The paper
Sean J. Westwood, Shanto Iyengar, Stefaan Walgrave, Rafael Leonisio, Luis Miller, and Oliver Strijbis (2018). The tie that divides: Cross-national evidence of the primacy of partyism. European Journal of Political Research 57(2), 333–354.
doi:10.1111/1475-6765.12228 (opens in a new tab) · No public replication archive
Main text
p. 337, case selection
Published
Belgium and Spain have linguistic fractionalisation scores (Alesina et al. 2003) exceeding 0.5, while the United Kingdom and the United States rank near the bottom with scores of 0.05 and 0.25, respectively.
Correction
Belgium has a linguistic fractionalisation score (Alesina et al. 2003) of 0.54 and Spain one of 0.41, while the United Kingdom and the United States rank near the bottom with scores of 0.05 and 0.25, respectively.
Basis Language fractionalisation in the Alesina et al. (2003) dataset. Spain's ethnic (0.42) and religious (0.45) fractionalisation scores in the same dataset are also below 0.5.
p. 337, Great Britain
Published
In the 2015 elections, the combined vote for the Liberal Democrats, Scottish Nationalist Party, the UK Independence Party and the Greens amounted to more than 31 per cent of the electorate.
Correction
In the 2015 elections, the combined vote for the Liberal Democrats, Scottish Nationalist Party, the UK Independence Party and the Greens amounted to 29.0 per cent of the vote.
Basis UK shares of the vote in the official 2015 results: Liberal Democrats 7.9, SNP 4.7, UKIP 12.6 and Greens 3.8 per cent. With turnout at 66.4 per cent, these parties received the votes of about 19 per cent of the electorate.
p. 339, hypotheses
Published
From this, we derive the following hypotheses: xat [...] H3: Partisan bias will be larger where partisanship overlaps underling social divides.
Correction
From this, we derive the following hypotheses: [...] H3: Partisan bias will be larger where partisanship overlaps underlying social divides.
p. 341, Partisan affect in our cases (Figure 1)
Published
The average level of party polarisation is greatest in the United Kingdom, followed by the United States, Belgium and Spain. None of the between-nation differences are significant.
Correction
The average level of party polarisation is greatest in the United Kingdom, followed by Belgium, the United States and Spain (mean in-party minus out-party rating of 46.8, 44.0, 39.6 and 34.0 points). Five of the six between-nation differences are significant; the difference between the United Kingdom and Belgium is not.
Basis Means and 95 per cent confidence intervals read from the published Figure 1, averaged over parties in proportion to the party sizes in SI Tables 4, 9, 12 and 16. Unweighted averages give the same order (45.0, 42.4, 39.5 and 34.7). The tests treat the four national samples as independent and give each in-party minus out-party gap the largest standard error its plotted intervals allow (the sum of the two standard errors). The United Kingdom-Belgium difference is 2.7 points (p = .15); the other five differences have p ≤ .005. The next sentence, that the thermometer data do not suggest party polarisation depends on the strength of social cleavages, is unaffected.
pp. 342-343, Great Britain
Published
When compared with the control (non-partisan) condition, the co-partisan bonus for Labour supporters amounted to 8 per cent, while the out-party penalty imposed on Conservative supporters was four times larger (approximately 34 per cent). [...] Compared to their allocations in the control condition, Liberal Democrats penalised Conservative supporters by 19 per cent and Labour supporters by 9 per cent.
Correction
When compared with the control (non-partisan) condition, the co-partisan bonus for Labour supporters amounted to 8 per cent, while the out-party penalty imposed on Conservative supporters was 3.2 times larger (approximately 25 per cent). [...] Compared to their allocations in the control condition, Liberal Democrats penalised Conservative supporters by 13 per cent and Labour supporters by 3 per cent.
Basis SI Table 20. Labour supporters gave 4.20 to the control player, 4.53 to Labour supporters and 3.13 to Conservative supporters. Liberal Democrats gave 3.86 to the control player, 3.37 to Conservatives, 3.76 to Labour supporters and 4.12 to fellow Liberal Democrats. The published 34 per cent divides the penalty (1.07) by the allocation to Conservatives, and the published 19 and 9 per cent are the gaps between the co-partisan and out-party allocations divided by the control allocation. The corrected figures use the control allocation as the base, as both sentences state.
p. 346, Belgium (Figure 4B)
Published
When we group parties based on their ideology, we find that participants are consistently more trusting of individuals who support more proximate parties (see Figure 4B). Leftists give more to fellow leftists, and rightists give more to fellow rightists, regardless of Player 2’s regional/ethnic identity. For example, PS and SP.A supporters, which are both left-wing but are on opposing sides of the regional border, donated more to each other than to supporters of opposing ideologies.
Correction
When we group parties based on their ideology, we find that participants are on average more trusting of individuals who support more proximate parties (see Figure 4B), and this pattern comes from the left-wing parties. Leftists give more to fellow leftists, regardless of Player 2’s regional/ethnic identity. For example, PS and SP.A supporters, which are both left-wing but are on opposing sides of the regional border, donated more to each other than to supporters of opposing ideologies. Rightists do not give more to fellow rightists. MR supporters gave NV-A supporters 3.22, less than they gave PS (3.35) or SP.A (3.34) supporters, and NV-A supporters gave MR supporters 3.92, less than the 4.01 they gave supporters of SP.A, the left-wing party of their own region.
Basis SI Table 32 (fitted means, in euros). Averaged over the cells of the two right-wing parties, allocations to fellow rightists (3.57) and to leftists (3.58) are about equal; for the two left-wing parties the corresponding averages are 4.20 and 3.55. The NV-A pattern matches p. 345, where region trumped ideology for NV-A voters.
p. 346, Spain (design of the Basque study)
Published
Participants were paired with a Player 2 who was similar on either ethnicity or partisanship (cross-cutting condition), or similar on both attributes (reinforcing condition).
Correction
Participants were paired with a Player 2 who was similar on either ethnicity or ideology (cross-cutting condition), or similar on both attributes (reinforcing condition).
Basis The case description (p. 339) and the description of Figure 5C (p. 348) define the reinforcing and cross-cutting conditions by ethnicity and ideology. See also the correction to SI Table 43.
p. 346, Spain (bonuses and penalties)
Published
Unlike supporters of the national parties, supporters of Basque political parties allocated a co-partisan bonus significantly above the allocation to players lacking a partisan affiliation. The bonus amounted to 32 per cent for Amaiur supporters and 17 per cent for Basque Nationalist Party (PNV) supporters. Out-party penalties, however, proved consistent across the regional and national parties, ranging between 12 per cent for PNV supporters playing with PP supporters, and 29 per cent for Amaiur supporters playing with PP supporters.
Correction
Supporters of Basque political parties allocated a co-partisan bonus significantly above the allocation to players lacking a partisan affiliation. The bonus amounted to 32 per cent for Amaiur supporters and 17 per cent for Basque Nationalist Party (PNV) supporters. Among supporters of the national parties, PSOE supporters gave co-partisans 12 per cent more than players lacking a partisan affiliation, and PP supporters gave co-partisans 5 per cent less. Out-party penalties, however, proved consistent across the regional and national parties, ranging between 3 per cent for PSOE supporters playing with PNV supporters, and 29 per cent for Amaiur supporters playing with PP supporters. PNV supporters penalised PP supporters by 18 per cent.
Out-party penalties between supporters of Basque and national parties, relative to each group's control allocation (fitted means from SI Table 42, in euros). Participant Player 2 Control Player 2 allocation Penalty (% of control) Amaiur PP 3.80 2.68 29.5 PP Amaiur 3.97 2.88 27.5 PSOE Amaiur 3.70 2.74 25.9 PNV PP 3.90 3.19 18.2 Amaiur PSOE 3.80 3.15 17.1 PP PNV 3.97 3.48 12.3 PNV PSOE 3.90 3.55 9.0 PSOE PNV 3.70 3.59 3.0 Basis SI Table 42 (fitted means, in euros); bonuses and penalties are relative to each group's control allocation. The published 12 per cent is the penalty PP supporters imposed on PNV supporters (12.3 per cent). PSOE supporters gave co-partisans 4.16 against 3.70 to the control player; an approximate within-respondent test based on the SI variance components (SI Table 41) gives p = .02, although the marginal confidence intervals in SI Table 42 overlap. The same page reports significant co-partisan effects in three of the four groups, with PP the exception.
p. 350, Discussion
Published
The nature of the underlying social cleavages can either exacerbate or weaken partisan animosity. When social and partisan cleavages are reinforcing, partisans are especially distrusting of their opponents. Conversely, under cross-cutting cleavages, partisan hostility diminishes, as in the case of the Basque Country. However, the nature of the underlying cleavages appears to alter only the intensity and not the existence of partisan hostility.
Correction
We find no evidence that partisan animosity grows where social and partisan cleavages reinforce each other: partisan animosity is not larger in the socially divided cases (H3), and in the Basque Country partisans are similarly indifferent to the ethnicity of opposing partisans. Shared identities do add to trust. A Player 2 who matched the participant on both ethnicity and ideology received a bonus, and one who matched on only one of the two received a penalty (Figure 5C). The nature of the underlying cleavages does not alter the existence of partisan hostility.
Basis The Results (p. 342) report no evidence for H3, and p. 348 reports that partisans are similarly indifferent to the ethnicity of opposing partisans. SI Table 46 shows this for each group of participants. Basque-party supporters of Basque ethnicity gave 3.22 to Spanish-party players with Spanish surnames and 3.30 to those with Basque surnames. Across the four groups, the allocation to an opponent who also differed in ethnicity minus the allocation to an opponent of the participant's own ethnicity averages −0.03 euros, and no gap exceeds 0.08. Figure 5C compares doubly and singly similar players with the control player and has no category for opponents who differ on both.
p. 352, note 10
Published
10. The results from Spain are marginal fitted means from experiments that crossed the social and political identities of Player 2. The American social divide results are from an experiment in which race was crossed with party. The American partisan divide data are from a second experiment in which race was not manipulated.
Correction
- See note 6.
Basis Notes 6 and 10 print the same text.
Supplementary information
SI section 2.1, p. 10, trust game instructions
Published
You will play this game four times, with four different people.
Correction
You will play this game four times, with four different people. [Four rounds is the length of the U.S. games (1,523 allocations from 381 respondents in SI Table 29). The games in the other countries had more rounds: the published models record 5.3 allocations per respondent in Great Britain (SI Table 23), 6.0 in Belgium (SI Table 35) and 9.0 in the Basque Country (SI Tables 39, 41 and 45).]
Basis Note 2 of the article describes the instructions as identical across countries, and SI Table 18 lists six Player 2 profiles for Great Britain, seven for Belgium and nine for the Basque Country.
SI p. 33, Table 43 (model for Figure 5C)
Published
Table 43: Spain - Trust Allocations (C) [...] Cross-cutting −0.21∗ [−0.38; −0.03] [...] Reinforcing 0.37∗ [0.12; 0.63] [...] Num. obs. 4944 [...] Num. groups: key 412
Correction
Table 43 contains 12 rows per respondent (4,944 rows for 412 respondents). Each Basque respondent played 9 rounds (eight party-by-surname profiles and a control; SI Table 18), and the other Basque models (SI Tables 39, 41 and 45) use 3,708 rows. The 1,236 additional rows, three per respondent, do not correspond to distinct rounds, so if they repeat rounds the standard errors and confidence intervals in Tables 43 and 44 and in Figure 5C are too narrow. Treating the additional rows as duplicates and scaling the standard errors by √(4,944/3,708) = 1.15 gives approximate 95 per cent intervals of [−0.41, −0.01] for the cross-cutting coefficient and [0.08, 0.66] for the reinforcing coefficient; both still exclude zero.
Basis No data or code for this study are public, so the model cannot be re-estimated on one row per round; the rescaled intervals are an approximation.