¹ Marion J. Levy, Jr. defines modernization as a positive function of “… the ratio of inanimate to animate sources of power” (Modernization and the Structure of Societies [Princeton, N.J.: Princeton University Press, 1966], p. 35Google Scholar). Karl W. Deutsch defines social mobilization as “something that happens to large numbers of people in areas which undergo modernization, i.e., where advanced, non-traditional practices in culture, technology and economic life are introduced and accepted on a considerable scale” (“Social Mobilization and Political Development,” American Political Science Review, 55 [September 1961], 493–514CrossRefGoogle Scholar, at p. 493).

² On this point, see among others Hurewitz, J. C., Middle East Politics: The Military Dimension (New York: Praeger, 1969), p. 117Google Scholar; and Welch, Claude E. Jr., “The Roots and Implications of Military Intervention,” pp. 1–59Google Scholar in Soldier and State in Africa: A Comparative Analysis of Military Intervention and Political Change, ed. Welch, Claude E. (Evanston, Ill.: Northwestern University Press, 1970), at p. 35Google Scholar. This theme is clearly dominant in news reports of military coups during the 1960s, as summarized in the sources cited in footnote 36 below.

³ Levy, , Modernization and Structure, p. 603Google Scholar.

⁴ Levy, p. 603.

⁵ Pye, Lucian W., “Armies in the Process of Political Modernization,” pp. 69–89Google Scholar in The Role of the Military in Underdeveloped Countries, ed. Johnson, John J. (Princeton, N.J.: Princeton University Press, 1962), at p. 78CrossRefGoogle Scholar. The term “politicians in uniform,” which I have used in the title of the present paper, comes from this analysis by Pye (p. 69).

⁶ Pauker, Guy J., “Southeast Asia as a Problem Area in the Next Decade,” World Politics, 11 (April 1959), 325–345, at pp. 339–340CrossRefGoogle Scholar; Halpern, Manfred, “Middle Eastern Armies and the New Middle Class,” pp. 277–315Google Scholar in Johnson, (ed.), Role of the Military, at pp. 291–299Google Scholar; and Halpern, , The Politics of Social Change in the Middle East and North Africa (Princeton, N.J.: Princeton University Press, 1963), p. 258Google Scholar. See also Shils, Edward, “The Military in the Political Development of the New States,” pp. 7–67Google Scholar in Johnson (ed.), Role of the Military.

⁷ See among others McAlister, Lyle N., “Recent Research and Writings on the Role of the Military in Latin America,” Latin American Research Review, 2 (Fall 1966), 5–36Google Scholar; the essays in Welch (ed.), Soldier and State in Africa; and Bienen, Henry, “The Background to Contemporary Study of Militaries and Modernization,” pp. 1–33Google Scholar in The Military and Modernization, ed. Bienen, Henry (Chicago: Aldine-Atherton, 1971), at pp. 17–19Google Scholar.

⁸ Lieuwen, Edwin, Generals vs. Presidents: Neo-Militarism in Latin America (New York: Praeger, 1964)Google Scholar; Needier, Martin C., “Political Development and Military Intervention in Latin America,” American Political Science Review, 60 (September 1966), 616–626CrossRefGoogle Scholar, at p. 619; Sohn, Jae Souk, “Political Dominance and Political Failure: The Role of the Military in the Republic of Korea,” pp. 103–121Google Scholar in The Military Intervenes: Case Studies in Political Development, ed. Bienen, Henry (New York: Russell Sage Foundation, 1968)Google Scholar; and Nordlinger, Eric A., “Soldiers in Mufti: The Impact of Military Rule Upon Economic and Social Change in the Non-Western States,” American Political Science Review, 64 (December 1970), 1131–1148CrossRefGoogle Scholar, at pp. 1134–1138.

⁹ Huntington, Samuel P., Political Order in Changing Societies (New Haven, Conn.: Yale University Press, 1968), p. 221Google Scholar.

¹⁰ For data on this point, see Taylor, Charles L. and Hudson, Michael C., World Handbook of Political and Social Indicators, 2nd ed. (New Haven, Conn.: Yale University Press, 1972), pp. 26–28Google Scholar.

¹¹ On this point, see the data on the “importance of the indigenous middle class” in Adelman, Irma and Morris, Cynthia T., Society, Politics, and Economic Development: A Quantitative Approach (Baltimore, Md.: The Johns Hopkins University Press, 1967), pp. 30–33Google Scholar. Excluding Greece, Japan, Rhodesia, and South Africa from the calculations (which leaves us with 70 countries), the mean value for middle-class size is 70.7 for the 21 Latin American countries while the mean value for the remaining 49 countries is 46.5 (the scale ranges from 5 to 95).

¹² Schmitter, Philippe C., “Military Intervention, Political Competitiveness and Public Policy in Latin America: 1950–1967,” pp. 425–506Google Scholar in On Military Intervention, ed. Janowitz, Morris and Van Doom, Jacques (Rotterdam, Netherlands: Rotterdam University Press, 1971)Google Scholar; and Weaver, Jerry L., “Assessing the Impact of Military Rule: Alternative Approaches,” pp. 58–116Google Scholar in Military Rule in Latin America: Function, Consequences and Perspectives, ed. Schmitter, Philippe C. (Beverly Hills, Calif.: Sage Publications, 1973)Google Scholar.

¹³ Schmitter, , “Military Intervention and Public Policy,” pp. 492–493Google Scholar.

¹⁴ Adelman and Morris, Society, Politics, and Economic Development.

¹⁵ Nordlinger, , “Soldiers in Mufti,” pp. 1143–1144Google Scholar.

¹⁶ On this point, see Suits, Daniel B., “Use of Dummy Variables in Regression Equations,” Journal of the American Statistical Association, 52 (December 1957), 548–551CrossRefGoogle Scholar. For general discussions of covariance analysis within the regression framework, see among others Cohen, Jacob, “Multiple Regression as a General Data-Analytic System,” Psychological Bulletin, 70 (December 1968), 218–231CrossRefGoogle Scholar; Kmenta, Jan, Elements of Econometrics (New York: MacMillan, 1971), 419–423Google Scholar; and Johnston, J., Econometric Methods, second ed. (New York: McGraw-Hill, 1972), 176–207Google Scholar.

¹⁷ The Adelman-Morris data were originally collected for and are described in their book, Society, Politics, and Economic Development. The sample of countries consists of 74 non-Western, non-Communist countries, “which, as of 1950, were underdeveloped with respect to social and economic structure” and which, if not independent by 1962, were at least “recognized national units” (p. 9).

¹⁸ For full details on this variable, see Adelman and Morris, pp. 74–76. The numerical scores for the variable are described on pp. 14–15.

¹⁹ Numerical scores for the following variables are all described in Adelman and Morris, pp. 14–15.

²⁰ Adelman and Morris, p. 31.

²¹ Ibid, pp. 50–51.

²² A parameter estimate is considered statistically “significant” in this analysis if it is at least twice the size of its standard error of estimate: this criterion is the same as examining the t-ratios associated with each coefficient and rejecting those not statistically significant at approximately the .05 level. Strictly speaking, this test is not fully appropriate given that the sample of nations under consideration was not chosen randomly: however, it does provide a useful a priori criterion in view of the fact that I am not treating equation (1) as a fixed-effects model.

²³ Note that no parameters are estimated for Z ₁ or XZ ₁ (i.e., γ₁ and δ₁). This is consistent with the specification in equation (1) where γ₁, and δ₁, were constrained to a value of zero in order to avoid a linear dependency.

²⁴ These calculations are made by summing (α + γ₂) and (α + γ₃), respectively, since the γ_i are adjustments to the constant. Thus, the mean change in human resources for the lowest category of middle-class size is estimated by the constant (18.031); that for the intermediate category is (18.031 + 32.734); while the mean for the highest category is (18.031 + 61.365). For a discussion of dummy-variable regression analysis in terms of estimating conditional expectations, see Goldberger, Arthur S., Topics in Regression Analysis (New York: Macmillan, 1968), pp. 3–38Google Scholar.

²⁵ The statistical test for the adjusted slopes (rather than the adjustments themselves) is described in Johnston, , Econometric Methods, at p. 179Google Scholar. Using the notation of equation (1) above, and focusing on the adjustment to β from δ₃ first sum the estimates for the two coefficients (), and divide this sum by the estimated standard error

where the three terms in the standard error are obtained from the variance-covariance matrix of coefficients. The standard error of the adjusted slope for GNP per capita growth rates in the high modernization category is .172, which is larger than the adjusted coefficient (−.150). Similarly, the standard error of the adjusted slope for rates of change in human resources in the high modernization category is .162, which is again larger than the adjusted coefficient (−.158).

²⁶ Thus, for the regional-effects models, Z has four categories. The Middle East (Z ₁) is taken as the “excluded” category.

²⁷ Again, the slope here is given by summing β and δ₁, which, since the estimate for β is not significantly different from zero, gives (0 + .693) = .693.

²⁸ Taylor, and Hudson, , World Handbook, pp. 326–328Google Scholar. Data on GNP growth rates have too many cases with missing data for our purposes.

²⁹ Ibid., pp. 225–228, for the 1965 country values. The 1960 series is available from the World Handbook data released by the Inter-University Consortium for Political Research (ICPR) at the University of Michigan. Average annual rates of change were calculated as follows:

change = [1n(X₁/X₂)] [100/K],

where X₁ is the 1965 score; X₂ is the 1960 score; and K is the number of years between the two scores (in this case, 5). For details on the calculation of average annual rates of change, see Barclay, George W., Techniques of Population Research (New York: Wiley, 1958), pp. 28–33Google Scholar.

³⁰ In addition, by retaining all of the observations in the analytic design, the number of degrees of freedom retained for these tests is maximized, as it would not have been had separate regressions been estimated within each level-of-development or regional category. On this point, see Kmenta, , Elements of Econometrics, at p. 421Google Scholar.

³¹ For general discussion on this point, see among others Tukey, John W., “Causation, Regression, and Path Analysis,” chapter 3 in Statistics and Mathematics in Biology, ed. Kempthorne, Oscar et al. (Ames, Iowa: Iowa State College Press, 1954)Google Scholar; and Blalock, H. M. Jr., “Causal Inferences, Closed Populations, and Measures of Association,” American Political Science Review, 61 (March 1967), 130–136CrossRefGoogle Scholar.

To illustrate the differences involved here, consider the following table, which reports the within-region relationships between the Adelman-Morris measure of “changes in human resources” (i.e., school enrollment ratios, designated here as y) and their measure of the political strength of the military (designated here as x):

Note that the ratio of the within-stratum standard deviations does in fact vary substantially across strata (from .961 to 1.676), and that this in turn substantially affects the relative magnitudes of the coefficients. Where the correlation coefficients suggest that the strongest (negative) effect of military rule on changes in enrollment ratios occurs in Latin America, the regression coefficients show the strongest effect occurring in Asia.

Incidentally, while I have been able to reproduce the Adelman-Morris zero-order correlations (reported in Society, Politics, and Economic Development at pp. 281–283), I have had more difficulty reproducing the within-stratum correlations reported by Nordlinger in his Tables 2, 3, and 4. My estimates are close to his in the cases of the third and fourth tables, and the mean differences between the (absolute) magnitudes of his correlations and mine are .010 and .011 for his tables 3 and 4 respectively. However, the mean difference between the absolute size of the correlations in his Table 2 and mine for this table is .079.

³² Nordlinger also notes this problem in his preliminary discussion (“Soldiers in Mulfti,” p. 138).

³³ This is also a problem with a more recent analysis that has come to my attention as this paper is going to press. See McKinlay, R. D. and Cohan, A. S., “A Comparative Analysis of the Political and Economic Performance of Military and Civilian Regimes: A Cross-National Aggregate Study,” Comparative Politics, 8 (October 1975), pp. 1–30CrossRefGoogle Scholar.

³⁴ Adelman, and Morris, , Society, Politics, and Economic Development, p. 75Google Scholar. Italics added.

³⁵ To be included in the analysis, countries had to be independent before December 31, 1962.

³⁶ The primary data source for this variable is Keesing's Contemporary Archives (London: Keesings, Ltd.)Google Scholar, which is published weekly. This source was also cross-checked against others, including the country profiles in Morrison, Donald G., Mitchell, Robert C., Paden, John N., and Stevenson, Hugh M., Black Africa: A Comparative Handbook (New York: Free Press, 1972)Google Scholar, the data in Ruddle, Kenneth and Gillette, Philip, Latin American Political Statistics (Los Angeles: Latin American Center, UCLA, 1972)Google Scholar, the data on coups in Thompson, William R., The Grievances of Military Coup-Makers (Beverly Hills, Calif.: Sage Professional Papers in Comparative Politics, 01–047, Sage Publications, 1973), at pp. 68–70Google Scholar, along with the papers in the following books: The Politics of the Coup D'État: Five Case Studies, ed. Andrews, William G. and Ra'anan, Uri (New York: Van Nostrand Reinhold, 1969)Google Scholar; Bienen (ed.), The Military Intervenes; and Janowitz and Van Doorn (eds.), On Military Intervention. The cross-checking indicates that the primary source is quite accurate: in no cases were the data in Keesing's Contemporary Archives contradicted by the other sources.

³⁷ The 1970 data are from United Nations, Statistical Yearbook 1971 (New York: Statistical Office of the U.N., 1972), pp. 336–339Google Scholar and United Nations, Statistical Yearbook 1974 (New York: Statistical Office of the U.N., 1975), pp. 359–362Google Scholar. The 1960 data are described in Taylor and Hudson, World Handbook, and reported in the series released by the ICPR at the University of Michigan. This second source was also cross-checked against the data in the United Nations Statistical Yearbook of 1961, pp. 278–280Google Scholar. Average annual rates of change were calculated as described in footnote 29, where X₁ is the 1970 score; X₂ is the 1960 score; and K is the number of years between the two scores (in this case, 10).

³⁸ Recall that these measures of social change were: GNP per capita growth rates; gross investment rates; industrial growth; and improvements in agricultural productivity.

³⁹ UNESCO, Statistical Yearbook 1971 (Paris: UNESCO, 1972), p. 99Google Scholar. Data for both 1960 and 1970 are from the 1971 Statistical Yearbook, pp. 101–119, and from the UNESCO Statistical Yearbook of 1972, pp. 93–114Google Scholar. Average annual rates of change were again calculated as described in footnote 29 above: in a few cases 1968 or 1969 data were substituted for (unavailable) 1970 scores, and in these cases K assumed a value of 8 or 9, instead of 10. Data for Taiwan are missing.

⁴⁰ Data for the total number of physicians (1960) are from the ICPR version of the World Handbook and from the United Nations Statistical Yearbooks of 1962, 1963, 1964 and 1965. Discrepancies between these two sources were resolved in favor of the second one. Population data for 1960 are also taken from the ICPR version of the World Handbook. The 1960 physicians per 1,000 population ratios were generated from these variables. The 1970 physicians per 1,000 population ratios are derived from the United Nations Statistical Yearbooks of 1971, 1972, and 1973 at pp. 711–714Google Scholar; 747–750; and 719–722 respectively. Average annual rates of change were calculated as described in footnotes 29 and 39 above.

⁴¹ See Jackman, Robert W., Politics and Social Equality: A Comparative Analysis (New York: Wiley-Interscience, 1975)Google Scholar, especially chapter 2.

⁴² Data for radio receivers per 1,000 population for 1960 and 1970 are from the UNESCO Statistical Yearbooks of 1972 and 1973 at pp. 840–849Google Scholar and pp. 732–737 respectively. Some 1960 values are from the ICPR version of the World Handbook. Average annual rates of change were calculated as described in footnotes 29 and 39 above.

⁴³ Cf., Deutsch, “Social Mobilization and Political Development.”

⁴⁴ Olson, Mancur Jr., “Rapid Growth as a Destabilizing Force,” Journal of Economic History, 23 (December 1963), 529–552CrossRefGoogle Scholar.

⁴⁵ See footnote 37 above for the 1960 energy consumption per capita data sources. This classification provides three categories of approximately equal size. In addition, the cutoff points coincide with “breaks” in the data, in the sense that no country falls within ±5 kilograms (of energy consumption per capita) of the cutoff points.

⁴⁶ The modified model produced the following estimates (with standard errors in parentheses): = 6.505 (.582); = −.013 (.008); = −2.462 (.776); = −3.672 (.792); R ² = .257; = .226. I also estimated a model with both β and the δ_i set to zero, for which the estimated is .212, which is very close to the for either of the more complete models (for a definition and discussion of , see, e.g., Kmenta, , Elements of Econometrics, at p. 365Google Scholar).

⁴⁷ The countries are classified by the four regions described in the Appendix.

⁴⁸ This expectation is borne out in the regional mean values for energy consumption per capita, 1960. The regional means are: Africa, 50; Asia, 126; Middle East, 336 (this figure excludes Kuwait: with Kuwait included, the Middle East mean is 714); Latin America, 582.

⁴⁹ A key assumption of the ordinary least squares estimates of equation (1) is homoskedasticity, which implies

for all j (where the subscript j refers to the j^th observation). When this assumption is not met, the disturbances are said to be heteroskedastic, which implies

Although the ordinary least-squares estimates are unbiased and consistent under heteroskedasticity, they are not efficient. This means that the estimated standard errors are biased (for a useful discussion of this problem, see Kmenta, , Elements of Econometrics, pp. 249–269Google Scholar).

⁵⁰ These estimates are not reported here for reasons of space. Note that the distribution of X (Duration of military rule) is approximately normal for the 36 countries with X>0.

⁵¹ This second outcome could indicate either that military governments were continuous but that every 25 months a new junta replaced the existing one through a coup, or that each 25-month period of military rule was followed or interrupted by (at least) a brief period of civilian government.

⁵² On this point, see among others the discussion of “durability” in Eckstein, Harry, The Evaluation of Political Performance: Problems and Prospects (Beverly Hills, Calif.: Sage Professional Papers in Comparative Politics, 01–017, Sage Publications, 1971)Google Scholar.

⁵³ A modified model for school enrollment growth rates of the form discussed above in the text and in footnote 46 (with the δ_i set to zero) produced an estimate of β that was considerably less than twice its standard error. Similarly, a modified model for physicians to population growth rates with β constrained to a value of zero produced insignificant estimates for both of the δ_i. Note also that an application of the checks for heteroskedasticity discussed above indicates again that none of these conclusions is due to biased standard errors of estimate.

⁵⁴ Schmitter, “Military Intervention and Public Policy,” passim; and Weaver, “Assessing the Impact of Military Rule,” passim.