Introduction

International events can have domestic consequences. For instance, the popularity of the head of government can surge suddenly following an unexpected high-profile international incident (Mueller 1970, p.19). This rally ‘round the flag effect has been studied extensively for the American President (Baum 2002; Baker and Oneal 2001). The extent to which this effect can ‘travel’ to other systems than the US is a matter of debate (Hetherington and Nelson 2003; Morgan and Anderson 1999). There is a limited number of studies outside of the US. Nearly all of these study systems where one political party monopolises power.Footnote 1 The question is whether the rally effect can be observed in a multiparty coalition system where political power is shared between different parties. The goal of this study is to examine how an international conflict influences the popularity of ministers of different political parties in a multiparty parliamentary system with coalition government. To this end, this study looks at the change in the approval of the Dutch Prime Minister, Mark Rutte of the Liberal Party (VVD) and Vice Prime Minister, Lodewijk Asscher of the Labour Party (PvdA), after a diplomatic incident between the Dutch and Turkish government in March 2017.

This paper offers three important contributions to the existing literature about rally effects. Firstly, we study a least likely case. Both where it comes to the country (a multiparty system) and the specific event (a diplomatic conflict), we are unlikely to see a great rally. We examine the Netherlands, a multiparty, parliamentary system where political responsibility is shared by cabinet ministers of the different political parties. In such systems, voters are unlikely to hold a single politician responsible for the government’s actions (Hetherington and Nelson 2003, p. 38; Hobolt et al. 2013). Specifically, we study a diplomatic incident. While it has been recognised that diplomatic events can cause a rally in support (Mueller 1970),Footnote 2 studies do show that diplomatic events elicit smaller rally effects than military events (Lai and Reiter 2005, p. 268).

We test a theory developed for the American President in a European setting characterised by a multiparty system, parliamentarism and coalition government. Our goal is not just to see whether this theory works in a parliament setting but how it works in that setting. There are two options: if the public recognises that the government responsibility is shared between the different parties in the coalition, the boost in support for the Prime Minister should also be visible for other members of his cabinet. Therefore, we also study the increase in support for Vice Prime Minister Lodewijk Asscher of the social-democratic Labour Party (PvdA). It could also be that in the increase in support of the Prime Minister is not shared by his colleagues, because all responsibility is attributed to the Prime Minister, despite the parliamentary system. In that case, the isolated increase in support for the Prime Minister may signal the presidentialisation of his office in the eyes of the voter (Fiers and Krouwel 2007). This has important implications to our understanding how citizens see their parliamentary government functioning.

Finally, this study has a matching quasi-experimental design (Li 2013). We take individuals who participated the same survey but who have answered it at different points in time: just before and just after the crisis hit its highest point. We compare their support for the Prime Minister and the Vice Prime Minister. In order to ensure that the differences we find are not because of differences in background variables, we use a matching approach: every respondent who answered the survey after the height of the crisis is matched with a ‘twin’ who is as similar as possible on relevant variables but answered the survey before the height of the crisis. The number of experimental or quasi-experimental designs in this field is limited (Kobayashi and Katagiri 2018; Lambert et al. 2010; Chatagnier 2012; Georgarakis 2017; Schubert et al. 2002; Feinstein 2018). Compared with the existing studies that tend to look at different samples over time (Mueller 1970) and between countries (Tir and Singh 2013; Singh and Tir 2018), the quasi-experimental design is better able to separate the stimulus from other factors that may influence support for the head of state. Compared with quasi-experimental studies of the same respondents at different points in time in one sample, our study has the advantage that we are able to compare respondents who experienced the stimulus with respondents who did not experience that stimulus.

This paper will have the following structure: the first section will discuss the existing theory on the rally effect and derive our hypotheses. The next section will consider the case selection in greater detail and discuss the specific events to put our case study in context. The third section will introduce our matching quasi-experimental approach. The fourth section will discuss the results of our analyses. The final section will draw a conclusion.

Theory and expectations

A rally effect is a sudden increase in the approval of a national leader in times of an international conflict (Mueller 1970, p. 19). Mueller (1970) was the first to systematically theorise and operationalise this notion: he found that US presidents get a short-lived, sudden increase in public approval after a major high-profile international event. For an event to evoke a rally in support, Mueller (1970, p. 21) proposed that:

  1. (1)

    it must be international in nature, as opposed to domestic, because only developments that confront the nation as a whole are likely to unify it;

  2. (2)

    it must involve the leader and the country of the leader directly, because if it only involved other nations the loyalties and sympathies of citizens may be divided;

  3. (3)

    it must be specific, dramatic and sharply focused, as opposed to gradual.

Mueller’s original sample of rally events included both military and diplomatic events, such as meetings between the US president and the leader of the Soviet Union: he believed that even these events that are less spectacular than the use of force, generated a “let’s-get-behind-the-President” effect (Mueller 1970, p. 22). Further research has revealed that military events tend to elicit larger rally effects than non-military events (Lai and Reiter 2005, p. 268; Singh and Tir 2018).

On average, rally effects tend to be small and short-lived (Mueller 1970, p. 28; Baker and Oneal 2001). There have been events that have evoked a larger increase: for instance, after the 9/11 attacks, support for President Bush increased by more than 20 percentage points (Baum 2002; Schubert et al. 2002; Gaines 2002). Yet, there are also notable military events that have not elicited a rally effect (Feinstein 2016).

1. Rally ‘round the flag hypothesis: after a rally event, the approval of the head of government is greater than before that event.

In this specific case, the hypothesis implies that after the height of the Dutch-Turkish crisis, the approval of the Dutch Prime Minister Mark Rutte should be greater than before.

Mechanisms

The occurrence of a rally round the flag effect is unrelated to the success of the action, as “[i]nvariably, the popular response to a President during international crisis is favourable, regardless of the wisdom of the policies he pursues.” (Polsby 1964, p. 112). Scholars have identified different mechanisms behind the rally effect.

One set of mechanisms are social–psychological in nature (Feinstein 2016). The basic idea is that in response to external threat citizens respond by increasing support to the President in a reflexive, kneejerk fashion (Baker and Oneal 2001; Baum 2002, p. 264; Chowanietz 2011, p. 676). Several mechanisms have been identified for this reaction: in-group/out-group reactions, patriotism and emotions.

The first proposes is that when the in-group is threatened from the outside, individuals may feel greater attachment to members of the in-group (Lai and Reiter 2005, pp. 256–257). The stronger in-group attachment can percolate to the group’s leaders (Theiler 2018). Other authors have emphasised the importance of patriotism (Hetherington and Nelson 2003, p. 37; Mueller 1970): the crisis threatens the national honour and prestige and thus activates the sense of attachment that citizens feel towards their homeland (Feinstein 2016, p. 308). The public unites behind their “commander-in-chief”, forms a front against the opposing country and supports their actions and leadership (Baker and Oneal 2001, p. 667; Lee 1977, p. 253). A third mechanism emphasises the importance of anger: citizens that are angry are more likely to support hawkish, high-risk policies (Lambert et al. 2010, p. 888). Individuals that desire retaliation against the perceived aggressor out of anger, are more likely to support a President that takes these actions (Lambert et al. 2010, p. 888; Lambert et al. 2011, p. 344). These different approaches are not completely separate: for instance, the attachment to the nation in particular may elicit strong emotions (Feinstein 2020).

Opinion leadership has also gained currency as a mechanism connecting a rally event and the rise in approval (Hetherington and Nelson 2003, p. 37; Brody and Shapiro 1989). This emphasises the information that voters receive. During an international crisis, the media are likely to present a more one-side argument (Baker and Oneal 2001). As the events quickly unfold, the government administration has a virtual monopoly on information (Brody 1991, p. 63). The domestic political debate is muted after the event, opposition politicians are likely to be silent or supportive of the actions of the administration. The media therefore cannot report partisan conflict and the public can only base their opinion on the official statements of the administration. The tone of the media coverage after the event is important (Groeling and Baum 2008; Chowanietz 2011). Only if this debate is positive, the support for the head of government will grow. Over time as more information becomes publicly available, the opposition politicians can criticise the head of government. This criticism may undermine the approval rating of the head of government.

Partisanship

An important moderating factor that has been observed is support for the party in government. Different authors have diverging expectations: one group of authors sees a positive relationship between partisanship and the rally effect (Oneal and Bryan 1995, p.383; Edwards and Swenson 1997, pp.202–203; Singh and Tir 2018): a head of government will receive a greater increase in approval from supporters of their own party than from members of other parties. The underlying mechanism they observe is the importance of elite cues: those who support the party of the head of government are more likely to follow these cues than those who do not (Singh and Tir 2018).

In contrast, Baum (2002, p. 267) proposes a ceiling effect. Prior to the event, the head of government already tends to receive considerable support from their own party. The supporters of opposition parties who tend not to support the president beforehand, are more likely to change their mind about the president. Similarly, Kobayashi and Katagiri (2018, p. 303) argue that liberals are more likely to shift their opinion in reaction to an external threat than conservatives because conservatives constantly feel threatened already.

Other authors argue that such effect will not be visible immediately: Sigelmann and Conover (1981, p. 307) argue that partisan differences are likely to be observed in the mid-to-long term as in the early stages of the conflict. The lack of elite criticism in the media prevents early partisan differences (Jacobson 2003, p. 712). Similarly, research by Lambert et al. (2011) shows that the rally effect and the prior ideological orientation are separate predictors of support for leaders.

There are grounds to expect a positive, negative and no relationship. In order to proceed, we need a testable hypothesis. Given that the best comparative evidence suggests a positive relationship, we formulated it positively, cognisant of the other possibilities:

2. Partisan-hypothesis: after a rally event, the increase in the approval of the head of government is greater among those who supported the party of the head of government before than among other citizens.

In this specific case, the hypothesis implies that after the height of the Dutch-Turkish crisis, the increase in the approval of the Dutch Prime Minister Rutte should be greater among VVD voters than among other voters.

Coalition government

Rally effects have first been observed in the United States (Mueller 1970). The US President is both head of government and head of state. He is party leader, symbolic leader of the nation and chief administrator with considerable power over foreign policy. Hetherington and Nelson (2003, p. 38) propose that political institutions that do not channel popular attention to a single leader are less likely to see these rally effects. In a presidential system, the ‘buck stops’ at the President’s desk. The President is responsible for all policy decisions. This makes them “the most inviting target” to attribute responsibility to (Hobolt et al. 2013, p. 170). Still, the US Secretaries of State and Defense saw similar rally effects in support as the US President after 9/11 (Gaines 2002, p. 534).

In parliamentary systems, governing is the task of the cabinet, rather than of a single leader. Decision-making is a collective responsibility of the entire cabinet (Morgan and Anderson 1999, p. 811). The literature on parliamentary government emphasises that the cohesiveness of these parliamentary governments can differ (Hobolt et al. 2013). When a voter faces a single-party government, they are likely to attribute responsibility for political events to that party and its leader (Hobolt et al. 2013, p. 170). However, when there are multiple parties in government, it becomes difficult for voters to know who they should hold responsible for government policy (Hobolt et al. 2013, p. 170). Coalition government may disperse the rally effect: the effect could be observed for more than one member of the government. The mechanisms discussed above should not just apply to one group: the out-group threat is likely to increase attachment to all members of the in-group, including different members of the government. The lack of opposition criticism will benefit all government parties. At the same time, there is evidence that the European coalition systems are presidentialising: despite the constitutional sharing of power, in practice and in the eyes of the voters, the person who serves as Prime Minister functions more like a President (Poguntke and Webb 2007, p. 5). This comes with electoral benefits: the party of the Prime Minister performs better in elections than junior coalition parties (Klüver and Spoon 2020). Given that formally responsibility is shared, we propose that:

3. Ministerial hypothesis: after a rally event, the approval of government ministers is greater than before that event.

In this specific case, the hypothesis implies that after the height of the Dutch-Turkish crisis, approval of the Dutch Vice Prime Minister Lodewijk Asscher should be greater than before.

At the same time, coalition government may also suppress the effect. As it is more difficult to attribute full responsibility to a single person under parliamentary coalition government, voters are less likely to hold governments responsible if government cohesiveness declines (Hobolt et al. 2013, pp. 179–180); therefore, the rally effect is likely to be weaker than under single-party presidential government. This means that under such a system, one is less likely to find evidence for the rally hypothesis (H1).

Case selection and description

Below we will discuss our case, the diplomatic incident between the Dutch government and the Turkish government in March 2017, and our justification for selecting it.

Case selection

Most research on the rally effect has focused on the United States, where it was first observed. The rally effect has also been observed in studies of the UK Prime Minister (Sanders et al. 1987), the French President (Georgarakis 2017), the Russian President (Theiler 2018), the Japanese Prime Minister (Kobayashi and Katagiri 2018) and the Israeli Prime Minister (Feinstein 2018). With the exception of Israel, all these systems concentrate power in the hands of a limited number of political parties and politicians.

Table 1 lists the effective number of parties in governmentFootnote 3 that is the extent to which the executive power is fractionalised between different parties or concentrated in the hands of one party, in these countries and a number of West European democracies. Here, we can see that the countries where the rally effect has been studied differ systematically from the others. Israel forms an exception to this rule. Feinstein’s (2018) study, however, does not problematise the applicability of the rally effect to Israeli coalition government. The rally effect has been studied in comparative analyses (Tir and Singh 2013; Singh and Tir 2018), but these analyses did not include the cohesion of government into their models.

Table 1 Selected countries with different levels of government fractionalisation
Full size table
Table 2 Overview of the Dutch-Turkish diplomatic crisis
Full size table

The Netherlands with its parliamentary system and tradition of coalition politics is representative of the West European countries that differ from the cases that have been studied. This provides a good justification to study the effect of an international crisis on the support of the head of government in a multiparty, coalition system in the Netherlands. According to O’Malley’s (2007) survey of Prime Ministerial influence, the Dutch Prime Minister scores right in the centre of his international colleagues. The choice to study the rally effect in a multiparty, parliamentary system with coalition government in combination with the choice to study a diplomatic as opposed to a military conflict makes our case a least likely case study.

In the Netherlands, the Prime Minister has a position as first among his equals (Andeweg 1991). The Prime Minister manages broad multiparty coalition governments, as chairperson of the meeting of the council of ministers and as minister responsible for the coordination of government policy. The other government parties provide a Vice Prime Minister. These, together with the chairs of the parliamentary party groups, coordinate government policy. Specifically, in the realms of foreign policy and the media that the role of the Dutch Prime Minister has grown in recent decades (Fiers and Krouwel 2007, pp. 144–150). The role of the Prime Minister has grown in foreign policy because he is the ‘figurehead’ of the Netherlands in international forums. The Dutch media also focus on the person of the Prime Minister as the ‘personification’ of the Dutch government (Table 2).

The Dutch-Turkish crisis

In March 2017, the Netherlands and Turkey became embroiled in a diplomatic conflict. In both countries, politicians were campaigning: the Netherlands held national parliamentary elections and Turkey held a referendum on a new constitution. Turkish citizens in the Netherlands have the right to vote in the Turkish referendum. Turkish ministers were planning to come to the Netherlands to campaign in favour of the new constitution. The Dutch government held that “the Dutch public space is not the place for political campaigns of other countries”.Footnote 4 Therefore, The Dutch government tried to dissuade them from coming to the Netherlands. Despite this, the Turkish minister of Foreign Affairs, Mevlüt Çavuşoğlu, planned to come to the Netherlands to speak at the Turkish consulate in Rotterdam. On March 11, Çavuşoğlu threatened: “[i]f the Netherlands cancels my flight, [the Turkish government] will impose severe sanctions on them that will affect them economically and politically.”Footnote 5 The Dutch government stated they “would under no circumstances negotiate under blackmail”.Footnote 6 When Turkish authorities declined the Dutch governments’ terms and conditions concerning the reassurance of security standards, Rutte withdrew the landing rights for Çavuşoğlu's plane on that day.Footnote 7 At a public rally, the Turkish president Recep Tayyip Erdoğan responded that “[The Dutch] are Nazi remnants and fascists”.Footnote 8 That night, the Turkish minister of Family Affairs, Fatma Betül Sayan Kaya, went to Rotterdam to speak at the consulate. She was not allowed to leave her car and was escorted out of the country. The Dutch–Turkish diplomatic row then escalated with the Turkish president Erdoğan calling for international sanctions and withdrawing from diplomatic relations with the Netherlands.Footnote 9

The conflict coincided with the Dutch election campaign. During the conflict, Dutch Prime Minister Rutte, who also was the top candidate for the conservative-liberal VVD in the upcoming elections, stated how “a conflict of this nature takes countless hours to handle, while I should be busy campaigning right now!”Footnote 10 The events drew attention away from the campaign. On Saturday and Sunday, the televised news bulletins of the public broadcaster (NOS) and the commercial broadcaster (RTL) opened with these events and the NOS even interrupted its normal programming on Saturday with an additional new bulletin to report on these events. It is important to note that during this weekend, there were no televised debates or appearances of party leaders in evening talk shows. The media reports were positive for the government. The right-wing De Telegraaf wrote: “it is rarity in election time, but Dutch politicians from left and right have expressed their support for the just and principled stance of the cabinet against the Turkish insults and threats”.Footnote 11 The liberal NRC Handelsblad wrote Rutte “developed into leader who would not be trampled underfoot”.Footnote 12 The left-wing De Volkskrant wrote “Rutte and Koenders correctly saw [the actions of the Turkish government] as impermissible pressure for which one should not bow”.Footnote 13 In the days before the parliamentary elections, pollsters and pundits were unanimous in their expectation on the conflict: it would benefit Rutte, because “in a crisis, people tend to unite behind the crisis manager”.Footnote 14

The main opposition parties supported the actions of the government: Sybrand Buma, the leader of the Christian-Democratic Appeal said that the Prime Minister “has put a line in the sand together with the cabinet, with the whole of the Netherlands and all politicians”. The leader of the Freedom Party, Geert Wilders said: “It is good that the Turkish ministers were not allowed in and have been sent out”.Footnote 15 Jesse Klaver, the leader of the GreenLeft said “I think it is good to see that all political parties have closed ranks and support the cabinet in the chosen approach.”Footnote 16 The leader of the Socialist Party, Emile Roemer said “[a] limit has been crossed here [by the Turkish government]. Rutte has done well in this regard.”Footnote 17

This discussion of the context of the events points to a number of important factors that may contribute to the presence of a rally effect. Mueller (1970) notes that there needs to be an international conflict which involves the leader and that events must unfold at a specific time in a dramatic and sharply focused fashion. Indeed, there was an international conflict where the Dutch Prime Minister was involved in. It concerned a specific and sudden event. Public and media attention was focused on the decision not to let the plane with the Turkish minister land.

As we noted above, there are a number of mechanisms that may lead to rally effect. Of these, we want to highlight two. Firstly, patriotism: Erdoğan characterised the Dutch as remnants of Nazism and fascists. This may have activated a patriotic counter-response in the Dutch as they see themselves as a people that resisted the evils of fascism and came out of the Nazi occupation with their democratic and liberal values unscathed (Van Vree 1995, p. 64). Secondly, opinion leadership: conforming to the theory of opinion leadership, the leaders of major opposition parties supported the government and that the media reported positively on Rutte’s actions. All in all, important social–psychological and institutional mechanisms for a rally effect may be activated.

Methods

The goal of our research is to determine how the Turkish–Dutch crisis influenced the popularity of the Prime Minister and the Vice Prime Minister in the Netherlands. Existing studies of the rally effect have used different designs. These designs come with their own advantages and drawbacks. Most research compares the result of surveys with the same questions but different samples over time (Mueller 1970) and between countries (Singh and Tir 2018). While the advantage of these studies is that they allow us to assess the effect of a large number of events, a drawback of analysing this kind of data is that the effects found may be spurious (Kernell 1978). Mueller (1970, p. 27) emphasised that the rally effect was a parasite that explained smaller bumps and wiggles in support for the President. The major changes in support for the president are mainly explained by other factors, such as the economy. We can see this in the debate on the effect of the Falklands War on UK Prime Minister Margaret Thatcher’s approval. Norpoth (1987) saw a large effect, but Sanders (1987) argued that economic circumstances that already affected public opinion before the Falklands War, better explained this rise.

Kobayashi and Katagiri (2018) and Lambert et al. (2010, 2011) employ an experimental design. While the advantage of these studies is that they have true random assignment of the stimulus, a drawback of these studies is that they concern reactions to re-exposure about existing news about international incidents. They re-activate earlier reactions to a crisis, instead of modelling how the crisis activated them originally.

Finally, there are studies that use a pretest–post-test quasi-experimental design (Chatagnier 2012; Georgarakis 2017; Schubert et al. 2002; Feinstein 2018; Edwards and Swenson 1997). These studies compare the answers of the same respondents before and after an event. While an advantage of these studies is that they allow within-respondent comparisons, a drawback of these studies is that there is no control group that did not receive the stimulus. We cannot be sure that the observed change can be attributed to this specific stimulus instead of other factors that occurred in the time between surveys (Cranmer 2018, p. 2). This issue become particularly pressing if there is considerable time between the two surveys.

In this study, we use a matching quasi-experimental design. The advantage of this design is that it allows us to estimate the causal of effect of an international incident while drawing from an observational study (Li 2013). This method allows us to contrast a group of citizens that has been exposed to the news about the international incident (‘treated group’) with citizens who have not been exposed to this news, but who on all other relevant variables are as similar to the exposed group as possible (‘matched group’). This combines the advantages of looking at reactions to real-world events (like pretest–post-test studies) and has a control group (like experimental studies). The drawbacks are firstly that variables not included in the matching process may explain differences between the ‘treated group’ and the ‘matched group’ and secondly that because in our specific design there is still a small time difference between when the ‘treated group’ and the ‘matched group’ answered the survey, other events in this period may have caused the change (similar to pretest–post-test studies). We remedy the first issue by selecting matches on a sizeable number of relevant covariates and the second by spending particular attention to the political events that coincide with these events.

The design of the quasi-experiment

We use the responses from two surveys that I&O Research held during the campaign for the elections for the Dutch lower house of parliament in the Spring of 2017. I&O Research is a Dutch research firm specialised in researching societal issues. I&O Research has an online panel. The respondents for the I&O Research Panel are recruited via samples of municipal population registries and online self-enrolment; given the voluntary nature of participation, respondents are still likely to be more politically interested than a representative sample of the population.

We use a survey that happened to be in the field during the events. This has advantages and disadvantages. The advantage is that we can see the effect of the diplomatic crisis directly. Moreover, because the entire survey took place within a span of five days, it seems likely that the Dutch–Turkish events led to the observed changes. Note that in this period, there were no televised debates. The online campaign on Facebook only really ramped up in the last 48 hours before the election (Ruigrok et al. 2017, p.81). Of course, we cannot be certain that it is these events and not others have changed the minds of voters. This design also has some other disadvantages, because the survey was not purpose-built to track a rally event. Therefore, not all questions are optimal and we cannot include manipulation checks.Footnote 18 We use data from two surveys, Survey 1, in the field between 10 and 13 February 2019 and Survey 2, in the field between 9 and 13 March. These two surveys did not have the identical respondents, but have considerable overlap. The respondents from these surveys are drawn from the same I&O Research Panel with 25,000 respondents to construct separate representative samples of the Dutch population.

Our main approach is a matching approach (McCready 2006). This is a method to construct an experiment using observational data (Li 2013, p. 189). Survey 2 was put into the field in the Thursday before the Dutch–Turkish incident and respondents had the option of answering until the Monday morning after the incident. In our matching approach, we matched respondents who answered the survey after the height of the crisis (Saturday at 12h00) with respondents who answered the survey before. In this way, we create a group that was ‘treated’ with the ‘stimulus’, who had the opportunity to know the news about the crisis. We match the respondents on age (as a continuous variable), gender (as binary gender self-identification), education level (in three categories) and vote in 2012 (whether voters voted for the VVD or PvdA in this election). We use MatchIt to create the matched samples using propensity score matching. Specifically, we chose to create an “optimal sample”. MatchIt finds the match samples with a minimal average absolute distance on the relevant variables across all matched pairs (Ho et al. 2011, p. 11). This method stands contrast with the nearest neighbour matching that is used often in that the latter method is ‘greedy’, where the closest match is chosen for every individual treated unit one at a time, without trying to minimise a global distance measure (Ho et al. 2011).

We create two different samples using the matched approach. These samples differ in how we deal with respondents who give answers that one might categorise as missing answers. In each sample, we remove respondents who gave missing answers from the matching. The samples differ in how they treat those who answered ‘did not vote’ in the 2012 elections. We might treat this answer as meaningful, non-missing answers. We do so in Matched Sample 1. In Matched Sample 2, we treat these responses as missing answers. We use Matched Sample 1 in the paper and Matched Sample 2 as a robustness test in the appendix. 2490 respondents participated in wave 2 of the survey. 2018 of those have answered all questions used in this analysis. 440 answered the survey after 10 March 2019 12h00. This means that we construct a sample of 880 respondents.

A second robustness also presented in the Online Appendix employs a one-group pretest–post-test design. Here, we compare the responses from wave 1 with the responses from wave 2 (only for those who answered after 10 March 2019 12h00). This has the advantage that we are looking within respondent changes, instead of at a matched sample but the drawback is that other events that occurred between wave 1 and wave 2 may also affect the results.

Analytical strategy

The dependent variable is respondents’ assessment of the trustworthiness of Prime Minister Rutte and Vice Prime Minister Asscher. The survey asked respondents for different party leaders to answer whether they believed that “ <politician> is trustworthy (would be trustworthy) as prime minister”. They could use a five-point (dis)agreement scale.Footnote 19 While trust in a politician is distinct from the measure of approval that is common in the American literature on the rally effect, we believe that this indicator is relevant here. Trustworthiness is a good indicator of approval of the head of government, as trust is one of the main pillars of leadership (Hetherington 2005, p. 9; Gomibuchi 2004, p. 27). It is important to note here that the question does not pertain to trustworthiness of politicians in general, but ask respondents to consider whether they consider specific politician trustworthy.

In the matched approach, the key independent variable is whether respondents have answered the survey before (0) or after Saturday 11, at 12h00 (1). Our core analysis uses a paired sample t test to find out whether the events of the Dutch–Turkish crisis affected respondents’ assessment of Mark Rutte and Lodewijk Asscher. Table 3 shows the similarity in Treated and Control Group on the variables used to construct the matched sample. Moreover, in the Online Appendix, we test the robustness of the results with a stacked regression analysis.

Table 3 Treated and control group in matched sample 1
Full size table

To examine the Partisan Hypothesis, we run one regression analysis for Rutte and one for Asscher with the trustworthiness of the leader as a dependent variable and the stimulus and party choice in the 2012 election (differentiating between voting VVD, the PvdA and others) as dependent variables as well as an interaction between the stimulus and the party choices. This will allow us to see whether those who supported the PvdA and the VVD in the 2012 election show a larger increase in approval of the leaders than others.

Results

Our key question is whether the Dutch–Turkish crisis affected the views respondents have about the Dutch Prime Minister and Vice Prime Minister. To examine this, we employ a quasi-experimental design, comparing respondents who answered the survey after the height events of the crisis with respondents before this moment. To ensure that the difference we find is not the result of differences between the kind of respondents that answered the survey at different points in time, we construct a matched sample where every respondent in the treated group (who answered the survey after the height of the crisis) is matched to a respondent who is as similar as possible as them but who has not received the stimulus (answered the survey before the height of the crisis).

Our core hypothesis, the Rally Hypothesis, proposes that there is a difference in the assessment of the trustworthiness of Mark Rutte as Prime Minister before the height of the Dutch–Turkish crisis and afterwards. Table 4 shows the result of a paired sample t test. Here, we can see that there is a significant difference (at the 0.01 level) in voter judgement of the trustworthiness of Rutte before and after the height of the crisis. On a five-point scale, respondents in the treated group consider Rutte almost a third of point (a twelfth of the scale) more trustworthy. This a small but statistically significant effect.Footnote 20 Rutte was considered as trustworthy as Asscher before the events, but that after the events he was considered more trustworthy. In the Online Appendix (Table A.1), we look at two alternative operationalisations of the quasi-experiment: firstly, still using the matching sample but now treating those who did not vote as missing; secondly as a one-group pretest–post-test design. With these different approaches, we find the same pattern: a significant increase in the trustworthiness after the diplomatic incident. This evidence corroborates the Rally Hypothesis.

Table 4 Trustworthiness of Rutte and Asscher in matched sample 1
Full size table

The Ministerial Hypothesis proposed that as the leadership of Dutch coalition government is collective, a rally effect should also be observed for other ministers. In this study, we include the Vice Prime Minister Asscher. His trustworthiness is increased by a tenth of the effect we saw for Rutte (0.03 point on a five-point scale). This effect is not significant. In the Online Appendix, we look at a number of alternative strategies to construct the quasi-experiment. In those analyses, we again do not find a significant positive effect for Asscher. In the one-sample pretest–post-test design, we even find that in the period between the surveys the trustworthiness of Asscher declined significantly. In the Online Appendix (Table A.2), we also compare the differences for Rutte and Asscher more precisely in a stacked regression. We find the same pattern in each of them, a significant positive effect for Rutte but no such effect for Asscher. Therefore, we reject the Ministerial Hypothesis; it is not the case that the increased support for the head of government can also be observed for his ministers.

A critical question is whether the observed rally effect depends on the party preferences of the respondents. The Partisan Hypothesis proposed that the rally effect would be stronger for supporters of the party of the Prime Minister and Vice Prime Minister. Figures 1 and 2 show the effect of an interaction between party vote in 2012 and the stimulus on the support for Rutte and Asscher. (Expected values with 95% confidence intervals. Based on Model 4 (Fig. 1) and 5 (Fig. 2) in Table A.3 in the Online Appendix.)

Fig. 1
figure 1

Trustworthiness of Rutte according to VVD other voters before and during the crisis

Full size image
Fig. 2
figure 2

Trustworthiness of Asscher according to PvdA other voters before and during the crisis

Full size image

Figure 1 shows that those who supported the VVD in 2012 rate Rutte higher than those who did not and that those who support the PvdA rate Asscher higher than those who did not. It shows the effect of the stimulus for Rutte (and not for Asscher) but it also shows no interaction effect for either politician. Among those who supported other parties than the VVD, Rutte sees an increase of trustworthiness of 0.35 (significant at the 0.01-level); among those who supported the VVD, that increase is only 0.07 (not significant); among those who supported other parties than the PvdA, Asscher sees an increase of trustworthiness of 0.08 (not significant); among those who supported the PvdA, there is a decrease of − 0.09 (not significant). We do not find a consistent pattern in the robustness test in the Online Appendix (Table A.3): in the alternative matching strategy, we find a significant negative effect for Asscher among PvdA voters (but no significant positive effect among other voters). The one-group pretest-post-test designs show no significant pattern other than that those who voted for either party in 2012 trust their own leaders more in 2017. This does not lend credence to the Partisan Hypothesis in either direction.

Discussion and conclusion

We aimed to make three contributions to the literature about the rally effect in this article: in terms of case selection, theorising the rally effect in coalition systems and finally a methodological contribution. We will discuss those three elements in this conclusion.

External validity

The first contribution is that we proposed to examine the rally effect outside of the US and systems where responsibility can clearly be attributed to one political actor. We found a small but robustly significant, positive difference between judgements of the Dutch Prime Minister before and during the incident. We did not find evidence for a significant partisan effect: it is not the case that those who already support the party of the Prime Minister are more likely to support his policies. As Lambert et al. (2011) showed, the rally effect and the prior ideological orientation are separate predictors of support for leaders.

What do these results say outside of the case of the Dutch–Turkish crisis? An important contribution of this study is showing that the rally effect can occur in multiparty parliamentary coalition systems that are common in Europe and in reaction to diplomatic events. The rally effect cannot just be seen for US presidents engaging in militarised interstate disputes but also for European prime ministers engaging in a diplomatic kerfuffle. It seems likely that other diplomatic incidents, such as the 2005 Danish Cartoon crisis, have led to similar effects. These diplomatic incidents may seem small on the world stage, but this study shows that even these relatively small events can briefly increase the stature of a political leader.

Theoretical contribution

Our ambition was not just to see whether this phenomenon could be seen outside of the cases it had been studied, but also consider the theoretical implications of examining the rally effect in a multiparty, parliamentary, coalition government. Given that in a coalition system political responsibility is shared between parties, we expected that other members of the government may benefit from the surge in support for the head of government. To this end, we looked at how citizens felt about the Vice Prime Minister. We found no evidence for such a dispersion effect. This stands in contrast to the existing literature on government accountability, which emphasises that in the eyes of voters, coalition governments share the burdens and the benefits of coalition government (Hobolt et al. 2013).

The fact that the increase in support of the Prime Minister did not spill-over to his deputy may indicate that in the eyes of the voters the Dutch Prime Minister is closer to a president than one would expect on the basis of the Dutch constitution. Fiers and Krouwel (2007) emphasises that in particular in the realm of foreign policy and where it comes to media exposure, the Dutch Prime Minister has gained a more presidential aura. These are the two factors that come together in international crisis. This has implications for how political scientists should approach the position of Prime Minister in the voters’ eyes: when the public assesses foreign policy, it appears to be the case that responsibility mainly falls to the head of the government and not so much to other members of the government. The international profile of Prime Minister may, for instance, be one reason that senior government parties do better in elections than junior government parties.

Methodological innovation and limitations

The final contribution of this article is methodological. This is, to our knowledge, the first matching quasi-experimental design to examine the rally effect. Compared to most the literature that looks at changes in the approval of the head of government in successive surveys, the likelihood that our findings can be contributed to external factors is much smaller because we compare respondents who answer the survey with only hours in between. Our study also has advantages over the existing experimental designs that analyse the re-exposure of respondents to existing news, because we can show how respondents evaluate their leaders in the immediate aftermath of an international crisis.

This study, however, did have a number of limitations: firstly, our study relied on a matched sample that served as a control group for the treated group. The drawback of matching is that it is possible that an important predictor has been omitted in the matching process and that the two groups are not balanced on that variable. Secondly, there still is a short period between when the treated group and the matched group finished the survey: there is a chance that other factors contribute to the changed opinion but this likelihood is much smaller than other pretest–post-test quasi-experimental studies. Thirdly, we studied a diplomatic incident, instead of situation where countries used military force. While, from the first study of the rally effect, the possibility that diplomatic events instead of military events boost the support for a country’s leader has been recognised (Mueller 1970), research indicates that military events tend to elicit larger rally effects than diplomatic events (Lai and Reiter 2005, p. 268). The fact that this study concerns a relatively minor diplomatic scuffle further reinforces the least likely status of our case study. Fourthly, a quasi-experiment like this, where an existing survey in the field is repurposed to study events as they happen in reality, is not optimally designed for the task: for instance, in our quasi-experiment, we could not use a manipulation check to ensure that the ‘treated’ group was aware of the events. Our dependent measure (trustworthiness of a politician as Prime Minister) is also different from the more direct measure of approval used in the literature. These last two weaknesses, however, make it even more noteworthy that we do find a significant result.