¹ Axelrod, Robert, Conflict of Interest (Chicago: Markham, 1970).Google Scholar

² Taylor, Michael, ‘On the Theory of Government Coalition Formation’, British Journal of Political Science, 11 (1972), 361–86.CrossRefGoogle Scholar

³ Axelrod, , Conflict of Interest, p. 150.Google Scholar

⁴ Axelrod, , Conflict of Interest, p. 168.Google Scholar

⁵ Axelrod, , Conflict of Interest, p. 167.Google Scholar

⁶ Axelrod, , Conflict of Interest, p. 177.Google Scholar

⁷ Felker, Lon, ‘Conflict of Interest Theory and Specific Systems: Postwar Italy and Weimar Germany’, Comparative Political Studies, XIV (1981), 357–70.CrossRefGoogle Scholar

⁸ DeSwaan, Abram, Coalition Theories and Cabinet Formations (Amsterdam: Elsevier, 1973), p. 288Google Scholar, and Taylor, Michael and Laver, Michael, ‘Government Coalitions in Western Europe’, European Journal of Political Research, 1 (1973), 205–48.CrossRefGoogle Scholar

⁹ Axelrod, , Conflict of Interest, p. 178Google Scholar, and Felker, , ‘Conflict of Interest Theory’, p. 365.Google Scholar

¹⁰ Taylor, , ‘On the Theory’, p. 372.Google Scholar

¹¹ But see Warwick, Paul, ‘The Durability of Coalition Governments in Parliamentary Democracies’, Comparative Political Studies, XI (1979), 465–98, p. 468CrossRefGoogle Scholar, and Axelrod, , Conflict of Interest, p. 178.Google Scholar

¹² Some Finnish cabinets are constituted as ‘expeditionary’ or ‘non-political’ governments which are appointed to office during periods of extreme crisis. Even where such cabinets include prominent party politicians, party loyalties are not expected to motivate decision making during the period of appointment. Given this ‘non-political’ mandate and the expectation of cabinet resignation with a return to ‘normalcy’, such cabinets are excluded from analysis.

¹³ See Dodd, Lawrence, ‘Party Coalitions in Multiparty Parliaments: A Game-Theoretic Analysis’, American Political Science Review, LXVIII (1974), 1093–117, p. 1094CrossRefGoogle Scholar, and Warwick, , ‘The Durability of Coalition Governments’, p. 494Google Scholar. For a view supportive of this interpretation, see Sanders, D. and Herman, V., ‘The Stability and Survival of Governments in Western Democracies’, Acta Politica, XXVI (1977), 346–77, p. 354.Google Scholar

¹⁴ Dodd, , ‘Party Coalitions’, p. 1094Google Scholar; Warwick, , ‘The Durability of Coalition Governments’, p. 468Google Scholar; and Taylor, Michael and Herman, Valentine, ‘Party Systems and Government Stability’, American Political Science Review, LXV (1971), 28–37, p. 29.CrossRefGoogle Scholar

¹⁵ Steiner, Kurt, Politics in Austria (Boston: Little, Brown, 1972), pp. 240–2.Google Scholar

¹⁶ It is clear that ‘left’ and ‘right’ scale interpretations allow meaningful comparisons to be made only in reference to locations on a single dimension. Where policy space is multidimensional, the functional nature of interactor affinity is relatively more difficult to describe.

¹⁷ Here we are arguing for measurement validation based on the intersection of face validity and an appeal to scholarly authority. This approach can be viewed as an approximation of criterion validation where the criteria used are the judgements of the scholarly community, particularly area experts. Certainly this approach should be expanded and improved upon by those same scholars through the systematic isolation and combination of indicators, making objective and precise the measurement of the ideological configuration of political systems. Criterion validation is to be preferred over construct validation in this context. Construct validation renders the theory unfalsifiable as a result of the fallacy of denying the antecedent.

¹⁸ Dodd, Lawrence, Coalitions in Parliamentary Government (Princeton, N.J.: Princeton University Press, 1976).Google Scholar

¹⁹ Dodd, , Coalitions in Parliamentary Government, p. 100.Google Scholar

²⁰ Dodd, , Coalitions in Parliamentary Government, p. 101.Google Scholar

²¹ In assigning parties to locations in policy space, Dodd acknowledges his indebtedness to work by Janda and Thomas. For documentation, see Dodd, , Coalitions in Parliamentary Government, p. 102.Google Scholar

²² DeSwaan, , Coalition Theories, p. 135.Google Scholar

²³ Browne, Eric C. and Dreijmanis, John, Government Coalitions in Western Democracies (New York: Longman, 1982).Google Scholar

²⁴ DeSwaan, , Coalition Theories, p. 131.Google Scholar

²⁵ An alternative view on this problem is presented by Sanders and Herman (The Stability and Survival'), who propose a measure of a coalition's ‘survival value’. This measures the proportion of time served in office in relation to the maximum possible time remaining in an interelection period. While this procedure has the virtue of ‘saving N’, Warwick has pointed out that there is no certainty of how long such cabinets would have survived had not elections occurred. Therefore, to treat a coalition that lasts only two months until an election in the same manner as one lasting three years (i.e., they both have the same survival value) would distort the meaning of duration. (Warwick, , ‘The Durability of Coalition Governments’, p. 468).Google Scholar

²⁶ There is an additional incongruity manifested in our dataset which should be noted. This is the fact that the length of an interelection period is not uniform over all cases. The norm is four years but in Sweden after 1970 and in Finland before 1955 it is three years. Italy has a five year interelection period. In our study, these differences do not occasion difficulty inasmuch as the general instability of Italian and Finnish cabinets resulted in none of the cases exceeding three years and the single coalition cabinet in Sweden after 1970 lasted only twenty-four months.

²⁷ Axelrod's computational procedure, applied to his Italian data, is given in a footnote, Conflict of Interest, p. 177Google Scholar. ‘There are an average of 2·6MCW coalitions at any one time, so the chance of a winning coalition being an MCW coalition is (2·6)/128 = 0·020. The expected number of MCW coalitions is thus only (0·020) (17) = 0·34. Ten MCW coalitions occurred, making them (10)/(0·34) = 29 times more likely than the average winning coalition.’ In these computations, the value 128 in the first denominator is the total number of combinations of parties possible for an eight-party system, divided by 2 to give the total number of possible majority (winning) coalitions. Alternatively, the value is calculated as 2^n–1. This calculation assumes (erroneously) an equal weight for all parties in the system. The value of 17 in the second calculation is the total number of cabinets considered in the test.

²⁸ Taylor, and Laver, , ‘Government Coalitions’Google Scholar; DeSwaan, Abram and Mokken, Robert, ‘Testing Coalition Theories: The Combined Evidence’, in Lewin, L. and Vedung, E., eds, Politics as Rational Action (Dordrecht: D. Reidel, 1980), pp. 199–215.CrossRefGoogle Scholar

²⁹ Correlation ratio analysis is also appropriate to measure the relationship between a nominal and an interval level variable. Indeed, eta-squared and the canonical correlation coefficient-squared can both be interpreted as the per cent of shared variance between two variables. The advantage of the discriminant analysis approach is two-fold. The classification procedure of the method allows for a closer examination of the structure of an extant relationship and also the isolation of incorrectly classified cases. This latter ability, not directly available when using correlation ratio analysis, enhances the method's multivariate capabilities by maximizing empirical induction through deviant case analysis. Such a process makes possible respecification of a model through the addition of other relevant characteristics.

³⁰ Although discriminant analysis assumes that independent variables have multivariate normal distributions and equal variance-covariance matrices within each group, the robustness of the method allows for violation without major consequence (See Klecka, W. R., Discriminant Analysis (Beverly Hills and London: Sage Publications, 1980), pp. 61–2.)CrossRefGoogle Scholar

³¹ As a test of statistical significance, Wilks's lambda has questionable utility for our research owing to the fact that our dataset is taken as the universe of formations for the countries included during 1945–80. Hence, probability estimates associated with U are difficult to interpret. We have, none the less, gone ahead with the test for two reasons. First, as noted, the value of lambda is an indication of association such that as U approaches 1·0, the failure of the independent variable(s) to discriminate categories of the dependent variable becomes increasingly manifest. Thus, U may be used in conjunction with r*² for interpretive purposes (Klecka, , Discriminant Analysis, pp. 39–40Google Scholar). Second, we may evaluate the test of U conservatively by making an assessment of the derived probability estimates. Here, if p is found to be insignificant, we would have to conclude that the value of the observed relationship (U) is insufficient to reject the null hypothesis for some putative universe of cabinets if it were assumed that our set of cases constituted an appropriate sample.

³² Like r, r* must be treated mathematically as symmetrical and can only be considered asymmetrical in the context of a theory. An assumption of asymmetry is unnecessary in this research since the theory asserts that MCW coalitions minimize conflict of interest and the reduction of conflict of interest is expected to produce MCW coalitions.

³³ The variable which classifies coalitions based on their congruence with theoretical descriptions is merely categoric and fails to involve orderedness of any kind. The category of interest in this research is minimal connected winning coalitions, as described by Axelrod. A coalition fitting his criteria, ideologically closed and rendered not winning by the exclusion of any included actor, is classified as MCW. Unlike many nominal variables, MCW does not intuitively transform to an interval level dichotomous variable. This is because the operational definition includes both size and connection. Therefore, a non-MCW coalition is not simply less MCW, but it can be either open, closed but too large, or closed and not winning. Thus, the descriptive status of coalitions can only be measured nominally in either the dichotomous case (i.e., MCW, non-MCW) or in some extended form (i.e., including categories MCW, open, closed/winning-too large, closed/not winning). The nominal nature of this variable dictates the use of discriminant analysis when coalition type is being treated as a dependent variable. Readers who would be willing to treat this dependent variable as ordered might opt for the use of probit analysis. They should, however, recall that probit and discriminant analyses of such variables yield similar outcomes. See Aldrich, J. and Cnudde, C. F., ‘Probing the Bounds of Conventional Wisdom: A Comparison of Regression, Probit, and Discriminant Analysis’, American Journal of Political Science, XIX (1975). 571–608.CrossRefGoogle Scholar

³⁴ We also measured conflict of interest as mean deviation of the weighted policy locations of parties where absolute values of deviation from the mean are used instead of squaring to deal with the direction of deviation. This measure was very similar to CI₁, (r = 0·95) and performed in basically the same way throughout the analysis.

³⁵ Axelrod, , Conflict of interest, p. 178Google Scholar; Felker, , ‘Conflict of Interest Theory,’ p. 365.Google Scholar

³⁶ It should be noted that our consideration of the hypothesis makes use of the comparative term ‘have longer duration’ rather than following Axelrod's probabilistic rendering: ‘the more likely the coalition will have long duration.’ This construction is consistent with Axelrod's own use when he employs the comparative statement: ‘The MCW coalitions that formed lasted on average 14 months each compared to only 8 months each for the others.’ (Axelrod, , Conflict of Interest, p. 178.)Google Scholar

³⁷ The null hypothesis of the two-tailed test, H₀: r = 0·0 is inappropriate since there is no theoretical prediction of an inverse relationship between conflict of interest and cabinet duration. Based on Axelrod's discussion, we expect longer duration associated with less conflict of interest (either CI = 0 or an MCW coalition). The direction of the inequality assumes that coalition status is coded MCW = 1, non-MCW = 0, and duration in months. Such a system predicts only ‘r’ greater than 0·0 and defines the one-tailed test described above.

³⁸ Further questions may be raised with regard to the execution of the t test analysis by the two authors. Felker incorrectly reports his degrees of freedom to be 17, when df is defined as n — 2 for this test (n = 18). See Cohen, J. and Cohen, P., Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences (Hillsdale, N.J.: Lawrence Zulbaum Associates, Inc., 1975), p. 49Google Scholar. Felker reports Axelrod as having committed the same error, but we have not found specific corroboration in his text. Using the data in Axelrod's Table 8–1 (Conflict of Interest, p. 178Google Scholar) and Felker's Table 1 (‘Conflict of Interest Theory’, p. 363Google Scholar), we attempted to replicate their findings. Our results are as in the table.

³⁹ While this decision violates the assumption made about the nominal nature of the MCW variable, discussed above in footnote 33, we have compared the results of our regression analyses with those derived from analysis of variance, a procedure which allows a parallel analysis with nominal variables. Comparison of our reported rs and the corresponding etas disclosed, in each instance, little or no difference in the derived magnitudes of the coefficients.

⁴⁰ To facilitate interpretation of the regression results, the nominal variable MCW_d (dichotomous) was coded, MCW = 0, non-MCW= 1. The trichotomous nominal variable (MCW_t) was converted into two dummy variables, majority non-MCW = 1 and minority non-MCW = 1, with MCW = 0 for both. This coding makes possible the comparison of CI = 0 and MCW = 0, both of which are considered equivalent to the theoretical construct of minimization of conflict of interest.

⁴¹ With respect to expectations associated with minority coalition formation, the logic of conflict of interest theory seems to present a contradiction. First, since minority coalitions are reduced in size from those having majority status, we might expect a decrease in the level of conflict of interest, which should be associated with an increase in duration. Conversely, inasmuch as the duration of a coalition depends on an ability to satisfy the decision rule over time, we might expect the duration of minority coalitions to increase as the size of the coalition approaches the decision rule. Thus, both increasing and decreasing size of minority coalitions might be expected to be positively (negatively) related to cabinet duration.

⁴² Country-specific factors are, in effect, controlled by an approximation of the most similar systems comparative method, which is implicitly employed by Axelrod and Felker by their choice of an analysis site confined to a single country in a bounded time period. Besides fallaciously generalizing their results from the specific to the general, their analyses fail to deal with the possibilities indicated by the bivariate country findings presented in the table.

In no single country does the bivariate ‘r’ between duration and MCW status produce a ‘p’ small enough to reject the null hypothesis of no relationship in the universe of coalitions. The magnitude of the ‘r’ is in only one case sufficient to explain more thatn 7 per cent of the shared variance. The inconsistency of the signs of the coefficients also remains unexplained by conflict of interest theory.

⁴³ Taylor, and Herman, , ‘Party Systems and Government Stability’Google Scholar; Dodd, , ‘Party Coalitions’, and Coalitions in Parliamentary GovernmentGoogle Scholar; Sanders, and Herman, , ‘The Stability and Survival of Governments’Google Scholar; and Warwick, , ‘The Durability of Coalition Governments’.Google Scholar

⁴⁴ Dodd, , Coalitions in Parliamentary Government, pp. 247–55.Google Scholar