Use of force and civil–military relations in Russia: an automated content analysis

Abstract

Russia's intervention in the Georgian–South Ossetian conflict has highlighted the need to rigorously examine trends in the public debate over the use of force in Russia. Approaching this debate through the prism of civil–military relations, we take advantage of recent methodological advances in automated content analysis and generate a new dataset of 8000 public statements made by Russia's political and military leaders during the Putin period. The data show little evidence that military elites exert a restraining influence on Russian foreign and defence policy. Although more hesitant than their political counterparts to embrace an interventionist foreign policy agenda, Russian military elites are considerably more activist in considering the use of force as an instrument of foreign policy.

Keywords:

• content analysis
• document classification
• civil–military
• use of force
• Russia
• elites

Acknowledgements

The authors would like to acknowledge assistance and helpful comments from Muhammet Bas, Matt Blackwell, Jeff Friedman, Justin Grimmer, Gary King, Kyle Marquardt, Rich Nielsen, Dan O'Huiginn and Beth Simmons. Any remaining errors are the authors' own. Financial support for this research project has been provided by the Davis Center for Russian and Eurasian Studies at Harvard University. Replication data and Appendix with additional estimation details are available on request from the authors.

Notes

 1. Medvedev, ‘Vstrecha s voennosluzhashimi Vooruzhennykh Sil Rossi’.

 2. Notable contributions to this debate include Betts, Soldiers, Statesmen and the Cold War; Brodie, War and Politics; Ekirch, The Civilian and the Military; Huntington, The Soldier and the State; Gelpi and Feaver, ‘Speak Softly and Carry a Big Stick?’; Gelpi and Feaver, Choosing Your Battles; Sechser, ‘Are Soldiers Less War-Prone than Statesmen?’; Vagts, A History of Militarism.

 3. See Quinn et al., ‘How To Analyze Political Attention With Minimal Assumptions and Costs’; Shulman,‘Editor's Introduction’, for two excellent surveys of current work in political science.

 4. This view builds on Huntington's, The Soldier and the State classic thesis on professional military ethic, which holds that ‘grand political designs and sweeping political goals are to be avoided, not because they are undesirable, but because they are impractical. The military security of the state must come first.’ Huntington, The Soldier and the State, 68.

 5. See Betts, Soldiers, Statesmen and the Cold War.

 6. See Gelpi and Feaver, ‘Speak Softly and Carry a Big Stick?’ and Choosing Your Battles. For a discussion of traditional and non-traditional military missions, see the recent special issue of Small Wars and Insurgencies 19, no. 3 (September 2008).

 7. See Vagts, A History of Militarism; Ekirch, The Civilian and the Military; Brodie, War and Politics, 495.

 8. See Snyder, The Ideology of the Offensive; Van Evera, Causes of War; Sechser, ‘Are Soldiers Less War-Prone than Statesmen?’.

 9. Notable exceptions have included , ‘Soldiers, Statesmen, Strategic Culture and China's 1950 Intervention in Korea’ and ‘Show of Force’, which have found empirical support for the military conservatism hypothesis in, respectively, China's 1950 intervention in Korea and the Taiwan Strait Crisis of 1995–1996.

10. Azrael and Payin, ‘US and Russian Policymaking’; Golts, ‘Bremya Militarizma’; Gomart, Russian Civil-Military Relations.

11. See Gelpi and Feaver, ‘Speak Softly and Carry a Big Stick?’; Sechser, ‘Are Soldiers Less War-Prone than Statesmen?’. Another alternative is opinion surveys. Gelpi and Feaver, Choosing Your Battles analyzed survey data on civilian and military attitudes to test the hypothesis that US civilian elites are more supportive of intervention than military elites.

12. Manual thematic analysis of speeches and articles by party and military elites has been used to gauge trends in leadership commentary over a range of substantive areas, from education policy (Stewart, ‘Soviet Interest Groups and the Policy Process’) to defense expenditures (Zimmerman and Palmer, ‘Words and Deeds in Soviet Foreign Policy’), deception (Axelord and Zimmerman, ‘The Soviet Press on Soviet Foreign Policy’) and articulation of dissatisfaction with federal policies (Breslauer, ‘Is There a Generation Gap in the Soviet Political Establishment?’). While the statistical approach to content analysis is new for this literature, the underlying analytic assumptions follow from this previous tradition.

13. Frost, ‘A Content Analysis of Recent Soviet Party-Military Relations’.

14. Stewart, Warhola and Blough, ‘Issue Salience and Foreign Policy Role Specialization’; Frost, ‘A Content Analysis of Recent Soviet Party-Military Relations’; Wallander, ‘Third World Conflict in Soviet Military Thought’.

15. Zimmerman, The Russian People and Foreign Policy and ‘Slavophiles and Westernizers Redux’.

16. To be ‘salient’, an actor's statements on a given issue need not take the form of coherent policy proposals. These assumptions find precedent in previous work primarily in American politics on senate press releases (Grimmer, ‘A Bayesian Hierarchical Topic Model’) and senate floor debates (Quinn et al., How to Analyze Political Attention’; Monroe et al., ‘Fightin' Words’).

17. Such differences in expressed priorities can, of course, be dictated by bureaucratic portfolios, making it difficult to separate the salience of an issue from the number of personnel assigned to it. Yet since we are interested primarily in expressed opinions, it is of secondary importance whether a statement is informed by bureaucratic parochialism, privately held views, or some combination of the two. In either case, the statement conveys the relative importance a group (Military/Political) publicly assigns to a given policy area.

18. Gelpi and Feaver, Choosing Your Battles, 22–3.

19. Axelrod and Zimmerman, ‘The Soviet Press on Soviet Foreign Policy’.

20. We can classify these subtleties by reference to the document's vocabulary, which is a function of tone and content. This model of text analysis is well established both in sentiment analysis (Godbole et al., ‘Large-Scale Sentiment Analysis for News and Blogs’; Hopkins and King, ‘A Method of Automated Nonparametric Content Analysis for Social Science’) and bag-of-words classification (Manning and Schütze, Foundations of Statistical Natural Language Processing).

21. Huntington, The Soldier and the State.

22. Golts, ‘Bremya Militarizma’ and ‘“Grazhdanskiy kontrol” po-putinnski’; Gomart, Russian Civil-Military Relations; Miller and Trenin, Vooruzhennye sily Rossii; Golts and Putnam, ‘State Militarism and Its Legacies’.

23. Gelpi and Feaver, ‘Speak Softly and Carry a Big Stick?’. For a discussion of contested institutions and civil–military relations, see Dassel, ‘Civilians, Soldiers and Strife’.

24. The corpus of documents includes 7920 statements by 533 individuals in positions of formal authority and informal influence, published in government and independent periodicals. While the sample is neither random nor universal, it is a reasonable representation of the public debate. A full source list can be found in Appendix A (available on request from authors).

25. Full lists of all considered elites are available in Appendix B1.

26. Huntington, The Soldier and the State, 86–90.

27. Manning and Schütze, Foundations of Statistical Natural Language Processing, 237.

28. Only unigrams (single words) are considered. Full text processing details can be found in Appendix C1.

29. While Russian is more grammatically rigid and morphologically more complex than English, it is roughly as sensitive to word order as English, making the BOW assumption roughly equivalent across the two languages.

30. Quinn et al., ‘How to Analyze Political Attention’.

31. Grimmer, ‘A Bayesian Hierarchical Topic Model for Political Texts’, 4.

32. Blei et al., ‘Latent Dirichlet Allocation’; Grimmer, ‘A Bayesian Hierarchical Topic Model for Political Texts’; Quinn et al., ‘How to Analyze Political Attention’.

33. Clusters are often described by the words that most influence the placement of a document in a particular category. Further lists of key words are available in Appendix C3.

34. Examples of each type of statement are provided in Appendix B.

35. Whether in automated or manual document analysis, error rates or inter-coder reliability rates are rarely taken into account at the level of estimation. Due to our relatively small corpus of documents we are unable to adopt techniques like those described in Hopkins and King, ‘A Method of Automated Nonparametric Content Analysis for Social Science’, which rely on having tens of thousands of documents at a minimum.

36. Polikar, ‘Ensemble Based Systems in Decision Making’.

37. Specific settings, feature extraction techniques and package for estimation are available in Appendix C4.

38. Hechenbiechler and Schliep, ‘Weighted k-Nearest-Neighbor Techniques and Ordinal Classification’.

39. Breiman, ‘Random Forests’.

40. Hillard et al., ‘Computer Assisted Topic Classification for Mixed Methods Social Science Research’; Yang and Liu, ‘A Re-examination of Text Categorization Methods’.

41. The 95% confidence intervals are the 5th and 95th quantiles of the distribution.

42. Frost, ‘A Content Analysis of Recent Soviet Party-Military Relations’.

43. Wallander, ‘Third World Conflict in Soviet Military Thought’.

44. The authors are grateful to Kyle Marquardt for this insight.

45. Colton, Commissars, Commanders, and Civilian Authority.

46. See Bondaletov, ‘“Sotsial'no-protestnaya aktivnost” voennosluzhashchikh’.

47. Golts, ‘Bremya Militarizma’ and ‘“Grazhdanskiy kontrol” po-putinnski’; Gomart, Russian Civil-Military Relations; Miller and Trenin, Vooruzhennye sily Rossii; Golts and Putnam, ‘State Militarism and Its Legacies’.

48. Dubnov, ‘Tadjikistan’; Selivanova, ‘Trans-Dniestria’.

49. The bar plot shows expected probabilities and 95% confidence intervals derived from simulation.

50. The use of Euclidean distance allows us to measure disagreement between our groups on two different variables simultaneously. Operationalizing disagreement by simple spatial models is common in the study of parties (Poole, Spatial Models of Parliamentary Voting) and here does not make any additional scaling assumptions other than that the two dimensions are equally important.

51. The plot shows 20 simulated slope coefficients for a linear regression of disagreement on elite integration, controlling for annual press freedom scores (Freedom House, ‘Press Freedom Scores: 2009 Edition’). By providing simulations across 20 iterations of the model, we not only demonstrate the 95% confidence interval of the model (the upper and lower most lines), but also show that the relationship is robust across multiple draws of the Use of Force Classification. Regression lines substantively show the effect on disagreement of increasing integration while holding Press Freedom at its mean.

52. Domke, War and the Changing Global System; Gelpi, ‘Democratic Diversions’; Levy, ‘Declining Power and the Preventative Motivation for War’ and ‘The Diversionary Theory of War’; Ostrom and Job, ‘The President and the Political Use of Force’; Russett, ‘Economic Decline, Electoral Pressure, and the Initiation of Interstate Conflict’; Skocpol, States and Social Revolution; Wilkenfeld, ‘Domestic and Foreign Conflict Behaviour of Nations’ and Conflict Behaviour and Linkage Politics; Zinnes and Wilkenfeld, ‘An Analysis of Foreign Conflict Behavior of Nations’.

53. Coser, The Functions of Social Conflict; Mueller, War, Presidents and Public Opinion; Polsby, Congress and the Presidency; Simmel, ‘The Persistence of Social Groups’; Waltz, ‘Electoral Punishment and Foreign Policy Crisis’.