1
Bachelor Thesis Social Psychology
The Influence of Information on Attitude Polarization
N. van der Veer Universiteit van Amsterdam
Student number: 10000091
Supervisor: J. (Jonas) Dalege MSc Date of submission: 02-06-2017
Word count: 4336
Abstract
In this study the effect of information on attitude polarization was researched. According to the Attitudes as Constraint Satisfaction (ACS) model by Monroe and Read (2008) greater knowledge of an attitude subject leads to self-generated attitude polarization . The Causal Attitude Network (CAN) model (Dalege et al.,2015) adds that people search for consonant information when exposed to new information regarding the attitude subject. This leads to the hypothesis that information amount increases attitude polarization. Secondly, a positive relation between attitude importance and attitude polarization was hypothesized. The third hypothesis was that information positively influences attitude importance. Students from the UvA (N=30) were divided in two attitude subject groups. For five consecutive days they read five neutral articles about one of two subjects and filled in attitude extremity and attitude importance questionnaires. No significant effects were found and therefore no hypotheses were supported. Future research should have more elaborated study designs and point out if information has a positive effect on attitude polarization..
2 The Influence of Information on Attitude Polarization
People nowadays seem to become subject to more and more extreme opinions. With the rapid development of technology and the individualization of Western societies, like the United States, people in Western cities are more and more isolated in their own social groups, which causes them to be confronted less with information that is dissonant with what they think or feel (Musterd & Ostendorf, 2013, p.1) . While it looks like attitudes of people are increasingly polarizing, the daily amount of information that is processed by the human brain increases (The American Diet, 2009). This is an interesting phenomenon of which the
underlying mechanisms should be understood to be able to draw useful conclusions from it. In this study we therefore research the effect of information on attitude polarization.
A common mistake is made when thought about attitude change and attitude
polarization. Well-known theories like the Elaboration Likelihood Model (Petty & Cacioppo, 1986) and the heuristic-systematic model (Chaiken, 1987) elaborate on the factors involved when attitudes change direction (from positive to negative and vice versa). However, in this study we focus on attitude polarization, which means an attitude gets more extreme than it already was. In an experiment by Lord, Ross & Lepper (1979) was found that death penalty advocates became more convinced of the effect, while opponents of death penalty became more convinced of the lack of an effect, after being presented with the same articles. Other experiments, such as that of time and thought effects on attitude polarization of Tesser & Conlee (1975) support the thought induced polarization model of Tesser (1978). According to this model, thought tends to make a person's beliefs more consistent and since attitudes are a function of beliefs, attitudes will polarize as a result of thought (Tesser, 1978).
A more recent study on attitude and attitude related phenomena, such as attitude polarization, is that of Monroe and Read (2008). Based on the tripartite view on attitudes (Rosenberg, Hovland, McGuire, Abelson, & Brehm, 1960), they proposed an attitude network model. In their Attitudes as Constraint Satisfaction (ACS) model a network represents an attitude object with related cognitions and beliefs. In such networks related cognitions and beliefs can be seen as connected nodes in which activation flows from one to another and consequently changes the activation of individual cognitions (Monroe & Read, 2008). Because of this, there are multiple reasons to think of attitudes as connectionist networks. Networks are able to account for a great variety of psychological phenomena, such as automatic processing, implicit attitudes, conditioning and priming (Bassili & Brown, 2005). One of the phenomena Monroe and Read (2008) simulated in a network model was self-generated attitude polarization. As Tesser (1978) found earlier, when a person spends extra
3
thought on an attitude, the attitude is likely to become more extreme. Besides, experiments have shown that greater domain knowledge results in greater polarization (Millar & Tesser, 1986). In line with Tesser (1978) simulation of self-generated attitude polarization in a
network model led to greater polarization for people with greater knowledge about the attitude object. This was caused by the fact that greater knowledge leads to more information to think about, which in turn leads to more activation of the network as a result of more attention to the attitude (Monroe & Read, 2008).
Where the ACS model by Monroe and Read (2008) captures the dynamics of attitude formation and change, it is limited in the ability to use dominant statistical analyses on emprical attitude data (Dalege, Borsboom, van Harreveld, van de Berg & Conner; 2015). Therefore, Dalege et al. (2015) proposed the Causal Attitude Network (CAN) model, a further elaborated formalized attitude measurement model. Because the tripartite model falls short in the assumptions for the attitude construct, Dalege et al. (2015) extend the conceptualization of attitudes. In this model attitudes are also conceptualized as networks of causally connected evaluative reactions (beliefs, feelings and behaviour toward an attitude object) as nodes with causal influences between them as edges. Edges are either excitatory or inhibitory and vary in
weight (strength of the influence). For this matter, people strive to have consistent
representations of their attitudes (to align their evaluative reactions), because it reduces energy expenditure. Furthermore, people search for information that conforms their attitude. Therefore, evaluative reactions are primed by other evaluative reactions (Dalege et al. ,2015). This leads to the fact that evaluative reactions self-align when the person interacts with the attitude subject and the connectivity of the attitude network increases. From this the
prediction can be derived that interacting in any form with an attitude subject strengthens the attitude and therefore leads to attitude polarization.
However, the CAN model has not yet been able to explain the process of attitude polarization. The catastrophe model of attitudes (van der Maas, Kolstein, & van der Pligt, 2003) does describe this process. The model states that attitudes come in two forms,
dependent of involvement (or attitude importance). When involvement is low, attitudes can change more easily When involvement is high with the attitude subject, attitudes are a lot more resistant to change. However, when involvement is high and positive and negative information balance each other out, a relative small amount of additional information can cause a sudden jump in attitude. For this matter, when the involvement is low the attitude valence can be seen on a continous scale, but when involvement is high attitudes are split in two groups. Therefore, Latané, Nowak and Liu (1994) concluded that when involvement
4
increases attitudes polarize. Figure 1 shows this cusp model of attitude change. As the model predicts, involvement creates greater divergence in attitude.
Figure 1. The cusp catastrophe model of attitude change
Thus, in this study the effect of information on attitude polarization was researched. Participants were asked to read five neutral articles in five consecutive days about one of two selected subjects. Information was operationalized by the amount of information read in the neutral articles. Each day the participants filled in a questionnaire about attitude extremity and attitude importance on both subjects. Attitude polarization was operationalized in terms of the increase of the attitude extremity scores over time.
From the literature three hypotheses were derived. The first hypothesis is that
information has a positive effect on attitude polarization. Therefore, it is expected that when a person reads more articles on the attitude subject the scores on the attitude extremity
questions will become more extreme. The second hypothesis is that there is a positive relationship between attitude extremity and attitude importance, as Latané, Nowak and Liu (1994) found earlier. The expectation is that when scores on the attitude extremity questions become more extreme, the scores on the attitude importance questions increase. Finally, the last hypothesis is that information has a positive effect on attitude importance. It is therefore expected that when a person reads more articles about the attitude subject, the scores on the attitude importance questionnaire increase.
5 Method
Pilot Study
A pilot study was done to test whether the attitude of our participants was neutral about multiple subjects that are thought to be less-known by first year students. Because the study focuses on attitude polarization it is necessary that participants start research with neither positive nor negative attitudes toward the subjects of the articles that were chosen (i.e. GM and MID). The pilot study consisted of five 5 points Likert scale items per subject. An example item is 'How much do you personally care about the subject?' with score 1 representing 'not at all' and score 5 representing 'very much'. The subjects that were
researched for the original study are E-numbers, interest rate of the ECB, mortgage interest discount, genetic modification and automation of the labor market. Results showed that people had the most neutral attitude and lowest attitude importance for GM as well as MID. Therefore, those subjects were chosen. Then five neutral articles about GM and five neutral articles about MID were gathered from newspapers and online sources.
Participants
The participants were 32 students at the University of Amsterdam. Age varied from 18 to 24 years old with an average of 20.2 years old. Of 32 in total 27 respondents studied
Psychology. Among the participants 19 were woman and 13 were man. They were gathered from lab.uva.nl, the online domain where first-year students are able to find all research with which they can obtain their mandatory participant points. In this research there is a condition in which participants only read the articles about Genetic Modification (GM) and a condition in which participants only read the articles about the Mortgage Interest Discount (MID). The 32 participants were equally divided into the GM condition (M= 20,6 years ; 9 women) and the MID condition (M= 19,7 years; 10 women).
Materials
For this research existing questionnaires were adapted. It consisted of 3 control questions to check whether the participant truly read the article, 3 attitude importance items and 3 attitude valence items. The control questions per article were self-created and contained some information of the article that only true reading would acquire. An example question is 'What did Wiebes do to prevent the website of the tax authorities from getting overloaded?' with three multiple choice answers. Three attitude extremity questions were based on the American National Election Studies (ANES) feeling thermometers (Malhotra & Krosnick, 2007). An example question is 'How positive/negative do you think about the subject?' These questions were also measured on a continuous slider with answer score 1 respresenting 'very
6
negative' and answer score 100 respresenting 'very positive'. Attitude polarization was measured as the absolute difference between the question score and the neutral point (50) of the questionnaire. The internal consistency of the questionnaire was fairly good for GM (α between .694 and 902) and very good for MID (α between .867 and .939). Attitude importance was measured with three questions based on an article about attitude strength (Krosnick, Boninger, Chuang, Berent & Carnot; 1993). The questions were measured on a continuous slider with answer score 1 respresenting 'not at all' and answer score 100 representing 'very much'. An example question is 'How much do you personally care about the subject?' The internal consistency of the questionnaire was high for both GM (Cronbach's alpha ranging from .838 to .916) and MID (Cronbach's alpha ranging from .780 to .913).
Procedure
After recruitment the participants were randomly assigned to either the MG condition or the MID condition. Participants first came to the laboratory to receive instructions and fill in the attitude valance and importance questions. After selection participants read the first article of five in total and again rated the attitude and attitude importance items. Every
consecutive day they were required to read one article and fill in an online questionnaire with the attitude questions, the attitude importance questions and control questions about the read article. To stimulate people to read the articles carefully 20 euros were raffled among the people who scored best on the control questions.
Data analysis
Firstly, the slider question variables were transformed into new variables that showed the absolute deviance of the question scores from the neutral point to measure attitude polarization. Secondly, mean variables were created to compare the attitude extremity and attitude importance per day. Then the correlations between attitude extremity and attitude importance for both subjects were calculated to check whether there is an association between the two variables, which is to be expected. After correlation calculation a three-way ANOVA was performed to search for a three-way interaction effect between condition, time and subject on attitude extremity and attitude importance. Then, a repeated measures two-way analysis of variances is performed for both attitude extremity and attitude importance to see whether any changes occur between the time points. Then the analysis proceeded with a separate repeated measures ANOVA for attitude extremity and attitude importance for each condition. This was to determine whether there were also differences in attitude extremity and attitude importance scores over time in each condition and if an interaction effect between
7
attitude subject and time appeared. Finally, the main effects per condition were researched to be able to say something about which actual groups differed.
Results
The cut-off score for participants to be included in the analysis was a correct score of 67% of the control questions. The third control question of the questionnaire for day 1 in the GM condition was incorrect. Since the answer did not match the text in the article, the question was excluded. In total 9 questions had to be answered correctly for GM and 10 questions for MID to be included. Two participants scored beneath the cut-off score and were therefore excluded.
A Pearson's correlation was run to determine the relationship between attitude and attitude importance of both subjects for each day. For GM the six correlations were found to be moderately positive ranging from .07 to .34, but all were not significant ( p < .05) (see Table 1).
Table 1. Correlations between Attitude Extremity and Importance for each day for GM Day 0 Day 1 Day 2 Day 3 Day 4 Day 5 Pearson's r
Significance
.22 .34 .12 .07 .17 .23
.241 .068 .545 .734 .361 .218
There were also no significant correlations between attitude extremity and importance for MID, with correlation values ranging from .12 to .32 (see Table 2). Therefore, no support for the second hypothesis was found..However, all correlations were found to be moderately positive, which could indicate that there potentially is a weak positive relationship between attitude extremity and attitude importance.
Table 2. Correlations between Attitude Extremity and Importance for each day for MID Day 0 Day 1 Day 2 Day 3 Day 4 Day 5 Pearson's r
Significance
.18 .18 .12 .32 .21 .32
.354 .353 .530 .086 .265 .089
Attitude Extremity
A mixed design analysis of variances on attitude extremity was performed with attitude subject and time as within-subjects variables and condition as a between-subjects variable. Mauchly's test of sphericity indicated that the assumption of sphericity had been
8
violated, χ2(14)= 27.54, p < .05, and since ε < .75 a Greenhouse-Geisser correction was applied (Field, 2013). In contrast with the expectations, no significant three-way interaction effect between subject, time and condition on attitude extremity was found, F(3.36,101.65) = 0.91, p = .453, η² = .032. It implies that between the conditions there were no differences in attitude extremity among the time points.
Then two-way repeated measures variance analyses were run for attitude extremity, one for each condition. In the GM condition, Mauchly's test of sphericity indicated that the assumption of sphericity had not been violated, χ2(14)= 23.55, p = .56. As can be seen in Figure 2, no meaningful pattern in the attitude extremity scores arose The interaction effect between subject and time on attitude extremity was not significant, F(5,70) = 0.45, p = .809,
η² = .031. It indicates that there was no difference in change of attitude extremity between the
subjects over time in the GM condition. This was not in support of the first hypothesis.
Figure 2. Deviant mean scores on attitude extremity for the GM condition
In the MID condition, Mauchly's test of sphericity indicated that the assumption of sphericity had not been violated, χ2(14)= 16.45, p = .296. Figure 3 shows that in the MID condition the
9
attitude extremity deviance scores for MID were increasing in a zigzag movement, with each score on the odd time points being higher than the score on the previous odd time point and each score on the even time points being higher than the score on the previous even time point. This could indicate that attitudes on MID were polarizing. However, the interaction effect between subject and time on attitude extremity was not significant, F(5,70) = 2.03, p = .085, η² = .13. Therefore, there were no differences in attitude extremity among the time points in the MID condition.
Figure 3. Deviant mean scores on attitude extremity for the MID condition
Furthermore, a simple effects analysis was run to investigate the main effects of time on attitude importance for each subject in each condition. First the simple effects analysis for attitude extremity was run with GM as subject. In the GM condition, no main effect of time on attitude extremity was found., F(3.18,44.49) = 1.20, p = .322, η2 = .079. The sphericity assumption was violated, χ2(14) = 24.58, p = .042, and therefore a Greenhouse Geisser correction was applied. It indicates that no difference was found in attitude extremity scores for GM between the different time points among the participants in the GM condition. In the
10
MID condition, again no main effect of time on attitude extremity was found,. F(2.46,34.40) = 1.71, p = .190, η2 = .11. The sphericity assumption was violated, χ2(14) = 33.25, p = .003, and therefore a Greenhouse Geisser correction was applied. This indicates that there was no difference in attitude extremity scores for GM between the different time points among the participants in the MID condition.
Then a simple effects analysis for attitude extremity was run with MID as subject. In the GM condition, no main effect of time on attitude extremity was found, F(3.06,42.79) = 1.71, p = .178, η² = .11. Mauchly's test of sphericity indicated that the assumption of sphericity had been violated, χ2(14)= 36.71, p < .01 GG correction applied. This indicates that there was no difference in attitude extremity scores for MID between the different time points among the participants in the GM condition. In the MID condition, again no main effect of time on attitude extremity was found, F(5,70) = 1.78, p = .143, η² = .11. Mauchly's test of sphericity indicated that the assumption of sphericity had not been violated, χ2(14)= 20.28, p = .128. This indicates that there was no difference in attitude extremity scores for MID between the different time points among the participants in the MID condition. All these findings are not in support of the first hypothesis.
Attitude Importance
Then again a mixed design analysis of variances was performed for attitude importance with attitude subject and time as within-subjects variables and condition as a between-subjects variable.. Again Mauchly's test of sphericity indicated that the assumption of sphericity had been violated, χ2(14)= 46.65, p < .001, and therefore a Greenhouse-Geisser correction was applied. There was no significant interaction effect between subject, time and condition on attitude importance, F(2.72,76.08) = 1.63, p = .193, η² = .055. It means that between the conditions there were no differences in attitude importance among the time points. This is not in support of the third hypothesis.
Two more repeated measures variance analyses were run for attitude importance. In the GM condition, a Greenhouse-Geisser correction was applied for violating the sphericity assumption, χ2(14)= 39.48, p < .001 As can be seen in Figure 4 no meaningful pattern has developed in attitude importance scores for both subjects. The interaction effect between subject and time on attitude importance was not significant, F(2.87,40.19) = 1.13, p = .346, η² = .075. It indicates there were no differences in attitude extremity among the time points in the GM condition.
11 Figure 4. Deviant mean scores on attitude importance for the GM condition
In the MID condition, Mauchly's test of sphericity indicated that the assumption of sphericity had been violated, χ2(14)= 35.50, p < .01. Figure 5 shows that in the MID condition the attitude extremity deviance scores for MID were also increasing in a zigzag movement, with each score on the odd time points being higher than the score on the previous odd time point and each score on the even time points being higher than the score on the previous even time point. This could indicate that attitudes on MID were polarizing. However, the interaction effect between subject and time on attitude importance was not significant, F(2.30,32.15) = 1.30, p = .290, η² = .085. Again, this implies that there were no differences in attitude extremity among the time points in the MID condition.
12 Figure 5. Deviant mean scores on attitude importance for the MID condition.
Furthermore, a simple effects analysis was run to investigate the main effects of time on attitude importance for each subject in each condition. Firstly, the simple effects analysis for attitude importance was run with GM as subject. In the GM condition, no main effect of time on attitude importance was found, F(5,70) = 1.77, p = .130, η² = .11. There was no violation of the assumption of sphericity, χ2(14)= 22.13, p = .08. It indicates that no
difference was found in attitude importance scores for GM between the different time points among the participants in the GM condition. In the MID condition, again no main effect of time on attitude importance was found, F(2.30,32.13) = 0.73, p = .508, η² = .050. A
Greenhouse Geisser correction was applied, since the assumption of sphericity was violated,
χ2(14)= 45.28, p < .001. Again, it means that no difference was found in attitude importance
scores for GM between the different time points among the participants in the MID condition. Finally, a simple effects analysis for attitude importance was run with MID as subject. In the GM condition, the assumption of sphericity was violated, χ2(14)= 40.22, p < .001.
13
No main effect of time on attitude importance was found, F(1.88,26.33) = 0.89, p = .417, η² = .060. Therefore, no difference was found in attitude importance scores for MID between the different time points among the participants in the GM condition. In the MID condition, the assumption of sphericity was violated,χ2(14)= 24.75, p < .05. In the MID condition, again no main effect of time on attitude was found., F(2.52,35.25) = 2.09, p = .129, η² = .13. At last, no difference was found in attitude importance scores for MID between the different time points among the participants in the MID condition. All these results do not support the third
hypothesis. For both combinations of condition and subject no support for a relationship between time and attitude extremity as well as between time and attitude importance was discovered.
Discussion
This study aimed to give more insight in the effect of information on attitude polarization. According to the CAN model (Dalege et al. , 2015) attitudes polarize if
information increases and motivation is low. Furthermore, the catastrophe model of attitudes (van der Maas, Kolstein, & van der Pligt, 2003) states that when pro/contra information balance each other out, a small amount of additional information can cause sudden attitude polarization. By presenting frequent, neutral information to the participants in this study we tested if attitudes would polarize. However, the results do not support these expectations. No difference in attitude about both subjects between the time points was found. Besides, the attitude importance for both subjects did not change over time. This was expected for the subject they did not read articles about, but not for the subject they did read about. Although the results do not support the hypotheses, the presented theories need not to be rejected as explanations for attitude polarization. On the contrary, new research should be done to investigate the proposed explanations for attitude polarization.
One of the main causes for these unexpected results can be the small sample size (n = 30). This small amount of participants causes the power of the study to be too low to find a significant effect. Because of limitations in time and reach, we were unable to attract more participants to our research. In this case, the effect of information on attitude polarization could exist, but the low power of the study caused it to be unfound. Future research should have greater sample sizes to be able to find significant relations between the variables.
Another shortcoming of the study was that at the start of the research the attitude towards GM was more extreme than the attitude towards MID, in contrary to the findings of the pilot study. It could have been caused by the small sample in the pilot study. Because of the small sample size, the finding that attitudes of participants about GM were relatively
14
neutral was not generalizable. The fact that no attitude polarization was found in this study can be the consequence of these more extreme GM attitudes at the start, which caused that the differences between start and end were too small to be significant. Future researchers on this subject should be cautious for extreme attitudes when starting their study.
Furthermore, the articles were not interpreted as neutral by all participants. One of them wrote in the questionnaire that the goal of our study was to investigate the effect of positive and negative information on polarization, while the articles were meant to contain neutral information. A consequence of this could have been that the participants filled in the questionnaires according to the valence of the information in the articles. Besides, the articles contained similar contents about the attitude subject. This could have caused that the
manipulation of information was incorrect. This could explain the unexpected pattern in the figures. In a next study the neutrality of the information should be tested in a pilot study as well.
Finally, the generalizability of the findings is not very high. All participants were undergraduate students. The subjects that were chosen (genetic modification and mortgage interest discount) could have been too unfamiliar for them, since both are difficult subjects young people do not often (or even never) read about. Perhaps the proposed explanation does not account for attitude polarization with such unknown subjects. Future research should test for this and/or create a greater variety of topics to draw conclusions about subject difficulty.
This study does not provide support for the presented theories and therefore it cannot be concluded that information positively influences attitude polarization. However, following these limitations it can be said that new research should be performed on the effect of
information on attitude polarization. If sample size is increased, neutrality of the information is guaranteed and participants' attitudes at the start are not polarized yet, possibly the expected relationships can be supported.
References
Bassili, J. N., & Brown, R. D. (2005). Implicit and Explicit Attitudes: Research, Challenges, and Theory.
Chaiken, S. (1987). The heuristic model of persuasion. In M. P. Zanna, J. M. Olson, & C. P. Herman (Eds.), Social influence: the ontario symposium (Vol. 5, pp. 3-39).
Dalege, J., Borsboom, D., van Harreveld, F., van den Berg, H., Conner, M., & van der Maas, H. L. J. (2015). Toward a formalized account of attitudes: The Causal Attitude Network (CAN) model. Psychological Review, 123, 2-22.
15
Field, A. (2013). Discovering statistics using IBM SPSS statistics. Sage.
Krosnick, J. A., Boninger, D. S., Chuang, Y. C., Berent, M. K., & Carnot, C. G. (1993). Attitude strength: One construct or many related constructs?. Journal of personality
and social psychology, 65(6), 1132.
Latané, B., Nowak, A., & Liu, J. H. (1994). Measuring emergent social phenomena:
Dynamism, polarization, and clustering as order parameters of social systems. Systems
Research and Behavioral Science, 39(1), 1-24.
van der Maas, H. L. J., Kolstein, R., & van der Pligt, J. (2003). Sudden transitions in attitudes.
Sociological Methods & Research, 32, 125–152.
http://dx.doi.org/10.1177/0049124103253773
Malhotra, N., & Krosnick, J. A. (2007). The effect of survey mode and sampling on
inferences about political attitudes and behavior: Comparing the 2000 and 2004 ANES to Internet surveys with nonprobability samples. Political Analysis, 286-323.
Millar, M. G., & Tesser, A. (1986). Effects of affective and cognitive focus on the attitude– behavior relation. Journal of personality and social psychology, 51(2), 270.
Monroe, B. M., & Read, S. J. (2008). A general connectionist model of attitude structure and change: The ACS (Attitudes as Constraint Satisfaction) model. Psychological
Review, 115(3), 733.
Musterd, S., & Ostendorf, W. (Eds.). (2013). Urban segregation and the welfare state:
Inequality and exclusion in western cities (pp.1). Routledge.
Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model of persuasion. In Communication and persuasion (pp. 1-24). Springer New York.
Rosenberg, M. J., Hovland, C. I., McGuire, W. J., Abelson, R. P., & Brehm, J. W.
(1960). Attitudes organization and change (pp. 15-64). New Haven: Yale University Press.
The American Diet: 34 Gigabites a Day (2009). Perceived on May 8th 2017, from https://bits.blogs.nytimes.com/2009/12/09/the-american-diet-34-gigabytes-a-day/
