THESIS This is to certify that the dissertation entitled A CYBERNETIC MODEL OF THE U.S. DEFENSE EXPENDITURE POLICY-MAKING PROCESS presented by Robin Frank Marra has been accepted towards fulfillment of the requirements for Ph.D. degree in Political Science Charles W. Ostran, Jr. Date 22> “11,2084 [9‘54 MSU is an Affirmative Action/Equal Opportunity Insulation 0- 12771 MSU LIBRARIES -_ RETURNING MATERIALS: Place in book drop to remove this checkout from your record. FINES will be charged if book is returned after the date stamped below. A CYBERNETIC MODEL OF THE U.S. DEFENSE EXPENDITURE POLICY-MAKING PROCESS BY Robin Frank Marra A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Political Science 1984 \ flg (7 (7‘4 / ABSTRACT A CYBERNETIC MODEL OF THE U.S. DEFENSE EXPENDITURE POLICY-MAKING PROCESS BY Robin Frank Marra When focusing on the body of literature which explicitly attempts to explain or develop models of U.S. military expenditures there is an emerging consensus which suggests that the United States does not react to the military expenditure behavior of the Soviet Union. This growing consensus is counterintuitive, particularly when compared with the statements of U.S. defense planners. This dissertation explores possible theoretical and methodological deficiencies which may account for this set of puzzling empirical results. An Organizational Process approach is adopted, and the defense expenditure process is viewed as consisting of four temporally ordered steps corresponding to the four major decisions made within the pro- cess. The four decision-making groups are the Military Services, Presi- dent, Congress, and the Department of Defense. Their respective de- fense decision outputs are the: Services' request, the President's recommendation, Congressional appropriation, and actual expenditures. The actors within this process are viewed as cybernetic decision makers. This approach has the following features: a) decision makers are boundedly rational; b) they respond to a limited number of envi- ronmental factors in a hierarchical fashion; and c) the structure of the hierarchy is a function of survival considerations. Robin Frank Marra This dissertation develops a multi-actor, multi-stage model of the U.S. defense expenditure policy-making process which is then operationalized and tested for the period of fiscal years 1947-80. A relatively_novel approach to the estimation of this model is intro- duced. Identified as multi-stage residual analysis (MSRA), it in- volves the use of the residuals from one equation in the system as the dependent variable in the next stage. The results obtained by this approach are compared with those generated by a more traditional, single-equation, multiple regression (MR) procedure. This study concludes that the hierarchic search structure sug- gested by the cybernetic approach is a reasonable approximation to the behavior of the major actors in the defense expenditure process. It was found that these actors respond not only to domestic political and economic variables, but also to factors located in the international environment, including the behavior of the Soviet Union. For my parents who never doubted that I would finish, though they often wondered when. For Maude and Jackson who continue to delight me each day with their antics and affection. For Harold who gave me his love, trust, and nuzzles. Most of all, this dissertation is dedicated to Jill, my partner in life. Without her love and support this project would never have been completed-nor would it have been as rewarding. ii ACKNOWLEDGMENTS With a project of this magnitude there are so many to whom one is indebted. Of those debts which can never be repaid in full, the largest is clearly owed to Professor Charles W. Ostrom, Jr. The rela- tionship between a dissertator and his major advisor is like a roller coaster-there are the dizzying heights, the plummeting lows, and the wild curves which seem to come out of nowhere. Ours was all this and more. For, in addition to the sage professional advice which he gave me, Chuck has been and continues to be one of my closest friends. At times it was difficult to judge which was the more valuable, his friendship or his advice. I shall always be thankful that we survived this dissertation without having to sacrifice one for the other. Brian Silver, Michael Schechter, and John Aldrich read earlier versions and drafts of this manuscript. For their time, patience, advice, and encouragement I am eternally grateful. I also wish to thank David Rohde and Philip Marcus who served on my committee briefly, but ably. Harriet Dhanak provided me with the technical assistance neces- sary to cope with MSU's computing system. She is an invaluable member of the Political Science Department who is often taken for granted. I hope that I never made that mistake. I am also thankful for her friendship and her banana bread recipe. iii iv I wish to acknowledge the Department of Political Science for the funding support which made the computer analyses possible. Of the staff of the department, special thanks are in order for Iris Richard- son and Karen Albrecht. They have helped me in many ways, many times. Professors Robert N. Wells, Jr. and J. Ansil Ramsay of St. Lawrence University were responsible for getting me interested in political science. They also continue to serve as reminders of what excellence in teaching really means. Thank you both. Peter Paluch, Margaret Gilkison, and Michael Davis deserve special acknowledgments. They have shared in my successes, have alternated as examples that hurdles in graduate school can be overcome, and have offered the type of support which only those in a similar situation could provide. Over these many years my family has continued to be a needed source of love, encouragement, and strength. They know how I feel and how grateful I am. I have also been fortunate to have had these same gifts extended to me by my wife's family. From the beginning they have always made me feel that I was a part of them. I am even grate- ful for their cautious queries as to how my writing was going. It is said that music soothes the savage beast. It also comforts the troubled mind of a dissertator. I wish to especially thank Leo Kottke for the many hours of pleasure which his music has given me. The final copy of this manuscript was quickly and professionally typed by Marilyn Henry. Her careful proofreading uncovered many potential inconsistencies, and her assistance is gratefully acknowledged. v One person who deserves more thanks than I can possibly express in this short space is Kriss Taya Ostrom. Though much of her contri- butions to this dissertation were behind the scenes, they have not gone unnoticed. Her love and friendship will always be treasured by me. Finally, I wish to thank my wife Jill. This dissertation is as much hers as it is mine. Not only did she willingly type earlier drafts, she also contributed the graphics which appear in this final version. Throughout it all she has given me the strength necessary to write another page when my mind was too confused and my body too weary. TABLE OF CONTENTS LIST OF TABLES O O C O O O O O O O O O O O O O O O 0 LIST OF FIG‘JRES O O O O O O O O O O O O O O O 0 L18 T 0F ACRONYMS O O O O O O O O O O O O O O O I O 0 Chapter I INTRODUCTION AND REVIEW OF THE LITERATURE. Empirical Findings . . . . . . . . . . Reaction to USSR?. . . . . . . . Effects of War . . . . . . . . . International Tensions/Crises. . Macroeconomic Factors. . . . . . Political Factors. . . . . . . . Bureaucratic/Organizational Factors. Theoretical Approaches . . . . . . Conceptualization of the Process . . . . . Characterization of Human Decision.Making. Methodological Issues. . . . . . . Dependent Variable . . . . . . . Method of Analysis . . . . . . . Empirical vs. Theoretical Causality. What Is To Be Done?. . . . . . . . II A CYBERNETIC THEORY OF THE DEFENSE EXPENDITURE POLICY-MAKING PROCESS. o o o o o o o o o o o o The Structure of the Process . . . . . . . A Cybernetic View of Behavior. . . . . . . . vi xi xii \ONO‘MUO ll 12 13 16 18 18 l9 19 20 22 23 27 Chapter III IV V vii A Cybernetic Model of Defense Expenditure PO 1 1 cy Making 0 O O O O O O O O O O O O A.METHODOLOGY FOR THE ANALYSIS OF THE Methodological Concerns . . . . . Autocorrelation . . . . . . . . Multicollinearity . . . . . . . Specification Error . . . . . . Seemingly Unrelated Regressions Methodological Trade-Offs . . . The MSRA Methodology. . . . . . CYBER.MDDEL OPERATIONALIZATION AND ESTIMATION OF THE CYBER MODEL 0 O C O C C O O O O O O O O O O O O C O O 0 Definitions, Operationalizations, sources 0 I O O O O O O O O O O Dependent Variables . . . . . . First-Order Essential Variables and Second-Order Essential Variables. Third-Order Essential Variables Data Operational Hypotheses. . . . . . . . . . Presentation and Interpretation of Results. . . A Comparative Analysis of MSRA and Multiple Regression. . . . . . . . . . . . . . . . . . Variable Operationalizations. . . . . . Estimation Procedure. . . . . . . . . . Multiple Regression Estimation Results. A FORECAST EVALUATION OF THE CYBER MODEL. . Goodness of Pit of the CYBER Model. . . . A Comparative Forecast Evaluation . . . . 37 50 50 53 55 57 59 61 66 73 73 74 76 8O 91 94 99 113 114 116 117 124 125 130 Chapter viii VI CONCLUDING OBSERVATIONS . . . . . The CYBER Model and the Conventional Wisdom . . Loose Ends: Problems and Prospects . . . . . . APPENDIX A - LIST OF CRISES OF CONCERN TO BOTH SUPERPOWERS, 1946-1980 APPENDIX B - FARRARPGLAUBER.ANALYSIS OF MULTICOLLINEARITY . BIBLIOGRAPHY 144 144 149 154 157 160 Table Table Table Table Table Table Table Table Table Table Table Table Table 3.1 4.1 4.2 4.3 4.4 4.5 4.6 5.1 5.2 5.3 5.4 5.5 5.6 LIST OF TABLES Properties of OLS Estimators in the Presence of Different Methodological Problems . . . . . . . . Bivariate Correlations between DECt and DEct-l' . . Coefficient Estimates for First-Order Essential Variab 1e (USEt-l) O O O O I O O O O O O O O O O O Coefficient Estimates, Standard Errors, and t-ratios for Second-Order Essential Variables . . Coefficient Estimates for Third-Order Essential variables (CUES t) O O O O O I O O O O O O O O O O Coefficient Estimates for "White Noise" Regressions . . . . . . . . . . . . . . . . . . . Coefficient Estimates, Standard Errors, and t-ratios for Multiple Regression Analysis of U.S. Defense Expenditure Decisions. . . . . . . . Bivariate Correlations between Actual and Pre- dicted Values of ADECt with Associated Surrogate R28 . . . . . . . . . . . . . . . . . . Bivariate Correlations between Actual and Pre- dicted Values of DECt ‘with Associated surrogate R28 0 O 0 O O O O O O O O O O O O O O O Decomposition of Explained Variation in Aggregate Defense Expenditure Decisions (DECt). . . . . . . Actual and Forecasted Values of Military Services’ Annual Change Requests, 1950-80 . . . . . . . . . Actual and Forecasted Values of President's Annual Change Requests, 1950-80. . . . . . . . . . . . . Actual and Forecasted Values of Congressional Annual Change Requests, 1950-80 . . . . . . . . . ix 65 100 101 103 111 112 118 126 127 128 133 135 136 Table 5.7 Table 5.8 Table 5.9 Table 5.10 Table 5.11 X Actual and Forecasted Values of DoD's Annual Change Requests, 1950-80 . . . . . . . . . . . . Actual and Forecasted Values of Services Requests ’ 1950.80 0 O I O O O O O O O O O O O 0 Actual and Forecasted Values of President's Recommendations, 1950-80 . . . . . . . . . . . . Actual and Forecasted Values of Congressional Appropriations, 1950-80. . . . . . . . . . . . . Actual and Forecasted Values of DoD Expenditures, 1950.80 O O O O O O O O O O O O O O O O O O O O 137 139 140 141 142 Figure 2.1 Figure 2.2 Figure 2.3 Figure 4.1 LIST OF FIGURES Annual Defense Expenditure Decisions. . . . . Annual Defense Expenditure Decision Changes . Hierarchic Screening Procedure for Actors in the U.S. Defense Expenditure Policy-Making Process 0 O O O O O O O O O O O O O O O O O Buffered Inner Environment. . . . . . . . . . xi 44 45 48 98 BLUE CBA CBACH CYBER DOD EV2 EV3 GLS MEDM MSRA OLS PBR PBRCH mass SIPRI sop SPD SPDCH SRQ SRQCH LIST OF ACRONYMS Best linear unbiased estimator Congressional defense budget appropriations Congressional defense budget appropriation changes Cybernetic (model) Department of Defense First-order essential variab1e(s) Second-order essential variab1e(s) Third-order essential variab1e(s) Generalized least squares Military expenditure decisionvmaking (process) Multiple regression Multi-stage residual analysis Ordinary least squares President's defense budget recommendations President's defense budget recommendation changes Reactive Linkage Mbdel Root mean square error Stockholm International Peace Research Institute Standard operating procedure Department of Defense military expenditures Department of Defense military expenditure changes Military Services' defense budget requests Military Services' defense budget request changes xii CHAPTER I INTRODUCTION AND REVIEW OF THE LITERATURE For over twenty years considerable effort and attention have been directed toward the study of the build-up of military expenditure levels among nations. As a result, the contemporary researcher in this field is faced with a number of choices: which of several theoretical approaChes to adopt; what mix of internal and external factors to in? clude; what methodologies to utilize, etc. This chapter will provide a critical overview of the major options available in each area. By proceeding in such a fashion I will be able to delineate what mdght be called the conventional wisdom of contemporary research in the area of "armsebuilding models" (Moll and Luebbert, 1980). By "conventional wisdom" I am referring to a set of empirical findings, theoretical approaches, and methodologies that have become established by their general usage or acceptance. I do not mean to imply, however, that there is either universal agreement or disagree- ment over any facet of the research in this field. In articulating this conventional wisdom I will be able to accomplish two distinct, though related, tasks. The first will be to provide a "sense of where we are" in terms of the current understanding of the factors and processes which govern the military expenditure behavior of nations. The second task will be to establish the areas which need further 1 2 examination. In other words, I will raise a number of questions which suggest where research in this field ought to be directed. It should be stated at the outset that this chapter will not pre- sent a comprehensive review of all of the research which comprises this body of literature. Such a task would be Herculean in scape, and the potential for obscuring major trends in a sea of research would be ' quite high. Nor is this essential to the task at hand. There are a number of excellent reviews already in existence (e.g., Zinnes, 1976; Moll and Luebbert, 1980; Russett, 1983), and to repeat these efforts would be counterproductive. Instead, this chapter will begin.with a survey of the major empirical findings which seem to characterize the research in this area. Empirical Findings A complete listing of these findings would reveal that there is more disagreement concerning the major explanatory variables than there is consensus. Moreover, a survey of these findings suggests that there is very little cumulative research being pursued. The empirical results of one study often have little or no influence on subsequent research. Given the enormity of the literature in this field, some structure needs to be imposed if one is to isolate major trends in the body of empirical research that has been produced. One obvious dichotomy is between international factors and domestic factors. Such a dichotomy, however, may obscure more than it reveals, given the host of different‘ factors whiCh fall under each heading. I would suggest that both international and domestic factors can be subdivided further into categories which permit a more manageable discussion. Specifically, I shall examine the following sets of international factors: those related to the behavior of a nation's major adversary; those which examine the effects of war or combat participation; and those which deal with overall tensions or crises in the international environment. Domestic factors include: macroeconomic variables; political factors; and those associated with bureaucratic forces. It should also be stated at this point that the focus of this review will be on the body of literature which explicitly attempts to explain or develop models of U.S. military expenditures. This decision reflects not only my own research interests, but also the work on this topic represents the bulk of the research in this field. Mbre impor- tantly, the observations which follow would not be significantly altered if one were to focus on the "arms-building" behavior of some other nation or group of nations. Given this orientation, the next section will examine the current conventional wisdom with respect to the effects which the behavior of the Soviet Union has on U.S. mili- tary expenditures. Reaction to USSR? There is a growing body of empirical research which suggests that the United States does not react to the military expenditure behavior of the Soviet Union. The following are typical of this emerging cons 6118118 3 4 . . . the non-process input external threat, measured by Soviet military expenditures, appears to have no effect on the MEDM process. This result, while not unexpected given the results from earlier arms race analyses . . . still appears counterintuitive. However, it adds support to the claim that the inertia for military spending is a function of domestic, nonrprocess factors (Majeski, 1983: 512). and: . . . the principal feature of this model-the presumed reactivity of U.S. military spending to Soviet defense outlays-—is not empirically vindicated provides little compelling evidence that an "arms race" embodies the pri- mary dynamic underlying U.S. defense expenditures (Cusack and Ward, 1981:460). The first observation by Majeski is based upon total or aggregate ex- penditure levels of the United States and the Soviet Union. Cusack and ward, on the other hand, are unable to find any relationship between the changes in military expenditures of the U.S. and the USSR. Nor are these two studies the only ones to have reached this same general conclusion. Others who have been unable to find any action- reaction dynamic between the military expenditure behavior of the U.S. and the Soviet Union include: Domke, Eichenberg and Kelleher (1983); Majeski and Jones (1981); Fischer and Crecine (1981); and Nincic and Cusack (1979). Ostrom (1978a) is one of the few who has been able to demonstrate a significant reaction by the United States to Soviet defense spending. As Majeski suggests, this growing consensus is counterintuitive, particularly when compared with the statements of U.S. defense plane ners. Former Secretary of Defense Harold Brown has addressed the issue of the Soviet threat quite clearly: 5 The most obvious and most significant of these threats is the global challenge posed by the only nation that rivals us in military power-the Soviet Union (Report 2: Secre- tagy 2: Defense 52 the Congress gg_the F: 1982 Budget, January 19, 1981). How is one to evaluate the growing empirical consensus in light of such statements? In addition, how much credence can one give to models which generate such counterintuitive results? What is more surprising is that, given the above, little research has been initiated which investigates other aspects of Soviet behavior to which the U.S. might be expected to react. Domke, Eichenberg, and Kelleher (1983) find positive effects on military expenditures (meas— ured in percentage changes) associated with a rise in EastAWest tensions. Except for this finding, reactions to other types of Soviet behavior have not been systematically investigated with respect to U.S. military expenditures. Effects of War There is some empirical evidence which suggests that war or active combat activity exerts a positive effect on levels of defense spending. There is less agreement, however, as to how "war" is to be measured. It has been operationalized in terms of: U.S. battle casualties (Ostrom, 1978a); U.S. battle casualties adjusted by a defense price deflator (Fischer and Crecine, 1980); as a war mobili- zation factor (Nincic and Cusack, 1979; Cusack and Ward, 1981); and as a dichotomous dummy variable (Ostrom, 1978a). With respect to this last conceptualization, however, Ostrom was led to the conclusion 6 that war does not have an impact on any of the policydmaking rules (the military services excluded since their request was already found to be a function of a war-related variable). While the presence of war and its positive effects on U.S. defense spending enjoys both substantive plausibility and empirical support, systematic analyses have been limited largely to the post- Korean era. The reasons for the exclusion of the Korean War period have not always been made clear. There appears to be a general cone sensus that the period surrounding the Korean War is sufficiently unique, such that any attempts to construct a general model which would encompass this period are not likely to prove successful. It is true that the onset of the Korean and Vietnam conflicts can be dis~ tinguished from one another by the sudden, unexpected outbreak of the former and the slower, more gradual involvement in the latter-but there is no compelling g priori reason why the Korean War period should be excluded from empirical analyses of U.S. military expendi- tures 0 International Tensions/Crises Finally, the effects of tensions or crises in the international environment have received scant attention. As such, there is little that can be said at this point, other than that it seems plausible that military spending increases would be desirable and/or necessary during a period of heightened international tensions. Furthermore, it is obvious that Presidents are cognizant of the larger interna- tional environment: 7. I am.recommending a significant increase in defense spending for 1977. . . . My request is based on a careful assessment of the international situation and the contingencies we must be prepared to meet. . . . We dare not do less. (Quoted in The New York Times, January 26, 1976.) While President Gerald R. Ford was undoubtedly basing part of his assessment on the likely future behavior of the Soviet Union, it should be recalled that 1975 was also a year which posed a number of foreign policy crises to his administration, e.g., Turkish threats to the unity of the NATO alliance, the civil war in Angola, and, of course, the fateful Mayaguez operation. It can be seen that there is a growing body of empirical evidence which suggests that the primary forces which affect the military expenditure behavior of the United States are not to be found in the international environment. The presence of war appears to be the only external input upon which there is any agreement. The next few sec- tions will examine the domestic (internal) side of the U.S. defense expenditure picture. Macroeconomic Factors There is even less agreement among researchers as to which, if any, economic factors should be included in explanations of U.S. defense expenditures. Some models incorporate no such factors di- rectly (e.g., Ostrom, 1978a), while others are almost exclusively built around economic variables (e.g., Fischer and Crecine, 1980). Perhaps the earliest incorporation of an economic term or 8 variable can be found in the original Richardson equations (cf. Rich- ardson, 1960). Not only did Richardson believe that nations built arms in response to the arms acquisition behavior of an adversary, but he also felt that the cost of existing armament levels imposed an economic constraint on future armaments. Indeed, many of the revisions in the basic Richardson model focused on how to more accurately char— acterize this economic factor (e.g., Caspary, 1967). Another, more recent approach to incorporating economic factors can be seen in the work of Fischer and Crecine (1980), who build upon the logic of the "Great Identity" (Huntington, 1961). This identity supposedly highlights the tensions between fiscal policy goals on the one hand, and defense and domestic program goals on the other. Defense and nonrdefense expenditures in a given year cannot exceed available revenues plus/minus some budget deficit/surplus. Fischer and Crecine suggest that the defense budgeting process takes place within the larger context of decisions concerning both total federal spending and the allocation of that spending between defense and domestic uses. Others have also attempted to incorporate this logic into the study of U.S. defense expenditures (e.g., Fischer and Kamlet, 1981; Domke, Eichenberg, and Kelleher, 1983). Even researchers who do not directly adopt the logic of the Great Identity are reaching the conclusion that economic factors are the principal forces behind U.S. defense expenditure policy making. Ninr cic and Cusack argue that variations in U.S. defense spending are related to real and desired conditions within the economy. Their basic conclusion is that 9 [I]f one removes the effects of war-time mobilization, it is clear that for the United States the principal driving forces in military spending dynamics were (1) the perceived utility of such spending in stabilizing aggregate demand, (2) the political or electoral value of the perceived economic effects arising out of such spending, and (3) the pressures of insti- tutional—constituency demands (1979:101). In a similar fashion, Cusack and Ward (1981:460) find that . . . U.S. military expenditures may be best explained by the political economic cycles through which the economy is manipulated and support is sought from the populace. . . . Both studies also raise the interesting notion that defense expendi- tures may be manipulated not only in response to economic concerns, but also to alter economic conditions for political and electoral motivations. The impact of political factors has not been studied widely, as will be shown in the next section. Political Factors Though the use of political factors in the context of U.S. de- fense expenditure policy making has been relatively limited, there has been some attempt to test hypotheses of this nature. One such hypoth- esis holds that defense expenditures are manipulated in the sense that they are timed in order to enhance electoral support. The major assumption underlying this work is that . . . the knowledge that defense spending can produce short- term improvements in income and employment can be turned to an incumbent administration's advantage through imposition of a cycle in the movement of governmental outlays that corresponds to the salience of major elections (Cusack and Ward, 1981:435). 10 This hypothesis, which stems from some of the substantive literature dealing with the existence of a political business cycle (e.g., Tufte, 1978), has not met with much empirical support. Other political factors which have received attention include: public opinion, the party of the President, and the effects of certain administrations. What is most interesting about these three factors is that each was operationalized as a dichotomous dummy variable. The results have been mixed. Fischer and Crecine (1980) were not able to conclude that there was an administration effect. The party of the President appears to have some effect on Presidential defense budget behavior. Specifically, Ostrom (1978a) found that Democratic Presi- dents tend to impose deeper cuts on the budget requests of the mili- tary services while simultaneously being more responsive to the Ratio Goal than their Republican counterparts. Finally, Ostrom (1978a) also found that a high degree of negative public Opinion visJE-vis defense spending tended to have a dampening effect on both the military ser- vices' request and the Congressional appropriation. Russett (1972) has also investigated how public opinion is related to military expenditures. Research on the effects of political factors on U.S. defense expenditures has been sporadic and inconclusive. While many seem to agree that there is a "politics" which underlies U.S. defense expend- iture.policy making, few seem very sure as to how these political factors might operate-or even what these factors might be. 11 Bureaucratic/Organizationa1 Factors With respect to bureaucratic or organizational factors, the major finding is that the b§§g_is the primary determinant of U.S. military expenditures. What this base is, however, varies across studies. The base has been characterized as: the previous year's expenditure total; the previous actor's defense expenditure decision; and as a fixed percentage of total expenditures. Regardless of the base employed there is a growing consensus that much of defense budgeting is inertial, i.e., that defense expenditures at time 2 can be understood as a fixed percentage of some base at time “5:1. The dominance of this base cannot be explained solely in empiri- cal terms. There are both theoretical and methodological reasons why this result is not unexpected. Each of these points will be consid- ered in more detail in the sections that follow. Before proceeding to an examination of the major theoretical approaches that have been adopted it seems prudent to try to come to some conclusions about the wealth of empirical findings concerning U.S. military expenditures. First, it seems apparent that there are a number of issues over which no clear consensus exists. VSecond, it seems equally obvious that the following observation by Mbll and Luebbert (1980:178) is not without an empirical basis: Domestic political—bureaucratic influences are more important than international rivalries in many cases. Empirical research which has been published during the subsequent 12 three-year period has done little to alter this basic conclusion. If anything, the observation that U.S. military expenditures are predomi- nantly a function of internal, domestic factors is more firmly en- trenched than ever. Part of the reason for this growing consensus' can be traced to the theoretical approaches that dominate this literature. Theoretical Approaches Existing explanations of the U.S. defense expenditure policy- making process may be categorized along a number of dimensions. One might distinguish between explanations on the basis of their charac- terization of the relevant environmental stimuli as: primarily international, primarily domestic, or a combination of the two. Such a classification scheme would tend to reinforce the conclusions reached in the previous section. Another approach would be to examine existing explanations in terms of their conceptualization of the policy-making process. By proceeding in this fashion it would be possible to identify those explanations which characterize the defense expenditure process as unitary, organizational, and bureaucratic. one might also attempt to distinguish among models on the basis of their underlying assumptions about decision makers and the decisionrmaking process. This section will identify different theoretical approaches on the basis of their conceptualization of the structure of the process and on the basis of their characterization of the decision makers which operate within that process. It will be suggested that, with 13 respect to these two criteria, there is little ground upon which to distinguish one set of explanations from another. Moreover, it will be argued that these basic similarities are impeding progress in this field and are at least partially responsible for the growing body of counterintuitive empirical findings. Conceptualization of the Process When speaking of a model's conceptualization of the process it is important to note that there are four interrelated questions which must be addressed. The first has to do with the type of policydmaking process which is being offered. Allison (1971) offers three types, and his conceptualization of the options available seems to dominate the literature. Most of the existing explanations of defense expenditure policy making could be described as falling within one of these three alter- native theoretical perspectives: (1) Rational Actor; (2) Organiza- tional Process (Politics); and (3) Bureaucratic Politics. The Rational Actor and Bureaucratic Politics explanations are at the two extremes. The former characterizes the process as if it were the result of the actions of a single decision maker, whereas the latter views the process as consisting of the actions and decisions of a numr ber of individuals. Somewhere between the two lies the Organiza- tional Process approach which views the policy-making process as if it were a "conglomerate of semi-feudal, loosely allied organizations, each with a substantial life of its own," and which sees the resulting l4 decisions "less as deliberate choices and more as the output of large organizations" (Allison, 1971:71). Of the empirical explanations of U.S. defense expenditure policy making with which I am.familiar, all could be described as being either Rational Actor or Organizational Process explanations. Arms race models seem to characterize the former, while more recent expla- nations (e.g., Ostrom, 1978a; Majeski, 1983) are representative of the latter approach. To date no one has successfully formalized and tested a Bureaucratic Politics Model. In addition to these three basic characterizations of the process one may distinguish explanations on the basis of the number of major actors or steps which each attempts to model. While arms race models deal with the behavior of a single actor, Ostrom's (1978a) Reactive Linkage explanation models the behavior of four major actors or steps. Finally, one may wish to consider such multi—step models in terms of the between-steps and within-steps relationships which they specify. One type of between-steps model views the defense expenditure decision from one step as the major input into the next step. An alternative conceptualization would be to view the actors or steps as independent of one another. The Reactive Linkage Model (Ostrom, 1978a) is char- acteristic of the former, whereas the latter strategy has not been adopted. One possible exception to this observation might be the Great Identity/Constant Shares Model (Fischer and Crecine, 1980) which focuses on the generation of Presidential defense budget re- quests. Whether this step is truly independent of the other steps in the defense expenditure process is not answerable since Fischer and Crecine do not model any of the other steps. 15 The Great Identity/Constant Shares Model is relatively unique in that it takes a single step and attempts to model various sub—steps or components which comprise the President's defense budget request. It is one of the few models which attempts to deal with the relationships which operate within steps. In terms of the conceptualization of the defense expenditure policy-making process it is no longer accurate to say that the number of actors in this process is limited to that of a single, reified actor. The increasing complexity of recent explanations of this pro- cess is largely a function of the adoption of an Organizational Pro- cess approach which, by its very nature, introduces greater complexity. The criticism that researchers have largely "blacksboxed" the state (Moll and Luebbert, 1980:161) must now be modified. By intro- ducing more steps or actors into the process this "black-box" is slowly being replaced. The logical question, however, is: What is this "blackrbox" being replaced with? I would suggest that, for the most part, this larger "black-box" has been replaced with a number of smaller "black-boxes." Just as arms race models have tended to reify the state, there appears to be a trend toward the reification of actors comprising the Organizational Process approach. What is lack- ing in many explanations is an explicit characterization of a set of assumptions about decision making and decision makers. Defense budgets are, ultimately, the product of human minds. As such, some attention should be given to this aspect of U.S. defense expenditure policy making. 16 Characterization of Human Decision Making There are three related questions which could be addressed within the context of this topic. I will suggest that: (l) the answer to the first question is usually given, but the question and its impli- cations are seldom considered; (2) rarely has the second question even been asked, much less answered; and (3) preoccupation with this ques- tion has all but obscured the first two. The first question has to do with the type of actors that are involved in the process. Are they individuals, small groups, organi- zations, etc.? In terms of the decisionrmaking behavior, does it matter whether one is talking about groups or individuals? Organiza- tions? Some attention must be given to the implications of the answers to these questions. The second question deals with the characterization of decision making itself. Most explanations of the defense expenditure process ignore this question, or else they seem to accept a description of decision makers as being "rational." There are, however, other char- acterizations which are plausible and should be given consideration. Steinbruner (1974), for example, identifies three conceptualizations of decision making which he terms the Analytic, Cybernetic, and Cog- nitive Process paradigms. Each makes different assumptions about the abilities and limitations of decision makers, and I think that there are definite implications in terms of the way in which one models the defense expenditure process. The failure of most modelers of this process to explicitly consider the characteristics of decision makers is the single greatest weakness of existing explanations. 17 Finally, considerable attention has been directed toward the specification of a set of decision rules which describe the behavior of decision makers. Given the failure of most researchers to give consideration to the nature of decision making it is difficult to evaluate whether or not these decision rules are accurate representa- tions of the ways in which decision makers actually behave. In addi- tion, it is often.assumed that these decision rules remain constant over time. Only Ostrom (1978a) and Lucier (1979) have explored situ- ations under which these decision rules might be altered, though Lucier's analysis was not concerned with U.S. defense expenditure policy making. This section has been brief, largely due to the fact that the issues raised here have not received proper attention in the litera- ture in this field. I think it would be accurate to conclude that the study of the U.S. defense expenditure policy-making process has not taken place within the larger framework of the study of human decision making, gzgg though the process is dominated by decision makers. While I would not cite this as the sole reason for the set of inconsistent and counterintuitive empirical findings which charac- terize this literature, I would suggest that it is a contributing factor. To be sure, some of the problems may be a function of data availability, data reliability, certain methodological concerns that will be addressed below, etc., but the lack of explicit characteriza- tions of decision makers can only exacerbate the intellectual conunr drum.which modelers of this phenomenon face. 18 Methodological Issues In this section I wish to raise a number of concerns which I feel have not received adequate attention in the literature. As most of these concerns are taken up in greater detail in the chapters that follow I shall raise only the major dimensions of these issues at this time. It should also be pointed out that these issues are not exclusively methodological concerns, but also reflect substantive and theoretical considerations. Dependent Variable There is no universal agreement even on what it is that research- ers are trying to explain, though most focus on aggregate levels of defense expenditures. This characterization of the dependent variable is consistent with the prevailing "blackrbox" conceptualization of the process. In a few instances the dependent variable has been viewed as the annual ghgggg_in defense expenditures (e.g., Nincic and Cusack, 1979; Cusack and Ward, 1981) or the percentage ghgggg in defense expenditures (e.g., Domke, Eichenberg, and Kelleher, 1983). The sub- stantive and theoretical implications of these various forms, how- ever, have not been explored by the authors who have adopted them. This is essential since it could be that "incrementalism" is a meth- odological manifestation associated with the use of aggregate levels, rather than a theoretical characterization of decision making. The variance in expenditure levels which can be "explained" by the previous 19 year's expenditure level is considerable, but does this represent a statistical or a substantive explanation? Method of Analysis Empirical analyses of U.S. defense expenditures are dominated by traditional multiple regression procedures. Problems of specifica- tion error, autocorrelation, multicollinearity, etc., need to be given greater attention, particularly since they may affect the types of substantive explanations which are ultimately produced. Considera- tion should also be given to other types of estimation procedures or approaches, e.g., nonrlinear regression, residual analysis, etc. These techniques should be enployed, though, if there are substantive or theoretical justifications, rather than just for the sake of experimentation. Empirical vs. Theoretical Causality Techniques which infer causal relationships on the basis of statistical associations should be rejected in favor of those which have a sound theoretical basis. Some have concluded, for example, that the United States does not react to the Soviet Union on the basis of "statistical causality analysis" (Majeski and Jones, 1981). In a similar fashion, caution should be exercised in accepting variables as explanatory factors on the grounds of statistical rather than sub- stantive robustness. 20 I would argue that each of these issues is important to the ex- tent that the types of explanations of the U.S. defense expenditure policydmaking process are as much a reflection of the underlying methodologies as they are a result of prior theoretical development. Nor is the list of methodological concerns limited to the three mentioned above. Other issues, such as data reliability, for example, have received more attention than those raised above, and for that reason were not included in this discussion. At this point it seems reasonable to try to pull together some of the major themes brought out in this chapter. This will be done not by reiterating what has already been said, but rather by suggesting research directions. What Is To Be Done? The previous overview of the literature which attempts to explain the processes that govern U.S. military expenditures has demonstrated that there is more disagreement than consensus. The contemporary re— searcher is faced with a set of empirical results which are inconclu- sive and, in some cases, counterintuitive. Increasingly, scholars are coming to the conclusion that the United States does not react in any significant manner to the behavior of the Soviet Union, despite the contrary claims of defense policy makers. While there is a growing trend toward Organizational Process explanations of defense expenditure policy making, unitary or Rational Actor models still enjoy great popularity. The major shortcoming, how- ever, concerns the widespread absence of an explicit characterization 21 of the decisionrmaking process and the attributes of the decision mak- ers who reside within that process. These theoretical deficiencies may account for the observation that: [Allternative models contending to explain a given arms phenomenon usually are based on such divergent assumptions and criteria that they seldom have been reconciled with each other. This situation has hindered validation of all models and their findings and presents a virgin field for secondary research efforts (Moll and Luebbert, 1980:178). What Moll and Luebbert fail to mention is that with respect to the characterization of decision making the assumptions are seldom made explicit or are entirely lacking. The failure to explicitly incorpo- rate a conceptualization of decision making may, for example, account for Ostrom's (1977a) conclusion that an arms race and an organiza- tional process model were analytically indistinct from one another. Closer inspection would reveal that each shares a similar set of implicit assumptions about the characteristics of decision makers. In the chapters that follow I will offer a model of the U.S. defense expenditure policy-making process based upon an explicit char- acterization of decision makers. 'The underlying approach is derived from the literature on cybernetic behavior, and it will be used to develop the aforementioned model. It is anticipated that this model, when operationalized and estimated, will resolve many of the problems identified in this chapter. CHAPTER II A CYBERNETIC THEORY OF THE DEFENSE EXPENDITURE POLICY-MAKING PROCESS It was suggested at the end of the previous chapter that one of the basic problems with the current literature in the area of defense expenditure processes is that most models that have been developed suffer from the lack of an explicit theoretical framework. It will be argued that a complete explanation of the defense expenditure process requires a number of essential components: a conceptualization of the process itself; a view of the individual decision maker residing within that process; and a characterization of the set of relevant environmental stimuli. Using these three components as a starting point this chapter will develop a model of the U.S. defense expenditure policy-making process that is based upon behavioral organization theory. The individuals who operate within this process will be viewed as "cybernetic" decision makers. The principle features of this characterization of decision? makers are: (l) decision makers are "boundedly rational"; (2) decision makers respond to only a limited number of factors in their environment; (3) the search of the environment proceeds in a hierar- chic fashion; and (4) the structure of the hierarchy is a function of 22 23 "survival" considerations. Both the structure of the process and the view of the individual decision maker will be shown to be quite comple- mentary. The third component, the set of relevant environmental stimuli, will be developed and incorporated into the proposed model. The Structure of the Process The underlying structure of the process that will be employed is based on behavioral organization theory (e.g., Simon, 1945; March and Simon, 1958; Lindblom, 1959; Cyert and March, 1963) and has been re- ferred to as an "Organizational Process" explanation (cf. Allison, 1971; Ostrom, 1977a; Ostrom, 1978a). The theoretical framework under- lying the Organizational Process approach is largely derived from the work of Herbert Simon and his associates. Two essential components of this behavioral theory are: the cognitive limitations of decision makers and the complexity of the environment within which they oper- ate. At this point the distinction between individual decision makers and groups of decision makers (organizations) starts to blur. Simon's argument is that in order to understand the behavior of organizations we first need to have some comprehension about the individuals which comprise them. Toward this and Simon takes exception with the classical rational model of behavior. He observes that [Tlhe capacity of the human mind for formulating and solving complex problems is very small compared with the size of the problems whose solutions are required for ob- jectively rational behavior in the real world--or even for a reasonable approximation to such objective rationality (1957:198). 24 If humans are not objectively rational does this mean that they are irrational? Simon does not suggest this, but rather offers a view of rationality which, unlike objective rationality, does not place unrea— sonable cognitive demands on decision makers. Simon uses the term boundedly rational to describe this "limited" rationality which humans appear to possess. Even with a conceptualization of bounded or constrained rational behavior it is still possible to observe behavior patterns which are quite complex. For Simon, complex behavior need not preclude the pos- sibility of limited cognitive abilities. The solution lies in discov- ering the source of the observed complex behavior. In Simon's (1969: 25) view the source is external to the decision maker: . . . a man, viewed as a behaving system, is really quite simple. The apparent complexity of his behavior over time is largely a reflection of the complexity g; the environ- gggg;in which he finds himself. (Emphasis added.) There are a number of ways in which this environmental complexity can be reduced or managed by decision makers. One way is through the development of Eglg§.g£.§hggb. These are routinized and general solu- tions to recurring types of problems. In a related manner, the bound- edly rational decision maker rarely is able to calculate the best course of action, rather he satisfices. In other words, the first acceptable alternative encountered is the one that will be chosen as the solution to the particular problem. The interaction of these two factors, the complexity of the envi- ronment and the limitations of the individual decision.maker, has been built into a theory of organizational behavior. Cyert and March (1963) 25 view an organization as a coalition of boundedly rational individuals with diverse goals. In addition, they observe that an organization is constrained by the uncertainty of its environment, the prob- lems of maintaining a viable coalition, and the limitations on its capacity as a system for assembling, storing, and utilizing information (1963:99). Given these constraints, organizations, like their human components, tend to be satisficing rather than maximizing entities. What this means is that the organization makes decisions primarily in response to a limited number of variables in the environment, using fairly simple, well articulated decision rules. These decision rules, some- times called standard operating procedures (SOPs) or rules of thumb, are reflective of the following basic principles of organizational behavior as articulated by Cyert and March (1963:102): (1) Avoid uncertainty: look for procedures that minimize informational needs; (2) Maintain the rules: having adopted a feasible set of decision procedures, abandon them only under duress; and (3) Use simple rules: rely on individual judgment to provide flexibility around simple rules. The combined effect of these three principles is that the complexities and burdens of organizational decision making are reduced consider- ably. Informational needs are kept to a minimum, decision rules which govern behavioral responses remain stable except under extreme circum- stances, and these decision rules are kept simple enough to insure their adaptability to changing environmental demands. As was mentioned earlier, the theory of organizational behavior 26 developed by Cyert and March portrays organizations as being quite similar to the individuals which comprise them. The limitations of individuals are reflected in the limitations of the organizations they create. One of the primary reasons individuals form organizations is to try to reduce the complexity of the environment which surrounds them. This reduction in uncertainty is achieved by factoring a par- ticular problem into a number of less complex problems. These, in turn, are dealt with by the various subunits within the organization. Each subunit is able to engage a number of standard behavioral pat- terns in the attempt to solve the subproblem. The following set of assumptions reflect the key elements of the theory of organizational behavior developed by Simon, Cyert, and March, and is adapted from Ostrom (1978d): (1) Human decision makers are boundedly rational; (2) In order to deal more effectively with the complexity in their external environment, boundedly rational decision makers form organizations; (3) Each complex decision problen is factored into a numr ber of subproblems which are dealt with by the sub- units of the organization; (4) Each subunit has a separate goal or set of goals- these goals reflect the roles of the subunits in the overall decision process; . (5) Decision makers in an organization focus on a small number of decision variables--the environment is simplified and only the most obvious information is gathered; (6) Organizational decision makers will make use of stan- dard operating procedures to cut down on the burdens of decision making; (7) The standard operating procedures are simple; and 27 (8) The standard operating procedures are stable in that they will be changed only if they do not work or in response to a major shift in a key environmental variable. This set of assumptions represents an explicit theoretical foundation for the conceptualization of the U.S. defense expenditure policy-making process. This Organizational Process approach suggests that there are a number of actors in the process, that each deals with a part of the overall problem, that routine solutions to recurring problems will generally be employed, and that decision makers will carry on a limited search of their environments. Although the theory of organi- zational behavior presented above is largely based upon a theory of individual behavior, there are still some remaining questions that need to be answered. For example, although we are assuming that decision makers in an organization focus on a limited number of var- iables in their environments, are there any particular types of fac- tors that they are more or less likely to attend to? What are some of the behavioral implications of bounded rationality? What determines whether or not a major shift has occurred in a key environmental variable? The next section will present a view of decision-making that will provide answers to these questions. A Cybernetic View of Behavior Karl Deutsch has called cybernetics ". . . the systemmatic study of communication and control in organizations of all kinds." Moreover, he sees it as a 28 . . . shift in the center of interest from drives to steer- ing, and from instincts to systems of decisions, regulation, and contrOlo o o 0 (1963376) Cybernetics can thus be viewed as the study of adaptive behavior, whether by individuals or by groups of individuals, i.e., organiza- tions. The notion of adaptation, however, immediately raises the question: What is it that we are adapting to? Relatedly, one might ask what purpose is served by adaptive behavior. Implicit in the study of adaptation is the study of process and chang . The concept of process is aptly conveyed by Herbert Simon in his distinction between state descriptions and process descriptions. The former characterize the world as sensed; they provide the criteria for identifying objects, often by modeling the objects themselves. The latter characterize the world as acted upon; they provide the means for producing or gen- erating the objects having the desired characteristics (1969:111). In other words, state descriptions are more like blueprints or photo- graphs while process descriptions are like recipes or differential equations. What is more important than the distinction between the two, though, is the way in which they are related to each other. In order for process descriptions to have any meaning they must be linked to certain state descriptions-namely, a problem state description and a goal state description. The former is a characterization of "where we are" whereas the latter is a description of "where we would like to be." Process descriptions can then be viewed as those activi- ties which eliminate the differences between the two state descriptions. 29 A process model of U.S. defense policy making, for example, should attempt to specify an existing state of affairs, i.e., the cur- rent state of our national defense capabilities, as well as our de- sired level of capabilities. The recent debate as to whether or not a "window of vulnerability" exists between the United States and the Soviet Union, as well as what to do about it, is an example of the relationship between state and process descriptions. Moreover, and this is a crucial aspect in terms of our national defense, a process description should tell us how to respond to the perceived differences between the two states of affairs. Earlier a question was raised pertaining to the goal of adaptive behavior. Simon provides a very explicit answer--surviva1. The distinction between the world as sensed and the world as acted upon defines the basic condition for the survival of adaptive organisms. . . . [Gliven a desired state of affairs and an existing state of affairs, the task of an adaptive organism.is to find the correlating process that will erase the difference (1969:111-112). To be successful, i.e., to survive, requires a continual translation between state and process descriptions. Much behavior can be under- stood as a complex means-ends chain which is designed to take one from a particular problem state to a given solution state. Of course, all of this is predicated upon the ability to recognize problems in the first place and, secondly, to be able to design solutions to those problems. The dominant theme underlying all cybernetic models is that of the decision maker as a boundedly, or adaptively, rational entity. Within the cybernetic decision-making paradigm there are a number of 30 basic premises which guide adaptively rational behavior. The first of these can be termed stability maintenance, and is most closely associ- ated with the work of Ashby (1960, 1964). The clearest example of this aspect of cybernetic behavior is that of "Ashby's cat" (Ashby, 1960:62). Ashby is trying to account for the behavior of a cat sleeping near a fireplace. Ashby notes that the cat does not appear to try to find the optimal spot near the fire by giving consideration to such factors as the amount of heat energy given off by the fire or the heat conductivity properties of the sur- rounding atmosphere. Rather, Ashby argues, the cat maintains or moni- tors a set of "essential variables," changing its position only when those variables move outside of some tolerable range. Thus, in observing Ashby's cat we would see it move away from the fire as more wood is added and the blaze begins to roar. Likewise, as the fire burns down the cat should move towards it. Underlying this form of adaptive behavior is the notion of sta- bility maintenance. The cat is trying to maintain an external body temperature that lies within a narrow range. Whenever this tempera- ture falls outside this range the cat adjusts its position accord- ingly. This simple systen remains stable only as long as the essen- tial variable, in this instance the external skin temperature of the cat, remains within the tolerable range. The cat's behavior is a function of its attempt to maintain the stability of this system. An- other simple cybernetic mechanism with which most people are familiar is the home thermostat. Again there is a monitoring device which attends to a small number of variables, in this case a single one, and which activates a response mechanism whenever the essential variable 31 falls outside of the tolerable range. To develop a defense analogy, it might be argued that military expenditures will remain fairly con- stant, i.e., within narrow bounds, unless some factor in the environ- ment upsets this "stability." One plausible factor might be the presence of a war. In such a circumstance one would expect defense expenditures to increase to meet this environmental demand, and then to return back to previous levels once the threat had passed. That U.S. military expenditures have never returned to predwar levels sug- gests that such a simple conceptualization might not be appropriate. The problem with such simple cybernetic devices is that they avoid the issue of complexity. The organism merely monitors a few critical factors, and each time one of these falls outside some narrow range a simple behavior pattern is triggered which, in turn, attempts to move the factor back towards an acceptable level. The organism is "blind" to the bulk of the external environment that is not directly connected with the essential variables. While this simple cybernetic mechanism might work quite well in explaining the behavior of a cat or a thermostat, many have suggested that it is not an adequate repre— sentation of human decision makers who must continually interact with an extremely complex environment. Simon is one who has tried to grapple with this problem. As noted in the previous section, Simon views the complexity of human behavior as due to not the complexity of the internal structure and organization of man, but rather the complexity of the external envi- ronment in which man must operate. The key question then becomes: How do humans react to and manage complexity? Simon's explanation is that we employ certain design principles which allow us to view the 32 complex in much simpler terms. Chief among these are the principles of hierarchy, near decomposability, and redundancy. Generally speaking, complex systems are those which consist of a large number of parts that interact nonsimply. Simon provides one key to understanding such systems: . . . c lexity frequently takes the form of hierarchy and [that hierarchic systems have some common properties that are independent of their specific content (1969:87). A way to understand complex systems is to exploit the hierarchic struc- ture of such systems by decomposing them into their component systems. Near decomposability suggests that the forces or activities within a given subsystem are stronger or more important than the activ- ities between different subsystems. To the extent that this is true it is possible to examine subsystems one by one in virtual isolation from one another. Redundancy can express itself in many forms. Three which Simon identifies are: (l) Hierarchic systems are usually composed of only a few different kinds of subsystems in various combinations and arrangements. (2) Hierarchic systems are . . . often nearly decomposable. Hence only aggregate properties of their parts enter into the description of the interactions of those parts. (3) By appropriate "recoding," the redundancy that is present but unobvious in the structure of a complex system can often be made patent (1969:110). What is the relationship between these design principles and adap- tively rational behavior? For one thing, it provides a framework for understanding how human beings behave and survive in a complex envi- ronment-they strive to view the world in simpler terms. The 33 simplifying tools at their disposal are those which I have just identi- fied: hierarchy, near decomposability, and redundancy. Moreover, these same design principles can aid us in our attempt to understand complex phenomena: we identify the hierarchic structure, note any redundancies which exist, and treat the non-redundant subsystems as single entities. In addition, we are largely freed from having to worry about the potentially large web of interactions which might exist among the various subsystems. Given this characterization of complexity we can begin to view human decision makers as "active Ashby cats." The "activity" revolves around the attempt to structure a complex environment simply so that ‘we can continue to exist and function within it. We are still some- what "blind" to large segments of the external environment, but we attempt to provide some organization to it, thereby reducing this com- plexity. Closely associated with this strategy is the assumption of uncertainty control. Unlike the analytic paradigm, the cybernetic paradigm argues that the decision maker avoids the overwhelming task associated with direct outcome calculations of alternatives. How is this done? Steinbruner offers the following answer: cybernetic mechanisms which achieve uncertainty control do so by focusing the decision process on a few incoming vari- ables while eliminating entirely any serious calculation of probable outcomes. The decision maker is assumed to have a small set of "responses" and decision rules which determine the course of action to take once he has re- ceived information to which he is sensitive. That is, decision rules associate a given action with a given range of "values" for the critical variables in focus. The "responses" are action sequences, of the character of a recipe, established by prior experience. They are pro- grams which accept and adjust to very specific and very limited kinds of information (1974:66-67). 34 The preceding passage suggests that cybernetic decision makers strive to maintain some sort of internal simplicity. How does the complexity of the environment affect this principle? Ashby, for example, argues that the complexity of the environment threatens the survival of the adaptive organism, unless that organism is able to develop a response repertoire which is as varied as the surrounding environment. It would appear that there are contradictory forces affecting the cybernetic decision maker. On the one hand the requirements of uncer- tainty control argue for internal simplicity, while on the other hand the survival of such organisms in a complex environment demands fairly elaborate response repertoires. It is the tension between these two demands which forms what Steinbruner calls a ". . . major drama of cybernetic analysis" (1974:69). The solution to this problem is quite simple: in order to deal more effectively with the complexity in the external environment, boundedly or adaptively rational decision makers form organizations. In so doing each decision maker is able to maintain his or her own internal simplicity by focusing on some limited dimension of the ex- ternal environment. The organization as a whole has the adaptive capacity sufficiently diverse to match the variety found in the envi- ronment by means of an internal complexity which is not a property of the individuals of the organization, but rather a property of the collective. Essentially the response repertoire of the whole is much greater than any single individual repertoire. For that matter, given that these individual behavioral responses can be combined in a variety of ways, it is quite likely that the organization's response repertoire will be far greater than the sum of the individual ones. 35 It can be seen that there is natural dovetailing between the Organizational Process perspective and the cybernetic view of behavior. It should come as no surprise since there is quite a bit of overlap in terms of the relevant literature. The primary assumptions of the Organizational Process approach have already been enumerated. It would, however, be useful to summarize the major premises underlying the cybernetic paradigm. (l) Cybernetic decision makers are boundedly or adaptively rational. (2) Much of the external environment is ignored: cyber- netic decision makers conduct only limited searches. (3) "Survival" is the predominant goal of cybernetic organ- isms, and it also directs the environmental search. (4) Given the predominance of survival considerations, cybernetic decision makers react to stimuli in a hierarchic fashion; the nature of this hierarchy is a direct function of the extent to which a given factor or set of factors is "survival threatening." The notion of limited search, combined with the assumption of bounded rationality, implies that change does not occur merely for the sake of change. Invoking the ceteris-paribus assumption, one should not expect to witness a marked change in behavior unless one or more of the essential variables is perceived to be outside the tolerable range set by the decision maker. In other words, the behavior at time 5 should be very close to the behavior at time t-l. This is to be ex- pected, given the limited cognitive abilities and limited search envi- ronments posited by the cybernetic view of decision making. The range of possible reactions which defense decision makers display is a function of two factors: the spatial proximity of the 36 decisionrmaking unit to the external environment; and the perceived importance or salience of the environmental stimulus to the specific decisionsmaking group. By "spatial proximity" I mean to suggest that some actors in the defense expenditure policy-making process are more isolated from the external environment than other actors are. The differing degrees of isolation are, to a large degree, a function of two factors: the primary roles which each actor or set of actors has adopted; and the size of the specific decisionvmaking group. The salience of different environmental stimuli can also be attributed to the roles of each actor as well as to the "position" of each actor within the overall organizational structure. ‘What emerges from a com? bination of the concepts of spatial proximity and stimulus salience is what I will call a "funnel of constrained reactivity." It will be argued that the defense expenditure policy-making process is such that the range of possible behavioral responses is successively narrowed as one proceeds through the process. The next section will present a model of the U.S. defense expend- iture policy-making process which incorporates the essential features of both the organizational process perspective and the cybernetic theory of decision making. This model will then be examined in order to note differences and similarities with existing explanations of defense expenditure policy making. It will be argued that the model developed below is a more accurate formalization,of the theoretical principles outlined in this chapter. 37 A Cybernetic Model of Defense Expenditure Policy Making Before presenting the formal structure of this cybernetic model, hereinafter referred to as the CYBER.Model, a brief overview of the defense budgetary process is required. The major work on a defense budget begins approximately fifteen to eighteen months before the start of that budget's fiscal year. Prior to 1976 the federal gov~ ernment used a fiscal year which ran from July 1 through June 30. Then, beginning in 1976 and following a three-month transition period, the federal government changed the onset of the fiscal year to October 1. For example, the fiscal year for 1984 (FY 1984) began on October 1, 1983 and will end on September 30, 1984. This means that the initial work on the FY 1984 defense budget began sometime in July 1982. Although numerous individuals participate in this lengthy budget pro- cess, the behavior of only the major actors will be modeled. The conceptualization of the defense budgetary process offered below is not a novel one, and for that reason the discussion of it will be rather brief. Following Stromberg (1970), Ostrom (1978a), Majeski (1983) and others, the defense expenditure policy-making process will be viewed as consisting of four temporally ordered steps corresponding to the four major decisions made within the process. Each decision and the accompanying decision output is linked with a distinct actor or set of actors within the process. The behavior of each group can be seen as the result of the roles and responsibilities which each possesses. The four decisionemaking groups or units within the process are: 38 military services, President, Congress, and the Department of Defense (DoD). Their respective decision outputs are the Services' Request (SRO), the President's Budget Recommendation (PBR), the Congressional Budget Appropriation (CBA), and the DoD Expenditure (SPD). Several preliminary points should be made. First, the four actors identified above differ somewhat in terms of their internal composition. The number of participants ranges from a small number (e.g., the President and his key advisers) to potentially several hundred (e.g., the Congress). The number of decision makers responr sible for the generation of the Services' Request is not widely known, nor is the size of the decision—making group in the Department of Defense. I would suggest that, in the absence of information to the contrary, one could assume that a reasonably small number of indi- viduals are intimately involved with the generation of the various decision outputs. Moreover, I do not expect that the size of each of these groups varies considerably--either within groups or across groups. Implicit in what I have just said is a conceptualization of the second decision-making unit which contains more than one indi- vidual. The President is undoubtedly the primary actor in this group, but given the overall complexity of the annual defense budget it does not seem reasonable that he is solely responsible for its production. Second, while the size of the four decisionrmaking groups may be taken to be relatively similar, their respective roles are quite dif- ferent. The implication of this statement is that observed behavioral responses can be expected to differ, given similar environmental stim- uli, and that these differences can be attributed for the most part to the roles which each group possesses. In addition, the roles are such 39 that they effectively constrain the range of behavioral responses which each set of actors can be expected to display. The military services are primarily responsible for seeing that defense spending is kept at a level sufficient to meet the security needs of the United States. This is a rather broad organizational goal, and the latitude in behavior which it suggests for the services is equally broad. Defense requirements as interpreted by the services include the ability to meet various levels of conflict ranging from large-scale war to more limited combat activities. Moreover, the ser- vices are expected to maintain a level of defense spending which will deter and/or contain the behavior of the U.S.‘s major adversaries. In addition to these overt roles, the services have also adopted the role occupied by agencies in a budgetary context (Wildavsky, 1964). Their position as an agency or group of agencies within the larger federal structure suggests that they have more flexibility to pursue "agency interests," even to the virtual exclusion of the interests of other, nonrdefense agencies. Presidents do not, however, enjoy such a luxury. Fraposed budget expenditures for defense must be made within the context of the entire federal budget. As such, Presidential latitude is constrained to a greater degree than is that of the military services. As chief execu- tive officer, the President must consider the overall fiscal climate within which his budget proposals must be made. Such a situation imposes additional constraints on the range of reactions which Presi- dents display. The President also has a conflicting role, that of Commander-ianhief, which requires that he be able to weigh the trade- offs between fiscal and strategic priorities. This provides him with 40 more flexibility, but still not to the degree of that possessed by the services. There is some question as to whether Congress adopts a program? matic (Kanter, 1972) or a fiscal (Korb, 1973) role toward defense expenditure policy making. What is important is not whether Congress adopts either of these roles to the exclusion of the other, for examples of both strategies can be found, but rather that Congress has the additional responsibility for setting appropriations levels each year. The constraints imposed by this requirement are such that Congress is able to demonstrate even less variability in its behav- ioral response repertoire than either of the other two actors. The Department of Defense is even.more constrained given that it has the legal obligation to spend most of its appropriated funds. From a bureaucratic budgetary perspective, there is an additional incentive which suggests that subsequent increases in appropriations will be difficult to justify if the Department starts to amass large budgetary reserves. That DoD has both obligated and unobligated funds in a given fiscal year does provide some discretionary expenditure power, though one should expect this group to be the most constrained by its actual and perceived roles. Finally, while the groups themselves are temporally ordered, this does not imply that their respective decisions are temporally depend- ent upon each other. This represents a fundamental shift from the logic of the Reactive Linkage Model. There the decision output from each actor fed directly into the decision calculus of the next actor in the process, e.g., the President's budget recommendation resulted from his application of a cutting rule to the initial Services' request. 41 Such a conceptualization, while quite plausible, implies a temporal dependence among the various groups such that each is largely re- strained from acting until the previous decision maker has completed his or her task. I am.not entirely convinced that this is the only plausible interpretation of the defense expenditure policy-making process. Furthermore, it seems equally reasonable to assume that each group begins an independent assessment of defense needs, if only in a rough, crude fashion. The model deve10ped below does not completely reject the logic of the Reactive Linkage Model, but instead suggests a shift in the priority which each actor gives to the decision output of the previous actor in the process. Based upon the discussion above, the following cybernetic model of the U.S. defense expenditure policy-making process is posited. DECit - DECit_l + u* (2.1) [chit - DECit_l] = u* (2.1a) u* = 311EV1t + v* (2.2) v* a Bij 2 EV2t + w* (2.3) w* = Bij+lEV3t + e* (2.4) e* = Idiosyncratic Behavior (2.5) DEC1t - DECit_1 + 311mt + 511 2 EV2t + Bij+lEvat + e* (2.6) 42 Equation (2.1) states that a particular defense expenditure deci- sion at time 5 should, ceteris paribus, be very similar to or in the vicinity of the decision made at time 5:1. In the absence of addi- tional information one would anticipate that "u*" would be random, i.e., non-systematic. This first equation reflects the notion that decision makerssearch in the region of their previous behavior for clues as to subsequent responses on their part. There is a very important theoretical distinction underlying the conceptualizations which appear in equations (2.1) and (2.1a). The focus in (2.1a) is on the change in a defense expenditure decision from one time period to the next. There are several reasons for this shift in attention. The first has to do with the manner in which the previous decision is taken into account, i.e., it is viewed as a base from.which decision makers deviate only slightly, except under extra- ordinary circumstances. Second, given the conceptualization of the decision maker as a boundedly rational individual, one would expect the behavioral search to be in the region of the previous decision. Finally, the complexities of the defense budget are such that decision makers do not possess the cognitive capabilities to "re-think" the budget each year nor do they "re-build" the budget each year starting from zero. The implication of this theoretical shift in attention is that if there are significant deviations in behavior from one time period to the next, then these should be observable in.ADEC1t and attributable to factors contained in u*. This is a fundamentally different explanation than is offered by those who suggest that budgeting proceeds in an incremental fashion, i.e., the budget at time 5 is a fixed mean percentage of the budget at 43 5:1. Rather than suggesting that there is an inexorable growth pat- tern, the CYBER.Model posits that, in the absence of mitigating fac- tors, the budget decision should remain relatively unchanged from year to year. Proponents of incrementalism suggest that the base remains relatively unchallenged. There is some face validity to the contention that budgets demonstrate a growth pattern, and that this growth is the result of a behavioral response which applies a simple, stable, linear decision rule. Figure 2.1, which contains the aggregate defense expenditure decisions for the four previously identified actors, sug- gests that there is a definite upward trend in these decision outputs. The strong trend supports the incrementalist notion of the dominance of the base in budgetary decision making, i.e., an inertial factor is the driving force behind each defense expenditure decision. There is a substantive trap to be avoided here, namely that the acceptance of an incremental explanation suggests a uniformity of behavior from one year to the next-a uniformity which may not exist. The size of the base relative to the annual changes is such that the changes themselves become masked. Figure 2.2, however, shows that _ there is a considerable amount of variation in the behaviors of all four actors, with the Services emerging as the most volatile. This is consistent with earlier observations concerning the range of behav- ioral responses available to each actor. From both a substantive and theoretical perspective the behaviors demonstrated in Figure 2.2 would appear to be the more challenging enterprise. Explanations which account for the annual changes can be used to describe the aggregate series, but the converse is not true. 44 or. mcoaowooa manuaosumxu snowmen Hmscs< up me D I «am» an ’ ‘ no. "~.N muawum cacti: cacao: cgmlw. ocular 7. .—0 .6609 ...~.G 2+. «— .le—vu N0181330 380110N3dx3 BSNBJBU 45 someone mosaics: ououwvsoexu woodman Hmsss< ”N.~ ouomfim zcm» - 2.. t z. . 2... . S. . an. L 2.3 l. a . . _. . . i . q . 1 .Bauo.dr J :38 .0. 3 ; .85.. .0. l. .. :83 .0. 1 ., \ l as. ._ v. .. , p _ . a a! 1. ,2 5.3%.! a \._5.w.,a..., a? . .1, .7 In .600— ...PO SBflNUHO N0191330 3801ION3dx3 BSNBJBO 46 The cybernetic view of decision making suggests that changes in behavior are a response to perceived stimuli (essential variables) in the environment. Furthermore, the decision maker, given limited infor- mation processing capabilities, attends to only a small portion of this environment. Since survival is the predominant goal of the adaptively rational organism, attention is focused sequentially to those factors which most immediately threaten that survival. It should be possible, therefore, to hierarchically decompose the set of essential variables into subsets which reflect varying degrees of "potential threat" or concern to decision makers. It should be noted that in the context of the U.S. defense ex- penditure policy-making process, the concept of "survival" is neither limited to nor primarily concerned with the physical continuation of the various decisionsmaking groups. Survival can and does encompass many different dimensions-national survival, political survival, fis- cal survival, survival in a bureaucratic sense, etc. Caution should also be taken in reading the term too literally. Taken together, equations (2.2) through (2.4) reflect the notion that decision makers sequentially attend to stimuli in their environ! ments. The breakdown of the set of essential variables into first- order (EVl), second-order (EV2), and third-order (EV3) groupings sug- gests that: a) some factors are perceived as being more important than others; and b) decision makers are able to attend to these fac- tors in a hierarchic fashion. This implies that defense policy makers possess the ability to assign priorities to different sets of stimuli. The operationalizations of the specific factors that are either a first-, second-, or third-order essential variable will be left to 47 Chapter IV. What is important to note at this point is that the cyber- netic view of decision making implies that such a behavioral pattern is not only plausible, but that it, in fact, is an accurate character- ization of human decision making. The logic of the analysis is as follows. First, the strong inertial factor represented by the base is eliminated in that the the- oretical focus is on the change in the annual defense expenditure decision. What follows is a hierarchic analysis involving a series of decisions. Figure 2.3 presents a diagram outlining the hypothe- sized behavioral sequence. Any changes or deviations from the base can be viewed as a series of adjustments, reflecting the decision maker's response to what he deems to be essential variables. Clearly some of these factors demand immediate attention, hence they are dealt with first. Then the next most salient variables are attended to, followed by those at the third level. If one views last year's defense expenditure decision as representing a particular actor's response to the perceived demands imposed by the environment, and if that environment should remain largely unchanged, then it should be expected that the behavior of that actor would also remain essentially the same. In terms of the empirical analysis that will take place in Chap- ter IV, at each level I allow each essential variable or set of essens tial variables to "explain" all that it can before proceeding to the next level. Equation (2.5) states that after the effects of the first three sets of essential variables have been taken into consideration any remaining adjustments can be attributed to idiosyncratic behavior. The ultimate defense expenditure decision, represented by equation 48 wmoooum wcfixozlzofiaom ousufiwsomxm omcowon .m.: sou a“ muouu< mow ousomooum wsficoouom ofiaoumuofim "m.m ouomam n>m + ~>m _>m + _-uum= mm» ~>m + _>u Ia + _ can mus ~>m + _>u ..1 oz + _-uumo m>u + _>m I _-u mm» s H u + um: m a 3 a w W a a u _>m + a e a s S s .135 2. n n a m m m M I I 3 m m [IE 9“. + N: U n. u. m o + _-uomo mm> w M m A A m . ~>u + m m m ' oz ulu .6 IV _ noun m>m + _-n mm» on: _-uoma oz 49 (2.6), can be retrieved through an additive recombination of all levels. The methodology proposed here is rather unique, and Chapter III will be devoted to an exploration of the statistical properties of the model. CHAPTER III A METHODOLOGY FOR THE ANALYSIS OF THE CYBER MODEL This chapter presents a general discussion of the major methodo- logical problems frequently encountered when one is using time-series data. Particular attention will be given to the issues of autocorre- lation, multicollinearity, specification error, and seemingly unre- lated regressions. After a consideration of the basic causes and consequences of each of these problems, as well as the relationships among them, the CYBER.Model of the U.S. defense expenditure policy- making process developed in the previous chapter will be examined to determine the extent to which it avoids these frequently encountered methodological pitfalls. Methodological Concerns The multivariate regression model can be represented as: Y1 ’ 81x11 + 82x12 + . . . + BKX1K + 61 (3.1) where Y1 and X1" are the dependent and independent variables, respec- tively, and 5 represents a stochastic error term. The 8K8 are the i regression parameters which one wishes to estimate. The complete specification of this model, however, requires an explicit statement 50 51 of several rather important assumptions (Kmenta, 1971:348): (l) is normally distributed, 51 (2) E(€1) 3 09 (3) 2(212) = .2, (4) E(sisj) a 0 for all i i j, (5) Each of the explanatory variables X is nonstochastic, (6) The number of observations (n) exceeds the number of coefficients (K) to be estimated, and (7) None of the explanatory variables can be expressed as a linear combination of the others. The first four assumptions deal with the specification of the error term, while the remaining three characterize the independent or explanatory variables. Assumptions (1) and (2) state that the disturbance (stochastic error term) is normally distributed around zero. Furthermore, it is assumed that this distribution is symmetric around its mean. The third assumption, also known as the assumption of homoskedasticity, states that the error term.has a constant variance. In other words, one does not expect the disturbances to be greater for higher values of the explanatory variables than for lower values. Assumption (4) implies that the error terms are uncorrelated with each other. When using time-series data this assumption suggests that the disturbance at time £_is unrelated to the disturbance at 2:1. The assumption of nonstochastic X means that, at least in theory, the values of X are either controllable or fully predictable. These first five assumptions are necessary, whether one is 52 estimating a simple bivariate relationship or a more complex multi- variate model. Assumptions (6) and (7), though, are peculiar only to multivariate regression models. Assumption (6) states that there must be sufficient "degrees of freedom” in order to estimate the regression parameters. In other words, there must be at least one more piece of information than the number of parameters (unknowns) one is trying to estimate. The final assumption requires that the independent varia- bles be independent of each other, i.e., that each explanatory varia- ble provide independent information about the behavior of the dependent variable. The violation of any of these assumptions can have potentially disastrous consequences for the estimation of a proposed model. In some cases corrective measures are available, but one does not always have this luxury. One problem frequently encountered when using time- series data is autocorrelation; multivariate models are also subject to the problems of multicollinearity. The first of these problems is the result of a violation of assumption (4), while multicollinearity is a problem resulting from the violation of assumption (7). While violations of the other assumptions have very real consequences, they are far less frequently encountered in time-series analysis. Hetero- skedasticity (a violation of the third assumption), for example, is much more commonly associated with cross-sectional data. This is because with time-series data, changes in the dependent variable and changes in one or more of the independent variables are likely to be of the same order of magnitude. As such, this methodological problem will not be discussed below. In addition to autocorrelation and multicollinearity a third methodological problem will be considered: 53 specification error. While not strictly resulting from a failure to meet any of the seven assumptions presented earlier, specification error is still a serious problem. As will be discussed later in this chapter, specification error is the result of an incorrect formulation of the regression model. Three major types of specification error are: the omission of a relevant explanatory variable, the inclusion of an irrelevant explanatory variable, and an incorrect mathematical formulation of the regression equation (Kmenta, 1971:392). The cone sequences of each of these specification errors, multicollinearity, autocorrelation, and the question of seemingly unrelated regressions will be examined below, followed by a consideration of the trade-offs among them. Autocorrelation This problem is frequently associated with the use of time—series data. As previously noted, autocorrelation is a violation of the assumption that the disturbances are mutually independent, i.e., E(e ) i 0 for all i f j. One explanation for why autocorrelation is 163 often associated with time-series data can be seen if one considers the disturbance term to represent, in part, omitted variables. It then seems quite plausible that if these omitted variables have an impact at time £_they also could be expected to have an effect on the dependent variables at time 5:1 as well. If one has a reason to be- lieve that the omitted variable is trended, then one is in a better position to make some assumptions about the form of the autocorrela- tion. As it will be demonstrated, this is quite important since the 54 implications for using Ordinary Least Squares (OLS) in the presence of autocorrelation depends, to a large degree, on the form of the auto- correlated disturbances. While the number of possible error structures is quite large, this discussion by way of illustration will be limited to one of the most frequent forms, that of a first-order autoregressive process.1 Formally, a regression model with a first-order autoregressive dis- turbance term can be written as: Yt = BXt + 8* (3.2) where *3 e pet_l + vt (3.3) and vt is assumed to meet the basic Gauss-Markov assumptions. In other words, Vt is normally distributed with a zero mean and constant variance, and E(vtvs) a O for all t i s. In addition, one assumes that E(vte ) a O for all 5 and that Ipl huoumsmamxo ooooauxo one omooaosw onu can: mono uo: ma mango .modomfium> huoumomaexm omoaausw osu sues omumaouuoo mum sagas one omumamuuoo haamfiuom mm>Hmmso£u mum nuana moanm Ifium> huoumsmaexm mouuuso on one ma souumamuuououoo was coo? mono so: ma munau .msgm .uo>mson .muo moumsaumo mmocH .ouo .mao>umusfi ousoowmcou mo sowuosuuoooo .wsaummu mwoonuomhn no nonsense man now o>wuomwm0sfi ammo oxma an no omumH om kHHouo loom one moocmuuo> wsfiHoamm ecu mo mouosaumm ecu uonu mason man as omwoouaaHo .manmaum> huouosmaoxu usm>uaou m we soamoaoxo onu hamsms .su0m oawwumam a we on Cu coxou ow uouuo sowumUfiwaoommo ummoam pummfim nauseoumaH Amvum> usmwoamwm uaofioawmm mHHmoHuoumahm< aaamowuoueshm< unusuammosH usofiowmmm m mucoumfiosoosH succumamsoo usoumfiooou < newsman caucuses: pounded: uouum coauoUHmHooem sowumflouuououo< huaummsaaaoowuanz @ mamanoum Hmowwoaooonumz usmuommwn mo mocomoum on» ma muoumafiumu mac mo mmfiuuoooum “H.m manna 66 The MSRA Methodolggz The CYBER.Model of the U.S. defense expenditure policy-making pro- cess was developed and presented in Chapter II. The general structure of this model is re-presented in equations (3.11) to (3.16): macit = mac“?1 + u* (3.11) [DECit - brand] a u* (3.11s) u* - BilEVIt + v* (3.12) v* - Bij Z EV2t +’w* (3.13) w* - Bij+lEVBt + e* (3.14) e* a Idiosyncratic Behavior (3.15) macit = mszcit_1 + 811E“: + Bij 2 EVZt + 81j+1EV3t + e* (3.16) where all the variables are as previously identified. The focus of this discussion will be on the methodological characteristics of this model. The structure of the CYBER Model is somewhat unique: the re- siduals from one equation in the system serve as the dependent vari- able in the next step. There were only two other models which I could locate in the literature which are of a similar form (Ostrom, 1977b; Rattinger, 1975). I shall call this type of analysis, where residuals from one step feed into the next step as the dependent variable, "multi—stage residual analysis" (MSRA). 67 Most econometric treatments of residuals are concerned with either their transformation so as to make them conform to the under- lying assumptions or their analysis in order to obtain results which are unbiased, more efficient and/or consistent. In a recent article, though, Kugler (1983) examines the use of residuals as proxies for variables that cannot be measured directly. Kugler posits four cone ditions which must be met in order for residuals to be used in such a fashion: a. The impact of the targeted variable must be suffi- ciently large to dominate the size and variations of the residual term. b. The hypothesized relationships among the dependent variable, the measured explanatory variables, and the variable associated with the concept to be tested must be accurately specified. c. The targeted variable must be statistically independ- ent of the measured elements in the estimating equation. d. The residuals must behave empirically in concordance with the theoretical expectations engendered by the concept to be tested (1983:106). Even though Kugler is discussing the use of residuals as independent variables, some of the logic underlying their use is still relevant to the MSRA methodology. Kugler's first condition is based on his observation that the main reason residuals are not routinely used as proxies for desired concepts is that other variables, excluded from‘analysis along with the targeted variable, have equal claim to the variations in 5* (1983:110). In other words, unless the targeted variable is large relative to the other elements which are "inhabiting" the residual term, it is unlikely 68 that one will be able to discern its individual effect. This condi- tion is not as crucial to the MSRA methodology since it is concerned with trying to "explain" the variation within the residuals rather than with using those residuals to explain variations in a dependent variable. The second and third conditions are directly related to the prob- lems of specification error and multicollinearity that were discussed earlier. The MSRA methodology is also based upon the assumption of a correct specification. The cybernetic theory of decisionemaking pre- sented in Chapter II, though, argues that decision makers respond in a hierarchical fashion, attending to those factors which are deemed most critical before examining lower-order variables. While one might argue that MSRA is intentionally introducing specification error be- cause relevant explanatory variables are omitted at each stage, the fact that the specification and form of the CYBER Model is fully cons sistent with a cybernetic theory of decision making obviates this problem. Whereas Kugler requires that the proxy variable represented by the residuals be unrelated to the other independent variables in the model, 7 the MSRA approach is based on the premise that such a relationship does exist. The difference between the two, of course, is due to the fact that Kugler treats the residuals as independent variables while MSRA views them as dependent variables. This difference, however, obscures the fact that there is substantial agreement between the MSRA approach and Kugler on the question of multicollinearity. Both recognize that multicollinearity can make parameter estimates so imprecise that it becomes difficult, if not impossible, to disentangle the individual 69 effects of multiple explanatory factors. Part of the underlying rationale of the MSRA methodology is that by including all of the relevant explanatory variables in a single equation model one would be introducing substantial multicollinearity. An empirical comparison of these two approaches will be presented in Chapter IV. Finally, Kugler's fourth condition is a restatement of the notion that there should be some connection between theory and data. Neither should rest independently of the other. Kugler suggests that theory alone is not enough to justify the use of residuals as proxies; rather, independent tests should be set up to provide some empirical corrobo- ration. While the MSRA approach does not disagree with the spirit of this condition, it does give primacy to theoretical development. Fol- lowing a few concluding observations about the MSRA.methodology, the question of empirical support will be addressed directly. At the beginning of this chapter four methodological concerns were identified and discussed: autocorrelation, multicollinearity, specifi- cation error, and seemingly unrelated regressions. To what extent does MSRA avoid, succumb to, or confound these basic problems? Autocorrelation becomes a problem only if the source or structure of the autocorrelation is ignored. The CYBER Model explicitly posits the source of the systematic behavior of the residuals at each step in the model. Ostrom's (1977b) three-tiered model of defense budgeting is the inspiration for the MSRA approach, and he is not unaware of the potential methodological pitfalls associated with such an approach. With respect to autocorrelation he notes that 7O . . . the residuals from equation la are expected to be sys- tematic and the cause of the nonrandomness has been posited. Hence there is no need to assume any of the usual error dependency processes are operating (l977b:27). The same observations are relevant to the CYBER.Mbde1 and the MSRA approach employed in the estimation. In the case of the CYBER Model there is substantial collinearity among the independent regressors. While MSRA.does not eliminate the multicollinearity, it does alleviate it. What I have in the CYBER Model of the U.S. defense expenditure policy-making process is the superimposition of several marginal considerations. A.conventional single-equation.model of the process might not be able to separate out the individual marginal calculations which are being made. This is largely an empirical question and will be investigated below. As was mentioned earlier, the most obvious methodological hazard which the MSRA approach suggests is that of specification error. One argument against this charge has already been offered-namely, that the form and structure of the model are directly derived from an ex- plicit theory of human decision making. Theoretical justifications aside, however, there is the question of whether one is able to toler- ate the bias associated with the exclusion of relevant explanatory variables. Recalling the discussion earlier about the trade-offs between specification error and multicollinearity it was argued that the unbiasedness was sacrificed in order to obtain estimates with greater precision. The decision as to which was preferable rested, in part, on what type of inferences one was interested in making. That no guidelines exist to assist in this decision is of little comfort. 71 In the attempt to determine the appropriateness of the current conventional wisdom of military expenditure models it seems that pri- ority should be given to obtaining estimates which are as precise as possible. Otherwise, I may be led to conclude that some factors are not significant when in fact they really are. If these estimates happen to be somewhat biased, that might be a small price to pay if I am able to demonstrate, for example, that the United States responds in some fashion to the behavior of the Soviet Union. Finally, I should note a potential methodological problem associ- ated with equation (3.14). From the discussion of the CYBER.Model in the previous chapter it was argued that at this stage the decision maker sought information from the previous actor in the process-a final check on one's own assessment of the environmental demands and the appropriate response to them. What is important to this discus- sion is the fact that decision makers are utilizing information from each other. This connection, however subtle, is important because if no such link existed between the various actors in the process the possibility exists that the disturbances might be correlated across blocks of equations. In such situations OLS produces estimates which are inefficient since not all of the information is being utilized, i.e., information about the mutual correlation of the disturbances is being ignored. These inefficient estimates are still, however, pp: ‘pigged_and consistent providing that none of the other methodological problems exist. The temporal recursivity of the CYBER Model also argues against this potential hazard. 72 NOTES 1For a discussion of this simple form and more complex autore- gressive processes, see Hibbs' (1974) "Problems of Statistical Esti- mation and Causal Inference in Time-series Regression Models." 2See, for example, Robert F. Engle, "Specification of the Dis- turbance for Efficient Estimation," Econometrics, 1973. CHAPTER IV OPERATIONALIZATION AND ESTIMATION OF THE CYBER.MODEL Chapter II posited a general form of a cybernetic model of the U.S. defense expenditure policy-making process. This chapter will present the functional form of this model-previously identified as the CYBER Model. A necessary first task will be the operationaliza- tion of the variables comprising the subsets of essential factors. Following this exercise several hypotheses concerning the direction and relative magnitude of the parameter coefficients will be derived, the coefficients will be estimated, and the derived hypotheses will be evaluated in light of the empirical evidence. The chapter will con- clude with a comparison of the MSRA methodology with a more straight- forward multiple regression estimation approach. Definitions; Operationalizations, and Data Sources For each variable in the CYBER Model the following strategy is employed. First, each variable is classified according to whether it is a dependent variable or whether it is one of three subsets of inde- pendent variables, i.e., the subsets of "essential variables." Within each of these four groupings the variables will be defined and oper- ationalized, and the data sources will be identified for each. 73 74 Additional comments pertaining to the operationalized variables will appear at the end of each section when necessary. Dependent Variables As noted in Chapter II, the dependent variables of interest are not, as is typical of most research in this area, the aggregate annual defense expenditure decisions. Rather, the focus of this analysis is on the yearly defense decision changes. The theoretical and substanr tive distinctions between these two forms of the dependent variable have already been noted and will not be repeated here. In the oper- ationalizations below, however, the aggregate form of the variable is presented first, followed by the transformed versions. DEC1t DEFINITION: proposal by military services for amount of defense expenditure each fiscal year OPERATIONALIZATION: Joint Chiefs of Staff October 1 submission to the Secretary of Defense SOURCE: Office of the Assistant Secretary of Defense (Comptroller), Director for Program and Financial Control1 ADEC1t DEFINITION: proposed annual change (increment or decrement) by military services for defense expenditures OPERATIONALIZATION: DEC1t - DEclt-l SOURCE: Office of the Assistant Secretary of Defense (Comptroller), Director for Program and Financial Control DEC2t ADEC2t DEC3t ADEC3t DEC4t 75 DEFINITION: Presidential defense budget request to Congress each fiscal year OPERATIONALIZATION: budget proposed by the Office of Management and Budget for "Department of Defense - Military" SOURCE: Congressional Quarterly Almanac (96th Congress, 2nd Session, Vol. 36, p. 186) DEFINITION: President's proposed annual change (increment or decrement) in the budget for "Department of Defense - Military" 2: ' DECZt-l SOURCE: Congressional Qparterly Almanac (96th Congress, 2nd Session, Vol. 36, p. 186) OPERATIONALIZATION: DEC DEFINITION: amount of money appropriated for de- fense each fiscal year by Congress OPERATIONALIZATION: Congressional appropriation for "Department of Defense - Military" SOURCE: Congressional Quarterly Almanac (96th Congress, 2nd Session, Vol. 36, p. 186) DEFINITION: annual change (increment or decrement) in Congressional appropriation for "Department of Defense - Military" OPERATIONALIZATION: DEC3t - DEC3t_1 SOURCE: Congressional Quarterly Almanac (96th Congress, 2nd Session, Vol. 36, p. 186) DEFINITION: total defense expenditures by the U.S. each fiscal year OPERATIONALIZATION: "Department of Defense - Military" expenditures SOURCE: 'The Budget pf_the United States Government (Washington, D.C.: 1947-1980) 76 ADEC4t DEFINITION: annual change (increment or decrement) in "Department of Defense - Military" expendi- tures OPERATIONALIZATION: DEC4t - DEC4t-1 SOURCE: The Budget pf the United States Government (Washington, D.C.: 1947-1980) First-Order Essential Variables First-order essential variables (EVl) consist of the set of fac- tors which most immediately threaten the "survival" of the cybernetic organism. In the context of the U.S. defense expenditure policydmaking process this statement should not be taken too literally. "Survival" does not necessarily refer to the physical continuation of the several decisionemaking groups in the process. It may refer to survival in a narrow bureaucratic budgetary sense, i.e., the budgetary "health" of the actor can be gauged by how well that actor fares in the annual allocation of finite budget resources. Under this conceptualization "survival" may be nothing more than the maintenance of a constant budgetary share from one year to the next. In order to understand the concept of survival as it is being used here, one must first specify the set of relationships implied by the terms "essential" and "first-order." Essentiality to the cyber- netic decision maker is taken to recall the notion that the cognitive limitations of said decision makers are such that they are only able to respond to a subset of all the potentially relevant environmental stimuli. From a theoretical perspective it should be possible to specify a set of general factors §_priori to which the decision makers 77 can reasonably be expected to attend. The empirical question as to which specific variables precipitate behavioral responses will be addressed in this analysis. Distinguishing between first-, second—, and third-order essential variables implies a hierarchical relation- ship among the various factors, i.e., some factors are clearly more important than others. Finally, there is the obvious question: Essential to whom? Given my focus on the annual defense expenditure policy-making process, the answer includes the major actors previously identified: the military services, the President, the Congress, and the Department of Defense. First-order essential variables also can be distinguished from lower-order essential variables by their immediacy. Responses to these factors cannot reasonably be delayed without the decision.maker assuming greater risks. In other words, the presence of such first- order factors argues against the defense planner being able to "con? duct business as usual." The immediacy of such factors is reflected in the primacy attributed to them by the decision makers themselves. The literature provides no clear guideposts as to what factor or factors might be included in the decision maker's set of first-order essential variables. A plausible argument may be made that periods of high combat activity, what the layperson might refer to as war, are clearly situations in which defense budget operations can no longer be conducted in a normal fashion. The most salient feature of these com- bat operations are the casualties to U.S. personnel which may occur. This salience extends not only to the defense decision makers, but also to the general U.S. public. The latter will be discussed in the context of negative public opinion via-bevis defense operations. 78 The salience of U.S. casualties to defense policy makers is actu- ally twofold. In the first place, casualties suffered by U.S. troops serve as a proxy for the magnitude of the combat operations them? selves. While other factors may be involved, e.g., the killing prow- ess of enemy troops, the greater the number of troops that become involved in the field the greater the number of potential casualties. The magnitude of the combat operations has, of course, a direct bear- ing on the expenditures necessary to maintain those operations. In the second place, the number of casualties gives defense planners some indication of likely replacement costs associated with the loss of men and material. Battle casualties, then, can serve as a reasonable proxy for the costs associated with U.S. combat operations. Although two conceptually distinct costs can be identified ("maintenance costs" vs. "replacement costs"), empirically they are difficult to isolate from each other.2 The true salience of U.S. battle casualties to defense decision makers is that they provide an indication of the mag- nitude of an immediate threat-a threat associated with active combat operations. This first-order essential variable has been termed USE and is defined and operationalized as follows: USEt_1 DEFINITION: intensity of combat operations involving U.S. troops OPERATIONALIZATION: lagged annual U.S. casual- ties (in 10008) which result from enemy action SOURCE: Statistical Abstract.gf_the United States (Washington, D.C.: 0.8. Government Printing Office, 1947-1980) 79 There is a general consensus in the literature that war is a sig- nificant exogenous threat which affects U.S. defense expenditures. It should be stressed that the USE variable is meant to serve as a surro- gate for the intensity of U.S. combat operations. In the time period covered by this analysis there are two intervals, corresponding to the Korean and Vietnamese conflicts, in which one observes high levels of U.S. casualties. These are also the two periods of highest combat activity. Russett (1970) has noted that, following the end of a war, U.S. defense spending has never returned to predwar levels. The additive form of the USE variable captures this "ratchet effect" during periods of escalating casualties. Two final points concerning this first- order essential variable are in order. First, it has been transformed by dividing each observation by 1,000. This transformation only has the effect of reducing the estimated coefficient by the same factor. Moreover, the USE factor is now expressed in a metric similar to that of the dependent variables, thereby facilitating the substantive inter- pretation of the estimated coefficients. Second, the USE variable is lagged by one time period. This is done to bring the time frames of the dependent and independent variables into closer agreement with each other. The former is expressed in fiscal years while the latter is reported in calendar years. Even lagging casualties by one time period does not fully synchronize the two variables, but as Fischer and Crecine (1980) have argued, combat activity during one time period can serve as a reasonable expectation for the level of combat in the subsequent time period. 80 Second-Order Essential Variables As previously mentioned, second-order essential variables (EV2) elicit behavioral responses, though they are attended to only after decision makers have dealt with questions related to U.S. combat oper- ations. In other words, in the absence of any direct combat activity these factors receive immediate attention from decision makers. In the hierarchy of relevant environmental stimuli these variables occupy the second tier. The immediacy of the response is triggered not so much by the immediacy of the stimulus, but rather by the presence or absence of other factors deemed to be more important by the actors in the defense expenditure policy-making process. An examination of the defense expenditure policy-making litera- ture has suggested four dimensions relevant to these actors: (1) Long—term threats posed by major adversaries; (2) International crises or tensions (short-term threats); (3) Domestic factors related to economic performance; and (4) Domestic political support. Each of these dimensions, along with the variables selected to repre- sent them, will be discussed below. Long-term threats are those which are posed by one's major adver- sary. With respect to the United States it is quite clear that the Soviet Union fulfills this role. The American public has been.warned consistently of the dangers associated with a failure to remain vigi- lant in opposing the Soviets. Secretary of Defense Casper Weinberger put the issue in no uncertain terms: 81 You're making a terrible mistake if you try to adjust your defense budget to food stamps, harbor dredgings and high- ways. It's the threat that makes the budget. You've got to build your budget on the Russian budget. (Quoted in the Sunday New York Times Magazine, February 6, 1983.) This passage suggests that one of the key elements of the Soviet threat is the Soviet defense budget. This notion is certainly consistent with the Arms Race literature which has used, almost without exception, expenditure data to operationalize and test the parameters of a US—USSR action-reaction model. There are a number of problems, however, associated with the use of Soviet data.3 In the first place, there is no agreement as to what is actually included in the budget figures released by the Soviet Min- istry of Defense. It is felt by some that the declared Soviet defense budget excludes such factors as military R&D, stockpiling, and civil defense. As the SIPRI 1980 Yearbook notes, upward adjustments to in- clude such expenditures are highly speculative. Secondly, Soviet pricing practices are quite different from those in the West (Holzman, 1980). Objectives are set in real terms with no requirement for money prices to coincide with the real costs of goods and services. The implications here are that U.S. decision makers may have consistently underestimated the true defense burden of Soviet military expenditures. A third problem arises when one attempts to convert ruble estimates into dollars, since the chosen exchange rate should reflect the pur- chasing power of the ruble in the Soviet Union to that of a dollar in the United States. The official exchange rate is considered inade- quate for this purpose, and there is no consensus on an alternative. These technical problems notwithstanding, the major shortcoming 82 of all existing Soviet defense expenditure series is that they reflect "current estimates" of past Soviet behavior, rather than the percep- tions of U.S. defense policy makers at the time decisions about our military expenditures were being made. What is sorely needed is such a data set-one which attempts to characterize the "Soviet threat" as perceived by U.S. decision makers at the time their defense budget decisions were made. The construction of such a series is beyond the scope of this analysis. Instead, I incorporate two dimensions of the Soviet threat: military expenditures and relative strategic strength. The former tries to capture the change in the magnitude of the Soviet defense effort, while the latter can be viewed as an indication of the relative danger to the United States posed by Soviet nuclear-capable delivery vehicles. SOVCHt DEFINITION: annual change in total military expenditures by the USSR OPERATIONALIZATION: (military expenditures by USSR in year 2) - (military expenditures by USSR in year 5:3) SOURCE: Stockholm International Peace Research Institute, Wbrld Armaments and Disarmament, SIPRI Yearbooks, 1968/69,'l979-1982. The Soviet defense budget is viewed as threatening not only be- cause of its sheer magnitude but also because of what those budgetary funds represent in terms of men and material. There is more than a symbolic threat posed by the vast numbers of Soviet troops, tanks, combat helicopters, strategic nuclear weapons, etc. The threat is also a function of the placement of these forces, their degree of combat readiness, their command structure and quality, and so forth. Our own 83 experience in Vietnam and the Soviet difficulties in Afghanistan serve as reminders that brute strength in numbers and technology alone are no guarantees of battlefield success. And yet, our decision makers cast a wary eye on the growing size of the Soviet military machine. Of all the various parts of this machine none has generated more concern in the minds of both policy makers and the general public alike than the Soviet strategic arsenal. Since September 1949 when the U.S. confirmed the first Soviet atomic test the perceived threat posed by Soviet nuclear weapons has steadily increased. Today the watchwords are "window of vulnerability"-in the past our attentions were di- rected toward a "bomber gap" and a "missile gap.‘ Our perceptions of and reactions to Soviet behavior in the area of strategic nuclear forces have been well documented elsewhere (cf., especially, Prados, 1982). Every administration since Truman's has voiced this concern. . . . the situation caused by the continuing rapid expan- sion of Soviet strategic offensive forces is a matter of serious concern. For some time, the Soviet forces which became operational in a given year have often exceeded the previous intelligence projections for that year. (Quoted in Prados, 1982:193.) This remark appeared in the Fiscal Year 1972 Posture Statement of Secretary of Defense Melvin Laird. It was used, in part, to justify increased funding requests for the modernization of U.S. strategic forces. At this point, questions of "quality vs. quantity," "launchers vs. warheads," "size vs. accuracy," and "land-based vs. sea-based" enter the debate as to whether the U.S. or the Soviet Union is more powerful in a strategic nuclear sense. This analysis does not attempt 84 to resolve these issues. Following Ostrom and Job (1982) the focus will not be on the absolute levels of Soviet strategic forces, but rather on the relative balance which exists between the United States and the Soviet Union in this area. Moreover, this balance will be examined in terms of the number of missile launchers (land-based, sea- based, and those bombers with intercontinental capabilities) which each side possesses. The periods of greatest concern to U.S. defense planners will be those in which the Soviet Union has caught up with or has passed the United States in numbers of intercontinental missile launchers. This variable, MISSILE, is operationalized below. MISSILEt DEFINITION: relative balance of U.S. and Soviet strategic nuclear strength which is based upon a comparison between U.S. delivery vehicles (USDV) or missile launchers and Soviet delivery vehicles (SOVDV) or missile launchers OPERATIONALIZAIION: (USDV - SOVDV)/USDV + SOVDV) SOURCE: Ostrom and Job, 1982 It should be noted that the above measure of strategic balance also indicates periods of imbalance. Prior to 1955 the Soviet Union was not thought to possess any intercontinental strategic capability. During this era of U.S. dominance the MISSILE ratio takes a value of 1.000 since the number of Soviet missile launchers is zero. During periods of relative equality or parity between the United States and the Soviet Union the MISSILE ratio will be close to 0.000-a period roughly corresponding to the early seventies. Finally, a negative value for the MISSILE ratio is indicative of a situation.where Soviet missile launchers outnumber their American counterparts-a period 85 covering 1973 to the present. The next type of second-order essential variable, international tensions or crises, represents threats which, unlike those posed by Soviet defense spending and strategic strength, tend to be of a less enduring concern to U.S. decision makers. Due to their sporadic and episodic nature these threats or crises do not lend themselves well to any sort of predictability by policy makers. Empirical research has suggested, however, that decision.makers £3552 to such events, and that these reactions have budgetary consequences (e.g., Domke, Eichenberg, and Kelleher, 1983). The exact nature and timing of these reactions, though, is less clear. The measure of international tensions/crises employed in this analysis, CRISIS, is defined and operationalized be- low. A listing of the specific events comprising this data set may be found in Appendix A. CRISISt DEFINITION: a period of increased military man- agement at the national level that is carried on in a sustained manner under conditions of rapid action and response resulting from une expected events or incidents that have occurred internationally or internally in a foreign country and that have inflicted or threaten to inflict violence or significant damage to the interests, personnel, or facilities of either the United States or the Soviet Union OPERATIONALIZATION: annual index of the number of events which satisfy the above definition and which are of concern to both the United States and the Soviet Union SOURCE: .éflélZ§£§.2£_the U.S. and Soviet Crisis Management'Experiences:~Technical‘Report,- CACI: Inc. - Federal 86 Several points concerning the variable CRISIS are worth noting. First, the index consists of events which are of concern to both super- powers. For the period l946-present CACI has catalogued over 400 Soviet crises and over 300 U.S. crises. The subset of joint crises consists of 95 events and was chosen since those crises involving both superpowers are the most salient in the minds of U.S. decision makers. Second, the index constructed is a simple additive measure of the num- ber of such events each year. Such a measure implicitly assumes that the events are of equal magnitude, importance, salience, etc. Such is clearly not the case. Looking at 1948, the first year in the analysis for which there is more than one joint crisis, one finds the following events: (1) Communist Party assumes power in Czechoslovakia; (2) Costa Rica invaded by Nicaraguanebased rebels; (3) Berlin blockade; (4) First Arab-Israeli war; and (5) Cominform expels Yugoslavia. One would be hard-pressed to successfully argue that the Yugoslav expulsion from Cominform had the same impact on U.S. decision makers as the Berlin blockade. Clearly some type of weighted index would be preferable, but on what basis would such weighting be done? By durar tion? Intensity? Subjective assessment of event salience? In the absence of any clear criteria such a weighted index, perhaps theoreti- cally more appealing but operationally more difficult to construct, was not developed-—opting instead for a simpler additive index. Finally, there is the question of the proper time frame within 87 which the CRISIS factor produces discernible effects. Some of the crises, e.g., the Berlin blockade and airlift, encompass more than one year while others are of a relatively brief, though intense, duration. The budgetary consequences of these events can also be expected to be spread out over a few years. Within a given year one could expect to see the effects of events from previous years as well as those of the current year. For this reason no lag structure is posited and the CRISIS variable is entered contemporaneously with the dependent vari- ables. CRISIS has been coded, however, on a fiscal year basis. Part of the conventional wisdom outlined in Chapter I suggests that the major determinants of U.S. defense spending are internal (do—~ mestic), bureaucratic, economic, and political factors. The last two types of second-order essential variables are derived directly from this assumption. There are a number of possible measures of economic performance that could be used. Nincic and Cusack (1979), for example, investigated the relationship between military spending and perceived aggregated demand. Fischer and Crecine (1980), however, suggest that the most salient economic factors are those related to the Great Identity, namely the revenue forecast and the projected budget deficit/ surplus. The CYBER.Model incorporates this same general logic with the variables PIE and BURDEN. PIEt DEFINITION: projected rate of increase/decrease in total federal revenues OPERATIONALIZATION: (REV t-REV t-l) / (REV t-l) where REVt and REVP1 are forecasted revenues for years_£ and‘tal, respectively 88 SOURCE: Treasury Bulletin, U.S. Department of Treasury, Office of the Secretary BURDENt DEFINITION: anticipated difference between planned current expenditures and forecasted revenues compared with forecasted revenues, i.e., the "burden" which deficits place on the federal budget OPERATIONALIZATION: (EXPt-REVt)/(REVt) where EXPt and REVt are projected total federal expenditures and revenues for year E SOURCE: Treasury Bulletin, U.S. Department of Treasury, Office of the Secretary Both of these variables are self-explanatory. PIE, for example, measures the rate at which the "revenue pie" is increasing/decreasing. Other things being equal, defense planners will be expected to seek budget increases which parallel revenue expansions. If revenue de- creases are forecast, then we should expect to see contractions in defense requests. Using the absolute level of a budget deficit/surplus is somewhat misleading. In 1952 an anticipated budget deficit of $8.2 billion represented 13% of the revenue forecast, but the deficit fig- ure itself was over $47 billion. The BURDEN variable is included to see whether there is any validity to the argument that defense spending can and has been used in a counter-cyclical fashion, i.e., to lead government spending as a means to combat an economy which is slowing down. The time frame of these two variables is also particularly impore tent. The basic information pertaining to revenues and expenditures was taken from monthly Treasury Bulletins. Periodically throughout the year the federal government adjusts its estimates of the receipts 89 (revenues) and outlays (expenditures) for the coming fiscal year, as well as for the current and prior fiscal years. Such estimates are known as current estimates and have been employed in the construction of both PIE and BURDEN. Current estimates reflect the information available to decision makers at the time that their decisions are being made. The final type of second-order essential variable pertains to domestic political support. Support can take many forms. Ostrom and Job (1982) examined the relationship between presidential support (measured in terms of popularity) and the use of force. Rohde and Simon (1983) have noted a relationship between presidential popularity and presidential success in avoiding veto overrides. Others (e.g., Tufte, 1974) have explored the relationship between support and suc- cess in Congressional midterm elections. As Ostrom.and Simon (1983) have shown, popular support can be an important resource which presi- dents can draw upon in seeking enactment of their legislative agenda. The measure of domestic political support employed in this analy- sis is directly related to public opinion toward U.S. military activ- ity. Based upon Gallup polls, the variable NEGOPIN attempts to tap public sentiment regarding expanded military activity on the part of the United States. NEGOPINt_1 DEFINITION: negative public opinion toward expansion of U.S. military activity either in terms of spending for defense or in terms of the continuation of specific military policies such as the wars in Vietnam and Korea OPERATIONALIZATION: yearly index of the per- centage of the public which feels that we 90 are spending too much for defense, or that we should get out of Vietnam or Korea SOURCE: Gallup Poll Reports4 Two comments are in order regarding NEGOPIN. The first has to do with the time element. This measure of negative public opinion, per- haps best construed as a measure of non-support, has been lagged by one year in the attempt to synchronize public attitudes toward defense with the time horizon of the defense budgetary process. Even so, the time frames are not completely parallel. The second point is perhaps a bit more crucial. It has to do with the fact that the index upon which NEGOPIN is based is actually a composite of two conceptually distinct types of opinion vis—E-vis national defense. For some years the index reflects public sentiment that too much is being spent for defense. For other years the index reflects opposition to continued U.S. presence in either Korea or Vietnam. The two dimensions have been combined largely owing to the infrequent and sporadic nature of the Gallup polls in this area. In years when neither question was asked I had to interpolate in order to arrive at the level of public opinion for those periods. The dimensions can be combined for a sub- stantive reason as well. Both sets of questions tap general public opinion in a negative fashion. Mbreover, defense decision makers can see the extent to which the public consensus favors an expansion, con- traction, or maintenance of the status quo vis-a-vis defense. Sub- stantial negative opinion is thereby translated into a lack of sup- port. For the President and members of Congress this lack of public support may have electoral consequences. For the groups located in the Department of Defense one would expect less direct effects since 91 they are not dependent upon overt electoral support. One would still expect them, however, to demonstrate some sensitivity to public opin- ion-if only to maintain their credibility in the eyes of the other actors in the process. Third-Order Essential Variables To this point the CYBER Model has been constructed in such a way that the actors in the defense expenditure policydmaking process have been treated as if they behaved completely independent of one another. This conceptualization of the process is quite different from that sug- gested by the Reactive Linkage Model (Ostrom, 1978a). The major stimp ulus for most actors is the behavior of the previous actor in the process, i.e., the "base" for actor 1_is the decision output from actor 12;. The major exception to this general rule occurs at the first step in the process-the services' request. Here the dominant stimulus is external to the process. The CYBER Model does not entirely disregard the logic of the Reactive Linkage Model. The third-order essential variables (EV3) offered here are in the spirit of the RLM. What is being suggested by the CYBER Model is that each actor looks to the previous actor in the process as a final check on one's own behavior. Decision makers take cues from each other, responding to large changes (increases or de- creases) in the previous actor's behavior. A failure to do so could, in the long run, be potentially hazardous from a bureaucratic budget- ing perspective; it may also reflect that a particular actor has failed to attend to important stimuli from the external environment. 92 The simplest way to explain the logic underlying these third- order essential variables is to treat them as Egg§_which, in turn, represent each decision maker's perception of the demands of the ex- ternal environment. Theoretically the effects of the first- and second-order essential variables should be sufficient to determine: a) whether to continue to seek funding at last year's level or whether changes in relevant environmental stimuli warrant a higher/ lower funding level; and b) if the latter is the case, how much of an increase or decrease is needed. On some occasions, however, there may be other environmental exigencies not specified by this model, or a combination of environmental effects may act in :3 non-additive fashion. Such occasions are easily identified-they are those times when a decision maker's response to a particular configuration of essential variables is greater than or lesser than one would normally expect. What this may indicate is a failure of the ceteris paribus clause. Viewed in this fashion abnormally large increases or de- creases in budget requests can be viewed as a signal that there are additional threats or concerns that should be noted. The logic and operationalization of the third-order essential var- iables is quite straightforward, though the actual construction of the variables is somewhat artificial. If the first- and second-order essential variables completely or perfectly "explained" annual defense budget decision changes, then one would expect w* (see equation 2.3 in Chapter II) to be equal to zero or to differ from zero in a random, non-systematic fashion. Similarly, the presence of systematic vari- ation in w* would lead one to infer that an actor's defense budget change request was not attributable entirely to the set of first- and 93 second-order essential variables. The third-order essential variables attempt to capture this additional information (cues) which decision makers may convey to each other. Temporal considerations, the structure of the defense expenditure policy-making process, and the empirical findings of the Reactive Link- age Model suggest the following: the President takes cues from the Military Services; the Congress responds to information supplied by the President; and the expenditure behavior of the Department of De- fense is responsive to Congressional activity. The defense budget cycle is such that as Congress is working on the budget for fiscal year t, the various agencies and departments are formulating their requests for fiscal year til. The Military Services, then, can be expected to respond to the previous year's Congressional action rather than the previous DoD expenditure behavior. The latter is rejected since Service requests for fiscal year 3:; are made prior to DoD expenditures for fiscal year 5. This is also consistent with Ostrom's (1978a) empirical observation that the Services respond to the trend in Congressional appropriation. The general form of the third-order essential variables, identified as CUES is offered below. t, CUESt DEFINITION: information from previous actor in the defense expenditure policy-making pro- cess which suggests that larger increments or decrements in the defense budget change requests are warranted; can be viewed as a "residual threat perception" on the part of the previous actor in the process. OPERATIONALIZATION: variation in defense budget change requests remaining after the effects of the first- and second-order essential variables have been removed; w* from equation 2.3 94 SOURCE: constructed from CYBER.Model Data for each of the above variables has been collected or constructed for fiscal years 1947 through 1980. Data for lagged variables have been collected as needed. Budget data for the transition quarter which occurred from July 1 to September 30, 1976 have not been in- cluded in the analysis. The next section of the chapter will present a number of hypotheses to be tested by the CYBER.Model. Operationalggypotheses This section presents a number of general research hypotheses, as well as a number of specific hypotheses as to the direction and rela- tive magnitude of the estimated parameter coefficients. Following this it will be shown that these hypotheses are consistent with the "funnel of constrained reactivity" developed in Chapter II. The cen- tral feature of this explanation is that as one moves through the defense expenditure policy-making process each successive actor is increasingly isolated from the external environment, and therefore can be expected to show smaller and fewer reactions to the set of essential variables. Hypothesis la: All actors respond to the first-order essential variable, USE. Hypothesis 1b: This reaction is directly proportional to _the degree of USE, i.e., as battle casual- ties increase, defense budget change requests will also increase. Hypothesis 2a: All actors respond to some or all of the second-order essential variables, though 95 it is expected that there will be less reactivity as one moves through the steps of the budget process. Hypothesis 2b: This decreasing reactivity will be evi- denced by fewer significant reactions to the set of second-order essential vari- ables as one moves through the steps of the budget process. Hypothesis 2c: Where more than one actor responds to the same second-order essential variable there will be a decreasing magnitude of effects as one moves through the process, i.e., the relative magnitude of the estimated parameter coefficients will be decreasing from one step to the next. Hypothesis 3a: Defense budget change requests are directly proportional to the annual change in Soviet military expenditures, SOVCH, such that ins creases in Soviet expenditures will produce additions to defense budget change requests and cutbacks in Soviet defense expenditures will lead to reductions in defense budget change requests. Hypothesis 3b: The reaction to the measure of US-USSR strategic balance, MISSILE, will be as follows: 'when the ratio is positive, a situation where the U.S. leads in the numr ber of missile launchers, there will be reductions in defense budget change re- quests; when the ratio is negative, a situ- ation where the USSR leads in the number of missile launchers, there will be increases in defense budget change requests; and when the ratio is zero, a situation of numerical parity between the US and the USSR, there will be no budgetary reaction to the vari- able MISSILE. Hypothesis 3c: Reactions to international tensions or crises are directly proportional to the nume ber of annual events in the index CRISIS. Hypothesis 3d: Defense budget change requests are directly proportional to the measure of the rate of revenue change, PIE, so that projected in— creases in revenues will produce additions to defense budget change requests and pro- jected decreases in revenues will lead to 96 reductions in defense budget change requests. Hypothesis 3e: Responses to the measure of deficits as a proportion of revenues, BURDEN, will be negative, i.e., defense budget change re- quests will be used in a counter-cyclical fashion. Hypothesis 3f: Defense budget change requests are nega- tively related to the measure of domestic political support, NEGOPIN, such that in- creasing opposition to expanding military activity results in reductions in defense budget change requests. Hypothesis 4: Actors respond to CUES provided by other decision makers regarding the presence of additional environmental concerns. These cues are either positive (suggesting ine creases to defense budget change requests) or negative (suggesting reductions in de- fense budget change requests). Hypothesis 5: After controlling for the effects of the first-, second-, and third-order essential variables, any remaining variation will be random and can be attributed to the idio- syncratic behavior of decision makers. This set comprises the research hypotheses that will be tested by the CYBER Model. Several comments are in order. First, given the hierarchic attentions to essential variables suggested by a cybernetic theory of behavior it is not surprising that all actors are hypothe- sized to respond to the first-order essential variable, USE, since this is the factor which poses the greatest threat. As noted earlier, the presence of active combat operations is a situation which is not "business as usual" and which demands immediate attention by defense policy planners. Second, the general thrust of hypotheses 2a-c is that, as one moves through the defense expenditure policy-making process, there is 97 a decreasing response or reactivity to the set of second-order essen- tial variables. The explanation for this phenomenon is based on the assertion that, within this policy-making process, there is a buffered inner environment (funnel of constrained reactivity) which increasingly shelters actors from potential stimuli in the external environment. Figure 4.1 visually presents the argument being offered here. It can be seen that the Military Services are most proximate to the external environment and, therefore, are assumed to be the most responsive to this environment. This is precisely the argument made by the Reactive Linkage Model (Ostrom, 1978a) where the initial stimulus from the external environment is mediated by the Services and then filtered throughout the rest of the process. At the other extreme is the expenditure behavior of the Depart- ment of Defense. Represented by the innermost circle in Figure 4.1, DoD is the most isolated from the external environment. Correspond- ingly, I would expect this actor to be the least responsive, either in terms of the number of variables to which DoD reacts or in terms of the relative magnitude of the coefficients of the variables which DoD reacts to. By the same logic I would expect Congress to be more reac- tive than DoD, the President to be more reactive than Congress, and the Military Services to be more reactive than the President. Each actor is progressively isolated from these external factors, i.e., each actor in the process acts as a buffer for subsequent actors. Some of this information from the external environment may be passed on to each actor through the third-order essential variables, though this is not their major function. For the most part, the process itself provides a "buffered inner environment" for the actors within the process. H>ZPUFJH>ZWNHNN HZMKZOWHoa no. as oo.o scum unouommao hauamoawaowfim mu:o«uammooo omuuoeou Haas AoH.~-V Amm.uv Ann.~V Ams.smv Ace.s~v Aos.v HN.NHH- mw.HmH . . Ho.H non Ame.~-v Aoo.nv amo.~v RNm.Huv Amm.meo Ame.mav Amm.amev Acm.HHm~V om.aman so.omm ce.sosa om.emflsu . mmmauzoo Ao~.m-v Awo.ov Ask.~v Ame.~uv a~a.mmv Amm.mmv Aem.haev Aom.aeomo -.H¢Hu mm.nnm om.soas om.aflomu u azuaummem aoo.nuv Aom.ev Ana.ev Ao~.~uv Ama.~mv Ans.mmsv Aam.aawv AoH.-an om.aflsu oo.~am om.ammm o~.mmmau . mmoHsamm zumouuz zmnmam mHm mHmHmo uaHmmHz. =u>0m mmmHnwwHw> Hufiuflummm HOvHOlvfloomm How moqumulu one .muouuu osmosmum .muuoaaumm uoowoammooo um.¢ manna 104 decreases as one moves from the Services to the President to the Con- gress. In addition, DoD responds to PIE and NEGOPIN. In both cases the estimated parameter coefficients are less than those for the Con- gress, and this is consistent with the general pattern. The empirical results provide strong support for the "buffered inner environment" explanation developed earlier. Table 4.3 can also be examined in light of Hypotheses 3a through 3f which deal with the individual factors comprising the set of second- order essential variables. As noted, all of the coefficients are in the hypothesized directions and have decreasing magnitudes relative to each other, also as predicted. There is one anomaly, however, which must be addressed. Only one actor, the Department of Defense, reacts to changes in Soviet military expenditures (SOVCHt). This result is not entirely consistent with the notion of a buffered inner environment. Logically, one would expect the military services to respond to Soviet expendi- ture behavior. Indeed, this is precisely the result obtained by Ostrom (1978a): the Services' request was found to be primarily a function of aggregate Soviet military expenditures. This reaction to the external environment was then filtered throughout the rest of the defense expenditure policy-making process in an indirect fashion. That other, more recent empirical analyses have been unable to demonr strate a reaction at any stage to Soviet defense expenditures is of little comfort. Part of the dilemma is undoubtedly related to the measurement problems outlined earlier. Moreover, given that this analysis employed a measure of the annual change in Soviet military expenditures, one 105 might expect results which vary from those utilizing aggregate defense expenditures. In addition, Cusack and ward (1981:446) note that whigh particular data set is used may be quite important. Not only are the actual magnitudes quite variant over time, but the manner in which these series change through time is markedly divergent, as evidenced by the low intercorrela— tion among the first differences. To the extent that measurement problems plague an aggregate expendi- ture series, first-differencing will probably exacerbate rather than ameliorate the situation. One final set of related observations may shed some light on this inconsistent result. Very little is known about the procedures used by SIPRI to generate estimates of Soviet defense expenditures. What seems likely, however, is that they employ some form of growth model, e.g., a first-order autoregressive process. The resulting expenditure series, when differenced, displays a monotonic form, i.e., few extreme changes from year to year. It will be recalled from Figure 2.2 that the-annual defense budget change requests by the military services were extremely volatile, and that the DoD expenditure changes showed the smallest variability. In using the aggregate Service Requests, though, much of the volatility of the differenced series is masked by the dominance of the inertial base. That one trended series has been found to vary systematically with another trended series should come as no surprise. Furthermore, given the four different defense budget change series, it seems quite clear that the U.S. expenditure changes most closely "line up" with the Soviet expenditure changes, more so than does any other U.S. change series.5 106 What I am left with is an empirical relationship which, given the data at hand, is not entirely unexpected. Nevertheless, this empirical observation fits neither the theoretical framework being offered nor any intuitive, common sense notion of the perceptions of U.S. defense planners vis-a-vis the military behavior of the Soviet Union. More than anything else, I think that this particular finding points to conceptual inadequacies which haunt existing estimates of the Soviet threat. Some solace can be found in that this result runs counter to the growing body of empirical analyses which find gg_link between the military expenditure behavior of the United States and the Soviet Union. While DoD is the only actor which responds to the first measure of long-term Soviet threats, each of the remaining three decisions making groups responds to MISSILE, the comparison of US-Soviet stra- tegic nuclear balance. As predicted by Hypothesis 3b, the coeffi- cient for MISSILE is negative, i.e., when the strategic balance is tipped in favor of the Soviet Union the MISSILE ratio will be nega- tive-thereby producing increments to U.S. defense budget change requests. During periods of U.S. numerical superiority in terms of missile launchers one would expect reductions in defense budget change requests since the MISSILE ratio will be positive. Once again, the Services react most sharply to the strategic balance, followed by the President and then Congress. These same three actors respond to short-term threats posed by international tensions or crises. Without sounding too repetitious, the same response pattern is observed once again. The Services are more than twice as reactive as either the President or Congress to 107 this index of short-term international threats. The caveats stated earlier, however, bear repeating. This index treats each event as being equivalent-clearly not the case. Knowledge that a crisis pro— duces increments in defense budget change requests of approximately $3.6 billion for the Services, $1.7 billion for the President, and $1.4 billion for the Congress may not be very meaningful given the obvious noneequivalence of events in the index. The fact that sig- nificant reactions can be detected, though, is important. The none reactivity of the expenditure change request decision is also worth noting. Given that the expenditure changes (SPDCH) show the least volatility (see Figure 2.2) in terms of the size of the annual changes, one would expect fewer significant reactions to the second-order essential variables at this step. Furthermore, given that the expend- iture stage occurs at the end of the defense budgeting process, one should expect less reactivity to relatively short-term events occur- ring much earlier. It is also felt by some that at this point in the process there is very little discretionary spending power, i.e., the ability to respond quickly to international events is greatly constrained. Hypothesis 3d is also supported by this analysis. Each of the four actors in the defense expenditure process makes adjustments in behavior in response to changing revenue conditions. Each actor seeks to.increase his defense budget change request in light of projected revenue gains. Similarly, projections of falling revenues should pro- duce reductions in defense budget change requests. The relative size of the coefficients is also important, but their absolute size may 5 seem somewhat artificial. For example, for every one percent increase 108 in projected revenues the Services adjust their budget change request upward by $812 million. For 1980 this represents about 17.62 of a projected revenue increase of $42 billion. For 1953, however, the corresponding figures are 1302 of a projected $6 billion increase. For the same year, though, the expenditure change represents 23.52 of this same $6 billion projected increase. This suggests that, while the Services seek more than their share of projected revenue increases, the actual expenditure change represents a more realistic share of the increase. One of the surprises of this analysis is that there are no sig— nificant responses to the variable BURDEN. It was hypothesized that defense budget change requests would be used in a counter-cyclical manner. In other words, as deficits increase and start to cause eco- nomic contractions, defense spending could be stepped up to reverse these trends. This argument is essentially the one raised by Cypher (1972) and others who examine the political economy of government spending, particularly military spending. There are three possible explanations for the inability of the CYBER Model to detect any reactions to this variable. The first is that the variable may be operationalized too crudely. Perhaps a measure of aggregate demand might more accurately reflect periods of expansion or contraction in the economy. Second, there are possible problems, discussed in Chapter III, associated with multicollinearity. {A Farrar-Glauber analysis of multicollinearity, found in Appendix B, does not rule out this interpretation. An examination of the matrix of partial correlations finds that BURDEN is correlated with SOVCH at -0.46 with an associated t-value of -2.54, suggesting that 109 multicollinearity is a potential problem. The only other partial cor- relation which is close to being statistically significant is between BURDEN and MISSILE (-0.39 with a t-value of -l.59). Finally, there simply may not be any counter-cyclical effects associated with military spending. Whatever the case, the present analysis does not provide any support for Hypothesis Be. The final second-order essential variable, NEGOPIN, produces sig- nificant reactions by all four actors in the model. The coefficients are also in the predicted direction and, once again, one observes a pattern of decreasing reactivity as one proceeds through the steps in the process. Notice, too, that the response of the Military Services is over twice as large as that of any of the other actors. In some ways this result is unusual, even though it is consistent with the buffered inner environment explanation. What is unexpected is that, of these groups, both the Services and DoD do not depend upon the direct electoral support of the public to maintain their position. One explanation, of course, is the general overall greater reactivity of the Services to all factors in the external environment, but the same cannot be said for DoD. Multicollinearity could also present a problem here as NEGOPIN shows significant partial correlations with SOVCH, MISSILE, and CRISIS. The partial correlation between.NEGOPIN and MISSILE, for example, is a substantial -0.91 with an associated t-value of -10.46. Finally, it is possible that the Services and DoD act in anticipation of the public pressures that will be brought to bear on the President and Congress and hence, indirectly, on them- selves. In any event, the apparent responsiveness to public opinion on the part of both the Services and DoD is certainly noteworthy. 110 Table 4.4 presents the results of the next stage of the analysis, where the residuals from the regression on the second-order essential variables are regressed on the set of EV3. Hypothesis 4 predicts that each actor in the process responds to information or cues provided by the preceding actor. It can be seen that the results are somewhat mixed. All of the coefficients are in the hypothesized direction, though there is no pattern to the relative size of the estimated coefficients. This latter finding may be due to the fact that the EV3 are stimuli from within the defense expenditure process itself. The buffered inner environment explanation pertained only to factors in the external environment. The results presented in Table 4.4 are also consistent with the notion that the actors in the defense expend- iture policy-making process are neither isolated from nor totally independent of each other. The final stage of the analysis deals with an examination of the residuals from the regressions on the third-order essential variables. These residuals (e* in equation 2.5) are regressed on their first, second, and third lags and the coefficients are examined to see if they are significantly different from zero. If these coefficients are not different from zero, then one can conclude that these residu- als, from a univariate perspective, are random. I can conclude that the residuals are "white noise" and that they can be attributed to the idiosyncratic behavior of decision makers. A summary of these regres- sions is presented in Table 4.5. While none of the coefficients is statistically significant, it should be reiterated that this procedure only tests for univariate error processes. The possible existence of 111 Table 4.4: Coefficient Estimates for Third-Order Essential Variablesa (CUESt) b18 Actor (Standard Error) (t-ratio) SERVICES 0.45 (0.21) (2.12) PRESIDENT 0.48 (0.10) (5.03) CONGRESS 1.04 (0.08) (13.17) DoD 0.48 (0.13) (3.77) “111 coefficients significantly different from 0.00 at .05 level. 112 Table 4.5: Coefficient Estimates for "White Noise" Regressions ( * R d * * d * 3 et egresse on et_1, et_2, an et_3) * * * et-1 et-Z et-3 “ct” b19 b110 b111 (Standard Error) (Standard Error) (Standard Error) (t-ratio) (t-ratio) (t-ratio) SERVICES -0.011 -0.295 -0.105 (.186) (0.179) (0.191) (-0.061) (-1.646) (-0.552) PRESIDENT -0.252 -0.046 -0.047 (0.189) (0.194) (0.198) (-l.336) (-0.236) {-0.236) CONGRESS -0.267 -0.305 0.161 (0.179) (0.184) (0.193) (-1.491) (-1.655) (0.835) DoD 0.210 0.099 -0.053 (0.217) (0.232) (0.236) (0.966) (0.426) (-0.223) aAll coefficients not significantly different from 0.00 at .05 level. 113 more complex error structures is not eliminated empirically by this analysis. By the same token, however, there is no reason to suspect that such processes exist. A systematic examination of the hypotheses generated earlier in this chapter finds considerable support for each except for Hypothe- sis 3e. All of the coefficients are in the hypothesized direction; moreover, the plausibility of a buffered inner environment is given empirical support by this analysis. Part of the test of any model is the extent to which hypotheses derived from that model can be empiri- cally validated. In this regard the CYBER.Mode1 of the defense expenditure policydmaking process performs quite well. The evaluation of the CYBER Model, though, is only partially complete. Chapter III developed and offered a novel methodology for the analysis of defense budget change requests. The final section of this chapter will com- pare the results generated by the MSRA methodology against those derived from a traditional multiple regression analysis. A Comparative Analysis of MSRA and Multiple Regression The CYBER Model and the accompanying MSRA methodology were devel- oped in response to theoretical and empirical deficiencies in existing explanations of the U.S. defense expenditure policy—making process. _ It was suggested, for example, that the classic arms race model and organizational process model were indistinguishable from one another in terms of their underlying conceptualizations of the policydmaking process and the decision makers who reside therein. 114 It has been suggested on more than one occasion that the CYBER Model is a more accurate and truer representation of the behavior of cybernetic decision makers. Implicit in this statement is the notion that the CYBER Model offers a better explanation of the behavior of policy makers in the defense expenditure budgetary process. The immediate question arises: Better than what? The answer is twofold. It is argued that the CYBER.Model is superior to existing explanations because of its explicit derivation from a body of literature dealing with theories of individual[organizational decision making. Moreover, it has been suggested that the MSRA.estimation approach avoids a mum? ber of methodological pitfalls associated with a more traditional, single-equation multiple regression (MR) procedure. This section will compare the results of the MSRA estimation of the CYBER Model with the results generated by a MR approach. While statistical comparisons between the two procedures will be drawn, the focus of this analysis will be on the different substantive explanae tions offered by each. Before doing so, I would like to note changes in the variable operationalizations that will be made in order to make them more compatible with a MR analysis. Points of difference in the underlying logic of the two estimation approaches will also be noted. Variable:Operationalizations The operationalizations for the first-order essential variable (USE) and all six of the second-order essential variables (SOVCH, 115 MISSILE, CRISIS, PIE, BURDEN, and NEGOPIN) will remain unchanged. The only alterations that will be made are to the set of dependent vari- ables and to the set of third-order essential variables. The dependent variables to be examined by the MR approach will not be the annual changes in defense expenditure decisions, but rather the 22531 defense requests, recommendations, appropriations, and expenditures. These variables were previously operationalized as DEclt’ DECZt’ DEC3t, and DEC4t' In addition, the previous year's defense expenditure decisions (DEC1t_1) will be incorporated into the model as explanatory variables. These two alterations will make the MR model consistent with most of the empirical analyses of the U.S. defense expenditure policy—making process. This formulation will also enable me to test the proposition that much of the "explanatory power" of plausible, relevant regressors is overshadowed by the dominance of the base (represented as last year's defense budget decision). Given the artificial construction of the third-order essential variables used in the CYBER Model, it is very difficult to find a direct translation of these factors that can be included in the MR equations. Since the EV3 were designed to represent cues provided by each actor in the process to the next actor, I have decided to oper- ationalize them as the annual defense budget changes (ADECit)' In other words, the size of the increase or decrease in defense budget decisions will be viewed as the additional information or cues which decision makers attempt to convey to each other. As with the CYBER Model it is assumed that the services react to the previous years' Congressional change, the President responds to the current change in the Services' Request, etc. 116 Estimation Procedure The general form of the equations to be estimated is given as follows: n k DEC + 2 Z 8 DEC -8 it 11 it 1 1_1 3-2 11 Ejt + Bi,j+l ADEC1_1’t (4.1) where Ejt are the set of exogenous factors previously identified as first- and second-order essential variables and all other factors are as before. The only exception to this general form was noted earlier, i.e., the services are expected to respond to the previous year's change in Congressional appropriation (ADEC3t_1). There are two aspects of equation 4.1 which will make the use of ordinary least squares (OLS) inappropriate. The first is the presence of a lagged value of a dependent vari- able on the right—hand side of the equation. Use of OLS in this situ- ation will produce estimates which are inconsistent. One can no longer make the assumption that the explanatory variables are uncorrelated with the error term. In order to obtain consistent estimates it is necessary to employ a two-stage instrumental variables technique. Second, this problem is attenuated when the disturbances are autocorre- lated. Unlike the CYBER.Mode1, where the form of the autocorrelation was specified, the MR.mode1 represented in equation 4.1 makes no such specification. For that reason, an iterative maximum likelihood pro- cedure which corrects for first-order autocorrelation will be used in conjunction with the two-stage instrumental variables technique. The 117 resulting estimators will have all of the desirable properties of OLS estimators under normal conditions. It should be noted that the estimation of equation 4.1 was con! ducted in two steps. First, the entire equation was estimated using the techniques outlined above. Variables whose estimated parameter coefficients were found 22$ to be significantly different from 0.00 at the .05 level were subsequently dropped from the analysis. The equa- tions, in the second step, were then re-estimated on this reduced set of regressors. Multiple Regression Estimation Results Table 4.6 presents the results of the multiple regression estima- tion of the annual defense budget decisions of the military services, President, COngress, and Department of Defense. Some general comments about the results displayed in Table 4.6 will be offered, followed by a closer examination of the substantive explanation which this analysis produces for each actor. Overall, it can be seen that the MR approach leads to a much more parsimonious explanation of the U.S. defense expenditure policy-making process. For three of the four actors the MR analysis yields a smaller set of regressors. In such situations it is typical for a researcher to invoke a criterion of parsimony as a means for deciding which of two or more competing explanations of a given phenomenon will be accepted. Parsimony should not, however, be the only criterion employed. Notions of plausibility and relevance to the real world should also be taken into consideration. As I have suggested before, "Occam's razor" should 118 .Hlum once an Haas menu umoa>uum «as you a o AUGOU UNHHmu—IQCOV HQ>0H “Do U” C900 BORN HAHUHUMMHU SAHHgUHWHame QUGQHUHNHQOU H2“ Ame.cv Amm.~Iv Ama.amv u Aoo.ov As~.hev Amo.ov A scene He.o m~.eaHI I I I I I I MH.H non AssHHNV Aam.HIV Amo.oeav um ano.ov Aem.mmv Aao.ov A once as o I I I I I I am.maHI koo.a mmmaozou Aam.av Asa.~v Ana.~oao u Aoonov Asa.mov Aao.ov A Nomnv as o I I oe.mo~ I I I I Ho.H ezmaHmmem Amm.~Iv Aan.NIv Am6.ov Ams.mv Aoa.HIv Ase.HIv Aom.mav 0H Ao~.ov Ama.oomc Amn.~mav Aom.mooav Ace.nHH6v AaH.m~oV Amo.ov A sumo as.oI so.emmI I oo.mnm oe.meam om.mmaoI I oH.Hooa mo.H muoHSmmm “a HIHouaa szouuz zmnaam HHS mHmHeo maHmmHz mo>om mm: H “some .hmmmmmw soammuuwom oaoauHsz How mowumulu one .muouuu oumoomum .moumaaumm usowowmmooo fl msowmwoon muouwosoexm omnuwon .m.= mo mwmhaon< “8.6 ounce 119 be a guide rather than a firm decision rule to be employed indiscrimi- nantly. Parsimonious models which generate counterintuitive and trivial results are of limited explanatory value. As a second general observation I would point to the signs (direction) of the estimated coefficients. Applying the operational hypotheses presented earlier in this chapter it becomes apparent that not all of the coefficient estimates are in the hypothesized direc- tion. This would seem to be a necessary condition for a model to meet if it is to have any useful explanatory relevance. Even the magnitude of some of the coefficients is questionable, a third point that will be discussed more fully below. A final observation is the relative robustness of the estimated coefficients associated with the previous year's defense expenditure decision, as evidenced by their large t-ratios. The MR analysis of the Services' Request yields an explanation very similar to that produced by the MSRA approach. Both find that the services react to the measure of U.S. combat activity (USE) as well as to the same set of previously identified second-order essen- tial variables (MISSILE, CRISIS, PIE, and NEGOPIN). Moreover, the two sets of coefficients are quite similar in terms of their relative size. The coefficient for the previous year's services request, i.e., the bags, suggests an average incremental growth of nine percent. This figure might not seem plausible were it not for the presence of the exogenous factors in the model, each of which may produce adjustments to this annual growth pattern. There are only two aspects of the MR analysis of the services 120 request which are at all disturbing. The first, and more serious, cone cerns the direction of the estimated coefficient for ADEC3t_1, the measure of additional information or cues supplied to the Services by the Congress. The negative sign, implying that the Services cut their requests in response to recent Congressional increases, is both coune terintuitive and not consistent with previous empirical results (cf. Ostrom, 1978a). In absolute value, however, this coefficient is almost identical with that associated with the third-order essential variable (CUE) in the CYBER.Mode1. The second feature worth.mentioning is the relatively large standard errors associated with the estimated coef- ficients for USE and MISSILE. These coefficients are only slightly larger than one and one-half times their standard errors. These coef- ficients are significantly different from zero at the .05 level under a one-tailed hypothesis test. For a corresponding two-tailed test one could not conclude that these coefficients are different from zero. The large standard errors should sensitize one to the potential insta- bility of these estimates. The results reported in Table 4.6 for the President's defense budget requests are perplexing for a number of reasons. The first is the relatively small incremental growth rate associated with the base (one percent per annum). Almost all of the yearly changes in Presi- dential defense budget requests can be attributed to projected revenue increases and cues provided by the Services. With respect to the latter finding it should be pointed out that this estimated coeffi- cient is virtually identical to that associated with the President's third-order essential variables. 121 Perhaps more important than what is included in this model are the factors which are excluded. This MR analysis suggests that the President is unresponsive to U.S. casualties, Soviet spending behavior, the strategic balance between the U.S. and the USSR, as well as to international crises. These results are not consistent with either a common-sense understanding of the President as Commander-in-Chief or with the typical rationale which Presidents have used to justify their sought-after defense budget increases, e.g., the dangers posed by the international environment. The exclusion of public opinion (at least as it is measured here) from the President's decision calculus is also somewhat surprising. The explanation suggested by Table 4.6 for Congressional defense budget appropriations is also unrealistic in some respects: the incremental growth rate appears to be less than one percent; Congress reduces defense appropriations in response to casualties Suffered by U.S. military personnel; and Presidents enjoy almost total success in getting Congressional approval for their proposed defense budget increases or decreases. This last result, when coupled with the almost nonexistent growth rate, is the most plausible aspect of this explanation. Ostrom (1978s), for example, found that Congress cuts Presidential defense budget requests by an average of slightly more than three percent. Most disturbing is the apparent response of Con- gress to U.S. battle casualties. Moreover, the unresponsiveness of Congress to changing economic conditions seems surprising. Finally, actual defense expenditures by the Department of Defense demonstrate the highest average annual growth rate-thirteen percent. 122 This seems reasonably plausible, though it should be stressed that this is an average rate which masks periods of low growth (e.g., 1957-61) and periods of high growth (e.g., 1978—80). Like the CYBER Model, DoD is found to be responsive to negative public opinion. Defense expend- itures do not, however, appear to be reactive to Soviet defense expenditures or to U.S. economic conditions. It should be noted that DoD takes cues from recent Congressional action, a result also con- sistent with the empirical findings of the CYBER.Model. A comparison of the two models, in terms of the explanations they produce, would favor the CYBER Model for a number of reasons. First, some of the coefficient estimates of the MR approach are not in the predicted direction. Second, some of the empirical results of the MR model are counterintuitive and inconsistent with previous empirical analyses. This is true not only in terms of the explanatory factors that are included in the model, but more importantly in terms of the factors that are excluded. Such comparisons, however, tap only one dimension of these two approaches-namely, the explanations which they produce. Another type of comparison, the focus of Chapter V, will examine the forecasting abilities of each model. 123 NOTES 1Data for FY 1948 and FY 1949 were obtained from Congressional testimony by the Services. The FY 1980 Services Request was provided to me by the Services themselves. For FY 1947 only the Air Force and Army supplied the requested data. The entire Services' Request for this year was extrapolated under the assumption that the Navy request was a constant share during this period. 2SIPRI (1973) has estimated that the "incremental costs" of the Vietnam War for the period 1965-1973 were in excess of $107 billion. Incremental costs are the additional expenditures above what would have been spent in peacetime for the base-line force. 3For an in-depth survey of the problems associated with estimat- ing Soviet defense expenditures, see Cockle (1978) or Burton (1983). 4Gallup has asked questions pertaining to defense spending with sporadic frequency. Data on defense spending were obtained from monthly reports #50, 71, 88, 93, 101, 112, 129, 146, 175, and from weekly reports prior to 1965. Gallup polls on public opinion concern- ing Vietnam.were conducted regularly from 1965-71. For years when no opinion data was available a surrogate measure was obtained by inter- polating from available data. For clarification and elaboration of the impact of public opinion on defense spending, see Russett (1972). 5The zero-order correlations between SOVCH and the defense expend— iture changes are: ADEC (.179); ADEC (.291); ADEC (.308); and 1t 2t 3t ADEC4t (.333). CHAPTER V A FORECAST EVALUATION OF THE CYBER MODEL The previous chapter initiated an evaluation of the CYBER.Model by examining the degree to which the empirical results provided sup- port for a number of hypotheses regarding the direction and relative magnitude of the estimated parameter coefficients. Based upon this preliminary evaluation I concluded that the CYBER Model demonstrated considerable potential as an explanation of the U.S. defense expendi- ture policy-making process. The CYBER.Mode1 and its underlying MSRA methodology were also evaluated through a direct comparison with a straightforward multiple regression (MR) approach. The MR analysis, while yielding a more par- simonious explanation of defense expenditure policy making, produced some estimates which were not in the hypothesized direction. In addi- tion, the magnitudes of some of the coefficients suggested incremental growth rates which were not plausible in that they were fairly small. As an explanatory device, then, the CYBER Model appears superior to its MR counterpart. Competing models may be judged, however, on criteria other than the explanations which they yield. The predictive capabilities of individual models may be equally important. It could be argued that a model which generated largely inaccurate forecasts was of limited 124 125 utility, regardless of the plausibility of its explanation. This chapter will evaluate the CYBER.Model in terms of the his- torical forecasts which it produces. In addition to the absolute forecasting ability of the CYBER.Mode1, this chapter will explore its relative success as a forecasting tool. As a basis for this compari- son, the forecasts of the MR.Model and a No—Change Naive Model will be generated. An absolute measure of the accuracy of these forecasts will be employed to conduct this test. As a prelude to this evalu- ative exercise, the goodness of fit of the CYBER Model will be briefly examined. Goodness of Fit of the CYBER Model The most frequently encountered measure used to assess the "fit" of a model represented by a regression line is R2 (r-square)--the proportion of variation in the dependent variable attributable to the independent variab1e(s). Unfortunately, the MSRA.methodology does not produce R23 that can be interpreted in the traditional sense. At each step in the analysis the R2 represents the proportion of explained variance, but of a partitioned set of variance. In other words, an R2 of 0.80 means that eighty percent of the available unexplained variance can be attributed to the particular set of essential variables. As a means of summarizing the total explanatory power of the CYBER Model I have generated a surrogate R2 which is based upon a comp parison between the actual defense expenditure decisions and those predicted by the.model. These predicted values are correlated with 126 the actual observed values. The square of these bivariate correla- tions provides an estimate of the proportion of the variance in the dependent variable that can be attributed to the combined subsets of essential variables. Tables 5.1 and 5.2 present the bivariate correlations and surro- gate st for each actor's defense expenditure change decision (ADECt) and aggregate defense expenditure decision (DECt), respectively. Table 5.1 suggests that the CYBER.Model meets with considerable suc- cess in terms of the amount of explained variation in the defense expenditure change decisions. The range is from 942 for the Congress to 57% for the Department of Defense expenditure change decisions. Table 5.1: Bivariate Correlations between Actual and Predicted Values of ADECt with Associated Surrogate R23 Actor Bivariate Proportion of a Correlation Explained Variation Services .88 .77 President .92 .84 Congress .97 .94 DOB .75 .57 aDerived by squaring the bivariate correlation between actual and predicted values of ADECt. 127 Table 5.2, on the other hand, indicates even greater success on the part of the CYBER Model in terms of its ability to predict aggre- gate defense expenditure decisions. The "poorest" fit is with the Services at 95%; at least 982 of the variation in the other three actors' aggregate defense expenditure decisions can be accounted for. Table 5.2: Bivariate Correlations between Actual and Predicted Values of DECt with Associated Surrogate R28 Actor Bivariate Proportion of a Correlation Explained Variation Services .970 .95 President .992 .98 Congress .997 .99 DOB .989 .98 aDerived by squaring the bivariate correlation between actual and predicted values of DECt. This phenomenal explanatory success, however, needsto be put into perspective. It should be noted that the previous year's deci- sion (DECt_1) or base is used in the generation of the predicted values of the aggregate defense expenditure decisions. In the pre- vious chapter it was shown how influential this base is (cf. Table 4.1). This "inertial dominance" makes it very difficult to detect the effects of other factors. In addition, this base all but 128 guarantees the extremely high level of explanatory power typically associated with such incremental models. Focusing on the Services Request it was demonstrated that the base accounted for 742 of the total variation. As indicated by Table 5.1, this figure is increased to 952 by the inclusion of the first-, second-, and third-order essential variables. While the increase in explained variation is, in an absolute sense, 21%, this figure rep- resents over 80% of the variation not already attributable to the base. Table 5.3 addresses this issue by decomposing the amount of ex- plained variation in the aggregate defense expenditure decisions into two portions: that which can be attributed to the bags and that which can be attributed to the combined effects of the essential variables. Table 5.3: Decomposition of Explained Variation in Aggregate Defense Expenditure Decisions (DECt) 2 Variation Attributed to Actor EV Increase (Z)c Basea Essential Variablesb Services 74 21 81 President 90 8 80 Congress 90 9 90 DoD 96 2 50 aPrevious year's defense expenditure decision (DECt_1). bCombined effects of first-, second-, and third-order EVs. c(Variation attributed to EVs)/(Total variation not attribut- able to Base) expressed as a percentage. 129 Table 5.3 also indicates the proportion of the unexplained variation which the essential variables capture once the effects of the base have been isolated. For example, the increase of 22 in the explained variation in the DoD expenditure series actually represents one-half of the variation not already accounted for by the large, inertial base. In trying to assess the explanatory power of the CYBER Model, several features emerge. First, in terms of the original dependent variables (ADECt) the model performs quite well. Across all four actors nearly eighty percent of the total variation can be attributed to the factors comprising the CYBER Model. The poorest fit is with the DoD expenditure change (572), while 942 of the annual change in Congressional appropriations can be explained by the model. Second, the combination of the essential variables with the previous year's decision is sufficient to explain nearly all of the variation in aggre- gate defense expenditures decisions for all four actors. Finally, even though the effects of the base appear quite large relative to those of the essential variables, Table 5.3 demonstrates that this come parison is somewhat misleading. Increases associated with the essen- tial variables tend to be fairly small in absolute terms, but are quite large when viewed in relative terms. Evaluations of the CYBER.Model based upon a simple goodness of fit criterion have found this approach to be quite fruitful. When this is coupled with the results of the hypothesis tests conducted in the previous chapter I am quite optimistic about the utility of the CYBER Model as an explanatory and predictive model of the U.S. defense expenditure policy-making process. The estimated parameter 130 coefficients are substantively plausible and empirically robust. In a comparison between the CYBER Model and a MR.Mode1 conducted in the previous chapter it was observed that the latter possesses many shortcomings as an explanatory vehicle. The question has not been addressed, however, as to the forecasting ability of this model. More importantly, the forecasting ability of the CYBER Model must be assessed relative to some other model. The next section will compare the forecasts generated by these two models of defense expenditure policy making. In addition, a No—Change Naive Model will be intro- duced as an evaluative baseline. A Comparative Forecast Evaluation As suggested earlier, the fact that the CYBER Model produces accurate predictions is a statement that cannot be judged in absolute terms alone. If another model yields predictions which are relatively more accurate, then the CYBER Model will have to be found wanting. One obvious candidate is the multiple regression (MR) model estimated in the previous chapter. First, an absolute measure of the accuracy of these forecasts must be introduced. One possible measure might be the bivariate cor- relation between the actual and predicted series. While this statis- tic is indicative of the degree to which the two series are associated with each other, it does not provide a measure of absolute forecast accuracy. For example, a model which consistently yields predictions that are 25% below the actual values will still demonstrate a high 131 bivariate correlation between the two series. This is because one series can be expressed as a linear combination of the other. In ab- solute terms, however, the model would not fare very well. As an absolute measure of the accuracy of the forecasts generated by the CYBER and MR Models I shall employ the root mean square error (RMSE). The RMSE is the standard deviation or dispersion around the line of perfect forecasts. The smaller the RMSE, the more accurate the forecasts. The RMSE can be computed as follows: RMSE - [1‘1 2 (Pt - 71921;“ (5.1) where Pt refers to the forecasted or predicted value, At is the actual value, and T is the number of forecasts. It can be seen that when the predicted values are identical with the actual values the RMSE will be zero. As Ostrom (1978::66) notes: Since [the RMSE] is a non-parametric measure, there are no objective standards (e.g., significance levels) with which to assess the forecast accuracy. A determination of the accuracy of the model forecasts can be made, however, by comparing their accuracy to that of a mechanistic alterna- tive-one which requires no prior theorizing (in the sense of having a prior conception of the dynamics of the policy- making process). One alternative which he suggests is the no-change naive model. This no-change model states that the value of a defense decision at time 5 is equal to the value of time t-l plus some random disturbance that is normally distributed with a zero mean and a constant but unknown vari- ance, i.e., 132 DECt =- mack1 + vit (5.2). Equation (5.2) can also be used to suggest a no-change naive model which accounts for the annual defense expenditure change decisions, i.e., ADECt=ADECt_1 + vit (5.3). Equation (5.3) says that the annual defense expenditure change deci- sion at time 5 will differ from the change decision at EZE by only a small, random factor. No-change naive forecasts will be generated using equations (5.2) and (5.3) and will serve as an evaluative baseline. Tables 5.4 through 5.7 present the actual defense expenditure change decisions for the Services, President, Congress, and DoD, respectively, as well as the predicted values for the CYBER, MR, and No-Change Models. The CYBER Model will be judged to be a more accu- rate forecasting model if its RMSE is less than the RMSEs of the other two models. In addition to the RMSEs, Tables 5.4 through 5.7 also present the correlations (rap) between the actual and predicted values for each model. Owing to the manner in which the CYBER Model is estimated the first three observations are lost. As such, fore— casts have been generated for the FY 1950-80 period. It can be seen that the historical forecasts produced by the CYBER Model for the Services' change decisions are more accurate than those of either the MR.Model or the No-Change Naive Model. Specifically, they have a smaller RMSE ($6.9 billion to $10.1 billion and $20.5 bil- lion) and are more highly correlated with the actual change requests 133 Table 5.4: Actual and Forecasted Values of Military Services' Annual Change Requests, 1950-80 ($ Millions) Forecasted Change Requests Fiscal Actual Year Change Request No-Change CYBER Model MR.Model Naive Model 1950 1415 4889 10013 1625 1951 3410 -1335 -318 1415 1952 60181 55609 64657 3410 1953 -l61211 7180 —6727 60181 1954 -l4890 -l6l69 -l3137 -l6121 1955 -14600 -l7659 -19495 -14890 1956 1500 1778 -3755 -l4600 1957 3200 -5818 -7382 1500 1958 14800 13949 24890 3200 1959 -1340 -7743 969 14800 1960 6400 16382 -5584 -13400 1961 —7697 -954 1620 6400 1962 9797 2386 -11481 -7697 1963 2000 5675 14925 9797 1964 6544 6834 4552 2000 1965 -160 5448 7992 6544 1966 -3936 -12158 -18577 -160 1967 16735 6849 12106 -3936 1968 20409 16525 21020 16735 1969 8654 16718 22831 20409 1970 -l683 223 322 8654 1971 -22134 -15223 -l6334 —1683 1972 -2427 -5655 -18947 -22134 1973 5277 -4540 -1598 -2427 1974 5194 5730 18425 5277 1975 -246 1506 4103 5194 1976 11916 5378 -6909 -246 1977 9691 9455 23266 11916 1978 8988 14895 11010 9691 1979 9887 8880 10549 8988 1980 9832 6690 5981 9887 RMSE 6873 10062 20490 0.879 0.804 -0.022 r 8? 134 (.88 to .80 and -0.02). While there is not a statistically significant difference between the CYBER and MR correlations, the smaller RMSE for the CYBER Model indicates that its forecasts are more accurate than those of the MR and No-Change Naive Models. The latter performs especially poorly, a fact borne out by its relatively large RMSE and low correlation between actual and predicted values. The CYBER Model does just as well in its forecasts of the Presi- dential change decisions. As presented in Table 5.5, the RMSE for. the CYBER Model is less than that of either of the other two models ($3.9 billion to $4.8 billion and $13.1 billion). Moreover, the cor- relations between actual and predicted values are largest for the CYBER Model (.92 to .87 and .03), though once again the difference between the CYBER and MR Models is not statistically significant. In terms of the magnitude of the RMSEs, the CYBER.Mode1's forecasts are more accurate for the President's change decisions than for those of the Services. As can be seen from Table 5.6, this forecasting accuracy improves when one examines the Congressional change decisions. More impor-s~ tantly, the CYBER.Mode1's forecasts are more accurate than those of either the MR or No-Change Naive Model. This is evidenced by both the RMSEs ($2.3 billion to $3.3 billion and $13.3 billion) and the corre- lations between actual and predicted values (.97 to .94 and -.04). Finally, the CYBER Model yields the most accurate historical fore- casts of DoD expenditure changes. .This is confirmed in Table 5.7 by its smaller RMSE ($4.1 billion to $4.4 billion and $5.6 billion). The predicted values generated by the MR.Mode1, however, are more closely correlated with actual values (.79 to .75 and .51). This only 135 Table 5.5: Actual and Forecasted Values of President's Annual Change Requests, 1950-80 ($ Millions) Forecasted Change Requests Fiscal Actual Year Change Request No-Change CYBER.Model MR.Mode1 Naive Model 1950 1316 -624 1505 735 1951 -ll92 2864 2640 1316 1952 43490 43265 40706 -1192 1953 -4214 -5421 -5139 43490 1954 -10671 -7387 ~5763 -4214 1955 -10833 -9242 -10035 -10671 1956 2346 -29 -l935 -10833 1957 1915 193 928 2346 1958 1980 6699 8818 1915 1959 2069 -6867 -8979 1980 1960 1051 4338 8426 2069 1961‘ 87 -3474 -2184 1051 1962 3607 4437 5015 87 1963 4965 2368 3453 3607 1964 1107 2624 3638 4965 1965 -1543 117 83 1107 1966 -2222 -1346 -3571 -1543 1967 12415 11388 8395 -2222 1968 13920 14626 12443 12415 1969 5490 12378 9888 13920 1970 -l796 4166 652 5490 1971 -6532 -8392 -12177 -l796 1972 4798 1801 2560 -6532 1973 6050 5007 8350 4798 1974 -2343 5274 7377 6050 1975 9806 1825 -2509 -2343 1976 10801 7165 1212 9806 1977 8904 7973 13776 10801 1978 7115 6595 8798 8904 1979 5423 6545 7193 7115 1980 13021 6419 5908 5423 RMSE 3858 4838 13091 0.915 0.868 0.029 r 8P 136 Table 5.6: Actual and Forecasted Values of Congressional Annual Change Requests, 1950-8O ($ Millions) Forecasted Change Requests Fiscal Actual Year Change Request No-Change CYBER Model MR.Mode1 Naive Model 1950 2197 1378 2122 1910 1951 -1554 -1204 -1344 2197 1952 42938 41384 41669 -1554 1953 -8373 -4700 -3005 42938 1954 -12239 -10125 -12764 -8373 1955 -5572 -9941 -12603 -12239 1956 3083 2555 5498 -5572 1957 2774 2801 3909 3083 1958 -897 1263 2829 2774 1959 5913 2928 267 -897 1960 -445 408 3011 5913 1961 769 210 162 -445 1962 6666 3699 3608 769 1963 1473 5409 7340 6666 1964 -916 499 -422 1473 1965 -468 -2078 -3943 -916 1966 135 -1l33 -2237 -468 1967 11180 12466 13461 135 1968 11870 12723 12756 11180 1969 1933 3660 2787 11870 1970 -2229 -2910 -4078 1933 1971 -3045 -6006 -6598 -2229 1972 3922 5725 7714 -3045 1973 3855 7122 6706 3922 1974 -658 -2528 -3922 3855 1975 8381 9946 10253 -658 1976 8371 10980 10263 8381 1977 13877 9063 6864 8371 1978 5409 6575 10101 13877 1979 7503 4912 5745 5409 1980 13725 13158 13948 7503 RMSE 2261 3333 13278 0.970 0.939 -0.044 r 8P 137 Table 5.7: Actual and Forecasted values of DoD's Annual Change Requests, 1950-80 ($ Millions) Forecasted Change Requests Fiscal Actual Year Change Request No-Change CYBER.Mode1 MR Model Naive Model 1950 433 -1421 -4679 1090 1951 7898 5349 5093 433 1952 19144 20055 25999 7898 1953 4694 -1l98 800 19144 1954 -3354 -8003 2685 4694 1955 -4887 88 -2849 -3354 1956 227 -1557 -3486 -4887 1957 2702 1795 1942 227 1958 1096 -2834 3130 2702 1959 2273 5139 2907 1096 1960 27 —2588 ~1077 2273 1961 1798 4272 187 27 1962 3534 4072 4763 1798 1963 1115 2570 1466 3534 1964 1529 -5044 -322 1115 1965 -3590 -4052 2335 1529 1966 8213 3086 -5809 -3590 1967 13264 8461 16775 8213 1968 9908 11342 14088 13264 1969 520 6832 3719 9908 1970 -715 963 -3259 520 1971 -2598 749 -448 -715 1972 604 2360 -572 -2598 1973 -1853 3367 1428 604 1974 4327 1857 -4723 -1853 1975 7350 6827 11576 4327 1976 2291 4717 8623 7350 1977 7666 8850 4138 2291 1978 7485 5703 8216 7666 1979 11971 6507 9422 7485 1980 17827 5361 17214 11971 RMSE 4098 4439 5583 0.753 0.785 0.508 r 8P 138 confirms the observation made earlier which suggested that such corre- lations were not infallible guidelines for determining the accuracy of a set of predictions. In terms of its ability to generate relatively accurate histori- cal forecasts the CYBER Model has been shown to do quite well in accounting for defense expenditure change decisions for all four actors. When compared with a MR.Model of the defense expenditure policydmaking process the differences are not always striking, but for each actor the RMSEs are smaller for the CYBER Model. When compared against a No-Change Naive Model the results are even.more impressive. With the exception of the DoD expenditure changes, the RMSEs of the CYBER Model are at least three times more accurate than those of the No-Change Model. The theoretical focus of the CYBER Model is on the annual chggge in defense budget requests, recommendations, appropriations, and expenditures, though I realize that considerable scholarly and public attention is usually devoted to aggregate defense expenditure levels. By adding the previous year's defense expenditure decision (DECt_1) to the change (ADECt) it is possible to construct the aggregate defense expenditure totals. Tables 5.8 through 5.11 present the his- torical forecasts for the Services' requests, President's budget recommendations, Congressional appropriations, and DoD expenditures for the FY 1950-80 period. As with Tables 5.4 through 5.7, the actual defense expenditure decisions are compared with forecasted levels gen- erated by the CYBER, MR, and No-Change Naive Models. Several important points are worth noting. First, with the exception of the Services' request, the MR Model yields historical 139 Table 5.8: Actual and Forecasted Values of Services Requests, 1950-80 ($ Millions) Forecasted Services Requests Fézzil Actual Request No Change CYBER Model MR Model Naive Model 1950 17920 21394 15757 16505 1951 21330 16585 15438 17920 1952 81511 76939 80095 21330 1953 65390 88691 73368 81511 1954 50500 49221 60231 65390 1955 35900 32841 40736 50500 1956 37400 37678 36981 35900 1957 40600 31582 29599 37400 1958 55400 54549 54489 40600 1959 42000 47687 55458 55400 1960 48400 58382 49784 42000 1961 40703 47446 51494 48400 1962 50500 43089 40013 40703 1963 52500 56175 54938 50500 1964 59044 59334 59490 52500 1965 58884 64492 67481 59044 1966 54948 46726 48904 58884 1967 71683 61797 61011 54948 1968 92092 88208 82030 71683 1969 100746 108810 104861 92092 1970 99063 100969 105183 100746 1971 76929 83840 88850 99063 1972 74502 71274 69903 76929 1973 79779 69962 68305 74502 1974 84973 85509 86730 79779 1975 84727 86479 90833 84973 1976 96643 90105 83924 84727 1977 106334 106098 107191 96643 1978 115322 121229 118200 106334 1979 125209 124202 128750 115322 1980 135041 131899 134731 125209 RMSE 6873 7317 14850 0.973 0.970 0.875 r 8P 140 Table 5.9: Actual and Forecasted Values of President's Recommendations, 1950-80 ($ Millions) Forecasted President's Recommendations Fiscal Actual Year Recommendation No-Change CYBER Model MR Model Naive Model 1950 13307 11367 11850 11991 1951 12115 16171 14490 13307 1952 55605 55380 55196 12115 1953 51391 50184 50058 55605 1954 40720 44004 44294 51391 1955 29887 31478 34260 40720 1956 32233 29858 32325 29887 1957 34148 32426 33253 32233 1958 36158 40847 42072 34148 1959 38197 29261 33092 36128 1960 39248 42535 41518 38197 1961 39335 35774 39334 39248 1962 42942 43772 44350 39335 1963 47907 45310 47803 42942 1964 49014 50531 51441 47907 1965 47471 49131 51524 49014 1966 45249 46125 47953 47471 1967 57664 56637 56348 45249 1968 71584 72290 68791 57664 1969 77074 83962 78679 71584 1970 75278 81240 79331 77074 1971 68746 66886 67153 75278 1972 73544 70547 69713 68746 1973 79594 78551 78063 73544 1974 77251 84868 85440 79594 1975 87057 79076 82931 77251 1976 97858 94222 92143 87057 1977 106762 105831 105919 97858 1978 113877 113357 114717 106762 1979 119300 120422 121910 113877 1980 132321 125719 127818 119300 RMSE 3858 3282 10222 0.992 0.994 0.949 r 8P Table 5.10: 141 Appropriations, 1950-80 ($ Millions) Actual and Forecasted Values of Congressional Forecasted Congressional Appropriations Fiscal Actual Year Appropriation . No-Change CYBER Model MR.Model Naive Model 1950 13600 12781 12449 11403 1951 12046 12396 11105 13600 1952 54984 53430 52775 12046 1953 46611 50284 49769 54984 1954 34372 36486 37006 ~ 46611 1955 28800 24431 24403 34372 1956 31883 31355 29901 28800 1957 34657 34684 33810 31883 1958 33760 35920 36640 34657 1959 39673 36688 36907 33760 1960 39228 40081 39918 39673 1961 39997 39438 40079 39228 1962 46663 43696 43687 39997 1963 48136 52072 51027 46663 1964 47220 48635 50604 48136 1965 46752 45142 46661 47220 1966 46887 45619 44424 46752 1967 58067 59353 57886 46887 1968 69937 70790 70642 58067 1969 71870 73597 73429 69937 1970 69641 68960 69351 71870 1971 66596 63635 62753 69641 1972 70518 72321 70467 66596 1973 74373 77640 77173 70518 1974 73715 71845 73251 74373 1975 82096 83661 83504 73715 1976 90467 93076 93768 82096 1977 104344 99530 100631 90467 1978 109753 110919 110732 104344 1979 117256 114665 116477 109753 1980 130981 130414 130425 117256 RMSE 2261 2226 10040 0.997 0.997 0.946 r aP 142 Table 5.11: Actual and Forecasted Values of Bob Expenditures, 1950-80 ($ Million) Forecasted DoD Expenditures Fiscal Actual Year Expenditures No-Change CYBER Model MR.Model Naive Model 1950 11674 9820 7511 11241 1951 19572 17023 12604 11674 1952 38716 41627 38603 19572 1953 43410 37518 39403 38716 1954 40056 35407 42088 43410 1955 35169 40144 39239 40056 1956 35396 33612 35754 35169 1957 38098 37191 37696 35396 1958 31194 35264 40825 38098 1959 41467 44333 43732 39194 1960 41494 38879 42655 41467 1961 43292 45766 42842 41494 1962 46826 47364 47605 43292 1963 47941 49396 49071 46826 1964 49470 42897 48748 47941 1965 45880 45418 51083 49470 1966 54093 48966 45275 45880 1967 67357 62554 62050 54093 1968 77265 78699 76138 67357 1969 77785 84097 79857 77265 1970 77070 78748 76598 77785 1971 74472 77819 76150 77070 1972 75076 76832 75578 74472 1973 73223 78443 77006 75076 1974 77550 75080 72283 73223. 1975 84900 84377 83859 77550 1976 87891 89617 92482 84900 1977 95557 96741 96620 87891 1978 103042 101260 104835 95557 1979 115013 109549 114258 103042 1980 132840 120374 131472 115013 RMSE 4098 3250 7054 0.989 0.993 0.978 r 8P 143 forecasts which are relatively more accurate than those of the CYBER Model. This is evidenced by both the lower RMSEs for the MR Model and its higher correlations between the actual and predicted series. Sec- ond, while not intending to deny the MR Model its "victory" over the CYBER Model it should be noted that the differences between the two are not striking. For example, the RMSE of the MR Model is only slightly lower than that of the CYBER Model for the Congressional appropriation series ($2.23 billion to $2.26 billion). When one con- siders that the average Congressional appropriation for defense during this period is nearly $60 billion, a difference in forecasting ability of $30 million seems relatively inconsequential. This observation, when coupled with the explanatory deficiencies of the MR.Mode1, should not be construed as particularly damning evidence against the CYBER Model. Third, both models easily outperform the No-Change Naive Model. Finally, given the inability of existing models based upon other theoretical approaches (e.g., Rational Actor and Organizational Process) to generate forecasts which are consistently more accurate than those generated by a No—Change Naive Model (Ostrom, 1977s), I con- ' clude that the CYBER Model represents an improvement over such models. When the substantive plausibility of the CYBER Model is combined with the reported empirical results and the ability of the model to generate accurate historical forecasts of both changes and 151315, I think it is reasonable to conclude that this approach represents a significant advance in the study of the U.S. defense expenditure policy—making process. The final chapter of this dissertation will expand on this observation and will offer suggestions for future research. CHAPTER VI CONCLUDING OBSERVATIONS This dissertation began with an overview of the current conven- tional wisdom suggested by models of the U.S. defense expenditure policy-making process. At this point I would like to review the major findings and features of the CYBER Model within the context of this conventional wisdom. In so doing I shall be able to demonstrate the extent to which the CYBER Model does or does not lend support to previous research efforts in this field. The CYBER Model and the Conventional Wisdom Earlier it was shown that there is an emergent consensus which holds that the United States does not, in terms of its defense expend- iture policy, react to the behavior of the Soviet Union. Furthermore, it was argued that the conventional wisdom suggested that economic, political, and bureaucratic factors were the primary determinants of U.S. defense expenditure levels. Even those studies which focused on the annual changes in military expenditures tended to reinforce the "primacy of domestic determinants" hypothesis. The CYBER Model provides little support for such arguments. With respect to the claim that the United States does not respond to the 144 145 behavior of its major adversary the CYBER.Mode1 was able to demonstrate significant reactions by all four actors in the defense expenditure policy-making process. The Military Services, President, and Congress are responsive to the strategic balance which exists between the U.S. and the USSR. Moreover, each of these actors was shown to react to international tensions/crises in which the interests of both super- powers were at stake. The Department of Defense exhibited sensitivity to changes in Soviet expenditure behavior such that these changes were matched on a one-to—one basis. The CYBER Model is not able to support arguments which suggest the primacy of domestic factors. It will be recalled that each of the four actors was'responsive to the first-order essential variable USE, a fac- tor residing in the external, international environment. In addition, three of the four actors (Service, President, Congress) were influenced by four second-order essential variables, two of which are interna- tional rather than domestic in nature. The expenditure behavior of DoD was found to be responsive to an additional factor in the external environment. Even if one were to consider the third-order essential variables CUES as reflecting only domestic concerns, eleven of the twenty-three significant reactions in the CYBER.Model are to variables in the international environment. Six types of variables were identified in Chapter I: war, behav- ior of the USSR, international tensions/crises, macroeconomic factors, political factors, and bureaucraticlorganizational factors. Of these six types, the CYBER Model found that defense decision makers were sensitive to all subsets, with the exception of the bureaucratic fac- tors. Given the predominant conceptualization of these factors as an 146 inertial base it was not incorporated into the CYBER.Mode1. In short, the CYBER Model does not support the trends which were identified as part of the current conventional wisdom. The estimated version of the model provides an accurate representation of changes in defense expenditure decisions for the period 1947 to 1980. On an equation-by-equation basis the CYBER Model explains a large proportion of the total variance. The parameter estimates are not only sharp and plausible, but also are fully consistent with the theoretical argument that the defense expenditure policy-making process takes place within a "funnel of constrained reactivity." The historical forecasts of defense expenditure change decisions generated by the CYBER Model are more accurate than those produced by either a multiple regression model or a no-change naive model. Taken together, the empirical results of the CYBER Model are quite encouraging. Chapter 1 also identified a number of theoretical approaches to the study of U.S. defense expenditures. In terms of its characteri- zation of the policy-making process the CYBER Model fits within the tradition of organizational process models. Four actors or steps were identified: the Military Services, President, Congress, and the De- partment of Defense. Unlike most multi-actor or multi-step models of defense expenditure policy making, the CYBER.Mode1 suggests that the steps or decision-making groups are largely independent of one another. Moreover, it was argued that successive actors in the process were increasingly isolated from the external (noneprocess) environment. The empirical analysis of the CYBER Model supports this theoretical conceptualization. 147 The major point of departure between the CYBER Model and the con- ventional wisdom is the explicit characterization of the behavior of decision makers which is integral to the CYBER.Model and which is so often absent from other explanations. Underlying the CYBER Model is a view of decision makers which holds that they attend to a limited number of variables in the external environment in a hierarchic fashion. The hierarchy is structured by the importance which decision makers attach to "essential variables" in their environments. "Sur- vival" was identified as a goal which adaptive organisms seek to pre- serve and which assists them in their determination of which variables are essential. This approach is distinctive for two reasons. First, a set of assumptions concerning the characteristics of decision.makers is used to derive a model of defense expenditure policy making. As pointed out earlier, such explicit characterizations are often absent. Sec- ond, those explanations which have addressed the issue of the nature of decision makers have usually done so within a rational actor frame- work. The cybernetic perspective offered here is relatively unique, both in terms of its conceptualization and its application to the de- fense expenditure policy-making process. This approach also provided a theoretical basis for focusing on changes in annual defense expend- iture decisions, rather than on aggregate levels. Given the relative lack of attention in previous research with respect to this question, such a theoretical basis is both welcomed and necessary. Finally, a number of methodological concerns have been addressed in this dissertation. The theoretical focus on changes, for example, 148 enabled the CYBER Model to detect reactions to a larger set of vari- ables which otherwise would have been masked or overshadowed by the large, inertial base. The cybernetic approach also suggested a formal- ization of the model which required a relatively untried methodologi- cal strategy. .A procedure, identified as multi-stage residual analysis (MSRA), was introduced and employed in the estimation of the CYBER Model. 'In a comparative analysis between MSRA and a more traditional multiple regression (MR) approach it was shown that the latter yielded a substantive explanation which contained a number of inconsistencies. I would also suggest that strategy followed in this dissertation epit- omizes a "theoretical causality analysis" rather than a "statistical causality analysis." When its theoretical development and focus are combined with the empirical estimation and forecasting results, the CYBER Model emerges as a challenge to much of the current conventional wisdom. The primacy of domestic determinants of U.S. defense spending cannot be supported, nor can the argument that the United States does not react to the be- havior of the Soviet Union be substantiated. The CYBER Model also stands as a potentially significant extension to the conventional wis- dom; it holds that decision makers and decision making should be the focus of analyses of the defense policy-making process. There have been costs associated with this enterprise, and I would be remiss if I did not mention some of them. In addition, there are a number of possible areas for extension, revision, and application. The final section will raise these issues as a means of offering sug- gestions for future research. 149 Loose Ends: 'Problems and Prospects One of the major costs associated with the theoretical perspective adopted in this dissertation is that by focusing on the decisionemaking process much parsimony is sacrificed. Whether this is inevitable is not entirely clear. My presumption is that a certain amount of rglgf ‘tiyg complexity is unavoidable. While more parsimonious explanations might exist none of these take an explicit decisionemaking focus. Moreover, given the absolute complexity of the U.S. defense expenditure policy-making process, the CYBER.Mode1 developed here is really quite simple. Still, if explanatory completeness could be maintained, then perhaps a more parsimonious formulation would be preferable. On a more concrete level, there are a number of areas that can be addressed immediately. Some of these are questions of better measure- ment or operationalization, while others may require a more rigorous theoretical development. The first-order essential variable USE, for example, is not entirely satisfactory. While the CYBER Model performs quite well throughout the entire sample period, it is true that the largest forecast errors are associated with the Korean demobilization period. The model is quite successful in predicting the massive in- creases connected with both the Korean and Vietnam War eras, but the ability of the model to capture the dynamics of the de-escalation period has not proved successful. Previous attempts to explicitly incorporate a de—escalation factor have been disappointing. Perhaps U.S. battle casualties need to be supplemented with or replaced by data 150 pertaining to the size of U.S. active duty military personnel. The results obtained for the measure of Soviet expenditure behav- ior, SOVCH, are also less than satisfying. Given the considerable measurement problems associated with existing estimates of Soviet defense expenditures, perhaps an entirely new approach is called for. An index based upon the perceptions of U.S. defense planners as to the magnitude and dimensions of the Soviet threat would be preferable to existing expenditure series. Such an index, with its emphasis on per- ceptions, would also be more consistent with the cybernetic framework that is being offered. By moving away from a reliance on expenditure data researchers might enhance their ability to discern a reactive com- ponent in U.S. defense expenditure policy making. With respect to the CYBER.Mode1 I would suggest that a better measure of the Soviet threat might eliminate the one empirical anomaly uncovered by this analysis, i.e., the reaction to SOVCH at only the DoD expenditure step. The failure of the CYBER Model to discern reactions to the eco- nomic factor BURDEN also represents a problem area. The solution may require better measurement, a different operationalization, or more rigorous theoretical justification as to why reactions to deficit levels should be expected in the first place. Perhaps the willingness of the current administration to tolerate sizable deficit levels in order to finance the expansion of defense programs is 222 a new phe- nomenon. Or, it may be that the possible counter-cyclical effects of federal expenditures are not well understood and, therefore, attempts to incorporate such effects into empirical models are doomed to fail- ure. It is even possible that the logic of the Great Identity bears re-examination as it is applied to the study of defense expenditure 151 policy making. The artificial construction of the third-order essential variable CUES also presents somewhat of a problem. While the argument that decision makers are responsive to information which they provide to each other is both substantively plausible and empirically supported, greater attention should be directed to the exact manner and form in which this information is transmitted and received. In addition to the concerns raised above, there are several issues which should be explored. The role of supplemental appropriations in the defense expenditure process is neither well understood nor has it been extensively modeled. Ostrom (1978a), for example, treats supple- mentals as part of an identity relationship and he does not specify an associated policy-making rule. Most analyses, including this one, largely ignore supplementals. Given that supplementals affect the expenditure total, and given that the CYBER Model's "poorest fit" is in its explanation of the DoD expenditure change, this would seem to be a fruitful area for further research. Greater attention should also be directed toward the number and size of the relevant decisionrmaking groups. This analysis has pro- ceeded on the assumption that the four groups were fairly similar, though this is an assumption which has not been put to any test. More- over, the implications of group size and group membership composition need to be examined. If it can be demonstrated that aggregate decision-making behavior is a function of the size and composition of group membership, then such considerations will have to be explicitly incorporated into models of this process. While it seems fairly obvious that the President does not prepare his defense budget request 152 by himself, the number of other relevant actors is unclear. By the same token, little is known about who or what in the Department of Defense is responsible for expending defense appropriations. It may be that the "funnel of constrained reactivity" narrows at this point to the extent that little discretionary expenditure power exists. This question cannot be answered properly without further research. Finally, the cybernetic perspective offered here could be tested to determine its applicability to other fields. If humans are, in fact, cybernetic decision makers, then this approach should be uni- versally applicable wherever and whenever humans are involved. Other budgetary phenomena appear to be the most likely candidates. The pos- sibility exists, however, that the so-called "uncontrollable" portions of the federal budget would not be amenable to this approach. Could other types of foreign policy decisions be understood within a cyber- netic perspective? Critics charge that the current administration lacks a coherent Middle East policy and that U.S. policy in Lebanon has been controlled by events in that region. Is this behavior cyber- netic? If so, is there a basic incompatibility between long-range planning and adaptive behavior? These are just a few of the questions that could be addressed within the perspective offered here. The CYBER Model is not being offered as Ehg_explanation of the U.S. defense expenditure policy-making process. My goals have been much more modest. I have attempted to imbue this process with an explicit characterization of decision making. I have proceeded in this fashion because I believed that it would address a major short- coming in this field and that it could help to resolve a number of 153 incongruities that have emerged. To the extent that I have been suc- cessful in these efforts the overall understanding of the U.S. defense expenditure process and its participants will have been enhanced. APPENDICES Crisis Number #UNH ‘OQNO‘UI 10 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Crisis Date 1946 1946 1946 1946 1946 1946 1947 1948 1948 1948 1948 1948 1949 1950 1950 1950 1950 1950 1950 1952 1953 1954 1955 1955 1956 1956 1956 1956 1957 1957 1958 1958 APPENDIX A LIST OF CRISES OF CONCERN TO BOTH SUPERPOWERS, 1946-1980 Crisis Name Soviet-Iranian disputes Greek civil war Chinese civil war Turkey rejects USSR demands to join in Dardanelles defense Status of Trieste Palestine-Establishment of State of Israel Italian CP ousted from government role (riots) CP assumes power in Czechoslovakia Costa Rica invaded by Nicaraguan-based rebels Berlin blockade First Arab-Israeli war Cominform expels Yugoslavia Federal Republic of Germany created, Germany divided USSR downs US bomber over USSR airspace - Latvia US backs France in Indochina Korean.War US 7th Fleet moves to Taiwan Straits Puerto Rico nationalist uprisings Yugoslavia tensions Burmese operations against KMT forces Workers riot in East Berlin Taiwan Straits crisis Egyptian-Israeli tensions Costa Rica fights Nicaraguanrbased rebels Jordanian crisis-continued Arab-Israeli conflict Mideast war; Suez Canal crisis Gomulka assumes power in Poland Hungarian Revolution Jordan survives dismemberment, ousts Egyptians Syria-Turkey dispute-USSR supports Syria US-Venezuelan tensions (Nixon visit) Members of French military join Algerian revolt--deGau11e returns 154 Crisis Number 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 Crisis Date 1958 1958 1958 1959 1960 1960 1960 1960 1961 1961 1961 1962 1962 1962 1962 1963 1963 1964 1964 1964 1964 1964 1964 1965 1965 1965 1965 1965 1967 1967 1968 1968 1968 1968 1969 1969 1970 1970 1970 1971 1972 1973 1974 1974 1974 155 Crisis Name Civil disorders in Lebanone-US Marines sent PRC shells QuemoyiMatsu-Taiwan Straits Berlin Crisis Sine-Indian border clashes U-2 incident Turkish military coup Congo crisis Cuba-US tens ions Bay of Pigs Dominican Republic crisis Berlin border crisis US-Cuban tensions US sends troops to Thailand Cuban Missile Crisis PRC-India border war Civil war in Laos; US 7th Fleet sent to Gulf of Siam Haitian crisis; conflict in Dominican Republic Cyprus troubles; Greece-Turkey war threat Panama Canal Zone flag riots British put down African mutinies Coup in Brazil Tonkin Gulf incidents Congo; US airlifts Belgian troops to Stanleyville IndonesiaeMalaysia border conflicts India-Pakistan‘War Southern Rhodesian independence Dominican revolt; US intervention New border incidents between Israel, Jordan, and Syria Sine-Soviet border clash in Ussuri Island Six Day War Czech crisis Seizure of USS Pueblo by North Koreans B-52 with four H-bombs crashes near Thule, Greenland FRG-GDR tension PRC-USSR border clashes Libyan coup Israeli-UAR conflict JordanePalestine guerrillas-Syria conflict US general officers accidently land in Armenia India-Pakistani conflict; Bangladesh formed Sadat expels Soviet advisors October Middle East War Ethiopian emperor overthrown Military coup in Portugal Cyprus civil war; Turkish invasion Crisis Number 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 Crisis Date 1975 1975 1975 1975 1976 1976 1976 1976 1977 1977 1978 1978 1978 1979 1979 1979 1980 1980 156 Crisis Name USSR rejects US trade deal US ends aid; Turkey closes US bases US Mayaguez Operation Civil war in Angola Lebanese civil war US accused of bombing Siem Reap Egypt abrogates Soviet Treaty The Aegean crisis First Zaire invasion; Western intervention Ethiopian.war Nicaraguan civil war Second Zaire invasion; Western intervention Unrest in Iran Somoza regime overthrown in.Nicaragua US Embassy seized in Tehran USSR invades Afghanistan Labor unrest in Poland; formation of Solidarity Escalating violence in El Salvador APPENDIX B FARRARPGLAUBER ANALYSIS OF MULTICOLLINEARITY This appendix presents the results of a Farrar-Glauber analysis of multicollinearity. The problems associated with multicollinearity were detailed in Chapter III and will not be repeated here. The original Farrar-Glauber procedure for detecting the presence and sources of multicollinearity in a set of independent regressors (variables) consists of three parts. Part 1 conducts a chi-square test of the hypothesis that the matrix X'X is invertible, i.e., that the determinant (det) of X'X.is nonezero. In the instance of perfect multicollinearity this matrix will not be invertible, and estimates of coefficients for the inde- pendent regressors cannot be derived. Part 2 of the analysis involves regressing each of the independ- ent variables on the set of all other independent regressors. Values of Rrsquared and associated F—statistics are reported for each regression. I Part 3 provides a matrix of the partial correlations between any pair of independent regressors keeping all other variables constant. Thus (r and X . Associated ) is the partial correlation between X1 j 1.1 t-statistics are also provided. Given the multi-stage character of the CYBER.Model, multicol- linearity will be a potential problem at only the step where the set 157 158 of second-order essential variables is introduced. For this reason, the Farrar-Glauber analysis conducted below is limited to this set of regressors. All variables are as defined and operationalized in Chapter IV. Part 1: Chi-square test of hypothesis that det (X'X) - 0 Test statistic . 65.898a aSignificant at .05 level with 15 degrees of freedom. Part 2: Values of Rrsquared and Associated F-statistics of Each Independent Variable Regressed on the Others Independent Variable _J§i_ SOVCH .740 MISSILE . 866 CRISIS .068 PIE .128 BURDEN .280 NEGOPIN .806 b F-statistic 14.798“ 33.624“ 0.382 0.763 2.021 21.245“ Significant at .05 level with (5, 26) degrees of freedom. 159 Values of Partial Correlation Coefficients and Associated t-statistics of Each Independent Variable on Every Other Independent Variable (All Other Independent Variables Held Constant) Part 3: SOVCH MISSILE -.824 (-7.43)c MISSILE CRISIS .191 .199 (0.99) (1.04) PIE .127 .143 (0.65) (0.73) BURDEN -.458 -.282 NEGOPIN -o 719 ‘0890 (-5.28)c (-9.95)c cSignificant at .05 CRISIS -0179 (-0.93) .024 (0.12) .216 (1.13) PIE -' 0151 (-0.78) BURDEN 0225 -0213 (1.18) (-1011) level with 26 degrees of freedom. B IBLIOGRAPHY BIBLIOGRAPHY Abelson, Robert P. (1963). "A Derivation of Richardson's Equations." Journal 2; Conflict Resolution 7:13-15. Abolfathi, F. (1978). "Defense Expenditures in the Persian Gulf." In W. Ladd Hollist (ed.), Exploring Competitive Arms Processes: Applications 2§_Mathematical Modelipg_and Computer Simulation 13 Arms Policy Analysis. New York: Marcel Dekker, pp. 99-131. Alcock, Norman 2. and Keith W. Lowe (1969). "The Vietnam War as a Richardson Process." Journal pf Peace Research 16:105-13. Alexander, Arthur J., Abraham 8. Becker, and William E. Hoehn, Jr. (1979). .332 Sigpificance g; Divergent U.S.-USSR Military Expend— iture, A Rand Note Prepared for the United States Air Force. N-lOOO—AF. Santa Monica, Calif.: Rand Corporation. Alexander, Christopher (1964). Notes gg_the Synthesis pf Form. Cambridge, Mass.: Harvard University Press. Alker, Hayward R., Jr. (1968). "The Structure of Social Action in an Arms Race." Prepared for delivery at the Sixth North American Peace Research Society (International), Sheraton Commander Hotel, Cambridge, Mass., November 10-12. and R. D. Brunner (1969). "Simulating International Conflict: A Comparison of Three Approaches." International Studies Quarter- 12,13:70-110. Allison, Graham T. (1971). Essence g£_Decision: Explaining the Cuban Missile Crisis. Boston: Little, Brown and Company. Anderson, T. W. (1971). The Statistical Analysis 2: Time Series. New York: John Wiley and Sons. Ashby, W. Ross (1960). Desigg for §_Brain, Second Edition Revised. New York: John.Wiley and Sons. (1964). ‘Ap_Introduction‘£g_Cybernetics. London: Methuen and Company. Ashley, R., C. W. J. Granger, and R. Schmalensee (1980). "Advertising and Aggregate Consumption: An Analysis of Causality." Economet- rica 48:1149-67. 160 161 Azar, Edward (1980). "The Conflict and Peace Data Bank (COPDAB) Pro- ject." Journal 25 Conflict Resolution 24:143-52. Bailey, J. J. and R. U. O'Connor (1975). "Operationalizing Increment- alism: Measuring the Muddles." Public Administration Review 33: 60-66. Bishop, William J. and David S. Sorenson (1982). "Superpower Defense Expenditures and Foreign Policy." In Charles W. Kegley, Jr. and Patrick J. McGowan (eds.), Foreigp Policy USA/USSR. Beverly Hills: Sage. Blechman, Barry M. and Stephen 8. Kaplan (1978). Force Without War: U.S. Armed Forces ggng Political Instrument. Washington, D.C.: Brookings Institution. Bobrow, Davis B. (ed.) (1965). Components gr Defense Policy. Chi- cago: Rand McNally and Company. Boulding, Kenneth E. (1962). Conflict and Defense: A_General Theory. New York: Harper and Row. Brito, Dagobert L. (1972). "A Dynamic Model of an Armaments Race." International Economic Review 13:359-75. and Muchael D. Intriligator (1972). "A General Equilibrium Model of the Stability of an Armaments Race." In Proceedings of the VI Asilomar Conference on Circuits and Systems Theory. Burton, Donald F. (1983). "Estimating Soviet Defense Spending." Problems 2; Communism (March-April):85-93. Busch, Peter A. (1970). "Appendix: Mathematical Models of Arms Races." In Bruce Russett, What Price Vigilance? The Burdens ‘2r_National Defense. New Haven: Yale University Press. Caspary, William.R. (1967). "Richardson's Models of Arms Races: Description, Critique and an Alternative Model." International Studies Quarterly 11:63-88. Center for the Study of American Business (1980). The Congressional Budget Process: Some Views From the Inside. Proceedings of a Conference Cosponsored by: Center for the Study of American Busi- ness and the Department of Political Science, Washington Univer- sity. Publication No. 32. Central Intelligence Agency (1981). Soviet and U.S. Defense Activi- ties, 1971-80: AzDollar Cost CompariSon, A Research Paper. SR 81-10005. Chase, P. E. (1969). "Feedback Control Theory and Arms Races." General Systems‘Yearbook 14:137-49. 162 Chatterjee, P. (1974). "The Equilibrium Theory of Arms Races: Some Extensions." Journal g£_Peace Research 11:203-11. Chatterji, Manas (1969). "A.Mode1 of Resolution of Conflict between India and Pakistan." Peace Research Societngapers 12:87-102. Cockburn, Andrew (1983). The Threat: Inside the Soviet Military Machine. New York: Random House. Cockle, Paul (1978). "Analysing Soviet Defence Spending: The Debate in Perspective." Survival 20:209-19. Cohn, Stanley H. (1973). "Economic Burden of Defense Expenditures." In Soviet Economic Prospects for the Seventies. Edited by Joint Economic Committee, Congress of the United States. Washington, D.C.: United States Government Printing Office, pp. 147-62. Collins, John M. and John S. Chwat (1976). The United States/Soviet Military Balance. 'Washington, D.C.: Library of Congress. Cook, Thomas D. and Donald T. Campbell (1979). Quasi-Experimentation: Desigp and Analysis Issues for Field Settingg. Boston: Houghton Mifflin Company. Crecine, John P. (1969). Governmental Problem Solvipg. Chicago: mm. (1971). "Defense Budgeting." In.W. W. Cooper, R. F. Byrne, A. Charnes, O. A. Davis, and D. Gilford (eds.), Studies rg_Budget- ing. Amsterdam: North Holland Publishing Company, pp. 210-61. and Gregory W. Fischer (1973). "On Resource Allocation in the U.S. Department of Defense."- In Cornelius P. Cotter (ed.), Political Science Annual 4:181-236. Cusack, Thomas R. and Michael D. Ward (1981). "Military Spending in the United States, Soviet Union, and People's Republic of China." Journal gr_Conflict Resolution 25:429-69. Cyert, Richard M. and James G. March (1963). .A Behavioral Theory 2: the Firm. Englewood Cliffs, N.J.: Prentice-Hall. Cypher, James (1974). "Capitalist Planning and Military Expenditures." Review 2: Radical Political Economics 6:1-19. Davis, Otto A., M. A. H. Dempster, and Aaron Wildavsky (1966a). "A Theory of the Budgetary Process." American Political Science Review 60:529-47. (1966b). "On the Process of Budgeting: An Empirical Study of Congressional Appropriations." In Gordon Tullock (ed.), Papers gp_NoneMarket DecisionrMaking. Charlottesville, Va.: Thomas Jefferson Center for Political Economy, University of Virginia, pp. 63-132. 163 Davis, Otto A., M. A. H. Dempster, and Aaron Wildavsky (1974). "Towards a Predictive Theory of Government Expenditures." British Journal g£.Political Science 4:419-52. Dempster, M. A. H. and Aaron'Wildavsky (1979). "On Change: Or, There is No Magic Size for an Increment." Political Studies 27:371—89. Deutsch, Karl W. (1963). The Nerves 2: Government: Models 2r.Politi- cal Communication and Control. New York: The Free Press. Domke, William K., Richard C. Eichenberg, and Catherine M. Kelleher (1983). "The Illusion of Choice: Defense and Welfare in Advanced Industrial Democracies, 1948-1978." American Political Science Review 77:19-35. Engle, Robert F. (1973). "Specification of the Disturbance for Effi- cient Estimation." Econometrics 41:225-38. Enthoven, Alain C. and K. Wayne Smith (1971). How Much I§_Enough? Shaping the Defense Program, 1961-1969. New York: Harper and Row. Fallows, James (1981). National Defense. New York: Random House. Fischer, Gregory W. and John P. Crecine (1980). "Defense Budgets, Fiscal Policy, Domestic Spending and Arms Races." Paper pre- sented at the Annual Meeting of the American Political Science Association, Washington, D.C. and John P. Crecine (1981). "Defense Spending, and the Need for Fiscal Restraint." Arms Control, pp. 65-106. and Mark S. Kamlet (1981). "Explaining Presidential Priori- ties: The Competing Aspiration Levels Model of Mbcrobudgetary Decision Making." Paper presented at the Annual Meeting of the Midwest Political Science Association, Cincinnati, April 16-18. Fisher, Louis (1975). Presidential Spending Power. Princeton: Princeton University Press. Freeman, John (1983). "Granger Causality and the Time Series Analysis of Political Relationships." American Journal 25 Political Sci- ence 27:327-58. Gillespie, John V., Dina A. Zinnes, G. S. Tahim, Phillip A. Schrodt, and R. Michael Rubison (1977). "An Optimal Control Model of Arms Races." American Political Science Review 71:226-44. Gist, John R. (1974). Mandatory Expenditures and the Defense Sector. Beverly Hills: Sage. (1977). "'Increment' and 'Base' in the Congressional Appro- priations Process." American Journal gerolitical'Science 21: 164 Granger, C. W. J. (1969). "Investigating Casual Relations by Econo- metric Models and Cross-Correlational Methods." Econometrics 37:424-38. Gray, Colin 8. (1971). "The Arms Race Phenomenon." world Politics 24:39-79. Gregory, P. (1974). "Economic Growth, U.S. Defense Expenditures and the Soviet Budget." Soviet Studies 24:72-80. Halperin, Morton H. (1974). Bureaucratic Politics and Foreigp_Poligy. Washington, D.C.: Brookings Institution. Hamblin, R. L., M. Hout, J. L. L. Miller, and B. L. Pitcher (1977). "Arms Races: A Test of Two Models." American Sociological Review 42:338-54. Hammond, Paul Y. (1962). "NSC-68: Prologue to Rearmament." In.Warner R. Schilling, Paul Y. Hammond, and Glenn H. Snyder (eds.), Strategy, Politics, and Defense Budgets. New York: Columbia Uni- versity Press. Hibbs, Douglas A. (1974). "Problems of Statistical Estimation and Causal Inference in Time-Series-Regression.Models." In Herbert L. Costner (ed.), Sociologrral Methodology 1973/1974. San Fran- cisco: Jossey-Bass, pp. 252-308. Hitch, Charles J. (1965). Decision-Making for Defense. Berkeley: University of California Press. and Roland N. McKean (1965). The Economics 2; Defense 12 the Nuclear Agg, New York: Atheneum. Hollist, W. Ladd (1977a). "Alternative Explanations of Competitive Arms Processes: Tests on Four Pairs of Nations." American Journal ‘2£_Political Science 22:313-40. (l977b). "An Analysis of Arms Processes in the United States and the Soviet Union." International Studies Quarterly 21:503-28. Holzman, Franklyn (1980). "Are the Soviets Really Outspending the U.S. on Defense?" International Security 4:86-104. (1982). "Soviet Military Spending: Assessing the Numbers Game." International Security 6:78-102. Hoole, Francis W. (1976). Politics and Budgeting in the World Health Organization. Bloomington: Indiana University Press. Hoppman, T. and T. Smith (1977). "An Application of a Richardson Pro- cess Model: Soviet-American Interactions in the Test Ban Negotia- tions, 1962-63." 'Journal gr Conflict Resolution 21:701—26. 165 Huntington, Samuel P. (1961). The Common Defense: Strategic Pragrams .32 National'Politics. New York: Columbia University Press. Intriligator, Michael D. and Dagobert L. Brito (1976). "Formal Models of Arms Races." Journal gr_Peace Science 2:77-88. Johnston, J. (1972). Econometric Methods. New York: Macmillan. Kanter, Arnold (1972). "Congress and the Defense Budget: 1960-1970." American Political Science Review 66:129-43. (1979). Defense Politics: A Budgetary Perspective. Chicago: University of Chicago Press. Kaplan, Stephen S. (1981). Diplomacy gthower: Soviet Armed Forces a_s_ 2 Political Instrument. Washington, D.C.: Brookings Institu- tions Kaufmann, William W. (1964). The McNamara Strategy. New York: Harper and Row. Kmenta, Jan (1971). Elements.gr_Econometrics. New York: Macmillan. Korb, Lawrence J. (1972). "The Secretary of Defense and the Joint Chiefs of Staff: The Budgetary Process." In Sam.Sarkesian (ed.), The Military-Industrial Corplex: A Reassessment. Beverly Hills: Sage, pp. 301-40. (1973). "Congressional Impact of Defense Spending, 1962- 1973." Naval War College Review 26:49-62. Kugler, Jacek (1983). "Use of Residuals: An Option to Measure Concepts Indirectly." Political Methodology 9:103-20. Lambelet, J. (1973). "Towards a Dynamic Two-Theater Model of the East- West Arms Race." Journal 2; Peace Science 1:1-37. and Urs LuterbaCher, with P. Allan (1979). "Dynamics of Arms Races: Mutual Stimulation vs. Self-Stimulation." Journal 2; Peace Science 4:49-66. Laurance, Edward (1976). "The Changing Role of Congress in Defense Policy-Making." Journal 2: Conflict Resolution 20:213-53. Lee, William T. (1977). The Estimation 2E Soviet Defense Erpendi- ’ tures: 1955-1975, Ag Unconventional Approach. New York: Praeger. “Lindblom, Charles (1959). "The Science of 'Muddling Through.'" Public Administration Review 19:79-88. Lucier, Charles E. (1979). '"Changes in the Values of Arms Race Paramr eters." Journal 2; Conflict Resolution 23:17-41. 166 Luterbacher, Urs (1975). "Arms Race Models: Where Do We Stand?" European JOurnal gerolitical'Research 3:199-217. Maddala, G. S. (1977). Econometrics. New York: McGraw-Hill. Majeski, Stephen J. (1983). "Mathematical Models of the U.S. Military Expenditure Decision-making Process." American Journal 22 Political Science 27:485-514. and David L. Jones (1981). "Arms Race Modeling: Causality Analysis and Model Specification." Journal 2E Conflict Resolution 25:259-88. March, James G. and Herbert A. Simon (1958). Organizations. New York: John Wiley and Sons. Marra, Robin F. (1981). "The Military Balance: Tension between Fiscal and Strategic Factors." Unpublished manuscript, Michigan State University. (1983). "Alternative Models of the U.S. Defense Expenditure Process: Forecasting the Reagan Budgets." Paper presented at the Annual Meeting of the International Studies Association/Midwest, East Lansing, Michigan, November 3-5. McGuire, M. C. (1965). Secrecy and the Arms Race: A Theory 22 the Accumulation gr‘Strategic'WeapOns and How Secrecy Affects 15. Cambridge, Mass.: Harvard University Press. Moll, Kendall D. (1974). "International Conflict as a Decision Sys- tem." Journal gr_Conflict Resolution 18:555-57. and Gregory M. Luebbert (1980). "Arms Race and Military Expenditure Models." Journal gr_Conflict Resolution 24:153-85. Natchez, Peter B. and Irvin C. Bupp (1973). "Policy and Priority in the Budgetary Process." American Political Science Review 67: 951-63. Newell, Allen and Herbert A. Simon (1972). Human Problem Solvi g. Englewood Cliffs, N.J.: Prentice-Hall. Nincic, Miroslav (1982). The Arms Race: The Political Economy 2E Military Growth. New York: Praeger. (1983). "Fluctuations in Soviet Defense Spending: A.Research Note." ‘Journal g£_Conflict ReSOlution 27:648-60. and Thomas R. Cusack (1979). "The Political Economy of U.S. Military Spending." Journal 3: Peace Research 16:101-15. 167 Ostrich, John T., Jr. and William C. Green (1981). "Methodological Problems Associated with the 1188 Military Balance." Comparative Strategy 3:151-72. Ostrom, Charles W., Jr. (1977a). "Evaluating Alternative Foreign Policy Decision-Making Models: An Empirical Test between an Arms Race and an Organizational Politics Model." Journal gr_Conflict Resolution 21:235-66. (1977b). "Organizations, Arms Races, and Bureaucracies: A Three-Tiered Model of U.S. Defense Policymaking." Unpublished manuscript, Michigan State University. (1978a). "A Reactive Linkage Model of the U.S. Defense Expend- iture Policy-Making Process." American Political Science Review 72:941-57. (1978b). Time Series Analysis: Regression Techniques, Sage University Papers in Quantitative Applications for the Social Sciences. Beverly Hills: Sage. (1978c). "An Empirical Evaluation of a Richardson-Type Arms Race Model." In W. Ladd Hollist (ed.), Exploripg Competitive Arms Processes: Applications gr_Mathematical Modeling and Computer Simulation rp,Arms Policy Analysis. New York: Marcel Dekker, pp. 65-88. (1978d). "'A Theory of the Budgetary Process' as a Research Program in Public Administration." Paper presented at "Political Science and the Study of Public Policy," at the Michigan State University Kellogg Biological Station, Hickory Corners, Michigan, may 15-17. and Brian L. Job (1982). "The President and the Political Use of Force." Paper presented at the Annual Meeting of the American Political Science Association, Denver, Colorado, Septem- ber 2-5 s and Dennis M. Simon (1983). "Managing Popular Support: The Presidential Dilemma." Paper presented at the Annual Meeting of the American Political Science Association, Chicago, 111., Sep- tember 1-4. Ott, David J. and Attiat F. Ott (1977). Federal Budget 961162, Third Edition. Washington, D.C.: Brookings Institution. Pindyck, Robert S. and Daniel L. Rubinfeld (1976). Econometric Models and Economic Forecasts. New York: McGraw-Hill. Prados, John (1982). The Soviet Estimate: U.S. Intelligence Analysis and Russian Military Strength. New York: The Dial Press. 168 Price, Thomas J. (1975). "Behavior Modes: Toward a Theory of Decision- Making." Journal 2: Politics 37:417—35. Rao, Potluri and Roger LeRoy'Miller (1971). Applied Econometrics. Belmont, Calif.: Wadsworth. Rapoport, Anatol (1957). "Lewis F. Richardson's Mathematical Theory of War." Journal 2: Conflict Resolution 1:249-99. Rattinger, Hans (1975). "Armaments, Detente and Bureaucracy: The Arms Race in Europe." Journal 2; Conflict Resolution 19:571-95. (1976). "Econometrics and Arms Races: A Critical Review and Some Extensions." European Journal 2; Political Research 4: 421-39. Richardson, Lewis F. (1960). Arms and Insecurity: A Mathematical Study g£_the Causes and Origins 2; War. Pittsburgh: Boxwood. Rohde, David W. and Dennis M; Simon (1983). "Presidential Vetoes and Congressional Response: A Study of Institutional Conflict." Paper presented at the Annual Meeting of the.American Political Science Association, Chicago, 111., September 1-4. Rosefielde, Steven (1982). False Science: Underestimatipg_the Soviet Arms Buildup: Ag Appraisal g; the CIA's Direct Costing Effort, 1960-80. New Brunswick, N.J.: Transaction Books. Rosenau, James (1971). "Toward the Study of National-International Linkages." In James Rosenau (ed.), The Scientific Study g£_For- eigg Policy. New York: The Free Press, pp. 305-38. Russett, Bruce M. (1970). What Price Vigilance? The Burdens 2; National Defense. New Haven: Yale University Press. (1972). "The Revolt of the Masses." In Bruce M. Russett (ed.), Peace, War and Numbers. Beverly Hills: Sage, pp. 299-319. (1982). "Defense Expenditures and National Well-Being." American Political Science Review 76:767-77. - (1983). "International Interactions and Processes: The Internal vs. External Debate Revisited." In Ada W. Finifter (ed.), Political Science: The State 2; the Discipline. Washing- ton, D.C.: The American Political Science Association, pp. 541- 68. and Miroslav Nincic (1976). "American Opinion on the Use of ‘Military Force Abroad." Political Science anrterly 91:411-31. Saaty, Thomas L. (1968). Mathematical Models 2£_Arms Control and Dis- armament. New York: John Wiley and Sons. 169 Saris, W. and C. Middendorf (1980). "Arms Races: External Security or Domestic Pressure?" British Journal g£_Political Science 10:121-28. Schilling, Warner R. (1962). "The Politics of National Defense: Fiscal 1950." In Warner R. Schilling, Paul Y. Hammond, and Glenn H. Snyder (eds.), Strategy, Politics, and Defense Budgets. New York: Columbia University Press, pp. 1-266. Schrodt, Philip (1976). "Richardson's Model as a Markov Process." In Dina A. Zinnes and John V. Gillespie (eds.), Mathematical Models 32 International Relations. New York: Praeger. (1978). "Statistical Problems Associated with the Richardson Arms Race Model." Journal gf_Peace Science 3:159-72. Shishko, Robert (1977). "Defense Budget Interactions Revisited." Rand Paper P-5882. Santa Monica, Calif.: Rand Corporation. Simaan, Marwan and Jose B. Cruz, Jr. (1973). "A.Multi-Stage Game Formulation of Arms Race and Control and Its Relationship to Richardson's Model." In William G. Vogt and Marlin H. Mickle (eds.), Modeling_and Simulation, Vol. 4. Pittsburgh: Instrument Society of America, pp. 149-53.‘ Simon, Herbert A. (1945). Administrative Behavior. New York: Macmillan. (1957). Models g£_Man. New York: John Wiley and Sons. (1959). "Theories of Decision Making in Economics and Behav- ioral Science." American Economic Review 49:253-83. (1969). The Sciences 2; the Artificial. Cambridge, Mass.: The Massachusetts Institute of Technology Press. (1979). Models 2; Thought. New Haven: Yale University Press. SIPRI [Stockholm Institute for Peace Research International] (1968- 1983). World Armaments and Disarmament: SIPRI Yearbook 1968/69, 1970/71, 1972, 1973, 1974, 1975, 1976, 1977, 1978, 1979, 1980, 1981, 1982. London: Taylor and Francis. Smoker, Paul (1963). "A.Mathematical Study of the Present Arms Race." General Systems Yearbook, Vol. 8:51-60. (1965). "Trade, Defense and the Richardson Theory of Arms Races: A.Seven Nation Study." Journal g£_Peace Research 4:161-76. Snyder, Richard C., H. W. Bruck, and Burton Sapin (eds.) (1962). Foreig2_Policy7Decision.Making, New York: The Free Press. Sorenson, David S. (1980). "Modeling the Nuclear Arms Race: A Search for Bounded Stability." Journal 2; Peace Science 4:169-88. 170 Steinbruner, John D. (1974). The Cybernetic Theory 2: Decision. Princeton: Princeton University Press. Stoll, Richard J. (1982). "Let the Researcher Beware: The Use of the Richardson Equations to Estimate the Parameters of a Dyadic Arms Acquisition Process." American Journal g£_Political Science 26: 77-890 Strauss, R. P. (1972). "An Adaptive Expectations Model of the East- West Arms Race." Peace Research Society Papers 19:29-34. (1978). "Interdependent National Budgets: A Model of U.S.- USSR Defense Expenditures." In.W. Ladd Hollist (ed.), Exploring Competitive Arms Processes: Applications 2£_Mathematical Modeling and Computer Simulation ig_Arms Policy Analysis. New York: Marcel Dekker, pp. 89-98. Stromberg, J. L. (1970). The Internal Mechanisms ngDefense Budget Processes: Fiscal 1953-1968. U.S. Air Force Project Rand, RM- 6243-PRt Santa Monica, Calif.: Rand. Taagepera, Rein (1979/1980). "Stockpile-Budget and Ratio Interaction Models for Arms Races." Papers, Peace Science Sociegyy(Interna- tional} 29:67-78. Tufte, Edward (1974). Data Analyses for Politics and Policies. Englewood Cliffs, N.J.: Prentice-Hall. (1978). The Political Control 2f_the Economy. Princeton: Princeton University Press. Wallace, Michael D. (1979). "Arms Races and Escalation: Some New Evidence." Journal gg Conflict Resolution 23:3-16. (1980). "Accounting for Superpower Arms Spending." In Patrick.McGowan and Charles Kegley (eds.), Threats, Weapgns,.ggd Foreigg Policy, Sage International Yearbook of Foreign Policy Studies, Vol. 5. Beverly Hills: Sage, pp. 259-74. Wanat, John (1974). "Bases of Budgetary Incrementalism." American Political Science Review 68:1221-28. Weil, Herman M. (1975). "Can Bureaucracies Be Rational Actors? Foreign Policy Decision-Making in North Vietnam." International Studies anrterly 19:432-68. Wildavsky, Aaron (1964). The Politics 2; the Budgetagprrocess. Boston: Little, Brown and Company. ‘Williamson, Oliver E. (1967). "A Rational Theory of the Federal Budgeting Process." Papers 23_Non-Market Decision Making 2:71-89. 171 Wolfson, Murray (1968). V"A.Mathematica1 Model of the Cold War." Peace Research Societnyapers, Vol. 9:107-24. Yarmolinsky, Adam (1972). "The President, the Congress, and Arms Control." In Sam Sarkesian (ed.), The Military—Industrial Complex: .A Reassessment. Beverly Hills: Sage, pp. 277-300. Zimmerman, William and Glenn Palmer (1983). "Words and Deeds in Soviet Foreign Policy: The Case of Soviet Military Expenditures." American Political Science Review 77:358-67. Zinnes, Dina A. (1976). Contemporagy Research i2 International Rela- tions. New York: The Free Press. and John V. Gillespie (1973). "An Analysis of Arms Race Models: USA vs. USSR and NATO vs. WTO." In William G. Vogt and Marlin H. Mickle (eds.), Modeling and Simulation, Vol. 4. Pitts- burgh: The Instrument Society of America, pp. 145-48. , John V. Gillespie, and R. Michael Rubison (1976). "A Rein- terpretation of the Richardson's Arms Race Model." In Dina A. Zinnes and John V. Gillespie (eds.), Mathematical Models ig_Inter- national Relations. New York: Praeger.