“This Unfavorable Poll Result for My Candidate Doesn’t Affect Me but Others”: Third-person Perception in Election Poll Coverage

Many political communication scholars have investigated whether news coverage of election polls affects voters’ perceptions and candidate choices (Farjam, 2021; Gattermann et al., 2022; Lazarsfeld et al., 1948; Myers et al., 1977). These studies have resulted in conflicted findings, with some studies showing a bandwagon effect (i.e., an increase in support for the leading candidate; Lazarsfeld et al., 1948; Myers et al., 1977) and others showing an underdog effect (i.e., an increase in support for the trailing candidate; Mendelsohn & Crespi, 1970; Navazio, 1977). In addition to research into the actual effect of polling news on voting, a number of studies have also examined how voters perceive the impact of pre-election poll results on their own opinions and others, given that people tend to incorporate their perceptions about polling effects (i.e., reported poll results) into their political beliefs (Chung et al., 2018; H. Kim, 2022; Perryman et al., 2020; Tappin et al., 2017).

It is worth noting that the perceived effects of polling news play a key role in public opinion and voting behaviors (H. Kim, 2022; Perryman et al., 2020). First, a perceived effect of polling news could lead to support for restrictions on election polls, such as banning election-night projections (de Vreese & Semetko, 2002; Lavrakas et al., 1995; Price & Stroud, 2006). Many countries already restrict the publication of poll results before an election day (Spangenberg, 2003) under the assumption that news coverage can affect voters’ opinions and even distort public opinions in some cases. Several studies examining the polling effect have based their claims on respondents’ self-reported perceptions about such an effect (Lavrakas, 2008; McAllister & Studlar, 1991; Myers et al., 1977). Some of those studies have also addressed that only a small number of voters admitted to being influenced by polls (Kavanagh, 1995; Worcester, 1996).

Many studies have brought attention to cognitive biases that influence estimations of the effects of polling news (H. Kim, 2022; Perryman et al., 2020; Price & Stroud, 2006). These studies have specifically shown evidence of a third-person effect (Davison, 1983), where individuals perceive that media messages—including election polls—will have greater effects on others than they do on themselves (Hyun & Seo, 2021; Jang & Kim, 2018; Perloff & Shen, 2022; Schmierbach et al., 2023). The findings of studies examining polling effect perceptions have generally been consistent with the third-person perception; that is, that voters tend to perceive that poll news has a greater effect on others than it does on themselves (Chung et al., 2018; Glynn & Ostman, 1988; Price & Stroud, 2006; Wei et al., 2011). However, these previous studies have examined the perceived impact of overall poll coverage, and it remains uncertain whether individuals perceive the impact of various poll news in the same manner.

Election polls provide specific information about who is leading at the time, which may affect the perceived impact of polling news. Existing studies have mainly examined the impact of poll coverage as a whole, so it has not been systematically studied whether third-person perception differs depending on the content of poll coverage. In the case of political ads, it was found that third-person perception varies depending on the content of the political ad (i.e., whether the ad is for or against a political party the recipient supports; Meirick, 2004; Price et al., 1998). The key content of election polling coverage is which candidate is ahead, and third-person perceptions are likely to differ accordingly. This study aims to investigate whether third-person perceptions vary depending on the content of the poll coverage (i.e., who is leading). Specifically, we systematically test whether third-person perceptions vary as a function of the desirability of the reported poll results from the recipient’s perspective (i.e., personal desirability of the poll result).

Third-Person Perception of Election Polls

The third-person effect involves both perceptual and behavioral hypotheses (McLeod et al., 1997; Perloff, 2009; Shen et al., 2018). The perceptual hypothesis states that the perceived media impact on the third person (PMI3) is greater than the perceived message impact on oneself (PMI1), also called third-person perception. Meanwhile, the behavioral hypothesis contends that the difference between PMI3 and PMI1 (i.e., the degree of the third-person perception, TPP hereinafter) affects relevant attitudes or behaviors.^[1] Thus, TPP is the key element in both the perceptual and behavioral hypotheses. Regarding election poll coverage, it has been found that the perceived impact of such coverage is greater on others than on oneself (i.e., perceptual hypothesis; Glynn & Ostman, 1988; H. J. Kim, 2013; Wei et al., 2011), and that a greater TPP was associated with a higher level of support for the prohibition of poll reporting (i.e., behavioral hypothesis; Price & Stroud, 2006; Yang, 1997). This study focuses on the perceptual hypothesis and whether the content of the poll coverage affects TPP of election poll coverage’s impact. Researchers have also investigated what factors affect TPP regarding election poll coverage. Noteworthy research on this topic includes Price and Stroud (2006). Based on the finding that third-person perception is predominantly observed in socially undesirable media content, Price and Stroud hypothesized that individuals who view poll reporting as socially undesirable would exhibit a greater TPP. They confirmed this hypothesis through a survey.

Desirability of Election Poll Results and Third-Person Perception

In third-person perception research, social desirability of media content is known to be a key moderator of third-person perception (Jensen & Hurley, 2005; Perloff & Shen, 2022; Shen et al., 2018). Previous studies have shown the existence of third-person perception for media content that is typically thought to be antisocial, such as pornography (Gunther, 1995; B. Lee & Tamborini, 2005). However, some researchers have pointed out the limitations of social desirability in explaining third-person perception. Andsager and White (2007) noted that “the way to define what is desirable and what is undesirable remains unclear” (p. 32). They pointed out that a message that is arbitrarily labelled as socially undesirable by the researchers may be viewed as desirable by the recipient, and if so, third-person perception may not be detectable in that person. Perloff and Shen (2022) have also raised concerns about the concept of social desirability, which they refer to as “message desirability.” They argue that there are uncertainties surrounding the definition of desirability and its underlying attributes. Furthermore, it has been observed that certain media messages, which are not inherently classified as socially undesirable, can still elicit third-person perception. For example, Meirick (2004) has shown the third-person perception about political advertisements, which are not simply categorized as socially undesirable. However, he found the third-person perception about a political advertisement not for all respondents, but only for those who might view the advertisement as undesirable (e.g., when conservative respondents read an ad from the Democrat candidate). Public campaigns addressing a controversial social issue are neither inherently socially desirable nor undesirable. However, third-person perception has been observed regarding the effect of those campaigns among those who had initially negative attitudes toward the issue (the nuclear power plants, Chung & Lee, 2012). Interestingly, they found the first-person perception (i.e., PMI1 > PMI3) among those who initially had a positive attitude toward it.

Similar to political advertisements and public campaign messages, there is no social consensus on the desirability of election poll coverage. However, third-person perception has been observed among those who viewed election poll coverage as undesirable (Price & Stroud, 2006). Price and Stroud measured individuals’ assessment about the social desirability about the poll coverage and used it as a moderator for third-person perception. A person’s assessment of the social desirability of poll coverage can be referred to as their attitude toward polls or their personal desirability about poll coverage. Price and Stroud demonstrated that personal evaluations of the desirability of a message (i.e., personal desirability), rather than a message-level variable of the social desirability of the message, can explain third-person perception. Chung and Lee (2012) show that the personal desirability of campaign messages explain third-person perception of persuasive campaigns. This study attempts to explain the third-person perception in election polling news by focusing on the personal desirability used in Price and Stroud (2006).

The Effect of Election Poll Results on Perceived Poll Impact on Oneself

In previous studies, perception of the impact of specific poll news has not been studied in terms of its content (i.e., who is leading) and the personal desirability of the content. Similar to Meirick’s (2004) finding that the personal desirability about a political ad affects perceptions of the ad’s impact on oneself and others, individuals may assess desirability of a specific poll results based on the poll result and the desirability may affect perceptions of the election poll’s impact on oneself and others.

Suppose an individual strongly supports a specific candidate (Candidate 1) over another candidate (Candidate 2) in an election race. When exposed to poll news showing that Candidate 1 (vs. Candidate 2) is leading the race, that individual is likely to view that poll news as being positive and desirable (vs. viewing it as being negative and undesirable). As noted above, individuals’ desirability of reported election poll results may influence their perceptions of the effect of those polling reports. For example, if the individual’s preferred candidate is losing in the poll, which is undesirable, they are more likely to perceive the effect of polling news on others to be greater than it is on themselves. Individuals’ personal desirability regarding the election poll result can be determined by the preferred candidate’s status in the poll or congruency between an individual’s attitude toward the candidates (i.e., the candidate preference) and the polling result (i.e., who is leading in the race).

According to impression management motivation theory (e.g., Douglas & Sutton, 2004; Tal-Or & Drukman, 2010), individuals are less likely to admit that they are gullible and persuadable. In other words, even if a persuasive message affects people, they tend to deny the influence of the external message to manage positive impressions of themselves (i.e., being strong and not susceptible). Similarly, voters who want their preferred candidate to win the election are likely to accept the influence of reported election poll results if the polling news indicates that their supporting candidate is leading in the race. By contrast, voters are less likely to admit the influence of reported election poll results if their preferred candidate is losing in the race. This study hypothesizes that PMI1 is greater when the preferred candidate leads in the poll (i.e., when a respondent’s pre-existing attitude toward candidates and the polling results are congruent) than when the candidate is losing (i.e., incongruent).

The Effect of the Preferred Candidate’s Status on Perceived Impact on Others

The personal desirability of reported election poll results can affect not only PMI1 but also PMI3. Individuals perceive a message argument quality to be higher when the given message position is congruent with their existing attitudes (Lord et al., 1979). Moreover, individuals who evaluate the quality of a message as being higher are likely to perceive the message’s impact on other citizens to be greater (Zhao et al., 2011). On the other hand, when the message position is incongruent with respondents’ existing attitudes, they are likely to underestimate the message quality and its effect on other message recipients. Similarly, when an individual’s preferred candidate leads in the poll, they wish that the poll result will influence other citizens (i.e., wishful thinking, Bastardi et al., 2011). Chung and Lee (2012) found that when the message position is congruent (vs. incongruent) with pre-existing attitudes, respondents tended to perceive the message quality to be greater, with this perceived message quality finally leading to an increase in PMI3. Following the results of prior studies, the present research predicts that PMI3 will be higher when the preferred candidate is leading than when the attitude and the result are losing in the poll.

Effect of the Preferred Candidate’s Status on TPP

TPP, the degree of the third-person perception has been typically operationalized by the difference gap between perceived message impact on others (PMI3) and perceived message impact on self (PMI1, Schmierbach et al., 2008). Most scholars examine PMI1 and PMI3 in a parallel way and subtract PMI1 from PMI3 to compute TPP (Gunther, 1995; McLeod et al., 1997; Price & Tewksbury, 1996). With this operationalization, PMI1 and PMI3 determine TPP mathematically (i.e., TPP = PMI3 – PMI1). In this context, the variation in TPP can be explained by variations in PMI3 and PMI1. More specifically, if a factor increases PMI3, that factor will also increase TPP, because as PMI3 increases, TPP increases. Similarly, if a factor increases PMI1, that factor will decrease TPP, because as PMI1 increases, TPP decreases (Chung & Lee, 2012).

Considering the mathematical relationships among PMI1, PMI3, and TPP (for a review, see Chung & Moon, 2016), the current study predicts that PMI1 and PMI3 will be greater when one’s preferred candidate is leading in the poll than when the candidate is losing. As discussed earlier, TPP is determined by PMI1 and PMI3 but it is associated with the two variables in opposite ways. Previous studies have shown that when media messages are desirable (vs. undesirable), PMI1 is higher, and that higher PMI1 leads to a lower TPP (Duck et al., 1995; Gunther & Mundy, 1993). Chung and Lee (2012) have shown that when one’s attitude was incongruent with the message position, PMI1 and PMI3 were lower whereas TPP was greater. Based on the findings of previous studies, this study predicts that TPP will be lower when the preferred candidate is leading than when the candidate is losing in the polling reports.

H1: Both (a) PMI1 and (b) PMI3 are greater when one’s preferred candidate is leading rather than losing in the polling news. However, (c) TPP is lower when one’s preferred candidate is leading rather than losing in the polling news.

Effect of Subjective Political Knowledge on PMI1 and TPP

Previous research examining the third-person perception has indicated that the perceived impact of media messages varies depending on subjective judgment about the level of knowledge they have on the topic (i.e., subjective knowledge, Chung & Lee, 2012; Driscoll & Salwen, 1997; Paul et al., 2000; Salwen & Dupagne, 2001). Salwan and Dupagne (2001) found that subjective knowledge is negatively associated with PMI1 regarding television violence. Price and Tewksbury (1996) measured the perceived impact of political scandal news such as “Whitewater Affair” and found that respondents’ subjective knowledge about an issue has a negative effect on PMI1.

Rucinski and Salmon (1990) argued that people with high levels of public information and knowledge tend to believe that they are better able to resist external persuasion attempts than others. People may use their existing knowledge to accept new information. However, as Rucinski and Salmon pointed out, the knowledge one possesses can be effectively used to resist external information that is deemed undesirable. Therefore, those who perceive themselves as being knowledgeable are more likely to see themselves as effectively resisting desirable messages from the outside and being less influenced compared to others who lack knowledge. Applying this to the perception of the influence of election result reporting, people who believe they have a lot of knowledge about elections and politics are more likely to perceive themselves as being less influenced by election result reports (i.e., lower PMI1) compared to others who do not have such knowledge. In particular, they are likely to perceive that they resist and are less influenced by poll reports suggesting that the candidate they support is losing. Thus, we predict that the amount of political knowledge that individuals think they have (i.e., subjective political knowledge) will be negatively associated with PMI1 when respondents’ preferred candidate is losing instead of leading in the poll news.

Subjective political knowledge decreases PMI1, and decreased PMI1 increases TPP. Driscoll and Salwen (1997) showed that perceived knowledge relevant to the issue was significantly associated with TPP about the O. J. Simpson trial. Salwen and Dupagne (2001) also found that self-perceived knowledge about television violence was negatively associated with PMI1 but positively associated with TPP regarding the effect of television violence. A higher education level among respondents has been shown to increase TPP (Driscoll & Salwen, 1997; Tiedge et al., 1991), and perceived knowledge of oneself has also been positively associated with TPP (Mutz, 1989; Price & Tewksbury, 1996). The results of these studies consistently suggest that increases in subjective knowledge are related to decreases in PMI1 and to increases in TPP.

H2: For individuals whose preferred candidate is losing in the poll (i.e., the incongruent-poll result), subjective political knowledge level is negatively related to (a) PMI1 but is positively related to (b) TPP.

Effect of Others’ Level of Political Knowledge on PMI3 and TPP

In addition to the respondents’ level of political knowledge, the information about others’ political knowledge may affect PMI3 and TPP. Salwen and Dupagne (2001) maintained that the tendency of people viewing themselves as smarter than others is crucial in the third-person perception, although they did not assess people’s perceptions about others’ knowledge. Several studies have found that respondents’ perceptions of media effect on others vary based on their perceptions of others’ level of knowledge (Chung & Lee, 2012; Eveland et al., 1999; Schmierbach et al., 2011). For example, Eveland et al. (1999) found that college students perceived that violent music would have a greater impact on younger target groups (e.g., adolescents) than adults. College students might view adolescents as lacking knowledge of the negative effects of violent rap music, and thus they perceive a high PMI3 of violent rap on adolescents. In a study of persuasive messages about nuclear power, Chung and Lee (2012) found that respondents evaluated others’ knowledge differently depending on who the others were, and that they estimated PMI3 differently depending on others’ knowledge. For instance, if the others were considered to be experts in nuclear power, then the perceived knowledge of others was higher than that of the self. These findings suggest that the target group’s level of knowledge about the issue is negatively associated with both PMI3 and TPP. Based on the findings of past studies, the following hypothesis is proposed.

H3: (a) PMI3 and (b) TPP are greater for the non-expert target group than for the expert target group.

Method

Participants and Procedure

To test the proposed hypotheses, we conducted an online experiment in the context of a gubernatorial election in South Korea. South Korea has been a politically polarized country with a strong partisan media system, and there is a growing concern about the political polarization between its liberal and conservative citizens (Han, 2022; Hyun & Seo, 2021; H. Kim, 2015; S. K. Lee et al., 2022). The 2014 Gyeonggi gubernatorial election in South Korea offers a good opportunity to test the proposed hypotheses because the ruling and opposing parties were competing in a tight race at the time of data collection. The 2014 Gyeonggi gubernatorial election was held on June 4, 2014, with Nam Kyung-pil (Saenuri Party, the ruling party) and Kim Jin-pyo (New Politics Alliance for Democracy, the opposing party) as the two main candidates. Candidate Nam won the election with 50.43% of the vote.

The survey was conducted from May 28 to June 2 in 2014, one week prior to the 2014 Gyeonggi gubernatorial election. The online survey was carried out with the assistance of a Korean polling agency that has a panel of approximately 276,000 members. For data collection, an email request was sent to 2,000 members using stratified sampling according to the age and gender distribution of voters. Respondents were allowed to participate in the survey if they indicated that they were willing to vote (i.e., either likely to vote or will vote) and if they had a clear candidate preference for either Candidate Nam or Candidate Kim. In total, 539 respondents initially participated in the online survey.

For a participant to be included in this study, they had to meet the following inclusion criteria: participants should (1) have not only a clear preference for one of the two major candidates, but also an intention to vote in the 2014 Gyeonggi gubernatorial election, (2) pass the attention check question, and (3) confirm the quality of their own survey data. Twenty-eight participants did not answer either attention or a quality check.^[2] For this reason, they were excluded from the analysis, therefore resulting in a final sample of 511 participants, including 277 (54.2%) males and 234 (45.8%) females. The participants’ ages ranged from 20 to 78 years (M = 42.36 years, SD = 12.13 years). In terms of support, 259 (50.3%) supported Candidate Nam whereas 252 (49.7%) supported Candidate Kim.

At the beginning of the online survey, respondents were presented with one of the two newspaper articles presenting pre-election poll results (i.e., Candidate Nam leading or Candidate Kim leading) that were manipulated in the online experiment. That is, participants were randomly exposed to one of the two fictional articles and were then asked to read the article carefully. They were also requested to answer and measure the study’s survey questions. Lastly, respondents were debriefed about the purpose of this study at the end of the survey.

Manipulation of Independent Variables

Status of the Preferred Candidate in the Poll

Participants were randomly assigned to either the preferred candidate leading condition or the preferred candidate losing condition. As noted earlier, participants were presented with a piece of online news about an election poll result, which indicated that either Candidate Nam was leading or Candidate Kim was leading. The title of the poll news was “Beyond the margin of error, [Candidate Nam or Candidate Kim] outruns [Candidate Kim or Candidate Kim].” The article then reported, “[Candidate Nam’s or Candidate Kim’s] support rate is 42.2% and [Candidate Kim’s or Candidate Nam’s] is 31.5%, widening the gap between the two candidates from last week.” Participants were randomly assigned to one of these two conditions.

In the survey questionnaire, participants were asked to indicate whom they wanted to win in the election. That is, they had to choose their preferred candidate (either Nam or Kim). Combining the manipulated leading candidate in the poll and (i.e., Candidate Nam leading vs. Candidate Kim leading) and candidate preference (i.e., Candidate Nam supporters vs. Candidate Kim supporters), four groups were created. The preferred candidate leading condition was formed by combining Group 1 and Group 4 whereas the preferred candidate losing condition was formed by combining Group 2 and Group 3.^[3]

Measures of Variables

Perceived Effect of Poll News on the Self (PMI1)

To measure PMI1, we used an 11-point Likert scale (0 = not agree at all, 10 = completely agree), where the questions were as follows (Chung & Lee, 2012): The article regarding the poll results (a) influenced my evaluation on the two candidates; (b) influenced my attitude toward the candidates; (c) influenced my choice of candidates. We calculated the average values for these three questions (Cronbach’s α = .97, M = 4.02, SD = 2.42) and used a composite variable for PMI1.

Perceived Effect of Poll News on the Public (PMI3_public)

We measured the perceived effect of polling news on the general other or public (PMI3_public). Participants were presented with three questions that were similar to the questions presented above for PMI1: (a) The polling news would influence other members of general public’s evaluations of the two candidates; (b) the polling news would influence other members of general public’s attitudes toward the candidates; (c) the polling news would influence others’ choice of candidates. For the PMI3_public variable, we calculated the averages of the responses to the three questions and used them as a composite variable (Cronbach’s α = .97, M = 5.74, SD = 1.75).

Perceived Effect of Poll News on Expert Group and Nonexpert Group

We additionally measured PMI3 for the expert group with a high level of political knowledge (PMI3_expert) and the nonexpert group with a low level of political knowledge (PMI3_nonexpert). Regarding the perceived polling effect on experts or nonexperts, we used questions regarding PMI3_public by replacing “general others” with “people who are well-informed about Korean politics, such as political experts” and with “people who are uninformed about Korean politics,” respectively. We created composite variables for PMI3_expert (Cronbach’s α = .98, M = 4.91, SD = 2. 00) and PMI3_nonexpert (Cronbach’s α = .98, M = 6.17, SD = 1.76).

Third-Person Perceptions

TPP for the public refers to the difference between the perceived polling effect on general others (i.e., public) and the perceived polling effect on oneself. We created TPP_public by subtracting PMI1 from PMI3_public (TPP_public = PMI3_public – PMI1). TPP for other types of others was created in the same way: TPP_expert = PMI3_expert – PMI1; TPP_nonexpert = PMI3_nonexpert – PMI1.

Subjective Level of Political Knowledge

We asked the following three questions to measure the respondents’ level of political knowledge: (a) How much do you think you know about Korean politics? (b) How much do you think you know about the candidates for the 2014 Gyeonggi gubernatorial election? (c) How much do you think you know about the procedures of the 2014 Gyeonggi gubernatorial election? The 11-point Likert scale used to capture the response to these three items ranged from 0 = not at all, 5 = moderate, to 10 = very much. The average values of the responses to the three items were used as a composite variable (Cronbach’s α = .85, M = 6.11, SD = 1.47).

Participant’s Interests in the Election

We also asked the following three questions to measure respondents’ interests in the 2014 Gyeonggi gubernatorial election: (a) How important do you think this election is to you?; (b) How much interest do you have in this election?; (c) How relevant do you think this election is to you? The 11-point Likert scale used to capture the responses for these three items ranged from 0 = not at all, 5 = moderate, to 10 = very much. The average values of the responses to the three items were used as a composite variable (Cronbach’s α = .91, M = 4.75, SD = 1.66).

Results

Effect of the Preferred Candidate’ Status in the Poll (H1a, H1b, & H1c)

Correlations among variables in the main analyses are presented in Table 1.^[4]

H1 predicted the effect of the preferred candidate’s status on PMI1 (H1a), PMI3_public (H1b), and TPP_public (H1c). Because PMI1, PMI3_public, and TPP_public were highly correlated (r [PMI1, PMI3_public] = .42, r [PMI3_public, TPP_public] = .32, both p < .001), we used a multivariate analysis of variance, in which PMI1, PMI3_public, and TPP_public were explained by the independent variable, question order, and other control variables. In particular, question order (self-question first vs. other-question first), subjective political knowledge, sex, age, and interest in the 2014 Gyeonggi gubernatorial election all served as covariate variables, and the analysis controlled for their effects on dependent variables. At the multivariate level, the preferred candidate’s status had a significant main effect on dependent variables (Wilks’ Lambda = .94, η_p² = .06, p < .001). Question order, election interest, and sex all had significant main effects on dependent variables (Wilks’ Lambda = .95, η_p² = .05, p < .001; Wilks’ Lambda = .98, η_p² = .02, p = .020 Wilks’ Lambda = .99, η_p² = .01, respectively, p = .031 for all three cases).^[5] No other variables or interactions were found to have a significant effect on the dependent variables (see Table 2 for the result of the multivariate ANOVA).

PMI1 was found to be significantly greater for the preferred candidate leading condition (M = 4.61, SD = 2.33) than it was for the preferred candidate losing condition (M = 3.44, SD = 2.34), F(1, 500) = 29.92, p < .001, η_p² =.06. The model’s R² = .42. Thus, H1a was supported. PMI3_public was found to be significantly greater for the preferred candidate leading condition (M = 6.07, SD = 1.63) than it was for the preferred candidate losing condition (M = 5.41, SD = 1.81), F(1, 500) = 15.90, p < .001, η_p² =.03. The model’s R² for PMI3 = .10. Thus, H1b was supported. TPP_public was found to be significantly greater for the preferred candidate losing condition (M = 1.97, SD = 2.37) than it was for the preferred candidate leading condition (M = 1.47, SD = 2.33), F(1, 500) = 6.95, p = .009, η_p² =.01. The model’s R² = .05. Thus, H1c was supported.

Effect of Subjective Political Knowledge (H2a & H2b)

H2a predicted the negative relationship between subjective political knowledge and PMI1 for the preferred candidate losing condition group. The result of regression analysis, where PMI1 was predicted by subjective political knowledge, election interest, preferred candidate’s status in the race, sex, and age, showed that the effect of subjective knowledge on PMI1 was not statistically significant for all participants (b < 0.01, standardized b = .003, t = 0.06, p = .956), the preferred candidate leading condition group (b = -0.03, standardized b = -.02, t = -0.24, p = .813), or for the preferred candidate losing condition group (b =0.03 0.01, standardized b = .02, t = 0.28, p = .782). H2a was not supported.

H2b predicted that there would be a positive relationship between subjective knowledge and TPP_public. The result of our regression analysis showed that the effect of subjective knowledge on TPP_public was neither statistically significant for all participants (b = 0.05, standardized b = .03, t = 0.61, p = .544) nor for the preferred candidate losing condition group (b = -0.10, standardized b = -.07, t = -0.91, p = .359). H2b was not supported.

Effect of Others’ Political Knowledge (H3a & H3b)

To test H3a, we conducted a repeated-measures ANCOVA. PMI3_nonexpert and PMI3_expert were used as repeated measures (i.e., others’ political knowledge). Variables of question order (self-question first vs. other-question first), sex, subjective political knowledge, and election interest were included as covariate variables in the analysis. The results showed that PMI3_nonexpert (M = 5.63, SD = 1.53) was greater than PMI3_expert (M = 5.27, SD = 1.55, F(1, 239) = 6.38, η_p² < .03, p = .012; see Figure 1). H3a was supported.

Figure 1.Perceived Message Impact on Oneself and Others (PMI1 & PMI3, Panel A) and Third-Person Perception (TPP, Panel B) by the Target Group’s Political Knowledge and the Preferred Candidate’s Status in the Poll

To test H3b, we conducted a repeated-measures ANCOVA with TPP_nonexpert, TPP_expert as repeated measures. Statistically significant effects were found for question order, F(1, 505) = 4.37, η_p² = .01, p = .037, as well as the interaction of other-poll congruence with election interest, F(1, 505) = 5.03, η_p² = .01, p = .025. For the low-level election interest group, the results showed that TPP_ nonexpert (M = 1.39, SD = 2.18) was greater than TPP_ expert (M = 1.03, SD = 2.06), F(1, 239) = 6.38, η_p² = .03, p = .012. H3b was supported only for the low level of election interest group.

Discussion

Previous studies have shown that the third-person perception of polling effect is more evident among those who viewed election polls as undesirable or socially harmful compared to those who had a positive attitude toward them. However, previous studies tested the third-person perception of election poll news not in terms of a specific poll with a result but rather as a whole. Therefore, the way in which individuals perceived the impact of specific poll news and whether the third-person perception varied depending on the poll results (i.e., whether their preferred candidate is leading or losing) remained unexplored. Whereas previous studies on third-person perception focused on the social desirability of messages, the current study addressed the limitations of social desirability to explain perceptions of media impact and highlighted individuals’ personal desirability toward the poll result. When the poll result is personally undesirable (i.e., when the respondent’s preferred candidate was presented to be losing the race), it was expected that individuals would perceive a reduced impact of the poll news on oneself and a greater gap in perceived impact of poll news between on others and on oneself. Moreover, the key factors of TPP, including the respondents’ political knowledge and information about the target group’s political knowledge (expert vs. nonexpert), were examined.

Findings and Implications

The main findings of the study are as follows. First, when the respondents’ preferred candidate was leading in the poll news, they reported a higher PMI1 and PMI3 than when their preferred candidate was losing. As predicted, TPP was higher when the respondents’ preferred candidate was losing in the poll than when the candidate was leading. These results clearly indicate that individuals’ personal desirability about the poll results is a key factor that determines TPP. Specifically, when individuals read the desired poll news (i.e., when their preferred candidate is leading in the race), they are likely to accept it without a critical mind because they have no reason to deny the reinforcing effect of the poll on their preexisting attitudes. By contrast, when they are exposed to undesirable results (i.e., when their preferred candidate is losing the race), they are likely to react and critically process the counter-attitudinal information. They are then less likely to admit the influence of undesirable poll news (low PMI1), and TPP is greater for undesirable poll news. These findings have implications not only in research into the perceived polling effect but also the theory of the third-person effect. Recently, Shen et al. (2018) mentioned that “our understanding of how and why TPP occurs is still muddled” (p. 400). Our finding of the importance of personal desirability as a determinant of third-person perception is expected to improve understanding about why TPP occurs.

Second, we found that respondents considered the target group’s political knowledge when estimating PMI3 and TPP. This tendency was only found for those with relatively low interest in the election. Moreover, those respondents reported a greater PMI3 and TPP when the target was a nonexpert than when the target was an expert group. These findings also help improve understanding of what determines the perceptions of the impact of not only election polls but also various types of media messages. When we estimate the impact of the persuasive force of media messages, we consider the cognitive ability of the target person’s (either oneself or the other) to resist the force. However, in contrast to our expectation, respondents’ subjective level of political knowledge was found not to affect either PMI1 or TPP among those whose attitudes were incongruent with the poll results. In this study, we measured subjective level of knowledge specifically regarding the gubernatorial election (e.g., How much do you think you know about the procedures of the 2014 Gyeonggi gubernatorial election?). Previous studies found a negative effect of subjective knowledge on the impact of messages on oneself but those studies typically measured subjective knowledge in a broad manner (Salwen & Dupagne, 2001; e.g., respondents’ knowledge about the issue of television violence). Biased perception is likely to be more engaged when we respond to general questions than specific ones, and therefore, overestimation of respondents’ own subjective knowledge might be better measured with general questions.

Lastly, the current study found that PMI1 and PMI3 varied depending on which question was asked first. Specifically, PMI1 was found to be greater when PMI1 was asked first than when PMI3 was asked first. On the other hand, PMI3 was greater when PMI3 was asked first than when PMI1 was asked first. These findings can be explained by the anchoring effect (David & Johnson, 1998). When PMI3 was asked first, the response to the question about others sets the anchoring point. That is, respondents may try to find a point for their PMI1 below the anchored point, thus making the respondents in that condition (PMI3 first) report their PMI1 as being lower than it was in the other condition (PMI1 first). For PMI3, when PMI1 was asked first, the response to PMI1 may set an anchoring point, after which respondents might try to find a point for their PMI3 near PMI1 (i.e., insufficient adjustment). Because the present research did not test these explanations, future studies are encouraged to test the effect of question order on perceptions of media effect.

Limitations

The third-person effect hypothesis has two components: perceptual and behavioral components (Gunther et al., 2008; Perloff, 2009; Sun et al., 2008). The current study is limited to the perceptual component, as it did not investigate the behavioral component involved in polling news. According to the third-person effect hypothesis, the gap in perception between others and themselves regarding the effect of election poll news can affect respondents’ attitudes toward regulations regarding election poll release (H. J. Kim, 2013; Yang, 1997). Future studies may investigate the behavioral hypothesis of TPP of election poll news.

Another limitation of our study lies in the nature of the data. Previous studies did not test the effect of the preferred candidate’s status in the poll on the third-person effect. We formulated hypotheses regarding the problem and then tested those hypotheses using data that we collected in the context of a gubernatorial election in South Korea in 2014. Even though our hypotheses are not closely tied to a specific election context (i.e., time, nation), there is a concern about data outdatedness. However, the pattern of results found in this study is consistent with findings of recent TPP studies. For example, data from this study consistently indicate that the third-person perception of election polling news is associated with voters’ desirability of poll results (i.e., whether their preferred candidate is leading in the race; Chung et al., 2018; Farjam, 2021; Gattermann et al., 2022; H. Kim, 2022). In addition, the present study has also shown that while voters tend to consider desirable news (i.e., their preferred candidate is leading) as unbiased, they are more likely to perceive undesirable news as biased (Jang & Kim, 2018; H. Kim, 2015; Perloff & Shen, 2022). However, it would be required to test hypotheses in different election contexts to extend the generalizability of the findings and to increase the confidence in the findings. Future studies are also encouraged to replicate the findings with more diverse and larger samples.

Perceptual biases may be involved in third-person perception. The gap between PMI3 and PMI1 could result from either overestimating election poll news on others or underestimating it on oneself (Douglas & Sutton, 2004; Perloff, 1999; Price et al., 1998). An important question is which bias (overestimation on others vs. underestimation on oneself) is the main cause of the third-person perception of election poll news. Our study has discovered that individuals perceive the influence of poll coverage differently based on the results of the polls. To advance the research, it is necessary to move beyond the discovery of effects and delve into an exploration of the process (Ewoldsen, 2022). We hope that future studies will not only uncover the causes of third-person perception but also shed light on the underlying mechanisms and psychological processes.

Funding

This work was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF-2021S1A5A2A03068831).