Communication Without Legislation? A Cross-National Field Experiment on Members of Parliament

Do legislators communicate with constituents about affairs that they cannot legislate? An influential literature underlines the communication of what legislators do in the legislature to their constituents. This article questions the hypothetical link between communication and legislation. I conduct a cross-national field experiment on members of national parliaments (MPs) to investigate how they behave when ordinary citizens require them to explain what they cannot legislate. Overall, the results show that MPs are evenly responsive to explanation requests within and outside their legislative competence. Likewise, their replies to inquiries on what they can and cannot legislate are equally quick and just as detailed. Further analysis reveals that electoral incentives motivate MPs to account for even those affairs that they cannot legislate. This suggests that, with regard to the content, legislators’ communicative accountability may be larger than previously supposed.


Introduction
Communicating with constituents comes with the job description of a legislator. They go all out for explaining their legislative activities to their constituents, not only because it builds trust and creates legitimacy (Fenno 1978) but also because it helps them get re-elected (Mayhew 1974). However, their job becomes complicated in democracies where legislators are not responsible for all legislative activities, yet where constituents are often unaware of the division of competences (Cutler 2004(Cutler , 2008Leó n 2011). Questioning the link between legislation and communication, this article analyses whether electoral incentives make legislators go out of their way to explain political affairs beyond their competence.
Legislators reach out to the voters indirectly through their speech in the legislature or to the media, and also directly by meeting many of them in their constituency and by interacting with increasingly many more on the Internet. Normatively, this serves democracy well because accountability requires a continuous communication between legislators and their constituents (Mansbridge 2004). A crucial subject of this communication is the personal accounts for their legislative activities that take place away from the eyes of ordinary people. This is why theorists define in chorus 'an obligation to explain and to justify his or her conduct' (Bovens 2007: 450) or 'an obligation to explain . . . his or her past actions' (Mansbridge 2003: 516) for a legislator as any representative needs to 'answer to another for what he does' (Pitkin 1967: 55). 1 Empirical studies of legislative behaviour confirm that legislative activities are indeed the main subject of the legislators' communication with their constituents. Explaining is one of the three components of the 'home styles' that Fenno (1978) observes among legislators working in their constituencies. There they provide information on and justification for 'what they have done while they have been away from home' (Fenno 1978: 136). Besides, the necessity to communicate explanations can also determine what they do in the legislature. Kingdon (1989Kingdon ( [1973: 47 -54) finds that legislators weigh their intended vote against the potential calls for explanations in their constituency. They might refrain from voting for or against a bill if they think they cannot meet the potential calls with the justifications they have.
The theoretical contribution of this article is to introduce the distinction between legislator's explanations of what they do and what they do not do. Compared to what we know about the communication of the decisions within their competences, it remains untested how legislators behave when ordinary citizens require them to explain affairs that are out of their control. Do legislators communicate with constituents about affairs that they cannot legislate? What (de)motivates them to do so? Answers to these questions have important implications not only for our understanding of legislative behaviour but also for the quality of democracy in multi-level systems.
The focus on the communication of what legislators do poses a limitation for our understanding of democratic accountability and legislative behaviour in political systems with more than one level of legislative decision-making. Legislators do not have authority and responsibility for all affairs at all times in an increasing number of political systems where the legislative power is shared among legislatures at different levels. As a result, there are affairs that fall partially or fully outside as well as inside what legislators can and cannot do-a distinction that may not always be clear to ordinary citizens.
Communication without legislation may enhance the overall democratic accountability in a political system as this could 'offer a remedy to the otherwise opaque procedures, the overwhelming complexity, and the difficult attribution of political responsibility in decisionmaking beyond the nation state' (Rauh 2015: 118). First, teaching the people 'what it does not know' is one of the classic functions of parliaments (Bagehot 2001(Bagehot [1867: 101), and legislators can explain the division of competences to their constituents in order to help them learn who is responsible. Second, legislators 'can make us hear what otherwise we should not' (Bagehot 2001(Bagehot [1867: 102). By providing justifications for their own position in these affairs and by blaming the others to force them to do the same, legislators can increase the public accountability of decisions in a political system even if they are not the ones making these particular decisions. Otherwise, a strict limitation of communication to legislation may not only mean these opportunities would be missed, but it also may lead to shirking among the legislators.
The empirical contribution of this study is to provide evidence from a cross-national field experiment on elected public officials: 2133 members of national parliaments (MPs) form five countries in the European Union (EU) were contacted by fictitious constituents about affairs that are now partially or fully transferred to the supranational EU level. This is to remedy an important methodological shortcoming of the existing research-the exclusive focus on formal channels of communication available to MPs. Earlier studies of national plenary debates and parliamentary questions find generally a poor communication effort in supranational affairs (Auel and Raunio 2014;Bergman et al. 2003;García Lupato 2014;Saalfeld 2003). 2 This article questions whether our existing knowledge on communicative behaviour of national MPs in supranational affairs is biased by the exclusive focus on formal channels of communication. Much of the communication between MPs and their constituents indeed takes place in an informal and interactive setting outside the parliament, but this type of communication is hard to capture due to the limitations of observational methods.
Contrary to the results from the studies of the plenary debates and parliamentary questions, this article finds that national MPs do not discriminate against supranational affairs in their communication to their constituents. The overall results show that MPs are evenly responsive to explanation requests within and outside their legislative competence. Moreover, their replies to inquiries on national versus supranational affairs are equally quick and just as detailed. Further analysis reveals that electoral incentives motivate MPs to account for even those affairs that they cannot legislate.

MPs and Communication Strategies
The theoretical assumptions of this study stems from the idea that political actors behave purposefully in pursuit of their preferences (Downs 1957). MPs might pursue various preferences while in parliament, but these all come down to one precondition: re-election as an MP (Cain et al. 1987;Epstein 1967;Mayhew 1974;Rae 1971). Therefore, MPs strategically choose their actions to secure their re-election. They have to choose, and they have to do so strategically, because they compete for parliamentary seats under the constraints of scarce resources and institutional rules.
Communication is no exception; MPs behave purposefully in their communication with their constituents in a way that contributes to their re-election prospects. The process of cultivating support for the election to the parliament is a discursively constituted process (Disch 2011;Mansbridge 2003;Saward 2010;Urbinati 2006). Top-down communication includes information, justification, and explanations in order to educate and persuade people to re-elect them for another term (Mansbridge 2003;Rawls 1993: 165;Urbinati 2006). However, representatives do not simply provide pure information in straightforward monologues, they instead use 'crafted talk' in their communication to change the perceptions of the citizens (Jacobs and Shapiro 2000).
Communication in this sense refers to, what Saward (2010: 10) calls, 'the subtle and deeper processes of constructing the represented' in representation. It is subtle because, as Disch (2011) argues, representatives communicate not only to inform and educate their electorates, but more often and more importantly so, they communicate to manipulate the preferences of the represented for their electoral favour. This mobilises, or 'constructs', a group of people around a political cause. One of the political causes that MPs strive to mobilise citizens around is their re-election to the parliament. This makes political communication 'competitive, self-interested, and oriented toward winning elections' (Disch 2011: 103). Therefore, representatives have the incentives to cultivate electoral support by communicating to their potential voters.

Strategies in Multi-level Settings
With the increasing number of multi-level democracies, the question arises as to how these re-election strategies are affected when constituents have multiple elected representatives at different levels. No study has yet examined the effects on communication strategies per se, but regional and federal studies on constituency service provide important insights for further investigation. The first of these insights is about the demand side: constituents expect any representative to help them with their grievances without taking jurisdictional COMMUNICATION WITHOUT LEGISLATION competences into account (André et al. 2014(André et al. , 2015Bradbury 2007). As Patzelt (2007: 64) puts it, 'most citizens do not bother about such distinctions. They simply turn to the most accessible MP' for their casework.
I theorise that constituents misdirect their calls for explanations as well, and replicate this in a field experiment detailed in the subsequent section. There is no indication that constituents would be more accurate with their calls for policy explanations than their calls for service. Besides, citizens do indeed tend to have a limited knowledge of the supranational level compared to domestic levels (Anderson 1998). Moreover, Saalfeld (2003) argues that the multi-level nature of politics makes it harder not only for the citizens but also for their representatives to locate where credit or responsibility is due. While this ambiguity could lead MPs to stay away from communicating supranational affairs in their plenary speeches or parliamentary questions (Saalfeld 2003), I argue that it may result in completely the opposite in informal and interactive forms communications. Unlike in a plenary debate, MPs have to react to the communications from their constituents in the informal and interactive forms, where misdirected demands put MPs in a position to respond.
A related insight is about the supply side, which defines the electoral competition as the main determinant for the legislators' response to the misdirected demands. Studies on constituency service in multi-level democracies show that demands from constituents are hard to dismiss for legislators under tougher electoral competition even when they are misdirected (André et al. 2014(André et al. , 2015Russell and Bradbury 2007). If legislators ignore misdirected demands, even for principled reasons such as the division of competences, they 'might develop a reputation for being unhelpful, and would fear an adverse effect on their personal vote' (Franks 2007: 39). Besides, as Cain et al. argue, 'whereas constituents attach considerable importance to policy making in the abstract, when it comes to evaluating their representatives, policy considerations decline sharply, and constituent assistance and district service considerations advance ' (1987: 51). Even if the subject matter might belong to the competence of another legislator, the same constituents have a vote for each legislature.
Electoral systems generate different levels of incentives for MPs to care about their personal reputation in their constituency. In elections where voters allocate seats directly to candidates rather than to parties, electoral systems encourage candidates and incumbents alike to cultivate a personal vote for themselves (Carey and Shugart 1995). Legislators put a greater emphasis on their relationship with their constituents in these systems  because their re-election prospects depends less on their party and more on their constituents. Therefore, I expect the likelihood for a legislator to communicate affairs outside their competence to increase with the incentives to cultivate a personal vote in their electoral system: Hypothesis 1: The higher the incentives to cultivate a personal vote in an electoral system, the more likely are the MPs to communicate affairs that they cannot legislate.
The value of a vote differs for legislators, depending on the safety of their seat. As the electoral safety of a seat decreases, each vote becomes more critical for legislators because the number of votes separating success from failure to secure a seat also decreases. Empirical analyses of legislative behaviour repeatedly show that MPs who only marginally won the previous election prioritise re-election seeking activities before anything else during their time in parliament (Cain et al. 1987;Gaines 1998;Heitshusen et al. 2005;Norton and Wood 1993). As electoral incentives to increase their vote shares are higher for electorally insecure MPs, I expect that MPs from electoral safe seats are more likely to ignore misdirected calls for explanations: Hypothesis 2: The higher the marginality of their seat, the more likely are the MPs to communicate affairs that they cannot legislate.
Finally, the two determinants of communication without legislation defined above are not mutually exclusive of each other. Incentives at the levels of electoral systems (Hypothesis 1) and individuals (Hypothesis 2) can interact. In particular, I expect the effect of personal electoral marginality to increase with the incentives to cultivate a personal vote in an electoral system: Hypothesis 3: The higher the incentives to cultivate a personal vote in an electoral system, the greater the effect of MPs' electoral marginality on their communication without legislation.

The Cross-national Field Experiment
In order to test these expectations on the communicative behaviour of individual legislators in multi-level democracies, I conduct a field experiment on national MPs in the EU. Experiments are one of the relatively new methods mushrooming in political science (for an excellent review, see Grose 2014), and this study builds upon the earlier field experiments on elected officials by introducing the cross-national aspect into the research design over five EU Member States-Finland, France, Germany, Ireland and the UK. This selection provides a large variety of electoral systems with highly divergent levels of personalisation. First, it includes all major types of electoral systems (plurality, proportional and mixed). 3 In addition, there is also a variation within some of these systems. For example, while Finland and Ireland both use open-list proportional representation systems, the former ensures proportionality with an open list while the latter safeguards it with a single-transferrable vote. Similarly, although both France and the UK use plurality voting systems, the former has a two-round system that guarantees a majority. These electoral systems yield significantly different levels of incentives to cultivate a personal vote (Carey and Shugart 1995;Johnson and Wallack 2012). Including cases that lie at separate ends of the scoring system is an advantage of this case selection; as Pilet et al. experienced in their own study of legislative behaviour, 'comparing countries with very contrasted electoral systems is extremely useful' (2012: 369).
The case selection provides two further advantages. First, it controls for the time of accession to the EU. This is important because, as Mattila and Raunio (2006: 442-443) show, the opinion gap between parties and their voters is smaller in the new member states. In this case selection, all cases classify as old member states. Second, the time period was away from any election date. There were no elections before May 2015 in any of the cases, which allowed data collection through field experiment between September 2014 and April 2015. For accessibility of parliamentarians as well as for the reliability of responses, the field experiments can best be carried out while parliaments are in their regular working terms.
This experiment provides an opportunity to explore the communicative behaviour of legislators in real-life settings, where constituents do not always contact the right legislator for the right affairs. By looking at the legislator-initiated formal communications, as indeed the existing literature on parliamentary communication of EU affairs exclusively does, we miss much of the actual communication between MPs and their constituents taking place outside the parliaments. This could potentially be a big miss due to the (often-misplaced) demand factor in the informal and interactive communications. However, these types of communication are hard, if ever possible, to document with observational methods of data collection. This is why the experimental design is crucial, and the importance of the research question outweighs the costs of field experimentation. 4

COMMUNICATION WITHOUT LEGISLATION
The Experimental Design I randomly assigned members of each parliament into two groups, who received an email from a fictitious constituent to their official e-mail address found on parliamentary websites, requesting information on their work as an MP. 5 Appendix D presents several tests on this randomisation, showing the process was successful with regard to the observable characteristics of MPs. There were two treatment groups and no control group, because a control group who receives no e-mail and therefore reveals no behaviour is not desirable in this design (for a similar situation, see Richardson and John 2012). The first treatment group of parliamentarians (n ¼ 1068) received an e-mail with an explicit enquiry about their legislative work on EU immigration. The second group (n ¼ 1065) received fundamentally the same enquiry on immigration without any explicit reference to the EU immigrants. Both e-mails mention social services, opening the opportunity-especially for the first group-to reply the e-mail without communicating the EU affairs. This makes it possible to analyse not only whether the subjects respond but also how they respond. The texts in the English version of the treatments, sent to MPs from Ireland and the UK, read as in Text Box 1. 6 In choosing the policy area and the framing for the treatments, I aimed to satisfy three criteria. The first was to imitate genuine communications that take place between MPs and ordinary citizens in real life. Field experiments are generally known to yield high external validity because interventions take place in real-life settings (De Rooij et al. 2009;Gerber and Green 2012) beside the internal validity intrinsic to all experiments. Intervention itself is also important to create this validity. Immigration policy ticks the right boxes in this sense. It is a popular policy area for both citizens and the representatives alike, and therefore one that is discussed between the two: citizens frequently contact their representatives about immigration . This is also important for a rather technical reason that subjects must find treatments plausible for the field experiment to work. Owing to the popularity of the policy area in all of the case countries, I expected that MPs must have received similar communications before and therefore that they would find the e-mail plausible.

TEXT BOX 1 E-mails in English
I took further practical measures to ensure plausibility and to prevent revelation. Texts were translated into the official languages of the parliaments by native speakers. I then tested the texts on a number of ordinary citizens from each case country to confirm the right balance between keeping treatments the same across the languages while adopting the language to each political system. Depending on the size of the parliaments, I created a number of e-mail accounts. 7 In order not to introduce undesirable variation, I chose typical male names from each country and registered credible addresses with a common e-mail service provider. I then randomly divided the e-mails into five, and sent them on 10 -14 November 2014 at 2 pm local time in the case countries. Bivariate tests show that neither the names nor the sending waves had any significant impact as to whether MPs replied to their e-mail. Table SI3 in Appendix D presents these tests.
The second criterion was to initiate a simple communication with a range of policy dimensions from supranational to local in all EU Member States. Immigration as a policy area provides important advantages for this end as well. It is a salient issue across the member states. The free movement of European citizens within the Union is fundamental to European integration, which is a supranational principle guaranteed by the EU law. However, immigration of EU nationals is only a part of the overall immigration, and each member state controls the non-EU immigration with their own legislation. Furthermore, immigration is also a local issue. Constituencies attract immigration in different numbers and types (Schö nwälder 2013).
Finally, immigration policy offered the room for meaningful variations, a central aspect in any experiment, with respect to the legislative competence. The treatment for the first group, who receive a question on EU immigration, explicitly puts them under a misdirected call to explain what they do not do as legislators: to control EU immigration. In contrast, the second group received a question on immigration, which covers both the EU and non-EU immigrants. Unlike the former, because the immigration of non-EU nationals is under the control of each Member State, and national MPs have considerably high decision-making power in this second type. Although EU immigrants make up an important portion of foreign nationals in every Member State, and therefore the EU dimension is crucial to answer the questions in both treatments, I expect MPs are more likely to explain what they do rather than what they do not do. As discussed above, incentives to bring up an issue outside their competence is lower in proactive communications, as evident from the parliamentary speeches and questions in EU affairs.

Results
The replies started to arrive in a matter of an hour while it took 141 days for the latest MP to answer their constituent's e-mail. Eventually, just under half of MPs replied to the treatments: there were 971 replies to 2133 successfully sent e-mails, amounting to 45.52% overall response rate. This means that just over half of MPs do not respond to all e-mail inquiries from their constituents. Nevertheless, this is comparatively high figure for similar audit studies on legislators, suggesting that the measures taken to imitate genuine communications were successful. Indeed, many MPs noted in their replies that they received similar communications before about the same issue from 'several other constituents'. Figure 1 shows the response rates for each parliament. It shows that the parliaments in Germany, Finland and Ireland are all positioned very close the mean with similar rates. However, visibly diverging from this group are the French Assemblée Nationale with fewer than 20% and the British House of Commons with over 70% response rate. To a certain COMMUNICATION WITHOUT LEGISLATION extent, these patterns might be a result of the overall differences in attitudes towards e-mails among MPs. Tellingly, for example, France was the case with highest number of e-mail failures by far: some e-mail addresses found on the website of the Assemblée Nationale did in fact not exist anymore while some others were full and incapable of receiving e-mails even if they existed. On the contrary, there were no such failures in the UK. The fact that French MPs can hold multiple offices might also have an effect on their response rate.  Notes: This table reports the results from the unpaired t-test of the replies to the treatments. The second and third columns show the response rates to EU immigration and immigration treatments, respectively, in each of the five national parliaments. The fourth column reports the differences in response rates, where positive values indicate differential treatment in favour of supranational affairs. The p-values, reported in the last column, are two-tailed and calculated using unpaired two sample t-tests.
Adding the dimension of treatments to the analysis, Table 1 reports the results from unpaired t-tests. How did the explicitly supranational aspect of EU immigration affect the communicative behaviour of the legislators at the national level? Overall, the tests show that there was no significant difference across the treatment conditions. The response rates to the question on EU immigration on the one hand and immigration on the other were almost exactly the same, with the latter attracting less than 1% higher rate of reply (45.04% versus 46.01%, where p ¼ .65). The differences in response rates, reported in the fourth column, are comparatively bigger at the level of individual parliaments but still statistically insignificant, all except for France. The results suggest that members of the Assemblée Nationale are the only group of MPs in this study to significantly shy away from EU affairs as they are significantly less likely to respond to the treatment on EU immigration (p ¼ .02).
The rates above exclude automatic replies to e-mail messages-a functionality that several MPs use. Nevertheless, it was evident in these automatic replies how e-mails are now an important channel of communication between MPs and their constituents. A typical automatic reply explains that 'e-mails are treated in the same manner and given the same priority as other forms of communication' and that the MP is working hard to deal with huge number of e-mails, which is claimed to be 'as many as 400 a day'. Such a volume of communications would force MPs to prioritise their communicative behaviour as they cannot answer every single communication, and the response rate confirms that they do so.
Therefore, another way to test whether the communication of what legislators can and cannot legislate differs significantly is to test their priorities. In order to do so, among those MPs who communicated back to their constituent with a reply, I compared the mean number of days it took for the replies to arrive and the mean number of words that these replies included in the two treatment conditions. Table 2 presents the results from these t-tests, demonstrating that there was no statistically significant difference across the treatment groups. This result holds for all MPs pooled as well as MPs in individual parliaments, this time also including the French Assemblée Nationale.
These results may raise a concern about the strength of the treatments. Here the key question is whether the EU aspect-the difference between the treatments-was strong enough to produce measurable effects. Specifically, for the experimental design to be effective, the EU Treatment had to signal the legislators that they were being approached about a matter beyond their control. It was evident in the responses that this was indeed the case. Among those who answered the question on EU immigration, 69.6% of the MPs openly mentioned that they or their parliament could not control EU immigration. In their justifications, many pointed to the EU level in general or particularly to the principle of free movement therein. As expected, however, none of the responses to the question on immigration suggested that the question was misdirected for jurisdictional reasons.

Determinants of Communication Without Legislation
To provide a further robustness check for the results as well as to test the theoretical expectations above, I next turn to the causal mechanisms behind communication without legislation. Why do legislators go out of their way to explain political affairs beyond their legislative competence? To answer this question, I first coded the responses into two categories. I classified a legislator as Ignores EU affairs if (a) they did not reply to their constituent's e-mail inquiry or (b) the EU aspect was completely missing in their response even though they did reply. I classified them as Explains EU affairs if legislators accounted for the supranational COMMUNICATION WITHOUT LEGISLATION policies that they cannot legislate. Overall, the results show that 20.2% of legislators explain while 79.8% ignore EU affairs. Figure 2 plots these results by the treatment conditions for each of the five parliaments. Table 3 presents the results from regression estimates of the legislators' communicative behaviour when ordinary citizens require them to explain what they cannot legislate. Because the dependent variable is binary, i.e., ignoring (y ¼ 0) and explaining (y ¼ 1) EU affairs, and the number of parliaments is small (n ¼ 5), I use logistic regression models with fixed-effects Notes: This table reports the results from the unpaired t-test of (a) the number of days it took MPs to reply to their treatment and (b) the number of words they used in their reply. The second and third columns show the mean number for the EU Immigration and Immigration treatments, respectively, in each of the five national parliaments. The values in brackets are the standard deviations. The fourth column reports the differences, where positive values indicate differential treatment in favour of supranational affairs. The p-values, reported in the last column, are twotailed and calculated using unpaired two sample t-tests.

FIGURE 2
Communication of EU affairs Notes: The dependent variable is whether the legislator ignored (y ¼ 0) or explained (y ¼ 1) the immigration policy of the EU. Both models estimated via binomial logistic regression. The Finnish Eduskunta is the excluded parliament. * p , .05 (two-tailed tests). * * p , .01 (two-tailed tests). * * * p , .001 (two-tailed tests).

COMMUNICATION WITHOUT LEGISLATION
approach. 8 In addition to the dummy variables for the parliaments, the models include several control variables for covariates at the level of political parties of the legislators as well as individual level of legislators themselves. 9 Model 1, the baseline model without any interaction terms, shows that the coefficient estimates for the treatment of EU Immigration and the incentives to cultivate personal vote (ICPV) are both positive and statistically significant. This demonstrates that constituent inquiry and electoral system-based incentives are strong predictors of communication without legislation. To start with the former, the legislators who received the explicit inquiry about EU immigration (x ¼ 1), compared to those who received the inquiry on immigration (x ¼ 0), were more likely to explain political affairs that they cannot legislate. This confirms that the noticeable differences in the bar heights between the facets of Figure 2 are indeed significant differences. Figure 3 plots the effect of constituent inquiry and ICPV on the predicted probability of communication without legislation based on Model 1. It shows that legislators are 9.4 percentage points more likely to explain affairs beyond their legislative competence if there is an explicit constituent inquiry about these affairs. Similarly, ICPV has a significant effect as it roughly doubles the average probability of legislators to provide explanations. According to Figure 3, a change from zero to five (out of possible six) in the measure of ICPV is associated with an increase this probability from 0.12 to 0.25 percentage points. This confirms Hypothesis 1.
The results provide conditional support for the expectation about the relationship between electoral safety of legislators and their communicative behaviour. Electoral Safety 10 is not significant in Model 1, suggesting that legislators with safe seats are no less likely to explain affairs beyond their legislative competence than legislators with electorally marginal seats. Hence, the results do not support Hypothesis 2. However, there is support for Hypothesis 3 that incentives at the level of electoral system on the one hand and individual legislators on the other are interactive. To test whether the effect of ICPV vary with Electoral Safety, Model 2 includes an interaction term of these variables. Here the results show a negative and statistically significant  Table 3, Model 1. Vertical lines indicate the 95% confidence interval.
relationship. This estimate measures the difference between the effect of ICPV when Electoral Safety is low and the effect of ICPV when Electoral Safety is high. Figure 4 plots the marginal effect of ICPV on the outcome that legislators explain the EU affairs to their constituents. It demonstrates that ICPV has a positive effect on the legislators' explanation of affairs beyond their control only when Electoral Safety is low. In other words, where electoral systems generate high levels of incentives for MPs to care about their personal reputation in their constituency, those MPs with unsafe seats are significantly more likely to explain political affairs even if they are not the responsible representatives for those particular affairs. On average, ICPV increases the probability of unsafe legislators to explain EU immigration by 4.3 percentage points while this figure is 2.5% for legislators with competitive seats and only 0.06 for electorally safe legislators.
In the presence of this interaction, the estimated effects of the EU immigration treatment and ICPV stay significant. In Model 2, the direct effect of Electoral Safety becomes significant while staying positive, indicating that seat safety has a significantly positive effect on communication without legislation in the absence of incentives to cultivate a personal vote. These are the systems where legislators have the incentives to cultivate votes for their party rather than for their personal name, such as the system electing half of the MPs to the German Bundestag. These German MPs may not be personally responsible for EU immigration, but their political party, as it is the case for many parties in the EU, competes for elections at all levels including the EU level. As safest places on party lists go to the important figures in each party, it is politically meaningful that those MPs safety has a significantly positive effect in these systems.

Conclusion
In an increasing number of democracies, legislative power is divided or shared among legislatures at different levels. This poses a challenge to the existing understanding of  Table 3, Model 2. Dotted lines indicate the 95% confidence interval around the fitted line.

COMMUNICATION WITHOUT LEGISLATION
communicative accountability in legislative representation-the understanding that suggests accountability is limited to legislative authority and legislators account for what they legislate. It is an academic challenge because we might be evaluating legislative behaviour of communication against the outdated benchmarks. More importantly, however, it is political challenge because, if the communicative accountability is indeed strictly limited to legislative competence, dividing and sharing legislative authority might weaken the overall accountability by creating grey areas of responsibility. As ordinary citizens are often unaware of or bemused by these divisions of responsibility among their various representatives, it becomes harder to locate and hold the right legislator to account. I conducted a cross-national field experiment on members of five national parliaments in the EU to test the borders of legislators' communicative accountability. The overall results demonstrate that, in their communications with their constituents, legislators do not shy away from affairs that they cannot legislate. MPs were no less likely to respond to constituent inquiries in EU immigration-an area which national legislators cannot control. Likewise, their replies to inquiries on what they can and cannot legislate were equally quick in terms of days and just as detailed in terms of words.
At the same time, the results also suggest that about half of MPs do not respond to all email inquiries from their constituents. This was particularly pronounced in France, where fewer than 20% of MPs sent a reply. The French MPs were also the only exception to the equal communication of what they can and cannot legislate as they returned a significantly lower response rate for the inquiry on EU immigration. But then again, their timing and wordiness did not differ significantly between the treatment groups.
These results contradict the observational studies of the plenary debates and parliamentary questions at the national level, which find a generally poor communication effort in supranational affairs in multi-level democracies. I argued that the interactive aspect of communication between representatives and the represented, which is missing in plenary debates and parliamentary questions, could bias the finding of these studies. The field experimental design, which replicates genuine communications and induces electoral incentives, proved essential in investigating why legislators would go out of their way to explain political affairs beyond their competence.
This study also provides important new evidence about the behavioural implications of electoral incentives for legislators in diverse settings. There is a strong support for the expectations on electoral incentives as the determinants of communication without legislation. First, compared to the treatment on immigration, the treatment on EU immigration was significantly more likely to generate explanations of EU policy on immigration although an important portion of MPs in the former group explained the supranational policy as well. The second source of incentives was related to the electoral systems under which legislators compete for their re-election. MPs were significantly more likely to disregard the legislative competences in their communications with their constituents under the electoral systems that create higher incentives to cultivate a personal vote. Finally, there was also a role, albeit a conditional one, for the incentives created by electoral safety of individual legislators. I found that MPs with unsafe seats were significantly more likely to explain EU immigration as long as their electoral system incentivised them to cultivate a personal vote for their name.
The EU is a stark example of division of legislative authority among different legislatures with its legislative arrangements at the supranational, national, and in some member states regional levels. How legislators adopt to these arrangements concern the parliamentary representation of about half a million people in the EU. Yet this is still an evolving system, and there are questions to be answered about whether the findings from the EU would hold in more stable systems such as multi-level nation states. Another viable direction is to bridge the gap between these findings on misdirected policy inquiries and the misdirected calls for constituency work in multi-level systems. Do the borders of legislative accountability differ between policy and constituency work? Only future research can tell. bounced back any communication or did not have an e-mail address on legislative websites at all.

6.
For treatments in other languages, see Appendix A of the online supporting information.

7.
I used one account roughly per 100 parliamentary seats. For example, I created two accounts for Eduskunta which has 200 members.

8.
An alternative way of estimation with random effects instead-as Table SI5 in the supporting information shows-yields very similar results, indicating that the results presented here are robust across the choice of modelling. 9.
I collected the data on majority of these variables from the official websites of the parliaments and the electoral authority of each country. However, the data on the incentives to cultivate a personal vote originate from the database created by Johnson and Wallack (2012) while the 2014 Chapel Hill Expert Survey (Bakker et al. 2015) provides the data on party positions for Salience of EU and Anti-Immigration Policy. The supporting information includes a detailed codebook for each of these variables.

10.
Following the approach developed by Heitshusen et al. (2005) for comparing electoral safety of legislators across various electoral systems, I coded Electoral Safety as 0 for unsafe, 1 for competitive, 2 for safe seats and treated this variable as an interval-level variable-just as I treated ICPV. Appendix H in the supporting information shows that the results reported here are robust to treating these variables as categorical variables as well.