1/15 Strategic Roll Call Vote Requests Fang-Yi Chiou Simon Hug Bjørn Høyland October 12, 2017
Overview 2/15 Motivation As roll call votes offer important behavioral information on parliaments and their members, we need a better understanding of when such votes are requested.
Overview 3/15 Introduction Votes may be taken by roll call in most parliaments The decision to record how individual legislators vote is unlikely to be random: 1 Disciplining tool 2 Signaling device
Overview 3/15 Introduction Votes may be taken by roll call in most parliaments The decision to record how individual legislators vote is unlikely to be random: 1 Disciplining tool 2 Signaling device How roll call? by procedural rule or by choice
Overview 3/15 Introduction Votes may be taken by roll call in most parliaments The decision to record how individual legislators vote is unlikely to be random: 1 Disciplining tool 2 Signaling device How roll call? by procedural rule or by choice We develop a theoretically informed statistical model to analyze roll call vote requests that takes strategy into account
Overview 4/15 Our contribution Quantal response equilibrium (QRE) model: actors take expectations of other actors behavior into account, as they can request, but not prevent roll call votes Our model out-preforms logit models in a Monte-Carlo study Replication study demonstrates the substantive importance of accounting for strategic behavior when studying roll call vote requests
Model 5/15 The Model Actors have preferences over whether or not to take the vote by roll call (but not over whether or not to request a roll call vote) When deciding whether to request such a vote, they take the expectations of the behavior of others into account
Model 5/15 The Model Actors have preferences over whether or not to take the vote by roll call (but not over whether or not to request a roll call vote) When deciding whether to request such a vote, they take the expectations of the behavior of others into account We assume a simultaneous move game where the probability of requesting a roll call vote on vote i by actor j is 1 Pr ij (r) = 1 + exp(τ λ j (U ij (R) h j (1 r (1) ih)))
Model 5/15 The Model Actors have preferences over whether or not to take the vote by roll call (but not over whether or not to request a roll call vote) When deciding whether to request such a vote, they take the expectations of the behavior of others into account We assume a simultaneous move game where the probability of requesting a roll call vote on vote i by actor j is 1 Pr ij (r) = 1 + exp(τ λ j (U ij (R) h j (1 r (1) ih))) λ j allows for actors to differ in their level of strategizing ( closeness to Nash, i.e. best response function) (λ j is only identified for j 1, provided U ij (R) comprises for all js some common elements (especially coefficients)). τ captures non-strategic costs of requesting
Monte-Carlo 6/15 Monte Carlo study: set-up Compare the estimates based on our QRE model with those from simple logit models (per party) Two summary statistics 1 Coverage 2 RMSE
Monte-Carlo 6/15 Monte Carlo study: set-up Compare the estimates based on our QRE model with those from simple logit models (per party) Two summary statistics 1 Coverage 2 RMSE 1000 votes 1000 simulations 2 parties: only party-specific effects 4 parties: 2 common variables, λ
Monte-Carlo 7/15 Monte Carlo results: 2 parties True βs rcr β mean cover mean rmse logit β logit cover logit rmse 1-2.00-2.36 0.45 0.18-1.78 0.02 0.06 2-0.50-0.45 0.72 0.02-0.46 0.07 0.01 3 0.20 0.18 0.64 0.02 0.18 0.09 0.01 4-2.50-2.79 0.50 0.17-2.12 0.00 0.15 5 0.30 0.25 0.68 0.03 0.27 0.12 0.01 6-0.20-0.16 0.62 0.02-0.18 0.18 0.01
Monte-Carlo 8/15 Monte Carlo results: 4 parties strategic (coverage) strategic (rmse) logit (coverage) logit (rmse) β 11 0.793 1.032 0.944 0.057 β.2 0.784 0.261 0.000 1.992 β.2 0.785 0.519 0.000 2.986 β 21 0.788 0.608 0.000 2.813 β 31 0.791 0.639 0.000 3.733 β 41 0.788 0.478 0.223 0.113 λ 1 0.796 0.552 λ 2 0.794 0.227 λ 3 0.791 0.154 Results from 1000 simulations of 1000 votes each with the following parameters: β 11 = 1, β.2 = 2, β.3 = 3, β 21 = -2, β 31 = -3, β 41 = 1, λ 1 = -1, λ 2 = 0, λ 3 = 1
Replication: Thierse 2016 9/15 Replication Thierse (2016) EUP: Overview Aim: Test monitoring / disciplining vs signaling as motivation for roll call vote requests Data on roll call vote requests in EP 6 (6001 votes)
Replication: Thierse 2016 9/15 Replication Thierse (2016) EUP: Overview Aim: Test monitoring / disciplining vs signaling as motivation for roll call vote requests Data on roll call vote requests in EP 6 (6001 votes) Variables Degree of dissent within committee (proposal) (RCV) Voting cohesion amongst political group committee members An extensive set of controls
Replication: Thierse 2016 9/15 Replication Thierse (2016) EUP: Overview Aim: Test monitoring / disciplining vs signaling as motivation for roll call vote requests Data on roll call vote requests in EP 6 (6001 votes) Variables Degree of dissent within committee (proposal) (RCV) Voting cohesion amongst political group committee members An extensive set of controls Cross-classified hierarchical logit (votes common across parties, votes nested in parties)
Replication: Thierse 2016 10/15 Replication: Key Results We estimate a model with common cost and keep 4 procedural controls common IPP: Committee split (negative and significant in Thierse): Positive and sign (at least at.10) for EPP, PES and Verts, while negative and sign for IND
Replication: Thierse 2016 10/15 Replication: Key Results We estimate a model with common cost and keep 4 procedural controls common IPP: Committee split (negative and significant in Thierse): Positive and sign (at least at.10) for EPP, PES and Verts, while negative and sign for IND Committee vote: winning side (negative and significant in Thierse): Positive (and sign) for EPP, PES, UEN, while negative (and sign) for GUE, IND and Verts
Replication: Thierse 2016 11/15 Differences in controls Policy position (no effect in Thierse): Negative for EPP and PES, while positive for ALDE, GUE, IND, UEN and Verts
Replication: Thierse 2016 11/15 Differences in controls Policy position (no effect in Thierse): Negative for EPP and PES, while positive for ALDE, GUE, IND, UEN and Verts Policy salience (negative sign effect in Thierse): Negative for EPP and Verts, while positive for GUE.
Replication: Thierse 2016 11/15 Differences in controls Policy position (no effect in Thierse): Negative for EPP and PES, while positive for ALDE, GUE, IND, UEN and Verts Policy salience (negative sign effect in Thierse): Negative for EPP and Verts, while positive for GUE. Reading (positive sign effect in Thierse): Negative for EPP, IND and Verts.
Replication: Thierse 2016 12/15 Substantive interpretation 1 The assumption of common effects for theoretically relevant variables is ambitious, and may mask real and interesting results. 2 In the EP is masks the differences between the large and smaller parties. 3 In particular, the two large groups, EPP and PES request roll calls for a different reason than the smaller groups.
Replication: Thierse 2016 12/15 Substantive interpretation 1 The assumption of common effects for theoretically relevant variables is ambitious, and may mask real and interesting results. 2 In the EP is masks the differences between the large and smaller parties. 3 In particular, the two large groups, EPP and PES request roll calls for a different reason than the smaller groups. The two big groups request roll call votes then the committee is divided and when their committee members are on the winning side. This may suggest that they request roll call votes in order to help pass votes when the committee is divided.
Replication: Thierse 2016 12/15 Substantive interpretation 1 The assumption of common effects for theoretically relevant variables is ambitious, and may mask real and interesting results. 2 In the EP is masks the differences between the large and smaller parties. 3 In particular, the two large groups, EPP and PES request roll calls for a different reason than the smaller groups. The two big groups request roll call votes then the committee is divided and when their committee members are on the winning side. This may suggest that they request roll call votes in order to help pass votes when the committee is divided. In contrast, the smaller parties request roll calls when they are on the losing side, regardless of the degree of division in the committee. They may do this to signal opposition to the votes passed.
Summary and further work 13/15 Summary It is very likely that roll call vote requests contain a strategic element We develop a statistical game-theoretic model for roll call vote requests Initial Monte-Carlo studies suggest that our model outperforms logit models
Summary and further work 13/15 Summary It is very likely that roll call vote requests contain a strategic element We develop a statistical game-theoretic model for roll call vote requests Initial Monte-Carlo studies suggest that our model outperforms logit models Replication study shows that this may be an issue in empirical research on roll call vote requests in EP
Summary and further work 14/15 To do list: A broader range of Monte-Carlo studies Easy-to-use software for estimating these models Applications in other settings
Summary and further work 15/15 References Thierse, Stefan. 2016. Going on record: Revisiting the logic of roll-call vote requests in the European Parliament. European Union Politics 17(2):219 241.