Long Live the Past: A Multiple Correspondence Analysis of People’s Justifications for Drawing Historical Analogies Between the Paris Attacks and Past Events

Comparing present and past situations by means of historical analogy is prevalent in political and public discourses. But when researching this phenomenon, scientists often use reception paradigms, where they ask people which past event is most applicable to a current situation or issue. In these paradigms, analogies are treated as unequivocal—rather than flexible—in their meanings. In this paper, we use a production paradigm to examine why European citizens (in France, Belgium, and Germany) selected historical analogies and justified their meanings following the two 2015 terrorist attacks in France. We find that most participants tend to mention a relatively small number of past events, characterized by similarities in time (recent), space (geographically close) and type (terrorist attacks) with the current attacks. However, a multiple correspondence analysis indicates that, even when they overwhelmingly agree about the relevance of a particular event (the attacks of September 11th 2001) for the present situation, participants confer widely varying—even conflicting—meanings to the “same” analogy, which align with different socio-political attitudes. We suggest that these variations do not just represent the emphasis that different participants place on particular sets of similarities between the past and the present attacks: They also embody specific, and conflicting, stances on salient and controversial issues surrounding the topic of contemporary terrorism (e.g., why were ‘we’ attacked, who deserves to be grieved, how should the government respond). Results are discussed in light of the literature on social representations of both history and terrorism.

the result of the process by which historical events, as social objects, become part of lay common sense and fulfill social and political functions in the present.
Historical analogies can be viewed as one process by which such representations of history are mobilized in the present. Representations of current situations are anchored in the social representations of specific past events in order to give meaning to the present (Gibson, 2012). Indeed, social representations of history involve at their core "explanatory narratives, which often form analogies that guide responses to present situations" (Liu & Hilton, 2005, p. 15;emphasis added). Historical analogies are an instance of more general analogical, or case-based, reasoning, where selected elements of a current situation-the target-are compared to selected elements of one or more past events-the source (Gentner, 1983). This comparison process involves an "inference that if two or more events agree in one respect, then they may also agree in another" (Khong, 1992, pp. 6-7). Thus, making a historical analogy implies the past and present share some meaningful similarities that can be used to make further inferences about the target.
Invoking historical analogies can result in various effects: constructing a specific representation of the current situation, defining the roles of the actors involved and their identities (e.g., who is with "us" or with "them"), planning and making decisions, as well as persuading others of the legitimacy of a given course of action (see Ghilani et al., 2017).
While previous research has shown how important historical events are remembered, constructed and used in the present (for reviews, see Hilton & Liu, 2017;Hirst et al., 2018), they often focus on history and how it retains general relevance in the present. Here, we study the reverse process in focusing on historical analogies per se: How will people faced with new situations in the present invoke representations of specific historical events, which ones, and why. In doing so, we also examine social representations of history but from a different perspective.
We zoom in on the particular process in which representations of past events are actively mobilized to anchor the representations of specific current events; thereby giving them meaning.

Selection of Analogies
In the process of drawing a historical analogy, at least one particular past event must be selected among all other possible alternatives. Accordingly, source selection and retrievability in memory are typically regarded as the earliest steps in analogical reasoning, be it historical (Shimko, 1994) or otherwise (Gentner & Forbus, 2011). In general, (non-historical) analogical reasoning, source selection has been found to rely mostly on the presence of for the past to be of continued relevance to new situations requires its meaning to never be entirely settled. For historical analogies, which mobilize not just particular past events but also a particular interpretation of those events (and of the present), such flexibility is both a source of their meaning making power and a source of argument (Gibson, 2012)-resulting in "metaphor wars" as people argue about the 'appropriate' analogy and its meaning (Paris, 2002). This is congruent with a rhetorical approach to social representations (Billig, 1996;chap. 8). Billig proposes that contrary themes are integral to the fabric of lay common sense. This "dilemmatic nature" is precisely what allows people to argue with one another, even (or especially) when they rely on shared knowledge (see Gibson, 2012;Voelklein & Howarth, 2005). For example, the importance of a historical figure (Robert the Bruce) for Scottish identity turned it into a valuable rhetorical resource recruited by political leaders in support for opposite political projects (Reicher & Hopkins, 2001).
Yet, such interpretative flexibility of history and its consequences has often been disregarded in studies focusing specifically on the use of historical analogies by lay people. Indeed, in both experimental (Gilovich, 1981;Smeekes et al., 2014) and survey studies (Michelson, 1998;Schuman & Corning, 2006;Schuman & Rieger, 1992), respondents are asked to choose from preselected analogies and cannot explain what meaning they give to those analogies. For instance, one survey study conducted during the 1991 Gulf war found that more than fifty percent of Americans who thought the Vietnam analogy applied to the current case expressed support towards US military involvement (Schuman & Rieger, 1992)-against the typical meaning given to this analogy. This paradox suggests that those respondents, if asked, may have justified their analogical choices in different, even contrasting, ways.
Thus, by using mainly reception paradigms (Blanchette & Dunbar, 2000), where participants are exposed to preselected analogies, scholars adopt a research posture that reduces people's uses of historical analogies and their entire political reasoning to a matter of external influence. As a result, people are only portrayed as passive recipients of externally imposed analogies and we end up with a partial understanding of the phenomenon (see Bougher, 2012, for a similar argument about political metaphors). We address this gap by considering people as active sense-makers (Elcheroth, Doise, & Reicher, 2011;Moscovici, 1981), and by using a production instead of a reception paradigm (see Blanchette & Dunbar, 2000), where participants are asked to choose and explain their own analogies.

Present Research
This paper examines the historical analogies produced by three samples of participants in the immediate aftermath of the two 2015 terrorist attacks in France. The data from the first sample were collected among French and French-speaking Belgian participants shortly after the Charlie Hebdo and Hypercacher attacks in January ii (Jan-Moreover, social representational processes (such as historical analogies) are best seen at work when social issues are threatening or controversial (Moscovici, 1981), resulting in their increased prevalence in public, media, and political communications (see also Joffe, 2003). Such a situation was observed in the aftermath of both terrorist attacks in Belgium (Crijns, Cauberghe, & Hudders, 2017), France (Ernst-Vintila & Macovei, 2016) and Germany (Vollmer & Karakayali, 2018). Focusing on citizens in these three countries was therefore relevant to our investigation.

Overview of the Analysis
We asked participants to report in open-ended questions (1) which past event(s), if any, those terrorist attacks reminded them of, and (2) why. Only complete responses were considered in this paper. The answers to both questions were coded to identify the specific sources that they selected, and the comparison categories they used to explain their selection. The analysis proceeded in three steps: (1) We examined and compared the sources selected across all three samples based on their frequency and their similarities to the 2015 terrorist attacks. (2) We conducted a multiple correspondence analysis (MCA; Husson, Lê, & Pagès, 2017) on the comparison categories to explore how the current attacks were related to past ones. Open-ended questions, like the ones we used, produce qualitative data that can often be coded in more than one category. MCA has the advantage of handling non-independent categorical variables (Hwang, Tomiuk, & Takane, 2009 Michelson, 1998;Schuman & Rieger, 1992). However, participants could not generate their own analogies, nor justify them.
In our study, we decided to examine if participants' freely generated analogies would also align with sociopolitical variables. For comparability purposes, we only selected the variables that had been collected in all three samples (see Table Sup1 in the Supplementary Materials for the list of all other collected variables). These variables related to three dimensions: how participants reacted to the attacks (negative emotions, online and offline engagement with the events), how they viewed Muslims especially in the context of terrorism (Islamophobia), and how supportive they were toward the kind of counterterrorist policies implemented both in the past (after the 9/11 attacks in the US) and in the present (after the 2015 attacks in France). The questionnaires, data, code for reproducing the reported analyses, and supporting material are provided online on the Open Science Framework (see Supplementary Materials).
sense of highly emotional and mediatized events. Participants completed the questionnaire on a voluntary basis.
Data collection was started two weeks after each terrorist attack and lasted 1 to 2 months, until participation waned down considerably. The Jan-FR sample was collected between January 22 nd and February 17 th 2015; the Nov-FR and Nov-GE samples were collected between November 25 th 2015 and the end of January 2016.

Samples Characteristics (Total N = 831; 9/11 N = 518)
Within the originally collected samples, this study focused only on participants who answered "yes" to the question "Do the [current attacks] remind you of a past event?" and also answered the two following questions about which event(s) the current attacks reminded them of and why. 29 participants were excluded a priori from the originally collected samples (N Jan-FR = 501; N Nov-FR = 430; N Nov-GE = 376), for being underage and/or missing attention checks. An additional 81 (N Jan-FR = 5; N Nov-FR = 7; N Nov-GE = 69) were dropped for answering only one of the two relevant questions. Table 1 shows the characteristics of the final samples, and the 9/11 subsamples. Note. Total samples = participants who answered all three questions about whether, which and why the current attacks reminded them of a past event. The 9/11 subsamples = only participants who mentioned being reminded of the 11 September 2001 attacks. a Other nationality includes holders of multiple nationalities (e.g., French and Belgian). b Other professional status includes self-employed, unemployed and (pre-)pensioned.

Source Selection
Answers to the question which event(s) participants were reminded of were categorized into either specific sources (bearing the name of a past event or group), countries/non-specific events, or general statements. We identified specific sources by attributing the same code (e.g., "9/11") to different references to the same event (e.g., "attack on the Twin Towers", "attack in New York 2001"). Some recurrent ambiguous references ("attacks in London", "attacks in Madrid") were coded as the best-known related events London bombings, 2004.
Despite our asking for events specifically, participants also mentioned individuals and groups (e.g., the Nigerian terrorist group Boko Haram), which we also coded as instances of specific sources.
General references to countries (e.g., "Mali" or "Afghanistan"), broad and non-specific events (e.g., "Syrian War", "attacks in Pakistan") and unrecognizable events (e.g., "subway attack") were coded as "countries/non-specific events". In this category, we also included ambiguous responses (e.g., "attacks in Tunisia" could refer to at least three different attacks in the year 2015 alone).
Finally, broad instances ("all attacks that happen in the world", "all attacks against journalists") were categorized as "General statements". We coded multiple general statements by the same participant only once.
We focused in this paper only on specific sources. After establishing a list of specific sources mentioned at least once, each answer was dummy coded on this list (1 = mentioned; 0 = not mentioned). Participants produced on average 2 specific sources (Ms = 2.07, 2.19 and 2.7; SDs = 1.48, 1.31 and 1.49; respectively for the Jan-FR, Nov-FR and Nov-GE samples), with a range of answers that varied between 6 (for the Jan-FR and Nov-FR samples) and 8 (for the Nov-GE sample). Table 2 shows the most frequent sources (for all sources, see Table Sup2 in the Supplementary Materials).
We coded specific sources into the following categories: (a) Terrorism, (b) (civil) wars and state violence, (c) attacks on specific targeted groups (journalists, activists-including censorship attempts and attacks on media outlets), (d) assassinations (attempts) of politicians and (e) other or non-specific type (see Table 3).
Finally, we categorized the sources according to geographical location based on -but singled out sources specifically located in Belgium, France and Germany because of their relevance to our participants (see Table Sup4 in the Supplementary Materials).

Note.
Only sources with a frequency ≥ 3 appear in this table. There were multiple mentions per participants (Total N = 831). "Countries / Non-specific sources" refers to answers like "Mali", "Syrian war" and "Subway attack". "General statements" refers to answers like "all attacks against journalists". Numbers in parentheses indicate percentages relative to the total of each column.
Note. The numbers in this table do not take into account responses referring to countries/non-specific sources (e.g., "Mali", "Subway attack") or to general statements (e.g., "all attacks against journalists"). Because of multiple mentions of sources per participant, the arithmetic sum of respondents across types exceeds the actual sample size in each sample. Percentages in parentheses correspond to the value in the cell divided by the total of the corresponding column.

Comparison Categories
We manually coded participants' reasons for choosing those analogical sources and identified different comparison categories between the past and target attacks in four steps: (1) The lead author first coded the Jan-FR and Nov-FR samples.
(2) A research assistant fluent in both German and French coded the Nov-GE sample after training on a small sample from the French-speaking datasets. Uncertainties in the coding were resolved during discussions with the first coder. (3) The first coder recoded the entire data again 3 times to ensure coding consistency. Finally (4) a third French-German bilingual (the 5 th co-author) independently coded a randomly selected subset of 204 answers (24% of the total). Interrater agreement as measured by Cohen's Kappa ranged from κ = .45 to κ = .95 depending on the categories; which represents a "moderate" to "almost perfect" agreement (Landis & Koch, 1977).
Three categories had lower agreement but their extremely low frequency (n < 7) made this computed statistic unreliable (see supplementary analysis 1 in the Supplementary Materials).
We coded answers without a recognizable topic as "uncategorized". Between 72% and 85% of participants produced at least partially coded statements (see Table Sup3 in the Supplementary Materials).
A total of 18 comparison categories were identified excluding "Uncategorized" answers. The final categories selected were present in at least two of the three samples. These categories were non-independent since each answer could be coded along several categories and the mean number of categories per participant was 1.79 (SD = 1.01 in the Jan-FR sample), 2.12 (SD = 1.24 in the Nov-FR sample) and 1.98 (SD = 1.08 in the Nov-GE sample). The total range of categories across participants varied between 6 (for the Jan-FR & Nov-GE sample) and 7 (for the Nov-FR sample). The content of those categories showed that the 2015 terrorist attacks were compared to past sources on various dimensions: What the attacks were (their nature), who was involved (actors), why it happened (motives), what will happen next (aftermath), who was targeted (targets), but also how (modus

Socio-Political Variables
Descriptive statistics and correlations between the socio-political variables ii in each sample appear in Tables Sup5-Sup7 in the Supplementary Materials. Here, only items collected in all samples were used (see full list in Table Sup1 in the Supplementary Materials). All items were measured with 7-point scales.

Online and Offline Engagement
Four items measured engagement online (e.g., "liked, retweeted or shared a content related to [the current attacks]") and offline ("exchanged with family and friends" about the attacks; α Jan-FR = .71; α Nov-FR = .62; α Nov-GE = .56).

Uniqueness of the Current Attacks
Participants indicated on four items the extent to which they "would qualify [the current attacks] as absolutely unique / incomparable / comparable to others (reverse coded)" (α Jan-FR = .81; α Nov-FR = .79; α Nov-GE = .79).

Islamophobia
Ten items (adapted from Imhoff & Recker, 2012) measured support for a negative view of Muslims (e.g., "In general Muslims are intolerant"; "Muslim dignitaries don't do enough to fight terrorism") and Islam in general ("Islam is a retrograde religion which encourages violence"). All ten items were averaged in a single scale (α Jan-FR = .75; α Nov-FR = .76; α Nov-GE = .76).

Support for Counterterrorist Policies (Self-Designed Scale)
Participants were asked to what extent they believed the French government should implement twelve policies following the attacks. Eight items related to restrictive policies (e.g., "Use physical and/or psychological torture against presumed terrorists if it allows to save lives"; α Jan-FR = .90; α Nov-FR = .85; α Nov-GE = .86). Four items tapped into preventive policies (e.g., "Encourage discussions and debates around democratic values in schools"; α Jan-FR = .55; α Nov-FR = .54; α Nov-GE = .55).

Support for US Patriot Act
Participants indicated how legitimate they found seven policies among those implemented by the US following the 9/11 attacks (e.g., "Increase measures allowing detection, prevention and legal pursuits of terrorists"; α Jan-FR = .86; α Nov-FR = .81; α Nov-GE = .81). (The measure was adapted from Abdolian & Takooshian, 2002).

Political Orientation
Participants reported their political orientation on one item, ranging from 1 "extreme-left" to 7 "extreme-right".
Journal of Social and Political Psychology 2020, Vol. 8 (2) Historical Analogies After the Paris Attacks

Source Selection
In each sample, between 17% and 20% of all answers were coded as referring to either countries, non-specific events or general statements. Less than 1% consisted of autobiographical events (e.g., the intrusion of an armed man in a participant's workplace or the death of a loved one); suggesting that the 2015 terrorist attacks were a matter of collective, rather than individual, sense making (see also Pelletier & Drozda-Senkowska, 2016).
For specific sources, we identified in total 76 (Jan-FR), 51 (Nov-FR) and 55 (Nov-GE) distinct ones but most (between 69% and 79%) were mentioned only 3 times or less; resulting in a "long tail"-shaped frequency distribution (see Table 2). All three samples were dominated by references to the 9/11 attacks (mentioned by 55% to 70% of participants). However, after the November attacks, the previous January attacks became a second favorite (mentioned by 37% to 55% of participants). This was especially true in the Nov-FR sample which consisted mostly of French participants.

Temporal and Geographical Proximity
Overall, the sources mentioned were recent. Most sources that could be dated took place post-2001 (Jan-FR = 48%; Nov-FR = 60%; Nov-GE = 71%). This skewness was even more pronounced after the November attacks: between 21% (Nov-FR) and 24% (Nov-GE) of the sources had taken place the previous year (2015). The common perpetrator (ISIL) of many of those 2015 attacks and the temporal proximity with the Paris attacks have likely increased their salience.
A Eurocentric/Western-centric tendency was observed in the geographical location of sources selected ( Overall, source choices tended to overestimate one's ingroup (national, European or 'Westerner'), with sources located in or closest to one's home (geographically or culturally) being the most frequent.

Types of Sources
Even though comparisons to different types of events were found (e.g., wars and state violence, assassinations of politicians), between 83% and 95% of participants across samples mentioned at least one terrorism-related source (see Table 3). References to religious, specifically "Islamist", type of terrorism were by far the most frequent-though fewer mentions also related to far-right and far-left terrorism. However, the Jan-FR sample was notable in that more than a third (34%) of the cited sources referred to attacks (including censorship attempts) on journalists, media outlets, activists and public figures of various kinds-though a few participants (11%) were responsible for this variety. In the Nov-FR and Nov-GE sample, only 2% of the cited sources fell in this category.

Source Interpretation
Given that several comparison categories could be mobilized by a single participant, we needed a way to examine whether the meanings attributed to a given analogical source varied across participants. Thus, we selected the most frequent source across samples, the 9/11 attacks; which mobilized almost every comparison category (with one exception; see Table Sup3 in the Supplementary Materials). We then conducted a multiple correspondence analysis (MCA) on the subgroup of participants who had mentioned the 9/11 attacks in all three samples (total N = 518).

Multiple Correspondence Analysis (MCA)
MCA is an exploratory multivariate method that summarizes the relationship between multiple categorical variables via a set of dimensions (or factors), based on their co-occurrences (Husson et al., 2017). Its output provides a dimensional space on which both variables and participants are projected. In this space, the relative position of these points can be interpreted as a measure of their association. Given the number of categories to consider, their binary nature and their non-independence, MCA allowed us to explore both the association between the categories of comparison themselves and with the (quantitative) socio-political variables.
Because low frequency variables can bias MCA results, three of the comparison categories (western responsibility, impact for Muslims and escalation; see Table Sup3  In the MCA, each participant is described by their coordinates on each dimension. To interpret the dimensions, we selected the variables that were (1) best represented on each dimension (i.e., whose position on the dimensional plane can be reliably interpreted; Husson et al., 2017) and (2)   Note. The labels of the comparison categories are colored based on their quality of representation (cos 2 ) on the dimensional plane, with lighter blue representing a higher quality of representation compared to the deeper shade of blue, as represented by the color gradient on the right. The circles correspond to the three clusters that emerge after analyzing the comparison categories most associated with the opposite poles of each dimension (i.e., the two axes).
We then applied a hierarchical, agglomerative clustering (HCPC) on the first two MCA dimensions. The HCPC was applied using the Ward's method with Euclidian distance matrix, followed by K-means consolidation to increase the quality of the partition (Husson et al., 2017). As in the MCA, the three-cluster solution also emerged in HCPC as relatively 'optimal', i.e., a solution that minimizes the inter-cluster inertia gain (see Figure Sup3 in supplementary analysis 2 in the Supplementary Materials). Since the dimension-based interpretation was congruent with the cluster-based interpretation, we only present the latter (Figure 2). Table 4 shows which supplementary (sociopolitical) variables were most associated with each cluster.
Finally, we identified representative participants of each cluster. We selected participants who had the best quality of representation and contributed most to the overall variance of the two MCA dimensions, and used their answers to illustrate the analysis provided in the next section. Note. The three clusters can be distinguished by the color and shapes of the points: Green cluster = circles; Red cluster = triangles; Purple cluster = squares. Figures 1 and 2) -The first cluster (N = 307; 59% of the MCA sample) was characterized by references to the "terrorist" nature of the attacks (9/11 and the current ones), their "religious and ideological motives", and the "random targets" they involved.

Cluster 1 -Focus on terrorism and (religious/ideological) motives (in green in
"Random target" included references to the general death toll and casualties, but also to the notion that the targets-either human or countries-were innocent and did not fit a particular "profile".
Consider these two representative answers from the Jan-FR sample on why they thought of 9/11 after the series of attacks in January 2015:

An attack that terrorizes a whole nation, political grievances and innocent victims
(Extract 2, participant Jan-FR-521) Both 9/11 and the January attacks embodied "terrorism" and "terrorizing" events, for the direct targets and the "whole nation"; and were underpinned by religious or political motives. In both cases, references to the targets emphasized their innocence ("innocent victims", "deaths of innocents"), and their killing was portrayed as a mere strategic mean ("collateral damage") in the pursuit of (political or religious) objectives.
Others construed the indisputable innocence of the victims and their unwarranted targeting by highlighting specific elements in the profiles of the victims like their age; as illustrated by this extract from the Nov-FR sample:

It is about the premeditated murder of children, teenagers or innocent adults, perpetrated under the influence of an ideology
(Extract 3, participant Nov-FR-1554) The victims' innocence was emphasized by reference to categories of people whose innocence was deemed indisputable and therefore went unqualified ("children", "teenagers"), or to people whose innocence was explicitly asserted ("innocent adults").
Some answers pointed to the similarities in target locations (rather than populations), as in the following extract from a German participant:

Because all of them were probably terrorist attacks [committed] by Islamist groups on the western world (Europe & USA)
(Extract 4, participant Nov-GE-804) The 9/11 analogy here construed Europe and the US as part of a specific category ("the Western world"), whose commonality was constituted by their targeting at the hand of specific perpetrators ("Islamist groups"). A similar take was found after the January attacks:

Because it is an Islamist terrorist attack with the purpose of attacking democracy-related values (Extract 5, participant Jan-FR-286)
Terrorism was portrayed as a conflict of "values" relating to Islam ("Islamist") and to democracy ("democracy-related"). Others emphasized the opposition of values by referring to the distinct ethnic or religious profile of the perpetrators: Arabs/Muslims, demands stemming from their religion (Extract 6, participant Jan-FR-446) Here, the terrorist impulse was explicitly located in religion, within Islam specifically; with an emphasis on the 'otherness' of the perpetrators ("their religion") who were constituted as a distinct outgroup.
In summary, this first cluster of participants put forth a specific type of narrative about the 9/11 attacks, and by analogy, about the current ones. They embodied "mainstream" views, highlighting the terrorist nature of the attacks, their religious and ideological motives as well as the targeting of innocent and randomly selected victims. When examining the quantitative variables most associated with this first cluster and compared to the whole sample means (see Table 4), these participants tended to score higher on Islamophobia, supported more restrictive counterterrorist policies by the French government, and perceived the set of counterterrorist policies implemented by the US after the 9/11 attacks (i.e., Patriot Act) as more legitimate. They also leaned more to the right politically and showed less support for preventive counterterrorist policies by the French government-although these associations were weaker. Figures 1 and 2) -The second cluster (N = 160; 31% of the sample) was dominated by references to the "collective" and "personal impacts" of the attacks, and the use of negative, value-laden-and sometimes emotional-descriptors for them ("negative evaluation" category, see Table Sup3 in the Supplementary Materials). The "collective" and "personal impacts" categories included answers where events were compared because they were perceived as equally impactful-especially emotionally-either in a collective or in a personalized fashion. They also included references to the unexpectedness of the attacks (past and present) and the sense of vulnerability or powerlessness they induced, as one participant stated in reference to the comparison between 9/11 and the January attacks:

Cluster 2 -Focus on collective and personal impact (in red in
It's a matter of violence and of surprise of the events that have affected many people (Extract 7, While the differences between the two events were sometimes recognized, the emotional and personal impacts were deemed equivalent and central to the meaning of the analogy: The collective (emotional) impact, and the strong sense of identification with the victims was also a recurrent theme on this dimensional pole:

They attacked a democratic country and targeted a population similar to us. They surprised us and have
terrified an entire country (Extract 9, participant Nov-FR-1467) In summary, this second cluster was formed by participants who used the 9/11 analogy to highlight the impact (both personal and collective) of the attacks and their negative evaluations. Moreover, two supplementary variables were most associated with this cluster (see Table 4): These participants tended to portray the 2015 attacks as more unique and incomparable to other events, and to search and exchange more information about the attacks both in online and offline settings iv . Figures 1 and 2) -The third cluster (N = 51; 10% of the sample) was characterized by references to the "instrumentalization of the attacks" (past and present), to the "excessive or unequal impact" they generated and to the (potential) "conspiracy" surrounding them. For one representative participant on this pole, the January attacks shared with 9/11 the following similarities:

Cluster 3 -Focus on disproportional reactions, political instrumentalization and conspiracy (in purple in
The movement of fear and the security measures that ensued. The fact that attacks in the West are deemed of crucial importance on the international scale whereas similar tragedies elsewhere go unnoticed.

The fact of using shock to put in place security measures and establish [one's] power
(Extract 10, participant Jan-FR-1016) The January attacks and 9/11 were both depicted here as events noticeable for their collective displays of emotions ("movement of fear") and their policy impacts ("security measures"); where the latter instrumentalized the former ("using shock to put in place security measures"). The account therefore presented emotions not just as reactions to events, but also as entities that can be manipulated to "establish power". Moreover, an implicit criticism was raised of the special treatment reserved for "attacks in the West" compared with "similar tragedies elsewhere" that "go unnoticed".

Political instrumentalization sometimes carried a hint of conspiracy:
Having unfolded in circumstances that are still very mysterious, they serve as pretext to launch previously planned wars against Afghanistan and Iraq (Extract 11, The account sheds doubt on the official 9/11 narrative because of its "still very mysterious" circumstances, and because the Afghanistan/Iraq wars were said to have been "previously-planned"; suggesting that instrumentalization, or even a conspiracy was involved.
Likewise, after the 13 November attacks: The event was experienced almost live and made accessible through the media, similar to 9/11, and the political reactions to it were just as similar and quick as they were then, the solution should be retaliation, violence and war. The attacks were once again used as a pretext for pursuing further military, power politics and economic goals, which will, however, only fuel further terror and completely destroy the attacked countries (Extract 12, participant Nov-GE-400) More than a "pretext" for other military, political and economic goals, instrumentalization of the attacks constituted a reinforcing element for perpetuating "further terror". Thus, terrorism was portrayed here not only as the product of religious or ideological forces (as in the first cluster) but also as a consequence of specific (foreign or domestic) policies.
In summary, this last cluster took a more "critical" stance on the attacks (past and present): emphasizing their political instrumentalization, the excessive and/or unequal amount of attention and reactions they generated, as well as the "mysterious" (or even conspiratorial) circumstances of the attacks. Participants on this "critical" cluster (compared to the whole sample mean; see Table 4) tended to lean more to the left politically, to perceive the current attacks as less unique, to score lower on Islamophobia, to support the implementation of less restrictive-and more preventive-counterterrorist policies by the French government, and to deem the 2001 US Patriot Act as less legitimate. Participants in this cluster were also slightly more often from the January sample (v-test = 2.37, p = .018).
Journal of Social and Political Psychology 2020, Vol. 8(2), 721-747 https://doi.org/10.5964/jspp.v8i2.1212 Note. All scales were measured on 1 (low/not at all/extreme left) to 7 (high/strongly/extreme right) scales. c The test value (v-test) is considered statistically significant when its absolute value is bigger than 2. Its sign indicates the direction of the association.

Discussion
The purpose of this paper is to examine how ordinary people in three European countries (Belgium, France, Germany) generated their own historical analogies and justified them in the aftermath of the two series of terrorist attacks that hit France in in January and November of 2015. Overall, most participants mention a relatively small cluster of past events, characterized by similarities in time (recent), space (geographically close) and type (terrorist attacks) with the current situations. While sources located in Europe and North America gather the highest consensus across samples (see  for a similar finding), this tendency seems especially pronounced in (p. 679). This is vividly illustrated in our samples where the largest cluster of participants (Cluster 1) justify their 9/11 analogy simply by the fact that "this is terrorism". Thus, the September 11 th attacks do seem to constitute an anchor for European citizens' representations of the Paris attacks (see also Truc, 2016); echoing Kilby (2016)'s claim that '9/11' has become synonymous with terrorism itself, guiding the sense-making of contemporary terrorist attacks in Europe.
The general patterns observed in source selection fit both with research on analogical cognition (Holyoak & Koh, 1987) and with the policymaking literature on historical analogies (Jervis, 1976;Khong, 1992) showing that the selection of analogical sources is sensitive to the presence of surface similarities with the target. It also fits with the type of historical analogies that are most featured in newspapers (Axelrod & Forster, 2017). The correspondence in the patterns of analogical selection between newspapers and citizens' responses may suggest that citizens are simply reproducing the historical analogies to which they were exposed in the media. Conversely, journalists might select analogical sources that they expect to resonate with their audience, because these events happened recently or close by. Moreover, to the extent that recent, European and terrorist events are generally overrepresented in social representations of history , the correspondence in analogy selection between media and lay people might also be the product of collective sense-making attempts in which the meaning of current events is derived by mobilizing those shared representations. In any case, the dominance of such a small cluster of events in our study suggests a relative consensus regarding which past events are deemed relevant to the Paris attacks by citizens of different European countries (France, Belgium and Germany).
Notably, the January sample produced an overall larger number of sources compared to the two November samples. In line with "symbolic coping" (Wagner, Kronberger, & Seifert, 2002), the November attacks-as opposed to the January ones-might have generated fewer analogical sources because people had become by then somewhat familiarized to this kind of events given that ISIL had claimed dozens of other terrorist attacks in the year 2015 vi . As collective sense-making tools, the breadth of historical sources produced might therefore indicate a familiarization process in the public sphere.
However, the January sample is also distinctive in the type of sources produced: A third of sources relate to past events in which people had been specifically targeted for what they said and/or what they did (attacks on journalists, media outlets, and activists). In contrast, this type of sources is practically absent in both samples surveyed after the November attacks. In other words, according to a particular subsample of participants, the relevant type of categorization for the January attacks was not in terms of terrorism but rather as a targeted, physical or symbolic, attack against specific categories of people.
We believe that this difference has to do with the particularities of the two attacks. The January attacks consisted of a series of attacks targeting the headquarters of the satirical magazine Charlie Hebdo, a police officer, and a Jewish supermarket, the Hypercacher. Public discourse revolved heavily around the definition and boundaries of liberal democratic values (freedom of speech and of the press, secularism), including outside of France (Castelli Gattinara, 2017). Arguably, the salience of these themes at the time made previous targeted attacks against specific categories of people (like journalists and activists) particularly relevant anchors for the January attacks.
In contrast, the targets of the November attacks were public places not associated with a particular category of people. Accordingly, our findings suggest that there was less consensus regarding the categorization of the January attacks and their defining features than there was for the November attacks.
But these "metaphor wars" (Paris, 2002) do not lie only at the source selection level (about which past event constitutes the most applicable analogy) but also at the source interpretation level (about what a given past event means for the present; Elcheroth et al., 2011). Even when people agree about the relevance of a particular past event (like 9/11) for the present, they can relate this "same" event to the present in diverse-sometimes even conflicting-ways, which align somewhat with their current concerns. While some evoke 9/11 mainly to emphasize the analogous terrorist nature and the religious or ideological underpinnings of the Paris attacks (Cluster 1), others point to how both 9/11 and the Paris attacks have affected them and/or the world around them in various ways (emotionally, socially, politically; Cluster 2). A third cluster of people uses the analogy to question the excessive or unfair amount of reactions that those events (9/11 and the 2015 attacks) generate, their political instrumentalization, and even to challenge the official narrative about the attacks in favor of conspiratorial ones.
Thus participants who draw analogies between the Paris and 9/11 attacks seem to do so for different reasons, with each cluster mobilizing different representations not only of the present, but also of the past. Furthermore, these representations seem to form implicit arguments about specific issues surrounding the attacks. For instance, implicit arguments are raised about who deserves 'our' empathy and grief. Those pointing to the collective and personal impacts of the attacks (Cluster 2) stress the role of (cultural and geographical) proximity in generating a greater sense of identification with the victims. Interviews after the Paris attacks found a similar emphasis on personal connections and cultural proximity when people discussed expressions of online solidarity (e.g., displaying the French flag on Facebook; Tóth, 2017). In our case, we find that participants drawing these analogies also tend to report more engagement with current events (both online and offline), and to portray the current attacks as more unique and incomparable to other events-a remarkable stance given that this cluster, like all participants included in this study, does in fact draw analogies with other past events, and to 9/11 in particular. They appear to simultaneously proclaim the uniqueness of their experience while comparing it to how they, and others, have felt and reacted after 9/11. For some, emphasizing the similarities between the populations targeted in 2001 and 2015 (e.g., "a population similar to us") may have helped derive a sense of uniqueness precisely through the analogy. Indeed, one of the common stances against the use of historical analogies is that the comparison downplays the uniqueness of either event (see Todorov, 2004). Here, it is the opposite, as the association with a better-established event like the 9/11 attacks-one that epitomizes contemporary terrorism (Kilby, 2016)-seems to magnify the importance of the present situation; lending further credibility to the uniqueness of one's experience.
As if in reply to this 'exceptionalist' argument for who deserves "our" grief, Cluster 3 questions the 'excessive' reactions to events that take place in the 'West', as opposed to events that happen elsewhere in the world, maintaining that reactions should be equivalent for similar events regardless of their location. The argument is taken further at times to portray those 'excessive' emotional reactions as the very instrument of political control, expressed both domestically (e.g., increased security measures) and abroad (military interventions in the Middle-East)-with political consequences that can themselves perpetuate the cycle of terror (for a similar narrative in the media, see Friedersdorf, 2015;Timm, 2015). In keeping with this view, Cluster 3 tends to be less supportive of the restrictive counterterrorist policies implemented by both the French government, and the US (Patriot Act after 9/11). It also scores lower on the Islamophobia scale, thus lending less support to portrayals of Muslims as retrograde, terrorist sympathizers whose values are incompatible with 'Western' ones.
In contrast, Cluster 1 stresses the "terrorist" nature of past and present attacks, and their religiously or ideologically, often supposedly Islam-inspired, motives. Some of these participants point to the commonality of values between the populations targeted by 9/11 and the Paris attacks-just as those in the Cluster 2 who highlighted how excep-tional these attacks are (in terms of impact and reactions). However, in this case, the commonality of values is presented as the cause, rather than the consequence, for being attacked: "We" were both targeted because "our" values are similar ("democratic", "Western"). "Our" difference with the perpetrators, and "their" ideology or religion further reinforces the sense of shared values between the American population in 2001 and the French one in 2015. Accordingly, participants in Cluster 1, in contrast with those of Cluster 3, tend to show more support for restrictive counterterrorist policies, both in the present and in the past, and to score higher on the Islamophobia scale. This is in line with experimental evidence showing that using the "terrorist" and "Islamist" labels-as was the case for many answers in Cluster 1-changes how an event is construed and how people think it should be dealt with (Baele, Sterck, Slingeneyer, & Lits, 2019). The narrative in Cluster 1 also maps with the narrative found in the media coverage of the 2015 Paris attacks, where those attacks were construed as an "outsider" phenomenon, perpetrated by domestic or foreign Muslims pictured as a hostile 'Other' (see Połońska-Kimunguyi & Gillespie, 2016).
Thus, we see here how commonalities between the past and the present are also used to define the boundaries of social identities (Ghilani et al., 2017): By constructing who is part of "us" vs. "them", delineating who deserves "our" grief and attention, and by labeling the perpetrators in ways that emphasize their otherness. Indeed, social identities are central to social representations of history (Elcheroth et al., 2011;Liu & Hilton, 2005). And those categorizations in turn are action oriented (Leudar, Marsland, & Nekvapil, 2004), (de)legitimizing past (Patriot Act) and present actions (counterterrorist measures to be implemented by the French government). Thus, when using historical analogies, "speakers are not simply representing history, but they are also using the analogy to construct other entities" (Gibson, 2012, p. 23; emphasis in the original).
Through the 9/11 analogy, participants are not simply emphasizing various sets of similarities between the past and the present attacks. Instead, just like the politicians in the televised debate studied by Gibson (2012), our participants use constructions of the past and the present to advance various positions on salient public and political issues surrounding the 2015 Paris attacks (e.g., why were 'we' attacked, who deserves to be grieved, how should the government respond). These findings are reminiscent of Billig's (1991) view that people's attitudes constitute "stances taken in matters of controversy: They are positions in arguments" (p. 143; see also Billig, 1996). The implicit argumentative nature of our respondents' answers remains mostly implicit, but is revealed when examining their collective production: We find then the 'seed' needed to both justify their own position and (more or less explicitly) criticize the opposite stance. Had participants been directly exposed with someone voicing that opposite stance, those seeds would likely have bloomed into an explicit, full-fledged argument. This suggests that historical analogies can be regarded not just as the product of externally imposed influence on a passive audience, but also as a rhetorical resource that people can wield to construct-and argue-the meanings of the past, and by analogy, of the present.
To conclude, we have argued that historical analogies, as meaning-making tools, are part of the broader process of collective sense-making, which helps explain their continued prevalence in public and political discourses (Ghilani et al., 2017) and the fierce "metaphor wars" that often erupt around them (Paris, 2002)-given that they carry implications for the meaning of the present and how best it should be dealt with. Thus, historical analogies allow to study how people construct representations of new events (by anchoring them in existing social represen-studying historical analogies could help shed light on the complex, dynamic interplay of consensus and disagreements in social representations (see Voelklein & Howarth, 2005).

Limitations and Future Directions
While considering the findings of this study, it is important to keep in mind some of its limitations. The most important is the representativity of the three clusters of participants who used the 9/11 analogy: Cluster 3 was smaller compared to the other two. In an MCA, categories with low frequency can disproportionately contribute to the variance of a dimension. This effect is less of a concern when the focus of the researchers is on the social relevance of a phenomenon rather than on its numerical importance (Cibois, 1997). To address this issue nonetheless, we ran several MCAs where less frequent categories were alternatively included as active and supplementary variables (see supplementary analysis 2 in the Supplementary Materials). The overall pattern of results remained unchanged.
However, we do not expect that every terrorist attack would yield analogies to the same sources identified here or that they will result in the same comparison categories. The content of those answers and the patterns of correlations will likely differ. But we do expect that the observed variations between specific past events and the meanings attributed to them will hold across contexts-especially since similar flexibility has been found even for older past events, whose meanings appear better established than 9/11 (like WWII; see Spellman & Holyoak, 1992).
In addition, the generalizability of our findings is restricted by the relatively young age of our participants; a consequence of relying on student and social media sampling. This offered us crucial comparability given that we set out to study different events in time (the January and November attacks) and their representations across national contexts (Belgium, France and Germany). Indeed, cross-cultural research on social representations of history has often explicitly sought out university students, precisely because they tend to have equivalent educational backgrounds and to be similarly exposed to a "globalized culture" , minimizing cultural differences.
However, the relatively small range of ages represented in our samples did limit our ability to investigate generational-effects in the production of historical analogies (e.g., Schuman & Corning, 2006). This calls for a conceptual replication of our study with more representative samples.
Finally, it is worth noting that the small size of the associations found in the MCA between comparison categories and socio-political variables might relate to the richer content of open answers (that were coded to extract comparison categories) compared to the restrictiveness of close-ended variables (used to measure socio-political variables). Indeed, open questions, while particularly adapted to exploring novel facets of a phenomenon without unduly restricting participants' answers, often result in a variability that can prove challenging to interpretation (see Doise, Clemence, & Lorenzi-Cioldi, 1992, chap. 2). Future studies could alleviate this issue by asking participants to generate historical analogies for a particular aspect of the target situation (e.g., "which past event reminds you of the way the government responded to the terrorist attacks?") rather than for the situation as a whole like we did here. This strategy, by eliminating some of the variability, might result in clearer patterns and stronger links between freely generated historical analogies and sociopolitical attitudes.

Notes
i) The questionnaire referred specifically the "Charlie Hebdo attack", because it was the first event in the series of attacks in January 2015, and was connected to the subsequent attacks. Participants also seemed to assume that the label "Charlie Hebdo attack" included the whole series of January attacks, as illustrated by the fact that several drew analogies to anti-Semitic events; suggesting that the Charlie Hebdo and subsequent Hypercacher attacks were viewed as a single event.
ii) Reliability indicators are only reported for the 9/11 subsamples.
iii) Several MCA's were run on top of the final analysis reported here and the first included all comparison categories as active variables. The overall pattern of results remained unchanged (see supplementary analysis 2 in the Supplementary Materials). The final selection of the active vs. supplementary variables was based on the purpose of reducing the dependence of the plane on low-frequency categories while keeping the highest number of categories in order to meaningfully distinguish groups of participants from one another (see Cibois, 2014).
iv) Negative emotions also tended to be more associated with the negative side of dimension 2 (i.e., where Cluster 2 is located; see supplementary analysis 2 in the Supplementary Materials). However, this association did not emerge in the cluster analysis and was thus left aside.
v) We are grateful to Gerôme Truc for this insight.
vi) We wish to thank Inari Sakki for this alternative explanation.

Funding
The first, second and fifth authors were supported by grants from the Belgian Funds for Scientific Research (FRS-FNRS).