Inside the Voter’s Mind:
An Online Experiment on the Effects of Congruence between Voters’ Personality Profile and Candidate Message
Master’s Thesis by
Lennart Josef Krotzek Student ID: 11368098 lennart.krotzek@student.uva.nl
Graduate School of Communication
Master’s Programme Communication Science Supervisor: dr. Joost van Spanje
January 26, 2018
Abstract
The 2016 U.S. presidential election caused turmoil in the media and in academia alike. Both scholars and the media ever since have been discussing the impact of technology-driven
campaign tools, like online microtargeting, on the election outcome. While traditionally political campaigns have mainly relied on demographic data to target potential voters, technological developments allow campaigners to analyze voters’ psychological profiles and to adapt political advertisements accordingly. Previous research on commercial advertising has shown that
psychometric microtargeting can improve attitudes and purchases. However, little is known about the underlying processes behind the persuasiveness of these messages. Hence, this thesis sheds light on the effects of psychometric microtargeting and aims to answer the question why candidate messages which are congruent with the receiver’s personality profile can be more persuasive than non-congruent ads. A U.S.-based online experiment (N = 199) reveals that congruent candidate messages improve the feeling towards the candidate while not significantly affecting the propensity to vote for the candidate. The proposed mediators cognition, emotion and trust are not significant. These results inform the debate on the potential of psychometric microtargeting to manipulate voters.
Keywords: political communication, election campaigns, psychometric microtargeting, voter persuasion, media effects, advertising
Inside the Voter’s Mind: An Online Experiment on the Effects of Congruence between Voters’ Personality Profile and Candidate Message
When in December 2016 the Swiss magazine Das Magazin released a story about the role of the data analysis company Cambridge Analytica in the 2016 U.S. presidential election,
reactions ranged from shocked to skeptical (Grassegger & Krogerus, 2016; Grassegger & Krogerus, 2017). In short, the article claims that the data company employed a sophisticated system of data exploitation and microtargeting based on personality profiles, and thereby
decisively changed the election outcome in favor of Donald Trump (Cadwalladr, 2017). Research shows that it is indeed possible to predict people’s personality profiles from their likes and shares on Facebook (Kosinski, Stillwell, & Graepel, 2013). It was also found that such psychometric microtargeting improves purchase rates (Matz, Kosinski, Nave, & Stillwell, 2017). However, only few studies have investigated the effects of this technique on voting behavior (Kruikemeier, Sezgin, & Boerman, 2016). The present study fills this gap in the field of psychological
persuasion (Matz et al., 2017) by answering two research questions. The first research question refers to the overall effects of psychometric microtargeting:
To what extent does a political advertisement for a candidate which is tailored towards a recipient’s personality profile affect their feeling towards the candidate and the propensity to vote for the candidate compared to a non-tailored advertisement?
While the first research question as well as previous research on microtargeting in the commercial sector has, in most cases, focused on the outcome of such microtargeting techniques, this thesis not only argues that psychometric microtargeting is an effective tool to persuade
voters. It also analyzes the underlying processes of these effects. Three variables which are predicted to mediate the effect of microtargeted messages on feeling towards the candidate and propensity to vote for the candidate are therefore tested. The second research question asks:
To what extent is the effect of political advertisement for a candidate which is tailored towards a recipient’s personality profile mediated by cognition, emotional response to the ad, and trust in the candidate?
Political microtargeting becomes an increasingly interesting topic for campaigners and scholars alike as the recently published special issue of the Internet Policy Review on political microtargeting and many other publications demonstrate (Bodó, Erickson, Katzenbach, Musiani, & v. Hoboken, 2017). And yet, recent research in this field has, in most cases, taken macro- or meso-level approaches, focusing on recent developments (Dobber, Trilling, Helberger, & de Vreese, 2017; Chester & Montgomery, 2017), democratic implications (Bodó, Helberger, de Vreese, 2017; Kreiss, 2017) and discussions of practical feasibility (Kruschinski & Haller, 2017).
The present study, on the contrary, adds to the existing literature by assessing the effects of psychometric microtargeting in campaigns on the individual level. This micro-level approach is an important contribution to the public and academic debate about potential voter
manipulation: by exploiting voters’ individual weaknesses campaigners could, for example, persuade voters to take decisions which are contradictory to their actual beliefs (Matz et al., 2017). Furthermore, the the assessment of underlying processes can start a debate about the implication of campaign regulations. After all, this research attempts to open the black box of
microtargeting effects to lay the foundation for updated advertising effect models in times of advanced technology-driven campaign tools.
Theoretical Framework Political Microtargeting
Traditional targeting can be defined as “the process of identifying groups of voters based on some characteristics and sending messages to those groups” (Hersh & Schaffner, 2013, p. 522). Hence, targeting is a superordinate concept which neither specifies the size of the actual target nor the level of detail of the data on which the segmentation is based. This practice has long been employed in political campaigns, for example by selecting a specific channel and medium for campaign messages, or by targeting specific voting districts (Franz, 2013; for a historical overview of political targeting see Fulgoni, Lipsman, & Davidsen, 2016; Turow, Delli Carpini, Draper, & Howard-Williams, 2012).
In recent years, political targeting has been further developed to microtargeting.1 The advent of microtargeting was enabled by technological developments which allow campaigners to collect and analyze vast amounts of voter data (Fulgoni, Lipsman, & Davidsen, 2016) and to address voters directly (Magin, Podschuweit, Haßler, & Russmann, 2017). In the U.S. in particular, with its limited financial campaign regulations and lenient privacy protection rules,
1 Microtargeting should not be confused with personalized tailoring, individual targeting or nano-targeting which imply personalization tailored to a single individual (Barbu, 2014; Dijkstra, 2008; Franz, 2013).
targeting was therefore able to develop rapidly in the last two decades (Bennett, 2016; for a legal overview for the American context see Rubinstein, 2014).
The data which inform the microtargeting process are gained by tracking citizens’ online behavior (e.g., Farahat & Bailey, 2012; Schumann, v. Wangenheim, & Groene, 2014; Yan et al., 2009) as well as by employing traditional market research tools like surveys (Franz, 2013; Schumann et al., 2014). The collected data are then used to compute predictive algorithms (Barbu, 2014; Rubinstein, 2014).
Since the 2016 U.S. presidential election the media focus, as well as to a certain extent the academic focus, lies on online microtargeting, specifically the use of Facebook ads (e.g., Kreiss & Mcgregor, 2017). Further, the election campaigns revealed that microtargeting has become increasingly complex due to the growing amount of available data as well as the increasing knowledge about voter segments. Therefore, the 2016 campaigns were another step towards ever-more technology-intensive campaigns (Kreiss, 2016; for an overview of data driven campaigning see Kreiss & Welch, 2015).
The increasing relevance and complexity of these techniques call for an empirical substantiation of microtargeting effects. Research in the fields of health communication or commercial advertising generally support the notion of a higher persuasiveness of microtargeted ads (e.g., Kreuter, Bull, Clark, & Oswald, 1999; Noar, Benac, & Harris, 2007; Yan et al., 2009). However, more critical voices claim that the effects of microtargeting are overestimated (Farahat & Bailey, 2012), and that mistargeting can even lead to negative effects (Hersh & Schaffner, 2013). To assess the effectiveness of this technique it is therefore necessary to not only look at the outcome of microtargeting but also to develop a theoretical framework which explains differential susceptibility to microtargeted ads and their underlying processes.
More specifically, this research examines psychometric online microtargeting; a form of targeting which is based on psychological user profiles. The general assumption of psychometric microtargeting is that the assessment of social media profiles allows us to infer users’ personality profiles (Kosinski, Stillwell, & Graepel, 2013). This psychometric information is often presented as the Big Five personality trait dimensions, a widely accepted framework to measure personality profiles. This hierarchal framework suggests that the multitude of specific personality
characteristics can be categorized in five broad personality dimensions, namely extraversion, neuroticism, agreeableness, openness and conscientiousness (Gosling, Rentfrow, & Swann Jr., 2003; for a review and historical overview of different personality trait scales see Eysenck, 1991; John & Srivastava, 1999). Extraversion can be associated with characteristics like sociability and activeness. Neuroticism is often linked to anxiety and anger. Agreeableness describes traits like tolerance and cooperativeness. Conscientiousness is characterized by achievement-orientation and hard-working. Finally, openness can be associated with intelligence and originality (Barrick & Mount, 1991).
Psychological targeting can be conducted in different ways, for example by adapting the message’s content and formulation to the receiver’s personality profile (Dijkstra & de Vries, 1999). Data analysis companies claim that messages which are tailored towards specific
personality profiles increase their persuasiveness (Cadwalladr, 2017). This claim finds support in two experimental studies in the commercial context in which persuasive messages were adapted to the receiver’s personality profile. The studies show that congruence between the ad’s message and the receiver’s personality improve the ad evaluation (Hirsh et al., 2012) and increase clicks and purchases (Matz et al., 2017).
In the field of political campaigns this tailoring strategy has not been tested yet. However, there is a wider body of research on voters’ personality trait profiles and their correlations with political attitudes and behavior. Personality traits can, for example, predict ideology (Bakker, 2017), the strength of party identification (Bakker, Hopmann, & Persson, 2015), electoral
volatility (Bakker, Klemmensen, Nørgaard, & Schumacher, 2016) and voting for populist parties (Bakker, Schumacher, & Rooduijn, 2016).
The Differential Susceptibility to Media Effects Model
While most of the research in the field of microtargeting focuses on the outcome of the messages, the question remains: what are the underlying processes behind the effectiveness of microtargeting?, or: why do people react more positively to a candidate ad which is particularly targeted towards their personality trait profile?
The Differential Susceptibility to Media Effects Model (DSMM) by Valkenburg and Peter (2013) is used as a structural framework to answer this question. The DSMM attempts to model media effects and their underlying processes. It claims, among other things, that media response states, namely cognition, emotion and excitation, mediate media effects. Furthermore, it suggests that differential susceptibility variables, e.g., personality traits, function as both predictors of media use and as moderators of those response states (Valkenburg & Peter, 2013). This research examines the effect of dispositional susceptibility, in this case personality, on the relationship between media exposure and the emotional and cognitive response state. Further, the mediation effect of those response states on the effectiveness of the media content are examined. The cognitive response state describes “the extent to which media users (…) invest cognitive effort to comprehend media content”, while the emotional response state indicates the emotional reactions to the media content, and the excitative response state “refers to the degree of physiological
arousal” (Valkenburg & Peter, 2013, p. 228). For this study only the emotional and the cognitive response states are further discussed.2
Persuasiveness of Psychometric Microtargeting
This research assumes that the higher the fluency with which a receiver processes a message, the more easily they can extract information from the message and the more positive reactions are elicited (Winkielman & Cacioppo, 2001). Fluency can be increased by adapting the message’s content to the receiver’s dispositional characteristics, e.g., their personality profile (Alter & Oppenheimer, 2009; Schwarz, 2004).3 To improve fluency based on psychological considerations, the persuasive message has to be adapted to the motivational concerns of the receiver. This can be achieved by, for example, addressing issues and concerns which appear to be important for the receiver. Fluency is assumed to be high when the sent message content is perceived as salient. The sender of the message intends to make the receiver believe that they
2 Not only is physiological excitement difficult to measure with the available methods for this thesis, excitement is also not expected to be as much affected by the stimuli as the other response states. To substantiate this argument arousal was measured in the experiment. As expected, the stimuli yielded only little excitement, here operationalized as arousal (M = 2.97, SD = 1.90, 9-point scale, 1 = lowest levels of arousal and 9 = highest level of arousal). 38.19% of the participants even indicated the lowest level of arousal.
3 Fluency can be established in many different ways. For linguistic fluency, prospective imagery fluency or conceptual fluency see for example Alter and Oppenheimer, 2009, and Reber, Winkielman, and Schwarz, 1998.
care about their concerns and issues as much as the receiver. The candidate, thus, aims to
establish a perceived similarity between themselves and the potential voter.4 Personality profiles can be used to predict concerns and salient issues of voters. Here, personality traits are not conceptualized as merely descriptive categories but as predictors of certain motives and needs in social interactions (Denissen & Penke, 2008). This enables the candidate to address issues and concerns, which are characteristic to a specific personality trait, to thereby establish perceived similarity and finally a more fluent processing of the ad.
As with other forms of fluency (see Alter & Oppenheimer, 2009), this form of personality fluency leads to a more positive judgment of the content through a higher cognitive response, a more emotional response, and increased trust, among others. Based on these theoretical
considerations and the aforementioned experimental studies, a higher propensity to vote and a more positive feeling towards the advertised candidate in a political ad can be predicted:
H1a: Candidate messages which are congruent with the receiver’s personality trait profile elicit a higher propensity to vote for the candidate than advertisements which are less congruent.
4 Similarity can be established in many ways, for example by imitating behavior or physical appearance. For the effects of similarity between sender and receiver see Bailenson, Iyengar, Yee, and Collins (2008), Byrne (1961), Heider (1958), and Patton and Kaericher (1980). In this case, a mere focus on imitation does not seem to be promising. A neurotic person, for example, is not expected to prefer a candidate who is neurotic themselves but rather a candidate who addresses the concerns of neurotic recipients (Denissen & Penke, 2008).
H1b: Candidate messages which are congruent with the receiver’s personality trait profile elicit a more positive feeling towards the candidate than advertisements which are less congruent.
Further, this research suggests that the feeling towards the candidate is positively related to the propensity to vote for the candidate. Experimental research shows that likeability is an important, but not the only factor for voting decisions (e.g., Lodge, Steenberge, & Brau, 1995; Patton & Kaericher, 1980). In a study among UK citizens Shephard and Johns (2008) find, for example, that when voters associate the characteristic warmth with the candidate they are more likely to vote for them. Hence, feeling towards the candidate is expected to mediate the effect of the candidate message:
H1c: The more positive the feeling towards the candidate the higher the propensity to vote for the candidate.
H1d: Feeling towards the candidate mediates the effect of congruence on the propensity to vote for the candidate.
Figure 1
Cognitive response.
The DSMM predicts that the cognitive response state mediates media effects (Valkenburg & Peter, 2013). Presumably, congruent candidate messages trigger a higher cognitive response state than non-congruent messages. This can be explained by the Limited Capacity Model of Motivated Mediated Message Processing (Lang, 2017). This model assumes that media users have a limited capacity to process information and that “the processing of congruent content requires less cognitive effort, which leaves more resources available for the processing of less salient content” (Valkenburg & Peter, 2013, p. 232). Congruence between the advertised candidate and the receiver, i.e., higher fluency, is, therefore, expected to require less capacities for the orienting response and to leave more capacities for cognitive evaluations. This model finds support in research on self-schemata which show that processing is faster, and retrieval from memory is better when the external stimuli are in line with one’s own self-schema due to a higher perceived relevance of the message (Hong & Zinkhan, 1995).
Following on from this, high cognitive involvement is expected to result in a more immediate persuasive effect while the effects in a low-involvement condition only appear after repeated exposure (Krugman, 1965). Since this experiment does not measure longitudinal effects, one can expect higher persuasive effects in line with the persuasive message when participants are more involved. This notion finds support in Social Cognitive Theory (Bandura, 2009), Cultivation Theory (Shrum, 2009), and Uses-and-Gratification Theory as Valkenburg and
Peter (2013) summarize.5 Based on these theoretical considerations, the following hypotheses can be formulated:
H2: The cognitive response state mediates the effect of congruence on feeling towards the candidate.
H2a: Candidate messages which are congruent with the receiver’s personality trait profile elicit a higher cognitive response than messages which are less congruent.
H2b: The higher the cognitive response state the more positive the feeling towards the advertised candidate.
Emotional response.
According to the DSMM emotional responses also mediate media effects (Valkenburg & Peter, 2013). An adapted candidate message is expected to yield stronger and more positive emotional involvement. This prediction is supported by the Hedonic Fluency Model which claims that the positive effect of fluency on evaluation judgments is mediated by a positive affective response (Winkielman & Cacioppo, 2001). According to this model, a fluent message
5 One could also expect a higher message scrutiny when participants are more involved which can result in a stronger resistance to the persuasive message (Sherif & Hovland, 1961; for an overview of the effects of involvement see Greenwald & Leavitt, 1984). A boomerang effect – which occurs when high involvement leads to counterarguing because the message conflicts with pre-existing opinions (Greenwald & Leavitt, 1984) – is not expected in this experiment since the persuasive messages are designed in such generic manner that they are not expected to elicit negative attitudes
elicits a spontaneous affective response before the evaluation takes place and increases the positive evaluation of the message due to perceptual fluency. Multiple studies support this assumption (e.g., Reber, Winkielman, & Schwarz, 1998; Winkielman & Cacioppo, 2001).6 Transportation Theory (Green, Brock, & Kaufmann, 2004) further supports the prediction that a more emotional response leads to more positive feelings towards the advertised subject. Based on these theoretical and empirical assumptions this study hypothesizes:
H3: The emotional response state mediates the effect of congruence on feeling towards the candidate.
H3a: Candidate messages which are congruent with the receiver’s personality trait profile elicit a higher emotional response than messages which are less congruent.
H3b: The higher the emotional response state the more positive the feeling towards the advertised candidate.
Trustworthiness.
A third factor which is expected to be positively affected by congruent personality traits is not part of the DSMM yet seems relevant for the study of the persuasiveness of political ads. Fluency increases trust in the source, i.e., “the honesty, integrity and believability of an endorser” (Erdogan, 1999, p. 297; McCroskey & Mehrley, 1969). In addition, according to the Source Credibility Model (Hovland & Weiss, 1951) higher perceived trust as one dimension of
6 The positive effect of positive emotions is even stronger when the receiver is unfamiliar with the stimuli as Machleit and Wilson (1988) show in a study on attitudinal change towards brands.
credibility of the source (McCroskey & Mehrley, 1969) increases the effectiveness of a message (Erdogan, 1999). Furthermore, similarity leads to higher trustworthiness (Erdogan, 1999). As mentioned before, similarity occurs when the candidate’s message matches the prevalent issues and concerns of the receiver. Research on patients’ trust in care providers shows, for example, that an important predictor of trust is fidelity, i.e., “caring and advocating for the patient’s interests or welfare and avoiding conflicts of interest” (Hall et al., 2002, p. 298).
Scientific evidence on the effect of trust on feeling towards a candidate is less clear (for a critical review of the role of trust in politics see Levi & Stoker, 2000). However, studies which establish trust in politicians as an independent variable suggest a positive relationship to
candidate evaluations (e.g., Hetherington, 1998; Parker, 1989). Psychological studies also reveal that trust mediates political decisions (e.g., Marzi, Righi, Ottonello, Cinotta, & Viggiano, 2014). Olivola and Todorov (2010), for example, show that even a very short exposure to a candidate’s image can form voter’s beliefs about the candidate’s personality traits like trustworthiness which then affects important evaluation categories such as competence. Generally, interpersonal trust in a candidate is one of the most important attributes on which voters base their voting decision in the U.S. (e.g., Miller, Wattenberg, & Malanchuk, 1986). Hence, it can be expected that trust is positively related to feeling towards the candidate.
H4: Trust mediates the effect of congruence on feeling towards the candidate.
H4a: Candidate messages which are congruent with the receiver’s personality trait profile yield higher trust in the candidate than messages which are less congruent.
H4b: The higher the trust in the candidate the more positive the feeling towards the candidate.
Method Research Units
The sample population of this empirical research are American persons who are 18 or older. The choice for American participants stems from considerations concerning the stimuli design, i.e., the creation of a suitably realistic ad, and concerning the cross-national differences in campaign communication and electoral systems (Swanson & Mancini, 1996). The U.S. was selected due to its pioneering role in campaign tools, i.e., Americanization (Swanson & Mancini, 1996), the high degree of personalization in majority systems (Garzia, 2011), and the presence of a feasible election case in the near future, i.e., the congressional elections in November 2018.
205 participants from the U.S. took part in the experiment. Six were excluded because they did not meet the age criterion.7 The questionnaire was disseminated via social media channels. The data collection took place between November 16 and November 30, 2017.
Research Design
This study used a between-subjects experimental research design. An online experiment with five experimental conditions was conducted. Participants were informed that the survey is related to the field of communication science. They first filled out a personality test. After some buffer questions, each participant was randomly assigned to one of the five conditions. Next, the response states, the level of trust, the attitude toward the candidate as well as the propensity to
7 Five participants were too young while another case was excluded because they indicated an age of 120 years.
vote for the candidate were assessed. Before demographic data were collected a manipulation check had been conducted. The questionnaire closed with a short debriefing.
Independent Variable: Congruence between Participant’s Personality and Candidate Message
Stimuli.
Each participant was randomly assigned to one of five tailored Facebook ads for a political candidate.8 For each personality dimension one tailored ad was designed. Only the textual claims in the ads differ from each other in order to ensure comparability. All ads were designed in the style of a sponsored post on the social media platform Facebook. The posts advertise the fictional candidate Bob V. Waterman and were tailored in a way that they are more appealing to the
recipient by adapting the messages’ formulations and contents to the receivers’ characteristics, i.e., the motivational concerns of their personality profile (Dijkstra, 2008).9 The message for the extraversion ad focuses on rewards and social attention (Denissen & Penke, 2008; Lucas, Diener, Grob, Suh, & Shao, 2000), the text for agreeableness on cooperation and interpersonal harmony
8 Participants (N = 199) were distributed among conditions as following: openness (n = 40), conscientiousness (n = 37), extraversion (n = 41), agreeableness (n = 41), neuroticism (n = 40).
9 Before participants were exposed to the ad they were briefed about the political context of the ad. They were told that the candidate will run as an independent candidate in the fifth congressional district in Oregon (U.S.) in the November 2018 congressional elections.
(Denissen & Penke, 2008; Graziano & Eisenberg, 1997), the ad for conscientiousness on goal pursuit, order, and efficiency (Denissen & Penke, 2008; Roberts, Chernyshenko, Stark, & Goldberg, 2005), the message for the neuroticism ad on threats, uncertainty, and sensitivity to social exclusion (Carver, Sutton, & Scheier, 2000; Denissen & Penke, 2008; Hirsh & Inzlicht, 2008), and the openness message on creativity, innovation, and reward value of cognitive activity (Denissen & Penke, 2008; McCrae & Costa, 1997). Hirsh, Kang and von Bodenhausen (2012) conducted a similar experiment with tailored ads for a cellular phone. Their manipulations worked successfully which is why their general tailoring strategy was adapted for the political context in this study (the stimuli can be found in appendix A).
A candidate ad, in contrast to a party ad or issue-related ad, was selected to increase the experiment’s external validity. Participants were supposed to believe the candidate was actually running for office. In majority systems, such as the U.S., political parties play a less important role in the electoral competition than in proportional electoral systems. Further, campaigns primarily evolve around individual candidates who often strongly focus on personal
characteristics (Swanson & Mancini, 1996). In addition, on social media personalized messages are more common than, for example, party messages (Enli & Skogerbø, 2013). The introduction of a new independent candidate without strong indications about his position on policy issues was therefore deemed the most realistic and hence most believable as well as the most feasible scenario. In order to further increase the experiment’s external validity a 30-second time limit for the exposure to the stimulus was set. In this way, a social media experience was simulated. The stimuli were pilot-tested and adjusted in line with the received feedback.
For the manipulation check participants were asked to rate the candidates on a Ten-Item Personality Inventory scale (Gosling et al., 2003). The original wording of the items was
adjusted for the evaluation of a different person (e.g., “Waterman seems like someone who is reserved”). Participants could indicate how much they agree with the statements on a 5-point scale (1 = disagree strongly and 5 = agree strongly). This scale was chosen due to its shortness and its adaptability to an interpersonal evaluation (in contrast to self-evaluation). For each dimension the mean of both items was calculated resulting in a scale ranging from 1 to 5 with nine possible values.
Personality trait measure.
There is a large variety of personality trait measures. The selection of Mini-IPIP10
(Donnellan, Oswald, Baird, & Lucas, 2006) is based on the criteria of reliability and feasibility. Some very short measures with only two items per dimension did not seem adequate but at the same time long measures can take up to 60 minutes which is not feasible for an online
experiment. Hence, participants’ personality profiles were assessed with a 20-item scale introduced by Donnellan et al. (2006) based on Goldberg’s 50-item IPIP (1999). Participants could indicate their answer on a 5-point answer scale (1 = disagree strongly and 5 = agree strongly). Based on the answers for the four items of each personality dimension an index score between 1 and 5 was computed for each dimension (1 = low score on dimension X and 5 = high score on dimension X).
Congruence.
The independent variable congruence describes the congruence between the receiver’s personality profile and the candidate message (1 = low congruence and 5 = high congruence; M
= 3.58, SD = 0.99). The variable was calculated by selecting the participant’s personality score of the dimension which corresponds with the experimental condition to which they were assigned. For participants who, for example, were exposed to the neuroticism ad, their score on the neuroticism dimension was selected.
Dependent Variables
Emotional response state.
A Self-Assessment Manikin (SAM) test was employed to measure the emotional state (Bradley & Lang, 1994). A SAM is a measurement for all three dimensions of emotional states: (dis)pleasure, (non-)arousal, dominance/submissiveness. The (dis)pleasure category indicates the emotional response state. The dimensions and their levels are presented in a graphic character which yields 9-point scales on each dimension (Bradley & Lang, 1994; 1 = displeased/non-aroused/submissive and 9 = pleased/aroused/dominant). A SAM has been used in multiple studies on advertising effects (e.g., Meagher, Arnau, & Rhudy, 2001; Morris & Boone, 1998, Morris & McCullen, 1994; Morris, Woo, Geason, & Kim, 2002) and is a quick and easy measure of affective response that takes participants’ incapability to judge their own emotional reaction into account (Bradley & Lang, 1994; M = 5.30, SD = 1.62).
Cognitive response state.
To assess the cognitive response state, participants were asked to write down all thoughts which were elicited by the ad as proposed by Cacioppo and Petty (1981). No time limit was set and participants could use as much as 25 open field boxes to write in their thoughts. A
dimensional distinction as suggested by Cacioppo and Petty (1981) was not deemed necessary for this experiment since the only relevant factor is the quantity of thoughts triggered by the stimuli. A scale from 0 to 25 indicates the cognitive state. The more thoughts a participants wrote
down the more cognitive processing took place (see appendix B3 for the introduction to the thought list, M = 4.54, SD = 3.06).
Trustworthiness.
To assess the level of trustworthiness participants were asked to indicate how much they trust the candidate on a 7-point scale (1 = not at all trustworthy and 7 = very trustworthy; M = 3.24, SD = 1.51; see for example Garramone & Smith, 1984).
Feeling towards the candidate.
Feeling towards the candidate was measured with a feeling thermometer as, for example, was used in the American National Election Study Time Series Studies (ANES, 2017).
Participants were asked to indicate their attitude towards the candidate by moving a slider from 0 to 100 (0 = not feeling favorable towards the candidate and 100 = feeling most favorable towards the candidate; M = 48.09, SD = 24.83; see appendix B5).11
Propensity to vote.
The propensity to vote for the candidate was measured by asking participants how probable it is that they will ever vote for this candidate (0 = not at all probable and 10 = very probable; M
11 In the original data set 16 cases were coded as missing. These cases have been recoded as the value 50. The slider for this scale was automatically set on 50. There was no forced response set for this item. It can be assumed that participants who did not change the position of the slider – and therefore were first coded as a missing value – actually intended to evaluate the candidate with the value 50.
= 4.17, SD = 2.42). This measurement was used, among others, in the European Election Studies (Schmitt, Popa, & Devinger, 2015).
Control Variables
In order to control for demographic characteristics participants were asked to indicate their gender (80 male, 105 female and 14 participants identify as neither male nor female), age in years (M = 37.72, SD = 18.28, min. = 18, max. = 91), highest reached level of education (17 answer categories, Mo and Mdn = Bachelor’s degree, see appendix C1), interest in politics on a 7-point scale (1 = not interested at all and 7 = very interested, M = 5.24, SD = 1.69, see appendix C2), political orientation on an 11-point scale (0 = left and 10 = right, M = 3.61, SD = 2.60, see appendix C3), and party affiliation (87 Democrats, 33 Republicans, 18 third party12, 61 do not feel close to any party).
Results
The analysis of the theoretical model introduced above has been conducted in two steps. For both steps, model four of Hayes’ PROCESS macro for SPSS (2017) was used. First, the relationship of congruence and feeling toward the candidate and the three hypothesized
mediators cognition, emotion and trust were assessed. Through this, the total and direct effect of congruence on feeling towards the candidate as well as the mediation effects could be tested.
12 Nine participants specified to be Libertarian, one to be a Democratic Socialist, four to be affiliated with the Green Party, one with the Socialist Party of America, one indicated
Some control variables have been included in order to rule out third factors.13 In a second step, the model was tested for the relationship between congruence and propensity to vote for the candidate including the predicted mediator feeling towards the candidate and the same control variable as in the first model.14 The choice for bootstrapping to test for the significance of mediations instead of Baron and Kenny’s (1986) causal steps approach is based on research conducted by Shrout and Bolger (2002) who claim that a non-significant effect of the
independent on the dependent variable should not be used as a reason to assume no mediation effects of predicted mediators.
In the following, I will first report the results of the manipulation check. Next, the results for each hypothesis will be presented. For an overview of the results see Figure 2.
Manipulation Check
For each condition an independent t-test was used to compare the means of the candidate evaluation on the dimension toward which the ad was tailored between those exposed to this condition and those who were not in this condition. The manipulation check reveals that
13 Gender (dummy), age, political orientation, party affiliation (dummy), and interest in politics.
14 The same tests were also conducted for each experimental condition individually revealing no significant effects.
generally the manipulations worked as desired, with the exception of the conscientiousness condition, but that the difference is not significant in any of the five cases (see appendix D1).15,16 The Effect of Congruence on Feeling towards the Candidate
H1b predicted a positive main effect of congruence on feeling towards the candidate. This hypothesis was tested by employing the PROCESS procedure in SPSS which reveals a significant positive total effect, b = 3.66, t = 2.03, p = .044, 95% CI = [0.10, 7.22]. This means that for each additional point on the 5-point congruence scale the feeling towards the candidate increases by 3.66 points on the 101-point feeling thermometer, while not controlling for the mediators but for the control variables. A participant whose personality is highly congruent with the candidate message therefore evaluates the candidate on average approximately 15 points better on the feeling thermometer than a participant whose personality is not congruent at all (see appendix D4 for a visualization of an exemplary case). The effect can be described as moderate. Hence, the result is in line with the predicted effect in H1b: congruence has a significant positive
15 Low group sizes as well as the very brief measurement of the personality traits might be the reason for the insignificant results.
16 In the following sections the results for all cases including the conscientiousness condition are reported. A second analysis excluding the conscientiousness condition (n = 162) reveals that the effect for H1b is slightly larger, b = 4.42, t = 2.17, p = .031, 95% CI [0.30, 8.44]. Further, the effect of emotion on feeling towards the candidate (H3b) is slightly larger when excluding the conscientiousness condition, b = 3.78, t = 1.22, p = .002, 95% CI [1.34, 6.19]. For all other hypothesis tests no substantial differences were found.
effect on feeling towards the candidate. The model also shows that the direct effect of
congruence on feeling towards the candidate, that is the effect when controlling for the tested mediators, is not significant, b = 2.22, t = 1.60, p = .111, 95% CI [-0.54, 4.94].17
Figure 2
Psychometric targeting effects model with unstandardized effect sizes (b) and significance ( a confidence interval does not contain the null hypothesis value. * p <.05. ** p <.01. *** p <.001)
Mediators: Cognitive Response State, Emotional Response State, Trustworthiness The effect of congruence on cognition is, holding all other variables constant, not significant, b = 0.33, t = 1.45, p = .148, 95% CI = [-0.12, 0.77]. The result is, therefore, not in line with H2a.18 Congruence neither has a significant effect on the emotional response state
17 The model is not significant, F(8, 190) = 1.565, p = .138. 18 The model is not significant, F(8, 190) = 1.430, p = .186.
controlling for the other variables, b = 0.12, t = 1.06, p = .302, 95% CI = [-0.109, 0.349].19 Hence, H3a has to be rejected. Congruence does not affect emotions significantly. The effect of congruence on trust is neither significant, b = 0.09, t = 0.82, p = 0.415, 95% CI = [-0.13, 0.31].20 Therefore, H4a has to be rejected, too.
A multiple regression model with the independent variables cognition, emotion and trust and the dependent variable feeling towards the candidate has been conducted to test H2b, H3b and H4b (see appendix D2). Forty-seven percent of the variation in the dependent variables can be explained by the variation in the independent variables (adj-R2 = .47). Therefore, the strength of the prediction is strong. All three predictors significantly affect the feeling towards the
candidate positively, holding the other variables constant. Trust has the largest effect: An increase of one point on the 7-point trust scale increases the feeling towards the candidate by 8.87 points on the thermometer scale while holding the other variables constant (b = 8.87). The effects of cognition and emotion are comparatively small: for each thought generated by the ad the feeling goes up by 1.06 points on the thermometer (b = 1.06), and for each increase on the emotion scale the feeling positively increases by 2.59 points (b = 2.59). Therefore, H2b, H3b and H4b are supported by the experiment.
However, the PROCESS mediation analysis reveals that none of the three variables cognition, emotion and trust mediate the effect of congruence on feeling towards the candidate
19 The model is significant, F(8, 190) = 2.459, p = .015. 20 The model is not significant, F(8, 190) = 1.401, p = .198.
significantly (see appendix D3). This result is not in line with the predicted mediation of these three variables. Thus, H2, H3 and H4 have to be rejected.
The Effect of Congruence on Propensity to Vote for the Candidate
The second PROCESS mediation test with congruence as an independent variable and propensity to vote for the candidate as the dependent variable, including control variables as well as the predicted mediator feeling towards the candidate, shows that the relationship between congruence and propensity to vote is not significant, b = 0.07, t = 0.38, p = .707, 95% CI [-0.27, 0.40].21 Therefore, H1a cannot be confirmed. Yet, feeling towards the candidate positively affects propensity to vote as predicted in H1c on a 99.9% significance level, b = 0.06, t = 10.43, p < 0.001, 95% CI [0.05, 0.07].22 For each increasing point on the feeling thermometer scale the propensity to vote goes up by 0.06 on the 10-point scale, meaning that an increase on the thermometer by approximately 17 points yields a full point higher on the variable propensity to vote for the candidate, which is a moderate effect (see appendix D5 for an exemplary
visualization).23 Therefore, feeling towards the candidate affects the propensity to vote for the candidate positively and significantly. This result supports H1c.
Further, the indirect effect between congruence and propensity to vote was tested by employing a bootstrap estimation approach with 5,000 samples, revealing that the variable
21 The model is significant, F(9, 189) = 2.745, p = .005.
22 Further, in this model congruence also affects the feeling towards the candidate significantly, b = 3.77, t = 2.09, p = 0.038, 95% CI [0.20, 7.35].
feeling towards the candidate is a significant mediator, b = 0.22, SE = .105, 95% CI [0.01, 0.43]. The completely standardized indirect effect is weak, b* = .09, SE = 0.043, 95% CI [0.01, 0.17]. H1d therefore finds support in this experiment.24
Conclusion
This research evaluated the effectiveness of candidate ads which are targeted towards the receivers’ personality trait profiles and their underlying processes. The results of this experiment yield four main insights which are presented in this section.
First, congruence between the receiver’s personality and the candidate message can improve their feeling towards the candidate significantly. At the same time, the propensity to vote is not affected by congruence. It appears to be more difficult to change people’s voting preferences – which normally depend on many factors such as political attitude, political context and others – than an interpersonal opinion on a candidate. In addition, this study suggests that the feeling towards the candidate mediates the (insignificant) effect of congruence on the propensity to vote for the candidate. This supports the interpretation that the underlying processes of
psychometric targeting mainly work on the emotional level: hence, microtargeting has great potential to sway candidate evaluations on the personal level, but not as much potential to change
24 All hypothesis tests have also been conducted with those cases excluded who needed less than five or more than 30 minutes to complete the experiment (n = 180). These tests reveal no substantial differences for any hypotheses but one. In the case of H1d the exclusion of cases yields an insignificant mediation effect, b = 0.18, SE = .102, 95% CI [-0.02, 0.39].
political convictions – at least not after a single exposure.25 Thereby, this study reveals an important difference between susceptibility of customers and voters: microtargeting has an immediate effect on buying behavior (e.g., Matz et al., 2017), but no significant effect on voting decisions.
Second, no positive effects of congruence on cognition and emotion could be found. The emotional and cognitive response states did not function as mediators as predicted by the DSMM. The accuracy of at least the second proposition of the DSMM26 should therefore be put into question.27
Third, this research contributes to the mixing console analogy introduced by Valkenburg and Peter (2013). The analogy concerns the combination of the three response states: cognition, emotion, and excitement. Based on the present results, the effect is the highest when cognitive response state (H2b) and emotional response state (H3b) are high as suggested by Valkenburg and Peter (2013).
25 Research shows that the effects of familiarity and similarity on person evaluation increase after repeated exposure (Moreland & Zajonc, 1982).
26 “Media effects are indirect; three media response states mediate the relationship between media use and media effects” (Valkenburg & Peter, 2013, p. 227).
27 On the other hand, an operationalization of emotional response as feeling towards the candidate would indeed lead to a confirmation of the role of the emotional response state as a mediator.
Fourth, trust was not significantly influenced by congruence. More background information about the candidate or a repeated exposure to contents concerning the candidate could increase the weight of this factor. Despite this insignificant effect of congruence on trust, the predicted effect of trust on the feeling towards the candidate was confirmed.
In conclusion, this study gives a mixed answer to the first research question, which asked about the effect of tailored advertising on the feeling towards the candidate and the propensity to vote for the candidate. Candidate messages are more effective in improving the voter’s feeling towards the candidate when they are congruent with the voter’s personality profile, but they do not result in a higher propensity to vote for the advertised candidate. The second research question asked about the mediation effect of cognition, emotional response and trust in the candidate. This study could not find any significant mediation effects of these variables. Therefore, the question about underlying processes behind the effects of congruent messages remains unanswered.
Discussion
To the best of my knowledge, this research is the first to assess differential effects of personality targeting in the field of political communication using manipulations for all Big Five personality dimensions. By testing for the underlying processes of microtargeting effects this research can be a starting point to develop updated theoretical models of microtargeting effects on the individual level, taking recent technological advancements into account.
One of the main weaknesses of this study is the small size of the subsamples. A higher number of participants per condition or a focus on two or three conditions instead of five would have allowed for more differentiated findings for different personality dimensions. With a larger sample size one could have examined how personality profiles differ in terms of susceptibility to
microtargeting effects. Future research should also evaluate how susceptibility differs by age, level of political interest and other variables. The results could shed light on the unanswered question of underlying processes behind psychometric microtargeting.
Another limitation of this research concerns its measurements. Both emotion and cognition were measured by using simple written single-item assessment techniques. Mechanical and electrophysiological measurements might be more reliable and could better detect actual unconscious effects (see Potter & Bolls, 2012). Further, since the effect of high cognitive
involvement compared to low involvement is expected to increase after a certain period of time, a second assessment of cognition could reveal a larger effect (Valkenburg, Peter, & Walther, 2016). Due to feasibility considerations these techniques were not employed in this experiment.
In addition, this research did not directly test for the effect of congruence on how fluently the message was processed. Fluency is one of the main theoretical concepts of this study. Therefore, a direct assessment of this variable would have increased the explanatory power of this research.
Notwithstanding these limitations, this research was a successful attempt to show the persuasive potential of psychometric microtargeting. Limited textual adaptions of the stimuli were sufficient to increase the messages’ effectiveness, while candidate image and the general design of the sponsored post did not differ. It can be expected that a tailoring of design elements sways voters’ attitudes even more towards the desired direction (for a successful adaption of design elements see Matz et al., 2017).
Moreover, although the uncontroversial stimuli design might have led to the high number of neutral emotional responses (45.23%, N = 199), this approach has the advantage that pre-existing political convictions, presumably, did not affect participants’ opinions on the candidate
substantially. Especially in a highly polarized political environment as the U.S., this factor should not be underestimated (Twenge, Honeycutt, Prislin, & Sherman, 2016).
On a final note, it should be mentioned that the initial claim by Cambridge Analytica – that psychometric microtargeting helped Trump become president – was debunked quickly (Taggart, 2017). The concept of psychometric microtargeting is, however, realizable as research shows. In addition, this study reveals the potential of this tool to influence voters’ feelings. In the long term, campaigners might be able to transfer this effect to actual behavioral change.
This research does not provide a definite answer to the question of the manipulative potential of microtargeting. But if future research can confirm that campaigners are able to exploit voters’ psychological weaknesses, e.g., by triggering neurotic people with fear-promoting messages, campaign regulations should be considered. A use of online tracking data for political purposes could, for example, be restricted to prevent psychological manipulation (Rubinstein, 2014).28
28 An actual restriction of these techniques in the U.S. is unlikely due to high standards of freedom of speech (Rubinstein, 2014).
References
Alter, A. L., & Oppenheimer, D. M. (2009). Uniting the tribes of fluency to form a metacognitive nation. Personality and Social Psychology Review, 13(3), 219-235.
doi:10.1177/1088868309341564
ANES 2016 Time Series Study (2017). Inter-university Consortium for Political and Social Research (ICPSR). doi:10.3886/ICPSR36824.v2
Bailenson, J. N., Iyengar, S., Yee, N., & Collins, N. A. (2008). Facial similarity between voters and candidates causes influence. Public Opinion Quarterly, 72(5), 935-961.
doi:10.1093/poq/nfn064
Bakker, B. N. (2017). Personality traits, income, and economic ideology. Political Psychology, 38(6), 1025-1041. doi:10.1111/pops.12349
Bakker, B. N., Hopmann, D. N., & Persson, M. (2015). Personality traits and party identification over time. European Journal of Political Research, 54(2), 197-215. doi:10.1111/1475-6765.12070
Bakker, B. N., Klemmensen, R., Nørgaard, A. S., & Schumacher, G. (2016). Stay loyal or exit the party? How openness to experience and extroversion explain vote switching. Political Psychology, 37(3), 419-429. doi:10.1111/pops.12257
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(2), 302-320. doi:10.1111/1475-6765.12121 Bandura, A. (2001). Social cognitive theory: An agentic perspective. Annual Review of
Barbu, O. (2014). Advertising, microtargeting and social media. Procedia - Social and Behavioral Sciences, 163, 44-49. doi:10.1016/j.sbspro.2014.12.284
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51(6), 1173-1182.
doi:10.1037/0022-3514.51.6.1173
Barrick, M. R., & Mount, M. K. (1991). The big five personality dimensions and job
performance: A meta-analysis. Personnel Psychology, 44(1), 1-26. doi:10.1111/j.1744-6570.1991.tb00688.x
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. doi:10.1093/idpl/ipw021
Bodó, B., Erickson, K., Katzenbach, C., Musiani, F., & van Hoboken, J. (Eds.) (2017). [Special issue]. Internet Policy Review, 6(4).
Bodó, B., Helberger, N., & de Vreese, C. H. (2017). Political micro-targeting: A Manchurian candidate or just a dark horse? Internet Policy Review, 6(4). doi:10.14763/2017.4.776 Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: The Self-Assessment Manikin and the
semantic differential. Journal of Behavior Therapy and Experimental Psychiatry, 25(1), 49-59. doi:10.1016/0005-7916(94)90063-9
Byrne, D. (1961). Interpersonal attraction and attitude similarity. The Journal of Abnormal and Social Psychology, 62(3), 713-715. doi:10.1037/h0044721
Cacioppo, J. T., & Petty, R. E. (1981). Social psychological procedures for cognitive response assessment: The thought listing technique. In T. V. Merluzzi, C. R. Glass & M. Genest (Eds.), Cognitive assessment, (pp. 388-438). New York (NY): Guilford Press.
Cadwalladr, C. (2017, May 7). The great British Brexit robbery: How our democracy was hijacked. The Guardian. Retrieved from
https://www.theguardian.com/technology/2017/may/07/the-great-british-brexit-robbery-hijacked-democracy
Carver, C. S., Sutton, S. K., & Scheier, M. F. (2000). Action, emotion, and personality: Emerging conceptual integration. Personality and Social Psychology Bulletin, 26(6), 741-751. doi:10.1177/0146167200268008
Chester, J., & Montgomery, K. C. (2017). The role of digital marketing in political campaigns. Internet Policy Review, 6(4). doi:10.14763/2017.4.773
Denissen, J. J., & Penke, L. (2008). Motivational individual reaction norms underlying the Five-Factor model of personality: First steps towards a theory-based conceptual framework. Journal of Research in Personality, 42(5), 1285-1302. doi:10.1016/j.jrp.2008.04.002 Dijkstra, A. (2008). The psychology of tailoring-ingredients in computer-tailored persuasion.
Social and Personality Psychology Compass, 2(2), 765-784. doi:10.1111/j.1751-9004.2008.00081.x
Dijkstra, A., & de Vries, H. (1999). The development of computer-generated tailored
interventions. Patient Education and Counseling, 36(2), 193-203. doi:10.1016/S0738-3991(98)00135-9
Dobber, T., Trilling, D., Helberger, N., & de Vreese, C. H. (2017). Two crates of beer and 40 pizzas: the adoption of innovative political behavioural targeting techniques. Internet Policy Review, 6(4). doi:10.14763/2017.4.777
Donnellan, M. B., Oswald, F. L., Baird, B. M., & Lucas, R. E. (2006). The mini-IPIP scales: Tiny-yet-effective measures of the Big Five factors of personality. Psychological Assessment, 18(2), 192-203. doi:10.1037/1040-3590.18.2.192
Enli, G. S., & Skogerbø, E. (2013). Personalized campaigns in party-centred politics: Twitter and Facebook as arenas for political communication. Information, Communication &
Society, 16(5), 757-774. doi:10.1080/1369118X.2013.782330
Erdogan, B. Z. (1999). Celebrity endorsement: A literature review. Journal of Marketing Management, 15(4), 291-314. doi:10.1362/026725799784870379
Eysenck, H. J. (1991). Dimensions of personality: 16, 5 or 3? Criteria for a taxonomic paradigm. Personality and Individual Differences, 12(8), 773-790.
doi:0.1016/0191-8869(91)90144-Z
Farahat, A., & Bailey, M. C. (2012). How effective is targeted advertising? Proceedings of the 21st International Conference on World Wide Web, 111-120.
doi:10.1145/2187836.2187852
Franz, M. M. (2013). Targeting campaign messages. Good for campaigns but bad for America. In T. N. Ridout (Ed.), New directions in media and politics (pp. 103-117). New York (NY) and London: Routledge.
Fulgoni, G. M., Lipsman, A., & Davidsen, C. (2016). The power of political advertising: Lessons for practitioners. Journal of Advertising Research, 56(3), 239-244. doi:10.2501/JAR-2016-034
Garramone, G. M., & Smith, S. J. (1984). Reactions to political advertising: Clarifying sponsor effects. Journalism Quarterly, 61(4), 771-775. doi:10.1177/107769908406100404 Garzia, D. (2011). The personalization of politics in Western democracies: Causes and
consequences on leader–follower relationships. The Leadership Quarterly, 22(4), 697-709. doi:10.1016/j.leaqua.2011.05.010
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. J. Deary, F. De Fruyt & F. Ostendorf (Eds.), Personality psychology in Europe (pp. 7–28). Tilburg: Tilburg University Press.
Gosling, S. D., Rentfrow, P. J., & Swann, W. B. (2003). A very brief measure of the Big-Five personality domains. Journal of Research in Personality, 37(6), 504-528.
doi:10.1016/S0092-6566(03)00046-1
Grassegger, H., & Krogerus, M. (2016, December 3). Ich habe nur gezeigt, dass es die Bombe gibt. Das Magazing. Retrieved from https://www.dasmagazin.ch/2016/12/03/ich-habe-nur-gezeigt-dass-es-die-bombe-gibt/
Grassegger, H., & Krogerus, M. (2017, January 28). The data that turned the world upside down. Motherboard. Retrieved from https://motherboard.vice.com/en_us/article/mg9vvn/how-our-likes-helped-trump-win
Graziano, W. G., & Eisenberg, N. (1997). Agreeableness: A dimension of personality. Handbook of Personality Psychology, 18(2), 185-213. doi:10.1016/B978-012134645-4/50031-7 Green, M. C., Brock, T. C., & Kaufman, G. F. (2004). Understanding media enjoyment: The role
of transportation into narrative worlds. Communication Theory, 14(4), 311-327. doi:10.1111/j.1468-2885.2004.tb00317.x
Greenwald, A. G., & Leavitt, C. (1984). Audience involvement in advertising: Four levels. Journal of Consumer Research, 11(1), 581-592. doi:10.1086/208994
Hall, M. A., Zheng, B., Dugan, E., Camacho, F., Kidd, K. E., Mishra, A., & Balkrishnan, R. (2002). Measuring patients’ trust in their primary care providers. Medical Care Research and Review, 59(3), 293-318. doi:10.1177/1077558702059003004
Hayes, A. F. (2017). Introduction to mediation, moderation, and conditional process analysis. Second edition: A regression-based approach. New York (NY): Guilford Press. Heider F. (1958). The psychology of interpersonal relations. New York (NY): Wiley.
Hersh, E. D., & Schaffner, B. F. (2013). Targeted campaign appeals and the value of ambiguity. The Journal of Politics, 75(2), 520-534. doi:10.1017/S0022381613000182
Hirsh, J. B., & Inzlicht, M. (2008). The devil you know: Neuroticism predicts neural response to uncertainty. Psychological Science, 19(10), 962-967.
doi:10.1111/j.1467-9280.2008.02183.x
Hirsh, J. B., Kang, S. K., & v. Bodenhausen, G. (2012). Personalized persuasion: Tailoring persuasive appeals to recipients’ personality traits. Psychological Science, 23(6), 578-581. doi:10.1177/0956797611436349
Hetherington, M. J. (1998). The political relevance of political trust. American Political Science Review, 92(4), 791-808. doi:10.2307/2586304
Hong, J. W., & Zinkhan, G. M. (1995). Self-concept and advertising effectiveness: The influence of congruency, conspicuousness, and response mode. Psychology & Marketing, 12(1), 53-77. doi:10.1002/mar.4220120105
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
John, O. P., & Srivastava, S. (1999). The Big Five trait taxonomy: History, measurement, and theoretical perspectives. In L. A. Pervin & O. P. John (Eds.), Handbook of personality: Theory and research, second edition (pp. 102-138). New York (NY) and London: Guildford Press.
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. doi:10.1073/pnas.1218772110
Kreiss, D. (2016). Prototype politics: Technology-intensive campaigning and the data of democracy. New York (NY): Oxford University Press.
Kreiss, D. (2017). Micro-targeting, the quantified persuasion. Internet Policy Review, 6(4). doi:10.14763/2017.4.774
Kreiss, D., & Mcgregor, S. C. (2017). Technology firms shape political communication: The work of Microsoft, Facebook, Twitter, and Google with campaigns during the 2016 US Presidential cycle. Political Communication, 00, 1-23.
doi:10.1080/10584609.2017.1364814
Kreiss, D., & Welch, C. (2015). Strategic communication in a networked age. In V. A. Farrar-Myers & J. S. Vaughn (Eds.), Controlling the message: New media in American political campaigns (pp. 13-31). New York (NY): New York University Press. Kreuter, M. W., Bull, F. C., Clark, E. M., & Oswald, D. L. (1999). Understanding how people
process health information: A comparison of tailored and nontailored weight-loss materials. Health Psychology, 18(5), 487-494. doi:10.1037/0278-6133.18.5.487
Krugman, H. E. (1965). The impact of television advertising: Learning without involvement. Public Opinion Quarterly, 29(3), 349-356. doi:10.1086/267335
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
Kruschinski, S., Haller, A. (2017). Restrictions on data-driven political micro-targeting in Germany. Internet Policy Review, 6(4). doi:10.14763/2017.4.777
Lang, A. (2017). Limited capacity model of motivated mediated message processing (LC4MP). In P. Rössler, C. A. Hoffner & L. van Zoonen (Eds.), The international encyclopedia of media effects (pp. 1-9). Hoboken (NJ): Wiley.
Levi, M., & Stoker, L. (2000). Political trust and trustworthiness. Annual Review of Political Science, 3(1), 475-507. doi:10.1146/annurev.polisci.3.1.475
Lodge, M., Steenbergen, M. R., & Brau, S. (1995). The responsive voter: Campaign information and the dynamics of candidate evaluation. American Political Science Review, 89(2), 309-326. doi:10.2307/2082427
Lucas, R. E., Diener, E., Grob, A., Suh, E. M., & Shao, L. (2000). Cross-cultural evidence for the fundamental features of extraversion. Journal of Personality and Social Psychology, 79(3), 452-468. doi:10.1037/0022-3514.79.3.452
Machleit, K. A., & Wilson, R. D. (1988). Emotional feelings and attitude toward the
advertisement: The roles of brand familarity and repetition. Journal of Advertising, 17(3), 27-35. doi:10.1080/00913367.1988.10673121
Magin, M., Podschuweit, N., Haßler, J., & Russmann, U. (2017). Campaigning in the fourth age of political communication. A multi-method study on the use of Facebook by German
and Austrian parties in the 2013 national election campaigns. Information,
Communication & Society, 20(11), 1698-1719. doi:10.1080/1369118X.2016.1254269 Marzi, T., Righi, S., Ottonello, S., Cincotta, M., & Viggiano, M. P. (2012). Trust at first sight:
Evidence from ERPs. Social Cognitive and Affective Neuroscience, 9(1), 63-72. doi:10.1093/scan/nss102
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.
McCrae, R. R., & Costa, P. T., Jr. (1997). Conceptions and correlates of openness to experience. In R. Hogan, J. A. Johnson & S. R. Briggs (Eds.), Handbook of personality psychology (pp. 825-847). Cambridge (MA): Academic Press Inc.
McCroskey, J. C., & Mehrley, R. S. (1969). The effects of disorganization and nonfluency on attitude change and source credibility. Communications Monographs, 36(1), 13-21. doi:10.1080/03637756909375604
Meagher, M. W., Arnau, R. C., & Rhudy, J. L. (2001). Pain and emotion: Effects of affective picture modulation. Psychosomatic Medicine, 63(1), 79-90. doi:10.1097/00006842-200101000-00010
Miller, A. H., Wattenberg, M. P., & Malanchuk, O. (1986). Schematic assessments of presidential candidates. American Political Science Review, 80(2), 521-540. doi:10.2307/1958272 Moreland, R. L., & Zajonc, R. B. (1982). Exposure effects in person perception: Familiarity,
similarity, and attraction. Journal of Experimental Social Psychology, 18(5), 395-415. doi:10.1016/0022-1031(82)90062-2
Morris, J. D., & Boone, M. A. (1998). The effects of music on emotional response, brand attitude, and purchase intent in an emotional advertising condition. Advances in Consumer Research, 25, 518-526. Retrieved from
http://www.acrwebsite.org/volumes/8207/volumes/v25/NA-25
Morris, J. D., & McMullen, J. S. (1994). Measuring multiple emotional responses to a single television commercial. Advances in Consumer Research, 21, 175-180. Retrieved from http://www.acrwebsite.org/search/view-conference-proceedings.aspx?Id=7581
Morris, J. D., Woo, C., Geason, J. A., & Kim, J. (2002). The power of affect: Predicting intention. Journal of Advertising Research, 42(3), 7-17. doi:10.2501/JAR-42-3-7-17 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), 673-693. doi:10.1037/0033-2909.133.4.673
Olivola, C. Y., & Todorov, A. (2010). Elected in 100 milliseconds: Appearance-based trait inferences and voting. Journal of Nonverbal Behavior, 34(2), 83-110.
doi:10.1007/s10919-009-0082-1
Parker, G. R. (1989). The role of constituent trust in congressional elections. Public Opinion Quarterly, 53(2), 175-196. doi:10.1086/269502
Patton, G. W., & Kaericher, C. E. (1980). Effect of characteristics of the candidate on voter's preference. Psychological Reports, 47(1), 171-180. doi:10.2466/pr0.1980.47.1.171 Potter, R. F., & Bolls, P. (2012). Psychophysiological measurement and meaning: Cognitive and
emotional processing of media. New York (NY): Routledge.
Schmitt, H., Popa, S.A. & Devinger, F. (2015). European Parliament Election Study 2014, Voter Study, First Post-Election Survey. doi:10.4232/1.5161.
