American Political Science Review (2018)112.4,1090-1095 doi:10.1017/S0003055418000461 American Political Science Association 2018 Letter Quantifying Political Relationships SIMON WESCHLE Syracuse University 'n this article,I introduce a method that uses large-scale event data and latent factor network models to provide a new comparative measure of cooperation and conflict in public relationships among politicians,nonpartisan political actors,and societal actors.The approach has a number of advan- tages over existing techniques:It captures public relationships in a multitude of venues on a continuous basis,incorporates both partisan and nonpartisan actors,allows quantifying the relationship between any pair of actors,reflects that communication is not unidirectional but rather a back and forth,and can be applied to a large number of countries over time.I apply the method to 13 Western European countries from 2001 to 2014 and demonstrate that party relationships are determined by coalition status as well as policy differences.The measure is publicly available and can be incorporated into standard research designs. here is a large discrepancy between the impor- tion of machine-coded news reports that chronicle tens tance attributed to the public communication of of thousands of interactions from many venues among political actors and the (lack of)empirical anal- hundreds of politicians,nonpartisan political actors ysis devoted to it.The way political elites interact with (e.g.,bureaucracy,police,or judiciary),and societal ac- each other as well as with members of society may tors(e.g.,citizens,unions,companies,religious groups) shape whether voters like or dislike them,how citizens I then infer the underlying networks that give rise to perceive policy positions,or the degree to which they these interactions by estimating latent factor network feel represented by the political system(Fenno 1978; models (Hoff and Ward 2004;Hoff 2005,2015).They Mansbridge 2003:Strom 2008).However,a lack of data place all actors in a low-dimensional "social space," has hampered the empirical study of the causes and where those who have a cooperative relationship are consequences of public communication involving polit- placed in the same direction,while those who have a ical elites.This has begun to change recently due to ad- conflictual relationship are placed in different direc- vances in automated text analysis methods.The focus tions.From the latent positions,it is possible to com- so far has been on examining parliamentary speeches pute scores that reflect the relationship of any pair of (e.g.,Martin and Vanberg 2008;Eggers and Spirling actors in the data. 是 2014;Herzog and Benoit 2015;Proksch and Slapin This approach to quantify political relationships has 2015;Lauderdale and Herzog 2016)and politicians' a number of advantages over existing techniques.First, press releases (e.g.,Grimmer 2013;Grimmer,West- it is not restricted to a single venue,but captures in- wood,and Messing 2015;Sagarzazu and Kluver 2017). teractions in many of them:press releases,parliamen- These approaches have yielded valuable insights but tary speeches,interviews,campaign events,and so on. have focused exclusively on the unidirectional commu Second,instead of examining only one-sided commu- nication by politicians to constituents in a single venue. nication by politicians,it incorporates information on and typically in a single country. communication from and to nonpartisan political and In this article,I introduce a method that uses large- societal actors as well.The approach also makes it scale event data and latent factor network models to possible to locate political and societal actors in a provide a new comparative measure of the public re- common space.Importantly,because the latent fac- lationships among politicians,nonpartisan political ac- tor model captures third-order network dependencies tors,and societal actors.Instead of relying on source (e.g.,a friend of a friend is a friend),the relation be- data from a single setting,I use an extensive collec- tween two actors can be inferred even ifno direct inter- action between them is reported.Finally,the approach can be estimated for a large number of countries over many years. Simon Weschle is an Assistant Professor,Department of Political This new measure opens up the possibility to study Science,Maxwell School of Citizenship and Public Affairs,Syra cuse University,100 Eggers Hall,Syracuse,NY 13244 (swweschl@ the causes and consequences of cooperative or con- maxwell.syr.edu). flictual relationships between partisan political,non- For helpful comments and advice,I thank the APSR editor Ken- partisan political,and societal actors in a comparative neth Benoit,four anonymous reviewers,James Adams,Ben Bar- manner.The relationship scores are publicly available ber,Pablo Fernandez-Vazquez.Sebastian Lavezzolo,Michael Ward and can be used by researchers to answer questions in and Christopher Wlezien.Replication materials are available at the American Political Science Review Dataverse:https://doi.org/ areas such as coalition politics,polarization,or demo- 10.7910/DVN/AOTVAU.The Ouantified Political Relationships data cratic representation and satisfaction.In this article,I are available at www.simonweschle.com/data. describe the details of the approach and demonstrate Received:February 8,2017;revised:January 8,2018;accepted:July 5, its value by applying it to 13 Western European coun- 2018.First published online:August 23,2018. tries from 2001 to 2014. 1090
American Political Science Review (2018) 112, 4, 1090–1095 doi:10.1017/S0003055418000461 © American Political Science Association 2018 Letter Quantifying Political Relationships SIMON WESCHLE Syracuse University I n this article, I introduce a method that uses large-scale event data and latent factor network models to provide a new comparative measure of cooperation and conflict in public relationships among politicians, nonpartisan political actors, and societal actors. The approach has a number of advantages over existing techniques: It captures public relationships in a multitude of venues on a continuous basis, incorporates both partisan and nonpartisan actors, allows quantifying the relationship between any pair of actors, reflects that communication is not unidirectional but rather a back and forth, and can be applied to a large number of countries over time. I apply the method to 13 Western European countries from 2001 to 2014 and demonstrate that party relationships are determined by coalition status as well as policy differences. The measure is publicly available and can be incorporated into standard research designs. There is a large discrepancy between the importance attributed to the public communication of political actors and the (lack of) empirical analysis devoted to it. The way political elites interact with each other as well as with members of society may shape whether voters like or dislike them, how citizens perceive policy positions, or the degree to which they feel represented by the political system (Fenno 1978; Mansbridge 2003; Strom 2008).However, a lack of data has hampered the empirical study of the causes and consequences of public communication involving political elites. This has begun to change recently due to advances in automated text analysis methods. The focus so far has been on examining parliamentary speeches (e.g., Martin and Vanberg 2008; Eggers and Spirling 2014; Herzog and Benoit 2015; Proksch and Slapin 2015; Lauderdale and Herzog 2016) and politicians’ press releases (e.g., Grimmer 2013; Grimmer, Westwood, and Messing 2015; Sagarzazu and Klüver 2017). These approaches have yielded valuable insights but have focused exclusively on the unidirectional communication by politicians to constituents in a single venue, and typically in a single country. In this article, I introduce a method that uses largescale event data and latent factor network models to provide a new comparative measure of the public relationships among politicians, nonpartisan political actors, and societal actors. Instead of relying on source data from a single setting, I use an extensive collecSimon Weschle is an Assistant Professor, Department of Political Science, Maxwell School of Citizenship and Public Affairs, Syracuse University, 100 Eggers Hall, Syracuse, NY 13244 (swweschl@ maxwell.syr.edu). For helpful comments and advice, I thank the APSR editor Kenneth Benoit, four anonymous reviewers, James Adams, Ben Barber, Pablo Fernández-Vázquez, Sebastián Lavezzolo, Michael Ward, and Christopher Wlezien. Replication materials are available at the American Political Science Review Dataverse: https://doi.org/ 10.7910/DVN/AOTVAU. The Quantified Political Relationships data are available at www.simonweschle.com/data. Received: February 8, 2017; revised: January 8, 2018; accepted: July 5, 2018. First published online: August 23, 2018. tion of machine-coded news reports that chronicle tens of thousands of interactions from many venues among hundreds of politicians, nonpartisan political actors (e.g., bureaucracy, police, or judiciary), and societal actors (e.g., citizens, unions, companies, religious groups). I then infer the underlying networks that give rise to these interactions by estimating latent factor network models (Hoff and Ward 2004; Hoff 2005, 2015). They place all actors in a low-dimensional “social space,” where those who have a cooperative relationship are placed in the same direction, while those who have a conflictual relationship are placed in different directions. From the latent positions, it is possible to compute scores that reflect the relationship of any pair of actors in the data. This approach to quantify political relationships has a number of advantages over existing techniques. First, it is not restricted to a single venue, but captures interactions in many of them: press releases, parliamentary speeches, interviews, campaign events, and so on. Second, instead of examining only one-sided communication by politicians, it incorporates information on communication from and to nonpartisan political and societal actors as well. The approach also makes it possible to locate political and societal actors in a common space. Importantly, because the latent factor model captures third-order network dependencies (e.g., a friend of a friend is a friend), the relation between two actors can be inferred even if no direct interaction between them is reported. Finally, the approach can be estimated for a large number of countries over many years. This new measure opens up the possibility to study the causes and consequences of cooperative or conflictual relationships between partisan political, nonpartisan political, and societal actors in a comparative manner. The relationship scores are publicly available and can be used by researchers to answer questions in areas such as coalition politics, polarization, or democratic representation and satisfaction. In this article, I describe the details of the approach and demonstrate its value by applying it to 13 Western European countries from 2001 to 2014. 1090 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:56:49, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000461
Quantifying Political Relationships APPROACH TO QUANTIFY POLITICAL categories:partisan political,nonpartisan political(e.g., RELATIONSHIPS bureaucracy,police,judiciary),and societal (e.g.,cit- izens,corporations,unions,religious groups).For the Data first category.I code actors'partisan affiliations,taking party switches into account.I also assign partisan affil- The event data of interactions among political and soci- etal elites I use stem from the ICEWS project(Boschee iations to institutional actors (e.g.,head of state,min- et al.2015).ICEWS is an early warning system that istry of defense,ruling party)when they can be clearly inferred.Second,I dichotomize the event type cate- was designed in conjunction with several academic re. gories into cooperative and conflictual.For example, search teams to help U.S.policy analysts predict violent "make optimistic comment"is a cooperative category, and nonviolent political crises (for a detailed introduc- tion,see O'Brien 2013).A central idea of the project whereas "bring lawsuit against"is a conflictual one.? is that the tone of interactions between the sociopolit- This dichotomization further increases coding accu- racy.If,say,the example sentence above had been mis- ical actors of a country can help predict such crises.It classified as "use unconventional mass violence"rather 4 has therefore developed an extensive event collection than as "criticize or denounce,"it would still correctly that documents the activities of countries'political and be considered a conflictual interaction. societal elites as comprehensively as possible. The ICEWS project takes a large collection of news stories and machine-codes them into dyadic events re- Relationship Scores porting the event source,target,and type.Its source material stems from the media repositories of the The next step is to summarize this wealth of informa- Open Source Center and Factiva,which collect inter- tion on the relations among political and societal actors national and national news reports from a large num- in a way that can be easily interpreted and is amenable ber of publications.The first six sentences of each to further quantitative analysis.The key insight is that the interactions arise from networks of relations among 4 report are coded by BBN ACCENT,a natural lan- guage analysis system.It employs a number of linguis- sociopolitical actors.My approach for a measure of po- tic models that were trained using a sample corpus to litical relationships is to infer the structure of those net- extract structured information from text (for details works using a latent factor approach(Hoff and Ward see Boschee et al.2015).Consider the following sen- 2004:Hoff2005,2015) Denote the number of cooperative interactions be- tence from a 2008 report from Germany:"Economics Minister Michael Glos,of the government's conserva- tween i and j by m,and the number of conflictual in- tive CDU/CSU coalition partner,attacked a draft pro- teractions with m.The interactions between the two posal from Justice Minister Brigitte Zypries."This is m+1 machine-coded into an event described by three vari- actors can be summarized by yi=yi=In m+ ables:The source is Michael Glos,Brigitte Zypries is These are then aggregated into an n x n sociomatrix the target,and the type is "criticize or denounce."For Y.the cells of which are modeled as follows: the type,the coding scheme developed by the Conflict and Mediation Event Observation(CAMEO)project. 5795.801g containing roughly 350 categories,is used (Gerner, yij a +ai+aj +eij+u;Auj, Schrodt,and Yilmaz 2009). a1,...,an ~iid.N(0,od) The goal of the ICEWS event data is to chroni- cle the activities of countries'main sociopolitical ac- (en)~ii.d.N(0.2) tors.Events are therefore extensively screened and fil- tered to exclude historical events,those unrelated to The latent factor approach decomposes the dependent sociopolitical activities (e.g.,sports or entertainment), variable into several components.The intercept is des- and exact duplicates.Validation studies find that the ignated by a.Overall differences in the tone of inter- machine-coded information triplets were judged to be actions by actors are captured by the random effects correct in around 75 percent of cases (Boschee,Natar- a;and aj.These coefficients are larger for actors whose jan,and Weischedel 2013;Boschee et al.2015),exceed- interactions are more cooperative in general.The ran- ing the performance that is typically achieved by hu- dom effects term ei captures the correlation of ac- man coders (King and Lowe 2003).The event collec- tions between a dyadic pair of actors (e.g.,reciprocity). tion has been made publicly available (Boschee et al Finally,the remaining variance in yi is absorbed by 四 2015).In the Online Appendix,I provide extensive fur- uAui,the multiplicative effects term that captures la- ther information on the data,including technical de tent nodal characteristics. tails on the coding algorithm,descriptive statistics,a list The basic idea of the approach is to represent the of media sources,frequency tables of actors,and a dis network by giving actors positions in an unobserved cussion of potential objections and limitations. I further process the event data in two ways.First, 2 See the Online Appendix for the full list of cooperative and con- hand code every actor as belonging into one of three flictual categories. 3 The ratio is logged to make it linear,so,for example,In(7/10)and In(10/7)have the same distance to neutral. 1 http://reuters.com/article/2008/03/05/autoshow-porsche-idINL057 4 The terms in Auare estimated using an eigenvalue decomposi- 5794620080305,accessed January 5,2017 tion.See Online Appendix for details. 1091
Quantifying Political Relationships APPROACH TO QUANTIFY POLITICAL RELATIONSHIPS Data The event data of interactions among political and societal elites I use stem from the ICEWS project (Boschee et al. 2015). ICEWS is an early warning system that was designed in conjunction with several academic research teams to help U.S. policy analysts predict violent and nonviolent political crises (for a detailed introduction, see O’Brien 2013). A central idea of the project is that the tone of interactions between the sociopolitical actors of a country can help predict such crises. It has therefore developed an extensive event collection that documents the activities of countries’ political and societal elites as comprehensively as possible. The ICEWS project takes a large collection of news stories and machine-codes them into dyadic events reporting the event source, target, and type. Its source material stems from the media repositories of the Open Source Center and Factiva, which collect international and national news reports from a large number of publications. The first six sentences of each report are coded by BBN ACCENT, a natural language analysis system. It employs a number of linguistic models that were trained using a sample corpus to extract structured information from text (for details, see Boschee et al. 2015). Consider the following sentence from a 2008 report from Germany: “Economics Minister Michael Glos, of the government’s conservative CDU/CSU coalition partner, attacked a draft proposal from Justice Minister Brigitte Zypries.”1 This is machine-coded into an event described by three variables: The source is Michael Glos, Brigitte Zypries is the target, and the type is “criticize or denounce.” For the type, the coding scheme developed by the Conflict and Mediation Event Observation (CAMEO) project, containing roughly 350 categories, is used (Gerner, Schrodt, and Yilmaz 2009). The goal of the ICEWS event data is to chronicle the activities of countries’ main sociopolitical actors. Events are therefore extensively screened and filtered to exclude historical events, those unrelated to sociopolitical activities (e.g., sports or entertainment), and exact duplicates. Validation studies find that the machine-coded information triplets were judged to be correct in around 75 percent of cases (Boschee, Natarjan, and Weischedel 2013; Boschee et al. 2015), exceeding the performance that is typically achieved by human coders (King and Lowe 2003). The event collection has been made publicly available (Boschee et al. 2015). In the Online Appendix, I provide extensive further information on the data, including technical details on the coding algorithm, descriptive statistics, a list of media sources, frequency tables of actors, and a discussion of potential objections and limitations. I further process the event data in two ways. First, I hand code every actor as belonging into one of three 1 http://reuters.com/article/2008/03/05/autoshow-porsche-idINL057 5794620080305, accessed January 5, 2017. categories: partisan political, nonpartisan political (e.g., bureaucracy, police, judiciary), and societal (e.g., citizens, corporations, unions, religious groups). For the first category, I code actors’ partisan affiliations, taking party switches into account. I also assign partisan affiliations to institutional actors (e.g., head of state, ministry of defense, ruling party) when they can be clearly inferred. Second, I dichotomize the event type categories into cooperative and conflictual. For example, “make optimistic comment” is a cooperative category, whereas “bring lawsuit against” is a conflictual one.2 This dichotomization further increases coding accuracy. If, say, the example sentence above had been misclassified as “use unconventional mass violence” rather than as “criticize or denounce,” it would still correctly be considered a conflictual interaction. Relationship Scores The next step is to summarize this wealth of information on the relations among political and societal actors in a way that can be easily interpreted and is amenable to further quantitative analysis. The key insight is that the interactions arise from networks of relations among sociopolitical actors.My approach for a measure of political relationships is to infer the structure of those networks using a latent factor approach (Hoff and Ward 2004; Hoff 2005, 2015) Denote the number of cooperative interactions between i and j by m+ ij , and the number of conflictual interactions with m− ij . The interactions between the two actors can be summarized by yij = yji = ln m+ ij+1 m− ij+1 . 3 These are then aggregated into an n × n sociomatrix Y, the cells of which are modeled as follows: yij = α + ai + aj + ij + u i uj, a1,..., an ∼ i.i.d. N (0, σ2 a ), (1) {ij} ∼ i.i.d. N (0, σ2 e ). The latent factor approach decomposes the dependent variable into several components. The intercept is designated by α. Overall differences in the tone of interactions by actors are captured by the random effects ai and aj. These coefficients are larger for actors whose interactions are more cooperative in general. The random effects term ij captures the correlation of actions between a dyadic pair of actors (e.g., reciprocity). Finally, the remaining variance in yij is absorbed by u i uj, the multiplicative effects term that captures latent nodal characteristics.4 The basic idea of the approach is to represent the network by giving actors positions in an unobserved 2 See the Online Appendix for the full list of cooperative and conflictual categories. 3 The ratio is logged to make it linear, so, for example, ln(7/10) and ln(10/7) have the same distance to neutral. 4 The terms in u i uj are estimated using an eigenvalue decomposition. See Online Appendix for details. 1091 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:56:49, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000461
Simon Weschle latent space.The K-dimensional vector u;characterizes can infer how well a party i represents a societal actor i's position in that network,and u;gives the same for no matter whether they directly interact or not (see j.A is a K x K diagonal matrix of (positive)scaling Weschle 2017).On the practical side,this advantage is constants(see Hoff 2015).5 The multiplicative effects especially relevant if the data are only a partial obser- term uAu;then represents the nature of the relation- vation of the network.While the ICEWS event col- ship between i and j.It places two actors that tend to lection is one of the largest and most detailed sources be cooperative with each other,or that interact in a there is,it nevertheless can only capture a sample of the similar way with third actors,in the same direction in universe of interactions the latent social space.Actors that are likely to be in conflict with each other.or that interact in different APPLICATIONS ways with third actors,are placed in opposing direc- demonstrate the value of my approach to quantify po- tions.The term uAui does not only reflect direct inter- litical relationships by applying it to 13 Western Euro- actions between two actors,it also takes into account pean countries for the years 2001 to 2014.5 There are a their relation through others.For example,if two politi- total of 250,316 domestic events involving 3,422 actors cians both have cooperative interactions with unions, reported by 232 media sources.I estimate a separate but conflictual ones with business representatives,they one-dimensional (K =1)latent factor model for each likely have a cooperative relationship with each.They country-year,which allows me to compute yearly re- are thus placed in the same direction,even if they did lationship scores for any dyadic pair of actors that are not directly interact with each other. present in the data.?The applications here are focused To understand the multiplicative effects term intu- on the relationships between political parties,but the itively,suppose K =1,so u;and A are simply scalars data can be used to analyze nonpartisan political and ui and A.Recall that yi is positive when interactions societal actors as well. are cooperative,and negative when conflictual interac- tions dominate.If A is positive and ui and uj are both Descriptive Example:Germany positive or both negative,then uiu;will be positive, To show what the relationship scores derived from the corresponding to a cooperative relationship between i latent factor models look like in practice,Figure 1 plots and j.If one of ui and uj is negative and one is pos- them for Germany from 2001 to 2009.In this case,I itive,so the actors are located in different directions have aggregated all partisan actors in the data,so par- in the space,then uiAuj is negative.Since this is not ties are unitary actors.8 Scores involving two parties are only done for i and j,but also for i and k,i and k, and so on,two actors will be located in the same direc- displayed by black dots,dark gray dots represent scores between one partisan and one nonpartisan actor,and tion if they have mostly cooperative interactions with light gray is two nonpartisan actors. each other(net of their overall level of cooperation) To highlight the face validity of the scores,I focus on or if they are connected through mutual cooperation three party-dyads over time.Social Democrats(SPD) with third actors.The latent factors thus summarize the complex network of relations into a low-dimensional and Grune formed a coalition government until 2005. Their relationship scores during these years are con representation that is easily interpretable and can be used for quantitative analysis.In particular,Auj pro- sistently large,indicating a positive relation both in di- rect interactions as well as through third actors.After 5795.801g vides an estimated relationship score between i and i based on their positions in the latent communication 2005,their scores go toward the neutral point.The most conflictual party-dyad relation during the SPD-Grune network. Focusing on the overall relationship uAuj rather coalition is between the SPD and the largest opposition party,the Christan Democratic Union/Christian Social than simply on yii,the balance of the dyadic direct inter- Union(CDU/CSU).This changes once these two par- actions,has a number of advantages.Equation (1)fil- ters out actor-specific idiosyncrasies.Some actors tend ties form a "grand coalition"in 2005,when they be- come the party dyad with the highest scores.Their re- to be more conflictual or more cooperative in general. lationship deteriorates over time,and the CDU/CSU no matter who they are interacting with,which would becomes friendlier with the Free Democratic Party contaminate a simple count or ratio measure.The la- tent factor approach also partitions out the correlation (FDP),foreshadowing the coalition they would form after the 2009 elections.However,this is not reflected of actions between a dyadic pair of actors,such as reci- in the scores for 2009.the year in which the Great Re- procity.Furthermore,relationships are not only deter- cession dominated the agenda and all party dyads ex- mined by direct interactions,but through communica tion with others as well.Using u'Au;has the advan- hibit neutral relationships.Notice that this is also true tage of taking these third-party communications into for the scores involving one party and one nonpartisan account.This makes it possible to analyze the relation- actor,while relations for nonpartisan dyads become more polarized (for a detailed analysis see Weschle ship between two actors even if no direct interaction 2017 between them is observed.For example,political repre- sentation does not require direct communication,so we Austria,Belgium,Denmark,Finland,France,Germany,Greece Ireland,Italy,Netherlands,Portugal,Spain,United Kingdom. The values of A are negative in the (unlikely)case that nodes with See Online Appendix for details on the Bayesian estimation similar characteristics are more likely to have conflictual interactions See Online Appendix for a list of parties.Relationships scores in See Online Appendix for details. which political actors are not aggregated are available as well. 1092
Simon Weschle latent space.The K-dimensional vector ui characterizes i’s position in that network, and uj gives the same for j. is a K × K diagonal matrix of (positive) scaling constants (see Hoff 2015).5 The multiplicative effects term u i uj then represents the nature of the relationship between i and j. It places two actors that tend to be cooperative with each other, or that interact in a similar way with third actors, in the same direction in the latent social space. Actors that are likely to be in conflict with each other, or that interact in different ways with third actors, are placed in opposing directions. The term u i uj does not only reflect direct interactions between two actors, it also takes into account their relation through others. For example,if two politicians both have cooperative interactions with unions, but conflictual ones with business representatives, they likely have a cooperative relationship with each. They are thus placed in the same direction, even if they did not directly interact with each other. To understand the multiplicative effects term intuitively, suppose K = 1, so ui and are simply scalars ui and λ. Recall that yij is positive when interactions are cooperative, and negative when conflictual interactions dominate. If λ is positive and ui and uj are both positive or both negative, then uiλuj will be positive, corresponding to a cooperative relationship between i and j. If one of ui and uj is negative and one is positive, so the actors are located in different directions in the space, then uiλuj is negative. Since this is not only done for i and j, but also for i and k, j and k, and so on, two actors will be located in the same direction if they have mostly cooperative interactions with each other (net of their overall level of cooperation) or if they are connected through mutual cooperation with third actors. The latent factors thus summarize the complex network of relations into a low-dimensional representation that is easily interpretable and can be used for quantitative analysis. In particular, u i uj provides an estimated relationship score between i and j based on their positions in the latent communication network. Focusing on the overall relationship u i uj rather than simply on yij, the balance of the dyadic direct interactions, has a number of advantages. Equation (1) filters out actor-specific idiosyncrasies. Some actors tend to be more conflictual or more cooperative in general, no matter who they are interacting with, which would contaminate a simple count or ratio measure. The latent factor approach also partitions out the correlation of actions between a dyadic pair of actors, such as reciprocity. Furthermore, relationships are not only determined by direct interactions, but through communication with others as well. Using u i uj has the advantage of taking these third-party communications into account. This makes it possible to analyze the relationship between two actors even if no direct interaction between them is observed. For example, political representation does not require direct communication, so we 5 The values of are negative in the (unlikely) case that nodes with similar characteristics are more likely to have conflictual interactions. See Online Appendix for details. can infer how well a party i represents a societal actor j, no matter whether they directly interact or not (see Weschle 2017). On the practical side, this advantage is especially relevant if the data are only a partial observation of the network. While the ICEWS event collection is one of the largest and most detailed sources there is,it nevertheless can only capture a sample of the universe of interactions. APPLICATIONS I demonstrate the value of my approach to quantify political relationships by applying it to 13 Western European countries for the years 2001 to 2014.6 There are a total of 250,316 domestic events involving 3,422 actors reported by 232 media sources. I estimate a separate one-dimensional (K = 1) latent factor model for each country-year, which allows me to compute yearly relationship scores for any dyadic pair of actors that are present in the data.7 The applications here are focused on the relationships between political parties, but the data can be used to analyze nonpartisan political and societal actors as well. Descriptive Example: Germany To show what the relationship scores derived from the latent factor models look like in practice,Figure 1 plots them for Germany from 2001 to 2009. In this case, I have aggregated all partisan actors in the data, so parties are unitary actors.8 Scores involving two parties are displayed by black dots, dark gray dots represent scores between one partisan and one nonpartisan actor, and light gray is two nonpartisan actors. To highlight the face validity of the scores, I focus on three party-dyads over time. Social Democrats (SPD) and Grüne formed a coalition government until 2005. Their relationship scores during these years are consistently large, indicating a positive relation both in direct interactions as well as through third actors. After 2005, their scores go toward the neutral point.The most conflictual party-dyad relation during the SPD-Grüne coalition is between the SPD and the largest opposition party, the Christan Democratic Union/Christian Social Union (CDU/CSU). This changes once these two parties form a “grand coalition” in 2005, when they become the party dyad with the highest scores. Their relationship deteriorates over time, and the CDU/CSU becomes friendlier with the Free Democratic Party (FDP), foreshadowing the coalition they would form after the 2009 elections. However, this is not reflected in the scores for 2009, the year in which the Great Recession dominated the agenda and all party dyads exhibit neutral relationships. Notice that this is also true for the scores involving one party and one nonpartisan actor, while relations for nonpartisan dyads become more polarized (for a detailed analysis see Weschle 2017). 6 Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, Spain, United Kingdom. 7 See Online Appendix for details on the Bayesian estimation. 8 See Online Appendix for a list of parties. Relationships scores in which political actors are not aggregated are available as well. 1092 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:56:49, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000461
Quantifying Political Relationships FIGURE 1.Relationship Scores,Germany,2001-2009 4号元 CDU/CSU-SPD SPD-Griine . CDU/CSU-FDP 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year Black:two parties.Dark gray:one party,one nonpartisan actor.Light gray:two nonpartisan actors Public Relationships among Political Parties tive relations with business and conflictual ones with unions. Having shown the face validity of the measure,I To test these conjectures,I estimate a set of now demonstrate how it makes a systematic compar- models in which the dyadic relationship scores in- ative analysis of public political relationships possible volving two parties are the dependent variable.I Specifically,I focus on the subset of relations between political parties.What determines whether they have regress this on a dummy variable indicating whether two parties are in a coalition together,as well cooperative or conflictual relationships? Figure 1 suggests that coalition status matters.We as a dummy for opposition-opposition dyads (mak- would expect parties in government to work together ing government-opposition dyads the baseline).Re- garding the effect of policy,I use the Compara- to achieve common goals,and to have similar inter- tive Manifestos Project (CMP)data and compute actions with other actors in society.Opposition par- the absolute distance between the positions of two ties,in contrast,are supposed to hold the government parties.In addition to the standard left-right dif- accountable,which means direct criticism and coop- ference,I also include a measure of parties'dif- eration with societal actors opposed to the govern- ference on issues of nationalism and multicultural- ment.The expectation is therefore that parties have ism as a second policy dimension in European party a more cooperative relationship when they are in a competition.9 coalition. Table 1 shows the results of three specifications.The Second,given the central role that policy plays in pooled model (1)and the one with country and year structuring political competition in European coun- fixed effects (2)include the mean relationship score tries,one can expect that it affects parties'public relationships as well.If two parties advocate sim- for each country-year to address potential concerns ilar policies,they should interact in a cooperative manner with each other,and interact similarly with third actors.For example,two economically con- Nationalism and multiculturalism position:(per601+per607)- (per602 per608).Since CMP positions can only be measured in servative parties both are likely to have coopera- election years,I use the most recent available CMP position. 1093
Quantifying Political Relationships FIGURE 1. Relationship Scores, Germany, 2001–2009 Black: two parties. Dark gray: one party, one nonpartisan actor. Light gray: two nonpartisan actors. Public Relationships among Political Parties Having shown the face validity of the measure, I now demonstrate how it makes a systematic comparative analysis of public political relationships possible. Specifically, I focus on the subset of relations between political parties. What determines whether they have cooperative or conflictual relationships? Figure 1 suggests that coalition status matters. We would expect parties in government to work together to achieve common goals, and to have similar interactions with other actors in society. Opposition parties, in contrast, are supposed to hold the government accountable, which means direct criticism and cooperation with societal actors opposed to the government. The expectation is therefore that parties have a more cooperative relationship when they are in a coalition. Second, given the central role that policy plays in structuring political competition in European countries, one can expect that it affects parties’ public relationships as well. If two parties advocate similar policies, they should interact in a cooperative manner with each other, and interact similarly with third actors. For example, two economically conservative parties both are likely to have cooperative relations with business and conflictual ones with unions. To test these conjectures, I estimate a set of models in which the dyadic relationship scores involving two parties are the dependent variable. I regress this on a dummy variable indicating whether two parties are in a coalition together, as well as a dummy for opposition-opposition dyads (making government-opposition dyads the baseline). Regarding the effect of policy, I use the Comparative Manifestos Project (CMP) data and compute the absolute distance between the positions of two parties. In addition to the standard left-right difference, I also include a measure of parties’ difference on issues of nationalism and multiculturalism as a second policy dimension in European party competition.9 Table 1 shows the results of three specifications. The pooled model (1) and the one with country and year fixed effects (2) include the mean relationship score for each country-year to address potential concerns 9 Nationalism and multiculturalism position: (per601 + per607) – (per602 + per608). Since CMP positions can only be measured in election years, I use the most recent available CMP position. 1093 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:56:49, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000461
Simon Weschle TABLE 1.Determinants of Party Relationship Scores () (2) (3) Coalition 0.127 0.152 0.179 (-0.018,0.272) (0.005,0.299) (0.021,0.336) Opposition -0.041 -0.061 -0.065 (0.097,0.015) (-0.133,0.011) (-0.154,0.024) l△CMP Left-Rightl 0.001 0.001 0.001 (0.001,0.002) (-0.001,0.003) (-0.001,0.003) A CMP Nationalism/Multiculturalism -0.006 -0.008 -0.011 (-0.013,0.001) -0.015,-0.001) -0.022,-0.001) Control A Vote Shares] Country-Year Controls Country and Year FE Country-Year FE N 1150 1150 1150 R 0.060 0.090 0.216 Ninety-five percent confidence intervals in parentheses (based on standard errors clustered at the dyad level) about the comparability of the separately estimated la- CONCLUSION:USING QUANTIFIED 4号元 tent spaces.In addition,I add a set of controls at the POLITICAL RELATIONSHIPS IN country-year level (election year,number of parties, COMPARATIVE POLITICS log population,number of events).Specification (3) & includes country-year fixed effects and thus only uses The public relationships among political actors are im- variation from within each latent space,which ensures portant,but a lack of comparative data has largely pre- comparability.All three specifications control for the vented their empirical study.Using a combination of difference in vote shares between the parties.I take the machine-coded news reports and latent factor models,I uncertainty of the estimated relationship scores into have introduced a way to quantify public relationships account by deriving the dependent variable separately among political and societal actors.The scores for 13 for all posterior draws of a latent factor model,running Western European countries from 2001 to 2014 involv- the regression and simulating coefficient draws for each ing almost 3,500 actors are publicly available and can (standard errors clustered at the dyad level),combining be incorporated by researchers into their analyses in a the results from these estimations,and calculating the straightforward way. confidence intervals. Of course,there are some limitations.First,the rela- The regressions provide evidence supporting both tionship scores are based on media reports,which tend conjectures.Depending on the specification,a coalition to focus on high-level actors.The measure does not re- partnership is associated with an increase in the rela- flect the actions of,for example,backbench members tionship score by 0.30-0.42 standard deviations,com- of parliament or societal actors at the local level.A pared to the government-opposition baseline.If both second limitation is that the scores are based on pub- parties are in the opposition,the relationship score is lic interactions,which may be different from private somewhat lower than in the baseline,although the con- ones.Which interactions happen in public may also fidence intervals include zero. differ between countries depending on their laws.cul- Policy distance also has an effect on what kind of re- ture,or institutions.It is therefore important to address lationship parties have.Interestingly,it is not the dif- potential comparability issues in any analyses,for ex- ference in left-right positions that affects the relation- ample,through country or country-year fixed effects. ship scores.Instead,parties with a greater difference Third,the measure in its current form treats relation- in their positions towards nationalism/multiculturalism ships as symmetric.It is possible to extend the latent have lower scores.A one standard deviation increase in factor approach to differentiate between sender and this independent variable is associated with a decrease receiver behavior.0 Finally,I treat all interactions as in the dependent variable by 0.06-0.10 standard devi- equal,whereas it is likely that some of them are more ations.The public relationships of political parties are important for a relationship than others.However,this therefore not as much driven by whether they are left concern should be alleviated by the large number of wing or right wing.but instead by the cleavage between interactions I am able to use. "mainstream"and other parties about nationalism and Keeping these limitations in mind,the quantified po- multiculturalism-a fact that has only been made more litical relationship approach opens the door for a host salient in recent years.Overall,a systematic analysis of new research.For example,it provides new data to thus shows that the public relationships among political parties in Western Europe follow predictable patterns. 10 See Online Appendix. 1094
Simon Weschle TABLE 1. Determinants of Party Relationship Scores (1) (2) (3) Coalition 0.127 0.152 0.179 (–0.018, 0.272) (0.005, 0.299) (0.021, 0.336) Opposition −0.041 −0.061 −0.065 (–0.097, 0.015) (–0.133, 0.011) (–0.154, 0.024) | CMP Left-Right| 0.001 0.001 0.001 (–0.001, 0.002) (–0.001, 0.003) (–0.001, 0.003) | CMP Nationalism/Multiculturalism| −0.006 −0.008 −0.011 (–0.013, 0.001) (–0.015, −0.001) (–0.022, −0.001) Control | Vote Shares| √√ √ Country-Year Controls √ √ Country and Year FE √ Country-Year FE √ N 1150 1150 1150 R2 0.060 0.090 0.216 Ninety-five percent confidence intervals in parentheses (based on standard errors clustered at the dyad level). about the comparability of the separately estimated latent spaces. In addition, I add a set of controls at the country-year level (election year, number of parties, log population, number of events). Specification (3) includes country-year fixed effects and thus only uses variation from within each latent space, which ensures comparability. All three specifications control for the difference in vote shares between the parties. I take the uncertainty of the estimated relationship scores into account by deriving the dependent variable separately for all posterior draws of a latent factor model, running the regression and simulating coefficient draws for each (standard errors clustered at the dyad level), combining the results from these estimations, and calculating the confidence intervals. The regressions provide evidence supporting both conjectures.Depending on the specification, a coalition partnership is associated with an increase in the relationship score by 0.30–0.42 standard deviations, compared to the government-opposition baseline. If both parties are in the opposition, the relationship score is somewhat lower than in the baseline, although the confidence intervals include zero. Policy distance also has an effect on what kind of relationship parties have. Interestingly, it is not the difference in left-right positions that affects the relationship scores. Instead, parties with a greater difference in their positions towards nationalism/multiculturalism have lower scores.A one standard deviation increase in this independent variable is associated with a decrease in the dependent variable by 0.06–0.10 standard deviations. The public relationships of political parties are therefore not as much driven by whether they are left wing or right wing, but instead by the cleavage between “mainstream” and other parties about nationalism and multiculturalism—a fact that has only been made more salient in recent years. Overall, a systematic analysis thus shows that the public relationships among political parties in Western Europe follow predictable patterns. CONCLUSION: USING QUANTIFIED POLITICAL RELATIONSHIPS IN COMPARATIVE POLITICS The public relationships among political actors are important, but a lack of comparative data has largely prevented their empirical study. Using a combination of machine-coded news reports and latent factor models, I have introduced a way to quantify public relationships among political and societal actors. The scores for 13 Western European countries from 2001 to 2014 involving almost 3,500 actors are publicly available and can be incorporated by researchers into their analyses in a straightforward way. Of course, there are some limitations. First, the relationship scores are based on media reports, which tend to focus on high-level actors. The measure does not reflect the actions of, for example, backbench members of parliament or societal actors at the local level. A second limitation is that the scores are based on public interactions, which may be different from private ones. Which interactions happen in public may also differ between countries depending on their laws, culture, or institutions. It is therefore important to address potential comparability issues in any analyses, for example, through country or country-year fixed effects. Third, the measure in its current form treats relationships as symmetric. It is possible to extend the latent factor approach to differentiate between sender and receiver behavior.10 Finally, I treat all interactions as equal, whereas it is likely that some of them are more important for a relationship than others. However, this concern should be alleviated by the large number of interactions I am able to use. Keeping these limitations in mind, the quantified political relationship approach opens the door for a host of new research. For example, it provides new data to 10 See Online Appendix. 1094 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:56:49, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000461
Quantifying Political Relationships study elite polarization,and the relationship scores be- Jacob Bercovitcch and Scott Sigmund Gartner.New York,NY: tween politicians and societal actors should be of in- Routledge. terest for students of political representation.The po- Grimmer,Justin.2013.Representational Style in Congress.What Leg- tential of the new data is perhaps even stronger as an islators Say and Why It Matters.Cambridge:Cambridge University Press. explanatory variable.Public relationships are likely to Grimmer,Justin,Sean J.Westwood,and Solomon Messing.2015.The have an effect on voters,for example,on how they per- Impression of Influence:Legislator Communication,Representa- ceive parties'policy positions,how polarized they are, tion,and Democratic Accountability.Princeton.NJ:Princeton Uni- or what their level of satisfaction with democracy is. versity Press. Herzog,Alexander,and Kenneth Benoit.2015."The Most Unkind- In the past years,scholars have increasingly made est Cuts:Speaker Selection and Expressed Government Dissent use of novel,large-scale data to address questions during Economic Crisis."Journal of Politics 77(4):1157-75. in comparative politics.The approach and data in- Hoff,Peter D.2005."Bilinear Mixed Effects Models for Dyadic troduced in this article contribute to this trend Data."Journal of the American Statistical Association 100 (469): 286-95. and make it possible to revisit long-standing ques- Hoff,Peter D.2015."Dyadic Data Analysis with amen."http://arxiv. tions on the causes and consequences of political org/abs/1506.08237. relationships Hoff,Peter D.and Michael D.Ward.2004."Modeling Dependen- cies in International Relations Networks."Political Analysis 12(2): 160-75. SUPPLEMENTARY MATERIAL King,Gary,and Will Lowe.2003."An Automated Information Ex. traction Tool for International Conflict Data with Performance as To view supplementary material for this article,please Good as Human Coders:A Rare Events Evaluation Design."In- ternational Organization 57:617-42 visit https:/doi.org/10.1017/S0003055418000461 Lauderdale,Benjamin E,and Alexander Herzog.2016."Measuring Replication materials can be found on Dataverse at: Political Positions from Legislative Speech."Political Analysis 24 https://doi.org/10.7910/DVN/AOTVAU. (3:374-94. Mansbridge,Jane.2003."Rethinking Representation."American Po- litical Science Review 97 (4):515-28 Martin,Lanny W,and Georg Vanberg.2008."Coalition Government REFERENCES and Political Communication."Political Research Quarterly 61(3): 502-16. Boschee,Elizabeth,Jennifer Lautenschlager,Sean O'Brien O'Brien,Sean P.2013."A Multi-Method Approach for Near Real Steve Shellman,James Starz,and Michael D.Ward.2015."ICEWS Time Conflict and Crisis Early Warning."In Handbook of Compu- Coded Event Data."https://doi.org/10.7910/DVN/28075. tational Approaches to Counterterrorism,ed.V.S.Subrahmanian. Boschee.Elizabeth.Premkumar Natarian,and Ralph Weischedel. New York,NY:Springer. 2013."Automatic Extraction of Events from Open Source Text Proksch,Sven-Oliver,and Jonathan B.Slapin.2015.The Poli for Predictive Forecasting."In Handbook of Computational Ap- tics of Institutional Debate.Cambridge:Cambridge University rrrrsmed.V.S.Subrahmanian.New York. Press. Sagarzazu,Inaki,and Heike Kluiver.2017."Coalition Governments Eggers,Andrew C,and Arthur Spirling.2014."Ministerial Respon- and Party Competition:Political Communication Strategies of at Common Debatemec aPoeal Coalition Parties."Political Science Research and Methods 5 (2): 333-49. Science58(4):873-87 Strom,Kaare.2008."Communication and the Life Cycle of Parlia- Fenno,Richard F.Jr.1978.Home Style:House Members in Their Dis- mentary Democracy."Political Research Quarterly 61 (3):537- tricts.Boston,MA:Little.Brown. 42. Gerner,Deborah J,Philip A.Schrodt,and Omur Yilmaz.2009 Weschle,Simon.2017."The Impact of Economic Crises on Politi- "Conflict and Mediation Event Observations (CAMEO):An cal Representation in Public Communication:Evidence from the Event Data Framework for a Post-Cold War World."In Inter- Eurozone."British Journal of Political Science.https://doi.org/10. national Conflict Mediation:New Approaches and Findings,eds 1017/S0007123417000023. 1095
Quantifying Political Relationships study elite polarization, and the relationship scores between politicians and societal actors should be of interest for students of political representation. The potential of the new data is perhaps even stronger as an explanatory variable. Public relationships are likely to have an effect on voters, for example, on how they perceive parties’ policy positions, how polarized they are, or what their level of satisfaction with democracy is. In the past years, scholars have increasingly made use of novel, large-scale data to address questions in comparative politics. The approach and data introduced in this article contribute to this trend and make it possible to revisit long-standing questions on the causes and consequences of political relationships. SUPPLEMENTARY MATERIAL To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055418000461 Replication materials can be found on Dataverse at: https://doi.org/10.7910/DVN/AOTVAU. REFERENCES Boschee, Elizabeth, Jennifer Lautenschlager, Sean O’Brien, Steve Shellman, James Starz, and Michael D.Ward. 2015. “ICEWS Coded Event Data.” https://doi.org/10.7910/DVN/28075. Boschee, Elizabeth, Premkumar Natarjan, and Ralph Weischedel. 2013. “Automatic Extraction of Events from Open Source Text for Predictive Forecasting.” In Handbook of Computational Approaches to Counterterrorism, ed. V. S. Subrahmanian. New York, NY: Springer. Eggers, Andrew C., and Arthur Spirling. 2014. “Ministerial Responsiveness in Westminster Systems: Institutional Choices and House of Commons Debate, 1832–1915.” American Journal of Political Science 58 (4): 873–87. Fenno, Richard F. Jr. 1978. Home Style: House Members in Their Districts. Boston, MA: Little, Brown. Gerner, Deborah J., Philip A. Schrodt, and Omur Yilmaz. 2009. “Conflict and Mediation Event Observations (CAMEO): An Event Data Framework for a Post-Cold War World.” In International Conflict Mediation: New Approaches and Findings, eds Jacob Bercovitcch and Scott Sigmund Gartner. New York, NY: Routledge. Grimmer, Justin. 2013. Representational Style in Congress.What Legislators Say and Why It Matters. Cambridge: Cambridge University Press. Grimmer, Justin, Sean J.Westwood, and Solomon Messing. 2015.The Impression of Influence: Legislator Communication, Representation, and Democratic Accountability. Princeton,NJ: Princeton University Press. Herzog, Alexander, and Kenneth Benoit. 2015. “The Most Unkindest Cuts: Speaker Selection and Expressed Government Dissent during Economic Crisis.” Journal of Politics 77 (4): 1157–75. Hoff, Peter D. 2005. “Bilinear Mixed Effects Models for Dyadic Data.” Journal of the American Statistical Association 100 (469): 286–95. Hoff, Peter D. 2015. “Dyadic Data Analysis with amen.” http://arxiv. org/abs/1506.08237. Hoff, Peter D., and Michael D. Ward. 2004. “Modeling Dependencies in International Relations Networks.”Political Analysis 12 (2): 160–75. King, Gary, and Will Lowe. 2003. “An Automated Information Extraction Tool for International Conflict Data with Performance as Good as Human Coders: A Rare Events Evaluation Design.” International Organization 57: 617–42. Lauderdale, Benjamin E., and Alexander Herzog. 2016. “Measuring Political Positions from Legislative Speech.” Political Analysis 24 (3): 374–94. Mansbridge, Jane. 2003. “Rethinking Representation.”American Political Science Review 97 (4): 515–28. Martin,Lanny W., and Georg Vanberg. 2008. “Coalition Government and Political Communication.”Political Research Quarterly 61 (3): 502–16. O’Brien, Sean P. 2013. “A Multi-Method Approach for Near Real Time Conflict and Crisis Early Warning.” In Handbook of Computational Approaches to Counterterrorism, ed. V. S. Subrahmanian. New York, NY: Springer. Proksch, Sven-Oliver, and Jonathan B. Slapin. 2015. The Politics of Institutional Debate. Cambridge: Cambridge University Press. Sagarzazu, Inaki, and Heike Klüver. 2017. “Coalition Governments and Party Competition: Political Communication Strategies of Coalition Parties.” Political Science Research and Methods 5 (2): 333–49. Strom, Kaare. 2008. “Communication and the Life Cycle of Parliamentary Democracy.” Political Research Quarterly 61 (3): 537– 42. Weschle, Simon. 2017. “The Impact of Economic Crises on Political Representation in Public Communication: Evidence from the Eurozone.” British Journal of Political Science. https://doi.org/10. 1017/S0007123417000023. 1095 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:56:49, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S0003055418000461