When far apart becomes too far apart: Evidence for a threshold effect in coalition formation

British Journal of Political Science

Volumn 35, Issue 3, 2005, Pages 383-401

  Warwick, Paul V ^a  

^a SIMON FRASER UNIVERSITY   (Canada)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 22544464729     PISSN: 00071234     EISSN: None     Source Type: Journal    
DOI: 10.1017/S0007123405000220     Document Type: Review

References (44)

1. This assumes that policy distance affects the utility of all parties. Most formal models make this assumption. If policy costs only accrue to government parties, however, a party might reject an offer if it does not include sufficient office benefits to offset the policy costs that joining would entail. This alternative characterization of policy costs anchors a model proposed by Sened (Ita Sened, 'A Model of Coalition Formation: Theory and Evidence', Journal of Politics, 58 (1996), 350-72). Like the approach pursued in this article, it implies that voters hold parties responsible for what they do in office; the problem is that it is unclear why these costs can be offset by office benefits, which mainly profit party leaders.
2. This factor has figured in the US voting literature since William Riker and Peter Ordeshook, 'A Theory of the Calculus of Voting', American Political Science Review, 62 (1968), 25-42.
3. Geoffrey Luebbert, Comparative Democracy: Policy Making and Government Coalitions in Europe and Israel (New York: Columbia University Press, 1986), p. 46.
4. Kaare Strøm, Minority Government and Majority Rule (Cambridge: Cambridge University Press, 1990), p. 45.
5. A policy located where horizons just touch could also have been considered acceptable for the purposes of forming a government, but it would mean that we would have to deal with the linguistic inelegance of talking about horizon intersections or tangencies throughout the article. Moreover, in creating the software to identify intersections, it proved easier to treat tangencies as non-intersections.
6. Not only might they indicate less flexibility than actually exists, but they might declare that there are strict bounds when in fact there are none.
7. The alternative of estimating limits on individual dimensions is not a viable strategy. Because the amount of compromise a party can accept on a dimension is likely to depend on how much it must yield on other dimensions, one can only get an accurate estimate of its limit on any one dimension if the party has been required to yield as much as it can on that dimension and nothing at all on the others. This situation is extremely unlikely to occur.
8. h, where p̂ is the predicted probability of government membership (given by standard computer programmes) and D is the distance between government and party ideal points.
9. Not only are there few cases where a party joins a government whose policy point is outside the party's horizon but, surprisingly, it turns out that parties typically join governments whose positions fall inside their horizons. A failure to do so would not be a violation of the horizon hypothesis even if it did occur frequently, but it would tend to undermine the viability of the logit method. The way in which the method is applied here, especially the use of the weighted mean positions of government parties to estimate government positions (as discussed below), partly creates this pattern: the governments a party belongs to will tend to appear closer because the party's own position contributes to the estimates of the governments' positions.
10. Horizons could, of course, be estimated on the basis of a different set of governments from the ones that are to be analysed. Further research indicates that a significant horizon effect does emerge under this approach. Unfortunately, to preserve enough cases for the test, horizon estimates must be based on small numbers of cases, which makes their measurement extremely imprecise. In addition, it is not clear that the approach solves the problem: if the use of horizon estimates from one subsample to explain the government formations of that subsample is illegitimate and if behaviour is essentially unchanged between the two subsamples, then one could argue that the use of those horizons to explain formations in the second subsample is also illegitimate.
11. Technically, this is an argument that the error term in the logit equation used to estimate horizons is not independent of policy distance. By extension, it applies also to the MJD method.
12. In a sense, this is akin to forecasting models in economics, where the concern is less with the correctness of the coefficient estimates than with the ability of the model to make correct predictions. Here, we pay no attention to what influences party horizons; we are simply concerned with determining where the horizons lie.
13. The initial coordinates for the parties are (3, 4), (4.5, 6.6), and (6, 4). They are altered by adding terms drawn from a normal distribution with mean of 0 and standard deviation of 1.
14. The decision to have six minority governments is arbitrary. It represents a smaller proportion than does occur in West European systems but, since the policy distance spanned by these governments is 0 (again, unlike reality), their presence tends to favour the policy distance hypothesis.
15. This usage is standard in the literature. Although usually justified on the basis of face validity alone, there is some empirical evidence to support it (Paul V. Warwick, 'Coalition Policy in Parliamentary Democracies: Who Gets How Much and Why', Comparative Political Studies, 34 (2001), 1212-36).
16. Horizons 3D © is a Java 2 SDK, Standard Edition, Version 1.4.1 programme for creating policy spaces and calculating distances and horizon intersections among parties. It is available free of charge to interested scholars from my website (http://www.sfu.ca/~Warwick).
17. D. McFadden, 'Conditional Logit Analysis of Qualitative Choice Behavior', Frontiers of Econometrics, edited by P. Zarembka (New York: Academic Press, 1973).
18. As Martin and Stevenson note, this model has the advantage that, since the formation situation (not the proto-coalition) is the unit of analysis, the existence of different numbers of proto-coalitions in different situations does not complicate matters (Lanny Martin and Randolph Stevenson, 'Government Formation in Parliamentary Democracies', American Journal of Political Science, 45 (2001), 33-50, pp. 38-9).
19. The countries are Austria (1949-90), Belgium (1954-95), Denmark (1945-96), Finland (1946-95), France (1945-95), Germany (1961-91), Iceland (1947-91), Ireland (1961-97), Italy (1958-91), Luxembourg (1945-84), Netherlands (1956-94), Norway (1961-97), Portugal (1976-95) and Sweden (1948-96). All formation situations for which we have adequate information (as specified below) are included, except for those that had a majority party or that did not involve a real choice of government, such as when a prime minister died or retired and the government continued in office.
20. These data are distributed in Ian Budge et al., Mapping Policy Preferences: Estimates for Parties, Electors, and Governments 1945-1998 (Oxford: Oxford University Press, 2001).
21. The analysis was conducted across all countries, with weights applied so that the number of cases contributed by each country was equalized. The decision to retain two dimensions is arbitrary, although not unprecedented (e.g. Ian Budge, David Robertson and Derek Hearl, eds, Ideology, Strategy, and Party Change: Spatial Analyses of Post-War Election Programmes in Nineteen Democracies (Cambridge: Cambridge University Press, 1987);
22. Christophe Crombez, 'Minority Governments, Minimal Winning Coalitions and Surplus Majorities in Parliamentary Systems', European Journal of Political Research, 29 (1996), 1-29). It can be justified on other grounds as well. The first two principal components account for approximately equal amounts of variance (5.8 and 4.9 per cent) and thus provide reasonable compliance with the assumption of equal dimension saliences. They are also readily interpretable. The items loading highest (0.4 and above) on the first principal component ('left-right') consist of negative mentions of free enterprise, incentives and economic orthodoxy, and positive mentions of peace, anti-militarism, labour groups and nationalizations. The highest loading items on the second component ('postmaterialism') involve positive mentions of internationalism, environment protection, culture, educational expansion and the protection of underprivileged minorities, and negative mentions of political authority. The principal component (factor) scores from this analysis provide the party positions.
23. The general pattern is one in which the more often a party participates in government, the larger its horizon (ceterisparibus). The effect is moderate except when the party always participates in government; for these parties, the horizons 'explode'. The adjustment was performed by regressing the parties' estimated horizons on their participation rates and a dummy variable identifying parties with a 100 per cent participation rate. The coefficient estimated for the dummy variable was then subtracted from these parties' horizons to move the latter to the trend established for the other parties. Note that, even with this adjustment, these parties still have very large horizons.
24. Charles Ragin, Fuzzy-Set Social Science (Chicago: University of Chicago Press, 2000), p. 111.
25. p-1, where p is the number of parties.
26. Since party positions are altered in the simulations, a proto-coalition's policy distance becomes its mean distance across the simulations. Little is changed in this process; in fact, the correlation between this version and the original distance variable is r = 0.995.
27. Of all governments in the dataset, 60.1 per cent are majoritarian. The policy distance spanned by majority coalitions exceeds that spanned by minority ones by an average of about one-half (0.58) a standard deviation, a difference significant at p < 0.0001.
28. See, e.g., Martin and Stevenson, 'Government Formation in Parliamentary Democracies' ;
29. Paul V. Warwick, 'Coalition Government Membership in West European Parliamentary Democracies', British Journal of Political Science, 26 (1996), 471-99;
30. and Paul V. Warwick, 'Policy Distance and Parliamentary Government', Legislative Studies Quarterly, 23 (1998), 319-45.
31. Because smaller scores indicate greater closeness to majority status, the expected effect of this variable on the odds of formation is negative.
32. The IIA assumption means, roughly speaking, that the choice between any two alternatives does not depend on what other alternatives are in the choice set. Following a suggestion by McFadden (D. McFadden, 'The Measurement of Urban Travel Demand', Journal of Public Economics, 3 (1974), 303-28),
33. Martin and Stevenson ('Government Formation in Parliamentary Democracies') tested this assumption by dropping a random 10 per cent of choices from each choice set and applying the Hausman test to determine if the estimated coefficients change significantly. This test was performed twenty times on each model and the average p-value for rejecting the IIA assumption for that model was calculated. I have followed the same procedure here and report the mean values in the tables.
34. Martin and Stevenson, 'Government Formation in Parliamentary Democracies', p. 47.
35. Martin and Stevenson, 'Government Formation in Parliamentary Democracies', p. 45.
36. The last-mentioned variable captures the hypothesis that minority coalitions should only be disfavoured if a proposed government must pass a formal parliamentary vote of investiture in order to take office.
37. Martin and Stevenson ('Government Formation in Parliamentary Democracies', p. 40)
38. measure a party's anti-system score by its total on CMP categories classified by Laver and Budge (Party Policy and Government Coalitions, p. 24) as 'anti-establishment views' and take the score of the most anti-system party in the proto-coalition as the proto-coalition's anti-system presence score.
39. Michael Laver and Kenneth Shepsle, Making and Breaking Governments (Cambridge: Cambridge University Press, 1996).
40. Martin and Stevenson ('Government Formation in Parliamentary Democracies') include Canada and Israel and exclude Finland, France and Portugal. In general, their data end in the 1980s, whereas the present data mostly terminate in the mid to late 1990s. Because this analysis uses their coding of pre-electoral pacts, Finland, France and Portugal are also missing here. Further testing shows that this does not bias the results in favour of the horizon hypothesis (if Models 2 and 3 of Table 3 are re-estimated without these three countries, both policy distance and the intersection rate receive slightly stronger coefficients).
41. Martin and Stevenson, 'Government Formation in Parliamentary Democracies', p. 47.
42. Policy distance is, in fact, reduced to insignificance. This is partly due to the fact that majority status and closeness to it are not included in the model specifications; as noted earlier, a correct assessment of the policy distance effect requires their presence. When they are included, policy distance approaches (MJD) or achieves (logit) significance. Interestingly, however, the intersection rate also becomes stronger in both models.
43. The correlation between MJD and logit-based intersection rates is r = 0.76.
44. An intersection is likely to exist in nearly twice as many proto-coalitions under the MJD method (31.2 per cent of all proto-coalitions) as under the logit method (17.9 per cent).