Personality Targeted Political Communication
The effect of targeting extraversion on voter persuasion and perceived attractiveness of the message
by
Tomás de Vilas-Boas Lebreiro
10210911
Master’s Thesis
Master’s Communication Science: Political Communication
Amsterdam Graduate School of Communication
Word Count: 7485 Supervisor: Erika van Elsas
1 Abstract
After the latest US election (2016) there were many unconfirmed stories in the media indicating that certain campaign strategies were used to target people in a psychographic way. Supposedly personality profiles of potential voters were composed, using online social media data, which were subsequently targeted in personality-specific sophisticated ways. There is a lot of skepticism from both journalists and scholars whether such methods were actually used or can be used in a successful way. Indications from other research fields such as health communication and commercial marketing indicate that they might work. This study sought to test if tailoring a political message from a specific party to a personality dimension
(extraversion in this case) would increase voting intention and perceived attractiveness of the message. This was done by conducting an experiment among 91 respondents who were either shown a neutral campaign poster or a extravert campaign poster and then asked about the attractiveness of the poster and their voting intention. Subsequently they filled in a personality questionnaire which measured their level of extraversion (subdivided into
enthusiasm and assertiveness dimensions). Results showed no evidence for a positive relation between extrovert messages and the reaction of extrovert people, the results pattern actually indicating the opposite. Assertiveness was found to have a significant negative relation with voting intention. None of the other relations was found to be significant except for one: people in the experimental group scored significantly higher on perceived attractiveness of the message. These results suggest that personality targeting in the political context might not work or calls for a much more complex process of stimulus development, along with a deeper understanding of how the human personality is related to persuasion.
Keywords: Political communication, personality-targeting , micro-targeting, Big Five, extraversion, voting intention, perceived attractiveness of message
2 The election of Donald Trump as president of the USA was an unforeseen result and has as a consequence led to a great deal of discussion and theorizing on how it was possible. Some journalists have picked up the idea that one of the possible drivers of that unexpected success was the use of psychographic micro-targeted data driven political advertisements (Grasseger & Krogerus, 2017; Anderson & Horvath, 2017; Doward & Gibbs, 2017) while others dismiss these ideas saying they are exaggerated and part of a credit claiming contest by political consultants (Taggart, 2017; Confessore & Hakim, 2017).
We do know that through the use of data mining, “a data analysis methodology that employs specialized algorithms to extract information from extensive data sets” (Murray & Scime, p.144, 2010), one can “model target audience groups and predict the behavior of like-minded people”, allowing for the tailoring of advertisements in a way that targets specific subgroups of people (Doward & Gibbs, 2017, p1; Murray & Scime, 2010). This is the idea of political micro-targeting, and it is something that has been done in US elections at least since the 2000 Bush campaign (Schipper & Woo, 2014). Using public data and computerized models, political campaigns can identify relatively homogeneous groups with similar points of view or primary issues and then target them specifically (Murray & Scime, 2010). The innovation attributed to the Trump campaign is that it supposedly used psychological characteristics of voters as one of the segmenting and targeting variables (Grasseger & Krogerus, 2017). This besides the “traditional” known ones such as demographics, voter history, stances on certain policy issues and consumer data (Schipper & Woo, 2014).
It is questionable whether personality-based micro-targeting actually happened and/or was effective (Doward & Gibbs, 2017) as there is no scientific evidence to back those claims up. As Daniel Kreiss, one of the leading scholars in the political micro-targeting field, puts it: “I haven’t seen any data that has suggested that this form of modeling does or doesn’t work” (Kreiss as quoted in Talbot, 2016). Nonetheless we do know that it is possible to create
3 a psychological personality profile of people based on their social media activity (Kosinski, Stillwell & Graepel, 2013) and that it is possible to successfully adapt persuasive commercial messages to certain personality traits, something called “psychographic persuasion (Matz et al., p.1, 2017; Moon, 2002; Hirsch, Kang & Bodenhausen, 2012). Studies that analyzed the efficiency of personality tailored advertisements for commercial products, in different ways and settings, found that matching ads to the consumers’ personality increases their impact, attractiveness and willingness to buy (Matz et al., 2017; Hirsch, Kang & Bodenhausen, 2012). If this is possible with commercial messages then it might be possible to apply these same methods, albeit with slight differences, to political messages. The question that arises here is whether these methods work and are potentially a future cornerstone of political communication. Especially as no research has been published into the specific dynamics of personality targeting in the political context, nor has there been any science-based theorizing and discussion on the possible consequences of such a development. This combination of a scientific knowledge gap and the potential societal and democratic implications lead to the research question that shall be at the heart of this study: “How effective is the tailoring of political messages to components of people’s personality in terms of (perceived) message persuasiveness and attractiveness?”
To answer this question an experiment was conducted among 91 respondents using an independent measures design which sought to analyze the relation between extraversion and perceived attractiveness and actual persuasiveness of political messages, within and between experimental groups (extravert condition vs neutral condition). The main objective is to see whether extrovertly framed political messages are more appealing to extrovert people.
This is relevant from a scientific point of view because, to my knowledge, there has not been any research conducted on personality targeted political communication. Conducting this experiment and adding its results and outcomes to the literature is therefore
4 a valuable contribution to the field of political marketing/communication. From a societal point of view the research is relevant considering the fact that the technological possibilities appear to allow for personality targeting and there are indications from the commercial marketing field that these techniques might work. If we consider the wider democratic implications that this would have, such as the creation of a public sphere where every individual gets his/her own version of political ideas, tailored to persuade him by means other than rational and/or ideological ones, this also adds to the societal relevance. Looking at these developments in a more optimistic way, one could also see them being used to increase political participation and interest. This by targeting voters on the issues that are important to them in a personalized way. Whether one or both these scenarios turn out to be true, it is relevant to study the effectiveness of these methods and to discuss the ethical boundaries that may come with them.
The development of political micro-targeting
The gathering of data on the electorate is central in political campaigning as one needs to know what the audience wants and how to cater to those needs. Political micro-targeting is about using data to segment the potential electorate into highly homogeneous groups and then cater to those segments’ needs in specific ways (Kreiss & Welch, 2015; Murray & Scime, 2010; Schipper & Woo, 2014).
This is a development that really started in the 2004 U.S. electoral cycle, when parties began measuring and using online data about internet users to make their interactive online advertisements more effective, something we can already consider political micro-targeting (Kreiss & Welch, 2015; Schipper & Woo, 2014). This really “exploded” with the 2008 Obama campaign, which was (by then) the “fullest realization of trends in the political field toward crafting better means of collecting, storing, analyzing, and acting upon data about
5 citizens, their online behavior, and their social relationships.“ (Kreiss & Howard, p.1033, 2010). This was the case because this was the first big campaign with massive resources where social media and new technologies played such a pivotal role (Kreiss & Welch, 2015; Kreiss, 2010; Lutz, 2009). The 2008 Obama campaign was also the first one where more ‘sophisticated’ online micro-targeting took place; it used and synchronized data from social media such as Facebook, MyBarackObama.com (the campaign’s own social media platform), website click troughs, “on the ground data” and the Democratic party’s national voter database. By applying statistical computer models to this different content was targeted at different demographic groups in the form of personalized persuasive appeals (Kreiss & Welch, 2015; Kreiss & Howard, 2010; Lutz, 2009).
Ever since that campaign the speed of developments in data-driven political communication has been high. In the 2012 US presidential campaign for example there were already more technological capabilities and data available, aspects that are at the heart of contemporary campaigning, that allowed for “individualized information flows” (Kreiss & Welch, p.2, 2015). With techniques such as “voter matching” political campaigns use browsing histories and online behavior of known supporters to find other computer users with similar traits, which subsequently become targeted voter groups (Kreiss & Welch, 2015; Schipper & Woo, 2014
So, as is clear from the existing literature, online political micro-targeting is a process in continuous development which started around 2000 and saw innovations with each US electoral cycle.
Looking at the case of Europe, Bennett (2016, p.274) notes: ” So far European parties and candidates cannot campaign in Europe, as their counterparts do in North America. Cultural, legal, institutional, financial, and other constraints will continue to block the more intrusive campaigning practices now seen in the USA.” It is important to note that this is the
6 case “so far”, as these developments and techniques are slowly making their way into European countries, which are “drawing lessons from US experience (Bennett, p.261, 2016). The Netherlands’ last general elections for example, held in March 2017, were the first ones where political parties performed “serious data operations’ for on- and offline targeting of smaller more homogeneous groups (Dobber, T. as cited in Martijn, 2017). This is the start of a trend that is expected to “grow with each electoral cycle” (Dobber, T. as cited in Martijn, 2017) and which makes research in the Dutch context relevant.
The effectiveness of political micro-targeting
The extent to which political micro-targeting is effective is unknown, as: “The gaps in this area of research are troubling” especially if we consider that “this was the most significant trend in online advertising over the last decade” (Barnard & Kreiss, 2013, p.2059). Even though there are no studies that have looked at the effectivity of political micro-targeting as done in ‘real life’ campaigns, as political parties and consultants are very reluctant in sharing this kind of information, there are studies that look at the effectivity of targeting groups with a shared identity in an experimental setting. Targeting on less homogeneous grouping variables such as gender for example was found to be effective in affecting votes (likelihood of voting for a certain candidate), particularly for female candidates priming female voters’ gender identity (Holman, Schneider & Pondel, 2015). Hersch & Schaffner’s study (2013) on the persuasive effects of targeting social groups; such as “middle class”, “gun owners” and “unions” for example, in campaign appeals found that this type of micro-targeting in most cases did not lead to significant benefits for the candidate. There were some exceptions however (appeals to Latinos and born again Christians). This could indicate that only specific subgroups are persuaded by targeted messages or perhaps those two groups mentioned before were more homogeneous in their
7 needs and ideals. It is also safe to assume that current developments in political campaigning allow for targeting based on groups that are more homogeneous than those two, as being Latino or born again Christian can be combined with other variables related to political ideas and socio-economic status, for example.
These two studies are the only ones I could find that test the effectiveness of group/micro-targeted campaign strategies; a scarce amount considering it is something crucially important for the field of political communication
Personality based targeting
Research on micro-targeting in general and personality based targeting in particular in the political field is lacking. It has however been conducted in other fields and found to be effective in many cases, an overview of which can be found in Kalyanaraman & Sundar (2006). These studies found that messages that are personalized to match aspects of the receiver are more persuasive as people, especially in western “individualistic” cultures, prefer experiences, ideas and objects that are tied to themselves personally (Barnard & Kreiss, 2013; Kalyanaraman & Sundar, 2006). When it comes to health communication for example studies found that the extent of tailoring in communications (to demographic, behavioral, cultural and psychosocial characteristics) was directly related to how well messages were remembered, read, considered relevant and grasped attention. This subsequently led to them having a bigger chance of changing lifestyle and behavior; at the same time the existing evidence suggests that they may be cost-effective (Noar, Benac & Harris, 2007; Skinner et al., 1999; (Noar et al., 2011).
This idea of “tying” messages to the receiver personally seems to me to be a logic predecessor of personality based targeting, as that would be a way to further increase the level of personalization and subsequently the effectiveness of the message.
8 To tailor a (political) message to the recipient’s personality one must firstly be able to describe or define that personality in a measurable way. For the assessing and describing of a person’s personality one of the most widely used models is that of the Big Five, a “hierarchical organization of personality traits in terms of five basic dimensions: extraversion, agreeableness, conscientiousness, neuroticism, and openness to experience” (McCrae & John, p.175, 1992). These factors can be inferred from self-reporting, theoretically based questionnaires and the natural use of language. They were also found to be the fundamental dimensions of personality, capturing the most significant variation in human personality in an essentially correct way, by researchers from different traditions (McCrae & John, 1992; Moon, 2002). This is the personality model allegedly used in the recent US elections and the one that will be used as a basis for this research.
Personality based micro-targeting consists of adapting messages in a way that is congruent with an individual’s personality. This was found to be effective in a number of studies, with the underlying mechanism being that individuals prefer communications styles that are in line with their personality, something that in this research will be referred to as ‘congruence’ (Kalyanaraman & Sundar, 2006; York, Brannon & Miller, 2012; Hirsch, Kang & Bodenhausen, 2012; Moon, 2002).
This ‘congruence’ has been described in a variety of ways. Moon (2002) argues that personality-tailoring is effective because people are more attracted to ‘similar’ others and also more influenced by them. Research in the health communication field suggests that tailored messages are more relevant and effective in influencing behavior because they ‘match some aspects of the self’, the underlying mechanism being that individuals want to accentuate their unique characteristics and positively affirm their identity and distinctiveness (Kalyanaraman & Sundar, 2006; York, Brannon & Miller, 2012). This ‘congruence’ makes tailored messages more effective according to (self) schema correspondence theory (York, Brannon & Miller,
9 2012). Self-schemas are the “structured cognitive organization of information about the self” (York, Brannon & Miller, p.51, 2012), meaning the image one has of one’s personality and individually related traits or “the individual’s representation of themselves” (York, Brannon & Miller, p.51, 2012). Messages that correspond to that self-schema are found to have a bigger and more effective impact as individuals prefer that ‘congruence’ (York, Brannon & Miller, 2012). In the same line Hirsch, Kang & Bodenhausen consider that when messages are congruent with an individual’s distinct motivational system, they are processed more fluently and evaluated more positively than incongruent messages. This is reflected by variations in each Big Five dimension and leads to what they in turn named “regulatory fit” (Hirsch, Kang & Bodenhausen, p.1, 2012).
So, summarizing, these studies point to ‘congruence’ of individuals’ personality with the message as a factor in increasing effectiveness, attention, relevance and memory of that message. This because individuals want to accent and act in line with the self and because they are attracted to and more persuaded by messages/objects that are similar to that self, much in the way a car salesman can be more successful by mimicking the language and gestures of the potential buyer (York, Brannon & Miller, 2012). Considering this I expect this positive effect of “congruence” on message effectiveness to translate into to the political communication field and to work for personality targeted political messages.
Personality based targeting in commercial marketing
From the field of commercial marketing comes some evidence suggesting congruent messages are more persuasive and effective. Matz et al. (2017), for example, conducted three real-world experiments that used online psychographic persuasion to influence people’s consuming behavior. Different ads were created to target Extraversion and Openness-to-Experience, considered to be personality dimensions more predictive of behavior and actual
10 preferences (Matz et al., 2017). The results showed that across the campaigns users in the personality congruent condition were more likely to purchase the advertised product as opposed to the non-congruent condition (Matz et al., 2017). Furthermore these matched persuasive messages resulted in far more clicks and purchases than mismatched or un-personalized content, being significantly persuasive (Matz et al., 2017). Hirsch, Kang & Bodenhausen’s found in their experimental study (2012) that an advertisements’ rated effectiveness increased when the respondents’ score on the targeted trait also increased. This correlation was found to be highest and most significant for extraversion and agreeableness (out of the Big Five) (Hirsch, Kang & Bodenhausen, 2012). Moon’s study (2002) suggests that computers are more effective influencers when they adapt their message style to be congruent with the participants personality type, specifically for consumer goods, news and entertainment messages in this case. The chosen trait was extraversion, which was chosen because it is the most “psychologically prominent” and the most useful in understanding and predicting individuals’ behavior [(Williams et al., 1995) as quoted in (Moon, p.314, 2002)]. When it comes to the targeting of people’s personality with a persuasive intent there is no research conducted so far in a political environment; something that seems to be relevant and called for considering the previously mentioned studies. Considering that, this article will focus on the political potential of persuasive personality targeted messages.
Personality targeted political communication: focus of the study
The intention of this research is to be a first step in understanding the full potential of personality based micro-targeting in political campaigning. As it is limited in its resources and previous field specific knowledge to build on, it will focus on the impact of tailoring for one personality trait: extraversion. This will be a limitation for eventual results’ generalizability, as there are four more personality traits, out of the Big Five, that could be
11 targeted as well as combinations of them which could have different implications. The intention here however, is not to build a fully comprehensive framework but to provide a first assessment of whether the indications from other communicative fields translate to the political communication field.
Extraversion as a trait was chosen because, as seen before, it is the trait most widely used across the different studies mentioned and considered to be on the higher end of predictability for behavior and preferences and for the judging of others (Moon, 2002; Williams et al., 1995; Hirsch, Kang & Bodenhausen, 2012; Matz et al., 2017).
The purpose of this research is to find out whether tailoring messages to respondents’ levels of extraversion will lead to higher levels of (perceived) persuasion, measured by voting intention in (Holman, Schneider & Pondel, 2015; Hersch & Schaffner, 2013) and (perceived) attractiveness (as measured in Hirsch, Kang & Bodenhausen, 2012). The following hypotheses were derived from that:
H1. An extrovertly framed political message will have a positive effect on perceived attractiveness of that message, compared to a neutrally framed message, but only for people with higher levels of extraversion.
H2. An extrovertly framed political message will have a positive effect on voting intention, compared to a neutrally framed message, but only for people with higher levels of extraversion.
These are the two main hypotheses that this research sets out to test and the ones most directly related to the research question. Looking at the “general” trait of extraversion in the literature however, the notion emerges that it is a trait with at least two subdivisions. Depue and Collins (1999), for example, noted in their literature review on Extraversion a primary division between agency and sociability. DeYoung, Quilty, & Peterson (2007) classified
12 these two dimensions of extraversion as assertiveness and enthusiasm, in their frequently cited and built on article. Assertiveness is considered to be related to the concepts of agency and dominance, meaning that people scoring highly on that trait feel they can accomplish goals, have a sense of potency and tend to be socially dominant (Depue & Collins, p. 491, 1999; DeYoung, Quilty, & Peterson, p.883, 2007). Enthusiasm in turn is related to positive emotions, sociability, anticipation, enjoyment of social rewards and impulsivity (Depue & Collins, p. 491, 1999; DeYoung, Quilty, & Peterson, pp.883-884, 2007). Most previously mentioned studies appear to me to have targeted the personality trait of extraversion through the enthusiasm factor. This appears to be easier and more straightforward to target, besides the fact that assertiveness is a factor which due to its agency component could possibly impede persuasion. It is not known if this effective targeting of extraversion in the previously mentioned studies was driven by the dimension of extraversion as a whole or more related to the enthusiasm ‘sub-dimension’. I expect extraversion as a whole to be a trait that will be positively related with an extrovert message, but I also expect that effect to be driven mostly by the enthusiasm component and despite the assertiveness component. This study intends to be an addition to the literature by looking at both the extraversion trait and its sub-dimensions, as targeted personality components. Considering this and the design of our stimulus this leads to the following hypotheses:
H3. An extrovertly framed political message will have a positive effect on perceived attractiveness of that message, compared to a neutrally framed message, but only for people with higher levels of enthusiasm.
H4. An extrovertly framed political message will have a positive effect on voting intention, compared to a neutrally framed message, but only for people with higher levels of enthusiasm.
13 The agency and sense of control factors in the assertiveness trait lead to the assumption that highly assertive people will be relatively more difficult to persuade to vote for a new party as they probably have stronger previous political ideas. This results in the following hypothesis:
H5. Voting intention will be negatively impacted by level of assertiveness.
Data and Methods
Data
To measure the efficiency of personality based targeting of political messages for the two experimental conditions an experiment amongst 108 respondents was conducted. After eliminating respondents that did not answer all the relevant questions this number lowered to N=91. Respondents were randomly allocated to the control condition (N=45) or the experimental condition (N=46). Considering the fact that the experiment was in part about (future) voting behavior and the Dutch local elections of March 2018, the decision was made to conduct the experiment in Dutch. These were the only two requirements that participants
Condition Voting Intention Rated Attractiveness / Enthusiasm) (Assertiveness EXTRAVERSION
Fig. 1 - Conceptual Model of Extraversion Targeting
+
14 needed to fulfill in order to be allowed to participate (speaking Dutch and being legally allowed to vote). This because personality targeted political advertisement can be directed at any member of the electorate; each individual has a personality that can be translated into the Big Five with a measurable level of extraversion ingrained in that. As this is an online experiment and looking at the inclusiveness of the criteria mentioned before, the sampling method used was that of a non-probability convenience sample. Using social media channels such as Facebook and direct communication respondents were recruited in a snowball way.
Variables
The manipulated independent variable is the level of extraversion of the political advertisement. This is high for the experimental group and neutral for the control group. To create the stimuli the fundamental characteristics of extraversion were taken into account, described as “a personality trait reflecting the extent to which people seek and enjoy company, excitement and stimulation. People scoring high on extroversion are described as energetic, active, talkative, sociable, outgoing, and enthusiastic” (Matz et al., p.2, 2017; McCrae & John, p.178, 1992). Hirsch, Kang & Bodenhausen (p.579, 2012) in turn mention that ”extraverts are especially sensitive to rewards and social attention”.
The targeting of extraversion has been translated into stimuli in different ways. Segalin et al. (2016) found that extroverts preferred images containing people and faces and with high contrasts in hue and saturation, and this is something that was used in the design of our stimulus as well as in the Matz et al.’s study (2017), for example. When it comes to the text and word use most studies use words such as “excitement”, “energetic”, “active”, “bold” (Matz et al., p.2, 2017; Hirsch, Kang & Bodenhausen, 2012)
The stimulus that was designed for the extravert condition took these things into account, resulting in a political poster with more color, more saturation, more faces and more
15 energetic, enthusiastic phrases, when compared to the control group’s neutral poster (see appendix A). This poster was in black and white (see appendix B), neutralizing the effect of contrast and saturation, and contained more neutral phrases than the one in the experimental condition (For example: “For a city where everyone gets a chance” vs “For a vibrant and active city where everyone gets a chance!”).
To measure extraversion the ‘Big Five Aspect Scales’ from DeYoung, Quilty, & Peterson (2007) was used, just as in Hirsch, Kang & Bodenhausen’s (2012) experiment (see Appendix E). This scale has been widely used in the literature and measures extraversion using twenty 5 point likert type scale questionnaire items and was confirmed to be highly reliable in this study (M=3.62, SD=.43, α= .837). This scale was subsequently subdivided into two parts, one related to assertiveness (M=3.67, SD=.49, α= .766) and one related to enthusiasm (M=3.57, SD=.48, α= .741) each variable measured by ten of the previously mentioned twenty 5 point likert type scale questionnaire items. This was done in the same way as in the DeYoung, Quilty, & Peterson study (2007). The dependent variables measured in this experiment were voter persuasion and rated attractiveness of the message. To measure voter persuasion the question “How likely are you to vote for this party based on this advertisement?” was posed using a scale ranging from 0 (highly unlikely) to 100 (highly likely) (M=19.8, SD=19.09) (see appendix D). This is the same item used in Hersch & Schaffner’s (2013) study about the effectiveness of identity targeting in politics. To measure perceived attractiveness of the message, the six item 5-point likert scale used in Hirsch, Kang & Bodenhausen’s (2012) study was slightly adapted to fit the current study (see appendix C). One item was eliminated, resulting in the following five items: “I find this advertisement to be persuasive,” “This is an effective advertisement,” “Overall, I like this advertisement,” “This advertisement has made me more interested in the candidate/party” and “I am interested in learning more about this candidate/party after seeing this advertisement”, all
16 with a 1 to 5 response scale (M=3.62, SD=.43). A reliability analysis showed that this 5 item index was highly reliable (α= .846).
Research Design
Participants in the experiment were first asked a number of standard questions (age, sex, political orientation and interest, education level, intention to turnout to vote in the upcoming elections) and then randomly allocated to either the control group or the experimental group. They were subsequently shown the stimulus; a political advertisement from a non-existing political party (AP) running for the Dutch municipal elections. These elections where chosen because they are the upcoming ones (March 2018) The party is fictional because it is important that participants do not hold previous thoughts or positions on the party/candidate used, as this could influence the results. After being shown the stimulus, which was either extravert or neutral as mentioned before, participants were asked a series of questions about it, as was mentioned above in the variables section. Following this, participants took the ‘Big Five Aspect Scales’ personality test so as to assess the level of extraversion for each one of them. This test was conducted after the experiment so as to keep it from possibly influencing results. This because taking the personality test before seeing the stimulus and answering the DV related questions could have made the participants conscious about their personality traits and in that way influence the rest of the answers.
Finally, a manipulation check was ran (see appendix F) by giving participants four options containing a sentence and asking them which of the options contained the sentence that was in the political ad they saw, as well as asking whether the poster they saw was in color or black and white. Following this the participants were debriefed and informed about the purpose of the experiment and about the fact that the mentioned party is fictional and created solely for this research.
17
Results
Randomization Check
The randomization check showed that randomization was successful for all but one characteristic (education). The experimental group scored significantly higher in education (M=4.33, SD= .6) than the control group ((M=4, SD=.74), t(89)=2.31, p=.023). The analyses in this study therefore control for that variable. Apart from this the experimental group was not significantly different then the control group in terms of age (M=30.35, SD=13.27) vs (M=29.78, SD=14.3), sex (M=1.35, SD=.48) vs (M=1.49, SD=.51), political ideological leaning (M=3.52, SD=1.62) vs (M=3.47, SD=1.25) and political interest (M=2.02, SD=.62) vs (M=1.89, SD=.71).
Manipulation Check
For the control condition 88.9% of respondents chose the correct sentence and also 88.9% correctly identified the poster as black and white. In the experimental condition 80.4% correctly identified the sentence that they saw on the poster and 97.8% of respondents
correctly identified the poster as being in color. A t-test showed that there was a significant difference between groups for both sentence choice (M. difference=1.36, t(88.89)=8.39, p<.001, 95% CI[1.04,1.69]) and color of the image choice (M. difference=-.91, t(60.57)=-16.94, p<.001, 95% CI[-1.02, -.80]). Therefore, the manipulation was successful.
Hypotheses Testing
To test the hypotheses of this study a number of moderation regressions (using the Hayes PROCESS macro in SPSS) were ran. The independent factor is condition (neutral vs extravert), the dependent factors are perceived attractiveness and voting intention and the moderator is level of extraversion of the participant (which was subsequently divided in
18 enthusiasm and attractiveness). The following tables present the results from the models that were ran, firstly for predicting the effect on voting intention and secondly for predicting the effect on perceived attractiveness of the message. All models are controlled for education.
Table 1. Regression models to predict perceived attractiveness of message
Model 1** Model 2* Model 3**
Constant 2.98*** Extraversion -.27 Condition .44* Interaction .64 Constant 2.95*** Enthusiasm -.18 Condition .45* Interaction .41 Constant 2.93*** Assertiveness -.23 Condition .43* Interaction .59 Overall Model R2 .13** .11* .14* F 4.33** 3.24* 3.42* ΔR2 .03 .01 .03 N 91 91 91
19
Table 2. Regression models to predict voting intention
Model 4* Model 5 Model 6*
Constant 48.12*** Extraversion -9.51 Condition 1.78 Interaction -2.52 Constant 46.36*** Enthusiasm -4.24 Condition 1.01 Interaction 4.5 Constant 48.66*** Assertiveness -10.64* Condition 2.52 Interaction -7.97 Overall Model R2 .1* .07 .13* F 2.6* 1.75 3.34* ΔR2 .0008 .003 .01 N 91 91 91
20 Looking at the results from the regression it is now possible to test the hypotheses derived from the theory.
The first hypothesis is the expectation that an extrovertly framed political message will have a positive impact on perceived attractiveness of the message, but only for participants with higher levels of extraversion. The overall model testing this hypothesis is significant (model 1, table 1) and can therefore be used to predict perceived attractiveness of the message. The strength of prediction is low: 13% of the variation in perceived attractiveness can be predicted based on the condition to which participants were allocated and their level of extraversion (R2=.13). The stimulus has a significant positive impact on perceived attractiveness, in general, (B = 44, t = 2.39, p=.019, 95% CI [2.03, 3.92]). This means that subjects in the experimental condition, on average, rated the poster as .44 points (on a 1 to 5 scale) more attractive than subjects in the neutral condition. According to the overall model, a one point increase on the extraversion scale (1 to 5 scale) is associated with a .27 point decrease in perceived attractiveness, but this result is not significant. The interaction effect is not significant and the same is true for the direct effect of extraversion on perceived attractiveness of the message. Looking at the model and above mentioned results, the first hypothesis is not confirmed. In fact, looking at the relation between these two variables (see fig. 2), the pattern is that extraversion actually leads to a lower level of perceived attractiveness in the experimental condition; running the reversed model showed it not to be significant (R2 = .12, p =.165). In the neutral condition there is a very small increase in attractiveness for higher levels of extraversion. These patterns that emerge from the analysis are actually the opposite of what was expected. This means that hypothesis 1 is rejected. The relations between the two variables are plotted in the following graph:
21
Fig.2- The relation between extraversion (low, medium, high) and perceived attractiveness of the message.
The second hypothesis is the expectation that an extrovertly framed political message will have a positive impact on voting intention, but only for participants with higher levels of extraversion. The overall model testing this hypothesis is significant (model 4, table 2) and can therefore be used to predict voting intention of the message. The strength of prediction is low: 10% of the variation in voting intention can be predicted based on the condition to which participants were allocated and their level of extraversion (R2=.1). The interaction effect is not significant, as well as the direct effect of extraversion on voting intention of the message and the same is true for the effect of condition, with the first two being negative and the latter positive. This means that being in the experimental condition is associated with a 1.78 increase in voting intention (1 to 100 scale), but this effect is not significant. The association between level of extraversion and voting intention is actually very close to significance (B = -9.51, t =3.9, p= .058, 95% CI [-19.38, .36] and indicates that a 1 point increase in extraversion is associated with a -9.51 decrease in voting intention. This means that
0 0,5 1 1,5 2 2,5 3 0 0,5 1 1,5 2 2,5 3 3,5 Per ce iv e d at tr ac tiv e n e ss
Average level of extraversion
Extravert C. Neutral C.
22 hypothesis 2 is not confirmed. The relationship between the variables is in fact the opposite of what was expected. These relations are plotted in the graph below. There we see that higher levels of extraversion are associated with lower voting intention in both conditions, the decline being larger for subjects in the extravert condition:
Fig.3- The relation between extraversion (low, medium, high) and voting intention.
As there is no evidence found for the first two hypothesis, it could be that it was only the component of enthusiasm that was targeted with the extrovert campaign poster. This leads us to the third hypothesis, which is the expectation that an extrovertly framed political message will have a positive impact on perceived attractiveness, but only for participants with higher levels of enthusiasm. The overall model testing this hypothesis is significant (model 2, table 1) and can therefore be used to predict perceived attractiveness of the message based on condition and level of enthusiasm. The strength of prediction is low: 11% of the variation in perceived attractiveness can be predicted based on the condition to which participants were allocated and their level of enthusiasm (R2=.11). The condition has a significant positive effect on perceived attractiveness, meaning that subjects found the extrovert stimulus to be more attractive than the neutral stimulus. The interaction effect is
0 5 10 15 20 25 30 0 1 2 3 4 Vo ting i n te n tion
Average level of extraversion
Extravert C. Neutral C.
23 slightly positive (B=.41) and non-significant, whereas the direct effect of enthusiasm on rated attractiveness of the message is non-significant but negative (B=-.18) This means that a 1 point increase in enthusiasm is associated with a .18 decrease in perceived attractiveness, but this effect is countered by the condition. Being in the experimental group and scoring one point higher on the enthusiasm scale is associated with a .41 increase in perceived attractiveness, but this effect is not significant. Considering the lack of significant results hypothesis 3 is not confirmed. Looking at the graph below the patterns do appear to be in line with the third hypothesis as we see a (very) slight increase in the extravert condition and a slight decrease in the neutral condition of perceived attractiveness, for higher levels of enthusiasm.
Fig.4- The relation between enthusiasm (low, medium, high) and perceived attractiveness of the message.
Looking at the same model for voting intention (model 5, table 2), we see that the overall model testing this hypothesis is not significant and the same is true for the effect of enthusiasm on voting intention. However, if would consider the model as a good predictor despite the lack of significance (R2= .07, p= .15), the observed relations between variables are
0 0,5 1 1,5 2 2,5 3 0 0,5 1 1,5 2 2,5 3 3,5 Per ce iv e d at tr ac tiv e n e ss
Average level of enthusiasm
Extravert C. Neutral C.
24 again not as expected. Enthusiasm appears to be negatively related to voting intention in both conditions (see fig.5). This makes sense for the neutral condition as there is a bigger mismatch there for increasing levels of enthusiasm, leading to penalization. Looking at the experimental condition however hypothesis 4 is not confirmed, as higher enthusiasm appears to lower voting intention. This negative effect is weaker than in the neutral condition but still the opposite of what was expected, meaning that the fourth hypothesis is not confirmed. The negative effect of enthusiasm on voting intention for both conditions is visible in the graph below:
Fig. 5- The relation between enthusiasm (low, medium, high) and voting intention.
The fifth hypothesis leads us to the extraversion component of assertiveness. The expectation is that subjects with higher assertiveness levels will be harder to persuade, as they have a stronger sense of agency and higher levels of dominance. This is expected to translate into a negative impact of assertiveness on voting intention. The overall model testing this hypothesis is significant (model 6, table 2) and can therefore be used to predict voting intention of the message based on condition and level of assertiveness. The strength of prediction is low: 13% of the variation in voting intention can be predicted based on the
0 5 10 15 20 25 0 1 2 3 4 Vo ting i n te n tion
Average level of enthusiasm
Extravert C. Neutral C.
25 condition to which participants were allocated and their level of assertiveness (R2=.13). The interaction effect is not significant, meaning that assertiveness has no significant effect as a moderator on the relation between condition and voting intention. The direct effect of assertiveness on voting intention is significantly negative though, meaning that H5 is confirmed (b= -10.64, t(86)= -2.29, p=.02, 95% CI [- 23.78, -1.41]) . The results indicate that for every one unit increase in assertiveness (1 to 5 scale) there is an associated decline of 10.64 points on voting intention (0 to 100 scale). Being in the experimental condition is associated with a 2.52 increase in voting intention, an effect which appears to be negatively moderated by assertiveness as the interaction effect B= -7.97, being non-significant. This negative effect appears to be stronger in the experimental condition where, as the graph below shows, the decline in voting intention is largest. Highly assertive subjects score lowest on voting intention, with the score being similar in both conditions. Participants with low levels of assertiveness appear to respond better to both stimuli in general and particularly well to the extrovert stimulus. This suggests that being assertive is an obstacle to being persuaded, whereas not being assertive makes persuasion easier, an effect which seems to be stronger for extrovertly framed messages.
Fig. 6- The relation between assertiveness (low, medium, high) and voting intention.
0 5 10 15 20 25 30 0 0,5 1 1,5 2 2,5 3 3,5 Vo ting i n te n tion
Average level of assertiveness
Extravert C. Neutral C.
26 In sum, this study finds that for both perceived attractiveness of the message and voting intention the experimental condition stimulus had an overall positive impact, meaning it was considered both more attractive and more persuasive than the control group stimulus. Contrarily to what was expected a higher level of extraversion was negatively associated (albeit not significantly) with voting intention and perceived attractiveness, an effect which appears to be stronger for the experimental condition. A higher level of enthusiasm was positively associated (albeit not significantly) with perceived attractiveness for participants in the experimental condition, which aligns with what was expected. This was however not the case for voting intention, where the pattern suggests a negative (non-significant) impact of enthusiasm for both conditions, being stronger for the experimental one. Looking at the results it could be that the significant negative impact of the assertiveness component of extraversion is one of the driving forces behind this, particularly for voting intention.
Discussion and Conclusion
The use of micro-targeting in political communication is a relatively new development but also one that has emerged and spread very quickly. One of the newest possible applications for these methods is by some claimed to be that of personality targeted messages. The aim of this study was to assess the effect of messages intended to speak to a specific personality trait on voting intention and perceived attractiveness of the message. The chosen trait was extraversion, as this emerged from the literature as the most predictive of behavior.
This research did not find that messages that were targeted at extrovert people had a bigger persuasive impact, for subjects with high levels of extraversion, than messages that were not. This is something which contradicts the evidence from the fields of commercial marketing and health communication (Matz et al., 2017; Kalyanaraman & Sundar, 2006;
27 York, Brannon & Miller, 2012; Hirsch, Kang & Bodenhausen, 2012; Moon, 2002). The response pattern of extrovert subjects to the extrovert message actually suggests a negative impact of congruence, both for voting intention and perceived attractiveness of the message (albeit not-significantly). It is possible that this negative effect was up to some extent driven by the assertiveness component of extraversion, which was found to have a significant negative association with voting intention. On the other hand the expectation was that the enthusiasm component of extraversion would have a positive impact on voting intention and perceived attractiveness, as that was the component I thought the extrovert stimulus would most appeal to. This relation was not strong enough for perceived attractiveness and actually looked to follow the opposite pattern for voting intention.
There are a few possible explanations for why relations were not found to be as expected or even appeared to counter those expectations. The main one simply being that personality targeted political communication does not work. As previously discussed there is a lot of skepticism about its efficiency and there is no scientific evidence so far to back up the claim that it “does or doesn’t work” (Kreiss as quoted in Talbot, 2016). There have been no scientific studies about these methods and journalists that looked further into the claims about their use in the 2016 US election are also not convinced (Taggart, 2017). Various anonymous sources, from within the campaign teams that were interviewed for the Taggart article (2017), have claimed that psychographic persuasion was never proven to work and express doubts about whether it was actually even used at all. The evidence from the commercial marketing field suggesting it might work looks relatively promising, but the fact is that also in that field studies are scarce. Not only this but selling a product may well be very different from selling a political candidate or idea. The lack of support for the hypotheses in this study, as well as the findings that did emerge, suggest personality based targeting is not effective for political messages as it is for commercial messages; the indications pointing towards low or no
28
effectiveness. `
Further research will be necessary to determine that more precisely, something which is partially due to the many limitations of this study. The first one being that the stimulus possibly did not sufficiently speak to the trait of extraversion. This is a distinct possibility especially considering the fact that there is no theoretical framework in place on how to tailor political messages to certain personality traits. The stimulus was developed based on previous experiments from the commercial field (Matz et al., 2017; Hirsch, Kang & Bodenhausen, 2012; Moon, 2002) and on the preferences expressed by extroverts in terms of images posted on social media (Segalin et al., 2017). Possibly the application to political messages is different or the stimulus itself was not sufficiently speaking to those preferences, as due to limited resources it was not pre-tested. Besides the limitations about the creation of the stimulus there is also a number of other ones; including the sampling method, the size of the sample and the fact that only one trait was targeted.
The sampling method led to the sample not being representative of the electorate as a whole. Not only this but the fact that it was collected mainly from my social network may have led to a biased sample. Furthermore its size was quite limited (n=91) if compared to the studies mentioned before, limiting generalizability and reducing statistical power: something which leads to a realistic possibility of non-causal coincidental findings. Even considering these limitations of the sample, the fact is that the two groups were found not to be significantly different for all variables except education (for which the models were corrected). The manipulation check was successful and the scales were all found to be reliable (α > .7), giving the findings at least some relevance.
Another important limitation is that this study only looked at one of the Big Five personality traits. It could be that when it comes to political messages, extraversion is not the trait most predictive of behavior, as it was found to be in other fields. Possibly there are other
29 traits for which a targeted message will have a more successful impact or perhaps it is more effective to target a combination of traits.
Considering these limitations it is important that future research works with samples that are representative of the electorate and whose subjects are more thoroughly examined in terms of personality dimensions: testing for different traits and combinations of them. It is also important that a comprehensive framework is constructed of how personality traits are related and respond to different political messages, language use and visual characteristics. Looking at how different personality types respond to different uses of color, saturation, word usage and tone of the message (or type of emotion it speaks to) would be the first step in that direction. Besides this I suggest future research also looks into different types of media and messages, as this study only looked at political campaign posters. It could also be interesting to look at the effect of personality tailored messages on political interest and participation.
Despite the many limitations of this research, its indications are still relevant, both considering the lack of research in the area and the potential implications that personality based micro-targeting could have. The fact that this study found no evidence that personality based targeting works is only a first step, as the subject is new and can be developed and researched in many different ways. It is important that science keeps up with these possible developments, not only to test them for validity and effectiveness but also to facilitate the discussion about its ethical implications and possible dangers. One of the potential associated dangers is that of the erosion of the public sphere: as each person gets served its own tailor made version of the political reality this will have a negative impact on the exchange of ideas. On the other hand it is possible that it will improve political participation as people may feel more involved when they are presented with issues that are relevant to them personally in a way that speaks to them. It is also possible that these applications will coexist in the future or that there is no future for personality targeting in political communication at all. Either way,
30 the fact is that these developments call for research and discussion as they could greatly impact the field in particular and even at some point democracy and society in general. Furthermore, looking at the widespread societal discussion concerning the huge amounts of data that are gathered about us, as well as the associated privacy issues, it is crucial that campaigning methods which delve into the electorate’s data are transparent, public domain and scrutinized by the associated scientific field(s).
31
References
Barnard, Lisa, and Daniel Kreiss. 2013. “A Research Agenda for the Effects of Online Political Advertising: Surveying Campaign Practice 2000–2012.” International Journal of Communication 7:2046–2066.
Bennett, C. J. (2016). Voter databases, micro-targeting, and data protection law: can political parties campaign in Europe as they do in North America?. International Data Privacy Law, 6(4), 261-275.
Brock, T. C., Brannon, L. A., & Bridgwater, C. (1990). Message effectiveness can be increased by matching appeals to recipients’ self-schemas: Laboratory demonstrations and a national field experiment. In S. J.Agres & J. A.Edell (Eds.), Emotion in advertising:
Theoretical and practical explorations (pp. 285–315). New York: Quorum Books.
Dibb, S., & Simkin, L. (1991). Targeting, segments and positioning. International Journal of Retail & Distribution Management, 19(3).
DeYoung, C. G., Quilty, L. C., & Peterson, J. B. (2007). Between facets and domains: 10 aspects of the Big Five. Journal of personality and social psychology, 93(5), 880.
Depue, R. A., & Collins, P. F. (1999). Neurobiology of the structure of personality: Dopamine, facilitation of incentive motivation, and extraversion. Behavioral and Brain Sciences, 22, 491–569.
Grossmann, M. (2009, October). Do the strategists know something we don't know? Campaign decisions in American elections. In The Forum (Vol. 7, No. 3).
Hersh, E. D., & Schaffner, B. F. (2013). Targeted campaign appeals and the value of ambiguity. The Journal of Politics, 75(2), 520-534.
32 Hirsh, Jacob B, Sonia K Kang, and Galen V Bodenhausen (2012), “Personalized Persuasion: Tailoring Persuasive Appeals to Recipients’ Personality Traits.,” Psychological science, 23(6), 578–81,.
Holman, M. R., Schneider, M. C., & Pondel, K. (2015). Gender targeting in political advertisements. Political Research Quarterly, 68(4), 816-829.
Kalyanaraman, S., & Sundar, S. S. (2006). The psychological appeal of personalized content in web portals: does customization affect attitudes and behavior?. Journal of
Communication, 56(1), 110-132.
Kosinski, M., Stillwell, D., & Graepel, T. (2013). Private traits and attributes are predictable from digital records of human behavior. Proceedings of the National Academy of Sciences, 110(15), 5802-5805.
Kreiss, D. & Welch, C, (2015) Strategic Communication in a Networked Age In: Farrar-Myers, V. A., & Vaughn, J. S. (Eds.). (2015). Controlling the message: New media in American political campaigns. NYU Press.
Kreiss, D., & Howard, P. N. (2010). New challenges to political privacy: Lessons from the first US Presidential race in the Web 2.0 era. International Journal of Communication, 4, 19. Lutz, M. (2009). The social pulpit. Barack Obama’s social media toolkit, Edelman - Digital Public Affairs.
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
33
201710966.McCrae, R. R., & John, O. P. (1992). An introduction to the five‐factor model and
its applications. Journal of personality, 60(2), 175-215.
Moon, Y. (2002). Personalization and personality: Some effects of customizing message style based on consumer personality. Journal of Consumer Psychology, 12(4), 313-325.
Murray, G. R., & Scime, A. (2010). Microtargeting and electorate segmentation: data mining the American National Election Studies. Journal of Political Marketing, 9(3), 143-166.
Noar, S. M., Grant Harrington, N., Van Stee, S. K., & Shemanski Aldrich, R. (2011). Tailored health communication to change lifestyle behaviors. American Journal of Lifestyle Medicine, 5(2), 112-122.
Noar, S. M., Benac, C. N., & Harris, M. S. (2007). Does tailoring matter? Meta-analytic review of tailored print health behavior change interventions. Psychological bulletin, 133(4), 673.
Segalin, C., Cheng, D. S., & Cristani, M. (2017). Social profiling through image
understanding: Personality inference using convolutional neural networks. Computer Vision and Image Understanding, 156, 34-50.
Skinner, C. S., Campbell, M. K., Rimer, B. K., Curry, S., & Prochaska, J. O. (1999). How effective is tailored print communication?. Annals of behavioral medicine, 21(4), 290-298. Shipper, Burkhard C.; Woo, Hee Yeul (2014) : Political awareness, microtargeting of voters, and negative electoral campaigning, Working Papers, University of California, Department of Economics, No. 14-8
Willams, J. E., Munick, M. L., Saiz, J. L., & FormyDuval, D. L. (1995). Psychological importance of the" Big Five": Impression formation and context effects. Personality and
34 Social Psychology Bulletin, 21(8), 818-826.
Valerie K. York , Laura A. Brannon & Megan M. Miller (2012) Marketing responsible drinking behavior: comparing the effectiveness of responsible drinking messages tailored to three possible “personality” conceptualizations, health marketing quarterly,
29:1,49-65,doi:10.1080/07359683.2012.652578
References (Non-Scientific)
Anderson, B. & Horvath, B. (02-09-2017) The rise of the weaponized AI propaganda machine. Available at: https://scout.ai/story/the-rise-of-the-weaponized-ai-propaganda-machine
Confessore, N. & Hakim, D. (2017) Data Firm Says ‘Secret Sauce’ Aided Trump; Many Scoff. Ny Times, retrieved from: https://www.nytimes.com/2017/03/06/us/politics/cambridge-analytica.html
Doward, J. & Gibbs, A. (04-03-2017) Did Cambridge Analytica influence the Brexit vote and the US election. Available at: https://www.theguardian.com/politics/2017/mar/04/nigel-oakes-cambridge-analytica-what-role-brexit-trump
Grassegger H, Krogerus M (2017) The data that turned the world upside down. Available at: https://motherboard.vice.com/en_us/article/how-our-likes-helped-trump-win
Martijn, M. (24-02-2017). Zo gebruiken politieke partijen jouw data om je stem te winnen. Retrieved from: https://decorrespondent.nl/6244/zo-gebruiken-politieke-partijen-jouw-data-om-je-stem-te-winnen/192040464-fa359e58
Neal, M. (2016). You're Going to Need an Ad Blocker for Your Next TV. Retrieved from: https://motherboard.vice.com/en_us/article/mg7ek8/youre-going-to-need-an-ad-blocker-for-your-next-tv
35 Taggart, K. (2017) The Truth About The Trump Data Team That People Are Freaking Out About. Buzzfeed, retrieved from: https://www.buzzfeed.com/kendalltaggart/the-truth-about-the-trump-data-team-that-people-are-freaking?utm_term=.toz5778yV#.nd8Z779PD
Talbot, D. (15-04-2016). How political candidates know if you are neurotic. Technology Review, retrieved from: https://www.technologyreview.com/s/601214/how-political-candidates-know-if-youre-neurotic/
36 Appendix B: neutral stimulus
37 Appendix C: components of the scale measuring perceived attractiveness
38 Appendix E: components of the scale measuring extraversion
Ik maak gemakkelijk vrienden (E) Ik neem de leiding (A)
Ik ben moeilijk te leren kennen*(E) Ik heb een sterke persoonlijkheid (A) Ik hou anderen op een afstand* (E)
Ik heb niet het talent om anderen mensen te beïnvloeden* (A) Ik onthul weinig over mezelf *(E)
Ik weet hoe ik anderen moet boeien (A) Ik ben snel op mijn gemak bij mensen (E) Ik laat anderen het voortouw nemen* (A)
Ik laat me zelden door de opwinding bevangen* (E) Ik zie mezelf als een goeie leider (A)
Ik ben niet een heel enthousiast persoon*(E) Ik kan anderen overhalen om dingen te doen (A) Ik laat mijn gevoelens zien wanneer ik gelukkig ben(E) Ik hou mijn meningen voor mezelf* (A)
Ik heb heel veel lol (E)
Ik ben de eerste die actie onderneemt (A) Ik lach veel (E)
Ik heb geen assertieve persoonlijkheid* (A)
Note: *= items that were recoded into reversed scale. (E) = Enthusiasm components. (A)=
39 Appendix F: manipulation check items
