Hunting for Voters
How political ads targeting voters’ personality traits affect ads’
persuasiveness and source credibility.
Lisa Bringhenti
Student number: 11817712
Master’s Thesis
Graduate School of Communication
Master’s Programme Communication Science
Political Communication Track
Supervisor: Bert Bakker
Date of Completion: 1
stFebruary 2019
2
Abstract
Political campaigns have increasingly become more personalized, as politicians try as hard as they can to target the desires of specific groups of voters. One way to do this, is by tailoring their messages to the personality of their voters. This study assesses how the personalization of political campaign ads tailored to people’s personality traits affect the persuasiveness and credibility of the ads and of the politicians sponsoring them. On one hand, the study theorizes that the persuasiveness and credibility of political ads and their sources increase as the ads get more congruent with the voters’ personality. On the other hand, a backfiring effect is expected: when the voters become aware of the message being personalized, the persuasiveness and credibility of the ad and its source decrease. Finally, the positive effect of the congruency of the ad is expected to be moderated by the negative effect of the awareness of the personalization. An online experiment (N = 327) was
conducted to investigate how tailoring political ads to people’s personality traits of Agreeableness and Extroversion, affects the persuasiveness of such ads and the credibility of their source. Findings show strong evidence supporting the congruency hypothesis. The awareness hypothesis, instead, is not supported, meaning that there is no evidence of people’s scepticism towards personalized messages. Finally, some suggestive evidence shows that the difference in the perceived
persuasiveness of congruent and incongruent messages is slightly larger for people who are aware of the personalization of the message than for people who are not. Such findings clearly
demonstrate the efficacy of messages tailored to voters’ personality traits. Employing such microtargeting techniques proves to represent an efficient way to gain voters’ support, but also a danger for democracy and the public discourse.
Introduction
Referring to the increasing popular trend of the use of political microtargeting in political
campaigns, Hal Malchow, one of the advisors and consultants to the US Democratic party, stated that “people want information, they do not want advertising. When they see our fingerprints on this
3
stuff, they believe it less.” (Tufekci, 2004, p.11). Today, we are undoubtedly living in the ‘big data’ age, where the amount of information we make available about ourselves is growing exponentially. Consequently, political parties are becoming increasingly skilled at analysing this data to predict electoral behaviour. For instance, the United States’ spending for digital political advertising grew from 212 to 1,878 million US$ in the last three years and is expected to reach even higher levels in the near future (Statista, 2018). Indeed, the last US presidential election saw both Republican and Liberal parties exploiting huge amounts of digital data to shape their campaigning decisions (The Economist, 2016). Also, the related scandal of the political consulting firm, Cambridge Analytica, played a pivotal role in Donald Trump’s victory. By illegally collecting Facebook users’ data they generated personality profiles of millions of individual voters, which were used to send narrowly targeted political advertisements (Gonzalez, 2017). This clearly demonstrated both the power and danger of the employment of big data in politics (The Guardian, 2018). In a political era
characterized by the increasing personalization of political campaigns (Bennett, 2012) political microtargeting becomes an extremely relevant topic in nowadays’ society.
We refer to political microtargeting as the act of finding and combining information about individuals to enable political actors to efficiently reach potential voters at the individual level at relatively low costs (Kruikemeier, Sezgin, & Boerman, 2016). Previous studies mostly focused on the way political microtargeting can help campaigns reach out to voters with the most appropriate messages (Barocas, 2012). Instead, despite political microtargeting being increasingly employed during political campaigns, little research has focused on the negative aspects of such techniques. Indeed, a major gap in research on election advertising exists in understanding the psychology of voters in response to personalized election ads. Most voters are still in the dark about the nature and extent of political microtargeting (Rubinstein, 2014). However, if they knew, the exploitation of their personal information for political microtargeting would likely be perceived as invasive and thus evoke resistance (Kruikemeier et al.,2016).
4
This study aims at analysing both the benefits and the harms of the use of voters’ personality traits in political microtargeting campaigns. On one hand, a message which is congruent with the voter’s personality is expected to be more persuasive than an incongruent one. On the other hand, when the voter becomes aware of the message being personalized, he might perceive it as fake and forged, and thus less credible. Hence, this study aims at answering the following research question: To what extent does the personalization of political campaign ads, tailored to people’s personality traits, affect the persuasiveness and credibility of the ads and of the politicians sponsoring them? Understanding how the personalization of political messages is affecting citizens’ attitudes could contribute to the democratization of the political process by allowing citizens to have more control and responsibility over the complex political environment. Furthermore, it may be relevant for political scientists, who are interested in the analysis of the interaction between personality traits, political campaigning, and voters’ decision-making (Caprara & Zimbardo, 2004). Indeed, the normative implications of the use of personality traits as means to target voters during political campaigns are of fundamental significance and will be further addressed in the final sections. This study begins with an outline of the theoretical background of the use of microtargeting in politics, of the concepts of personality traits and of the notion of credibility. The following sections will focus on the methods and results of a survey-embedded experiment and will end with a
discussion of results, limitations and normative implications of the study. Theoretical framework
Political microtargeting
Voter microtargeting is a relatively new form of political marketing in which political actors target personalized messages to individual voters by applying predictive modelling techniques to massive troves of voter data aiming at activating the base and persuading undecided voters (Rubinstein, 2014). Today, online microtargeting and the use of personalized ads have become an important
5
campaign strategy during election times in the United States as well as in Europe (Borgesius et al., 2018). In the US, several companies, such as Cambridge Analytica, offer online microtargeting services to politicians, enabling them to target people with ads on the web. For instance, Cambridge Analytica claims to have collected data by surveying hundreds of thousands of people to determine their psychographic profiles – measured through the Big Five personality traits - and to predict what message would most likely persuade them (Gonzalez, 2017).
Indeed, the contemporary information environment has made it easier to target messages to narrow segments of the population, allowing candidates to increasingly micro-target their voters. By narrowly communicating issue messages, candidates reduce the risk of alienating other voters, thereby broadening the range of issues on the campaign agenda (Schipper & Woo, 2012). In fact, in a campaign message that will reach all voters, candidates prefer to make broad appeals that could alienate the smallest share of the population. Instead, when messages can be hidden from all but the intended recipients, they might craft more pointed and effective messages (Hersch & Schaffner, 2013). Therefore, we can say that voter microtargeting enables campaigns to allocate their field resources very efficiently (Rubinstein, 2014).
Nevertheless, personalized online advertising can have a great variety of influences on people. Some studies note that people perceive personalized ads as useful because they reduce information overload, serve users’ needs, and provide aids for decision-making (Kruikemeier et al., 2016). Other studies have shown that targeted appeals can increase electoral turnout (Green & Gerber, 2008). Indeed, microtargeting is able to reach citizens who are difficult to reach through mass media, such as politically uninterested voters and people opting out of traditional media exposure (Borgesius et al., 2018). This study will focus upon the psychological persuasion approach (Matz, Kosinski, Nave, & Stillwell, 2017), according to which tailored messages are found to be considerably more effective than normal campaigns, and the effectiveness of tailoring increases with greater
6
On the other side, this form of persuasion could also be used to exploit weaknesses in people’s character and persuade them to take actions against their own best interest (Matz et al., 2017). In politics the positive consequences of microtargeting seem to be outnumbered by the negative ones. Personalized ads are found to lead to lower support for politicians, lower engagement in political behaviour, negative attitudes, lower source trustworthiness and more ad scepticism (Kruikemeier et al. 2016). Previous studies even discuss a chilling of political participation due to the perceived violations of voters’ privacy caused by microtargeting (Barocas, 2012), as well as potential harms to civic discourse (Tufekci, 2014). For instance, Tufekci (2014) argues that by targeting the specific wishes of potential voters broadly relevant political topics are often excluded or left ambiguous during campaigns. Similarly, many of the claims and promises made by the candidates through targeting messages remain hidden from the non-targeted majority of the public (Hersch & Schaffner, 2013), hence showing diverse images of the same candidate according to differently targeted voters.
Moreover, people perceive personalized ads as an invasion of their privacy, which evokes their resistance towards the targeted message and its source. In her paper on voters’ privacy in the digital era, Rubinstein (2014) exemplifies this by citing a 2012 national telephone survey which found that 86% of adult Americans do not want political campaigns to tailor advertisements to their interests and that 64% of voters would be less likely to vote for a candidate who engages in the practices of microtargeting. Similarly, in their experimental study, Hersch and Schaffner (2013) found that voters rarely prefer targeted messages to general ones and that ‘mistargeted’ voters penalize candidates enough to erase the positive returns of microtargeting.
Therefore, drawing both on the promises and threats of microtargeting, this paper aims at
investigating whether the personalization of a political message increases its persuasiveness and/or whether it has a backfiring effect on the credibility of its source, due to people’s perception of the customization of the message as an invasion of their privacy.
7
Personality traits as moderators of personalized communication
The present study focuses upon personality as a factor to which political ads can be tailored. In the political domain, previous studies (e.g. Caprara, Barbaranelli, & Zimbardo, 2002; Bakker,
Hopmann, & Persson, 2015; Matz et al., 2017) employed the ‘Big Five’ model as a measure of personality traits. The model builds upon five traits, namely Extroversion, Agreeableness, Conscientiousness, Emotional Stability and Openness to Experience (Gerber, Huber, Doherty, & Dowling, 2011). The popularity of such measure lies in its ability to encapsulate a wide variety of individual differences, and since it develops in early childhood, it is often considered a relatively stable measure over time (Bakker, Rooduijn, & Schumacher, 2016; Dinesen, Norgaard, &
Klemmensen, 2013). It represents an excellent starting point for the inclusion of personality traits in models of political and social behaviour, also thanks to its reliability across cultures, contexts and measurement instruments (Mondak & Halperin, 2008).
In particular, this study focuses upon two of the Big Five personality traits, namely Agreeableness and Extroversion. Indeed, the personality psychology literature (e.g. Amsalem,Zoizner, Sheafer, Walgrave, & Loewen, 2018) conceptualizes interpersonal behaviours according to two major axes, which in case of the Big Five models are represented by Extroversion (enthusiasm and
assertiveness) and Agreeableness (politeness and compassion). Also, previous studies on personality traits found that Extroversion and Agreeableness are primary anchors for evaluating politicians (Caprara et al., 2002). For instance, Caprara, Barbaranelli, Consiglio, Picconi and
Zimbardo (2003) found that only Extroversion and Agreeableness had significant results in terms of congruency between politicians’ and voters’ personality traits. Moreover, since these two traits largely affect interpersonal relations and people’s perceptions (Amsalem et al., 2018) they can also be influential when it comes to persuasion. Indeed, targeting voters with messages tailored to their agreeableness and extroversion is likely to increase the persuasiveness of such messages.
8
Moving to a more in-depth description of the two personality traits, Agreeableness is related to a person’s social disposition and refers to cooperative and altruistic tendencies (Mondak & Halperin, 2008). For instance, individuals who score high on agreeableness like other people, are usually liked by others, are cooperative, sympathetic, pleasant, and harmonious in their relationships, whereas people with little agreeableness are competitive, suspicious, conflicting, and manipulative (Amsalem et al., 2018; Dinesen, 2013; Matz et al., 2017). This personality trait is also linked to political trust, since people with high levels of agreeableness are usually exhibiting high levels of trust (Mondak & Halperin, 2008). In the present study, messages tailored to agreeable individuals will therefore value communal goals, interpersonal harmony, trust, and likeability, whereas those tailored to non-agreeable ones will value competition, conflict, and opposition.
Extroversion, instead, is the most widely discussed, researched and cited personality trait and has been defined with alternative names, such as ‘sociability’, ‘activity’ (Mondak & Halperin, 2008) or ‘energy’ (Caprara & Zimbardo, 2004). Extroverted people are outgoing, sociable, energetic, and excitement-seeking, whereas introverted people are reserved, quiet, and inhibited in their
relationships (Amsalem et al., 2018; Dinesen, 2013; Matz et al., 2017). Links between extroversion and political behaviour are very common, since extroversion has been found to be related to aspects of group-based political participation and to opinionation (Mondak & Halperin, 2008). In this study, messages tailored to extroverted individuals will therefore be energetic, outgoing, and seeking attention, contrary to those tailored to introverted ones.
Aside from being associated with political attitudes and behaviours (Dinesen et al., 2013; Gerber, Huber, Doherty, Dowling, & Panagopoulos, 2013), the Big Five can also be used to predict political persuasion. Indeed, recent research (e.g. Hirsch et al., 2012; Matz et al., 2017) found that the use of personality traits as means to target people with personalized messages can make such messages more persuasive for the targeted public. For instance, in their study on personalized persuasion, Hirsh et al. (2012) constructed five advertisements, each designed to target one of the five major
9
traits. Each advertisement showed the same picture, but each text highlighted a different
motivational concern associated with one of the Big Five. By analysing the respondents’ attitudes towards the advertisements, they found that the ads were evaluated more positively the more congruent they were with participants’ personality (Hirsch et al., 2012).
This psychological persuasion approach (Matz et al., 2017) has been associated to the context of the personalization of politics, the trend that recently saw individual characteristics of both voters and candidates assume greater importance in the political discourse. In this political system, and according to the so-called congruency model of political preference, people vote for candidates whose personality traits match their own traits (Caprara & Zimbardo, 2004). In other words, the more apparent the similarity with a political candidate, the more positive the voter’s perception of that candidate. Therefore, drawing both on the congruency model of political preference and on the psychological persuasion approach, it can be assumed that the more congruent a message is with the voter’s personality, the more persuasive it is perceived. Hence, this study aims at proving that political messages tailored to and congruent with the voters’ personality traits scores of agreeableness and extroversion are perceived as more persuasive than incongruent messages: H1 (or congruency hypothesis): Exposing people to a personalized political campaign ad, which is congruent with their own personality traits’ scores of agreeableness and extroversion, will lead them to perceive the ad as more persuasive than people exposed to an ad which is incongruent with their own scores of agreeableness and extroversion.
Perceived credibility of the source
As stated in its research question, this study aims at investigating the effects of personalized
messages on ads’ persuasiveness and source credibility. While an ad’s persuasiveness can be easily rated by its reader judging its content and style, source credibility is a more complex concept and needs to be assessed more thoroughly. Source credibility (or ethos) refers to the attitude towards a source of communication held at a given time by a receiver (McCroskey & Young, 1981).The
10
concept of source credibility has been analysed in the political communication literature and has been found to increase a message’s persuasiveness (Hovland & Weiss, 1951), to develop favourable attitudes and to influence vote intentions (Mondak, 1995). Also, in political campaigns, higher source credibility predicts more positive global candidate evaluations and higher assessments of leadership ability (Householder & LaMarre, 2014).
In this concern, research offers only a few empirical studies on individuals’ evaluations of political candidates’ source credibility in the digital era. Recently, due to the increasing use of social media during political campaigns, questions regarding the reliability, trustworthiness and credibility of sources have been rising in the electorate. However, most of the studies have been focusing on the credibility of media channels and little is known about the mechanisms underlying people’s
judgement of politicians’ credibility in online environments (Householder & LaMarre, 2014). Since source credibility has been found to increase the persuasiveness of a message (Hovland & Weiss, 1951), it becomes an extremely relevant concept in this study. Analysing whether the
personalization of political messages affects their persuasiveness and source credibility is of fundamental importance for research on personalized communication and political microtargeting. Most studies on source credibility have been conceptualizing it in two main dimensions, namely trustworthiness (or character) and expertise (or competence), whereas others have added a few dimensions, such as goodwill (Householder & LaMarre, 2014), attractiveness (Ohanian, 1990), dynamism and objectivity (Whitehead, 1968). Based on the characteristics that are most commonly considered in evaluating political candidates, the present study considers three specific dimensions of the theoretical construct of source credibility, namely trustworthiness, expertise and goodwill (Teven & McCroskey, 1997). Trustworthiness refers to voters’ perceptions of the politician’s
assertions as valid; expertise refers to the knowledge and competence of the politician; and goodwill refers to the degree to which voters perceive that the politician cares about their well-being
11
This study will investigate to what extent source credibility is affected by the personalization of political messages. As previously mentioned in the section on political microtargeting, extant research found that people perceive personalized messages as biased and invasive and that this may lead to lower support for politicians, negative attitudes, lower source trustworthiness and more ad scepticism (Kruikemeier et al. 2016; Barocas, 2012). People are usually unaware of being targeted during political campaigns and, if they knew, they might avoid campaign information and be resistant and suspicious towards the message and its source (Kruikemeier et al., 2016). Translating such reasoning to the concept of credibility, instead of increasing the persuasiveness of the message, we expect the personalization of the message to originate a backfiring effect:
H2 (or awareness hypothesis): The exposure to a personalized political campaign ad will lead people, who are made aware of the personalization of the ad, to perceive the candidate sponsoring the ad as less credible than people who are not made aware of its personalization.
No previous research investigated the relationship between the opposing effects theorised by the first two hypotheses. Therefore, this paper aims also at examining the interaction existing between the positive effect of the congruency of the message on its persuasiveness and the negative effect of the awareness of the personalization on the source credibility.
Drawing on the first hypothesis, exposing someone to an ad which is congruent with his/her
personality will very likely persuade him/her more than someone who is exposed to an incongruent ad. Therefore, a person in the condition of congruency and who is not aware of the ad being
personalized would likely perceive the message as very persuasive, because he/she would identify him/herself in that message. However, drawing on the second hypothesis, if that person would become aware of the ad being personalized and tailored to his/her personality, this could instil a sense of concern for his/her privacy and evoke resistance towards the ad. Therefore, a person in the condition of congruency of the message and awareness of the personalization would likely perceive the message as less persuasive than the person being unaware.
12
Considering the condition of incongruence of the ad, Hersch and Schaffner’s experimental study (2013) found that ‘mistargeted’ voters penalize candidates and erase the positive returns of microtargeting. This means that people exposed to an incongruent ad will not identify themselves with that ad and will therefore not be very persuaded by it. The negative effect of the incongruence of the ad would be even stronger if people were made aware of the personalization. Indeed, these people would likely await a congruent message. Being exposed to an incongruent ad when expecting a congruent one, would not only instil a sense of concern for their privacy due to the awareness of the personalization, but would also make them very confused about the dissimilarity of the content of the ad relatively to their personality. Therefore, people in the condition of
incongruence of the message and awareness of the personalization would very likely perceive the message as not persuasive at all.
Since the two dependent variables of this study, the persuasiveness of the message and the source credibility, have been previously found to be associated (Hovland & Weiss, 1951), we extend the arguments made on the concept of persuasiveness to the concept of source credibility too.
Therefore, the interaction hypothesis states the following:
H3 (or interaction hypothesis): The positive effect of the congruency of a political ad with the voters’ personality on the persuasiveness of the ad and on the credibility of its source will be moderated by the negative effect of the awareness of the personalization.
Method
Sample
Participants were recruited through a convenience sample, because of the need to collect easily accessible respondents in a quick and inexpensive way. Due to the need of maximizing the amount of respondents, snowballing sampling was employed, meaning that some participants were asked to recruit other participants. This also meant that a large part of the sample was going to come from
13
the same country as the researcher’s background, namely Italy. Therefore, the survey was created in English, to include European participants, and translated into Italian, in order to render its
completion easier for Italian respondents. Participants whose mother tongue was Italian were
assigned to the Italian version of the questionnaire, whereas all other participants to the English one. A power analysis executed with the G*power software (Faul, Erdfelder, Lang, & Buchner, 2007) was performed to illustrate the minimum sample size. Previous literature on political microtargeting and on personality traits (e.g. Kruikemeier et al., 2016; Gerber et al., 2011) manifested the
expectation of small to medium effect sizes. In this study, the plan was to create four different treatment conditions and employ both regression analyses and analyses of variance. The power analysis, assuming four conditions, an alpha of .05, and beta of 0.95, illustrated the need of a minimum sample size of 275 participants.
The study was conducted online and respondents were invited to participate through a link both posted on Facebook and sent via WhatsApp, email or Facebook Messenger. The survey was fielded between November 23rd and December 8th 2018 and was completed by 395 respondents. Since the
study was carried out among European citizens and in the foresight of the 2019 European Parliament elections, participants were included only if more than 18 years old and if European citizens. One underage participant was excluded from the analysis, as well as 67 other participants who either did not give their informed consent or dropped out from the survey before being exposed to the experimental manipulation. The final sample was N = 327, which rendered this a well
powered study.
The sample consisted of a diverse group of people in terms of the respondents’ age (M = 37.12, SD = 15.70, Min = 18, Max = 82). The majority of the sample was female (63.6%) and Italian (77.4%). The sample was slightly left-leaning on its political self-placement, which was measured on a scale from 1 (left) to 10 (right) (M = 4.38, SD = 2.08, Min = 1, Max = 10) and had rather pro-European
14
Union attitudes, which were measured on a scale from 1 (against the EU) to 10 (pro EU) (M = 8.04, SD = 2.05, Min = 1, Max = 10).
Design
Experiments
The present study builds on an experimental approach embedded with a survey. Two experiments were conducted in the same survey in order to separately investigate the personality traits of
Extroversion first and Agreeableness after. An experimental design allowed us to have control over the independent variables: the congruency of the message and the awareness of the personalization. By randomly varying these, this study can establish the causal effects of congruency and awareness on the persuasiveness of the message and the credibility of its source. The controlled environment of the experiment also enabled us to exclude the effect of external factors on the dependent
variables (Neuman, 2012).
The design employed for both experiments was a 2 (personalization: high and low personality traits) x 2 (awareness of the personalization: yes, no) between-subjects experimental factorial design. Indeed, in both experiments, participants were exposed to one of four different stimuli, varying both on the awareness of the personalization and on the personality traits incorporated in the messages. The stimuli were created for the purpose of the study and displayed a tweet sponsored by a fictitious European political candidate. In all conditions the tweets varied only in terms of the message, whereas the candidate’s name and image did not change. The Italian version of the stimuli differed from the English one only in terms of the name of the political candidate, which was chosen to be Italian to render the tweets’ artificiality less evident. For each experiment, two different messages were created in order to portrait the same candidate with varying personalities, namely an
extroverted one and an introverted one for the Extroversion experiment, and a low-agreeable one and a high-agreeable one for the Agreeableness experiment.
15
Measuring the participants’ scores on extroversion and agreeableness and then comparing these to the personality traits incorporated in the tweets they were exposed to, enabled us to calculate the congruency of the tweets with the respondents’ personality. The congruency between the
personality traits of the respondents and those incorporated in the tweets was later employed as an independent variable in our analyses.
Concerning the content of the tweets, the extroverted message combined positive emotions,
leadership and an out-going tone, whereas the introverted one presented a rather quiet and reserved political candidate (see Figure 1). The high agreeable message combined the elements of
cooperation, trust and sympathy, whereas the low agreeable one portrayed a competitive, suspicious and untrustworthy candidate (see Figure 2). As previously mentioned in the theory section of the study, such elements are in line with the description of behaviours and features of extroversion and agreeableness in the previous literature on personality traits.
Instead, the awareness of the personalization was explicated by telling the participants that the tweet they were going to read was tailored to their personality and by adding the label ‘Promoted’ under the tweets’ messages.
The messages were pre-tested in order to evaluate if respondents correctly perceived the personality traits embedded in the messages. A sample of 15 people was asked to evaluate the extroversion and agreeableness of the tweets on a scale from 1 to 7. Results showed that respondents were better able to distinguish an introverted message (M = 2.62, SD = 1.12) from an extroverted one (M = 5.85, SD = .55) than to distinguish a low agreeable message (M = 2.62, SD = 1.56) from a high agreeable one (M = 4.85, SD = 1.28). Therefore, the tweets varying on the level of agreeableness were slightly changed after the pilot-test, in order to incorporate more extreme messages (see Appendix A for both English and Italian tweets).
16
Figure 1 - Tweets portraying extroverted (above) and introverted candidate (below).
17
Procedure of the study
The survey started by investigating participants’ demographics and political preference, as well as their European attitude. Next, their personality traits of extroversion and agreeableness were measured with the valid and reliable but brief (ten items per trait) IPIP-NEO scale (Goldberg, 1999). Following, participants were exposed to the treatment condition of the Extroversion experiment. Respondents were then asked about their perceptions of the persuasiveness of the message they were shown and of the credibility of its source. After being exposed to a manipulation check, participants were shown the treatment condition of the Agreeableness experiment. Again, their perceptions of persuasiveness and credibility were investigated. Participants were then exposed to the second manipulation check and were finally thanked for their participation and debriefed about the purpose of the study.
Measures
The respondents’ Extroversion was measured using the ten items of the Neo-IPIP scale (e.g. ‘I make friends easily.’ see Appendix B for the other items), each answered on a 7-point Likert scale ranging from “strongly disagree” (1) to “strongly agree” (7). The negatively phrased items were reversed, and an exploratory principal axis factor analysis was computed by extracting only one factor (eigenvalue = 3.87) explaining 38.7% of the total variance in the items. The extroversion scale was created by summing the scores of the items and proved to be internally consistent (Cronbach’s alpha = .82). Values theoretically ranged from 10 to 70, where small values
represented low extroversion and large values represented high extroversion (M = 45.57, SD = 8.86, Min = 23, Max = 68).
Participants’ Agreeableness was also measured using ten items of the Neo-IPIP scale (e.g. ‘I respect others.’, see Appendix B for the other items). The negatively phrased items were reversed, and a factor analysis was performed by extracting a single factor (eigenvalue = 2.56) explaining 25.6% of the total variance in the items. The agreeableness scale was computed by summing the scores of the
18
items and proved to be internally consistent (Cronbach’s alpha = .65). Values theoretically ranged from 10 to 70, where small values represented low agreeableness and large values represented high agreeableness (M = 53.19, SD = 6.36, Min = 35, Max = 70).
Independent variables
As briefly mentioned in the design section, in order to assess the congruency of the message shown in the experimental condition with the actual personality of the respondent, an index of the
congruency of the message was created. This was done by measuring the participants’ scores on extroversion and agreeableness and then comparing these to the extroversion and agreeableness scores incorporated in the tweets they were exposed to. For instance, a new dichotomous variable was created, indicating which message the respondent was exposed to. To render this variable comparable to the above-described agreeableness and extroversion variables, the value 70 was chosen to indicate the exposure to a high agreeable message whereas the value 10 indicated the exposure to a low agreeable message. The same was done for the introverted (value 10) and
extroverted messages (value 70). An incongruence index was created by subtracting the personality variable (theoretically ranging from 10 to 70) from the variable indicating the exposure to the message (presenting values 10 and 70). For instance, a respondent scoring high on agreeableness (70) who was exposed to a low agreeable message (10), presented high values of incongruence (70 – 10 = 60). To make the data more understandable, the index was then reversed, by subtracting the incongruence value to 60 (i.e. the difference between the highest and lowest values of the index). In this way congruency values theoretically ranged from 0 to 60, where large values represented higher congruency and small values lower congruency. In the extroversion experiment participants were exposed to an average congruency of M = 29.06 (SD = 10.42, Min = 5.00, Max = 58.00), whereas in the agreeableness experiment the average congruency was M = 30.06 (SD = 14.67, Min = 2.00, Max = 60.00).
19
Finally, the awareness of the personalization was measured by looking at whether participants were made aware of the personalization of the message during the treatment conditions. This
dichotomous variable displayed the value 1 in case of aware respondents and the value 0 in case of unaware respondents.
Dependent variables
The persuasiveness of the message was measured with a 7-point agreement scale ranging from “strongly disagree” to “strongly agree” and asking the respondent “How persuasive do you think was the message of the tweet you saw?” (adapted from Hirsch et al., 2012). A new variable was then computed by transforming the 7-point scale into values ranging from 0 (not persuasive) to 1 (very persuasive). The average persuasiveness of the message was M = .37 (SD = .27, Min = .00, Max = 1.00) in the extroversion experiment, and M = .34 (SD = .27, Min = .00, Max = 1.00) in the agreeableness experiment.
Lastly, the source credibility was measured using the fourteen items of the credibility scale by Teven and McCroskey (1997), each answered on a 7-point Likert scale ranging from “strongly disagree” to “strongly agree” (e.g. ‘I think the politician sponsoring the ad is intelligent’, see Appendix B for the other items). No item needed to be reversed. An exploratory principal axis factor analysis was computed for both experiments. Only one factor (eigenvalue credibility extroversion experiment = 8.15, eigenvalue credibility agreeableness experiment = 9.11) was extracted, which explained 58.2% of the total variance in the fourteen items in the extroversion experiment and 65.1% in the agreeableness experiment. The source credibility scale was created by computing the mean scores of the items and transforming them into values ranging from 0 to 1. The scale also proved to be internally consistent (extroversion experiment: Cronbach’s alpha = .94; agreeableness experiment: Cronbach’s alpha = .96). In the extroversion experiment the average source credibility was M = .50 (SD = .17, Min = .00, Max = .89), whereas in the agreeableness experiment M = .40 (SD = .19, Min = .00, Max = .92).
20
Results
Randomization and manipulation checks were performed and showed that participants were correctly randomized across conditions and that the manipulation was perceived as intended (see Appendix C for complete report of results).
To test the effects of the congruency of the message and the awareness of the personalization on both the persuasiveness of the message and the source credibility two linear regressions were computed. A first linear regression was computed with the persuasiveness of the message as dependent variable and the congruency of the message and the awareness of the personalization as predictors (see Table 1). The second linear regression was computed with the source credibility as dependent variable and the congruency of the message and the awareness of the personalization as predictors (see Table 2)1.
Table 1 - Linear regression models predicting the ‘persuasiveness of the message’. Persuasiveness of the message Extroversion model
b*
Agreeableness model b*
Constant .18*** .16***
Congruency of the message .20*** .30***
Awareness of the personalization .14* .03
R² .05 .08
F 9.80*** 15.75***
N 327 327
Note: * p < .05. ** p < .01. *** p < .001.
1 Robustness check: By decomposing the variable of source credibility into its three clusters – competence,
trustworthiness and goodwill – and employing them as dependent variables in a regression model with the congruency of the message and the awareness of the personalization as predictors, results showed no difference relatively to the analyses computed with the variable of source credibility as a whole (see Appendix C).
21
Table 2 - Linear regression models predicting the ‘source credibility’. Source credibility Extroversion model b* Agreeableness model b* Constant .48*** .25***
Congruency of the message -.04 .39***
Awareness of the personalization .20*** .03
R² .03 .15
F 6.77*** 28.90***
N 327 327
Note: * p < .05. ** p < .01. *** p < .001.
In the remainder of this results section, results will be presented hypothesis by hypothesis. Before that, the fit of the models will be briefly observed. When it comes to the explained variance, the model predicting the persuasiveness of the message in the Extroversion experiment explains 5% of the variance and the model predicting the source credibility explains 3%. Concerning the
Agreeableness experiment, instead, the model predicting the persuasiveness of the message explains 8% of the variance and the one predicting the source credibility predicts 15%.
Coming to the hypotheses testing, results showed that the congruency hypothesis (H1) was supported in both experiments. Concerning the extroversion experiment, as stated in the first
hypothesis, the congruency of the message has a significant association with its persuasiveness, b = .005, b* = .20, t = 3.63, p < .001 (see Table 1, column 1). Per unit of increase in the congruency of the message, its persuasiveness improves by 0.5%, when keeping the other variable constant. The congruency of the message, instead, was found not to be a significant predictor of the source credibility, b = -.00, b* = -.04, t = -.71, p = .477 (see Table 2, column 1). Also in the agreeableness experiment, the congruency of the message is a significant predictor of its persuasiveness, b = .006, b* = .30, t = 5.59, p < .001 (see Table 1, column 2). Hence, per unit of increase in the congruency of the message, its persuasiveness improves by 0.6%, keeping the other variable constant. In this
22
case the congruency of the message was a significant predictor of the source credibility too, b = .005, b* = .39, t = 7.59, p < .001 (see Table 2, column 2), meaning that per unit of increase in the congruency of the message, the credibility of the source improves by 0.5%, when keeping the other variable constant.
Moving to the awareness hypothesis (H2), results of the Extroversion experiment showed that the awareness of the personalization is a significant predictor of the source credibility, b = .07, b* = .20, t = 3.61, p < .001 (see Table 2, column 1). However, contrary to what was expected in the awareness hypothesis, the positive sign of the prediction suggests that the awareness of the personalization of the message increases the source credibility. Moreover, the awareness of the personalization was found to be a significant positive predictor of the persuasiveness of the
message, b = .07, b* = .14, t = 2.52, p < .05 (see Table 1, column 1). Concerning the Agreeableness experiment, results were rather different and showed that the awareness of the personalization is not a significant predictor of the source credibility, b = .01, b* = .03, t = .59, p = .556 (see Table 2, column 2) and neither of the persuasiveness of the message, b = .02, b* = .03, t = .58, p = .562 (see Table 1, column 2). Hence, the second hypothesis cannot be supported.
Finally, to test the interaction hypothesis - which stated that the positive effect of the congruency of the message is moderated by the negative effect of the awareness of the personalization - two univariate analyses of variance were conducted with the congruency of the message and the awareness of the personalization as independent variables. The first univariate ANOVA was computed on the dependent variable source credibility (see Table 3) and the second one on the persuasiveness of the message (see Table 4). In order to be used as a categorical variable in the analyses of variance, the interval variable of the congruency of the message was recoded into a dichotomous one by computing a median split (incongruent = 0, congruent = 1).
In the Extroversion experiment, results of the univariate analysis of variance with the source credibility as a dependent variable showed no significant interaction effect between the awareness
23
of the personalization and the congruency of the message on the source credibility, F(1,325) = .10, p = .751. The same applies to the Agreeableness experiment, F(1,325) = .00, p = .989 (see Table 3). Table 3 - Results of a univariate analysis of variance on the ‘source credibility’.
Sum of Squares df Mean Square F p η2
Extroversion experiment Awareness .38 1 .38 13.46 < .001 .04 Congruency .03 1 .03 1.12 .290 Interaction .00 1 .00 .10 .751 Error 9.00 323 .03 Total 9.39 326 Agreeableness experiment Awareness .00 1 .00 .18 .668 Congruency 2.18 1 2.18 71.66 <.001 .17 Interaction .00 1 .00 .00 .989 Error 9.84 323 .03 Total 12.04 326
Results of the univariate ANOVA with the persuasiveness of the message as dependent variable showed no significant interaction effect in the Agreeableness experiment, F(1,325) = .07, p = .799 nor in the Extroversion experiment, F(1,325) = .1.69, p = .195 (see Table 4). However, in the Extroversion experiment’s case, the interaction plot (see Figure 3) displays how the difference in the persuasiveness of congruent and incongruent messages is stronger in case of participants who were aware of the personalization of the message. Indeed, in the awareness condition, the difference in persuasion between congruent (M = .46, SD = .26) and incongruent messages (M = .33, SD = .26) is larger than the difference in persuasion between congruent (M = .36, SD = .25) and incongruent messages (M = .30, SD = .26) in the unawareness condition. This provides suggestive evidence for the interaction hypothesis, although results are not statistically significant, possibly due to a too small sample size.
24
Table 4 - Results of a univariate analysis of variance on the ‘persuasiveness of the message’. Sum of Squares df Mean Square F p η2
Extroversion experiment Awareness .37 1 .37 5.54 < .05 .02 Congruency .72 1 .72 10.68 < .001 .03 Interaction .11 1 .11 1.69 .195 .00 Error 21.76 323 .07 Total 23.02 326 Agreeableness experiment Awareness .01 1 .01 .20 .654 Congruency 2.39 1 2.39 35.37 <.001 .09 Interaction .00 1 .00 .07 .799 Error 21.78 323 .07 Total 24.19 326
Figure 3 - Interaction effect between the ‘awareness of the personalization’ and the ‘congruency of the message’ on the ‘persuasiveness of the message’.
Conclusion and Discussion
Findings of the present experiment showed strong evidence supporting the congruency hypothesis, meaning that political ads which are congruent with the voters’ personality are perceived as more
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 Congruent Incongruent
Persuasion of the message (%)
25
persuasive and credible than those which are incongruent. Such evidence is in line with Caprara and Zimbardo’s congruency model of political preference (Caprara & Zimbardo, 2004) and with the psychological persuasion approach (Hirsch et al., 2012; Matz et al., 2017). Indeed, it confirms that when politicians tailor their campaigns to their potential voters’ personality, the persuasiveness of their ads increases. Hence, voter microtargeting enables campaigns to allocate their field resources more efficiently than traditional campaigning, since it allows politicians to send out customized and persuasive messages.
The awareness hypothesis, instead, was not supported. Indeed, being aware of the personalization of the message did not have any backfiring effect on the credibility of the source of the message. On the contrary, only in the Extroversion experiment, findings showed that being aware of the
personalization brought people to perceive the message as more persuasive and the source as more credible. Previous studies assumed that people reject personalized messages, especially because of their concern for their privacy (e.g. Kruikemeier et al., 2016; Rubinstein, 2014). Instead, our findings cannot support such theory, since no sign of rejection of the message was found among respondents who were aware of the personalization. Perhaps, this could suggest that people have not yet developed a great concern for the use of political microtargeting in politics. On the opposite, they are probably getting used to the sponsored ads they see on their social media and to the personalized ads they receive as customers in their daily lives. This may restrain their scepticism towards personalized messages even in the political field.
Finally, some suggestive evidence of the interaction hypothesis was found only in the extroversion experiment, which showed that the difference in the perceived persuasiveness of congruent and incongruent messages is larger in case of people who are aware of the personalization of the message than in case of unaware respondents. A possible explanation is that, contrary to people who are unaware of the personalization, people who are aware of the message being tailored to their personality expect to receive a message congruent with their own personality. A very incongruent
26
message increases their confusion and makes the message less persuasive. This finding is in line with Hersch and Schaffner’s study, which stated that ‘mistargeted’ voters penalize candidates enough to erase the positive returns of microtargeting (Hersch & Schaffner, 2013).
Limitations
Coming to the limitations of this research, some may argue that a convenience sample limits the external validity of the results. However, extant research (e.g. Mullinix, Leeper, Druckman, & Freese, 2015) proved that representative samples are not indispensable in allowing the
generalization of findings in experimental research. Therefore, the employment of a convenience sample cannot be considered as a limitation of this specific study, but rather as a pondered choice. The first limitation of our study concerns the difference between the findings of the extroversion and agreeableness experiments. For instance, the agreeableness experiment showed no effect of the awareness of the personalization, whereas the extroversion experiment showed a positive effect on persuasiveness and source credibility. This could be due to the fact that people were first exposed to the extroversion experiment and then to the agreeableness experiment. Reading the first
manipulation check question - asking them whether they thought the message was personalized or not - might have alerted the respondents during the second part of the survey, making them more sceptical towards the messages of the agreeableness experiment. Therefore, a possible follow-up study could avoid this bias, by randomly assigning the order of the two experiments. Furthermore, future studies could investigate other personality traits as well, in order to compare their influence on persuasiveness and source credibility against those examined in the present study.
Secondly, the stimuli were artificially created for the purpose of this study in the form of a tweet. This might have been perceived as an odd choice among those who are not familiar with reading political messages on Twitter, and in particular among older people, who are probably less acquainted with social networks in general. This may have rendered the artificiality of the stimuli more evident. However, the choice of creating the political messages in form of tweets is justified
27
by the fact that Twitter is increasingly becoming a channel used by politicians to interact with their voters and to express political opinions (Bode & Dalrymple, 2016). Also, social networks in general are the main channels that allow the microtargeting of voters, since such techniques have not yet been developed on more traditional media such as television or newspapers.
Lastly, a final limitation is connected to the manipulation check, which asked respondents if they thought the message they read was personalized. People in the unawareness condition who were shown a message very congruent with their own personality might have thought it was personalized even if they were not supposed to be aware of the personalization. Therefore, the results of the manipulation check (see Appendix C) might have been slightly altered by those respondents. Discussion
In conclusion, the present study clearly demonstrated the efficacy of tailored campaign messages. Consequently, we can undoubtedly foresee a greater use of political microtargeting in the coming years, since it has been proven to increase ads’ persuasiveness and source credibility. This requires us to linger over the normative implications of political microtargeting.
For instance, when every voter is targeted with a different message according to his/her personality traits, different people get different messages by the same politician, instead of a single and clear message directed to the whole electorate. Indeed, many of the claims made by political candidates through targeting messages remain hidden from the view of the non-targeted majority of the population (Hersch & Schaffner, 2013). Therefore, the risk of keeping people shut inside an information “filter bubble” made of personalized ads is very high and represents a dangerous implication for democracy.
Furthermore, by targeting the specific wishes of potential voters, more broadly relevant political topics are often excluded or left ambiguous during campaigns (Tufekci, 2014). Therefore, we can
28
state that the employment of personalized political messages represents a potential harm to the civic discourse as well.
Finally, the fact that the average voter is apparently unaware of the implications and dangers represented by such microtargeting techniques renders the issue even more urgent and delicate. Unquestionably, the electorate should be made aware of the employment of their personal information, such as their personality traits, as means to target them with personalized messages. Only in this way, we could contribute to the democratization of the political process, by allowing citizens to have more control and responsibility over such a complex political environment.
References
Amsalem, E., Zoizner, A., Sheafer, T., Walgrave, S., & Loewen, P. J. (2018). The Effect of Politicians’ Personality on Their Media Visibility. Communication Research, 1-24. doi: 10.1177/0093650218758084
Bakker, B. N., Hopmann, D. N., & Persson, M. (2015). Personality traits and party identification over time. European Journal of Political Research, 54, 197–215. doi:
10.1111/1475-6765.12070
Bakker, B. N., Rooduijn, M., & Schumacher, G. (2016). The psychological roots of populist voting: Evidence from the United States, the Netherlands and Germany. European Journal of Political Research, 55, 302–320. doi: 10.1111/1475-6765.12121
Barocas, S. (2012). The Price of Precision: Voter Microtargeting and Its Potential Harms to the Democratic Process. PLEAD '12: Proceedings of the first edition workshop on Politics, elections and data, 31-36. doi: 10.1145/2389661.2389671
29
Bennett, W.L. (2012) The Personalization of Politics: Political Identity, Social Media, and Changing Patterns of Participation. The Annals of the American Academy of Political and Social Science, 644(1), 20-39. doi: 10.1177/0002716212451428
Bode, L. & Dalrymple K. E. (2016). Politics in 140 Characters or Less: Campaign Communication, Network Interaction, and Political Participation on Twitter. Journal of Political Marketing, 15(4), 311-332. doi: 10.1080/15377857.2014.959686
Borgesius, F. J. Z., Möller, J., Kruikemeier, S., Fathaigh, R., Irion, K., Dobber, T., & de Vreese, C. (2018). Online political microtargeting: Promises and threats for democracy. Utrecht Law Review. doi: 10.18352/.420
Cadwalladr, C., & Graham-Harrison, E. (published on 2018 March 17th). Revealed: 50 mln Facebook profiles harvested for Cambridge Analytica in major data breach. The Guardian. Retrieved from: https://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook-influence-us-election
Caprara, G. V., Barbaranelli, C., & Zimbardo, P. G. (2002). When Parsimony Subdues Distinctiveness: Simplified Public Perceptions of Politicians’ Personality. Political Psychology, 23(1), 77-95. doi: 10.1111/0162-895X.00271
Caprara, G. V., Barbaranelli, C., Consiglio, C., Picconi, L., & Zimbardo, P. G. (2003). Personalities of politicians and voters: Unique and synergistic relationships. Journal of Personality and Social Psychology, 84(4), 849-856. doi: 10.1037/0022-3514.84.4.849
Caprara, G. V., & Zimbardo, P. G. (2004). Personalizing Politics: A Congruency Model of Political Preference. American Psychologist, 59(7), 581-594. doi: 10.1037/0003-066X.59.7.581
30
DeYoung, C. G., Weisberg, Y. J., Quilty, L. C., & Peterson, J. B. (2013). Unifying the aspects of the Big Five, the interpersonal circumplex, and trait affiliation. Journal of Personality, 81, 465-475. doi: 10.1111/jopy.12020
Dinesen, P. T., Norgaard, A. S., & Klemmensen, R. (2013). The Civic Personality: Personality and Democratic Citizenship. Political Studies, 1-19. doi: 10.111171467-9248.12094
Faul, F., Erdfelder, E., Lang, A. G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39, 175-191. doi: 10.3758/BF03193146
Gerber, A. S., Huber, G. A., Doherty, D., & Dowling, C. M. (2011). Personality traits and the consumption of political information. American Politics Research, 39, 32-84.
doi:10.1177/1532673X10381466
Gerber, A. S., Huber, G. A., Doherty, D., Dowling, C. M., & Panagopoulos, C. (2013). Big Five Personality Traits and Responses to Persuasive Appeals: Results from Voter Turnout Experiments. Political Behavior. doi: 10.1007/s11109-012-9216-y
Goldberg, L. R. (1999). A broad-bandwidth, public domain, personality inventory measuring the lower-level facets of several five-factor models. In I. Mervielde, I. Deary, F. De Fruyt, & F. Ostendorf (Eds.), Personality Psychology in Europe, Vol. 7 (pp. 7-28). Tilburg, The
Netherlands: Tilburg University Press
González, R. J. (2017). Hacking the citizenry?: Personality profiling, ‘big data’ and the election of Donald Trump. Anthropology Today. doi: 10.1111/1467-8322.12348
Green, D. P., & Gerber, A. S. (2008). Get Out theVote: How to Increase Voter Turnout. Washington, DC; Brookings Institution Press
31
Hersh, E. D., & Schaffner, B. F. (2013). Targeted campaign appeals and the value of ambiguity. Journal of Politics. doi: 10.1017/S0022381613000182
Hirsch, J. B., Kang, S. K., & Bodenhausen, G. (2012). Personalized Persuasion: Tailoring
Persuasive Appeals to Recipients’ Personality Traits. Psychological Science, 23(6) 578–581. doi: 10.1177/0956797611436349
Housholder, E. E., & LaMarre, H. L. (2014) Facebook Politics: Toward a Process Model for Achieving Political Source Credibility Through Social Media. Journal of Information Technology & Politics, 11(4), 368-382. doi: 10.1080/19331681.2014.951753
Hovland, C.I., & Weiss, W. (1951). The Influence of Source Credibility on Communication Effectiveness. Public Opinion Quarterly, 15(4), 635-650. doi :10.1086/266350
Kruikemeier, S., Sezgin, M., & Boerman, S. C., (2016). Political Microtargeting: Relationship Between Personalized Advertising on Facebook and Voters’ Responses. Cyberpsychology, Behavior, And Social Networking, 19 (6), 367-372. doi: 10.1089/cyber.2015.0652
Matz, S. C., Kosinski, M., Nave, G., & Stillwell, D. J. (2017). Psychological targeting as an effective approach to digital mass persuasion. Proceedings of the National Academy of Sciences, 114(48), 12714-12719. doi: 10.1073/pnas.1710966114
McCroskey, J. C., & Young, T. J. (1981) Ethos and credibility: The construct and its measurement after three decades. Central States Speech Journal, 32(1), 24-34. doi:
10.1080/10510978109368075
Mondak, J. J. (1995). Competence, integrity and the electoral success of congressional incumbents. Journal of Politics, 57(4), 1043–1069. doi: 10.2307/2960401
32
Mondak, J. J., & Halperin, K. D. (2008). A framework for the study of personality and political behaviour. British Journal of Political Science. doi: 10.1017/S0007123408000173
Mullinix, K. J., Leeper, T. J., Druckman, J. N., & Freese, J. (2015). The Generalizability of Survey Experiments. Journal of Experimental Political Science, 2(2), 109-138. doi:
10.1017/XPS.2015.19
Neuman, W. L. (2012). Experimental Research. Basics of social research: qualitative and quantitative approaches (pp.193-215). Boston: Pearson Education.
Ohanian, R. (1990). Construction and Validation of a Scale to Measure Celebrity Endorsers'
Perceived Expertise, Trustworthiness, and Attractiveness. Journal of Advertising, 19(3), 39-52. doi: 10.1080/00913367.1990.10673191
Rubinstein, I. (2014). Voter Privacy in the Age of Big Data. SSRN. doi: 10.2139/ssrn.2447956
Schipper, B. C., & Woo, H. (2012). Political Awareness, Microtargeting of Voters, and Negative Electoral Campaigning. SSRN. doi: 10.2139/ssrn.2039122
Statista (2018). Digital political advertising spending in the United States from 2008 to 2020 (in million U.S. dollars). Retrieved from: https://www.statista.com/statistics/309592/online-political-ad-spend-usa
Teven, J. J., & McCroskey, J. C. (1997). The relationship of perceived teacher caring with student learning and teacher evaluation. Communication Education, 46(1), 1–9. doi:
10.1080/03634529709379069
Tufekci, Z. (2014). Engineering the public: Big data, surveillance and computational politics. First Monday, 19(7). doi: 10.5210/fm.v19i7.4901
33
The Economist (published on 2016 November 20th). The role of technology in the presidential election. Retrieved from: https://www.economist.com/united-states/2016/11/20/the-role-of-technology-in-the-presidential-election
Whitehead, J. L. (1968). Factors of Source Credibility. Quarterly Journal of Speech, 54(1), 59-63. doi: 10.1080/00335636809382870
34
Appendix A: Tweets
Figure 4 - English extroverted tweet (left) and introverted tweet (right) of awareness condition.
Figure 5 - English high agreeable (left) and low agreeable tweet (right) of awareness condition.
Figure 6 - Italian extroverted (left) and introverted tweet (right) of awareness condition.
Figure 7 - Italian high agreeable tweet (left) and low agreeable tweet (right) of awareness condition.
35
Figure 8 - English extroverted tweet (left) and introverted tweet (right) of unawareness condition.
Figure 9 - English high agreeable (left) and low agreeable tweet (right) of unawareness condition.
Figure 10 - Italian extroverted (left) and introverted tweet (right) of unawareness condition.
Figure 11 - Italian high agreeable tweet (left) and low agreeable tweet (right) of unawareness condition.
36
Appendix B: Scales
Table 5 - Source credibility scale by Teven and McCroskey (1997).
‘Source credibility’ scale adapted from Teven and McCroskey’s scale (see Table 1) and used in the present study:
I think the politician sponsoring the ad:
Is intelligent; Is expert; Is competent; Is bright; Is honest; Is trustworthy; Is honourable; Is ethical; Is genuine; Cares about me; Has my interests at heart; Is not self-centred; Is sensitive; Is
understanding.
NEO-IPIP Scale items (retrieved from: https://ipip.ori.org/newNEOKey.htm#Agreeableness) Agreeableness:
Have a good word for everyone.
Believe that others have good intentions. Respect others.
Accept people as they are. Make people feel at ease. Have a sharp tongue. Cut others to pieces.
37
Get back at others. Insult people. Extroversion:
Feel comfortable around people. Make friends easily.
Am skilled in handling social situations. Am the life of the party.
Know how to captivate people. Have little to say.
Keep in the background.
Would describe my experiences as somewhat dull. Don't like to draw attention to myself.
Don't talk a lot.
Appendix C: Randomization, manipulation and robustness checks Randomization checks
Respondents were correctly randomized across the experimental conditions. Indeed, in the
extroversion experiment, 49.8 % of respondents were exposed to the awareness condition, whereas in the agreeableness experiment 51.4% were.
To check if participants’ age, political self-placement and EU attitudes were comparable across the awareness condition, three independent-samples t-tests were conducted with age, political self-placement and EU attitude as dependent variables and the awareness condition as grouping
variable. The independent-samples t-tests showed that participants’ age, political self-placement and EU attitudes in the awareness condition were not significantly different from those of participants in the unaware condition: age: t(310) = .07, p = .945, 95% CI [-3.38, 3.63], political self-placement: t(322) = 1.37, p = .171, 95% CI [-.14, .77], EU attitudes: t(321) = -.41, p = .684, 95% CI [-.54, .36]. Therefore, the randomization of participants across the awareness condition was successful in terms of participants’ age, political self-placement and EU attitudes. The same tests were run for the second message participants were exposed to, and results in terms of significance were the same. To check if participant’s gender was comparable across the awareness condition, a cross-tabulation was conducted with gender as a dependent variable and awareness as independent variable. The Chi-square test was not significant in the case of the extroversion experiment, X² (2, N = 327) =
38
3.21, p = .201, showing that participants were correctly randomized across the awareness condition in terms of their gender. Instead the chi-square test was significant in the case of the agreeableness experiment, X² (2, N = 327) = 12.15, p = .002, showing that there is an association between gender and the awareness condition.
To check if participants’ age, political self-placement and EU attitudes were comparable also across the congruency condition, a linear regression was conducted with age, political self-placement and EU attitude as predictors and the congruency of the message as a dependent variable. None of the independent variables resulted in significant predictors: age, b = .00, b* = .00, t = .03, p =.975; political self-placement: b = .40, b* = .08, t = 1.31, p =.190; EU attitudes: b = .28, b* = .06, t = .91, p =.363. Hence, the randomization of participants across the congruency condition was successful in terms of participants’ age, political self-placement and EU attitudes. The same tests were run for the agreeableness experiment and results in terms of significance were the same: age, b = -.00, b* = -.00, t = -.10, p =.920; political self-placement: b = .53, b* = .08, t = 1.23, p =.218; EU attitudes: b = .21, b* = .03, t = .48, p =.630. Finally, to check if gender was comparable across the congruency condition too, a one-way analysis of variance was conducted between gender as a grouping variable and congruency as a dependent variable. The randomization occurred correctly both in case of the agreeableness experiment, F(2,324) = .60, p = .552, and in case of the
extroversion one, F(2,324) = 2.15, p = .118. Manipulation check
To check if participants perceived the manipulation as intended, a cross-tabulation was conducted to prove a correlation between the manipulation check variable (‘Do you think the message you saw was specifically tailored to you and your personality?) and the awareness of the personalization. A cross-tabulation was conducted for both messages and the association proved to be significant both in the extroversion experiment, X² (2, N = 327) = 7.53, p < .05 and in the agreeableness one, X² (2, N = 327) = 6.68, p < .05. Hence, in both experiments participants in the aware condition were aware of the message being personalized and participants in the unaware condition were not.
Robustness check
The variable of the source’s credibility was decomposed into three factors – competence, trustworthiness and goodwill – by computing the mean scores of the four items measuring
competence (extroversion experiment: M = .51, SD = .19, Min = .00, Max = .96, Cronbach’s alpha = .86; agreeableness experiment: M = .44, SD = .19, Min = .00, Max = .92, Cronbach’s alpha = .89), the five items measuring trustworthiness (extroversion experiment: M = .51, SD = .19, Min = .00, Max = 1.00, Cronbach’s alpha = .90; agreeableness experiment: M = .40, SD = .21, Min = .00, Max
39
= 1.00, Cronbach’s alpha = .91), and the five items measuring goodwill (extroversion experiment: M = .46, SD = .19, Min = .00, Max = .90, Cronbach’s alpha = .86; agreeableness experiment: M = .38, SD = .21, Min = .00, Max = 1.00, Cronbach’s alpha = .91).
By decomposing the variable of credibility into its three clusters – competence, trustworthiness and goodwill – and employing them as dependent variables in a regression model with congruency and awareness as predictors, results follow the same logic as the ones presenting source credibility as one whole variable. Concerning the extroversion experiment, awareness seems to be a positive predictor of the three clusters of credibility, whereas in the agreeableness experiment, congruency is a negative predictor of the perceived competence, trustworthiness and goodwill of the source (see Table 6).
Table 6 - Linear regression models predicting the source credibility. Competence b* Trustworthiness b* Goodwill b* Extroversion experiment Constant 3.52*** 4.11*** 3.94*** Congruency .09 .07 -.11 Awareness .24*** .16** .16** adjR² .06 .02 .03 F 11.19*** 4.82** 6.01** N 327 327 327 Agreeableness experiment Constant 3.04*** 2.51*** 1.98*** Congruency .25*** .34*** .46*** Awareness .02 .02 .04 adjR² .06 .11 .21 F 11.20*** 21.08*** 44.27*** N 327 327 327 Note 1: * p < .05. ** p < .01. *** p < .001.
Note 2: Correlations extroversion experiment: competence-trustworthiness: r(326) = .71, p < .001; competence-goodwill: r(326) = .66, p < .001; trustworthiness-goodwill: r(326) = .84, p < .001. Note 3: Correlations agreeableness experiment: competence-trustworthiness: r(326) = .72, p < .001; competence-goodwill: r(326) = .72, p < .001; trustworthiness-goodwill: r(326) = .88, p < .001.
