Negative Campaigning - ein Muss im modernen Wahlkampf (auch in Deutschland)?

Der US-Wahlkampf – besonders der Vorwahlkampf der Republikaner in South Carolina – gilt häufig und fast schon traditionell als schmutzig. Die Angriffe auf den politischen Gegner gelten entsprechend als probates Mittel, um erfolgreich zu sein. Nicht zuletzt muss sich Jeb Bush nach seinem Ausscheiden aus dem Rennen jetzt auch noch vorhalten lassen, dass der Großteil des Geldes seines SuperPACs für „positive Werbung" ausgegeben worden ist. Dabei schwingt natürlich mit, dass man mit „negative campaigning" erfolgreicher hätte sein können. Stimmt das? Funktioniert negative campaigning? Und vor allem funktioniert es auch in Deutschland?

Diesen und anderen Fragen ist Daniel Schmücking nachgegangen. Seinen lesenswerten Aufsatz gibt's exklusiv hier. Wer etwas mehr Zeit hat, sollte sich auch sein Buch zum Thema durchlesen.

The Impact of Negative Political Advertisement on Voter Behaviour in Germany – an Experiment

Autor: Dr. Daniel Schmücking

1    The Impact of Negative Campaigning in Germany – Current State of Research

2    Hypotheses and Operationalization

3    Study Design

4    Subjects

5    Results

6    Discussion

7    References

Negative campaigning has been a feature of German political campaigns from the very beginning of the Federal Republic. At the same time, it is extremely unlikely that political actors would ever admit to using negative campaigning. Indeed, public opinion views such methods in a negative light, considering it to be more associated with mudslinging, lying, and playing dirty tricks rather than a legitimate campaign tactic. Initially therefore, this calls for a precise definition of negative campaigning. A broad definition is used in this study: it is seen as a critique of the opponent, whether it is personal, issue-based, legitimate, aggressive, or with regard to the extent to which the campaign is conducted on a negative platform. This means the definition makes it much easier to distinguish between negative and positive campaigning.


There has been surprisingly little research on the impact of negative campaigning despite the number of examples attributed to German campaigning in the past. Indeed, low research interest over the last few decades has meant that no one can really say exactly how it works and how effective it is. Compared to the US, there is limited research in Germany on the effects of negative campaigning, however the research from the US may very well be used to break ground and analyse its impact here in Germany.


The central idea of this paper is to examine the considerable difference between negative campaigning in Germany compared with that in the US. Therefore, a psychological experiment with students was performed in order to explore the changing attitudes of voters in the event of a direct attack compared to a comparative attack and positive campaigning. As a stimulus three different types of newspaper advertisement were used.
The results yielded a number of interesting insights. The voters evaluated direct attacks most negatively, but they did not observe a difference between comparative attacks und positive campaigning. There is no evidence that a comparative attack is perceived as more credible than a direct attack and, moreover, no evidence of a direct attack having more negative effects for the attacker or the attacked than a comparative attack would be expected to have. In saying this, the observed aggressiveness of the attack does have an impact on the attacker. An aggressive attack triggers more negative effects for the attacker than a nonaggressive attack. Yet the level of aggressiveness has no effect on the attacked. There is also no evidence that an aggressive attack brings about more negative effects for the attacked than a nonaggressive attack. This leads us to our next point, namely the importance of party identification and its high importance for the evaluation of the stimulus, the attacker and the attacked respectively. Supporters of the attacker estimate attacks on political opponents more positive than the supporters of the attacked, or indeed independent voters. Moreover, attacks on political opponents have a higher impact on the attitudes of the supporters of the attacker than on supporters of the attacked and independent voters. Nevertheless, this calls for a distinction to be made: the supporters of the attacker are fond of the attack and they evaluate the attacker much more positively, but this does not mean that they form any worse evaluations of the attacked, and for the supporters of the attacked and independent voters it is seen as irrelevant if there is an attack. In fact, negative campaigning in Germany is much more risky for the attacker than the impact it may have on the attacked party. As far as the latter is concerned, it is much more important how the voters see and feel the attack than how the attacker thinks it should work.


This leads us to a brief description on the outline of this paper. Firstly, I shall present an account on the current state of research before moving on to state and explain the hypotheses of this study and how they were operationalized. Third, I will attempt to describe the design of the study and the stimulus and then turn to present and discuss the results.

Nach oben

 

1    The Impact of Negative Campaigning in Germany – Current State of Research

There is a great deal of research that can be found on the impact of negative campaigning in the US-context. However, it is not clear if the US results (Ansolabehere & Iyengar 1997, Finkel & Geer 1998, Wattenberg & Craig 1999, Pinkleton 1997, Niven 2006, Newhagen & Reeves 1991, Lau et al. 2007) are transferable to other countries, in particular because of the differences in the election environments between the countries. There are different party systems, different electoral systems, different media systems, and even political cultures differ. Furthermore, the results of most American studies are not clear. This is especially apparent when it comes to answering the question whether negative campaigning is mobilizing or demobilizing. The answers are unfortunately inconsistent. Ansolabehere and Iyengar argue that negative campaigning is demobilizing, particularly for party loyalists (Ansolabehere & Iyengar 1997). However, Finkel and Geer disagree and for them, negative campaigning is mobilizing because it gives important information to the voters, which leads to higher political participation (Finkel & Geer 1998: 573ff).
In the case of Germany, all that can be found are a few studies on negative campaigning in general, and about the impact of negative campaigning in particular. In saying this, however, it is clear that research interest on the impact of negative campaigning in Germany has increased since the European Parliament campaign of the Social Democratic Party of Germany (SPD) in 2009. The SPD attacked their opponents with some humorous comic style posters and TV advertisements. Yet the research methods used still retain a relative simplicity in comparison with US studies. The German studies paid particular attention to posters and commercials. To use an example, Christina Holtz-Bacha analysed the impact of TV advertisements in a student discussion group. She concludes that the participants of the group did not like the campaign. They stated that they could recall the commercials easier because of their comic style, but they did not enjoy the concept of negative advertisement. This means that the SPD was successful in stirring attention for their advertisement, but the party did not succeed in creating negative feelings towards their opponents. Instead the commercials were in danger of creating a boomerang effect (Holtz-Bacha 2010: 173).


The SPD campaign in the European elections also led to another experiment conducted by Melanie Leidecker. This dealt with the differences between direct attack and comparative attack posters. In her results, she concludes that the posters and the attacker are rated most negatively if a direct attack takes place. For the comparative attack, on the other hand, the poster is rated more positively than the direct attack, but not as the attacked party. Yet as Leidecker herself makes clear, her study lacks certain important aspects. For example it is unclear how the attack advertisement has an impact on the attacked party and how the attack changes attitudes towards issues, parties, and candidates (Leidecker 2010: 117ff). All in all, the study does not extend particularly deeply: in the main it tests how voters rate negative posters. Not surprisingly, these posters are rated negatively – this is common sense by politicians and political scientists.


Four years earlier, in the Federal Election of 2005, two commercials drew the attention of the researchers. Here, both advertisements attacked the opponent directly and both were analysed via experiments by Jürgen and Michaela Maier. The SPD commercial was a reaction to one broadcast earlier by the CDU. It can also be said that the situation in 2005 was special because in general commercials in Germany had had a positive tone and did not react to each other. Long-term studies showed that criticism of the opponent in commercials for the federal elections took place in just 30 per cent of the cases (Maier & Maier 2007: 329ff). The reaction of the recipients was negative: both adverts did not inflict any damage the image of the opponents. Rather, they led to a boomerang effect on the party responsible for the advertisement and the candidate himself. Maier and Maier assume the cause of the negative rating to stem from recipients' different ideals as concerns adverts in Germany. In the mind of the recipients, the ideal commercial should present credible and convincing information in an interesting setting (Maier & Maier 2007: 329ff).


Another study by Kaid and Postelnicu deals with the same commercial as studied by Maier & Maier. In an online experiment with students, either the CDU ads or the SPD ads were shown separately, or both together. The researchers concluded that the CDU commercial negatively influenced the opinions on Chancellor Gerhard Schröder. Yet in watching the SPD ad or both ads, no effect on the opinions of the recipients about the candidates was found. This would indicate that the results are in opposition to the results of Maier and Maier (Kaid & Postelnicu 2006).
Finally, Klimmt et al. dealt with an aspect of negative campaigning in the German context, which has not really been touched on in US. This involved analysing the effect of humour on negative campaigning. Using an online experiment they analysed the differences between humorous and humourless negative posters and were able to demonstrate that humour has an influence on the perception of attacks. In particular, humour has positive effects for the attacker (Klimmt et al. 2007: 390ff) so we might say that humour can be a helpful tool in preventing the boomerang effect.

Nach oben

 

2    Hypotheses and Operationalization

The main focus of the experiment is based on the following question: are there any differences in the effectiveness between direct attacks, comparative attacks, and positive advertisement? In this paper, I shall argue that most of the results culled from American literature are transferable to Germany, but not all of them.


In US literature, negative campaigning is seen as something that is not very popular, but very effective. In contrast, I will argue that in Germany the recipients do not respond to negative campaigning as positively as their US counterparts. A direct attack will result in the most negative feelings towards the stimulus and a comparative attack will result in more negative feelings than positive campaigning.
I also expect that a comparative attack will be viewed as more credible than a direct attack, because it also demonstrates the position of the attacker. The US literature shows that a comparative attack is a useful tool which can be used to soften the danger of a boomerang effect. This is why I assume the direct attack has worse effects for the attacker then the comparative attack. A further aspect concerning the boomerang effect is the aggressiveness that is felt about the attack. I consider that the more aggressive the attack is perceived by the recipients, the worse the effects for the attacker will be.


Increasing or decreasing the boomerang effect is not the only important question, however. It is also interesting to ask: which variation of the attack has what effect on the attacked? I will embrace the idea that these variations, which serve to soften the danger of the boomerang effect, will hurt the attacked in a less harsh manner. This means that a direct or an aggressive attack is more effective against the attacked than a comparative or a nonaggressive attack.
In the US, the impact of negative campaigning differs due to the party identification of the voters. The more republican and the more conservative the voters are, the more effective negative campaigns are. In an analysis of the 2002 TV-Debate in Germany, Marcus Maurer and Carsten Reinemann achieved similar results (Maurer & Reinemann 2003): attacks against the opponent are evaluated much more positively by the supporters of the attacker. Supporters of the attacked and independent voters evaluate the attack much more negatively. I expect that this effect will also be found in this experiment. I also expect that attacks will have a higher impact on the attitudes of supporters of the attacker.
Based on these thoughts I have developed the following hypotheses:

Hypothesis H1: A direct attack is perceived more negatively than a comparative attack or positive campaigning.

Hypothesis H2: Positive campaigning is perceived more positively than a comparative attack.

Hypothesis H3: A comparative attack is perceived as more credible than a direct attack.

Hypothesis H4: A direct attack has more negative effects for the attacker than a comparative attack.

Hypothesis H5: An aggressive attack has more negative effects for the attacker than a nonaggressive attack.

Hypothesis H6: A direct attack has more negative effects for the attacked than a comparative attack.

Hypothesis H7: An aggressive attack has more negative effects for the attacked than a nonaggressive attack.

Hypothesis H8: Attacks against the political opponent are rated more highly by the supporters of the attacker than by supporters of the attacked and independent voters.

Hypothesis H9: Attacks against the political opponent have a higher impact on the attitudes of the supporters of the attacker than on supporters of the attacked and independent voters.

The hypotheses are operationalized as follows: to test the hypotheses H1 and H2, the subjects are asked if they like the stimulus and the results are then compared between the three groups. In a second step, an index is developed using several questions dealing with different feelings towards the stimulus such as whether it is considered to be funny, informative, credible, objective, positive, aggressive, serious, relevant or important. To test hypothesis H3, respondents are asked how credible they think the stimulus is. These results are then also compared between the three groups. To test hypotheses H4, H6 and H7 the effect of the stimulus towards the attacker is analysed. This is accomplished using an index which is made up of different aspects concerning the attacker such as sympathy, competence, and credibility. To test hypotheses H5 and H7 it is first asked how aggressive the stimulus is seen by the subjects. Then the correlation between aggressiveness and effectiveness perceived is analysed.


To test the hypotheses H8 and H9 the subjects are asked if they are supporters of a certain party in the questionnaire before the stimulus. With this information the subjects can be divided into three groups: supporters of the attacker, supporters of the attacked and independent voters. To make the setting easier to understand, I do not differentiate between the two types of the attack (comparative and direct). I divide the stimulus groups into attack and no attack. This enables me to see whether party identification has an impact on the evaluation of the attacker, the attacked, and on the attack itself.

Nach oben

 

3    Study Design


A constant research environment and variation of the independent variable is a primary factor for an experiment. An experiment consists of different treatment groups, with the subjects are assigned to these groups randomly (Druckman et al. 2011: 20ff).
For the online experiment, the subjects were randomized in three groups: two experimental groups and one control group. In the two experimental groups, the subjects either witnessed a direct attack or a comparative attack. In the control group, the subjects were exposed to a positive advertisement. The advertisements were the same, except for a variation in the text. A positive advertisement was shown in the control group to study the important differences between different types of negative campaigning and positive campaigning.
To analyse what effect the stimulus has on a voter's attitudes and their electoral behaviour, the subjects responded to a series of questions from a questionnaire after having seen the stimulus. This method allows the impact of the stimulus to become clear, the idea being that due to statistical probability and because of randomization, all characteristics of the subjects are the same in the groups. If there are differences, it is because they have seen a different stimulus.


The research design is based on the study of Melanie Leidecker (Leidecker 2010: 117ff). The aim was to improve the methodological weaknesses of this study and to enlarge the research interest to cater for some other aspects. The main goal of her study was to find out how the subjects evaluate the stimulus. Her focus did not extend to the effects of the stimulus as to the evaluation of the attacker and the attacked, nor did it address the effects attributed to characteristics of the recipients.
In Leidecker's experiment the stimulus was a poster, whereas in the current experiment the stimulus used was a newspaper advertisement. Newspaper advertisements as well as posters are part of the standard repertoire for German campaigns. However for this experiment the newspaper advertisement contained some advantages. In general, more text could be placed in it, because the attention span while reading is higher than with posters. Especially when dealing with the question of how issue attacks work, this is an important advantage. In saying this there is another reason that a similar setting was used. Many scholars claim that experiments are not generalizable. It is important for the generalisation of experiments, that they are repeated at different times and at different places (Hamenstädt 2012: 112).
The concrete experimental set-up was as following:

Figure 17.1: Experimental set-up
Source: Own Illustration

 

M – measurement before stimulus: political interest and media behaviour
O – measurement after stimulus: attitudes towards politics, politician and to the stimulus
R – randomisation
Xd – experimental group 1 – direct attack
Xc – experimental group 2 – comparative attack
Xp – control group – positive campaigning


The subjects took part in the experiment online. The advantage of this was that subjects could participate in the experiment independently of time and place. Firstly, general aspects on media behaviour, political interests etc. were measured. After the first measurement the subjects were randomized into three groups. The randomization took place with a random variable in the online questionnaire tool. The subjects were not informed as to which group they were in. Following randomization, subjects were exposed to a different type of stimulus (direct attack – comparative attack – positive campaigning) in each group. The second measurement was taken directly after subjects had viewed the stimulus. The logic of the experiment is that if the results in this measurement are different, it is because of the differences in the seen stimulus (O1<>O2<>O3).

As mentioned above, there were three different types of stimulus – only the text was different. The visual design, the issue, the attacker, and the attacked were the same in every stimulus. The stimulus was a typical newspaper advertisement dealing with the issue of unemployment. The attacker was the Christian Democratic Union (CDU). The Social Democratic Party (SPD) was attacked. The content of the attacks were as following:
Direct attack: Vote for good work. Do not vote for the SPD. The last time the SPD led the federal government, they promised to lower the unemployment rate to less than 3.5 million. In the end, 5 million people were unemployed – the highest rate since unification. Today, 3 million people are unemployed. Vote for good work. Do not vote for the SPD.
Comparative Attack: Vote for good work. Vote for the CDU. The last time the SPD led the federal government, they promised to lower the unemployment rate to less than 3.5 million. In the end, 5 million people were unemployed – the highest rate since unification. Since the Union has been in power, the rate has decreased steadily. Today, 3 million people are unemployed. Vote for good work. Vote for the CDU.
Positive: Vote for good work. Vote for the CDU. Since the CDU has led the federal government there are now 2 million unemployed people less. With 3 million unemployed people it is the lowest rate since unification. Vote for good work. Vote for the CDU.

 
 

4    Subjects

The experiment was performed using students of sociology and political science from the Friedrich-Schiller-University in Jena. 159 subjects participated and 136 finished the experiment. The drop-out rate was 14.5 per cent – considered to be fairly low. The following data analyses the cases that completed the experiment only.


The data set was cleaned before any analysis could be undertaken. The cleansing followed the suggestions of Delev Lück (Akremi et al. 2011: 66). The subjects who were not from Germany were excluded (one subject), as well as those who did not complete more than ten per cent of the questionnaire (five subjects) and those, whose answers could not be right (two subjects)1.  128 cases were analysed all together.
Firstly, it was necessary to confirm whether randomization was successful. In general, randomization is successful when there are no significant differences in the following aspects: age, gender, education, origin, political interest, and party identification.


The three groups in this experiment did not have the same size because randomization was a simple non-restrictive randomization. This problem can be solved with a randomization with urns or much more cases only. Nevertheless, all three groups were still larger than 30 subjects, so all the typical significance tests could be used. The subjects were distributed to the three stimuli as follows:

 

 

Frequency

Per cent

Direct

51

39.8

Positive

40

31. Mrz

Comparative

37

28. Sep

Total

128

100.0

Table 17.1: Distribution of the subjects in the groups
Source: Own calculations.

 

The distribution of gender in the three groups was not significant (Cramer's V: p>0.052). Altogether, more men (61.4 percent) attended the experiment. The average age of the subjects was 21.3 years. This means that there were also no significant differences between the groups in the mean of age (Anova: p>0.05). Taking party identification into consideration (Table 17.2 below), it is clear that there were no differences between the groups, but that the distribution was not the same as in the German population. CDU/CSU and the FDP were underrepresented. Over 90 percent of the subjects were not CDU supporters. This may have an effect on general evaluation of the advertisement, because the attacking party in the experiment was the CDU. I expect that all three groups rated the advertisement poorly. Nevertheless, the results show that randomization with regard to party identification was also successful (Cramer's V: p>0.05).

 

 

 

CDU/ CSU

SPD

FDP

Grüne

Linke

Piraten

None

Sum

Direct

Frequency

5

12

0

7

7

5

15

51

Percentage

9.8

23.5

0.0

13.7

13.7

9.8

29.4

100.0

Positive

Frequency

4

11

1

4

7

2

11

40

Percentage

10.0

27.5

2.5

10.0

17.5

5.0

27.5

100.0

Comparative

Frequency

4

11

0

2

5

3

12

37

Percentage

10.8

29.7

0.0

5.4

13.5

8.1

32.4

100.0

Total

Frequency

13

34

1

13

19

10

38

128

Percentage

10.2

26.6

0.8

10.2

14.8

7.8

29.7

100.0

Table 17.2: Distribution party identification
Source: Own calculations, Cramer's V: p>0.05

 

Another aspect is the very high political interest of the subjects. Table 17.3 demonstrates that 67.2 percent of the subjects had a high or very high political interest. I expect that it is much more difficult to influence voters with a high political interest with political advertisement. Nevertheless, in this case randomization was also successful. There were no significant differences between the groups concerning political interest (Cramer's V: p>0.05).

 

Political interest

Frequency

Percentage

Very strong

42

32.8

Strong

44

34.4

Moderate

32

25.0

Weak

6

4.7

Very Weak

4

3.1

Total

128

100.0

Table 17.3: Political interest of the subjects
Source: Own calculations, Cramer's V: p>0.05

 

In total, it can be said that randomization was successful because there were also no significant differences between the groups concerning origin (Cramer's V: p>0.05) and education (Cramer‘s V: p>0.05).3

Nach oben

 

5    Results

Perception of the Stimulus

First, I will present the differences in the evaluation and perception of the stimulus. The likeability of the stimulus saw a general rejection towards it. This might be due to the very low CDU party identification of the subjects4.  Moreover, independent from the stimulus, 72.7 percent of the subjects did not like the advertisement they had seen5.  Despite this basically negative tendency there were differences between the groups: 86.3 percent did not like the direct attack, 67.7 percent did not like the comparative attack and 60.0 percent did not like the positive stimulus. There was a small but significant effect (Cramer's V: V=0.196, p<0.05) between the type of stimulus and how the subjects liked it. Table 17.4 below highlights that the effect could also be seen in the analysis of the arithmetic mean.6

  Mean N
Positive Feb 23 40
Comparative Feb 14 37
Direct Jan 67 51
Total Jan 98 128

Table 17.4: Means liking Stimulus
Source: Own calculations; p<0.05 eta-squared=0.0697 ; range of values: 1 (do not like the advertisement at all) to 5 (like the advertisement very much).

 

In general, the differences between the groups were significant, but there was no substantial difference between positive advertisement and the comparative attack. The effect size was medium.
To take a closer look at the feelings of the subjects towards the stimulus, other emotional aspects were asked and these items were used to build an additive index.

Figure 17.2: Distribution of means of the perception of the stimulus
Source: Own calculations; Anova: p<0.001; eta-squared=0.128; range of values 9 (very negative perception) to 36 (very positive perception).

Figure 17.2 above indicates quite clearly that the direct attack was perceived much more negatively than the comparative attack and the positive advertisement. There was only a very small difference between comparative attack and positive advertisement. The differences were highly significant and the effect size was medium. The post-hoc analysis Bonferroni and Scheffé show that there were no significant differences between the comparative attack and the positive advertisement.


With these results hypothesis H1 can be confirmed. Hypothesis H2 must be rejected, however. This means, that a direct attack was perceived most negatively while there was no difference between the comparative attack and the positive advertisement.
Hypothesis H3 expects that a comparative attack is perceived more credibly than a direct attack. The data shows that this was not the case. Although 33.3 percent considered the direct attack to be credible, only 27 percent thought the comparative attack is credible.8
It is remarkable that only 15 percent thought that the positive advertisement is credible. In the three groups were the following mean differences:

  Mean N
Comparative 2.03 37
Direct 2.02 51
Positive 1.83 40
Total 1.96 128

Table 17.5: Means credibility
Source: Own calculations; p>0.05; range of values: 1 (the advertisement is not credible) to 4 (the advertisement is credible).

In this case it can be considered that differences were not significant. This means, that although there were indications, there was no evidence that a positive advertisement was perceived less credibly compared to the direct attack.
The experiment not only polled how credible the subjects find the stimulus, several other emotions were also asked. In particular, it was important to ascertain which aspects were significant differences and which were not. Following a variance analysis it could be recognized that the following aspects were not significant between the groups: funny, informative, and credible.


Significant differences could however be found in the following aspects: objective, positive, aggressive, serious, relevant, and important. Figure 17.3 below shows the exact means of these aspects in the groups.

Figure 17.3: Significant means reception of the stimulus per group
Source: Own calculations; p<0.05; range of values: 1 (very negative evaluation of the stimulus) to 4 (very positive evaluation of the stimulus).

As regards the aspects of objectivity, positive-negative, and aggressiveness the positive stimulus was evaluated in a mostly positive manner. In these aspects, the comparative stimulus was evaluated more positively than the direct attack. The seriousness of the comparative attack was rated best, but the differences between the groups ‘positive' and ‘comparative' were not significant. In the aspects of relevance and importance, which are quite equal, the comparative attack was rated mostly positive.

Perception of the Attacked

After analysing the reception of the stimulus, our focus moves to the effect of the advertisement on the attacked. Here, the data shows no significant differences between the groups concerning the attacked. No significant difference can be found in the general evaluation or in the evaluation of the credibility, competence or sympathy. There was no evidence that the advertisements have an influence on the attacked, even where this extended to the index built of the afore-mentioned items or in the evaluation of the competence in solving the problem of unemployment. Because of this, hypothesis H6 has to be rejected. It means that there was no evidence that a direct attack has more negative effects for the attacked then a comparative attack.

Perception of the Attacker

The results for the attacked were quite similar to those for the attacker. No evidence could be found to prove that the stimulus has an impact on the evaluation of the attacker. As a result, hypothesis H4 must also be rejected. This means that there was no evidence that a direct attack has more negative effects for the attacker than a comparative attack.

Influence of Perceived Aggressiveness

To test the hypotheses H5 and H7, the main focus was set on the perceived aggressiveness of the stimulus. The independent variable was not the stimulus itself, but the emotions caused by the stimulus. To analyse this, the variable aggressiveness was recoded in two categories: aggressive and non-aggressive. Firstly, it was necessary to see how perceived aggressiveness was distributed in the three groups.

    Direct Positive Comparative Total
Aggressive Frequency 45 13 20 78
% in the groups 88.2 32.5 55.6 61.4
Not-aggressive Frequency 6 27 16 49
% in the groups 11.8 67.5 44.4 38.6
Total Frequency 51 40 36 127
% in the groups 100.0 100.0 100.0 100.0

Table 17.6: Distribution of aggressiveness in the groups
Source: Own calculations, p<0.001; Cramer's V: V=0.487.

The direct attack was perceived to be most aggressive and the comparative attack was similarly seen to be more aggressive than the positive advertisement. There was a high significant and medium effect. Another question that arises here is: how did the perceived aggressiveness of the stimulus influence the evaluation of the attacker? In the analysis of the general evaluation of the CDU/CSU it becomes clear that there was a significant (p<0.05) difference in mean (aggressive=4.86; not-aggressive=5.90). Although the effect was small (eta-squared=0.032), it means that the CDU/CSU as the attacker was rated negatively by those who felt that the stimulus was aggressive. The same result could be found in an index built of different ratings of the attacker such as competence, credibility, and sympathy etc. With this index there was a non-significant (p>0.05) small effect (eta-squared=0.024), which means hypothesis H5 can be confirmed. Therefore, we can say that an aggressive attack had more negative effects for the attacker than a nonaggressive attack.


The next question relates to the effect the perceived aggressiveness of the stimulus had on the evaluation of the attacked. In the general evaluation of the SPD there was no significance between those who thought the stimulus was aggressive and those who thought it was not. Also, the index built using different evaluations of the SPD showed no significant differences. This is why hypothesis H7 has to be rejected. It means that there was no evidence that an aggressive attack had more negative effects for the attacked than a nonaggressive attack.
The results demonstrated that the perceived aggressiveness had no effect on the attacked, but on the attacker. These effects could be found in a skill index (Table 17.7 below), which also dealt with competences of the CDU and the SPD.

  Mean N
Aggressive 23.27 51
Not-aggressive 20.74 34
Total 22.26 85

Table 17.7: Means of the skill index per aggressiveness
Source: Own calculations; p<0.01; eta-squared=0.094; the range of values for this index is between 7 (high competence for the CDU) and 35 (high competence for the SPD).

There was an effect in the evaluation of the attacker, which was caused by the perceived aggressiveness but not by the intended effect of the stimulus. In this case it was necessary to afford the effect a closer look by using a multivariate analysis of variance. Figure 17.4 below shows which effect the stimulus and the perceived aggressiveness had on the attacker.

Figure 17.4: Effect of the felt aggressiveness and the stimulus on the attacked
Source: Own calculations; p>0.05; range of values: 4 (very negative evaluation of the attacker) to 26 (very positive evaluation of the attacker); mean for all subjects: 13.93.

The subjects evaluated a comparative (16.37) or a direct attack (16.17) in a mostly positive manner when they did not feel that the attack was aggressive. An aggressive and a comparative attack received the worst evaluations. All the same these results cannot be considered as significant. Indeed, the effects on the attacked reached an even lower level of significance. The mean differences and level of significance were so low, that interpretation of the results could not be taken into serious consideration.

Influence of Party Identification

To test hypothesis H8, it was necessary to ascertain to what extent the subjects liked the stimulus.

Stimulus Party identification Mean N
Attack Supporter CDU 3.33 9
Supporter SPD 4.22 23
Independents 4.23 56
Total 4.14 88
No attack Supporter CDU 3.25 4
Supporter SPD 4.00 11
Independents 3.76 25
Total 3.77 40
Total Supporters CDU 3.31 13
Supporters SPD 4.15 34
Independents 4.09 81
Total 4.02 128

Table 17.8: Impact of the seen stimulus and the party identification on the likability of the stimulus
Source: Own calculations; p<0.05, eta-squared=0.092; range of values: 1(like the advertisement very much) to 5 (do not like the advertisement at all).

Table 17.8 above shows that an attack was more liked by the supporters of the attacker than by the supporters of the attacked. It is interesting to see that there was just a small difference between the supporters of the attacked and the independents. If there was no attack the supporters of the sender of the advertisement also liked the stimulus more than independents and supporters of the SPD. This means that in general, supporters of the advertiser are much more receptive to its political advertisements than supporters of the opponent or independents.
In the next step, an index was analysed which consists of different attitudes of the subjects towards the stimulus.

Stimulus Party identification Mean N
Attack Supporter CDU 23.22 9
Supporter SPD 17.27 22
Independents 17.69 55
Total 18.16 86
No attack Supporter CDU 24.33 3
Supporter SPD 20.00 11
Independents 19.78 23
Total 20.22 37
Total Supporter CDU 23.50 12
Supporter SPD 18.18 33
Independents 18.31 78
Total 18.78 123

Table 17.9: Impact of the seen stimulus and the party identification on the evaluation of the stimulus
Source: Own calculations; p<.0.01, eta-squared=0.134; range of values 9 (very negative perception) to 36 (very positive perception).

The effects were the same as with the likeability of the stimulus. The attack was evaluated positively by the supporters of the attacker and not as much by the supporters of the attacked and independents. With these results, hypothesis H8 can be confirmed.
Testing hypothesis H9 requires a closer investigation of the effects of party identification and the shown stimulus on the evaluation of the attacker. Table 17.10 below illustrates the distribution of the attacker index.

Stimulus Party identification Mean N
Attack Supporter CDU 23.57 7
Supporter SPD 13.83 23
Independents 12.91 56
Total 14.02 86
No attack Supporter CDU 21.25 4
Supporter SPD 13.45 11
Independents 12.64 25
Total 13.73 40
Total Supporter CDU 22.73 11
Supporter SPD 13.71 34
Independents 12.83 81
Total 13.93 126

Table 17.10: Impact of the seen stimulus and the party identification on the evaluation of the attacker
Source: Own calculations; p<0.001, eta-squared=0.269, range of values: 4 (very negative evaluation of the attacker) to 26 (very positive evaluation of the attacker).

Unsurprisingly, the attacker's own supporters evaluated him best, as it does not matter if there was an attack or not. But the attacker was rated best by his own supporters if he attacked. As for all those who were not supporters of the attacker, it did not matter if there was an attack or not. In general, a lot of the effects of negative campaigning on the evaluation of the attacker could be explained by party identification.
The same strong effects can be seen in the evaluation of the attacked. The index in Table 17.11 below can be interpreted in the same way as the index in Table 17.10:

Stimulus Party identification Mean N
Attack Supporter CDU 16.57 7
Supporter SPD 20.91 23
Independents 15.93 56
Total 17.31 86
No attack Supporter CDU 16.00 4
Supporter SPD 20.36 11
Independents 16.12 25
Total 17.28 40
Total Supporter CDU 16.36 11
Supporter SPD 20.74 34
Independents 15.99 81
Total 17.30 126

Table 17.11: Impact of the seen stimulus and the party identification on the evaluation of the attacked
Source: Own calculations; p<0.001, eta-squared=0.269; range of values: 4 (very negative evaluation of the attacked) to 26 (very positive evaluation of the attacked).

The results in Table 17.11 show that the attacked was evaluated best by its supporters, but it did not matter if there was an attack or not. The intended impact of negative campaigning, to weaken the attacked, could not be found either in relation to supporters of the attacker or independents.
With these results, hypothesis H9 can be confirmed. In saying this, however, it is necessary to differentiate between the results. The supporters of the attacker liked the attack and they evaluated the attacker much more positively, but they did not evaluate the attacked any worse. For the supporters of the attacked and the independents, it did not matter if there was an attack or not.

Nach oben

 

6    Discussion

Two questions are important for the discussion of the results. Firstly, how do the results combine with and complement the current state of research and what results can be generalised? Second, which weaknesses can be found in the research design and what lessons can be learned for other experiments?
In the following the results of this study concerning the current state of research will be discussed. The study of Leidecker shows that in the evaluation of the poster there are significant differences between the direct attack, the comparative attack, and the positive poster in the aspects serious, aggressive, and sympathetic. There are no significant differences in the aspects of entertaining, eye-catching, graphical design, credible, and creative (Leidecker 2010: 130ff).


Compared to Leidecker, some other aspects were analysed in this study. As described above, I found significant differences in the aspects objective, positive, aggressive, serious, relevant, and important. No significant differences were found in how funny, informative, and credible the stimulus was evaluated.


This means that the aspects of seriousness, aggressiveness and credibility were treated to the same evaluation in both experiments. There were significant differences between the groups in the aspects of seriousness and aggressiveness and in both studies, a significant difference in the aspect of credibility could not be found.
An investigation of the differences between the aspects aggressiveness and seriousness in the two studies found that quite similar results could be observed. In both studies the direct attack was perceived as most aggressive and the comparative attack was evaluated as more aggressive than the positive advertisement. There was a small difference between the studies in the aspect of seriousness. In this experiment the comparative attack was evaluated a fraction more seriously than the positive advertisement. In Leidecker's experiment it was the opposite. Yet in both studies the differences in seriousness between the positive advertisement and the comparative attack were not significant.
In the evaluation of the attacking party, Leidecker observed significant differences how aggressive, provocative, and ruthless aspects led the subjects to evaluate the attacking party. Leidecker found no significant differences in how sympathetic, competent, convincing, credible, dynamic, authentic, trustworthy or strongly the subjects evaluated the attacker (Leidecker 2010: 132ff). In the three significant aspects Leidecker discovered that the advertiser was evaluated best if there was no attack (Leidecker 2010: 132ff).
In my experiment no significant differences in the evaluation of the attacker could be found. Certainly, fewer aspects were analysed than in Leidecker's experiment these being: sympathetic, competent, and credible. The aspects analysed in both studies (sympathetic, competence, credible) were not significant in both experiments.


On the level of the attacked, Leidecker could not find any significant differences in how sympathetic, competent, trustworthy, and dynamic the subjects evaluated the attacked (Leidecker 2010: 133ff). My experiment presents the same results. No significant differences could be found in the analysed aspects (competent, sympathy, credible) of the evaluation of the attacked.
In conclusion, the results of the experiment of Melanie Leidecker could be confirmed by the results of this study. In general it became clear that the newspaper advertisement was rated negatively, regardless of whether it was an attack or not. The reason for this is probably that only a few subjects identified with the CDU. Because there was no real control group, it was impossible to discover whether the subjects were simply opposing political advertisement.


To date, subjects have not been particularly receptive towards negative campaigning (Holtz-Bacha 2001: 670f) but it is held that this observation must be seen in a more differentiated manner. In the experiment, it became clear that the direct attack was much less liked than the positive advertisement. But no significant differences between the positive advertisement and the comparative attack could be found, so only the direct attack was not liked by the subjects. How much the subjects liked the stimulus had an impact on the attacked (Table 17.12 below), but not on the attacker. This means that a direct attack was likely to harm the attacker, while the comparative attack had no negative effect. I could not find any effect in all three variations of the stimulus on the attacked. The probable reason for this is discussed below.

Liking of stimulus Mean N
Like it 18.42 12
Moderate 14.82 22
Do not like it 13.13 92
Total 13.93 126

Table 17.12: Impact liking of stimulus on the attacker
Source: own calculations; p<0.01; eta-squared=0.089; Range of values: 4 (very negative) to 26 (very positive).

Other aspects show, that the comparative attack has many advantages compared to the direct attack. Indeed, subjects viewed the comparative attack as most relevant and most important, even more so than the positive advertisement.


As referred to above, if the stimulus is taken as the independent variable, no impact on either the attacker or attacked could be observed. At the same time this assessment omits the fact that there is a discrepancy between the intention of advertisement's sender (the party, the candidate or in this case the researcher) and the real perception of the receiver of the advertisement. The perceived aggressiveness was important for the boomerang effect on the attacker. If an attack was perceived as aggressive, the attacker was rated significantly more negative, however I could not find an effect on the attacked. It is interesting that the party identification had no influence on how aggressive an advertisement is perceived. In conclusion, the perception of an attack is more important than the intention of the attacker. This means, at least in reality, that campaign material should always be tested by focus groups. And the attackers should keep in mind that negative advertising is evaluated in a different way by their own supporters, supporters of the opponent and independent voters. Negative campaigning is more relevant for the party's own supporters rather than for the rest.


Furthermore, it also became apparent that the stimulus only had an impact on the attacked and not on the attacker. This brings forth the question as to whether negative campaigning is nothing more than a danger and brings no discernable benefit to the attacker. On this point, it became clear that it is very important as to how the attack is made. That is, is the attack direct where the party omits to mention its own position, or is the level of aggressiveness too high meaning the danger of the effect the boomerang goes viral. This is why attacks need solid planning and the materials thorough testing, otherwise attacks could likely result in more damage than benefit.


Finally, I would like to discuss the research design with regard to lessons learned for further experiments. The most important question here is: why is there no effect on the attacked? Is there no effect on the attacked in reality or was it not possible to measure in this case? First of all, it should be kept in mind that the stimulus was quite weak. One sole newspaper advertisement is probably not sufficient to change long-term attitudes towards parties. In US studies, the subjects viewed commercials (Ansolabehere & Iyengar 1997, Finkel & Geer 1998). These TV ads involve the subjects on a much higher level because of their audio-visuality. In further experiments, therefore, it would be advisable to test whether commercials have a higher influence on subjects' opinion forming.


A further point to note is that the subjects were rather specialised. Students of sociology and political science do have a high political interest and a high aversion towards the CDU. Consequently, it is suggested that further experiments work to minimize the effect of the pre-conditions of the subjects. This could be achieved with a more representative sample of subjects or perhaps where the negative campaign takes place in a virtual scenario, where pre-conditions have a low influence.

Daniel Schmücking

Nach oben

7    References

Akremi, Leila, Nina Baur, and Sabine Fromm. 2011. Datenanalyse mit SPSS für Fortgeschrittene 1: Datenaufbereitung und uni- und bivariate Statistik, 3rd edn. Wiesbaden: VS Verlag für Sozialwissenschaften.

Ansolabehere, Steven, and Shanto Iyengar. 1995. Going Negative: How Attack Ads Shrink and Polarize the Electorate. New York: Free Press.

Brosius, Hans-Bernd, Haas, Alexander, and Friederike Koschel. 2009. Methoden der empirischen Kommunikationsforschung: Eine Einführung, 5th edn. Wiesbaden: VS Verlag für Sozialwissenschaften.

Druckman, James, Donald Green, James Kuklinski, and Arthur Lupia. 2011. Experiments: An Introduction to Core Concepts. In: James Druckman, Donald Green, James Kuklinski and Arthus Lupia. eds. Cambridge handbook of experimental political science. Cambridge: University Press, 19–41.

Finkel, Steven E., and John G. Geer. 1998. "A Spot Check: Casting Doubt on the Demobilizing Effect of Attack Advertising." American Journal of Political Science 42 (No. 2): 573-595.

Hamenstädt, Ullrich 2012. Die Logik des politikwissenschaftlichen Experiments. 1st edn. Wiesbaden, Münster: VS-Verl.

Holtz-Bacha, Christina. 2001. Negative Campaigning: in Deutschland negativ aufgenommen. ZParl (32), 669–677.

Holtz-Bacha, Christina. 2010. Politik häppchenweise: Die Fernsehwahlwerbung der Parteien zur Europa- und Bundestagswahl. In: Christina Holtz-Bacha. ed. Die Massenmedien im Wahlkampf: Das Wahljahr 2009, Wiesbaden, 166–188.

Kaid, Linda, and Monica Postelnicu. 2006. The Impact of Television Advertising in the 2005 German National Elections: Paper presented at the Midwest Political Science Association Convention, Chicago.

Klimmt, Christoph, Petra Netta, and Peter Vorderer. 2007. Entertainisierung der Wahlkampfkommunikation: Der Einfluss von Humor auf die Wirkung negativer Wahlwerbung. Medien und Kommunikationswissenschaft 55 (3), 390–409.

Lau, Richard R., Lee Sigelman, and Ivy Brown Rovner. 2007. "The Effects of Negative Political Campaigns: A Meta-analytic Reassessment." The Journal of Politics 69 (No. 4): 1176–1209.

Leidecker, Melanie. 2010. Angreifende Plakatwerbung im Wahlkampf - effektiv oder riskant?: Ein Experiment aus Anlass der SPD-Europawahlplakate 2009. In: Christina Holtz-Bacha. ed. Die Massenmedien im Wahlkampf: Das Wahljahr 2009, Wiesbaden, 117–139.
Maier, Jürgen, and Michaela Maier. 2007. Audience Reactions to Negative Campaigns Spots in the 2005 German National Elections: The Case of Two Ads Called "The Ball". Human Communication (10), 329–344.

Maurer, Marcus, and Carsten Reinemann. 2003. Schröder gegen Stoiber: Nutzung und Wirkung der TV-Duelle, Wiesbaden.

Newhagen, John, and Byron Reeves. 1991. Emotion and memory responses for negative political advertising. In: Frank Biocca. ed. Television and political advertising: Psychological Processes, Hillsdale, 197–220.

Niven, David. 2006. A Field Experiment on the Effects of Negative Campaign Mail on Voter Turnout in a Municipal Election. Political Research Quarterly 59 (2), 203–210.

Pinkleton, Bruce. 1997. The Effects of Negative Comparative Political Advertising on Candidate Evaluations and Advertising Evaluations: An Exploration. Journal of Advertising 26 (1), 19–29.

Wattenberg, Martin P., and Craig Leonard Brians. 1999. "Negative Campaign Advertising: Demobilizer or Mobilizer?" American Political Science Review 93 (No. 4): 891-899.

Nach oben

 

1 In both cases there was a conflict between the graduation from school and the current study situation.

2 Cramer's V is a symmetric measure for the strength of a correlation between two or more nominal scaled variables, if (at least) one of both variables has more than two categories. The results can be interpreted as follows: 0.1 - 0.3 small effect, 0.4 - 0.5 medium effect, >0.5 large effect.

3 Results not shown, but available on request.

4 Independently from the kind of stimulus, CDU supporters liked the seen advertisement significantly more than the others (p<0.01).

5 The variable BA01 with the categories "like it very much", "like it", "partly", "do not like it", "do not like it very much" was recoded into the variable BA01_Re with the categories "like it", "partly", "do not like it".

6 The values are calculated as the arithmetic mean, although the scale is ordinal. A low value means that the subjects do not like the stimulus.

7 The size of the effect is shown by eta-squared. Eta-squared describes the ratio of variance explained in the dependent variable by a predictor while controlling for other predictors. Eta-squared can be interpreted as follows: eta-squared=0.01 small effect, eta-squared=0.06 medium effect, eta-squared=0.14 large effect.

8 The variable was recoded from four into two categories.

 

Download als PDF

Daniel Schmücking - The Impact of Negative Political Advertisement on Voter Behaviour in Germany – an Experiment (PDF, 530KB)