3.3 Methods
3.3.1 Data description
Theo van Gogh
The Dutch data for the second wave of the European Social Survey (ESS) (2012) were collected between 11 September 2004 and 19 February 2005. As a first step, I ordered the observations by date and used 2 November as a cut-off point. To avoid bias due to immigrant background, I removed responses by people that either did not have Dutch citizenship (51 cases) or were not born in the Netherlands (119 cases). In addition, I deleted responses of people who had been interviewed on the day of the murder, as it is unclear whether they had learned about it yet (17 cases). Finally, as mentioned, I limited the sample to 14 days before and after the murder. This way, a group of 282 observations before and 134 observations after the murder remained. Figure A.1 (Appendix) displays the distribution of the number of times participants were approached. The median number of the visit at which the interview took place was 3 for both groups.
Demographics and other relevant details of both groups are reported in Table 3.1. I include statistics for gender, age, education level, main activity (a dummy variable that expresses whether or not respondents are engaged in full-time work or education), political orientation, political interest, the degree of urbanization of the respondent’s living environment, whether or not the respondent lives in Amsterdam, and the amount of time the respondent spends per day reading newspapers. The codebooks for these variables are reported in Table A.1. As shown in the table, the participants in the post-murder sample display somewhat higher levels of political interest (expressed in a lower score on the scale). The two samples do not differ otherwise.
Table 3.1: Demographics for the pre-and post-event group in the 14 days before and after the murder of Theo van Gogh.
Variable Pre-murder Post-murder t df
% Male 44.33 50.75 −1.23 414 Mean age 49.09 47.72 0.75 413 Education level 2.91 3.10 −1.28 414 % Work or education 55.87 62.69 −1.32 413 Political orientation 5.25 5.25 0.02 387 Political interest 2.40 2.21 2.28* 414 Urbanization 3.05 3.21 −1.31 414 % Amsterdam 2.84 2.99 −0.08 414 Newspaper reading 1.58 1.63 −0.35 414 * p < .05, ** p < .01, *** p < .001
As for the heightened level of political interest after the murder, it seems that this is an effect of the event. Although differences on relevant demographics could of course be due to selection bias, it seems relevant to note that additional analyses show that the two groups do not differ on more objective measures of political interest, such as party membership and voter turnout (see Figure A.4). Moreover, the median for both groups is the same (2; quite interested). These results suggest that this increase in political interest is driven by the murder, making the more subjective measure inappropriate for the further analysis of this case. After all, we cannot know whether post-murder respondents who indicate high political interest would have given the same reply before the murder, rendering the comparison of groups with different levels of political interest before and after the murder invalid.
Stockholm bombings
To investigate the impact of the terrorist attack in Sweden, I use the fifth wave of the European Social Survey (ESS; 2012). For Sweden, these data were collected between 27 September 2010 and 1 March 2011 (included). I ordered the observations by date and used 11 December as a cut-off point. To avoid bias due to immigrant background, I removed responses by people that either did not have Swedish citizenship (49 cases) or were not born in Sweden (128 cases). There were no interviews collected on the day of the attack. As before, I limited the analysis to 14 days before and after the attacks. This way, a group of 152 respondents before and 60 respondents after the event remained. The small size of the post-attack group is due to the timing close to the holidays. Extending the bandwidth to three weeks does not lead to a larger post-attack group. Since further extending the bandwidth could imply the potential confounding effect of other events, I use the 14-day bandwidth for my analyses. Figure A.2 displays the distribution of the number of times participants were approached. The median number of the visit at which the interview took place was 3 for both groups.
Statistics for both groups on the same demographics as above are reported in Table 3.2. Again, codebooks are reported in Table A.1. As shown in the table, the participants in the post-attack sample are somewhat more likely to be engaged in full-time work or education, and display somewhat higher levels of political interest (expressed in a lower score on the scale). The two samples do not differ otherwise.
Again, while there is a significant difference in political interest before and after the attack, the median value for both groups is the same. Moreover, a χ2 test does not reveal
significant differences on more objective political variables such as voting behavior and party membership. Therefore, we cannot be sure whether there is a pre-existing difference in political interest between the two groups, or whether this was driven by the events.
Table 3.2: Demographics for the pre-and post-attacks group in the 14 days before and after the Stockholm bombings.
Variable Pre-attack Post-attack t df
% Male 39.47 46.67 −0.96 210 Mean age 43.58 42.9 0.24 210 % Higher educated 76.97 88.33 −1.48 207 % Work or education 73.68 86.67 −2.05 * 210 Political orientation 5.63 5.12 1.44 204 Political interest 2.36 2.07 2.15* 210 Urbanization 2.99 2.73 1.46 210 % Amsterdam 22.37 26.67 26.67 210 Newspaper reading 1.45 1.61 −1.17 210 * p < .05, ** p < .01, *** p < .001 Charlie Hebdo
Responses for the 2014 wave of the ESS (2014) were collected in France between 4 November 2014 and 23 February 2015. Again, participants who did not have French citizenship (97) or were not born in France (129) were dropped. In addition, I dropped respondents who were interviewed on the day of the attack (15). Given that the attacks took place towards the end of the data collection period, and right after the Christmas holidays, a 14-day bandwidth yields only 31 responses in the two weeks before the attacks, and 123 in the two weeks after. Due to the small sample sizes of the 14-day bandwidth, especially in the weeks before the attacks, I consider a bandwidth of 21 days before and 14 days after the attacks. This yields 119 and 123 observations, respectively.
Figure A.3 (Appendix) displays the distribution of the number of times participants were approached. The median number of visits was 6 for both groups, but the figure shows that the distributions are rather different. A χ2 test confirms that this difference is
statistically significant (χ2 = 30.77; p = .00). While Table 3.3 did not reveal significant
differences between the pre- and post-attack group on relevant demographics, the fact that those interviewed after the attack were approached more often may indicate a lower initial willingness to participate – or simply a lower reachability. In either case, this could imply a qualitative difference between the two groups, possibly leading to differences in attitudes toward immigrants.
Demographics for the two groups are reported in Table 3.3. Since this wave of the ESS (2014) did not include a variable measuring newspaper use, I operationalized media use as time spent watching political news on TV. The used scale is the same as for the newspaper use variable used in the previous cases. Codebooks are again displayed in Table A.1. As the table shows, other than their difference in reachability, the two groups do not differ on any of the demographics.
Table 3.3: Demographics for the pre-and post-attacks group in the 21 days before and 14 days after the Charlie Hebdo attacks.
Variable Pre-attacks Post-attacks t df
% Male 47.90 47.15 −0.12 240 Mean age 48.80 46.97 0.76 239 Education level 396.71 408.93 −0.48 239 % Work or education 61.34 68.29 −1.13 240 Political orientation 5.13 5.17 −0.15 223 Political interest 2.47 2.33 1.16 240 Urbanization 2.60 2.45 0.88 239 % Paris 11.67 13.82 −0.48 240 TV watching 4.05 3.80 0.89 240 * p < .05, ** p < .01, *** p < .001 Dependent variables
Similar to Finseraas et al. (2011), I consider the impact of the murder on preferences for immigration policy. I expand their measure (p. 399) by adding a third variable, thereby creating a scale consisting of the following items:
• To what extent do you think [your country] should allow people of the same race or ethnic group as most [your country] people to come and live here?3
• How about people of a different race or ethnic group from most [your country] people?
• How about people from the poorer countries outside Europe?
All items were measured on a four-point scale, ranging from 1 (allow many to come and live here) to 4 (allow none). The items were standardized and added to create one scale. Cronbach’s α was .85 for the Theo van Gogh case, .93 for the Stockholm case, and .88 for the Charlie Hebdo case. The scale was recoded such that a higher score indicates more liberal immigration policy preferences.
In addition, I consider a more cultural outcome. Again, I create a scale, which consists of the following items:
• Would you say it is generally bad or good for [country]’s economy that people come to live here from other countries?
• Would you say that [your country]’s cultural life is generally undermined or enriched by people coming to live here from other countries?
• Is [your country] made a worse or a better place to live by people coming to live here from other countries?
These items were measured on an 11-point scale, where a higher value expressed more positive attitudes. The items were standardized and added to create one scale. Cronbach’s α was .77 for the Theo van Gogh case, .87 for the Stockholm case, and .87 for the Charlie Hebdo case. A higher score on this scale means a more positive attitude toward
immigrants.
In all cases, standardization of the variables took place before limiting the samples to the restricted bandwidths. A factor analysis confirms that the six items included in the analysis load on two different factors, thereby speaking in favor of using two separate scales rather than combining all items into a single measure.