Micro targeting and data mining in online political communications: The
effects on citizen’s trust in political parties.
Eva Lodewijks - 10215611 6th of June 2016
Afstudeerproject Politieke Communicatie en Journalistiek Dr. Sanne Kruikemeier
Bachelor of Communication Science University of Amsterdam
1
Abstract
The purpose of this study is to identify the effects micro targeting in online political communications has on trust in political parties. In order to do this a 2x2 between subjects experiment has been constructed with trust in political parties as the dependent variable and “successfulness of micro targeting” and “awareness of micro targeting” as the independent variables. 206 Dutch students between the ages of 18 and 25 participated in this study. The results show that a successfully micro targeted message results in a lower level of trust in the political party. Awareness of being micro targeted leads to a higher level of trust in the
political party than not being aware. The results of this study can be used by Dutch politicians when implementing micro targeting in their campaign strategies.
Introduction
Recent new media trends have brought along with them a lot of new online marketing possibilities. These new possibilities are not only used in the commercial world but also by political parties and actors.
A political marketing tactic that has gained many new possibilities due to these new opportunities is micro targeting. Online micro targeting is defined as “advanced psycho-geographic segmenting which is based on an algorithm determining a series of demographic and attitudinal traits to distinguish individuals for each targeted segment” (Barbu, 2014). In the United States micro targeting is frequently used during election campaigns. Political parties receive help from companies that have huge databases containing information about potential voters (Barbu, 2014). Even presidential candidates’ apps use the data from users phones to make their promotional material more personal and more fitting for each user (“Ted Cruz volgt app gebruikers overal.” source: www.volkskrant.nl).
2 Personal information is often collected via social media, not just demographic
characteristics but also interests, relationship status, education, work information, contact information, Likes, etc. (Barbu, 2014). Politicians have the opportunity to get to know their audience better than the audience knows themselves and they seem to know exactly what their audience wants to hear. Potential voters can be approached with relevant messages that
communicate information about a candidate’s positions on the issues in which different groups of voters are most invested (Barocas, 2012).
The information stated above makes micro targeting sound like a brilliant idea. However in practice this is not always the case. Online privacy is a tending topic and is very important to most people. Micro targeting however, is a concept that does not seem too bothered with people’s privacy. It makes use of personal internet details and often people are not even aware these details are accessible to the companies whom collect them. After all, everyone knows that no one ever reads the terms and conditions! Another problem with online micro targeting is that politicians know everything about voters but media and the pubic do not know what politicians really belief because they communicate different messages to different target groups. As a result micro targeting could lead to a less informed electorate and a less representative government (Barocas, 2012).
A less informed electorate and less representative government should be of great concern for democracy. Previous research has pointed out that citizens are becoming more and more cynical when it comes to politics(de Vreese, 2005) and that current trends lead to distrust in politicians (Guggenheim, Kwak & Campbell, 2011). Micro targeting is a trend that could potentially also lead to cynicism and distrust because it uses personal details, of which voters are often not aware of that they are available to politicians.
3 Regardless of the concerns mentioned above, most previous research has focused on the positive effects of micro targeting (Bimber, 2014) and on the negative effects it has on the representativeness of policies (Barocas, 2012). However, there is a lack of research that focuses on the effects micro targeting and datamining have on citizen’s attitudes and beliefs about politics. For this reason the goal of this study is to see whether the issues mentioned above have an effect on citizens’ trust in politics. This leads to the following research question: “What is the effect of online micro targeting in political communications, specifically micro targeted messages vs. generic messages and awareness of being micro targeted, on trust in political parties?”
As mentioned above, the possibilities of the internet are still not fully discovered and new marketing techniques are a result. As previous research suggests, it is important to study this concept: “These rapid changes in technology and new media require scholars to update the design and direction of future research” (Towner & Dulio, 2012).
In addition to this, most research about political communications and micro targeting that has been mentioned above has been done in the United States. The American political culture and campaigns are more spectacular and outshine European ones, so people might think that strategies such as micro targeting and data mining are not very common in Europe. For this reason it is very interesting to see what the effects of online micro targeting are in a European political system.
Another important factor to consider in this research area is that the new media effects might differ per age group. Today’s young adults grew up with the internet so they might be less skeptical about concepts like micro targeting.
Previous research suggests that the new media effects might not only differ in age group but also vary in different industries (Jai, Burns and King, 2013). The use of micro
4 targeting in marketing and the effects is has on consumer trust has already been studied, but these results could be different for political communications.
This study will first look back on previous research done on the subject. This will lead to three hypothesis. These hypothesis will be tested in an experiment which will be described in the method. After this, the results will be discussed and that will lead to a discussion and conclusion of this study.
Theory
The internet offers a new public sphere. It gives citizens the opportunity to engage in politics in new ways and interact with politicians more. (Dahlgren, 2005; Towner & Dulio, 2012). Previous research suggests that potential impacts of online political communications could be positive and make especially young adults feel more engaged in politics. However, others say that we are or will be overwhelmed by the more negative impacts such as fragmentation, manipulation, consumerism, dominance of entertainment over public affairs and the paralyzing impact of information overload (Delli Carpini, 2010).
One of the effects of online political communications that can be positive but also negative is that the internet offers such a massive amount of information that you can almost always find what you are looking for. This also means that if you have an opposing political opinion, you will be able to find likeminded people on the web. People can sort into cocoons of homogeneous perspectives and this will polarize their attitudes even more. The
homogeneous media environment and polarized attitudes could lead to political extremism, a formation of a broader counter political culture and organized political anti-consumption (Dahlgren, 2005; Warner, 2010). Micro targeting could reinforce this as it targets certain people with the same messages. If people with extremists point of views are exposed to the
5 same messages and respond to these messages the same way, this could strengthen their extremist views.
The internet and trust in politicians also have a link with personalization. Personalization of political messages seems to be more common than ever before. The internet gives citizens the opportunity to watch these talk shows in YouTube clips or they can become “trending” on Facebook or Twitter. Some say these online trends increase political efficacy and political involvement. Adding the entertainment factor to politics makes it more interesting for
some(Jebril, de Vreese, van Dalen & Albæk (2013) and even increases trust in politicians for people with low political knowledge (Boukes & Boomgaarden, 2014). They could also lead to citizens believing that political actors do this with the intention of creating citizen
empowerment or to boost democracy. Opposing this point of view Bimber (2014) argues: “It is important to remember that political elites do not employ new communication channels with the aim of citizen empowerment. The goal of the candidate investment in media tools is to win elections”. The issue with micro targeting in a situation like this is that when people with low political knowledge are all targeted with mainly entertainment messages rather than political messages this could lead to higher levels of trust but also to a less informed
electorate.
Having discussed these possible negative aspects of the increasing role of the internet in political communications, the focus will now be on how this affects micro targeting and vice versa.
Even though research about the effects of online micro targeting in politics is lacking there are quite a few studies about online micro targeting in marketing and sales. These studies show
6 that personalized messages can be a useful marketing strategy to attract new users, enable sophisticated segmentation and are better predictors of purchasing patterns. It is worthwhile for firms to invest in data mining to analyze the transaction patterns among like-minded people (Ho & Ho 2008; Sivadas & Grewal, 1998; Ziliani & Bellini, 2007).
Ho and Ho (2008) specifically look at the effects of micro targeting in marketing on young adults. Their study suggest that frequent and involved interactions among web users and sites leads to users being more reluctant to switch to a new website. This could suggest that politicians who communicate with potential voters frequently and in a personal and involved manner, would need to worry less about these people voting for someone else.
Young adults in today’s society have a decreasing interest in politics (Delli Carpini, 2000). This is a great concern for democracy as today’s youth is the future. Younger adults are generally more computer literate than their counterparts ten years ago. So would this mean that because young adults are used to the internet and its micro targeting ways that the use of micro targeting in politics would not have an effect on the trust they have in political parties? Previous research suggest that young adults are concerned about their information privacy even when they are dealing with trusted online retailers. Young consumers care about whether their behavioral information will be broadly disseminated. The level of dissemination of information significantly increases young consumers’ perceived risk and perceived unfairness (Jai, Burns & King, 2013). This could mean that using online political micro targeting could lead to politics being perceived as more risky and unfair. This could potentially have a negative impact on trust in political parties.
A research article about trust and personalization of electronic medical information shows that trust is positively associated with behavioral intention (Guo, Sun, Yan, & Wang, 2012). Trust mediates the relationship between perceived personalization and privacy concern and behavioral intention. This could in turn mean that citizens who do not trust politicians
7 will be very concerned about privacy and think that messages are personalized to a high extent. This could decrease behavioral intention, in case of politics, vote intention.
As mentioned previously, research on online micro targeting in political communications is lacking. However it is not non-existent. In 2004 the Bush campaign specifically targeted Hispanic women with children in New Mexico, in 2000 New Mexico voted Democrat and in 2004 they voted Republican. This and other examples convinced political marketers that micro targeting was the new way to win elections (Bennett, 2014). Another campaign to actively use online micro targeting was the Obama campaign in 2008. This campaign successfully invested in acquiring personal data of many kinds about citizens resulting in crafting messages and making tactical decisions about how to allocate resources (Bimber, 2014). Previous research has pointed out that micro targeting in political communication can be very effective in predicting voting behavior (Bimber, 2014; Murray & Scime, 2010).
Research has also shown that online micro targeting in political campaigns may increase the effectiveness of political campaigns. It can also lead to more effective messages which in turn lead to better informed voters. Micro targeting can also mobilize voters who otherwise might have abstained from voting (Murray & Scime, 2010). This could mean that the decreasing political engagement for young adults could be solved by using micro
targeting. By segmenting the electorate in useful and meaningful ways political messages can have bigger effects on their receivers.
Although it might seem like this from the above mentioned articles, micro targeting is not all fun and games. Micro targeted campaign messages increase the level of unaccountable and misleading campaigning. Not everyone receives the same message so reporters and fact checkers might never know that some messages were distributed. Political parties spread
8 different messages to different target groups and it becomes more difficult to hold politicians accountable for their statements and to police deception. So as a result of micro targeting, the electorate becomes less representative (Barocas, 2012; Gottfried, Hardy, Winneg & Jamieson, 2013).
It is expected that these negative effects of micro targeting such as a less
representative electorate, privacy concerns and more misleading campaigning would lead to micro targeting having a negative image. This brings us to the first hypothesis. When you see an ad or post that perfectly suits your interests and opinions you could suspect this ad or post was micro targeted. In turn this could lead to a feeling of distrust towards the sender of this message due to micro targeting’s negative image. This study will test whether micro targeting has a negative effect on trust in political parties.
H1: Seeing a micro targeted online political ad vs. a generic online political ad will have a negative effect on citizens’ trust in political parties. (main effect)
Not only seeing an ad or post that fits you perfectly, but also knowing you are being micro targeted could also have a serious impact on trust. Micro targeting is a very subtle process and often people may not actually be aware that they are being micro targeted (Barocas, 2012; Jai, Burns & King, 2013). Previous research shows that 77% of Americans agree (including 35% who agree strongly) that ‘If I knew a website I visit was sharing information about me with political advertisers, I would not return to the site.’ (Barocas, 2012). Another study suggests that long term effects of micro targeting tactics in political communications will have a negative effect on trust in receiving an honest agenda of issues from candidates (Turow, Carpini, Draper & Howard-Williams, 2012). This shows that people do not appreciate their personal information being shared with political advertisers and this could also lead to distrust
9 towards a political party that includes micro targeting in their tactics. For this reason it is expected that when people are made aware that a certain political party makes use of micro targeted messages in their online communications they would feel less trusting towards this political party. This leads to the following hypothesis:
H2: Knowing you are being micro targeted vs. being unaware that you are being micro targeted has a negative effect on citizens’ trust in political parties. (main effect)
A combination of the two hypothesis mentioned above could also have an effect on trust. It is expected that seeing a generic ad will lead to a higher level of trust than seeing a micro targeted ad. It is also expected that when people are aware they are being micro targeted they will feel less trusting towards a political party than when they are not aware. It is interesting to see a combination of these two hypothesis as an interaction effect and to find out whether being aware that you are being micro targeting but then seeing a political ad that does not seems specifically for you would have a negative effect on trust. And if trust levels would be higher when you are aware you are being micro targeted and also see a micro targeted ad. It is expected that this effect does exist because it would confuse people to see a generic political ad after they had just been informed the ad is going to be specifically targeted towards them. When you are already skeptical about a concept like micro targeting (Turow, Carpini, Draper & Howard-Williams, 2012) and then also being confused by it would lead to a lower level of trust. This brings us to the third and last hypothesis:
H3: The effect that knowing you are being micro targeted has on trust citizens have in
political parties is moderated by the type of message (generic vs. micro targeted). (interaction effect)
10 This moderation effect means that knowing you are being micro targeted will have a negative effect on trust in the political party, however this effect is made stronger if participants are shown a generic political ad and made weaker when participants are shown a micro targeted political ad. An example of the conceptual model of this expected moderation effect can be found in the appendix.
The results of the last hypothesis will be interesting for politicians as they will show them that if they do decide to use micro targeting they need to actually make sure the message is
targeted properly so that the receiver of the message does not to get confused, as this is expected to lead to a decrease in their audience’s trust level.
Method
To test the above formulated hypothesis an experiment has been conducted. The goal of this research is to study the effects micro targeting in online political messages has on the trust citizens have in political parties. Micro targeting will lead to more or less trust as trust can’t lead to more or less micro targeting, therefore this is a causal relationship.
In any real life situation a situation where people would be exposed to two of the same messages where one was micro targeted and the other one was generic would never occur, that is why this situation needs to be manipulated in an experiment. In order to keep control over how many people were directly made aware of being micro targeted, an experiment is also a good solution as you can manipulate the experimental groups and give some groups different information than others. Therefore, in order to study this causal relationship an experiment is the best fitting research method.
11
Participants
Young adults in the United States under the age of 30 are less interested in politics, less knowledgeable about substance or processes of politics, less likely to register or vote and less likely to participate in politics beyond voting. These young adults also embrace new
technologies and are significantly more likely to have access to the internet than other age groups. So new technologies for online political communications might offer new
opportunities for increasing the civic engagement of young adults (Delli Carpini, 2000). For this reason the target group of this study were young adults.
There were 206 people that participated in this study. The participants are students aged between 18 and 25 years old (Mage=20.88, SDage=1.72). Of the 206 participants 52.9% is
male and 47.1% is female.
Design
To test the hypothesis a 2x2 between-subjects design experiment has been performed with one measuring moment. A between-subjects design is the best option for this study because it examines the differences between different experimental groups. If a within-subjects design would have been used, where participants are exposed to all different experimental
conditions, this would have ruined the manipulation.
Within the design two factors were manipulated on two levels. The stimulus material consists of four different Instagram posts. The first factor is micro targeting, this factor is manipulated by showing one group a generic post and the other group a micro targeted post specifically for students. The second factor is awareness. This factor is manipulated by showing one group an Instagram post with a “sponsored” label and an explanation of what this label means, so they know they are being micro targeted when they see the sponsored
12 label. The other group was not shown a label or an explanation. After collecting the data, the four conditions were compared to each other.
Conditions:
Generic post & sponsored label and explanation Specific post & sponsored label and explanation Generic post & no sponsored label and no explanation Specific post & no sponsored label and no explanation
Procedure
The participants were recruited via a convenience sample, participants were the researchers’ friends or family. To avoid any issues that could occur when there are more missing values than expected, the aim was to have 50 participants in each condition (N=200).
The participants were asked to fill out an online questionnaire. They first received a few questions about their demographic characteristics, political interest and their opinion on a few Dutch political parties.
After this the participants were asked about their general trust in politics, it is
important to see whether participants find the political system and the politicians in it equally trustworthy in all four experimental conditions.
Following these questions the participants were shown one of the four manipulations. After the manipulation participants were asked questions about their trust in D66 and lastly the manipulation check questions. All questions can be found in the appendix.
The online questionnaire was distributed via Qualtrics. Participants were forced to respond to each question to avoid missing values and random assignment to one of four manipulations was automatically done by Qualtrics.
13
Manipulation material
As mentioned above the stimulus material consists of four different Instagram posts. The posts are all posted by a Dutch political party, D66. Instagram has been chosen as the social media platform for this study because it is one of the biggest platforms and it has a lot less visual noise compared to other platforms. Respondents will only see the D66 post and nothing else besides the Instagram logo on top of the screen. This way the results will not be
influenced by any other messages or other external factors.
The message is about a neutral political subject: recycling. This subject has been trending in Dutch politics over the last few months. Since January 2016 shops in the Netherlands are required by law to charge consumers for plastic bags to decrease plastic consumption (Tweede kamer, 2015-2016). The D66 post will promote a re-usable canvas shopping bag as a response to this subject. The images in the different posts are the same but they will have different captures and hashtags.
The lettering on the bag in the condition with successful micro targeting states: “Don’t you want it as well?! #Studentrecycle”, with the following caption: “12.000 students already have it and recycle through the roof! You too?! #StudentRecycle #D66nowforward” .
The lettering on the bag in the generic condition states: “Don’t you want it as well?! #Recycle”. With the caption: “12.000 people in the Netherlands already have it and recycle through the roof” #Recycle #D66nowforward”. The posts are formulated in a way where it is either specifically addresses students, or formulated in a generic way addressing all Dutch people. Both bags have got a D66 logo on them. Examples of the manipulations can be found in the appendix.
The messages in the Instagram post are normative, they emphasize the fact that everybody does it and this way they increase social pressure. Research shows that normative messages are more effective (Cialdini et al., 2006) because people are more sensitive for
14 social pressure from normative messages. For this reason it is expected that the messages in the posts will be effective in activating the participants (Reno et al., 1993). Because all four conditions make use of a normative message this can not become a confounding variable.
Variables
The dependent variable trust was tested using different questions. These questions consist of statements and 7-point likert scales. The questions eventually constructed scales that represent the dependent variable trust. The questions can be found in the appendix.
First the variable for general trust was constructed, the three relevant items for this variable appeared to be consistent and reliable (Eigenvalue = 2.48; explained variance = 82.76%; α = 0.87; M = 3.79; SD = 1.09).
After, the variable for trust after manipulation was constructed, the three relevant items for this variable were also consistent and reliable (Eigenvalue = 2.38; explained variance = 79.3%; α = 0.87; M = 3.41; SD = 1.27).
Analyses
Checking for confounding variables
To check for possible confounding variables randomization checks have been used to test the differences between the different experimental groups for the independent variables sex, age, education level, political orientation, likeliness to vote for D66 and general trust in politics. These background variables have been kept constant to avoid any of these to become confounding variables.
Main and interaction effects
15 1. The effect a generic vs. a micro targeted post has on trust
2. The effect of knowing you are being micro targeted (awareness) vs. not knowing you are being micro targeted (not aware) has on trust
These main effects were tested using a One-way ANOVA’s.
The existence of an interaction effect was tested by preforming a regression analysis.
Manipulation checks
In order to test whether the manipulation had been successful a manipulation check has been performed. The manipulation check questions were asked at the end of the questionnaire to make sure they did not have an effect on the questions measuring the effects micro targeted messages have on trust.
A One-way ANOVA shows that the group who was successfully micro targeted (with a post specifically for students) and received an explanation, was under the impression that the post was specifically made for them (M=3.19, SD=2.00) less than the other three groups (M=3.37, SD=1.94).This means that the participants were not aware they were being targeted while some participants were specifically targeted as students. This result was not significant, F(1, 204) = 2.35, p = .127.
The group that was shown the ad for students and were made aware they were being micro targeted should have scored highest on the question whether they thought the post was
personalized. The results of the One-way ANOVA show that this is the case, this group scores higher (M = 3.28, SD = 1.97) than the other three groups (M=3.08, SD = 1.86). However this result is also not significant, F(1, 204) = 0.46, p= .510.
16 The participants were also asked if they had seen the “Sponsored” label. In order for the manipulation to be successful the groups that were made aware what the sponsored label means and were shown this label should have all answered Yes. The results show that 49 participants who were shown the “Sponsored” label answered “yes” to this question. 21 participants said they had not seen this label and 38 participants said they could not remember whether they had seen the label or not.
Of the experimental groups that were not shown the label 44 participants correctly answered “no” when asked if they had seen the label, 32 participants incorrectly said they had seen the label and 22 participants could not remember.
The participants were also asked what they thought the sponsored label means. The groups who received an explanation should have answered correctly in order for the manipulation to be successful. The results of the manipulation check show that of the 108 participants that received an explanation, 64 answered correctly. 28 participants believed a sponsored logo means that the ad is used to advertise to all Instagram users and 16 participants did not know the right answer.
In the groups that did not receive an explanation, out of 98, 57 participants answered correctly. 28 participants believed a sponsored logo means that the ad is used to advertise to all Instagram users and 12 participants did not know the right answer.
Validity
External validity
The external validity was increased by using a Dutch political party, D66, that actually uses sponsored messages on Instagram in real life. D66 is a political party that is neither left nor
17 right winged and is quite neutral in that point of view. The manipulation material in this study is an exact replica of what a real life post would look like including the “sponsored” label, this also increases external validity because in real life micro targeted Instagram messages always have the “sponsored” label. Instagram users automatically receive these sponsored micro targeted posts when they agree to Instagram’s general terms and conditions. The research situations is very similar to reality because of the way Instagram is used and the lay-out of the post. These factors increase the external validity of this study.
Internal validity
Internal validity has been increased by randomly assigning participants to groups. This means that every participant has the same chances of ending up in each group. This means that the groups will most likely be equal. However in case of one group having a higher level of certain independent variables (such as gender, education level, age etc.) these variables were checked and are kept constant using correlation matrixes. This has been done to avoid confounding variables that could have a different explanation for the causal relation(s) found in this study.
One of the reasons Instagram has been used in this study is because it barely has any visual noise that could distract participants from the manipulation material. For this reason no other factors could become confounding variables and have an effect on the causal
relationship micro targeting potentially has on trust. This increases the internal validity. By preforming an experiment with a between-subjects design the internal validity has also been increased. Because every participant has been exposed to only one condition, order-effects can not become confounding variables.
18
Results Randomization
To check for any differences between the different experimental groups randomization checks have been performed for the following variables: education level, political activity, political orientation, likeliness to vote for D66 and general political trust.
To test for education level, age and gender a chi square test has been performed. This shows that there is no significant difference in education level (χ2(15) = 16.01, p = 0.382), age (χ2 (21) = 23.41, p = 0.323) or gender (χ2(3) = 0.61, p = 0.895) between the experimental conditions.
The randomization check for political orientation, likeliness to vote for D66 and general trust in politics have been tested using One-way ANOVA’s. This shows that there is no significant difference in political orientation (F(3, 202) = 0.33, p = .805), likeliness to vote for D66 (F(3, 202) = 1.56, p = .199) or general trust in politics (F(3, 202) = 0.395, p = .756) between the different experimental groups.
Hypothesis 1
To test the first hypothesis a One-way ANOVA has been performed to test if the student groups differed from the generic groups on trust levels after being exposed to the
manipulation material. This shows that the groups that were successfully micro targeted had lower levels of trust (M = 3.34, SD = 1.20) than the groups that were shown the generic posts (M = 3.48, SD = 1.35). Unfortunately this result is not significant, F(1, 204) = 0.58, p = .446. This means the results do support H1 however this effect is not significant.
19 Figure 1: Trust in D66 after exposure to manipulation material, generic experimental groups vs. student experimental groups.
Figure 1 illustrates that there is a difference between the student experimental groups and the generic experimental groups. The student groups have a lower level of trust in D66 after being exposed to the manipulation material.
Hypothesis 2
Hypothesis 2 tests whether the independent variable “awareness of micro targeting” has a negative effect on dependent variable trust after the manipulation. In two experimental groups the respondents were first shown an explanation of the “Sponsored” label and after this label was shown in the Instagram posts. This way they were made aware of being micro targeted.
To test the second hypothesis a one way ANOVA compared the groups with and without explanation and label. This shows that the groups that were aware they were of being micro targeted had a higher level of trust (M = 3.47, SD = 1.26) than the groups that were not aware (M = 3.34, SD = 1.29). This result is also not significant, F(1, 204) = 0.50, p = .481. This means that H2 is not supported however this result is not significant.
The manipulation check has shown that 49 participants who were shown the “sponsored” label correctly recalled seeing the label. In the experimental groups that were not shown the
3,25 3,3 3,35 3,4 3,45 3,5
Generic groups Student groups
20 label 44 participants correctly answered “no” to the question whether they had seen the label or not.
Because the rest of the participants did not recall seeing the label correctly another one-way ANOVA was performed to test the second hypothesis, but this time only for the participants whom correctly recalled seeing the “sponsored” label. This shows that the people whom correctly recalled seeing the label and thus were aware they were being micro targeted had a higher level of trust (M = 3.74, SD = 1.15) compared to people whom correctly recalled not seeing the label and thus were not aware they were being micro targeted (M = 3.30, SD = 1.24). These results show the opposite of what was expected in H2, however it needs to be mentioned that this result was also not significant, F(2, 203) = 2.21, p = .112.
Figure 2: Trust in D66 after exposure to manipulation material, experimental groups that were made aware of micro targeting vs. groups that were not made aware of micro targeting.
Figure 2 illustrates that the participants that were made aware they were being micro targeted show a higher level of trust in D66 compared to the experimental groups that were not made aware they were being micro targeted.
Hypothesis 3
To test whether an interaction effect exists between successful micro targeting and awareness of micro targeting a linear regression analysis has been performed. Hypothesis 3 predicted that knowing you are being micro targeted will have a negative effect on trust in the political
3 3,1 3,2 3,3 3,4 3,5 3,6 3,7 3,8
Aware of micro targeting Not aware of micro targeting
21 party, however this effect is made stronger if participants are shown a generic political ad and made weaker when participants are shown a micro targeted political ad. This would be a moderating effect as awareness of micro targeting would not explain the relationship between trust in D66 and successfulness of micro targeting but only make it stronger or weaker.
The results show that there is no significant interaction effect, b* = 0,02, t = -0,145, p = .885. H3 is not supported, this means that the variable “awareness of micro targeting” does not make the effect “successfulness of micro targeting” has on trust any stronger.
Figure 3: The interaction effect where successfulness of micro targeting has an effect on trust in D66 and that effect is moderated by awareness of micro targeting.
Figure 3 shows that the two lines do not cross, this means that there is no interaction effect. The low successfulness of micro targeting variable represents the generic experimental groups and the high successfulness of micro targeting groups represent the student experimental groups. When you look at the lines closely you can see that they will eventually cross as they are not completely parallel to each other. This means that there is an interaction but this effect is not strong enough and not significant.
3,2 3,25 3,3 3,35 3,4 3,45 3,5 Low successfulness of micro targeting High successfulness of micro targeting T ru st in D66 Low awareness of micro targeting High awareness of micro targeting
22
Conclusion and discussion
The goal of this study was to paint a clearer picture of the effects micro targeting in online political communications has on trust citizens have in political parties. The results of this study show that a successfully micro targeted message indeed has a negative effect on trust. They show that if people are aware they are being micro targeted this increases their trust in politicians. The differences between the experimental groups were not big enough for these results to be significant. Nevertheless these results have important implications for the debate on micro targeting.
The first hypothesis predicted that successfully micro targeted messages can have a negative effect on trust. The data did show that the micro targeted Instagram posts had a slightly more negative effect on trust than the generic Instagram posts. This suggests that politicians who are looking to gain trust from potential voters would be best off to not use micro targeted social media posts as this could have a negative effect on trust.
The manipulation check in this study shows that the experimental groups with generic Instagram posts were more likely to think that the ad was made specifically for them than the groups that were shown the Instagram ad specifically for students. This means that the participants were not aware that the Instagram post was targeting them specifically. This is something future researchers of this subject should keep in mind and a few possible
explanations for this will now be discussed.
The first explanation for this is the paralyzing impact of information overload that has been discussed earlier (Delli Carpini, 2010). When you scroll through social media feeds you get so many ads that you might become blind to them. The manipulation in the different Instagram posts was very subtle, the only difference was “students” vs. “Nederlanders”.
23 People receive so many ads that they might not even notice these subtle manipulations or whether an ad is targeted towards students or towards Dutch people in general.
It could also be the case that Dutch students identify just as much with being Dutch as they do with being a student. An ad aimed at “Nederlanders” could feel just as targeted as an ad aimed towards students. This study has shown that students don’t feel more targeted when they are approached as students as they are when they are approached as “Nederlanders”. Politicians and future researchers could use pre-tests and focus groups to find out what “buzzwords” students identify with and make them feel like they are addressed personally by an ad or social media post when they want to target their campaign towards this group.
Another possible explanation is that the target group of this study is so used to the internet that they are not bothered by the fact that politicians use their internet details for their own interests. The study of Jai, Burns and King (2013) that was discussed earlier, shows that young adults are worried about their internet privacy. However this study was accomplished in 2013 and it is possible that society has become more accustomed to micro targeting and data mining over the last three years.
The results for the second hypothesis, that predict that being aware you are being micro targeted will have a negative effect on trust, showed the opposite effect of what was expected. The participants that were made aware of the micro targeting were actually more trusting than the participants that were not made aware. There are a few explanations for this that will be discussed now.
First, the ads that did not have a “sponsored” label could be perceived as less trustworthy because these types of messages usually do have a “sponsored” label in real Instagram feeds. Participants could have been more trustworthy in the conditions that were
24 aware of the micro targeting because they perceived this as honest because the micro targeting had been explained.
Second, a manipulation check where participants were asked what they thought the sponsored label meant, shows that all conditions scored relatively equal. Even the participants who did not receive an explanation still had a fair idea of what micro targeting on Instagram involves. In the conditions that did receive an explanation 59.3% answered correctly and in the conditions that did not receive an explanation 58.2% answered correctly. This means that young adults are more aware of micro targeting than initially expected. This could also mean that it bothers them less than expected because they are aware of it and maybe even got used to it.
For politicians this means that if they want to reach a younger audience with their campaigns they could use “sponsored” Instagram posts but it is important they are honest about their intentions. Young adults are very aware what sponsored messages mean in general and they are more trusting towards political parties when the party is honest about a message being micro targeted.
What is noticeable in general about the results is that the participants answered the question how likely it would be they would vote for D66 in the next election with M = 4.35, SD = 1.75 on a scale from 1 to 7. The participants were also asked how likely it would be they would vote for five other Dutch political parties (PvdA, VVD, GroenLinks, PVV and CDA) and the average score for D66 was definitely the highest (see appendix). Most participants said they would most likely vote for D66 and after this they were shown an ad by D66, so they could have assumed the micro targeting was based on what party they would most likely vote for. Previous research shows that because there is so much political information available
25 and opinions that they would prefer to avoid (Bennett & Iyengar, 2008). The results of this study show that the participants favored D66 over the other political parties so the ad perfectly matched their preferences and they did not have to wall themselves for this political message. This could explain why the participants already felt like they were being targeted based on what political party they were most likely to vote for in the next election. Bennett and Iyengar (2008) link the endless amount of political information to a new era of minimal effects in political communication. They state that because people will only consume the information they are interested in political communication messages will not have an effect unless they fit someone’s initial believes and interests. Micro targeting could amplify this effect as targeted messages only show one side of the story and allow citizens to tune out from messages and opinions that do not appeal to them.
Another interesting factor is that as mentioned previously, in the Netherlands the political landscape is a lot different from the United States. Politics is becoming more of a game in the US, a game that has winners and losers and not everyone plays fair. All the entertainment and spectacles around the elections in the United States could lead to people seeing it as less of a serious business and for this reason they start distrusting the players. When citizens then find out politicians are using their personal internet details in order to improve their game plan this could lead to negative effects on trust in political parties and politics in general. The results of this study however, show that general trust in politics before the manipulation is M = 3.79, SD = 1.09 on a 7 point-likert scale. This suggests that Dutch people are less distrusting of politics than initially expected because politics are less of a “game” in the Netherlands than it is in the United States. Previous research shows that people who are more trusting towards people with whom they have weaker social ties (like
politicians), were more likely to perceive online activities as political participation, use social media to interact with others about political issues and consume political information online
26 (Himelboim, Lariscy, Tinkham & Sweetser, 2012). The results also show that the participants have quite a high level of political interest, M = 4.07, SD = 1.52. This suggests that most participants would spend some of their spare time searching for political news and
information and would not be surprised if they would come across a targeted political ad in their own news feeds. This implies that because the participants already had high levels of trust they are very familiar with using social media to communicate about politics and find political information. This could explain why they got so used to seeing political ads in their daily social media and internet use that they did not notice the sponsored label and that they did not feel targeted by the posts. For Dutch politicians this means that they have to be very careful when using the results of American research studies for their own campaigns. This study shows that Dutch students mostly appreciate honestly about micro targeting and are not very bothered by politicians using their internet details to make their campaigns more suited for students as long as they are honest about it.
Another aspect that is worth mentioning is that previous research shows that social media are a great way for politicians to build interpersonal relationships with potential voters and to increase trust among voters (Himelboim et al., 2012; Painter, 2015). Voters are more trusting towards politicians when they are exposed to a candidates social media page
compared to being exposed to their campaign website (Painter, 2015). The fact that the post was placed on a social media forum could explain the relatively high levels of political trust after the manipulation amongst the participants.
In addition to some of the elements mentioned above there are a few limitations to this
research. Most experiments would do a pre-test in order to avoid surprises in the manipulation checks. Due to time constraints a pre-test was not a possibility in this study. Doing pre-tests would have made sure the manipulation was not too subtle and errors and misinterpretations
27 could have been ruled out. The use of a convenience sample was another consequence of the time constraints of this study, and it might bias the results when participants know the researchers personally
Future studies could explore the effect of online micro targeting on trust in politics even more. This study could be used as a starting point and could be worked out further. It would be interesting to see if being honest about a message being micro targeted and making this very clear would increase trust in the sender of the message. It would also be worthwhile to study whether people are getting used to internet trends like micro targeting and if this has an effect on their perceived violation of their privacy. Another interesting aspect of micro targeting would be to study whether micro targeting increases citizens’ one sided political views.
Online political communications is an area where a lot is still unknown and a lot of future research is still needed. In the world of social media and the internet new developments are established every day. New campaign opportunities for politicians are coming up so fast that the research industry is getting more and more behind as it takes a lot longer to research the effects of all these new applications than it does to develop them. It is very important for results of studies to be most accurate and for researchers to be able to keep up with fast developing technologies and new ways to communicate. This is going to be more and more of a problem in the future. For this reason another suggestion for future research is to optimize and speed up research processes in online communications in order for research results to stay relevant.
28
Appendix A. Manipulation materials
The explanation about the “sponsored” label is the same for both conditions.
29 Generic post with no sponsored label or explanation
30
B. Questions general trust
1. Hoeveel persoonlijk vertrouwen heb jij over het algemeen in de Nederlandse politiek?
Weinig vertrouwen 1 2 3 4 5 6 7 Veel vertrouwen
2. Politici houden zich aan de beloftes die ze tijdens verkiezingen maken
Niet mee eens 1 2 3 4 5 6 7 Mee eens
3. Politici zijn over het algemeen te vertrouwen
Niet mee eens 1 2 3 4 5 6 7 Mee eens
4. Politici denken ten eerste aan hun eigen belang
Niet mee eens 1 2 3 4 5 6 7 Mee eens
C. Question trust in D66 after manipulation
1. Hoeveel persoonlijk vertrouwen heb je na het zien van de advertentie in D66?
Weinig vertrouwen 1 2 3 4 5 6 7 Veel vertrouwen
2. Na het zien van de advertentie heb ik er vertrouwen in dat D66 Nederland zou kunnen besturen.
Niet mee eens 1 2 3 4 5 6 7 Mee eens
3. Na het zien van de advertentie denk ik dat D66 haar beloftes zal na komen als ze verkozen worden in een verkiezing.
Niet mee eens 1 2 3 4 5 6 7 Mee eens
D. Questions manipulation check
1. Weet u wat Instagram is? Ja/Nee
2. Gebruikt u zelf Instagram? Ja/Nee
31 3. In hoeverre denk u dat deze advertentie speciaal op jou gericht was?
Volledig niet op mij gericht 1 2 3 4 5 6 7 Volledig op mij gericht
4. Heeft u het bovenstaand logo gezien in de Instagrampost? Ja/Nee/Weet ik niet
5. Wat denk je dat het “gesponsored” label betekent?
a. Dat er geld betaald is door de bron van de boodschap om mij deze advertentie te laten zien.
b. Dat de bron van de boodschap mijn internet gegevens heeft gebruikt om deze advertentie zo goed mogelijk op mijn interesses af te stemmen.
c. Dat de bron van deze boodschap met dezelfde advertentie adverteert aan iedereen op Instagram.
d. Anders namelijk ….. e. Weet ik niet
f. Ik kan het niet verwoorden
E. Conceptual model of the expected moderation effect.
Awareness of micro targeting Trust in political parties
Generic vs. micro targeted ad
+
-
-
32
F. Likeliness of participants voting for a party in the next elections, measured on a 7-point likert scale.
M SD PvdA 2.95 1.73 VVD 2.87 1.91 GroenLinks 3.34 1.95 D66 4.34 1.75 PVV 1.77 1.50 CDA 2.14 1.30
33
References
Barbu, O. (2014). Advertising, Microtargeting and Social Media. Procedia–Social and Behavioral Sciences, 163, 44-49. doi:10.1016/j.sbspro.2014.12.284
Barocas, S. (2012). The price of precision: Voter micro targeting and its potential harms to the democratic process. In Proceedings of the first edition workshop on Politics, elections and data (pp. 31-36). ACM. doi: 10.1145/2389661.2389671
Bennett, W. L., & Iyengar, S. (2008). A new era of minimal effects? The changing
foundations of political communication. Journal of Communication, 58(4), 707-731. doi: 10.1111/j.1460-2466.2008.00410.x
Bennett, C. (2014). Voter Surveillance, Micro-Targeting and democratic politics: Knowing how people vote before they do. Stable URL: http://ssrn.com/abstract=2605183
Bimber, B. (2014). Digital media in the Obama campaigns of 2008 and 2012: Adaptation to the personalized political communication environment. Journal of Information Technology & Politics, 11(2), 130-150. doi: 10.1080/19331681.2014.895691
Boukes, M., & Boomgaarden, H. G. (2016). Politician Seeking Voter: How Interviews on Entertainment Talk Shows Affect Trust in Politicians. International Journal of Communication, 10, 22. doi: 1932–8036/20160005
34 Cialdini, R. B., Demaine, L. J., Sagarin, B. J., Barrett, D. W., Rhoads, K., & Winter, P. L. (2006). Managing social norms for persuasive impact. Social Influence, 1, 3-15. doi: 10.1080/15534510500181459
Dahlgren, P. (2005). The Internet, public spheres, and political communication: Dispersion and deliberation. Political communication, 22(2), 147-162. doi:
10.1080/10584600590933160
Delli Carpini, M. X. (2000). Gen. com: Youth, civic engagement, and the new information environment. Political communication, 17(4), 341-349. doi:
10.1080/10584600050178942
Guggenheim, L., Kwak, N., & Campbell, S. W. (2011). Nontraditional news negativity: The relationship of entertaining political news use to political cynicism and
mistrust. International Journal of Public Opinion Research, 23(3), 287-314. doi: 10.1093/ijpor/edr015
Guo, X., Sun, Y., Yan, Z., & Wang, N. (2012). Privacy-Personalization Paradox in Adoption of Mobile Health Service: The Mediating Role of Trust. PACIS, (27). Stable URL: http://www.pacis-net.org/file/2012/PACIS2012-054.pdf
Gottfried, J. A., Hardy, B. W., Winneg, K. M., & Jamieson, K. H. (2013). Did Fact Checking Matter in the 2012 Presidential Campaign?. American Behavioral Scientist, 57(11), 1558-1567. doi: 10.1177/0002764213489012
35 Himelboim, I., Lariscy, R. W., Tinkham, S. F., & Sweetser, K. D. (2012). Social media and
online political communication: The role of interpersonal informational trust and openness. Journal of Broadcasting & Electronic Media, 56(1), 92-115. doi: 10.1080/08838151.2011.648682
Ho, S. Y., & Ho, K. K. (2008). The Effects of Web Personalization on Influencing User’s Switching Decisions to a New Website. PACIS 2008 Proceedings, 67. Stable URL: http://aisel.aisnet.org/pacis2008/67
Jai, T. M. C., Burns, L. D., & King, N. J. (2013). The effect of behavioral tracking practices on consumers’ shopping evaluations and repurchase intention toward trusted online retailers. Computers in Human Behavior, 29(3), 901-909.
doi:10.1016/j.chb.2012.12.021
Jebril, N., Vreese, C. H., Dalen, A., & Albæk, E. (2013). The Effects of Human Interest and Conflict News Frames on the Dynamics of Political Knowledge Gains: Evidence from a Cross‐national Study. Scandinavian Political Studies, 36(3), 201-226. doi:
10.1111/1467-9477.12003
Van der Kolk, T. (2016, February 15th). Ted Cruz volgt app gebruikers overal [Newspaper article]. Retrieved from http://www.volkskrant.nl/buitenland/blog-usa16-kasich-onder-
36 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. doi: 10.1080/15377857.2010.497732
Painter, D. L. (2015). Online political public relations and trust: Source and interactivity effects in the 2012 US presidential campaign. Public Relations Review, 41(5), 801-808. doi: 10.1016/j.pubrev.2015.06.012
Reno, R. R., Cialdini, R. B., & Kallgren, C. A. (1993). The transsituational influence of social norms. Journal of personality and social psychology, 64(1), 104. doi: http://dx.doi.org/10.1037/0022-3514.64.1.104
Sivadas, E., Grewal, R., & Kellaris, J. (1998). The internet as a micro marketing tool:
targeting consumers through preferences revealed in music newsgroup usage. Journal of Business Research, 41(3), 179-186. doi:10.1016/S0148-2963(97)00060-X
Towner, T. L., & Dulio, D. A. (2012). New media and political marketing in the United States: 2012 and beyond. Journal of Political Marketing, 11(1-2), 95-119. doi: 10.1080/15377857.2012.642748
37 Turow, J., Carpini, M., Draper, N., & Howard-Williams, R. (2012). Americans roundly reject
tailored political advertising at a time when political campaigns are embracing it. Scholarly Commons, 1-28. Philadelphia, PA: University of Pennsylvania, Annenberg School of Communication. Stable URL: http://repository.upenn.edu/asc_papers
De Vreese, C. H. (2005). The spiral of cynicism reconsidered. European Journal of Communication, 20(3), 283-301. doi: 10.1177/0267323105055259
Warner, B. R. (2010). Segmenting the electorate: The effects of exposure to political extremism online. Communication Studies, 61(4), 430-444. doi:
10.1080/10510974.2010.497069
Ziliani, C., & Bellini, S. (2004). Retail micro-marketing strategies and competition. The International Review of Retail, Distribution and Consumer Research, 14(1), 7-18. doi: 10.1080/0959396032000154266
