Shaping voters? The effect of Voting Advice Applications on political knowledge and vote choice. Naomi Kamoen, Utrecht University/Tilburg University

Transcription

1 Shaping voters? The effect of Voting Advice Applications on political knowledge and vote choice Naomi Kamoen, Utrecht University/Tilburg University Bregje Holleman, Utrecht University André Krouwel, VU University Amsterdam Jasper van de Pol, University of Amsterdam Claes de Vreese, University of Amsterdam 1

2 ABSTRACT (174 words) In many Western democracies, a substantial part of the electorate experiences difficulty in deciding which party to vote for. It is in these countries that we see the rise of online Voting Advice Applications (VAAs). The current research investigated how these applications impacted political knowledge and vote choice of their users, and whether these effects were particularly large for specific groups of VAA users. To examine this, VAA users subjective perceptions were measured during the 2012 Dutch national elections. In addition, the effect of VAAs on vote choice was also measured with objective data. Results show that demographic characteristics of VAA users play only a minor role in explaining the effects of VAAs on knowledge and vote choice. Instead, respondent dispositions, such as political efficacy, and situational characteristics, such as the prime reason for VAA usage, are of much more importance. In addition, results show that there are discrepancies between subjective and objective measures of the effect on vote choice. This suggests that VAAs shape some types of VAA users more consciously than others. 2

3 1. INTRODUCTION The erosion of traditional social class structures since the 1970s has led to dramatic changes in Western societies. One major political consequence has been the reduced influence of class, religion and other such structural cleavages on vote choice (e.g. Franklin, Mackie & Valen 1992; Garzia, 2010). Instead of ideological class voting, voters rely more and more on issue voting nowadays, which means that they base their voting decision on the perceived proximity between their own issue positions and those of political parties (Downs 1957). This voting strategy assumes that voters have party stances of their own, as well as information about the stances of political parties (Walgrave, Van Aelst & Nuytemans 2008). However, in political landscapes with increasingly large numbers of parties competing for one s vote it is hard to become informed (Marshall & Garzia 2012). This is especially true since increasing numbers of new parties enter into the elections (Krouwel & Lucardie 2008), and as some of the new and traditional parties display more issue volatility than the average voter. Furthermore, due to the growing the distance between citizens and politics, many voters simply do not feel like spending their costly time on politics (Dumont & Kies 2012). That is why making a vote choice has become a difficult for a substantial part of the electorate (Dalton, 2000; Van der Brug & Pellikaan 2003), and superficial cues rather than issue content guide voting decisions. Voting Advice Applications (VAAs) attempt to bridge the gap between citizens and politics by providing political information in an accessible, time-efficient and user-friendly manner (Garzia 2010). VAAs are online tools containing party s issue positions on a number of topics relevant in a certain election. When users visit a VAA, they indicate their opinions towards these issues. Based on the match between the party s issue positions and the users opinions, a VAA produces a personalized voting advice. This can be done implicitly, by depicting the user in the political landscape, or explicitly, by showing a ranked list of those parties matching best. Supplying the user with an implicit or explicit voting advice is not the only goal VAAs strive for. VAAs also aim to increase their users knowledge about political 3

4 parties and their issue positions. This increase in information should then, in turn, have a positive effect on the likelihood of users going to the ballots to cast a vote (Ladner & Pianzola, 2010). To date, VAAs reach between 10 and 40 percent of the electorate in many European countries (Garzia & Marschall 2012). These usage figures indicate that VAAs may have a crucial impact on the political landscape. Therefore, research into VAAs is of utmost importance: can VAAs shape the knowledge and vote choices of their users? The goal of the current research is to answer these research questions in the context of the national Dutch elections of Drawing upon a unique dataset gathered in collaboration with one of the largest national VAAs, we examine whether VAAs affected their users perceived political knowledge and vote choice, and whether these effects are particularly large for specific groups of VAA users. All of this is investigated in different samples, and at various points in time, that is, immediately after VAA usage, and after election day. In this way, we are not only able to reflect on the effects of VAAs, but also on some of the measurement challenges VAA research is facing. In addition, the current study examines the effects of the VAAs on vote choice not only with self-perception measures, but also with objective data. This allows for a comparison between these measurement methods (see also Walgrave et al. 2008). 1.1 The effect of VAA usage on political knowledge The effects of VAAs on political knowledge have been investigated in several previous studies. In these studies, VAA users are generally asked to indicate in a post-vaa survey to what extent they feel that the VAA had increased their political knowledge either directly, or indirectly by functioning as a steppingstone to look for more information. For example, in a study by Ladner (2012), 84% of the respondents indicated that the VAA Smartvote improved their knowledge about the elections. In addition, about 50% of the respondents stated that Smartvote motivated them to look for information about specific candidates or issues, and over 65% reported that Smartvote encouraged them to discuss political issues with others. While comparable results were obtained in Germany (Marschall 2005), an Italian study on 4

5 cabina-elettorale.it were reported the effects of VAAs to be more moderate: only 21% of the respondents felt encouraged to search for information, and 41% indicated that they were motivated to discuss the results of the test with others (De Rosa 2010). Clearly, these studies show that VAA users experience a knowledge increase because of VAA usage. The reported size of the effect, however, varies between studies. This is probably due to various reasons, such as the focus on different countries with different voting systems (Ladner 2012), the low reliability of self-perception measures (Walgrave et al. 2008), the use of different question wordings (Garzia & Marschall 2012), and the timing of question asking (post-vaa or post-election). In addition, concerns have been expressed about limited generalizability of the results of perceptual studies, because of their reliance on nonrandom samples (e.g. Walgrave et al. 2008). To reflect on the extent to which the timing of question asking and the use of nonrandom sample are indeed a cause for fluctuating results, the effect of VAAs on political knowledge will be measured at different points in time and in different nonrandom samples in the current study. The main aim of our research, however, is to reflect on the extent to which particular groups of VAA users experience a relatively larger knowledge increase. The few previous studies investigating this issue have primarily focused on the difference in knowledge increase for younger and older VAA users. Younger voters are expected to experience a larger knowledge increase, because they are relatively least informed about politics and therefore their potential to learn is relatively large (Dumont & Kies 2012). And indeed several studies show that the knowledge increase is particularly large for the young (e.g. Ladner, Fivaz & Nadig 2009). In the current research we will go beyond the effects of age and other socio-demographic variables, and we will test four key hypotheses related to respondent dispositions and contextual characteristics of VAA users. The first of four factors we will analyze is Need for Cognition. This respondent characteristic has been defined as the tendency for an individual to engage in and enjoy thinking (Cacioppo & Petty 1982: 116). Research has shown that the Need for Cognition influences several behaviors; for example, individuals with a high Need for Cognition report a 5

6 more frequent consumption of news (Hallahan 2009). In a political context a high Need for Cognition has been associated with more interaction about politics, and more content-oriented usage of political media (Condra 1992). Therefore, it is to be expected that VAA users are in general associated with higher levels of Need for Cognition. In addition, it may be expected that VAA users with a high Need for Cognition will elaborate more on the VAA results. That is why we hypothesize that they will experience a larger increase in political knowledge than VAA users with a low Need for Cognition (H1.1). Second, the knowledge increase due to VAA usage may depend on an individual s political efficacy. This concept relates to the feeling of an individual to be able to influence the political process (Kaid, McKinney & Tedesco 2007). Previous research (e.g. Finkel 1985) has found that efficacy consists of two components: internal efficacy, which is defined as beliefs about one s own competence to understand, and to participate effectively in, politics and external efficacy, which is defined as beliefs about the responsiveness of governmental authorities and institutions to citizens demands (Niemi, Craig, & Mattei 1991: 1407). We will measure both internal and external efficacy in the current study, but we expect the internal dimension to be most important for our purposes. We expect individuals who have a low internal efficacy, and hence, consider their political competence to be low, to learn more from VAAs than individuals who feel quite competent already (H1.2). Third, we expect an individual s certainty of the vote choice to play a role. Research shows that respondents turn to a VAA with varying levels of certainty about their vote choice; some respondents do not know at all whom to vote for, whereas others already have clear preferences (Rosema & Ruusuvirta 2009). In general, uncertainty has found to be related to a lack of understanding politics (Popkin & Dimock, 1999). Therefore, we expect individuals who are quite uncertain about their vote choice to experience a larger increase in political knowledge due to VAA usage; their potential to learn from VAAs is probably larger than that of individuals who are certain about their vote choice already (H1.3). Fourth, the prime reason for turning to a VAA is expected to be a relevant characteristic for predicting the knowledge increase due to VAA usage. It has been argued 6

7 that VAA users probably have a variety of different reasons for turning to a VAA (e.g. Dumont & Kies 2012; Ruusuvirta Rosema 2009): some users may employ a VAA to learn something about politics, others may use a VAA just as a fun test. Based on the uses and gratifications approach (e.g. Farnsworth & Owen 2004; Rubin 1994), it is to be expected that the intended reason for the usage of an internet tool determines what one gets out of it. We therefore expect that users who regard the VAA to be just a fun test, will probably experience a smaller increase in political knowledge as compared to other VAA users who consider the tool to be a serious advice instrument (H1.4). 1.2 The effect of VAAs on the vote choice As VAAs provide users with knowledge about the political landscape and an advice about whom to vote for, VAAs may affect vote choice too (see also Ladner, Fivaz & Pianzola 2012). Self-perception studies indeed show that VAAs influence vote choices. For example, in a study by Ladner (2012), no less than 60% of the respondents reported that the candidatebased VAA Smartvote influenced their voting decision. In other countries with party based VAAs the self-reported effect is more moderate: in the Netherlands, between 10% (Kleinnijenhuis et al. 2007) and 15% (Aarts & Van der Kolk 2007) of the respondents indicated that they were affected, whereas in Germany and Finland these percentages lay around 6% (Marschall 2005) and 3% (Mykkänen & Moring 2006). Taken together, this shows that VAAs have a substantial impact on vote choice, but also that the exact size of the effect varies between studies. To remedy some of the measurement problems that may cause these differences between perceptual studies (see the previous section), objective measurements of the effects of VAAs on vote choice have also been applied. While these studies are hard to compare mutually due to different operationalizations for what counts as an effect on vote choice, they all clearly show that VAAs have an effect on the vote choice of their users (e.g. see Rosema & Ruusuvirta 2009; Walgrave et al. 2008; Wall, Krouwel & Vitiello 2012). 7

8 While several studies have examined whether there is a general effect of VAAs on vote choice, only few studies have focused on the specific groups of VAA users for whom the effect is especially large. These scarce studies have mainly focused on socio-demographic variables, showing that the voting advice has a relatively large effect on women (Ladner 2012; Mykkänen & Moring 2006), individuals with a lower educational level (Mykkänen & Moring 2006) and younger VAA users (e.g. Ladner 2012). In the current research we will go beyond these demographic effects and test four hypotheses related those subgroups introduced in the previous section. First, we expect individuals with a low Need for Cognition to display a larger effect on their vote choice than individuals with a high Need for Cognition (H2.1). While respondents with a high Need for Cognition probably think about the results of a VAA more thoroughly, and therefore, learn more from VAAs (H1.1), they are probably also more critical towards each piece of incoming information (Condra 1992). That is why they are less likely than individuals with a low Need for Cognition to adopt the advice given. Based on similar reasoning, we also expect individuals with a high rather than a low political efficacy (H2.2), and individuals who are certain rather than uncertain about their vote choice (H2.3) to be affected to a smaller extent. Just like individuals with a high Need for Cognition, these groups of individuals probably hold strong political attitudes and therefore they are less likely to alter their vote choice in response to new information (Zaller 1992). Finally, we expect the prime reason for turning to a VAA to play an important role. Based on the uses and gratifications approach (e.g. Farnsworth & Owen 2004; Rubin 1994), we expect that users who consider the VAA to be a serious advice instrument vote for the party recommended by the VAA more often than VAA users who consider the VAA to be just a fun test. 2. METHOD 2.1 Data collection and materials 8

9 For this study, we draw on unique data gathered in collaboration with one of the two largest VAAs in the Netherlands: Kieskompas. During 2012 Dutch national elections, KiesKompas attracted approximately 1.2 million visitors. About of them completed the vote test and received a voting advice. For each user who completed KiesKompas in the month prior to the 2012 elections (August 15 th September 12 th ) 1, we created a log file. In this log file, we registered the answers to the VAA statements and the voting advice given by the application. In addition, we registered answers to questions about demographic variables, vote certainty, and the voting likelihood for each party (the Propensity To Vote or PTV; see Van der Eijk, Van der Brug, Kroh & Franklin 2006) that VAA users could optionally answer before receiving the voting advice (see Appendix 1; N = about for each question). After users had received their implicit voting advice - their positioning in the political landscape - they were invited to participate in a survey about the use of the VAA. This post- VAA survey, among other things, contained questions measuring the subjective effect of VAAs on the respondents political knowledge and vote choice, as well as question on their Need for Cognition, internal efficacy, external efficacy, and main reason for VAA usage. Hence, we were able to measure a large number of theoretically interesting respondent variables in this set-up. Due to space limitations in the pop-up each concept was measured with only one question (see Appendix 2; N = about for each of these questions). The answers to all the questions in the post-vaa survey were added to the respondents log files. The last question of the post-vaa survey asked whether respondents were willing to contribute to additional research. The group of respondents who agreed (N = ), was sent an with an additional survey on September 13 th, the day after election day. This survey again addressed the effect of the VAA on the users political knowledge and vote choice. In addition, respondents in the post-elections survey were also asked to indicate whom 1 We did not create a log file if the Kieskompas user was under 18 at election day, when the user answered all 30 VAA statements within one minute, when the user exposed straightlining on the VAA statements, when the user did not answer the VAA statements at all, and when the VAA user indicated he or she was 100 years of age or older. 9

10 they actually voted for in the elections (N = about for each of the post-election questions). The answers to the questions in the post-election survey were also added to the respondents log files. Hence, the log files contain unique data reflecting answers to many relevant questions measured at different points in election time. 2.2 Respondents The respondents in the post-vaa sample and the respondents in the post-election sample are not representative for the KiesKompas population at large, due to non-response. As compared to the entire population of KiesKompas users, the respondents in the post-vaa survey, and respondents in the post-election survey even more so, are relatively often male, older, higher educated and more certain about whom to vote for in advance (see Appendix 3). The same biases apply when we compare our samples to the population of Dutch voters (also see Appendix 3). Nevertheless, we decided not to weigh our data, because of the disadvantages of weighting (Brick & Kalton 1996). The main strength of our different samples is that they are relatively large and that they contain a relatively large variety of respondents. This allows us to study groups of VAA users that are less easily accessible through traditional polling, such as younger voters and lower educated voters. 2.3 Analysis The effect of VAAs on political knowledge and on vote choice was assessed in three steps. First, for both dependent variables, three types of descriptive statistics are provided: one reflecting the answers to the perceptual questions for all respondents in the post-vaa survey, one reflecting the answers to the same questions but now asked in the post-election survey, and one reflecting the answers to the questions of interest in the post-vaa survey, but only for those respondents who also answered the same question in the post-election survey as well. Based on these descriptive statistics, we reflect on the self-reported effect of VAAs on 10

11 political knowledge and vote choice as well as on the bias introduced by smaller samples and different times of administration. Second, the perceived effects of VAAs on knowledge and vote choice are predicted in three regression models. In the first model, we only enter three demographic controls into the equation: age, gender and educational level. In the second model, we also add four personality characteristics: Need for Cognition, internal political efficacy, external political efficacy, and vote certainty. In the third model, we explore the effect of the prime reason to use the VAA. 2 Third, in a binary logistic regression we estimate whether (1) or not (0) the respondent voted for the recommended party. This variable is constructed based on the match between the actual voting advice given by the VAA and the party the respondent reported to have voted for in the actual elections. In estimating whether or not the respondent voted for the party recommended, we include the same predictor variables as in the earlier described models. In addition to those factors, we also included the PTV of the party the respondent voted for. In this way, our analysis provides insight into what kind of respondents are more likely to vote for the party recommended by the VAA, while controlling for the likelihood the respondent would already vote for the recommended party. It is important to control for the PTV, as both Wall et al. (2012) and Ruusuvirta and Rosema (2009) show that the effect of VAAs depends on the correspondence of the advice with the VAA user s preexisting preferences. 3. RESULTS 3.1 Political knowledge Descriptive statistics. Immediately after receiving a voting advice, about 35% of the respondents indicates that the VAA has increased their insight into the political parties issue positions (see Appendix 4). These results apply both to the sample as a whole, as well as to 2 We also ran regression models specifying not only all main effects, but also several interactions which the literature suggest might be of importance. In most cases, the interaction terms did not reach statistical significance, and adding them only slightly improved the predictive value the model. That is why we decided not to report the models including interaction terms. 11

12 only those respondents who also answered this same question in the post-election survey. 3 When respondents are asked to judge the impact of the VAA after the elections, this perceived effect on their political knowledge has grown. In the post-election survey, about 56% of the respondents reported an increase in political knowledge due to the use of the VAA (t = 20.35; df = 6985; p < 0.001; Cohen s d = 0.24). Hence, results show that the perceived effect of VAAs varies more between time points than between different nonrandom samples. The knowledge effect for different groups of VAA users. To estimate which kind of respondents perceive most benefits from VAA usage, we ran three linear regression analyses predicting the self-reported increase in political knowledge. Due to largely converging results 4 and space limitations, we only report about the results of the post-vaa survey consisting of respondents who also participated in the post-election survey. Because of the large sample size, we test our hypotheses against an alpha-level of As can be read from Table 1, demographic variables play only a minor role in predicting the effect of VAA usage on knowledge; together the demographic variables entered into the model explain 3.1% of the variance in scores. The only demographic variable that significantly contributed to the model is age: younger VAA users experience a larger increase in political knowledge than medium aged or than older VAA users. The respondent dispositions in model 2 explain even less variance than the demographic characteristics (1.9%). This explained variance can largely be attributed to effects of Need for Cognition and of vote certainty. In line our hypotheses, individuals who have a higher Need for Cognition, learn more from VAAs than respondents with a lower Need for Cognition. Furthermore, individuals who doubt about whom to vote for before 3 The mean scores to the question about political knowledge differ significantly, but caused by the large sample size, as the effect size is very small: t = 3.14; df = 6985; p = 0.002, Cohen s d = All effects found for the smaller post-vaa sample, also reached significance in the entire post-vaa sample. In the entire post-vaa sample, even more predictor variables made a significant contribution to the regression models. As these effects were often irrelevant in terms of effect size, we decided not to report these models here. More interesting is the comparison between the smaller post-vaa sample and the post-election sample, because these samples include the same respondents at different points in time. In the post-election sample the exact same subgroups reported a large knowledge increase; however, the effect of Need for Cognition was only marginally significant (p < 0.05). 12

13 starting the VAA, learn more from VAAs than those individuals who already know whom to vote for, but also as compared to those individuals who do not know at all whom to vote for. Table 1: Parameter estimates of the self-perceived issue position on political knowledge Model 1: Model 2: Model 3: Intercept (0.055) (0.010) (0.109) Gender: Female a (0.027) (0.028) (0.027) Age: b (0.035)** (0.035)** (0.033)** Age: 55+ b (0.034)** (0.034)** (0.033)** Educational level: medium c (0.050) (0.050) (0.047) Educational level: high c (0.047) (0.047) (0.045) Need for Cognition (0.018)** (0.017)** Internal political efficacy (0.016) (0.013) External political efficacy (0.015) (0.015) Certainty: doubts between parties d (0.041)* (0.041)* Certainty: certain about choice d (0.043) (0.044) Reason: gaining insight e (0.037)** Reason: checking initial idea e (0.028)** Reason: deciding content vote e (0.048)** N R R 2 change Note. Standard error between parentheses. a Reference category: males b Reference category: c Reference category: low educational level d Reference category: does not know whom to vote for e Reference category: fun test * p < 0.01 ** p < The only situational characteristic entered into the model is the prime reason for VAA usage. As can be read from Table 1, this variable is the most important predictor of knowledge increase; by itself this predictor explains about 8% of the variance in the knowledge scores. Results show that VAA users who regard the VAA to be just a fun test, learn much less from 13

14 their VAA usage than VAA users who turn to a VAA to gain knowledge, to check their initial preferences, or to decide about which party to vote for. In sum, in line with our hypotheses, our results show a larger knowledge increase for individuals with a high Need for Cognition (H1.1), individuals who are uncertain about whom to vote for (H1.3), and individuals who consider the VAA to be a serious advice instrument (H1.4). Our results provide no support for an effect of political efficacy (H 1.2). 3.2 Vote choice Descriptive statistics. In our surveys, we asked respondents to indicate what role their VAA usage played in their vote choice. Respondents could choose one from four answers: Kieskompas made me doubt about which party to vote for; Kieskompas did not affect my vote choice; Kieskompas has confirmed my party preferences; KiesKompas has determined my vote choice. In both our larger and smaller post-vaa sample, we found that most respondents report not to be affected by the VAA advice (about 52%) or to have received a confirming advice (about 34%; see Appendix 4). In the post-vaa survey, however, we find a smaller percentage of respondents reporting not to be affected (41%), and we find much more respondents stating the voting advice was in line with their initial preferences (47%). Hence, these results again suggest that the differences between different nonrandom samples are negligible as compared to the differences due to the timing of question asking; after some time has passed VAA users perceive the VAA advice to be more in line with their initial vote preferences, and they find the VAA advice to have a smaller effect on their vote choice. The perceived effect of VAAs on vote choice for different groups of VAA users. To estimate which kind of respondents perceive most effect of the VAA on their vote choice, we recoded our subjective measure of the effect on vote choice into a binary variable. The answering categories Kieskompas made me doubt about which party to vote for and KiesKompas has determined my vote choice were taken to indicate an effect on vote choice (code 1), whereas the answering categories Kieskompas did not affect my vote choice and Kieskompas has confirmed my party preferences were coded to reflect no effect (code 0). 14

15 This binary variable was predicted in three regression analyses. Table 2 shows the results for the post-vaa survey for those respondents who also answered this question in the postelection survey. Results for the other samples were comparable. 5 Table 2: Parameter estimates (in Logits, standard error and percentages between brackets) of whether (1) or not (0) the respondent perceived an effect of VAA on vote choice Model 1 Model 2 Model 3 Intercept (0.170; 15.4%) (0.346; 43.6%) (0.374; 19.7%) Gender: Female a (0.083)** (0.090) (0.092) Age: b (0.110) (0.116) (0.118) Age: 55+ b (0.110) (0.116) (0.119) Educational level: medium c (0.155) (0.163) (0.166) Educational level: high c (0.146) (0.155) (0.157) Need for Cognition (0.059) (0.060) Internal political efficacy (0.050)** (0.051)** External political efficacy (0.048) (0.049) Certainty: doubts between parties d (0.106)** (0.113)** Certainty: certain about choice d (0.164)** (0.173)** Reason: gaining insight e (0.150)** Reason: checking initial idea e (0.129)** Reason: deciding content vote e (0.139)** N Nagelkerke R Change in Nagelkerke R Note. Standard error and percentage between parentheses. a Reference category: males b Reference category: c Reference category: low d Reference category: does not know whom to vote for e Reference category: fun test * p < 0.01 ** p < Exactly the same predictor variables made a significant contribution to predicting knowledge scores in the post-election sample. In the entire post-vaa sample, more predictor variables reached significance, but the sizes of these effects were very small. 15

16 As Table 2 shows, 19.7% of the respondents in the reference category (males, who are young, lower educated, associated with a low need for cognition, etcetera) indicated that the VAA determined their vote choice or made them hesitant about whom to vote for. As demographic variables do not make a significant contribution to predicting vote choice, we may actually apply this percentage to all individuals of all genders, age groups, and levels of education. The respondent dispositions added in model 2, explain a large part of the total variance in scores (about 16%). The effects of internal efficacy and vote certainty are of particular importance, as these variables significantly contribute to the regression model. The higher an individual s internal efficacy, the less likely he is to indicate that KiesKompas influenced his vote. We may get an impression of the size of this effect by translating the regression weights in Table 2 to percentages. When all other parameters are set to the default, individuals who report to have a low efficacy score report to be affected by the advice in 19.7% of the cases, versus 6.9% of the respondents with the highest efficacy scores. Results also show that the more certain a VAA user is about his vote choice before turning to the VAA, the less likely he or she is to perceive the voting advice to have an effect. When all other parameters are set to the default, individuals who report to be uncertain about their vote choice, are affected more (19.7%) than individuals who doubt between only parties (14.7%) or those who are quite certain in advance already (2.8%). Finally, results show that prime reason for VAA usage added in model 3 is an important predictor. By itself, it explains 4.6% of the variance in scores. VAA users who consider the VAA to be just a fun test, are not affected so much as respondents who turn to the VAA for another reason. When all other parameters in the model are set to the default, the percentage of VAA users that reports to be influenced runs up to 47.9% for respondents who consult the VAA to determine whom to vote for. In sum, in line with our hypotheses, the perceived effect of VAA usage on vote choice can be said to be larger in general for individuals with a low internal efficacy (H 2.2), for individuals who are uncertain in advance about whom to vote for (H 2.3), and for 16

17 individuals who consider the VAA to be more than just a fun test (H 2.4). We found no evidence for an effect of Need for Cognition (H 2.1). The objectively measured effect of VAAs on vote choice for different groups of VAA users. In the post-election survey, we asked respondents to indicate which party they voted for in the actual elections. Based on this information and the actual voting advice given, we constructed a variable indicating whether (1) or not (0) the respondent voted for the party recommended by the VAA. This objective measure for the effect of vote choice was estimated using all predictor variables introduced earlier. In addition, we added the PTV of the advised party as a covariate. This is done to make sure that when a certain subgroup of VAA users shows to vote for the party recommended relatively often, this is not due to that group obtaining an advice that is more in line with initial preferences. Table 3 shows that it is indeed important to control for the PTV of the advised party. This PTV score determines to a large extent whether or not the respondent votes for the party recommended. The higher the PTV for the advised party, the higher the likelihood the respondent will vote for the party recommended. When all parameters in the model are set to the default, the likelihood of voting for the party recommended by the VAA fluctuated between 6% (PTV of 0) and 20.5% (PTV of 10). This suggests that the more serious a party was already contemplated by the VAA user, the higher the likelihood that he or she will vote for this party when it is recommended. When controlled for the PTV, we find that age is the only demographic variable that makes a significant contribution to the model: younger (18-34) and medium aged (35-54) VAA users are more likely to vote for the recommended party than older (55+) voters. When all other parameters are set to the default, the estimated percentage of older VAA users who votes for the party recommended, is as low as 3.8%, whereas the default value of 6% applies to the young and medium aged. 17

18 Table 3: Parameter estimates (in Logits, standard error and percentages between brackets) of whether (1) or not (0) the respondent voted for the party recommended Model 1 Model 2 Model 3 Intercept (0.175; 15.0%)** (0.344; 5.7%)** (0.358; 6.0%)** Gender: Female a (0.081) (0.083) (0.083) Age: b (0.095) (0.096) (0.096) Age: 55+ b (0.097)** (0.098)** (0.098)** Educational level: medium c (0.161) (0.162) (0.162) Educational level: high c (0.149) (0.151) (0.151) Need for Cognition (0.054) (0.054) Internal political efficacy (0.049)* (0.049)* External political efficacy (0.048) (0.048) Certainty: doubts between parties d (0.134) (0.140) Certainty: certain about choice d (0.140) (0.149) Reason: gaining insight e (0.119) Reason: checking initial idea e (0.085) Reason: deciding content vote e (0.130) PTV advised party (0.011)** (0.011)** (0.011)** N Nagelkerke R Change in Nagelkerke R a Reference category: males b Reference category: c Reference category: low d Reference category: does not know whom to vote for e Reference category: fun test * p < 0.01 ** p < Contrary to what we have seen for the subjective measurements, respondent dispositions added in model 2, play only minor role in explaining vote choice (1.4% variance explained). Internal efficacy is the only such predictor that shows an effect, and the effect is in another direction than expected: voters who have a high internal political efficacy are more likely to vote for the party recommended than voters with low efficacy scores. When all other parameters are set to the default, the estimated difference between voters with a low versus a 18

19 high efficacy is about 6.6%. This means that about 12.6% of the individuals with the highest internal efficacy score votes for the party recommended. Finally, results show that the prime reason for VAA usage does not make a significant contribution to explaining the effect on vote choice. Adding this predictor to the model does not increase the percentage of variance explained. In sum, results show that individuals more often vote in correspondence with the VAA advice if they were already seriously considering voting for that party in advance. In addition, when controlled for PTV, we find an effect of internal efficacy, but it is in another direction than expected (H2.1): individuals high efficacy scores are more likely to vote for the party recommended by the VAA. In addition, our data show no support for an effect of Need for Cognition (H.2.1), vote certainty prior to VAA usage (2.3), and prime reason for VAA usage (2.4). When we combine these results based on objective measurements with our subjective measurements, clear discrepancies arise. These will be elaborated on in the conclusion and discussion section. 4. CONCLUSION AND DISCUSSION In election time, millions of voters in West-European countries turn to an online Voting Advice Application (VAA). Therefore, these tools may have a crucial impact on the political landscape. The current research investigated whether VAAs shaped their users political knowledge and vote choice in the 2012 Dutch national elections, and whether these effects were larger for specific groups of VAA users. To investigate these issues, we measured VAA users subjective perceptions of these effects. In addition, the effect of VAAs on vote choice was also examined with objective measurements. Results show that, rather than the demographic characteristics, respondent dispositions and situational characteristics are important for understanding the self-reported effects of VAAs. The only demographic characteristic that showed to have an impact was age: younger VAA users perceive to learn more from VAAs than medium-aged or older VAA 19

20 users (compare Ladner, Fivaz & Nadig 2009). In addition, the effect of VAAs on knowledge appeared to be relatively large for individuals who have a high Need for Cognition, individuals who are uncertain about whom to vote for before using the VAA, and individuals who regard the VAA to be a serious advice instrument. These latter two groups also perceived the effect on their vote choice to be relatively large. In addition, we found that individuals with a low internal efficacy perceive the effect on vote choice to be larger than individuals with higher levels of efficacy. Objective measures of the effect of VAAs on vote choice, however, show a different picture. While individuals with low efficacy scores perceived the effect on vote choice to be large, individuals with high efficacy scores voted for the party recommended by the VAA relatively often. In addition, uncertain VAA users and VAA users who consider the VAA to be a serious advice instrument, do not vote for the recommended party more often. Instead, the young were found to vote for party recommended relatively often. All of these effects occur when controlled for initial vote preferences. An important question raised by these results is why there are such large differences between subjective and objective measurements. Measurement error may play a role, but as our perceptual results showed to be robust for some sources of error (see below), we rather take these results to indicate that some VAA users are more aware of the impact of VAAs on their vote choice than others. If this is true, our study shows that individuals with a high internal efficacy in particular underestimate the impact of a VAA. Internal efficacy reflects an individual s perceived political competence and knowledge (Niemi et al. 1991). It is known from psychological research on attitude change that individuals who actually have high levels of knowledge are less likely to be influenced by new incoming information (Petty & Krosnick 1995). Perhaps the impact of VAAs on highly efficacious individuals is larger than expected because this group overestimates their actual political competence and knowledge about politics. It would be interesting to test this explanation in a future study. Another suggestion for future research is to investigate the effects of VAAs on knowledge and vote choice in other countries with different voting systems. The reason is that 20

21 the current research rests on one specific case, that of the Dutch 2012 national elections. While this is only one case, it is theoretically a very interesting one: the Netherlands has large numbers of floating voters and a relatively long history in VAA usage at the same time (Krouwel 2012). Therefore, we expect to find the same user characteristics to contribute to a better understanding of the effects of VAAs in other countries, but at the same time, we expect the effects of VAAs to be a bit more moderate as compared to the Dutch case. In addition to investigating the same user characteristics in different countries, it is important to investigate the relation between the effects of knowledge and vote choice. In the current research, the effects on knowledge and vote choice were analyzed separately. However, it is to be expected that the effects of VAAs on knowledge and vote choice are related: the effect on vote choice is likely to be especially large for those VAA users who also learn a lot from VAAs. In order to test this prediction, mediated analysis should be performed. A prerequisite is that the amount of variance explained in the mediating dependent variable, i.e. knowledge, is substantial (Lord & Novick 1968). In order to obtain an even more rich understanding of the impact of VAAs on vote choice it is therefore also crucial to investigate which other user characteristics explain the effects of VAAs. A side goal of the current research was to reflect on some of the measurement challenges in VAA research. That is why the perceived effects of VAAs on knowledge and vote choice were measured in two different samples: one large self-selected sample drawn from the population and a second self-selected sample drawn from the first. Results appeared to be quite stable across these nonrandom samples. A second measurement issue of interest in the current research was the timing of asking perceptual questions. The perceived effects of VAAs on knowledge and vote choice were measured at two different points in time, that is, post-vaa and post-election. While the same subgroups of VAA users indicated a relatively large effect at both measurement moments, the average effect on knowledge and vote choice varied greatly over time. Postelection, VAA users judged the knowledge effect to be larger than before. In addition, when 21

22 asked after the elections, VAA users reported the voting advice to have less impact on their vote choices and for it to be more in line with their already existing preferences. This latter result raises the question how these differences due to the timing of question asking can be explained. One possibility is that effects of VAAs really vary over time. This explanation is especially plausible for knowledge: the knowledge increase may become larger over time, because VAAs do not only function as a source of information, but also as a steppingstone to look for additional information (e.g. Ladner 2012). Such an explanation, however, cannot account for why, after some time, more respondents perceived the VAA advice to be in line with their initial preferences. Based on post-hoc rationalizing, we think this finding indicates that some respondents fail to remember their voting advice correctly after a while. Additional analyses of our data indeed show that when asked in the post-election survey, a substantial part of the respondents admits to have forgotten the advice, or remembers the advice incorrectly (see also Wall et al. 2012). It is important that future studies investigate the phenomena of rationalizing choices and forgetting political information further. VAAs provide an excellent context for investigating these broader psychological and political phenomena. 22

23 REFERENCES Aarts, K. & Van der Kolk, H. (2007). The parliamentary election in the Netherlands, Electoral studies 26: Brick, J.M. & Kalton, G. (1996). Handling missing data in survey research. Statistical methods in medical research 5: Cacioppo, J.T., & Petty, R.E. (1982). The need for cognition. Journal of personality and social psychology 42: Cacioppo, J.T., Petty, R.E., & Kao, C.F. (1984). The efficient assessment of need for cognition. Journal of personality assessment 48: Condra, M.B. (1992). Link Between Need for Cognition and Political Interest, Involvement, and Media Use. Psychology: A Journal of Human Behavior 29: Dalton, R.J. (2000). The decline of party identifications. In R. J. Dalton & M. P. Wattenberg (eds.). Parties without partisans: political change in advanced industrial democracies (pp ). Oxford: Oxford university press. De Rosa, R. (2010). Cabina-elettorale.it (provides advice to Italian voters since 2009). In L. Cedroni & D. Garzia (Eds.). Voting advice applications in Europe: The state of the art (pp ). Napoli: Scriptaweb. Downs, A. (1957). An economic theory of democracy. New York: Harper Collins Publishers. Dumont, P., & Kies, R. (2012). Smartvote.lu: usage and impact of the first VAA in Luxembourg. International Journal of Electronic Governance 5:

24 Farnsworth, S.J., & Owen, D. Internet use and the 2000 presidential election. Electoral studies 23: Franklin, M.N., Mackie, T.T., & Valen, H. (1992). Electoral change: responses to evolving social and attitudinal structures in western societies. Cambridge University press: Cambridge. Garzia, D. (2010) The effects of VAAs on users voting behaviour: an overview. In L. Cedroni & D. Garzia (Eds.). Voting advice applications in Europe: The state of the art (pp ). Napoli: Scriptaweb. Garzia, D., & Marschall, S. (2012). Voting Advice Applications under review: the state of research. International Journal of Electronic Governance 5: Hallahan, K. (2009) Need for Cognition as Motivation to Process Publicity and Advertising. Journal of promotion management, 14: Kaid, L.L., McKinney, M.S., & Tedesco, J.C. (2007). Introduction: political information efficacy and young voters. American behavioral scientist, 50: Kaye, B.K., & Johnson, T.J. (2004). Web for all reasons: uses and gratifications of Internet components for political information. Telematics and Informatics 21: Kleinnijenhuis, J., Scholten, O., Van Atteveldt, W., Van Hoof, A., Krouwel, A., Oegema, D., De Ridder, J.A., Ruigrok, N., & Takens, J. (2007). Nederland vijfstromenland. De rol van media en Stemwijzers bij de verkiezingen van Amsterdam: Uitgeverij Bert Bakker. 24

25 Krouwel, A.P.M. (2012). Party transformations in European democracies. Albany: Suny press (state university of New York Press). Krouwel, A., & Lucardie, P. (2008). Waiting in the wings: new parties in the Netherlands. Acta Politica 43: Ladner, A. (2012). Voting Advice Applications Impact on voting decisions in the 2011 Swiss National Elections. Paper presented at the 62 nd Political Studies Association annual International Conference in Belfast 3-5 April Ladner, A., Fivaz, J., & Nadig, G. (2009). Voting assistance applications as tools to increase political participation and improve civic education. In M. Prim and H. Milner (eds.). Civic education and youth political participation (pp ). Sense publishers. Ladner, A., Fivaz, J., & Pianzola, J. (2012). Voting advice applications and party choice: evidence from smartvote users in Switzerland. International Journal of Electronic Governance 5: Ladner, A., & Pianzola, J. (2010) Do voting advice applications have an effect on electoral participation and voter turnout? Evidence from the 2007 Swiss federal elections. In E. Tambouris, A. Macintosh & O. Glassey (eds.). Electronic Participation. Berlin, Heidelberg, New York: Springer: pp Lord, F.M., & Novick, M.R. (1968). Statistical theories of mental test scores. Addison- Wesley: Reading Mass. 25

26 Mykkänen, J., Moring, T. (2006). Dealigned politics comes of age? The effects of online candidate selector son Finnish voters. Paper presented at the conference of politics on the internet: new forms of media for political action, November 25, Tampere. Niemi, R.G., Craig, S.C., & Mattei, F. (1991). Measuring internal political efficacy in the 1988 National Election Study. American Political Science review 85: Petty, R.E., & Krosnick, J.A. (1995). Attitude strength: Antecedents and consequences. Mahwah: Lawrence Erlbaum Associates. Popkin, S.L. & Dimock, M.A. (1999). Citizen competence and democratic institutions. In: S.L. Elkin & K.E. Sołtan (eds.), Citizen competence and democratic institutions. Pennsylvania, The Pennsylvania state university press: Rubin, A.M. (1994). Media uses and effects: a uses-and-gratifications perspective. In: J. Bryant, & D. Zillmann (eds.), Media effects: advances in theory and research. Lawrence Erlbaum Associates: Hillsdale, NJ. Ruusuvirta, O. & Rosema, M. (2009). Do Online Vote Selectors Influence Electoral Participation and the Direction of the Vote? Paper Presented at the ECPR General Conference. Potsdam: September. Van der Brug, W., & Pellikaan, H. (2003). Preface. Acta Politica 43: Van der Eijk, C., Van der Brug, W. & Kroh, M. (2006). Rethinking the dependent variable in voting behavior: on the measurement and analysis of electoral utilities. Electoral Studies 25: