Abstract
This study proposes the third-person perception (TPP) can be viewed as a type of comparative social judgment in the domain of media influence, in that it is a function of assimilation, contrast, and anchoring mechanisms in the process of social comparison. The derived hypotheses were tested with web-based experimental data (N = 511). Results showed some evidence that TPP was a function of assimilation and contrast effects. There was also evidence that there were anchoring effects, and such effects tended to emerge when self was the anchor. Implication and directions for future research on TPP were discussed.
The third-person perception (TPP) research has proliferated following Davison’s (1983) seminal work. Subsequent research over the past three decades indicates that the TPP is a robust phenomenon (Davison, 1996; Perloff, 2002; Sun, Pan, & Shen, 2008). Scholars have investigated the perceptual processes that are believed to undergird the TPP, uncovering a host of relevant factors. These factors include media effect schemas (e.g., D. M. McLeod, Detenber, & Eveland, 2001; D. M. McLeod, Eveland, & Nathanson, 1997; Price, Huang, & Tewksbury, 1997), undesirability of presumed influence (e.g., Gunther & Storey, 2003), characteristics of the other (e.g., geographic distance, psychological or relational distance, generality, and heterogeneity; for example, Brosius, & Engel, 1996; Cohen, Mutz, Price, & Gunther, 1988; Duck & Mullin, 1995; Tewksbury, 2002), and characteristics of self (e.g., self-esteem, paternalistic orientation, and self-protection or self-enhancement motivations; for example, Brosius & Engel, 1996; David & Johnson, 1998; David, Morrison, Johnson, & Ross, 2002; D. M. McLeod et al., 2001; White & Dillon, 2000). Several explanations can be found in the literature, among which, Perloff declared that (the motivational explanation) “self-enhancement has the most evidence in its behalf” (p. 495). In their meta-analysis, Sun, Pan, and Shen (2008) found strong evidence for this claim: Desirability of presumed media influence and perceived vulnerability of others significantly predicted magnitude and direction of the perceptual gap.
On the other hand, there have been efforts to provide such accounts in cognitive terms. Indeed, cognitive factors have been identified as significant predictors of TPP, such as presumed likelihood and/or frequencies of exposure to the media messages in question (e.g., Eveland, Nathanson, Detenber, & McLeod, 1999), social distances between self and others (e.g., Cohen et al., 1988; Eveland et al., 1999), and presumed self-other differences arising from social categorization (e.g., Duck, Hogg, & Terry, 1995; Reid & Hogg, 2005). Sun, Pan, and Shen’s (2008) meta-analysis also provided evidence supporting this view: Perceived similarity and others being more likely recipients of media content also predicted TPP. Although without any direct evidence, Gunther (1991) proposed that attribution error could be a cognitive explanation for TPP as well: It suggests that individuals tend to attribute media influence on self to what they regard as “the objective reality,” or external conditions, but attribute media effects on others to their personal characteristics, or internal dispositions.
TPP implies that individuals are engaged in estimating media effects on multiple referents, which can be considered as comparative social judgments on media effects, that is, comparing self versus others in the domain of media effects (David, Liu, & Myser, 2004). The perceptual gap in media effects resulting from such comparative social judgments is inevitably “relational,” “motivational,” and “selective” (Dunning, 2000, p. 374), that is, confounded with motivation. For example, the attribution explanation has a motivational element: Being influenced by the media has undesirable connotations. Attributing media effects on self to situational factors and effects on others to dispositional factors tends to portray a positive self-image. Although the exposure explanation is cognitive in nature, the estimate of exposure to media content tends to be motivationally driven (e.g., Peiser & Peter, 2000). The categorization explanation (e.g., Duck et al., 1995; Reid & Hogg, 2005) has comparative social judgments at its core. Yet social comparison might not be all cognitive: There are downward (e.g., Wood, Michela, & Giordano, 2000) as well as upward social comparisons (e.g., Alicke, LoSchivavo, Zerbst, & Zhang, 1997), which tend to be driven by self-enhancing motivations, although scholars have argued such motivations may not always be involved (e.g., Chambers & Windschitl, 2004; Kruger & Gilovich, 2004). Moreover, it should be safe to assume that, among available options, individuals tend to categorize themselves with the more desirable or appropriate group. One challenge for cognitive explanations of TPP, therefore, lies in identifying and demonstrating evidence for cognitive explanations that can arguably be separated from motivational forces.
Adopting the premise that estimating media effects on multiple referents involves a comparative social judgment process, we propose that the cognitive mechanisms of assimilation, contrast, and anchoring effects involved in the process of social comparisons can offer such a cognitive explanation for TPP. Following a theoretical articulation of this explanation, this article will derive a set of hypotheses and test them with data gathered from a web-based experiment with an undergraduate sample.
TPP as Social Comparisons
Although the message effect questions in most TPP studies are asked on each target alone, in a form of “absolute judgments” rather than direct comparisons, TPP is a comparative social judgment in that the message effect questions compel a person to place multiple referents on a scale of being influenced. Consequently, rendering such a judgment requires retrieving, selecting, using, and weighing information about, among other things, each target, some undifferentiated social aggregate as a whole, and oneself (see Dunning, 2000).
Assimilation and Contrast Effects
Because TPP is defined in terms of self-other differences, the notion of “egocentrism” is readily applicable. Some scholars have equated egocentrism to a motivational force, variously labeled “self maintenance,” “self protection,” “self enhancement,” and so on (e.g., Brosius & Engel, 1996; David & Johnson, 1998; White & Dillon, 2000). A number of social psychologists (e.g., Chambers & Windschitl, 2004; Dunning & Hayes, 1996; Kruger, 1999; Mussweiler, 2003; Pronin, Gilovich, & Ross, 2004), however, have explicated egocentrism as a type of cognitive bias characterized by privileging self during the social judgment process. The self-serving motive may or may not be involved (Chambers & Windschitl, 2004; Kruger & Gilovich, 2004). That is, regardless of the presence of motivational forces, there are cognitive mechanisms that underlie the comparative social judgments in TPP.
As in other comparative social judgment tasks, when individuals make judgments regarding message effects estimates on self and on target others, they may take one of the two hypothesis-testing procedures: similarity testing and dissimilarity testing (Mussweiler, 2003). There is always a hypothesis-confirmation bias involved in these two procedures since expectation-confirmation is a key mechanism in brain dynamics that helps individuals navigate through the changing world (e.g., Spiers & Maguire, 2006). This general hypothesis-confirmation bias will privilege the activation and access of cognitive ingredients congruent with the focal hypothesis in forming judgments (see also Mussweiler, Rüter, & Epstude, 2004). Consequently, the similarity hypothesis testing may produce an assimilation effect (i.e., perceived similarity between self and a referent in the domain of being influenced by the media), resulting in reduced TPP. On the other hand, the dissimilarity hypothesis testing may lead to a contrast effect (i.e., perceived divergence between self and a referent in the domain of being influenced by the media), resulting in an increased perceptual gap. Assimilation and contrast effects have been well documented in social judgment theory (Granberg, 1982). However, such effects might be subtle such that, for an assimilation or a contrast effect to occur, individuals need to be primed with the corresponding mindset, that is, similarity- and dissimilarity-hypothesis-testing modes, respectively (e.g., Mussweiler et al., 2004; Tormala & Clarkson, 2007).
In the context of TPP, comparative judgments and such assimilation and contrast effects can occur in at least two potential domains: First, in the overall estimates of media effects on self and on target others, and second, in the domain of one’s media effects schemas. Media effects scholars have suggested that lay individuals have their own explanations (as relative to formal social scientific models and theories) for how the media might influence people (e.g., D. M. McLeod et al., 2001; Price et al., 1997). Media effects schemas are the informational ingredients individuals apply in their explanations of media effects, such as exposure to the media content, one’s critical use of media, and susceptibility to media influence. 1 Along this line of argument, it is predicted that
Hypothesis 1 (H1): When individuals are primed with the corresponding mindset, there would be assimilation and contrast effects, respectively, in the domain of media effects estimation:
Hypothesis 1a (H1a): With assimilation priming, there will be assimilation effects in perceived effects estimation for multiple referents; that is, the magnitude of TPP will be reduced.
Hypothesis 1b (H1b): With contrast priming, there will be contrast effects in perceived effects estimation for multiple referents; that is, the magnitude of TPP will be increased.
Hypothesis 2 (H2): When individuals are primed with the corresponding mindset, there would be assimilation and contrast effects respectively in the domain of media effects schemas:
Hypothesis 2a (H2a): With assimilation priming, there will be assimilation effects in the three referent-specific media effects schemas; that is, the perceptual gaps in these domains will be reduced.
Hypothesis 2b (H2b): With contrast priming, there will be contrast effects in the three referent-specific media effects schemas; that is, the perceptual gaps in these domains will be increased.
Anchoring Effects
Anchors are always involved in comparative social judgments (Festinger, 1954). The stimulus presented to and judged by the individual first in the sequence becomes the anchor or the reference point. Assimilation and contrast effects in social judgments are always in reference to the anchor. Assimilation effects occur when one’s judgment of the target stimulus shifts toward the anchor. Contrast effects occur when one’s judgment of the target stimuli shifts away from the anchor. For example, in Sherif, Taub, and Hovland’s (1958) study, with a 141-gram anchor (which was similar to the target weights), the perceived lightness/heaviness of the target weights shifted toward the anchor (i.e., an assimilation effect). With a 347-gram anchor (which was different to the target weights), the perceived lightness/heaviness of the same target weights shifted away from the anchor (i.e., a contrast effect). The anchoring effect is considered to be strong, robust, and reliable (Strack & Mussweiler, 1997).
Both perceptions of self and of others can function as anchors in social judgments (Karylowski, 1990). Otten and van der Pligt (1996) suggested that in a typical TPP study, an individual is more likely to use self as the anchor in the comparative social judgment in a self-other question condition (i.e., questions concerning self are asked first, and questions concerning others are asked second in the sequence). On the other hand, the other tends to be used as the anchor in the other-self question condition (i.e., questions concerning the other are asked first, and questions concerning self are asked second in the sequence). Chapman and Johnson (1999) suggested that the anchoring effect can be influenced by the activation of anchor-related knowledge. Differences in anchor-related information should then impact possible anchoring effects.
Scholars have suggested that substantial differences exist in information about self and others that individuals retrieve in a comparative social judgment. Such information differentials can be in amount, type, depth, and accessibility of information (McGuire & McGuire, 1986; Prentice, 1990; Rogers, 1981). Such information differentials relevant to TPP can be found in two content domains: (a) Whereas people have a wider range of information on their own orientations toward media use (e.g., selectivity and critical interpretations, see Kosicki & J. M. McLeod, 1990; J. M. McLeod & McDonald, 1985), such information about others may be limited, vague and diffused, and (b) whereas people have more privileged knowledge about how they themselves process a message and the general good nature of their own inner world (e.g., Kruger & Gilovich, 2004; Pronin et al., 2004), information on other’s psychological functioning is an informed conjecture at best. Usually not motivated to seek more information or to evaluate the accuracy of existing information, people often simply rely on the heuristic cues in the immediate context and/or bring in their self-anchored lay explanations (Kruger, 1999; Mussweiler, 2003).
This has implications for anchoring effect and assimilation and contrast effects in TPP. When the other is the anchor and self is the target, because self-related information is perceived as more specific and valid, perception of self is therefore more rigid and not as easily affected. Hence, the anchoring effect will not be robust, and in turn assimilation and contrast effects are less likely to emerge. On the other hand, when self is the anchor and the other is the target, because other-related information tends to be vague and inaccurate, perception of others is more flexible and easily affected. Hence, anchoring effect will be robust, and in turn assimilation and contrast effects are more likely to take place when the individuals are primed with the corresponding mindsets (see Biernat, Manis, & Kobrynowicz, 1997; Karylowski, 1990). Therefore, it is predicted that in a TPP setting, an interaction exists between mindset priming and anchor:
Hypothesis 3 (H3): There would be a significant interaction between mindset priming and anchor on referent-specific media effects schemas (critical media use, susceptibility to influence, exposure, and effect) such that:
Hypothesis 3a (H3a): When self is the anchor, there will be a significant main effect of assimilation/contrast priming on perceptions of other-related media effects schemas (i.e., the stimuli).
Hypothesis 3b (H3b): When other is the anchor, there will not be a significant main effect of assimilation/contrast priming on perceptions of self-related media effects schemas (i.e., the stimuli).
Method
Experimental Design
The study was a web-based experiment with a 2 (mindset priming: assimilation vs. contrast) by 2 (anchor: self vs. other) by 2 (target other: peer vs. parent) by 3 (media content: Internet pornography, reality, and prosocial public service announcements [PSAs]) mixed design. Mindset priming, anchor, and target other were between-subjects factors, and media content was a within-subjects factor. Manipulation of target others and media messages was adopted from Sun, Shen, and Pan (2008).
Mindset priming
Priming of mindset followed similar manipulations used in past research to induce assimilation and contrast effects (e.g., Mussweiler, 2001; Mussweiler et al., 2004). In the assimilation condition, participants were instructed to first “list up to four similarities between you and an average person in this age group” (either peer or parent), and then “elaborate and explain each of these similarities.” Text boxes were provided on the webpage for the participants to list, elaborate, and explain these similarities. In the contrast condition, the participants were asked to list up to four differences and then elaborate and explain in the text boxes provided.
Target referents
The two target referents were “an average person in your own age group” (i.e., peer) and “an average person in your parents’ age group” (i.e., parent). The purpose of this manipulation was to induce difference in perceived similarities as well as to enhance external validity.
Anchor
In social judgment settings, the stimulus presented to and judged by the individual first in the sequence becomes the anchor or the reference point. Therefore, the anchor manipulation was achieved by question order. When questions specific to self were asked first in the sequence, the self served as the anchor; the target other became the anchor when questions specific to the target other were asked first in the sequence.
Media messages
Three types of media content were used: Internet pornography, reality TV show, and prosocial PSA. Following the procedure in Sun, Shen, and Pan (2008), to ensure the same messages were under consideration by the participants, other than Internet pornography, we referred to specific messages. For reality TV shows, we asked about The Apprentice. Respondents first read a brief synopsis about the TV show and then answered questions relevant to the show. The same approach was used to ascertain measures related to prosocial PSAs. We asked about a PSA entitled “In America,” which encourages donations to charity. The purpose of selecting these media content was to induce variations in social desirability of presumed influence from a particular media message.
Participants
There were 511 participants recruited from introductory undergraduate communication classes at the University of Georgia. For their participation in the study, students either fulfilled a course requirement or earned some extra credit. The participants’ ages ranged from 18 to 42 years (M = 19.87, SD = 1.91). The majority were female (n = 350, 68.5%). The participants were predominantly white/Caucasian (n = 414, 81.0%), with 8.8% (n = 45) indicating Asian descent, 5.3% (n = 27) African American descent, and 3.5% (n = 18) Hispanic descent. Seven participants did not disclose their race/ethnicity. There were slight fluctuations in the actual sample size in data analyses due to missing responses.
Procedure
Data collection took place in the form of a web-based experiment, administered through SurveyMonkey. A separate webpage provided the consent form and the survey would not start until the participants clicked on the “consent” button. Clicking the “consent” button at the same time activated an automated algorithm that randomly assigned the participant to one of the eight versions of the questionnaire, as determined by the three between-subjects factors (mindset priming, anchor, and target other). After being assigned to one of the eight conditions, participants reported their general media use, completed the mindset priming task before they went on to respond to measures on media effects schemas, perceived media effects on self versus the other (in different sequences across conditions), and listed potential actions they might take related to the specific media content. They reported their demographic information toward the end of the survey. Not all data from the survey were reported here.
Measures
Perceived message effects
For each type of media content, a specific set of perceived effect measures were asked on respondents themselves and on the target other. For all the perceived-effect questions, the same 7-point scale was used (1 = not at all to 7 = a great deal). For reality TV show, the questions asked how the media content would impact the referent’s perception of the business world, goals in life, works ethics, and definition of success. For internet pornography, the question asked about impact on the referent’s moral values concerning sex, ways of dealing with sexually involved relationships, and acceptance of sexually explicit talks and images. For PSA, the questions asked about impact on the referent’s altruistic values, perception of charity organizations, behavior of helping others, and contribution to charity. CFA confirmed that all the perceived effects measures were unidimensional. Across three referents, reliabilities of the effects measures ranged from .93 to .94 for reality TV show, from .86 to .88 for Internet pornography, and around .94 for PSA. Within each media message, an index for perceptual gap in media effects was constructed by subtracting the perceived effect on self from the perceived effect on the target other. Positive score indicates presence of TPP, and negative scores suggest first-person perception (FPP).
Message exposure 2
The participants estimated the amount of exposure to each type of media content for each referent. To dilute the intrusiveness of questions on exposure to Internet pornography and to increase respondents’ truthful reporting, we included a preamble acknowledgment that encountering such materials could be accidental. Then two questions were asked about exposure to Internet pornography: (a) how frequently (1 = never to 7 = quite often) the participants themselves and the target other encountered such materials on the Internet, and (b) if they accidentally encountered such materials, how likely (1 = very unlikely to 7 = quite likely) they would “take a look at them.” The two measures were correlated for each target (r = .59, p < .001, for self; r = .53, p < .001, for both peer and parent). For each referent, the two items were multiplied to form an exposure index, which was then transformed via square root, returning it to the original scale.
For reality TV show, the participants were asked to indicate how frequently they themselves and the target other watched (a)The Apprentice and (b) similar TV shows (1 = never to 7 = quite often). The two items were correlated for each target (r = .48, p < .001, for self; r = .48, p < .001, for peer, and r = .60, p < .001, for parent). A cumulative index of reality TV show exposure was created by taking the average of the two.
Exposure to PSA was measured in a similar way as Internet pornography. The rationale was that viewers do not actively seek prosocial PSAs to watch, and exposure to PSAs tends to be less frequent and accidental. The participants were asked to rate (a) how often (1 = never to 7 = quite often) they and the target other have encountered a PSA in the past 12 months, and (b) how likely (1 = very unlikely to 7 = quite likely) each referent would pay attention to it when they encountered one on TV. The two items were correlated for each target (r = .44, p < .001, for self; r = .43, p < .001, for both peer and parent). For each target, the two items were multiplied to form an index of PSA exposure. Each resulting index was transformed via square root to the original scale.
Self-other gaps in exposure to each message were created by subtracting self-exposure from other-exposure. Hence, a positive score indicates the target other was perceived to have higher exposure than self, and a negative score suggested otherwise.
Critical media use
Respondents were asked to rate the critical use of the media by themselves and by the target referent on six 7-point scale questions (1 = rarely to 7 = very often) adapted from Kosicki and J. M. McLeod (1990; see also Price et al., 1997). CFA showed that these items were loaded on a single factor. The six items were averaged into one scale (α = .88). An index of self-other gap in critical media use was created by subtracting the score for self from the score for the other. A positive score indicated the other was perceived to use media information in a more critical manner than oneself. A negative score suggested otherwise.
Susceptibility to influence
Susceptibility to influence was measured by six 7-point items (1 = very unlikely to 7 = very likely). The respondents indicated how likely they themselves and the target other would: “be swayed by others who seem to know more,” “take up the perspective of a media report,” “be persuaded by an ad to buy a product,” “switch positions on an issue in a debate,” and so on. CFA confirmed that for each referent, the set of items formed a single factor. They were then averaged into an index for each referent (α = .82). An index of self-other gap was created by subtracting the score for self from the score for the other. A positive score indicated the other was perceived to be more susceptible to influence than oneself. A negative score suggested otherwise.
Social desirability
Social desirability of presumed message effects was measured by four 7-point semantic differential scales. The word pairs were socially undesirable, beneficial to society, harmful to cultural values, and favorable to societal norms, versus their counterparts. CFA confirmed that these items were unidimensional within each message. These items were averaged into a single index within each message. The reliabilities of the measure were .77 for Internet pornography, .70for reality shows, and .62 for PSA.
Results
Data Analysis Strategy
Recall that all the key variables of theoretical interest had multiple indices. The multiple indices under one concept were necessitated either by multiple referents (i.e., measures repeated for “yourself” and the target other), such as critical media use and susceptibility to influence, or by multiple referents and multiple messages (i.e., measures replicated on self and the target other, and varying with media content, such as perceived message exposure and perceived message effects). Such repeated measures resulted in interdependence in the data, where measurement occasions (defined as media content by target referent) were nested within each respondent. To address the interdependence in the data structure introduced by the repeated measures, the multilevel modeling approach was adopted in the data analyses and two-level models were estimated using the xtreg procedure in Stata, with individuals at the higher level and measurement occasions as repeated measures clustered within individuals.
The basic model predicting a dependent variable specific to message j by respondent i, is as follows:
where α denotes the constant, or the regression coefficient of x1ij = 1, βp is the regression coefficient of the predictor xpij, where p = 1, 2, 3 …P, ξ ij denotes the disturbance. For clustered data from a repeated-measures design, ρξij,ξij ≠ 0. Rather, ξ ij can be decomposed into two components as in ξ ij ’ ≡ µ i + ϵ ij , where µ i is individual-specific deviation from the overall mean for participant i, and ϵ ij is the transitory error component that varies across measurement occasions and participants. This recognition leads to express Equation (1) in a random-intercept model:
where µ i is assumed to have a mean zero and variance σ2µ (between-individual variance in yij) and ϵ ij has a mean zero and variance σ2ϵ (unaccounted residual variance).
The model estimates individual-specific constant term as α + µ i . If σ2µ is zero, a null hypothesis can be assessed via a likelihood ratio test, then, there is a constant α across all respondents, and Equation (2) is reduced to a linear regression model. The predictive power of xpij may be assessed by additional variance explained, calculated as the proportion of residual variance in a reduced model, that is, the model with p-k predictors, explained by k predictors added to the model (see Gelman & Pardoe, 2006). Significance of the reduction is assessed via a likelihood ratio test of the difference in the log-restricted likelihood statistics between the two models with k degrees of freedom. A multilevel model is required when σ2µ is nonzero. In that case, there is also covariance between µ i and ϵ ij , which is called intraclass correlation (ρ∈,µ). (see Rabe-Hesketh & Skrondal, 2005, for discussions of multilevel models in general). Repeated-measures ANOVA/ANCOVA was not preferred due to three reasons. First, there is no restrictions on the errors structure (i.e., the components in the variance-covariance matrix of σ2ϵ, σ2µ and ρ∈,µ), which is often too conservative and leads to lower statistical power. Second, repeated-measures ANOVA/ANCOVA can only handle completely balanced data. Third, the relationship between repeated-measures ANOVA/ANCOVA and multilevel modeling is similar to that between ANOVA and regression, with the latter being in a more generalized form and the former a special case.
Preliminary Results on Social Desirability
Preliminary analyses were performed to assess the social desirability of media messages under consideration in this study. A two-level model was estimated to predict social desirability, using message target referent, mindset priming, and anchor as fixed-effects factors and age and sex as covariates. Results showed that the main effect of message was significant: F(2, 1041) = 1052.63, p < .001. The marginal means showed that PSA was perceived to be the most desirable (M = 5.70, SE = 0.05), followed by reality TV show (M = 3.95, SE = 0.04), and Internet pornography was perceived to be the most undesirable (M = 2.60, SE = 0.05). All pair-wise differences were significant at the level of p < .05. Assuming α = .05, with a sample size of 510, the statistical power exceeded .67 to detect an effect size equivalent to r = .10, and exceeded .99 to detect an effect size equivalent to r = .20. These results showed that, as intended, the three types of media messages successfully induced variability in perceived social desirability.
Predicting Perceptual Gap in Perceived Effects
H1 predicted that TPP would be a function of mindset priming such that assimilation priming would reduce (H1a), and contrast priming would increase the magnitude of TPP (H1b). To test this hypothesis, a two-level model was estimated to predict TPP using message topic, target other, and mindset priming as fixed-effects factors, and with age and sex as covariates. The results showed that there was a significant main effect of message topic: F(2, 1128) = 147.51, p < .001: The marginal means revealed that the magnitude of TPP was the largest for Internet pornography (M = 0.74, SE = 0.08), followed by reality TV show (M = 0.40, SE = 0.07). The perceptual gap in perceived effects was in the form of FPP for PSA (M = −0.12, SE = 0.05). All pair-wise comparisons were significant at p < .05. Although the main effect of message topic was not of theoretical interest, it is important to have variability in the perceptual gap in message effects.
There was also a significant main effect of mindset priming: β = .40, p < .001. The marginal means revealed that when there was more TPP when the participants were primed with the contrast mindset (M = 0.52, SE = 0.06) than with the assimilation mindset (M = 0.12, SE = 0.06). More importantly, z tests showed that the magnitude of TPP was not significantly different from zero in the assimilation mindset condition but significantly different from zero in the contrast mindset condition. This showed that indeed there were assimilation effects (i.e., the perceptual gap was reduced to being not significantly different from zero) when the individuals were primed with the assimilation mindset and contrast effects (i.e., the perceptual gap was significantly larger and significantly different from zero) when the individuals were primed with the contrast mindset. Possible interaction effects involving priming manipulation was also assessed: There was a significant interaction between priming and message topic: F(2, 1112) = 6.81, p = .002. Marginal means showed that the pattern was consistent with assimilation/contrast effects within each topic: For Internet pornography, the marginal means were M = 1.01 (SE = 0.11) for contrast and M = 0.46 (SE = 0.11) for assimilation. For reality TV show, the marginal means were M = 0.69 (SE = 0.10) for contrast and M = 0.11 (SE = 0.10) for assimilation. And within PSA, the marginal means were M = −0.13 (SE = 0.07) for contrast and M = −0.11 (SE = 0.07) for assimilation. The interaction between priming and order was also significant: F(1, 1279) = 4.92, p = .03. Marginal means were almost identical between the two orders within the assimilation condition: M = 0.14 (SE = 0.08). On the other hand, within the contrast condition, the gap was significantly larger when self was the anchor (M = 0.71, SE = 0.08) than when other was the anchor (M = 0.38, SE = 0.08). No other interaction terms involving priming manipulation was significant. Hence, both components in H1 received support.
Assimilation and Contrast Effects in Media Effect Schemas
H2 concerned the assimilation and contrast effects in the domain of media effects schemas that were specific to referents such as exposure, critical media use, and susceptibility to media influence. Specifically, it was predicted that the self-other gap in these variables would be reduced in the assimilation condition (H2a) but increased in the contrast condition (H2b). To test this hypothesis, a series of two-level models were estimated, using message topic, mindset priming, target other, and anchor to predict the self-other gaps in three referent-specific media effects schema variables. The demographic variables were entered in the models as covariates.
Results from these two-level models showed that there was no main effect of message topic. Mindset priming did not predict self-other gap in exposure (β = .08, p = .35) or self-other gap in critical media use (β = −.17, p = .17), but it did significantly predict self-other gap in susceptibility to influence (β = .27, p = .02): Others were perceived to be more susceptible to influence than oneself. But the perceptual gap was larger when individuals were primed with the contrast mindset (M = 0.48, SE = 0.05) than when primed with the assimilation mindset (M = 0.20, SE = 0.06). Overall, these results did not support H2.
Anchoring Effects in the TPP Setting
H3 concerned the anchoring effects in a TPP setting, which involves comparative judgments of media effects on multiple referents. It was predicted that there would be a significant interaction between mindset priming and anchor due to information differentials between self and others. Specifically, the anchoring effect would be robust with the self as the anchor, and in turn, the main effect of assimilation/contrast priming would be significant on other-related media effects schemas (H3a). On the other hand, the anchoring effect would not be robust with the target other as the anchor, and the main effect of assimilation/contrast priming would be nonsignificant (H3b). To test this hypothesis, two sets of two-level models were estimated. The first set of models used message topic, mindset priming (assimilation vs. contrast), anchor (self vs. other), the three two-way interaction terms, and the three-way interaction term to predict the three target self-specific media effects schemas (susceptibility to influence, exposure and critical media use) and perceived effects on other. Along with the demographic variables, the corresponding variables for self were entered in the models as covariates. The second set of models used the same terms to predict the four variables for self, with demographic variables and the counterparts for target other entered as covariates. These models were estimated with the pooled data, with the data for peer referent and with the data for parent referent. Results from the latter two were consistent with those from the pooled data. Only the results from the pooled data were reported here.
Predicting other-related variables
There were no significant main effects of message topic on others’ critical media use, susceptibility, or effect on others, but there was a significant main effect of message topic on other’s exposure: F(2, 1477) = 10.46, p < .001. Marginal means showed that others were perceived to have similar levels of exposure to Internet pornography (M = 4.30, SE = 0.07) and to reality TV show (M = 4.33, SE = 0.09), but significantly less exposure to PSA (M = 3.95, SE = 0.07). There was no significant interaction between anchor and message topic on any of these variables. Neither was there significant interaction between priming and message topic. The mindset priming by anchor interaction term emerged as a significant predictor for each of the four other-related variables: β = −.59, p = .006, for critical media use; β = .68, p = .001, for susceptibility to influence; β = .29, p = .03, for exposure; and β = .58, p < .001, for perceived effects. The three-way interaction term was nonsignificant.
Predicting self-related variables
There was no significant effect of message topic on self-critical media use or susceptibility to influence. However, the main effect of message topic was significant on exposure: F(2, 949) = 211.85, p < .001. The self-exposure was significantly higher to PSA (M = 4.07, SE = 0.07) than to reality TV show (M = 2.88, SE = 0.06) and to Internet pornography (M = 2.24, SE = 0.06); the latter two were not significantly different from each other. The main effect of message topic was also significant on perceived effects on self: F(2, 1064) = 70.95, p < .001. Marginal means showed that individuals perceived themselves to be most influenced by PSA (M = 4.21, SE = 0.08), followed by reality TV show (M = 3.47, SE = 0.09), and by Internet pornography (M = 3.24, SE = 0.09). All pair-wise comparisons were significant. None of the interaction terms emerged as significant predictors for any of the four self-related variables, including the priming by anchor interaction.
To test H3a and H3b, the significant priming by anchor interaction effects on the four other-related variables had to be decomposed for interpretation. In the other-anchor condition (i.e., other-related questions were asked first), there was no significant main effect of mindset priming for any of the four variables. On the other hand, the main effect of mindset priming was significant for all four variables in the self-anchor condition (i.e., self-related questions were asked first). Figures 1a-1d present the decomposed interaction between mindset priming and anchor on the four other-related variables. The mean scores of the corresponding self-related variables in self-anchor and other-anchor conditions were juxtaposed in each figure to demonstrate the presence/absence of assimilation/contrast effects.

Assimilation and contrast effects in other-related variables as a function of anchor.
In the self-anchor condition, there were assimilation and contrast across the four variables. For susceptibility to influence (Figure 1a), others’ score in the assimilation condition (M = 3.97, SE = 0.06) was closer to the self score (M = 3.80, SE = 0.05) than in the contrast condition (M = 4.54, SE = 0.06), with an effect size equivalent to r = .18. 3 For exposure (Figure 1b), others’ score in the assimilation condition (M = 3.78, SE = 0.05) was closer to the self score (M = 2.93, SE = 0.04) than in the contrast condition (M = 3.97, SE = 0.05), with an effect size equivalent to r = .06. For critical media use (Figure 1c), others’ score in the assimilation condition (M = 4.11, SE = 0.06) was closer to the self score (M = 4.24, SE = 0.06) than in the contrast condition (M = 3.45, SE = 0.05), with an effect size equivalent to r = .21. Finally for perceived effects (Figure 1d), others’ score in the assimilation condition (M = 3.98, SE = 0.05) was closer to the self score (M = 3.44, SE = 0.05) than in the contrast condition (M = 4.27, SE = 0.06), with an effect size equivalent to r = .09. All pair-wise comparisons were significant at p < .05. This means the simulation/contrast priming effects occurred as predicted. In the other anchor condition, the others’ scores in the assimilation and contrast conditions were not significantly different from each other, which means assimilation/contrast priming effects did not occur. These results showed that both H3a and H3b were supported.
Discussion and Conclusion
Conceptualizing TPP as an outcome of rendering comparative social judgments, we set out to propose a social comparison explanation for TPP. It is proposed that TPP shares the common cognitive mechanisms that underlie other types of comparative social judgments, which include assimilation and contrast effects as well as anchoring effect. We developed a set of hypotheses and tested them with data from a web-based, mixed-design experiment. The study had some strengths and limitations.
Strengths and Limitations
Three types of media messages distinctively located at the spectrum of social desirability, as well as two target others with variable social distance from the self, were utilized in this study. This feature contributed to increased external validity of the findings. It would be more difficult to conclude the same findings based on references to “the media” in general or general content category with only assumed rather than measured desirability of their presumed influences. The mixed design in this study means that key variables were measured multiple times across target referents as well as media messages. Such a design might be necessary for the detection of assimilation, contrast, and anchoring effects because they might be subtle and, therefore, require certain “volume” (i.e., repeated observations) for the patterns to emerge. Two features in the study reduced threats to internal validity that might be associated with interdependence in the data structure due to repeated measures. First, the multilevel modeling approach is more parsimonious, which reduces the inflation in Type I error due to multiple tests. Second, the procedure is more appropriate when the independence assumption is violated due to repeated measures. The multilevel modeling approach yields unbiased estimates of the error terms, hence produces less biased estimates and more robust significance tests.
There were also features in this study that reduced external validity. The findings might have low ecological validity because the participants in our study were not randomly selected from a larger definable population but from a convenience sample instead. The findings might not be generalizable beyond the three types of media messages or the two target others considered in this study. There were also limitations that might reduce the internal validity. As a web-based experiment, our study was cost-effective as far as resources and efficiency were concerned. However, we did not have the same levels of control in the data collection as in a laboratory study. Target other (peer vs. parent) was a between-subjects factor, which might have introduced biases into social judgments that were rather subjective in nature. There has been evidence for such biases due to the between-subjects design. For example, 9 might be perceived as significantly larger than 221 (Birnbaum, 1999). However, there is also evidence that TPP remains robust in a between-subjects design (David et al., 2004). With these strengths and limitations in mind, we turn to the specific findings.
Social Comparisons in TPP
Extending the cognitive explanations, we proposed that as comparative social judgments in the domain of media influence, TPP should share the same cognitive mechanisms that underlie social judgments in general. That is, there exist two hypothesis-testing (similarity and dissimilarity) procedures that underlie all social judgments. The similarity-hypothesis-testing procedure leads to assimilation effects and the dissimilarity-hypothesis-testing procedure to contrast effects. Consistent with previous research, findings from this study showed that assimilation and contrast effects might be subtle: They might require priming of the proper mindset to take place, as well as the presence of anchoring effect. Without considering anchoring effects, assimilation and contrast effects occurred for self-other perceptual gap in perceived effects (H1) and susceptibility to influence but not for critical media use or exposure (H2).
The comparative nature of TPP also means that anchoring effects are at play in social judgments. It is argued that there exists information differential between self and target others when individuals render comparative judgments in the domain of media effects. Individuals tend to have more information about self, which is also more valid and accessible. On the other hand, individuals tend to have less information about others (especially when they are socially distant from oneself), which is also less accurate, diffused, and vague (i.e., inaccessible). Consequently, perception of self is less likely to be affected by anchoring effect, hence, not subject to assimilation or contrast, while perception of others tends to be influenced by anchoring effect and more likely to shift toward (assimilation) or away from (contrast) the anchor. Across four domains in the context of perceived media effects (susceptibility to influence, exposure, critical media use, and perceived effects), results on perceptions of self and target other demonstrated clear patterns consistent with this line of argument (H3, Figures 1a-1d). The lack of support for H2 then was probably due to the interaction effect between mindset priming and anchor: The assimilation and contrast effects for exposure and critical media use only occurred when self was the anchor but did not emerge when the other was the anchor.
Synergizing Motivation and Cognition in TPP Research
These results showed evidence that TPP is a persistent cognitive bias (see also David et al., 2004), which reflects a much broader fallacy in social perceptions and judgments (Pronin et al., 2004). Arguably, among the host of existing cognitive explanations, the mechanisms of assimilation, contrast, and anchor effects are most separated from motivational elements. This is by no means to challenge or to deny the validity of the other explanations, including attribution error, exposure, and categorization explanations. Instead, they might well overlap with and be supplementary to each other: Comparative social judgments are at the core of the self-categorization explanation, and, as the results in this study showed, assimilation and contrasts effects do occur regarding exposure. To advance TPP research and further our understandings of cognitive explanations for TPP, then, requires scholars to develop and test theoretical frameworks that can integrate these different explanations yet remain cohesive and parsimonious.
To develop such cognitive models is not to deny the role of motivation in TPP. Motivation must be an integral part of the processes that produce such a fallacy. As a matter of fact, results regarding message topic and TPP were consistent with the motivational explanation: While there was TPP for Internet pornography and reality TV show, there was FPP for prosocial PSA. And the self-other gaps in the three media effects schemas clearly portrayed positive images (Figure 1). Media effects estimates, as in all social judgments, are “relational,” “motivational,” and “selective” (Dunning, 2000, p. 374). No theoretical model of TPP would be truly integrative until motivational factors are explicitly incorporated as among the key explanatory variables (see Kruglanski & Mayseless, 1990). The point is, as the results in this study clearly demonstrated, the cognitive mechanisms could be at work regardless of the presence/absence of the motivational forces. The challenge is to develop a clear theoretical framework to not only differentiate and incorporate but also explain judgments embedded with motivated and nonmotivated forms of biases (Chambers & Windschitl, 2004).
The implicit and explicit assumptions in this study suggest that developing such an integrated theoretical model must be accompanied with more innovative methods. The evidence from the current study was rather indirect due to these assumptions and the method applied in the study. First, the prerequisite informational differential regarding self and others was only assumed but not actually measured. It is also possible that there is no information differential, only that not all available information was applied in the social judgment tasks regarding self and others, or the same information was not applied in the same manner, as suggested by the dual-process models. Second, the finding in this study could have resulted from the characteristics of our measures, which might implicitly privilege the interpretive context casting media influence in a negative light (Price et al., 1997). Another possibility is testing effects. By asking about the media effects schemas and compelling the participants to render media effects judgments on multiple referents, we are at the risk of creating cognitions, instead of accessing cognition (Ogden, 2003). Future studies with more creative methods that assess cognitive processes more directly and without relying on self-report measures should provide better testing of the cognitive model.
What we can conclude from this study are three points. First, further development of the TPP research could benefit from a more systematic appropriation of the contemporary literature in social comparisons and social judgments. Second, the social comparison explanation for TPP has empirical credence but will need stronger tests on its core propositions and hypotheses with more innovative methods. Third, a remaining major weak spot in the extant TPP literature is the lack of an integrated framework of cognitive mechanisms and various motives relevant to message effects estimates that individuals render in realistic and socially consequential action situations.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
