The Moderating Effect of Connectivity
in a Causal Attitude Network (CAN)
Model on Potential and Felt
Ambivalence
Conducted on an Individual Level
Blaisse, A. N. M.
Master Thesis Social Psychology – University of Amsterdam
Supervisor: Jonas Dalege Student number: 10373365 December 2016 Abstract: 147 Words Total Word Count: 5072
1 Table of Contents Abstract ... 2 CAN Model ... 3 Connectivity ... 4 PCR Method ... 6 Present Research ... 7 Method ... 7 Participants ... 7 Procedure ... 8 Materials ... 8 Data Analysis ... 9 Results ... 10 Donor registration ... 10 Abortion ... 11 NIX<18 ... 12 Exploring Analysis ... 14 Discussion ... 15 References ... 17
2
Abstract
Attitude networks may differ in structure like connectivity and strength. One might find an attitude object interesting and form a clearly positive or negative attitude. This interest could also lead to mixed and torn feelings. Unknown is whether connectivity in a network has effect on the mixed and torn feelings which accompany felt ambivalence. The effect of
interest on connectivity, and the moderating effect of connectivity on potential and felt ambivalence was examined on an individual level. In this study, there were 106 individual participants. Participants filled in a survey assessing their attitudes about organ donation, abortion, and the new alcohol limit in the Netherlands. Results show a positive relation between interest and connectivity for one of the topics. Connectivity of the attitude network has no moderating effect in the relationship between potential and felt ambivalence. Future research may improve this analysis by collecting more representative data.
3
Life goes hand in hand with numerous choices, opinions, and attitudes. Some attitudes are clearly positive or negative, others can cause infinite struggles. For instance, are less alert people wronged by the active donor registration law, or is the benefit for those in need worth it? Is abortion a crime, even a murder, or is it ‘her body, her choice?’. And is the change in the alcohol limit in the Netherlands from the age of 16 to 18 a way to help adolescents to stay away from addictive substances while their brains are still developing, or is this new law leading to illegal drinking without supervision? Some people will support both sides of those questions, which is called potential ambivalence (PA; Newby-Clark, McGregor and Zanna, 2002). But they canevoke mixed and torn feelings (i.e. felt ambivalence [FA]), which individuals want to avoid. A new way to examine attitudes is to apply the Causal Attitude Network (CAN) model (Dalege et al, 2016). Every network might have its own structure and yet unknown is the effect of these structures on the aversive feelings of FA. This study examines the possibly moderating effect of connectivity in individual attitude networks between potential and felt ambivalence.
CAN Model
Borsboom and Cramer (2013) brought the Network Approach Psychology to light. In this approach symptoms of a disorder are mutually interacting and createa dynamical system. The symptoms are arranged in a network and form multiple pathways from one disorder to another. Every network consists of nodes (i.e. the symptoms) and edges between those nodes (i.e. the relation between two nodes). In this way, a small-world structure arises in which similar nodes tend to cluster (Dalege et al., 2016) and all nodes are interconnected and can be reached from every other node via shortcuts in just a few steps, even when big clusters are formed. Representing nodes in a network makes it visible to seewhich nodes are strongly connected or central. The Network Approach however, does not say anything about directionality of the causal relationship (Frewen, Allen, Lanius, & Neufeld, 2012).
Dalege et al (2016) adapted this approach and expanded it to the CAN model. In this model, an attitude is modeled as a network, consisting of evaluative reactions like beliefs, feelings, and behavior. In this model, the nodes function as the evaluative reactions and the edges form causal links from one node to another (i.e. the belief that organ donation is grateful causes the behavior to become an organ donor). A small-world structure is formed, and the CAN model works with empirical network models in which the nodes are part of a bigger system. It can deal with either actual data systems or even with complex systems like
4
attitudes (Barabási, 2001; Dalege et al., 2016). The edges between the different nodes represent a causal relation and can have an excitatory or inhibitory function. The influences can differ for different edges due to different weights on these edges. To conserve energy, evaluative reactions with the same valence are linked with an excitatory edge, and evaluative reactions with different valance with an inhibitory edge. In a network, similar evaluative reactions which are often simultaneously activated tend to cluster for a smooth information flow. When a node is activated, it causes readiness in the whole cluster for potential reactions.
The CAN model has proven to be a good model for understanding attitude networks, but more research is needed for a wider application. So far, the model focused on group level estimation of attitude networks and it could be a valuable addition to explore the model on an individual level. Group networks are interesting when exploring a general structure. To explore specific structures, it is important to view the networks at an individual level (Borsboom & Cramer, 2013). This way it is easier to target at a specific symptom or evaluative reaction to adjust the network.
Connectivity
Networks can differ in structure (i.e. centrality and connectivity). Connectivity tells something about the dynamics in an attitude network (Dalege, Borsboom, van Harreveld, & van der Maas, 2016a). The strength of an attitude can be expressed as the global connectivity in an attitude network (Dalege et al., 2016). High global connectivity implies that on average, all nodes are closely connected and are seen as strong attitudes. Dalege, Borsboom, Harreveld and van der Maas (2016b) conducted research on the connectivity hypothesis, which holds that strong attitudes correspond to highly connected attitude networks. When interacting with an attitude object, connections between nodes organize themselves to be able to deal with the object. When two nodes are connected with a strong positive edge, they will adopt the same valence. Activation in one of the nodes creates a state of readiness in connected nodes. This way, highly connected nodes rely on each other.
Key features of a strong attitude are stability (over time), resistance and impact on behavior. The aligned edges provide conviction to connected nodes they are having the right evaluative state: they match. In highly connected networks, this confirmation comes from many other nodes which convinces the attitude of its own strength. This way it is less affected by persuasion attempts and harder to change the attitude. Dalege et al (2016b) also found
5
support for the hypothesis that strength and connectivity in networks are positively related. High connectivity between nodes could be seen as closely spaced domino tiles influencing each other in a flow (Borsboom & Cramer, 2013). When activating node A, the connected node B will be activated as well. This way the nodes are dependent on each other in a high connected network, and less dependenton each other in a less connected network. Highly connected networks can predict behavior,whilst behavior will be better predicted by other factors (i.e. ideology) when connectivity in a network is low (Dalege et al, 2016c).
Strength could be well indicated by interest in an attitude object (Krosnick, Boninger, Chuang, Berent, & Carnot, 1993). Krosnick (1988) found that a link between attitudes and values, needs, and goals makes somebody interested in an attitude object. Attitudes that are important are frequently thought of and are often extreme, thus easy to retrieve. In Krosnick’s study on political choice it turned out that important attitudes had more impact on political choice than less important attitudes, which implies that important attitudes are better predictors of behavior than less important attitudes. In a follow-up study Krosnick and colleagues defined interest in an attitude object as the motivation to gather information about the subject (Krosnick, Boninger, Chuang, Berent, & Carnot, 1993). They examined the correlation between, among others, importance and interest and found strong positive
correlations. Still, they concluded that the two variables should be seen as independent. When often thinking about an attitude object, the attitude network will broaden and the knowledge about this object will expand and increase in strength. This does not imply a preference. It could be that there is interest to know a lot about an attitude object without a judgement, which can lead to an ambivalent attitude. Ambivalent attitudes are something the CAN model has not yet examined. Since connectivity and interest both predict behavior more as they increase, and a strong connected attitude network does not necessarily imply a univalent attitude, it is interesting to study the relation between interest and connectivity, and the moderating effect of connectivity on potential and felt ambivalence.
Ambivalence
In contrast to univalent attitudes (i.e. attitudes with strong positive or negative
evaluations), ambivalent attitudes have simultaneous strong positive and negative evaluations about the object (de Liver, van der Pligt, & Wigboldus, 2007). Newby-Clark, McGregor and Zanna (2002) conducted research on the difference between potential ambivalence (PA) (i.e. the presence of strong positive and negative evaluations) and felt ambivalence (FA) (i.e.
6
unpleasant feelings and mixed emotions due to the ambivalence). FA is present when contradictory evaluations are available at the same time. Until the contradictory evaluations are available, there is a lack of awareness of PA (van Harreveld, Rutjens, Rotteveel,
Nordgren, & van der Pligt, 2009). Whether both sides are available depends on contextual factors, which could make the PA salient (van Harreveld, van der Pligt, & de Liver, 2009). Ambivalent attitudes often have a longer reaction time, which was first thought to result from vague attitudes which makes it less facile to respond correspondinglyto the attitude. A more recent study conducted by de Liver and colleagues implies that this extended reaction time is a result of the assessment between the strong positive and strong negative evaluations in a highly-connected network (Bargh et al, 1992; de Liver, van der Pligt, & Wigboldus, 2007). Before a proper reaction concerning an ambivalent attitude can be given, an assessment must be made between the positive and negative components of the attitude, which extends the reaction time. Networks can still be strongly connected for potential ambivalent attitudes, but the evaluative reactions are less aligned than for univalent attitudes (Dalege et al, 2016c). When the attitude is unaligned and active but not in compliance, FA arises. For FA, all relevant nodes are active, so the urge to reduce the feelings of ambivalence will increase. It might be conceivable that the network of attitudes high in FA is more
connected than the network of an attitude high in PA because without connectivity it might not evoke significant feelings for individuals. The increased connectivity could lead to more activity in the network. In this study, we examine whether connectivity of attitude networks moderates the effect of PA on FA.
PCR Method
Previous studies examined the networks at the group level. A new method needs to be used to estimate individual attitude networks and the Perceived Causal Relations (PCR) (Frewen, Allen, Lanius and Neufeld, 2012) method is the one we selected to do so. In their research on causal risk factors for posttraumatic stress and anxiety disorder, Frewen and colleagues asked respondents about symptoms and causal associations. Thus, the PCR method allows people to indicate ideas about the causal links between present symptoms. Frewen and colleagues first asked respondents how often certain symptoms occurred. In the next set of items, the occurring symptoms were offered in pairs after which respondents had to assess the extent to which they thought one symptom causes the other, and the other way around. (i.e. to
7
of the causal effect on the relation between two symptoms was measured. Another study conducted by Frewen examined comorbidity with the presence of a third disorder Z (Frewen, Schmittmann, Bringmann, & Borsboom, 2013). In this more complex study PCR turned out to be a systematic approach to notice which variable correlates to which other variables, even in a complex web of causal chains (Frewen, Schmittmann, Bringmann, & Borsboom, 2013). This method is often used for clinical set ups however,this study examines the application to attitude networks. Attitudes can be seen as a complex system of beliefs, feelings and
behaviors which, just like symptoms, work in a structured causal network.
Present Research
In the present study, participants will fill in a questionnaire about ambivalent attitude objects in which the connectivity of their networks is measured using the PCR method. Next to this we examine whether FA is present. From the CAN model perspective, it is expected that higher connectivity in the network will lead to a need of consistency thus the network wants to adapt, which causes the mixed feelings of FA. In addition, the degree of interest will be measured to confirm if a higher amount of interest is related to a more connected network structure. It is expected that potential and felt ambivalence correlate, but the degree of this correlation will depend on the connectivity in the attitude network. More connectivity will then lead to a higher correlation between PA and FA.
Method
Participants
A power analysis was conducted using the program G* Power to determine the sample size of our study. It turned out at least 156 participants are needed to detect a moderate effect (alpha = 0.05, effect size = 0.33, Power = 0.8, Partial η2
= 0.1). Due to time constraints, a smaller sample was used in this study. One hundred and thirty-seven individuals participated in this study. Among them were 25 men and 111 women with an average age of 20 years old (SD = 2.01). The study was conducted among undergraduate students at the online UvA-lab (www.lab.uva.nl). Students from the University of Amsterdam (UvA) were asked to fill in the online questionnaire. In gratitude for participating in the study, participants received 0.5 participation credit. All participants read and signed an informed consent.
8
Procedure
Prior to this study a small pilot study (n=28) was conducted to understand the attitudes content of the target population on the selected topics. The instructions of the survey were placed above the items in an online setting using Qualtrics. The survey addressed three topics; abortion, donor ship and alcohol limit. A short explanation for every topic was placed prior to each topic. The main survey consisted of four sections; the two parts of the PCR method (concerning PA and connectivity), FA, and interest. After conducting the survey participants filled in demographic information (age, gender and occupation) and received compensation for participation.
Materials
Attitude Items. The pilot study provided an insight into the most commonly used
words for each topic. The five most commonly used positive words (i.e. grateful, precious, and helpful), and the five most commonly used negative words (i.e. scary, insecure, and laziness) concerning each topic where used in the main survey. Those ten words function as the attitude items in the PCR.
Potential Ambivalence. PA was measured in the first part of the PCR method by using
the non-partitioned dimensions technique. This technique was found the most comprising method in research conducted by Refling and colleagues (Refling et al, 2013). The ten attitude items, which had a satisfactory alpha for each topic (organ donation: alpha = 0.71, abortion:
alpha = 0.48, and alcohol limit: alpha = 0.54), were shown in a list and had to be rated on a
7-point scale as how well it fits with the topic (1 = strongly disagree and 7 = strongly agree). To work with these scores, positive and negative attitude items were matched into five positive-negative pairs. The five pairs were then used in the formula described by Newby-Clark, McGregor, & Zanna (2002): weak^2/strong. In all five pairs, the one with the lower score stands for the ‘weak’ part of the formula, and the higher score for the ‘strong’ part of the formula. For each attitude object an average score was calculated per participant. The higher this mean, the greater the PA.
Connectivity. The second part of the PCR examined connectivity. The ten attitude
items were matched with each other to form 45 combination items. Every combination was examined unilaterally. Participants rated whether a relation exists between the two items in the combination item. This could either be a positive or negative relation. The combination items were scored on a 7-point scale from 1 (strongly disagree) to 7 (strongly agree). To measure the connectivity in an attitude network, the Likert scale points were added. The
9
higher this score, the more related the items were seen to each other, and thus the more connectivity in an attitude network.
Felt Ambivalence. To which degree FA was present was measured using two items per
attitude object modeled after Newby-Clark, McGregor, & Zanna (2002). The two questions have a satisfactory internal correlation for each subject (organ donation: r = 0.71, p < .001, abortion: r = 0.55, p < .001, alcohol limit: r = 0.57, p < .001). The first item asked ‘I have
strong mixed emotions both for and against abortion (alcohol limit/donor ship), all at the same time’. The second item asked ‘I do not find myself feeling torn between the two sides of the issue of abortion (alcohol limit/donor ship): my feelings go in one direction only’ (reverse
scored). The response was rated on a 7-point scale (1 = strongly disagree and 7 = strongly agree). The mean for the two items was used as a measure of FA. A high score on these items implied a high degree of FA.
Interest. Four items were used to measure interest in the attitude objects, based on the
items used by Krosnick, Boninger, Chuang, Berent, & Carnot (1993), which had a satisfactory alpha (organ donation: alpha = 0.83, abortion: alpha = 0.81, alcohol limit: alpha = 0.85). For all three attitude objects, the items asked about paying attention to information on the issue, how interested participants were in obtaining information about the issue, how closely attention was payed to stories about the issue and how important information about the issue was for them. Participants responded on a 3-point scale from 1 (never) to 3 (always). Per topic, the scores on the four questions where added for each participant. The higher this score, the more interest is shown in the topic.
Data Analysis
In this study the data processing program Statistical Package for Social Sciences (SPSS) was used. All three topics were analyzed separately. First, Spearman’s rho was computed to analyze the relation between interest and connectivity. It was expected that interest and connectivity are positively related. After this a multiple regression analysis was conducted (dependent: FA, independent: PA, connectivity and an interaction between PA and connectivity). A moderating effect of connectivity and ambivalence was expected with PA leading to FA when connectivity is high. After this, some explorative analyzes were
computed: the same regression analysis was computed, screening out very slow or very quick participants to correct for any laxness. Finally, the correlation was measured between PA and connectivity.
10
Results
Some participants did not finish the survey completely. Only complete surveys were included in the analysis. Participants who admitted non-serious participation were excluded from the analysis. In the end 137 participants were included in the analysis. For the scores on PA and FA, mean scores were calculated for each participant. For the scores on interest and connectivity the sum was calculated for each participant.
Donor registration
Relation between interest and connectivity. Since interest turned out not to be
normally distributed, a Spearman analysis was conducted to analyze the correlation between interest and connectivity. Figure 1 visualizes that no significant correlation was found (r = 0.02, p = .828). The expectation of high interest leading to high connectivity was not confirmed.
Figure 1. The Effect of Interest on Connectivity. No Significant correlation was Found.
Effects of Potential Ambivalence and Connectivity on Felt Ambivalence. All relevant
assumptions were met. The degree of PA predicts the degree of FA significantly (b = 0.79, t = 6.91, p < .001). The coefficient b is positive; the more PA is experienced; the more FA will be experienced. The degree of connectivity was not significantly predicting FA (b = 0.001, t = 0.07, p = .943). R-square indicates that the degree of FA is predicted by degree of PA and degree of connectivity for 27,2%.
The estimated regression model does declare a significant proportion of the variance (F = 25.03, df = (2, 134), p <.001). PA has the most predictable value in this model (beta = 0.52). When including an interaction variable (PA * Connectivity) in the model, the
11
significantly 0.01% (F change = 2.25, df = (1, 133), p = .136). Thus, no interaction effect between PA and connectivity on FA was found (see Figure 2.).
Abortion
Relation between interest and connectivity. Since interest turned out not to be
normally distributed, a Spearman analysis was conducted to analyze the correlation between interest and connectivity. A significant correlation was found (r = 0.27, p = .002). The hypothesis was confirmed: a higher amount of interest is related to a higher amount of connectivity in an attitude network (see Figure 3.).
Figure 3. The Effect of Interest on the Degree of Connectivity in an Attitude Network
Effects of Potential Ambivalence and Connectivity on Felt Ambivalence. All relevant
assumptions were met. Just as for the topic of donor registration, the degree of PA predicts the degree of FA significantly (b = 0.66, t = 4.64, p < .001). The coefficient b is positive: the higher the degree of PA, the higher the degree of FA. Connectivity was also predicting FA
12
significantly (b = -0.01, t = -2.08, p = .039). The effect of connectivity on FA was negative: the more connectivity, the lower FA. Degree of PA and degree of connectivity predict 15,4% of FA.
This model does declare a significant proportion of the variance (F = 12,15, df = 2, 134, p = .001). And again, PA is the strongest predictor (beta = 0.37). In the second model, an interaction variable (PA * Connectivity) was added. The difference in explained variance in this new model is 1,2%, (F change = 1.87, df = (1, 133) p = .174). A main effect between PA and FA was found, and a main effect between connectivity and FA, but no moderating effect has resulted from this analysis (see Figure 4.).
NIX<18
Relation between interest and connectivity. Since interest was not normally
distributed, spearman’s rho was measured to find a correlation between interest and connectivity. No significant correlation was found between interest and connectivity (r = 0.08, p = .332). Spearman’s rho is positive so it implies that a higher degree of interest leads to a higher amount of connectivity. But this is not a significant effect so our hypothesis was not confirmed (see Figure 3).
Figure 4. The interaction effect of Potential Ambivalence and Connectivity on Felt Ambivalence.
13
Effects of Potential Ambivalence and Connectivity on Felt Ambivalence. All relevant
assumptions were met. As with the previous topics, a significant effect of PA on FA was found (b = 0.68, t = 5.66, p < .001). This is a positive effect: the more PA is experienced; the more FA will be experienced. This effect is not found between connectivity and FA (b = 0.002, t = 0.32, p = .747). This model predicts FA for 20%, and declares a significant proportion of the variance (F = 16.75, df = (2, 134), p < .001). PA has the most predictable value in this model (beta = 0.44).
Like with the other two topics, the second model does not have a significant
improvement in difference in explained variance (R square change = 0.01, F change = 1.49,
df = (1, 133), p = .224). The hypothesis was not confirmed: a main effect of PA on FA was
found, but there was no moderating effect of connectivity (see Figure 6.).
Figure 5. The Effect of Interest on Connectivity in a Causal Attitude Network
14
Exploring Analysis1
Effect of Interest and Connectivity on Felt Ambivalence. As an exploring analysis, a
multiple regression was computed to analyze the potential effect of interest and connectivity on FA. In the topic of donor registration, a negative effect was found from interest on FA (b = -0.14, t = -2.17, p = .032). This implies that more interest in the topic leads to less FA. This effect is not found between connectivity and FA (b = 0.01, t = 1.37, p = .173). This model predicts FA for 4,6% and declares a significant proportion of the variance (F = 3.22, df = (2, 134), p = .043). Interest has the most predictable value in this model (beta = -0.18). But no interaction effect was found between interest and connectivity on FA. For the other two topics, no main effects, nor interaction effects were found between interest and connectivity on FA.
Connectivity and Potential Ambivalence. During these analyses, another possible
influence was conceived. The main analysis did not show a moderating effect of connectivity on FA. Even though there were no preconceived expectations, to explore the data further, Pearson’s correlation between connectivity and PA was measured for all three topics. In the topic about donor registration, a positive relation resulted (Pearson’s correlation = 0.21, p = .016). This means an increase of connectivity in the attitude network can lead to more PA. This predicting value was not found for the other two topics of abortion (Pearson’s
correlation = 0.09, p = .299) and alcohol limit (Pearson’s correlation = 0.15, p = .074). 2
1 The data was further explored by computing the regression on a more selected database. This time
very slow and very quick participants were excluded: only surveys finished between 10-60 minutes were analyzed (N=119). All results remained the same.
15
Discussion
This study examined the effect of interest on connectivity in a causal attitude network, and the moderating effect of connectivity in a causal attitude network on FA. The
expectations of the effect of interest on connectivity was confirmed for one of the topics, but this confirmation was missing for the other topics. There could be a mild impact of interest on connectivity, but this effect is not robust. The expectations of the moderating effect of
connectivity in a causal attitude network on FA was again found for one of the topics,
however, this was not the positive effect we expected but it turned out to be a negative effect. For the other topics, no relation between connectivity and FA was found. Even when very slow and very quick participants were excluded from the database, results did not change. In further analysis, for one of the topics a relation was found between interest and FA and for one of the topics an effect was found of connectivity on PA. However, those effects were not found for any of the other two topics.
It was confirmed that a higher amount of interest is more likely to lead to a more connected network. But since this was the result for only one of the topics, this is still
considered as an unconfirmed hypothesis. Furthermore, unconfirmed is the direct and indirect effect of connectivity on FA. For all the analyses, mild effects were found, but no hypothesis was confirmed. This could imply that amount of interest in a topic does not affect the
connectivity in an attitude network, and that the connectivity in an attitude network has no effect on ambivalent attitudes. It may also imply that there have been some flaws in this study which will be discussed next.
The power analysis computed at the start of this study showed that 156 participants were needed to indicate a moderate effect in this study (alpha = 0.05, effect size = 0.33, Power = 0.8, Partial η2
= 0.1). Due to time constraints, data collection stopped at 137 participants. This lack of sufficient participants could have affected the data with a type 1 error. A strong effect is not found, but it could be that a moderate effect does exists but was not found. Next time, pursue data collection until all needed participants have filled in the survey could result in the expected results.
Next to this, FA had no high scores in any of the topics. To find the expected effects of any variable on FA, it is necessary in this survey that the topics are found ambivalent and that the corresponding mixed emotions are experienced. It could be that participants do not have ambivalent attitudes towards the topics, but it could as well be that the questions measuring
16
FA were too fierce. One of the questions referring to FA asked participants about torn feelings in a charged fashion. This could have led to the idea of choosing wrong topics, while instead people chose to not agree with the composition since they are not in the depressed state whilst thinking about the topic, which the composition does propose. In future research, the topics should be revisited and the questions concerning FA could be weakened to make it more relatable for participants, leading to more reliable data.
Furthermore, participants indicated that the questions concerning connectivity were quite hard. Despite the detailed explanation, it was not clear to everybody whether they had to indicate if the two evaluative reactions relate in any sense, or if the two evaluative reactions relate in their opinion and in their case. And the question remains, if participants do
understand the question, are they able to say whether two evaluative reactions influence each other, or might it be an unconscious process. Next to this, participants thought too many questions implied the same thing, which might have led to boredom. Previously mentioned research conducted by Krosnick, Boninger, Chuang, Berent, & Carnot (1993) and Krosnick (1988) did give support to expect an effect of interest on connectivity. Information that is found interesting is thought of more often and attention is high which implies that the network works more conveniently than for information that is found less interesting. Maybe the
adjusted PCR method used in the current study failed to measure connectivity in the right way. This could also explain why the expected moderating effect of connectivity on
ambivalence was not confirmed. In future research concerning connectivity in a CAN model, this could be resolved by conducting the survey in a lab setting. This way questions can be asked at any point of the survey if something is not understood. If it turns out that the relation between two evaluative reactions is an unconscious process, the lab setting will not solve this, in this case a new measurement should be developed. If so, a new measurement could be developed which could measure physical changes as heart rate and perspiration increase, which might result from the torn feelings of FA. This is only the beginning and it requires new research to build upon.
For now, the conclusion can be made that the measurement instrument used in this study did not work in the intended way. This study does provide new insights for future research. The PCR method might not be the way to measure connectivity in the way we used it, it could be adjusted for a better fit. It might even be better to use a different measurement instrument in a lab setting. Based on this study, new leads are explored and it brings us one step further in understanding the application of the CAN model.
17
References
Borsboom, D., & Cramer, A. O. J. (2013). Network analysis: An integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9, 91-121. Dalege, J., Borsboom, D., van Harreveld, F., van den Berg, H., Conner, M., & van der Maas,
H. L. (2016). Toward a formalized account of attitudes. The Causal Attitude Network (CAN) model. Psychological review, 123, 2-22.
Dalege, J., Borsboom, D., van Harreveld, F., van der Maas, H. L. J. (2016a). Network analysis
on attitudes: A brief tutorial. Manuscript submitted for publication.
Dalege. J., Borsboom, D., van Harreveld, F., van der Maas, H. L. J. (2016b). Network
perspective on attitude strength: Testing the connectivity hypothesis. Manuscript
submitted for publication.
Dalege, J., Borsboom, D., van Harreveld, F., van der Maas, H. L. J. (2016c). Network
structure explains the impact of attitudes on voting decisions. Manuscript submitted
for publication.
Frewen, P. A., Allen, S. A., Lanius, R. A., & Neufeld, R. W. J. (2012). Measuring perceived symptom causal relations: Examination with posttraumatic stress. Depression and
Anxiety Symptomatology Assessment, 19, 480-493.
Frewen, P. A., Schmittmann, V. D., Bringmann, L. F., & Borsboom, D. (2013). Perceived causal relations between anxiety, posttraumatic stress and depression: Extension to moderation, mediation, and network analysis. European Journal of
Psychotraumatology, 4:20656.
Van Harreveld, F., van der Pligt, J., & de Liver, Y. N. (2009). The agony of ambivalence and ways to resolve it: Introducing the MAID model. Personality and Social Psychology
Review, 13, 45-61.
Van Harreveld, F., Rutjens, B. T., Rotteveel, M., Nordgren, L. F., & van der Pligt, J. (2009). Ambivalence and decisional conflict as a cause of psychological discomfort: Feeling tense before jumping off the fence. Journal of Experimental Social Psychology, 45, 167-173.
18
Krosnick, J. A. (1988). The role of attitude importance in social evaluation: A study of policy preferences, presidential candidate evaluations, and voting behavior. Journal of
Personality and Social Psychology, 55, 196-210.
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, 1132-1151.
De Liver, Y., van der Pligt, J., & Wigboldus, D. (2007). Positive and negative associations underlying ambivalent attitudes. Journal of Experimental Social Psychology, 43, 319-326.
Newby-Clark, I. R., McGregor, I., & Zanna, M. P. (2002). Thinking and caring about cognitive inconsistency: When and for whom does attitudinal ambivalence feel uncomfortable? Journal of Personality and Social Psychology, 82, 158-166.
Refling, E. J., Calnan, C. M., Fabrigar, L. R., MacDonald, T. K., Johnson, V. C., & Smith, S. M. (2013). To partition or not to partition evaluative judgments comparing measures of structural ambivalence. Social Psychological and Personality Science, 4, 387-394.
