¹ For a review of the literature in the field and a useful bibliography see Bienen, H., ‘The Background to the Contemporary Study of Militaries and Modernization’, in Bienen, H., ed., The Military and Modernization (Chicago: Aldine Press, 1971), pp. 4–39Google Scholar. For a general view of the conflicting viewpoints expressed in the literature the following works may be useful: Nordlinger, E. A., ‘Soldiers in Mufti’, American Political Science Review, LXIV (1970), 1131–48Google Scholar; Feit, E., ‘Pen, Sword and People’, World Politics, XXV (1973), 251–73Google Scholar; Feit, E., Armed Bureaucrats (Boston: Houghton Mifflin, 1973)Google Scholar; Pinkney, R., ‘The Theory and Practice of Military Government’, Political Studies, XXI (1973), 152–66Google Scholar; Willner, A. R., ‘Perspectives on Military Elites as Rulers and Wielders of Power’, Journal of Comparative Administration, II (1970), 261–76Google Scholar; Kossok, M., ‘Changes in the Political and Social Functions of the Armed Forces in the Developing Countries, The Case of Latin America’, in Janowitz, M. and Van Doom, J., eds., On Military Intervention (Rotterdam: Rotterdam University Press, 1971)Google Scholar; and in the same book, Schmitter, P. C., ‘Military Intervention, Political Competitiveness and Public Policy in America: 1950–67’Google Scholar; and Welch, C. E. and Smith, A. K., Military Role and Rule (North Scituate, Mass.: Duxbury Press, 1974).Google Scholar

² For a critical review of the ways in which the literature on revolution, development, and instability has dealt with the phenomena of military coups and military regimes, see McKinlay, R. D. and Cohan, A. S., ‘Military Coups, Military Regimes and Social Change’ (paper presented at the annual meeting of the American Political Science Association in Chicago, 1974).Google Scholar

³ Data for the economic variables were collected on an annual basis. The main sources were United Nations, Statistical Yearbook (New York: United Nations, 1951–72)Google Scholar; United Nations, Yearbook of National Account Statistics (New York: United Nations, 1950–72)Google Scholar; and International Monetary Fund, International Financial Statistics (Washington, D.C.: International Monetary Fund, 1964–72).Google Scholar

⁴ Data for the military size and expenditure variables were drawn from The Military Balance (London: Institute for Strategic Studies, 1960–72)Google Scholar; Stockholm International Peace Research Institute, Yearbook of World Armaments and Disarmament (Stockholm: Almqvist, 1968–72)Google Scholar; Booth, Richard, The Armed Forces of African States, Adelphi Papers, No. 67 (London: Institute for Strategic Studies, 1970).Google Scholar

⁵ The main data sources for political variables were Europa Yearbook (London: Europa Publications, 1958–70)Google Scholar; Statesman's Yearbook (London: Macmillan, 1951–70)Google Scholar; Stebbins, Richard P. and Amoia, Alma, eds., Political Handbook and Atlas of World (New York: Simon and Schuster, 1970)Google Scholar; Whitaker's Almanack (London: J. Whitaker, 1951–70)Google Scholar; Keesing's Contemporary Archives (London: Keesing, 1950–70).Google Scholar

⁶ The type of coup is dichotomized into either (a) a coup in which a military government is replaced with a further military government, or (b) a coup in which a civilian government is replaced with a military one. The branch of the armed forces making the coup is categorized in terms of (a) the army alone, (b) the army in combination with either the navy or air-force, or (c) the navy or air-force without the army. The rank of the coup leader is categorized as (a) captain and below, (b) above captain to colonel, (c) above colonel to general. Civilian involvement is dichotomized in terms of either (a) direct participation by civilians in the execution of the coup, or (b) no participation. The degree of violence incurred during the coup is scored (a) none, (b) incidental (twenty or less killed), (c) substantial, (d) extensive (i.e. civil war).

⁷ The following types of executive may occur: (a) the single dominantly military executive (at least 90 per cent of positions are held by military personnel), (b) the single, mixed civilian-military executive, (c) the dual system in which there exists a dominantly military council together with a mixed executive. The type of civilian employed in the executive may be (a) none, (b) politicians, (c) non-politicians (i.e. technical experts) or (d) a mix of politicians and non-politicians. The use of an assembly or a party is dichotomized simply in terms of use or non-use. Additionally, on occasion we shall use an index of civilianization which is a composite index of the degree of civilian participation in the executive, in parties and in the legislative assembly. Each of these three variables is scaled from high to low participation. Each scale is of the same length, i.e. we assume parties, assemblies and executives are of equal importance. The composite index is the geometric mean of these three scales. The rationale for the multiplicative nature of the composite index is that we assume that high civilian participation on one scale will interact with high civilian participation on other scales.

⁸ The agent responsible for the termination may be (a) the military alone, (b) a coalition of military and civilian forces, or (c) civilians alone. The method of termination may be (a) an election, (b) a direct transfer to a civilian government, (c) a military coup, or (d) an uprising. The form of post-transfer government may be (a) another military government, (b) military leadership of a civilian regime, or (c) a civilian regime.

⁹ McKinlay and Cohan, ‘A Comparative Analysis of the Political and Economic Performance of Military Regimes’. In Table 1, MR refers to military regime, CRM refers to a civilian regime in a system which has experienced military rule, CR(900–) is a civilian regime in a system which has never experienced military rule with a per capita GNP of less than $900, and CR(900+) is the same type of regime system with a per capita GNP greater than $900.

¹⁰ A canonical correlation can be seen as representing a linear correlation between two sets of variables. Basically, the technique seeks to maximize a linear correlation between two variates representing each of two sets. The variates can be seen as being broadly analogous to principal components except that principal component analysis seeks; o produce variates with maximum intercorrelation. The canonical coefficients can be interpreted as a form of standardized regression coefficient in that they represent standardized contributions of each variable to the appropriate canonical variable.

¹¹ In all the multiple regressions we allow variables to enter the equation as long as the residual mean square is bejng reduced.

¹² R² = ·34; the regression equation is significant at ·008.

¹³ R² = ·56; the regression equation is significant at ·001.

¹⁴ R² = ·37; the regression equation is significant at ·01.

¹⁵ R² = ·20; the regression equation is significant at ·08.

¹⁶ R² = ·35; the regression equation is significant at ·05.

¹⁷ R² = ·78; the regression equation is significant at ·001.

¹⁸ The regression equation for the trade balance has an R² of ·33 (significant at ·01). The regression for the rate of growth of food production has an R² of ·17 (significant at ·03).

¹⁹ R² = ·60; the regression equation is significant at ·001 (military expenditure accounts for 42 per cent of the variance).

²⁰ For these T tests we have used a significance level of ·10.

²¹ For the canonical correlations, the variables were dichotomized (e.g. the method of termination is either election or non-election with coup and uprising becoming non-election).

²² Discriminant analysis can be viewed as a multivariate extension of ANOVA. Given a set of categoric groups and a profile of scores across n variables for each of these groupings, discriminant analysis attempts to produce composite linear functions consisting of the variables which maximally differentiate these groupings. Discriminant analysis will produce a number of functions equal to whichever is the smaller of the number of groups or the number of variables minus one. As a consequence of the discriminant analysis it is possible to allocate cases to particular groups from scores on a particular set of variables. The degree of success of the discriminant analysis can be gauged from the number of cases which given their scores on the functions would be allocated to a group to which in reality they do not belong. In our particular examples, were it to be the case that our critical variables were valid over time, then we would have a useful means for predicting the form of termination of current military regimes.

²³ The first discriminant function accounts for 86 per cent of variance and is significant at the 04 level. Seventeen per cent of cases are misallocated. The mean discriminant scores for the election, uprising, and coup are –·93, –·16 and 1·19 respectively.

²⁴ The discriminant function accounts for 86 per cent of variance and is significant at ·002. Thirty per cent of cases are misallocated. The mean discriminant scores for the uprising, election, and coup groups are – 1·45, –·49 and 1·37 respectively.

²⁵ The two discriminant functions account for 60 and 40 per cent of variance respectively. The mean discriminant scores on the first function for uprising, coup and election are –1·56, ·77 and ·06; the mean discriminant scores on the second function are –·51, –·65 and ·78.

²⁶ Mean Scores of Key Coup/Structure Variables by Termination Method

²⁷ This discriminant analysis misallocates only 7 per cent of cases. The mean discriminant scores on the first function for the uprising, election, and coup groups are –1·68, –·85 and 1·93 respectively. The scores on the second function are –1·41, –·22 and ·84.

²⁸ The discriminant function misallocates 17 per cent of cases and is significant at ·008.