Institutional change in parliament through cross-border partisan emulation

Abstract Institutional responses of parliaments to international developments are widely regarded as efficient changes because they tend to be unaffected by partisan preferences and benefit all members of parliament equally. This article challenges that common notion by providing evidence that the institutional responses of national parliaments to European integration are in large part the result of international partisan emulation. Spatial regression analyses robustly show that parliamentary EU oversight institutions diffuse across member states whose majority parties have similar constitutional preferences. A parliament is more likely to emulate the EU oversight institution of another parliament if their majority parties have similar ideas about the territorial distribution of power and institutional framework for policy making. This result has important implications for our understanding of institutional change in parliament. Responses of parliaments to external developments may appear non-partisan at first sight but unfold partisan characteristics if one looks beyond the domestic level.

In parliamentary democracies, political processes and outcomes are crucially affected by institutional rules within parliament. Rules influence voting behaviour in parliament and access to the parliamentary agenda (D€ oring 2001;M€ uller and Sieberer 2014;Rae and Taylor 1971;Tsebelis 2002). In short, they ultimately decide who holds power in parliament. For this reason, scholars pay close attention to institutional change in parliament and provide a range of theoretical explanations as to why and how parliamentary rules and procedures become modified (Binder 1996;H eritier 2006;Sieberer et al. 2016;Strøm et al. 2003). One of the most influential contributions on institutional change differentiates between 'efficient change' that benefits all members of parliament (MPs) and 'redistributive change' that favours MPs who belong to the majority group (Tsebelis 1990).
Institutional adjustments of parliaments to international developments are widely accepted to belong to the former (M€ uller and Sieberer 2014). This also applies to the most crucial international development in recent decades affecting European parliamentary democracies, namely European integration (Katz and Wessels 1999;Norton 1996). Parliaments of EU member states responded to European integration and installed oversight institutions, such as information rights and European affairs committees, that provide parliamentarians with the necessary means to keep an eye on EU policy making.
Scholars widely agree that little to no partisan considerations exist in EU oversight institutions. I present and empirically test an argument that challenges this common notion. My argument builds on the idea that EU oversight institutions are not only the result of domestic factors but are also affected by other EU oversight institutions abroad (Auel et al. 2015;Buzog any 2013;Karlas 2012). Because of uncertainty about their institutional response to new challenges resulting from authority transfers from the national level to the EU level, decision makers in national parliaments draw inspiration from national parliaments of other member states.
Thus, national parliaments observe each other and take information about institutional choices of other parliaments into account when they set up or reform their own EU oversight institution. Recent studies examine inter-parliamentary diffusion processes and find that EU member state parliaments most likely learn from best practices and culturally similar countries (Bormann and Winzen 2016;Buzog any 2013;Jungar 2010). While these studies contribute greatly to our understanding of institutional change in parliament, they do not consider that inter-parliamentary diffusion processes might entail partisan characteristics. In other words, majority groups in parliament that ultimately decide about EU oversight institutions might learn from abroad in a biased way, so that they do not look at the country that has the best parliamentary EU oversight institution but consider information from a country that has a majority group with similar political preferences. Such cases of biased diffusion are well known from public policy scholarship showing that partisan preferences of decision makers affect diffusion outcomes (Butler et al. 2017;Gilardi 2010;Grossback et al. 2004).
Which preferences do majority parties use to decide which other EU oversight institutions to draw on? Political parties differ in their support for centralisation in political and administrative procedures (Toubeau and Wagner 2015). Closely related, parties differ in their constitutional preferences (Winzen 2017). These preferences indicate parties' desired institutional framework for policy making as well as their preferred levels of state centralisation. In the context of EU oversight institutions, these preferences are key to understanding domestic institutional adaptation to authority transfers to the EU more generally and the involvement of national parliaments in EU affairs in particular. I suggest that these preferences are decisive for the diffusion of EU oversight institutions. To find relief from uncertainty as to how to respond to authority transfers from the domestic level to the supranational level, majority parties in national parliaments turn to previous decisions of their international peers with similar constitutional preferences.
I compile a time-series cross-sectional (TSCS) dataset, including a measurement of EU oversight institutions used as a dependent variable. I apply spatial regression models and consider both instantaneous and time-lagged spatial effects. I find that a parliament is more likely to emulate the EU oversight institution of another parliament if their majority parties have similar constitutional preferences. Thus, political views of majority groups affect EU oversight institutions installed in parliament. Yet, because of uncertainty about how to respond to authority transfers to the EU level, majority groups tend not to push for their preferred oversight institutions in a direct way but through partisan emulation from abroad. This research has important implications for our understanding of institutional change in parliament. It identifies a pathway of inter-parliamentary diffusion that challenges the common notion that parliaments' institutional responses to international developments are inherently efficient. Through diffusion, majority parties install EU oversight institutions that represent their constitutional preferences, the result being that these institutions represent the interest of MPs who belong to the majority group and not all MPs in parliament. As a result, if we want fully to understand the nature of institutional change in parliament, we have to consider factors that go beyond the domestic level.

EU oversight institutions
The increased impact of European integration on domestic politics created a demand for national parliamentary oversight of EU decision making. Regarding the institutional aspect of this demand, all EU member state parliaments responded in a similar way. They introduced information rights and installed European Affairs Committees (EACs). Information rights can take the form of government memorandums. Several EU member state parliaments receive such explanatory documents from the government providing background information about a given EU legislative proposal and the government's intended voting behaviour. However, in most parliaments, EACs are at the heart of national parliamentary EU oversight. Committee members confer EU proposals, hear ministers, and vote on resolutions and mandates. For example, in January 2018, the EAC of the Riigikogu (Parliament of Estonia) issued its support of the resettlement of 80 persons from Turkey to Estonia as part of the EU's 2016 arrangement with Turkey. While this is an example of ex post scrutiny, in some EU member states, the government is required to seek a negotiation mandate from the EAC before a significant decision is taken in the Council of the European Union. An example of this is Denmark, where the finance minister sought a mandate for Denmark to join the Euro Plus Pact prior to the European Council meeting in March 2011. In sum, parliamentary EU oversight institutions comprise a wide range of rules and procedures, mainly within EACs, that support parliamentarians in scrutinising EU policy making.
A large body of literature compares the organisation of EU oversight institutions across EU member states resulting in rankings of national parliamentary strength in EU affairs (Bergman 1997;Bergman et al. 2003;Hefftler et al. 2015;Maurer and Wessels 2001;Norton 1996;Raunio 2005;Winzen 2012). Scholars draw on different factors to explain variation between countries, including public opinion, political culture, general parliamentary strength, party preferences, and the salience of the EU in the domestic political discourse (Hamerly 2012;Winzen 2010). Two factors are robustly associated with strong EU oversight institutions. First, generally powerful legislatures are more likely to have strong EU oversight institutions. Second, the higher the anti-EU sentiment among the public, the stronger are the EU oversight institutions in parliament. Various measurements of EU oversight institutions exist in the literature. The most comprehensive one takes access to and processing of information and mandating power into account (Winzen 2012(Winzen , 2013. 1 A parliament is considered to be strong if it has relevant information about EU proposals, enough time and manpower to process this information, and an opportunity to influence the government's negotiation position. Figure 1 presents the average level of EU oversight institutions . 2 The figure shows that over time national parliaments became more powerful examiners of EU policy making. This is in line with research emphasising that national parliaments strengthened their EU oversight institutions over the course of European integration (Blom-Hansen and Olsen 2015;Raunio 2009). In addition, it seems that new EU member states raised the average level of EAC strength, as we observe significant increases in years that mark accessions to the EU. Figure 2 gives a more detailed picture. The three panels represent the level of EU oversight institutions in 1990EU oversight institutions in , 2000EU oversight institutions in , and 2010. Colour shades from light grey to dark grey indicate ascending levels of strength. The figure shows that most oversight institutions of EU member state parliaments became stronger over time. 3 Furthermore, it shows that new EU member states from Central and Eastern Europe (CEE) predominantly established strong EU oversight institutions. Estonia, Lithuania, and Latvia score particularly high. Cyprus and Poland are the only two member states from this cohort having very weak oversight institutions at the time of accession, with values below one.

Inter-parliamentary diffusion
Several scholars attribute the development of national parliamentary EU oversight institutions to international exchange of information and experiences between parliaments through inter-parliamentary networks such as COSAC (Auel et al. 2015;Buzog any 2013;Karlas 2012;Raunio 2011). 4 COSAC is a biannual conference that brings together national delegations  from EU member state parliaments to discuss current EU matters. The meetings allow MPs to exchange views and learn about EU oversight institutions of other parliaments. For example, there is anecdotal evidence that some of the new EU members from CEE countries used the knowledge exchange between COSAC participants to learn about the EAC design of experienced parliaments such as Denmark (Buzog any 2013; Jungar 2010; Raunio 2009). 5 The first comparative study that empirically tests the claim of diffusion of EU oversight institutions finds no support for learning from and emulation of strong northern parliaments (Bormann and Winzen 2016). Instead, the authors find that EU oversight institutions diffuse across EU member states that are culturally similar. 6 The study offers an important starting point, but I suggest expanding the study of diffusion of EU oversight institutions in two ways. First, demand for inspiration from abroad is not only sought by young democracies but is universal across EU member states. Because of drastic authority transfers from the domestic level to the EU level with regard to both decision-making power and policy degree, uncertainty about the design of functional EU oversight institutions concerns all national parliaments to a similar extent. Second, the actors deciding about EU oversight institutions are partisan. Generally, decisions about parliamentary institutions are made by majority groups in parliament that want to enforce their political ideas and preferences. These usually contrast with what opposition parties want. In the following, I develop an argument that shows how political preferences of majority parties characterise the international diffusion of EU oversight institutions.

Institutional change through partisan emulation
Institutional rules crucially determine who holds power in parliament. Rules affect who is winning votes, influencing policy, and securing office (D€ oring 2001;M€ uller and Sieberer 2014;Strøm et al. 2003). As a result, political party groups in parliament usually aim to set and change rules in such a way that they serve their own interests at the expense of other party groups. Such institutional changes reflecting partisan interests are commonly known as 'redistributive changes' (Tsebelis 1990). Yet not all institutional rule changes are contested. This applies to rule changes that benefit the interests of all MPs and party groups to the same extent. The literature refers to such rule changes as 'efficient changes'. Responses to international developments and technological change usually fall under that category (M€ uller and Sieberer 2014). The most significant example of a response to an international development is presumably the institutional response of national parliaments to European integration, which introduced EU oversight institutions (Hefftler et al. 2015;Katz and Wessels 1999;Norton 1996;O'Brennan and Raunio 2007).
As the described theoretical classification would suggest, a review of the literature points to the non-partisan nature of parliamentary rule changes introducing EU oversight institutions. In short, party preferences do not explain variation in EU oversight institutions (Winzen 2013). A common explanation for this is that mainstream parties have no interest in making the EU a salient political topic in parliament and therefore decide not to pursue the topic further. While this is certainly the case for general questions about the EU, the explanation is not hugely satisfying in the case of EU oversight institutions. These institutions are far from being discussed in the media and carry no electoral incentives. Thus, from the perspective of a majority group, it would be harmless to install the preferred institutional design.
Might it be the case that parties do not differ in their views about EU oversight institutions? This seems unrealistic given that parties differ in the positions that are used to decide about political institutions related to the EUtheir constitutional preferences (Winzen 2017). These preferences carry parties' ideal conception of the political order and its institutional design. Constitutional preferences reflect opinions about classical divides between centralisation and decentralisation or unitary and federal states. Party groups that favour centralisation and unitary political systems tend to be sceptical toward transfers of power to federal actors. When in power, such parties might initiate significant institutional changes that not only decide who holds power but also how the parties in power are held to account for their actions. This implies that institutional choices about state organisation are not merely functional responses to challenges but consumedly political (Marks and Hooghe 2000). As a result, knowing the mindset of parties is necessary for understanding the processes of institutional change that shape the territorial distribution of authority. A growing body of literature is dedicated to describing and explaining party preferences regarding constitutional preferences and questions about decentralisation (Toubeau and Wagner 2015;Winzen 2017).
I suggest that parties' preferred distribution of power between national parliaments and EU institutions is likely to be explained by parties' constitutional preferences. A party that favours centralisation is likely to prefer national parliaments with strong EU oversight institutions so that it can keep an eye on what is decided at the federal level. A party that favours decentralisation is likely to prefer national parliaments with weak EU oversight institutions so that federal actors are free in their decision making. 7 Apart from the question about the source of parties' preferences for EU oversight institutions, there is also anecdotal evidence that party groups within the same parliament differ in their preferences for EU oversight institutions. When the Austrian Nationalrat in 2011 voted for legislation that changed parliament's information rights in EU affairs, two major opposition parties declined their support. In the ensuing debate, the parties expressed their dissatisfaction with the design of the EU oversight reform. Now, why do the differences between parties in their preferred EU oversight institution not become visible in previous research? I argue that the reason why partisan differences remain undiscovered is that we have not been searching for them in the right place. Decisions about EU oversight institutions are not only affected by domestic factors but also by factors from abroad. As mentioned above, national parliaments consider information about institutional choices of other national parliaments. The reason is that parties face uncertainty about what the most adequate response to authority transfer may be. From the early 1990s onwards, the powers of the EU were drastically increased, with regard to both supranational decision making and the degree of policies decided at the EU level. From the perspective of parties in national parliaments, this situation created uncertainty about how to deal with these significant changes.
One strategy would be to rely on information from the domestic level (e.g. other oversight institutions). However, uncertainty about whether previous institutional choices are appropriate remains because procedures that work in domestic affairs might not work in EU affairs. Therefore, majority groups keep a lookout for what other national parliaments are doing. The critical task that they face is the decision on which foreign parliaments to turn to. An obvious choice would be to observe the EAC design of the closest neighbours. However, the notion of geographical proximity as a defining condition of diffusion is increasingly challenged. As a result, scholars focus on theoretical linkages that highlight cultural, economic, and ideological similarities between actors (Beck et al. 2006;B€ ohmelt et al. 2016;Butler et al. 2017;Gilardi 2010;Grossback et al. 2004;Shipan and Volden 2012). Similarly, I suggest that majority parties turn to actors with similar preferences. Again, the constitutional preferences of parties come into play. As described above, majority groups in parliament favour certain institutional responses to European integration over others based on their constitutional preferences. However, because of uncertainty, they do not push for their preferred institutions directly but draw on information from other national parliaments abroad. As institutional choices are political, majority parties are not interested in the most successful and best EU oversight institutions, but they want to know what their international peers with similar constitutional preferences have done. Knowing about their institutional choice produces relief regarding how to respond to authority transfer and, at the same time, assurance that information from abroad builds on similar constitutional preferences. As my proposed diffusion mechanism is based on partisan preferences, it is best described by emulation. In marked contrast to learning, emulation occurs irrespective of the success or failure of EU oversight institutions elsewhere. Instead, it takes place when actors seek to adapt their behaviour towards a group with similar opinions. In sum, I argue that the diffusion of EU oversight institutions is a partisan process in which constitutional preferences of majority groups in parliament condition how information from abroad is used. This motivates the following hypothesis: Parliaments are more likely to emulate EU oversight institutions of other parliaments if their majority groups have similar constitutional preferences.

Research design
To test my hypothesis about the diffusion of parliamentary EU oversight institutions, I rely on spatial-econometric analysis (Anselin 1988;Darmofal 2015;Franzese and Hayes 2008;LeSage and Kelley Pace 1998). In the following, I introduce my data and outline my spatial modelling strategy.

Data
I compiled a dataset including 23 EU member states for the period from 1992 to 2010. 8 The time period is chosen to capture the exchange between national parliaments in EU affairs beginning in the early 1990s (Bormann and Winzen 2016;Buzog any 2013). My unit of analysis is at the country-year level. The dependent variable, the strength of parliamentary EU oversight institution, was already introduced in the section about EU oversight institutions. The variable is standardised (bounded at 0 and 2) and composed of two parts. The first part considers the strength of EACs regarding their access to information and the processing thereof. This includes a parliament's level of access to documents and government memorandums, status of EU committees and involvement of specialised committees. The second part indicates the power of EACs to constrain government. It considers scrutiny reserves and the right to issue mandates (Winzen 2012(Winzen , 2013. A scrutiny reserve stipulates that government ministers may not agree to a proposal, either formally or informally, as long as parliament has not completed its examination of the proposal. I understand the diffusion of EU oversight institutions to be complementary to domestic-level factors. Therefore, I consider several variables measured at the domestic level. These include government parties' constitutional preferences. The inclusion of this variable considers the direct association between parties' constitutional preferences and the strength of EU oversight institutions. The data on government parties' constitutional preferences stem from Seki and Williams (2014). Their data aggregate the Manifesto Project dataset and construct annual positions of government parties weighted by their percentage of seats in government for all available policy codes. To identify constitutional preferences, I use quasi-sentences (units of manifesto text) on centralisation (per302) and decentralisation (per301) in parties' electoral manifestos. Quasi-sentences that fall into the category of centralisation express opposition to political decision making at other political levels and support centralisation in political and administrative procedures. In contrast, statements in manifestos annotated to decentralisation reflect support for federalism or decentralisation of political power.
In order to identify the constitutional preference of majority parties for a given country-year unit, I subtract the percentage of quasi-sentences about decentralisation from the percentage of quasi-sentences about centralisation. In sum, this procedure gives a government's overall constitutional preference, where positive values indicate that parties favour centralisation.
In addition, I include public opinion towards the EU, the strength of other oversight institutions in parliament, and the depth of European integration. To measure the annual level of public opinion towards the EU, I rely on the 'European mood' (Guinaudeau and Schnatterer 2017). The variable combines several indicators that capture general and specific support for European integration. The original European mood variable is a bi-annual measure. I therefore compute the annual mean for each country-year observation. The final variable indicates the overall level of EU support and ranges from 37.80 (representing the most negative 'European It is important to note that my dependent variable, the strength of EU oversight institutions, concerns EU-specific dimensions only and is therefore independent of the criteria that define the strength of other parliamentary committees. This implies that no material used to measure the power of parliamentary committees is used to measure the strength of parliamentary EU oversight institutions. Furthermore, related literature emphasises that EU oversight institutions developed in parallel to the depth of European integration. I therefore include the average level of EU authority across policy areas. I rely on the same measure as Bormann and Winzen (2016). The variable ranges from 1.8 to 3.9 and is the same for all EU member states. Please consult Table A1 in the online appendix for descriptive statistics.

Spatial modelling
Before proceeding with the modelling of the data-generating process, I conducted tests to account for serial autocorrelation and stationarity. Testing for serial correlation reveals that the null hypothesis of no serial correlation is strongly rejected. In fact, my dependent variable is heavily determined by its levels from the year before. I decided to use lagged response models that treat one-year lagged dependent variables (LDV) as covariates to account for serial correlation (Wilkins 2017). Because of this modelling strategy, I effectively predicted short-term changes in the response variable. In addition, I tested if the time series in question are stationary. An augmented Dickey-Fuller test provides support for the alternative hypothesis of stationarity.
My hypothesis suggests interdependence between EU member states based on the constitutional preferences of majority parties. An important matter when modelling dependence between units is to decide how values of other units enter the model. 9 I rely on spatial lag models because my theory outlines a mechanism for the diffusion of the dependent variable across units of analysis (Beck et al. 2006).
Apart from model specification, it is necessary to decide how to calculate spatial effect estimates. I follow recent contributions in the literature and apply spatial maximum likelihood (S-ML) models that have proved to perform well with regard to bias, efficiency, and standard error accuracy compared to non-spatial ordinary least squares (OLS), spatial OLS (S-OLS), and spatial two-stage least squares (S-2SLS) models Hays 2007, 2008). Another relevant decision to take is whether response values of units immediately affect other units or whether it can be assumed that it takes a time lag to affect them. Bormann and Winzen (2016) suggest that government parties consider contemporaneous institutions of other member states (y kt ), that is, the EAC of other units within the same year. I agree with the authors and estimate instantaneous spatial effects. 10 This leads to the following modelling specification: where i ¼ 1,2,. N, t ¼ 1,2,. N and k ¼ 1, 2,. N and y it is the value of the dependent variable in unit i at time t: X k W ikt P k W ikt y kt is a row-standardised spatial lag variable composed of a spatial weights matrix W (N Â N Â T) that captures the relative connectivity of units and a matrix (N Â T) of the dependent variable of other units k; t: The term y it À1 is the one-year temporally lagged dependent variable of unit i; X it is a vector of unit-specific variables, and e it is an identically and independently distributed error process. The spatial autoregression parameter q states the estimated effect of spatial dependence. My decision to row standardise the spatial weights matrix implies a homogeneous total exposure to spatial stimulus across all units. For each unit i; the sum of connectivities to all units k equals 1 (Neumayer and Pl€ umper 2016). 11 In addition, all models include region and period fixed effects. This helps to rule out the possibility that what appears to be a diffusion process across EU member states is actually the result of common exposure. In addition, it reduces potential omitted variable bias. 12 In sum, obtaining significant results for my spatial effect variable while including period and regional fixed effects and a LDV will increase the confidence in true diffusion effects across member states because the LDV already captures prior spatial effects (Beck et al. 2006). The following subsection introduces the operationalisation of the spatial weights matrix.

Defining spatial lags
Connectivity between my units of analysis is based on government parties' constitutional preferences. I therefore need a spatial lag variable that carries information about the dependent variable weighted by the distance in constitutional preferences between member states' majority parties. The connectivity matrix of this spatial lag variable is defined so that each element W i;k;t receives a value of the absolute distance in constitutional preferences between majority parties (i.e. government parties) in country i and k: The variable indicating constitutional preference of majority parties was already described in the data section. Empirically observed distances in constitutional preferences between member states' government parties range from 0.00 between Finland and Sweden in 1996 to 30.83 between Austria and Italy in 2009. The mean distance between member states' majority parties is 2.58. To create spatial lag variables, the connectivity matrix is multiplied with the dependent variable and time-lagged dependent variable, respectively. Descriptive information about the spatial lag variables is to be found in the online appendix (Table A1).

Results
The two models testing my hypothesis are presented in Table 1, columns 2 and 3. Column 2 shows that the coefficient of the spatial lag variable q (Constitutional preference) is negatively signed and statistically significant. This result supports my hypothesis. EU member state parliaments are less likely to emulate the EU oversight institutions of other parliaments whose majority parties have dissimilar constitutional preferences. If all EU member state parliaments with an average absolute distance increase their EU oversight institutions by one, this would lead a focal parliament to decrease its EU oversight institution by 0.58. Model comparisons based on likelihood ratio tests and Akaike information criteria show that the model accounting for dependences between units fits the data better than the model with domestic-level factors only. This is a sign that the spatial model is superior and that the assumption of spatial independence across member states tends to be inaccurate. Hence, the spatial lag model based on distance in government parties' constitutional preferences adds important information to explain changes in the strength of EU oversight institutions. In addition, results from a spatial lag model with a timelagged spatial variable are presented in column 3. 13 As mentioned, diffusion of EU oversight institutions may take time. Therefore, the model considers that parliaments emulate EU oversight institutions from the past year instead of contemporaneous institutions. The spatial lag variable based on absolute distances between parties' constitutional preferences is again negatively signed and statistically significant. If all EU member state parliaments with an average absolute distance had increased their EU oversight institutions by one in the previous year, this would lead a focal parliament to decrease its EU oversight institution by 0.3. Model performance tests show that the model with the time-lagged spatial lag variable fits the data even better than the model considering instantaneous spatial effects. In sum, the two models (together with additional results in the online appendix) provide robust support for my hypothesis.
In addition, I present results from a model that regresses the response variable on the domestic-level variables described earlier. The model applies a generalised least square fit by maximum likelihood estimation, which means that errors are allowed to be correlated or have unequal variances. It includes a lagged response as well as region and period fixed effects. Results are presented in column 1 in Table 1. It shows that the European mood indicator is negatively associated with the strength of EU oversight institutions and thereby supports previous findings. To further substantiate the role of public opinion towards the EU, I rely on statistical simulation techniques (King et al. 2000). I re-estimated the parameters in the regression equations and drew 1000 sets of simulated coefficients from their posterior distribution. Then, I fixed the European mood indicator at two values (43.7, i.e. the 5% percentile, and 68.9, i.e. the 95% percentile), holding all other variables at their means, and calculated 1000 expected values of the strength of EU oversight institutions. Figure 3 presents the distributions of the simulated expected values. It clearly shows that EU oversight institutions are stronger when the public mood towards the EU is more negative. The means of the two distributions are different from each other at conventional levels of statistical significance. I also find support that the strength of EU oversight institutions is positively associated with the strength of other parliamentary committees. The coefficient for the variable measuring the depth of European integration is positive but does not reach conventional levels of statistical significance. The model also provides evidence that constitutional preferences of majority parties in parliament have no direct influence on the strength of EU oversight institutions. I now return to the interpretation of the spatial lag variables.
In models 2 and 3, the effect of dependence between EU member state parliaments is only described by the spatial parameter q; which reflects changes in the response variable based on changes in the dependent variable of all neighbours given average levels of distances in constitutional preferences. As such, the exact interpretation of the parameter is difficult to understand as it does not provide information about specific diffusion patterns. Therefore, substantial relations between units of analysis should be investigated to better understand the mechanism at work (Darmofal 2015). To do so, I analyse the impact of changes in the dependent variable of one EU member state parliament on the strength of EU oversight institutions in other member states. In spatial regression analysis, such an impact is called equilibrium impacts (Ward and Skrede Gleditsch 2008).
My illustration focuses on changes in the dependent variable. I manipulate the variable in Denmark. The country is chosen because of its generally high level of EU oversight institutions, which means that further increases in EU oversight institutions are probably more likely in Denmark than in other countries. I estimate differences in predicted values of the strength of EU oversight institutions between model 2 in Table  1 and a model in which the value of EU oversight strength is changed from its empirically observed value (1.36) to the maximum value 2 in 1994. The manipulation thereby signifies a potential increase in EU oversight strength that is even higher than the actual observed change after agreement to the Treaty of Maastricht. Table 2 shows how the manipulation in Denmark would diffuse across other countries. The difference in predicted values stems from models based on instantaneous spatial effects. This implies that the output values are equilibrium impacts within the same year. The manipulation results in positive differences in all EU member states. However, we see that some parliaments responded more strongly than others. The results are based on the differences in constitutional preferences between countries but also differences in the response variable, as parliaments with similar EU oversight institutions react less strongly. In sum, the manipulation study gives substance to the theoretical mechanism at work and shows that the design of parliamentary EU oversight institutions diffuses across member states whose government parties have similar constitutional preferences. Although predicted changes in EU oversight strength are rather small, we should keep in mind that the dependent variable is bounded at 0 and 2. In addition, the model considers contemporaneous effects in one single year after a manipulation of EU oversight institutions in one single EU member state.

Conclusion
This study extends recent research that aims to understand international diffusion of EU oversight institutions. I develop a new theoretical argument that addresses the importance of political preferences of majority groups in parliament. More precisely, I say that when majority groups of EU member state parliaments look for inspiration from abroad to design and reform EU oversight institutions, they consider similarities in constitutional preferences between them and the government parties of the country to emulate. My empirical analysis based on spatial regression analyses supports my proposed diffusion process. I find that EU oversight institutions diffuse across member states whose majority parties have similar constitutional preferences. My finding sheds new light on the mechanism behind the diffusion of EU oversight institutions. Recent research focused on rather broad concepts (e.g. learning from cultural peers or best practices) without considering that the actors who decide about EU oversight institutions have distinct political preferences that are likely to affect their institutional choices. Beyond EU oversight institutions, my research challenges the widespread notion that institutional responses of parliaments to international challenges are inherently efficient. It follows from the fact that majority groups rely on their constitutional preferences when they emulate the institutional design of other parliaments that parliamentary institutions represent partisan preferences. Yet this would go unnoticed if we would focus on domestic-level factors only to explain institutional change in parliament. Given my contribution to the research area of institutional change in parliament, there are several important questions to explore for future research. Most obviously, as this study focuses on EU oversight institutions, it would be worthwhile to investigate whether other institutional responses related to international events represent partisan preferences. These could include the rules and procedures of parliamentary institutions that deal with issues such as digitalisation, globalisation, and foreign affairs, that is, parliamentary institutions that engage in policy issues affected by increasing levels of international cooperation. Such an approach would allow expansion of the topic into other geographic areas other than the EU and integrate it into broader programmes that aim to map and explain parliamentary rule change (Sieberer et al. 2016). In addition, it would be interesting to explore the conditions under which institutional responses of parliaments follow the mechanism of partisan emulation and when they follow non-partisan strategies (e.g. learning from best practice). Bringing in the power of opposition parties might be a fruitful endeavour. In addition, recent developments in methodology, that aim to combine spatial analysis with case-study research, are promising tools for researchers to answer these questions (Harbers and Ingram 2017 Figure A1 in the online appendix. 4. COSAC is a French acronym and stands for Conference of Community and European Affairs Committees of Parliaments of the European Union. 5. Likewise, decisions by national parliaments about EU policies as part of the 'early warning system' are not made independent of each other (Malang et al. 2017). 6. Cultural similarity is operationalised on the basis of central tendencies in the religious orientation of the majority population (majority Catholic, majority Protestant, majority Orthodox). 7. Scholars also rely on parties' EU support to explain the relationship between political parties and EU oversight institutions. While there is certainly an overlap between parties' constitutional preferences and EU support, the latter is solely directed at processes at the EU level (EU institutions and policies). As such, parties' preferences for domestic institutional responses to authority transfers to the EU level remain unreported. Constitutional preferences, on the contrary, capture broader questions about parties' desired institutional framework for policy making as well as their preferred levels of centralisation, which can be applied to cases of domestic institutional responses to European integration. 8. As mentioned before, Romania is missing because no data on the dependent variables are available. In addition, Cyprus, Bulgaria, and Malta are missing because data on independent variables (constitutional preferences and strength of other parliamentary committees) are not available. My data cover EU member states from 1992, with additional data becoming available with the accession of new member states. 9. One can differentiate between several common approaches: (1) spatial error models that include dependence in a second error term in addition to the independent error component; (2) spatial X-models that identify dependence in the weighted values of one or more independent variables; and (3) spatial lag models that regress the dependent variable on the weighted values of the dependent variable of all other units (Franzese and Hayes 2008;Pl€ umper and Neumeyer 2010). 10. However, it could also be argued that government parties need time to react to institutional choices of other member states. Even though parliaments have much more contact and exchange more information than ever before, it is difficult to say whether government parties always know what is happening in other parliaments and anticipate their reform steps. In addition, the implementation and reform of EU oversight institutions might be a tedious process. Therefore, I also consider models that allow for a oneyear time lag of the spatially lagged variable. From a methodological point of view, this erases the simultaneity problem of S-OLS models because the spatial lag is time lagged and therefore does not carry the same information as the dependent variable on the left-hand side of the equation (Beck et al. 2006;Franzese and Hayes 2008). 11. As described in the next subsection, the spatial weights matrix carrying absolute differences in parties' constitutional preferences connect all units i with the same number of units k: In this case, row-standardisation is convenient because it gives the same metric or units to the dependent variable and the spatial lag (Neumayer and Pl€ umper 2016). However, more importantly, the decision to row-standardise the weights matrix is theoretically justified. First, all EU member states have the same number of connections to other member states and observe the same pool of parliaments during international cooperation. Second, there is no argument suggesting that countries that have overall smaller absolute distances in constitutional preferences should receive a weaker spatial stimulus than countries with overall larger absolute differences in constitutional preferences. For a comprehensive discussion of row-standardisation of weights matrices, see and Pl€ umper (2012 2016). 12. Member states belong to one of the regions: North, South, Central, West.
The time period from 1992 to 2010 is divided into the following subperiods: < ¼1994 , 1995-2003, 2004-2006, > ¼2007. Opting for regional and period fixed effects implies that, similar to Bormann and Winzen (2016), I do not include unit-fixed effects, as I believe that not only changes but also different levels in EU oversight institutions of other member states matter. However, in the online appendix, regression results from models that replace regional fixed effects with country fixed effects are presented (Table  A2). The findings and interpretation regarding the spatial lagged variables are robust to this alternative model specification. 13. Time-lagged spatial effect models can be expressed in the following way: