Scaling the Commons: Using MPs Left-Right Self-Placement and Voting Divisions. to Map the British Parliament,

Transcription

1 Scaling the ommons: sing MPs eft-right Self-Placement and Voting ivisions to Map the British Parliament, Simon Hix ondon School of Economics and Political Science Abdul Noury New York niversity Paper prepared for presentation at the annual meeting of the American Political Science Association, in Washington,, September 2-5, 21.

2 Abstract In this paper we examine the performance of three ideal-point estimation methods to analyze voting behaviour in a parliamentary systems legislature characterized by multiple political parties and high level of party discipline. sing roll call votes ( divisions ) from the British House of ommons, we apply IEA, NOMINATE, and Optimal lassification to scale the positions of the British MPs. We find that two dimensions are needed to explain voting patterns in this parliament, and whilst the first dimension captures a government-opposition divide, a second dimension exists and captures left-right preferences of MPs both within and between political parties. We also find that although these three methods use different estimation strategies they produce almost identical estimates of the spatial locations of the British MPs. 2

3 Introduction There has been explosion of research on legislative voting in the last decade. This new research has been driven by several factors, such as the availability of new data (e.g. on the internet) and the new computer power that allows us to analyse large voting datasets in more sophisticated ways. Above all, this new research has been inspired by the development and dissemination of new methods for scaling legislative votes and using these data to estimate the ideal points of legislators. The three most prominent methods are NOMINATE, developed by Keith Poole and Howard Rosenthal (1997), Optimal lassification, developed by Keith Poole (2), and IEA, developed by Joshua linton, Simon Jackman and ouglas Rivers (24). These methods tend to work well in settings where there are a large number of roll call votes and where legislators are free to vote independently from their political parties, which allows for sufficiently heterogeneous cut-points to allow legislators ideal-points to be identified with some accuracy. However, there is some scepticism about the applicability of these methods in parliaments were there are cohesive political parties or where the dynamics of voting are driven by strategic behaviour, such as government-opposition alignments, rather than by relative policy positions vis-à-vis the status quo the alternative in votes. For example, Rosenthal and Voeten (22) argue that Poole s (2) Optimal lassification method is preferable to parametric methods, such as NOMINATE, for studying voting in parliaments outside the S ongress. Specifically, party discipline, near-perfect spatial voting and parliamentary institutions (where the government controls the legislative agenda) lead to severe violations of the error assumptions underlying parametric methods. Spirling and Mcean (26, 27) go even further, arguing that in parliamentary systems, such as the British House of ommons, roll call votes might be able to identify differences between 3

4 parties but cannot be used to extrapolate the relative ideal-points of parliamentarians within parties. The reason for this, they contend, is that radical members of governing parties often vote against their party leaders and with opposition parties to signal the general direction of party policy. When scaling the positions of these members, such behaviour leads them to be identified as ideologically more moderate than their leaders (closer to the opposition parties than their leaders) when in fact they are more ideologically extreme (further from the opposition parties than their leaders). This observation seems to be supported by other research on voting in parliamentary systems, by Kam (29) and Benedetto and Hix (27) for example, which suggests that backbench members of governing parties are driven as much by career incentives, such as the carrot of promotion to ministerial office, than by policy preferences on particular bills. Zucco (29) finds similar strategic incentives at work in the Brazilian ongress, where allegiance to the government and the prospect of ministerial office are more powerful determinants of roll call voting in this chamber than ideological preferences. More generally, arroll et al. (29) compare NOMINATE and IEA to analyze the sources of observed differences between those methods. They examined the cases of the 1994 to 1997 Supreme ourt and the 19th Senate. They show that while some observed differences in the estimates produced by each model stem from fundamental differences in the models underlying behavioral assumptions, others arise from arbitrary differences in implementation. They conclude that neither model has a clear advantage over the other in the recovery of legislator locations or roll-call midpoints in either large or small legislatures. linton and Jackman (29), on the other hand, argue that it is preferable to choose the more flexible Bayesian approach. However, they focus on the case of the S ongress and on a one-dimensional policy space. In this paper we look at a different setting, where there are more than two parties and there is a parliamentary system, and we hence argue that two 4

5 dimensions are needed to provide a meaningful map of the ideal positions of legislators in this type of parliament. By focusing on voting in the House of ommons we try to address two main questions in this paper, as follows. First, can we apply the existing models of spatial voting to a multiparty parliamentary system where legislators vote cohesively and strategically? In other words, using the existing methods do we obtain reasonable results when estimating ideal positions in such a setting? And, can the ideological preferences of British MPs be correctly identified from divisions in the House of ommons; and, if so, which is the best estimation strategy for extracting these ideal-points? One can also examine other hypothesis such as a divergence between moderates and extremists when comparing the various estimation methods (arroll et al., 29). They have argued that it is likely that moderates are better identified by IEA and extremists are better estimated by NOMINATE. To answer these questions the rest of the paper is organized as follows. We first briefly discuss the three ideal point estimation methods, before presenting how we apply them to votes in the and 21-5 sessions of the British House of ommons. We then set out and apply a second-stage estimation strategy to try to identify the substantive meanings of the first and second dimensions uncovered by the two methods, and whether either or both of these dimensions relates to the ideological positions of British MPs both within and between political parties. Statistical Models of Ideal Point Estimation: NOMINATE, O and IEA To study legislative behaviour, for decades scholars have used roll call data and statistical models of voting. The statistical models, often called spatial models of voting or ideal-point 5

6 estimators, have provided new insights about legislative behaviour in the S ongress and beyond. The most influential model for decades has been Poole and Rosenthal s (1997) NOMINATE. Building on NOMINATE, and in reaction to it, researchers have developed a variety of other methods to estimate the ideal positions of legislators. The most significant alternatives to NOMINATE are Optimal lassification (O) and IEA. All these models are developed in the context of the spatial theory of voting. To formally describe the basic spatial theory as applied in legislative votes, consider the behaviour of n MPs voting on m bills. When MP i=1,,n votes on bill j=1,,m she has a choice c j { yes, no }: that is, a choice between Yes and No votes on each bill. For simplicity, we ignore abstention and assume that a No vote corresponds to the status quo. When deciding whether to vote for or against a bill, legislator i has a utility function that is based on the Euclidean distance between her ideal position and the yes or no outcomes, as follows: u = f (δ ) + e, ijc ijc ijc where δ ijc is the Euclidean distance between her ideal position x R i d and the d outcome z jc R with d denoting the dimension of policy space, and where e ijc is the stochastic portion of the utility function. 2 wx 2 NOMINATE assumes that f ( x) = β e, where β and w are constants, while IEA assumes that 2 =. The error term in these functions is either assumed to have f ( x) x a logistic distribution, in NOMINATE, and a normal distribution, in IEA. tility maximization implies that legislator i votes Yes on proposal j if and only if u, otherwise she votes No. That is Pr( y = 1) = Pr( u > u ). ijyes u ijno ij ijyes ijno Just like NOMINATE and IEA, O assumes that legislators have single-peaked preferences. The main differences between NOMINATE and IEA, on the one hand, and 6

7 O, on the other, are that O is non-parametric and makes no explicit assumption about the shape of the utility function of MPs or about the distribution of the stochastic component of the utility function. As already discussed by arroll et al. (29) and linton and Jackman (29), there are some differences between IEA and NOMINATE. espite the differences, however, the three different methods share some important similarities. For instance, NOMINATE and IEA are both based on additive utility functions (random utility). Although they use different utility functions, those functions are almost identical, especially for non-extremists (arroll et al., 29). More importantly, NOMINATE, O, and IEA use exactly identical staring values extracted from the legislator-by-legislator correlation matrix. This is a crucially important similarity that explains why the results produced by those methods are highly correlated. ata and Estimation So, what happens when these three statistical estimations are applied to voting in the British House of ommons? The House of ommons is a good laboratory for testing the models. The British system of government is, of course, the classic Westminster model of a parliamentary system: where the government is dependent on a parliamentary majority, there is single party government, and the government has a virtual monopoly on the legislative agenda (cf. ijphart, 1999). This institutional set-up usually means that parties are highly cohesive in parliamentary votes, not least because governing parties can use the carrot of promotion to ministerial office or the threat of parliamentary dissolution to enforce party discipline (Huber, 1996; Kam, 29). Nevertheless, rebellions in the House of ommons surfaced in the 197s, and increased dramatically in the 199s, particularly towards the end of 7

8 the second term of the abour government, in the term (Norton, 1975, 198; Kam, 21; owley, 22, 25; Benedetto and Hix, 27). In addition to the generalisable significance of the British institutional set up and the increased frequency of rebellions, data on divisions in the House of ommons is now widely available on the internet. As a result of these factors, there is a growing body of research which analyses voting in the House of ommons using spatial statistical models (e.g. Spirling and Mclean, 27; Spirling, 21; Spirling and Quinn, 21). Building on this research we look at the and sessions of the House of ommons: in other words, the first two parliamentary terms of Tony Blair s abour government. This is an interesting period to analyse, as there were a number of high-profile rebellions against the government, for example on increasing university tuition fees, the introduction of foundation hospitals, and the decision to go to war with Iraq. We use the voting data on avid Firth s website. 1 The dataset contains 1,285 divisions in the term and 1,246 in the term. We excluded MPs who participated in less than 25 votes, which lead to four MPs being dropped from the data and two from the data. Following the standard practice, we also excluded all lopsided votes: votes with fewer than 5 percent on the minority side. This reduced the first sample from 1,285 to 1,245 votes and the second sample from 1,246 to 1,198 votes. It is well documented that the estimated ideal positions with the standard statistical models are not identified (Bafumi et al. 25). In NOMINATE, identification is obtained by fixing the positions of the extremists (the most leftwing member is assumed to be at -1 and the most rightwing at +1), as well as by normalizing the maximum distance between members to 2. In IEA, in contrast, identification is obtained by normalizing ideal positions to have a mean of zero and a standard deviation of one, and specifying the polarity of the dimension: 1 (see Firth and Spirling, 23). 8

9 such as abour MPs on the left and onservative MPs on the right. These identification strategies work well when there is a one-dimensional policy space. In addition, given the stability of ideal positions in many legislatures, together with low-dimensionality, the lack of identification is not usually a serious problem. As noted by Rivers (23), identification is more complicated in multidimensional models, however. Typically, for a d-dimensional model, one should fix the positions of d+1 legislators. An alternative solution would to use priors to achieve identification. We use the latter strategy to identify the ideal positions. We estimate a twodimensional model. Following the research that identifies the first dimension in the House of ommons as primarily relating to government-opposition dynamics rather than ideological positions of parties or MPs (Spirling and Mcean, 27), we leave the first dimension open in all three models. We specify the polarity of this dimension, though, so that the most leftwing party is mapped as the left-hand side of this dimension. We do something rather different for the second dimension, however. For NOMINATE and O we identify the polarity of this dimension by placing a well-known leftwing MP on the left on this dimension. We use a different strategy for IEA. Assuming that legislators in parliamentary systems generally follow the voting instructions of their party leaders, we identify the second dimension using data on the self-placement of party leaders on a left-right dimension: entering the position of the party leader as prior position of each MP in a party. That is, we assume that prior distribution of ideal positions is normal centred on the party leaders positions. We obtain the data on party leaders self-placement from Pippa Norris and Joni ovenduski s British Representation Studies in 1997 and This strategy, of using an exogenous measure taken from survey data, is similar to the one used by Zucco and auderdale (29) for scaling the Brazilian ongress, who use the average left- 2 see 9

10 right self-placement score of the survey respondents of each ongress member s party as the identifier for each member. In addition, we normalize the positions of legislators to have a mean of zero and a standard deviation of one. Results: imensionality and orrelations of the Estimates In terms of the results, the first things to note are the goodness-of-fit statistics: in other words, how well do the models explain the variance in voting in the House of ommons. The first two dimensions (Eigenvalues) capture approximately 68 percent and 8 percent of the variations, respectively, in all three models. Eigenvalues associated to higher dimensions were less than 2 percent. In terms of the percent of correct classification votes, NOMINATE produced scores of 97.5 percent the first dimension and 98.5 percent for the second dimension in the parliament, and 97.5 percent for the first dimension and 97.8 percent for the second dimension in the parliament. The corresponding numbers produced by O are higher: 99.2 percent and 99.4 percent for the two dimensions in the first parliament and 98.8 percent and 98.9 percent for the two dimensions in the second parliament. This is not that surprising, given that O focuses on classification optimization while NOMINATE maximizes a likelihood function. [Figure 1 about here] As an illustration of the outputs of the models, Figure 1 presents the two-dimensional ideal-point estimates using IEA. Note that in both figures the abour MPs appear on the left on the first dimension, with the onservative MPs on the right, and the iberal emocrat 1

11 MPs in the middle, although closer to the onservatives in the parliament than in the parliament. On the second dimension, however, the abour and onservative Parties are both at the top in both figures, while the iberal emocrats are at the bottom. Note also, that despite the high levels of party cohesion in the House of ommons, there is some dispersion within all three of the main parties, particularly in the parliament. [Table 1 about here] Our identification strategy produces highly correlated results from the three models, as Table 1 shows. Note, however, that the correlations are higher on the first dimension than on the second dimension. A more detailed comparison of the results of the three models is presented in Figures 2 and 3. [Figures 2 and 3 about here] Interpreting the imensions: Government-Opposition, Parties and MPs Ideology So, how do we interpret the substantive meaning of the dimensions produced by the three models, and is the interpretation the same for each model. Recall that the standard view of voting in the House of ommons is that it is primarily related to government-opposition dynamics, that parties are highly cohesive, and that any variance in voting behaviour of MPs within the parties is primarily driven by strategic behaviour rather than ideology or policy preferences. Put another way, how far do the government-opposition status of parties, the 11

12 parties policy positions, and individual MPs preferences explain the locations of the MPs on the two dimensions produced by applying IEA, NOMINATE, and O? To answer this question we estimate a series of linear regression models. To test the effect of government and opposition we use dummy variables for the four main parties: abour, onservative, iberal emocrat, and the Scottish National Party. abour, of course, is in government in both parliamentary terms. To test the effect of parties ideological positions we use estimates of the left-right location of British parties in the Manifestos Research Group/omparative Manifestos Project (MRG/MP) dataset (Klingemann et al., 26). The MRG/MP project involves the handcoding of party manifestos, using a 54 category coding frame, and then the construction inter alia of a the location of parties on an integration left-right dimension. This dataset is widely used in comparative political science as estimates of party positions across country and over time. In the British case, the dataset includes estimates of the parties positions, as set out in their manifestos, for both the 21 and 25 elections. Table 2 presents the estimates of the British parties positions in the MRG/MP dataset in comparison with the average left-right self-placement of each party s group of MPs who answered the questionnaires in the Norris and ovenduski British Representation Studies in 1997 and 21. According the manifesto data, the left-right line-up of the parties in both elections was the same: with the SNP furthest to the left, and iberal emocrats to the left of abour although the iberal emocrats move to the right and closer to the position of abour in the 25 election. The data from the self-placement of the elected MPs is slightly different, in that the average positions of the SNP, iberal emocrat and abour MPs are very close together, yet all are considerably to the left of the onservative MPs. [Table 2 about here] 12

13 Finally, for the ideological preferences of the individual MPs we use the data on MP left-right self-placement from the 21 and 25 British Representation Studies. In 1997, 228 MPs answered the left-right self-placement question in the survey, which is 34 percent of the MPs we were able to scale in the term. And, in 21, 22 MPs answered the left-right self-placement question, which is 31 percent of the MPs we were able to scale in the term. Tables 3 and 4 present the results for the first dimensions in the and terms, respectively. The first dimension in both terms of the parliament is clearly a government-opposition split. First, abour MPs are on the left-hand side on this dimension. However, this is less a result of this party s ideological position than the fact that these MPs are from the party in government, as revealed by the fact that no party appears to the left of the abour MPs yet are positioned either to the left or very close to the abour Party. In fact, the SNP and iberal emocrat MPs are in the middle on the first dimension, as a result of the fact that they usually voted with the opposition onservatives MPs rather than with the governing abour MPs. [Tables 3 and 4 about here] Second, individual MPs ideological preferences do not account for much. They are significant in some specifications, but this is because they are correlated with party positions, and the effect of MPs left-right self-placement disappears when party affiliation is controlled for. Put another way, MPs ideological positions do not explain variation within the parties on the first dimension. Furthermore, these findings are robust across both parliamentary terms and across all three ideal-point estimation models. 13

14 Tables 5 and 6 reveal that the second dimension identified by the ideal-point estimation models is in fact an ideological dimension relating to left-right positions of parties and individual MPs. The positions of MPs on the second dimension are again mainly determined by the positions of their parties. And, the positions of the parties on this dimension are determined by left-right policy preferences. This is clearest in the IEA results for the two parliamentary terms: where the iberal emocrats are at the bottom on the second dimension (and to the left of abour in policy terms in this period), abour are next up (in the centre in policy terms), and the onservatives are at the top (the furthest right in policy terms of the main parties). [Tables 5 and 6 about here] Then, the individual left-right policy preferences of MPs explain variations within the parties on the second dimension. In other words, within the cluster of each group of MPs in Figure 1, above, the more rightwing MPs within each party are towards the top of each cluster while the more leftwing MPs within each party are towards the bottom of each cluster. It is worth noting, however, that the explanatory power of the independent variables is considerably lower for the second dimension than for the first dimension, which suggests that there is simply a lot more noise in the behaviour of MPs when voting is not primarily along government-opposition lines. 14

15 onclusion Spatial models of voting in legislatures do not work as well where there is a strong government-opposition dynamic and powerful parties, which tend to produce strategic rather than policy-driven behaviour amongst politicians. We find evidence of such strategic voting in the British House of ommons. Nevertheless, spatial models are not necessarily incompatible with strategic voting in the form of government-opposition splits. If a onedimensional model is estimated, a government-opposition dynamic will be identified. However, if the one-dimensional assumption is relaxed, and a two-dimensional model is estimated, then the second dimension captures some aspects of political conflict. In the British case, our results indicate that the second dimension, if carefully identified, can be regarded as a left-right policy dimension, both between and within political parties. In a parliamentary system characterized with strong political parties and strategic voting, focusing on just one dimension of politics in parliaments is not enough and can provide a misleading image of political conflict. Furthermore, when comparing different methods for estimating the positions of parliamentarians from their voting behaviour, we find that these methods produce highly correlated estimates. This confirms the conclusion of arroll et al. (29) that NOMINATE and IEA are comparable in recovery of legislator locations. The observed similarity in outputs is due to the similarity between the methods, notably the use of exactly identical starting values. 15

16 References Bafumi, Joseph, Andrew Gelman, avid Park, and Noah Kaplan. 25. Practical Issues in Implementing and nderstanding Bayesian Ideal Point Estimation. Political Analysis 13(2) Benedetto, Giacomo, and Simon Hix. 27. The Rejected, the Ejected, and the ejected: Explaining Government Rebels in the British House of ommons. omparative Political Studies 4(7) linton, Joshua., and Simon Jackman. 29. To Simulate or Nominate? egislative Studies Quarterly 34(4) linton, Joshua., Simon Jackman, and ouglas Rivers. 24. The Statistical Analysis of Roll all ata. American Political Science Review 98(2) arey, John M. 21. arey ata Archive. arroll, Royce, Jeffrey B. ewis, James o, Keith T. Poole, and Howard Rosenthal. 29. omparing NOMINATE and IEA: Points of ifference and Monte arlo Tests. egislative Studies Quarterly 34(4) owley, Philip. 22. Revolts and Rebellions: Parliamentary Voting nder Blair. ondon: Politico s Publishing. owley, Philip. 25. The Rebels: How Blair Mislaid His Majority. ondon: Politico s Publishing. Firth, avid, and Arthur Spirling. 23. ivisions of the nited Kingdom House of ommons, from 1992 to 23 and Beyond. Mimeo, Huber, John The Vote of onfidence in Parliamentary emocracies. The American Political Science Review 9(2) Kam, histopher. 21. o Ideological Preferences Explain Parliamentary Behavior? Evidence From Britain and anada. Journal of egislative Studies 7(1) Kam, hristopher. 29. Party iscipline and Parliamentary Government. ambridge: ambridge niversity Press. Klingemann, Hans-ieter, Andrea Volkens, Judith Bara, Ian Budge, and Michael Mconald. 26. Mapping Policy Preferences II: Estimates for Parties, Electors, and Governments in Eastern Europe, the European nion and the OE, Oxford: Oxford niversity Press. ijphart, Arend Patterns of emocracy: Government Forms and Performance in Thirty-Six ountries. New Haven, T: Yale niversity Press. Martin, Andrew., and Kevin Quinn. 29. MMpack: Markov hain Monte arlo 16

17 Package. Norton, Philip issension in the House of ommons: Intra-Party issent in the House of ommons ivision obbies, ondon: Macmillan. Norton, Philip issension in the House of ommons, Oxford: larendon Press. Poole, Keith T., and Howard Rosenthal A Political-Economic History of Roll all Voting. Oxford: Oxford niversity Press. Poole, Keith T. 2. Nonparametric nfolding of Binary hoice ata. Political Analysis 8(2) Rivers, ouglas. 23. Identification of Multidimensional Spatial Voting Models. Mimeo, Stanford niversity. Rosenthal, Howard, and Erik Voeten. 24. Analyzing Roll alls with Perfect Spatial Voting: France American Journal of Political Science 48(3) Spirling, Arthur. 21. Rebels with a ause? egislative Activity and the Personal Vote in Britain, Mimeo, Harvard niversity. Spirling, Arthur, and Iain Mcean. 26. The Rights and Wrongs of Roll alls. Government and Opposition 41(4) Spirling, Arthur, and Iain Mcean. 27. K O OK? Interpreting Optimal lassification Scores for the.k. House of ommons. Political Analysis 15(1) Spirling, Authur, and Kevin Quinn. 21. Identifying Intraparty Voting Blocs in the.k. House of ommons. Journal of the American Statistical Association 15(49) Zucco Jr., esar. 29. Ideology or What? egislative Behavior in Multiparty Presidential Settings. The Journal of Politics 71(3) Zucco Jr., esar, and Benjamin auderdale. 29. Making Votes Talk: Ideology and Government Influence on egislative Behavior. Paper prepared for presentation at the Annual Meeting of the American Political Science Association, in Toronto. 17

18 18 Figure 1. Estimates from IEA, with party leaders positions as priors for the second dimension a I I S p S I p I p P K d P S d d S S p I S p First imension Second imension b d d N S I K S S d K N d d N I K N K N First imension Second imension Note: = abour, = onservative, = iberal emocrat, S = Scottish National Party, d = emocratic nionist Party, K = K Independence Party, I = Independent. We estimated the ideal positions with 5 iterations.

19 Table 1. orrelations Between IEA, NOMINATE and O Estimates IEA NOMINATE O IEA NOMINATE O IEA NOMINATE O Note: pper diagonal entries are the correlation coefficients for the first dimension and those on the lower diagonal are the coefficients for the second dimension. 19

20 Figure 2. Scatter Plots of (orrelations Between) IEA, NOMINATE and O Estimates First imension 1 First imension.5.5 NOMINATE Second imension.6 First imension O Second imension.2 Second imension IEA Note: The scatter plots show the relationship between the estimates of IEA, NOMINATE and O. pper diagonal figures show the first dimension; lower diagonal figures correspond to second dimension. 2

21 Figure 3. Scatter Plots of (orrelations Between) IEA, NOMINATE and O Estimates First imension 1 First imension.5.5 NOMINATE Second imension.5 First imension O Second imension.5 Second imension IEA Note: The scatter plots show the relationship between the estimates of IEA, NOMINATE and O. pper diagonal figures show the first dimension; lower diagonal figures correspond to second dimension. 21

22 Table 2. Estimates of eft-right Positions of Parties in the House of ommons Parties (ordered left to right, based on manifestos data) parliament oding of party manifestos, integrated left-right scale Mean MP left-right self-placement, 1-1 scale (no. of respondents) Plaid ymru 4. (1) Scottish National Party (3) Sinn Fein -8.1 iberal emocrats (32) lster nionist Party 1.15 emocratic nionist Party 1.6 abour (149) onservatives (58) parliament Plaid ymru 2.33 (4) Scottish National Party (3) Sinn Fein -8.1 iberal emocrats (14) lster nionist Party 1.15 emocratic nionist Party 1.6 abour (148) onservatives (61) Note: ata on MP left-right self-placement from Pippa Norris and Joni ovenduski s British Representation Studies in 1997 and 21 ( ata on party locations from Manifestos Research Group/omparative Manifestos Project (MRG/MP) project data (see Klingemann et al., 26). 22

23 Table 3. Explaining the Estimated Positions of British MPs First imension (1) (2) (3) (4) (5) (6) (7) (8) (9) IEA IEA IEA NOMINATE NOMINATE NOMINATE O O O MP left-right self-placement *** *** ***.469 (.923) (.181) (.866) (.599) (.131) (.56) (.185) (.378) (.172) onservative 1.15*** 1.35***.759***.662***.235***.236*** (.277) (.273) (.151) (.167) (.667) (.519) abour -.546** -.164*** -.368** -.163*** * (.273) (.151) (.148) (.123) (.659) (.27) iberal emocrat *** *** *** (.275) (.28) (.147) (.127) (.661) (.398) SNP *** *** *** (.275) (.315) (.147) (.193) (.69) (.633) Party left-right location.264*.176*.53* (.149) (.913) (.284) onstant *** -.922*** *** -.588*** *** -.144*** (.275) (.787) (.123) (.148) (.551) (.75) (.661) (.157) (.234) Observations R-squared Note: Robust standard errors in parentheses. *** p<.1, ** p<.5, * p<.1. 23

24 Table 4. Explaining the Estimated Positions of British MPs First imension (19) (2) (21) (22) (23) (24) (25) (26) (27) IEA IEA IEA NOMINATE NOMINATE NOMINATE O O O MP left-right self-placement *** **.248*** -.18** *** (.395) (.145) (.395) (.533) (.14) (.533) (.281) (.357) (.281) onservative.477***.454***.872*** (.193) () (.236) () (.322) () abour *** *** -1.17*** *** -.312*** -.353*** (.838) (.794) (.123) (.982) (.33) (.175) iberal emocrat **.337*** -.496*** (.235) (.228) (.117) (.17) (.32) (.246) SNP -.756** -.127*** -.151** -.642*** * (.312) (.312) (.596) (.567) (.31) (.269) Party left-right location.461***.439***.842*** (.187) (.228) (.311) onstant.357*** ***.83***.459*** ***.195***.178*** -.26***.127*** (.959) (.771) (.619) (.126) (.752) (.53) (.39) (.191) (.475) Observations R-squared Note: Robust standard errors in parentheses. *** p<.1, ** p<.5, * p<.1. 24

25 Table 5. Explaining the Estimated Positions of British MPs Second imension (1) (11) (12) (13) (14) (15) (16) (17) (18) IEA IEA IEA NOMINATE NOMINATE NOMINATE O O O MP left-right self-placement.828**.223***.92**.538***.86***.525***.24***.25***.213*** (.414) (.394) (.411) (.13) (.946) (.128) (.523) (.374) (.488) onservative.228* *** ** (.125) (.312) (.762) (.163) (.213) (.19) abour.218*.441***.545***.576***.398*.986*** (.124) (.725) (.321) (.276) (.211) (.83) iberal emocrat -.311** -.419*** -.477*** (.125) (.25) (.31) (.122) (.212) (.842) SNP *** *** (.129) (.462) (.483) (.186) (.223) (.144) Party left-right location.47***.332***.177*** (.122) (.839) (.596) onstant -.219* -.131*** -.863*** -.57*** -.275*** -.87*** -.436** -.16*** -.286*** (.124) (.262) (.126) (.461) (.555) (.697) (.212) (.218) (.491) Observations R-squared Note: Robust standard errors in parentheses. *** p<.1, ** p<.5, * p<.1. 25

26 Table 6. Explaining the Estimated Positions of British MPs Second imension (28) (29) (3) (31) (32) (33) (34) (35) (36) IEA IEA IEA NOMINATE NOMINATE NOMINATE O O O MP left-right self-placement.957***.292***.957***.53***.65***.53***.185***.316***.185*** (.336) (.388) (.336) (.126) (.929) (.126) (.544) (.437) (.544) onservative.419***.531***.428*** (.224) () (.62) () (.547) () abour.381***.181***.592***.339***.418***.214*** (.171) (.11) (.358) (.277) (.52) (.275) iberal emocrat -.156*** -.233*** -.429*** -.526*** (.26) (.182) (.275) (.22) (.62) (.548) SNP *** -.77** -.134*** (.169) (.153) (.313) (.296) (.557) (.512) Party left-right location.45***.513***.414*** (.216) (.6) (.528) onstant -.376*** -.11*** -.62*** -.629*** -.183*** -.937*** -.46*** -.13*** -.78*** (.174) (.246) (.248) (.363) (.558) (.357) (.51) (.266) (.77) Observations R-squared Note: Robust standard errors in parentheses. *** p<.1, ** p<.5, * p<.1. 26