adaptive behaviors based on calculations of how others’ possible behaviors may affect the chances to achieve their own goals. For example, Cohen and Tsfati (2009) showed that sophisticated voters switched their vote from their preferred, smaller party to a bigger one that appeared to be more favored by the media and therefore more influential on other voters (Study 2). In an experimental study conducted by Tal-Or, Cohen, Tsfati, and Gunther (2010), they showed that when respondents believed that the article they read on a sugar shortage had more influence on others, they indicated a stronger intention to rush to the stores to purchase sugar.
On the other hand, there are situations where perceived behavioral reactions by others are regarded to inflict harm or sustain some less-than-optimal conditions. Under such circumstances, individuals may be motivated to take up actions to fix the problems or deficiencies and improve their surroundings. Sun, Shen, and Pan (2008) use the term “rectification” to designate such behaviors. It includes, but is not confined to, restrictive or regulatory reactions toward media messages (labeled as “prevention behaviors,” Gunther et al., 2006, Tal-Or et al., 2009). Rectifying behaviors can also include other actions designed to redress situations deemed problematic. Lim and Golan (2011) showed that respondents were more likely to participate in “social media activism” behaviors, including posting comments or their own countering video online, when they believed that a YouTube parody video on Al Gore and global warming negatively influenced others. Such “corrective” behaviors (also see Rojas, 2010) are aimed at dispelling potential misperceptions or correcting biases that may be propagated by the media. Even in the context of media content with positive influences such as public service announcements, Sun et al. (2008) argued, “rectification” could take the form of promotional behaviors to further disseminate the messages and amplify their influence. All of these behaviors, restrictive, corrective, or promotional, share the goal of improving the less-than-desirable social conditions due to perceived excessive or insufficient media influences on others.
Unpacking “Presumed Influence”
“Presumed influence” (PI in short) denotes one’s subjective perceptions of the exerted or potential impact of the given media content on some referent others. Though seldom explicitly explicated, PI entails two connected aspects. The primary aspect is the subjective estimate of the extent or likelihood of message influence on the referent others. Operational measures are typically variants of this more general formula: “How much influence do you think [the media message] has on [the referent others]?” The second, ensuing aspect is the presumed collective responses from others (deemed either potentially possible or already actualized) resulting from such influence. For instance, Gunther and colleagues’ (2006) study employed the perceived smoking prevalence among peers as a proxy measure of presumed influence of smoking-related messages. The first aspect is an estimate of message influence in terms of its extent and magnitude, whereas the second aspect captures speculations of the substantive effects of the message on others, that is, how it may have made others (re)act in certain ways.
Both aspects are important to the construct validity of PI. Without including the first aspect, the notion of media as a source of perceived norms would be missing. Perceptions of peer smoking prevalence, for example, can be a result of direct observation or communication, instead of an inference based on presumed media influence. On the other hand, not measuring the second aspect, the presumed responses from others, can also be inadequate. Sheer perception of the magnitude of influence does not necessarily prompt behavioral responses. Rather, “how one responds to a message depends largely on what the message is thought to do to [others] (italics added)” (Tewksbury et al., 2004, p. 140). Jensen and Hurley’s (2005) study included what they called “presumed behavior” to explicitly capture the likelihood that others were thought to do something (such as talking about the issue or acting on the issue), and found that such presumed behaviors associated with different referent others had varied roles in motivating respondents to engage in behavioral responses.
Emphasizing the second aspect is also to highlight that inherent to PI is a media-referent relationship that is context-bound and referent-specific. The presumed message influence is a relational assessment, not a context-free evaluation of the message content or other message properties alone. This difference distinguishes PI from other related notions, such as perceived effectiveness/argument quality of a message, or perceived utility/gains of a message system/tool, the focal assessment of which is the properties of the evaluated object (though such evaluation inevitably evokes some referents in the mind of the respondents, Dillard & Sun, 2008).
Delimiting borderlines between PI and these other notions serves to maintain the theoretical identity of IPI. Take as an illustration Tsfati, Cohen, and Gunther’s (2011) study, where “presumed media influence” was measured in terms of how “published research featured in the mass media” is believed to give scholars more publicity, help their academic careers, get research funding, and so on (pp. 152–153). Strictly speaking, these items capture individuals’ beliefs about the benefit or utility of publicizing research on the media outlets, an assessment not bound to specific referents or contexts. As such, they can very well be measures of the “belief” component in the Theory of Attitude (Attitude= Σbiei; Fishbein & Ajzen, 1975) in predicting attitude toward the behavior in question. The study could be regarded as a test of the direct linkage between beliefs (about media as a tool to advance some relevant goals) and attitudes (toward using media as such a tool), instead of an indirect route between media messages and behaviors via reasoning about referent others as postulated in IPI.
Self, Others, and Messages
Self and Referent Others
Though the perceived influence on self is no longer a critical element in the theoretical formulation of IPI (Gunther & Storey, 2003), the construal of self-other
relationship is nonetheless intrinsic to the perception-behavior process. The self-other relationship in IPI studies can be broadly put in two categories. One type is nested, where self (the respondent) is part of the referent group on the dimension evaluated, such as friends or other college students (Gunther et al., 2006), or other voters in the country (Cohen & Tsfati, 2009). The other type of self-other relationship is juxtaposed, where the referent others belong to an out-group on the characteristics defined by the context of the study. Such in-group and out-group distinction can be based on demographic characteristics, such as gender (e.g., female respondents vs. “other men in general,” Park, 2005), race (e.g., Israeli Arabs vs. Israeli Jews, Tsfati, 2007), or party affiliation (e.g., Democrats, Independents, and Republicans, Hoffner & Rehkoff, 2011). Groups can be sociologically or institutionally defined as occupants of different positions in a specific social system (i.e., physicians vs. clients in the DTC advertising context, Huh & Langteau, 2007; or congressmen vs. the public, Cohen et al., 2008). Perceptions of group boundaries can also be created through media portrayal, such as the “featured group” of media reports on some issues vs. the rest (for example, the Gaza settlers vs. other audience members in Tsfati & Cohen, 2005).
Do all the referent others weigh the same on one’s decision-making? The existing evidence suggests not. Presumed influence of anti-smoking messages on distant peers was not related to one’s own smoking attitudes or intention, but that on close friends was (Paek, 2009; Paek & Gunther, 2007). The indirect effect of advertising on materialism was mediated by teenagers’ perception regarding their friends, but not that regarding their parents (Chia, 2010). These findings have shown that “Not all others are equal.” The literature to date, though, does not yet offer compelling arguments as to why they are not equal. Although the relevance of the referent group seems to be an easy explanation to evoke, the ad hoc usage of such an explanation borders on tautology if self-other relationships are not theoretically explicated a priori and examined as an empirical question on its own. A combination of individual, interpersonal, and contextual factors may be responsible for differential judgment processes involving different referent others.
Message
Message tends to be the “backgrounded” element in IPI research. In most IPI studies, researchers usually provided the respondents with a general description of media messages in a broad topic area, such as “news media coverage of the elections” (Cohen & Tsfati, 2009), “anti-smoking messages on TV” (and magazines, billboards, etc.; Gunther et al., 2006), or “media content that includes talk about sex, sexual behavior, and sexual relationship” (Chia & Lee, 2008). Respondents were asked to recall their own exposure to these messages before estimating others’ exposure to such messages and the presumed influence on others. Message characteristics and individuals’ own perceptions and interpretations of such messages are rarely measured.
Such operational practice can marshal some defense. That is, when the goal of the study is to explain the formation, reinforcement, or change in one’s attitudes or behavior in a given message environment resulting from a cumulative process involving constant exposure to such messages, a vague, broad measure has face validity in terms of capturing the immersion of the individual in the message environment. In a theoretical light, however, such self-report exposure measures without attention to message characteristics are problematic in at least two ways. First, using exposure as the antecedent factor presupposes that individuals use the “exposure is effect” heuristic to make judgments about influence on others. This assumption does not necessarily hold. Lim and Golan’s (2011) experimental study just demonstrated the opposite: The perceived likelihood of exposure, manipulated as high versus low numbers of views on YouTube, had no significant effect on presumed influence on others, whereas perceived persuasive intent of the message (manipulated through source intent) did. Broken linkages between exposure variables and presumed influence were also shown in a few other studies, especially when the referent others were regarded as distant (e.g., distal peers in Paek & Gunther, 2007; Paek, 2009) or different (i.e., parents, Chia, 2010; male others, Park, 2005) from self.
Second, without examining conceptual characteristics of messages, the theoretical processes between message construal and judgment-making remain opaque. The problem of the lack of specific message explication looms large when unexpected results turn up and require further explanations. As Paek et al. (2011) lamented, “our global measure of exposure … does not allow further explication of the reasons for the unintended association” (referring to the positive association between exposure to anti-smoking messages and smoking outcomes; p. 141). Though ad hoc explanations could be summoned up, they remain uncompelling speculations.
Directions for Future Research
* * *
Based on the review of empirical evidence and the conceptual analysis, I will make three critical observations of problems or challenges that face IPI research.
(I) Despite a sizable body of research studies on the process of IPI, its empirical credence is not yet quite established due to inconsistencies in extant findings and a general lack of causal evidence.
Inconsistencies in Empirical Findings
Though most studies show satisfying model fit indices, inconsistencies in specific findings should not be overlooked. For example, in the context of anti-smoking messages, the direct effect of self-exposure to anti-smoking messages on smoking susceptibility was shown to be nonsignificant in Gunther et al. (2006), but counterintuitively, positive in other analyses (Paek & Gunther, 2007; Paek et al., 2011). In Paek et al. (2011), the overall indirect effect from anti-smoking message exposure to smoking susceptibility at Time 1 was negative (−.02, p < .05), but positive at Time 2 (.04, n.s.). Such results, both internally inconsistent and in contradiction with some external literature (i.e., meta-analytic findings on the effectiveness of anti-smoking campaigns, Sussman, Sun, & Dent, 2006), call for more investigations in this context, especially as the findings reported in these three studies were all based on the same data source.
Some other inconsistencies include Gunther and Storey (2003), where the predicted process received support from the self-report data, but not when actual measures of observed interactions were used as an outcome variable, or Park (2005), which showed positive indirect effect but negative direct effect of presumed influence on other women on one’s desire to be thin. Though the authors made ad hoc explanations for these inconsistencies, more empirical investigations are needed in future research to replicate or explain such findings.
The Lack of Causal Evidence
As extant studies rely heavily on cross-sectional self-report survey data (Tal-Or et al., 2009), a prominent concern with IPI research is that the evidence does not translate to causal interpretations. Studies that use SEM analysis seldom test out alternative models. When reversed causal links did get tested, the evidence tended to be equally favorable for the alternative models. For example, reversing the causal path between self-exposure and smoking attitudes/susceptibility produced a model fit as good as (Paek & Gunther, 2007), or even a slightly better fit than (Gunther et al., 2006), the original model. Chia and her colleagues’ research on sexual norm perceptions also showed equivalent support for the alternative explanation, the projection effect.
Simply acknowledging the lack of causal evidence as a weakness in discussion sections, which most research papers do, is not enough. How to parse out causal processes poses methodological as well as theoretical challenges that should be taken up by future IPI research. Randomized experimental studies and longitudinal studies, as effective ways to establish causal evidence, should be conducted more often. So far, only two experimental studies (Lim & Golan, 2011; Tal-Or et al., 2010) and one longitudinal study (Paek et al., 2011) bespeak such efforts.
A Cautionary Note About SEM
As SEM has been a popular technique used in IPI studies, a note of caution should be made emphatic. One common misuse of SEM, as SEM scholars hav
e alerted us to, is to prioritize “adjudging” fit over theory-testing (Hayduk, Cummings, Boadu, Pazderka-Robinson, & Boulianne, 2007). A symptom of such misuse in IPI studies is that variables are sometimes added to or removed from the originally posited IPI process without theoretical justifications. For example, in Park (2005) presumed influence on self was inserted between presumed influence on others and attitudinal outcome, a modification of the original IPI model without sufficient theoretical justification. In Paek et al. (2011), peer exposure was removed from the model solely based on model trimming procedures without any discussion of the theoretical reasons and implications.
Moreover, another problem of overemphasis on model fitness indices (FI) is that the conventionally used FIs have less bona fides than usually credited with. Saris, Satorra, and Van Der Veld (2009) showed with simple examples that the conventional FIs could lead “substantively relevant misspecification” (e.g., imposing wrong restrictions on certain parameters) to be retained and “substantively irrelevant misspecification” (e.g., good enough for practical purposes though not exactly the same as the “true” model) to be rejected. FIs are also unable to detect common perils to SEM analysis (such as common method variance and simultaneity, to which IPI studies are particularly vulnerable) that can inconsistently bias path coefficients and invalidate causal inferences (Antonakis, Bendahan, Jacquart, & Lalive, 2010; Cole, Ciesla, & Steiger, 2007). IPI scholars should use SEM with more discretion, prioritize theory-testing, and interpret the results with great care.
The SAGE Handbook of Persuasion Page 77
