Manpower and counterinsurgency: Empirical foundations for theory and doctrine

Security Studies

Volumn 20, Issue 4, 2011, Pages 556-591

  Friedman, Jeffrey A ^a,b  

^a HARVARD UNIVERSITY   (United States)

^b HARVARD UNIVERSITY   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 84859014736     PISSN: 09636412     EISSN: 15561852     Source Type: Journal    
DOI: 10.1080/09636412.2011.625768     Document Type: Article

References (157)

1. John A. Nagl, "Forward," in The U.S. Army/Marine Corps Counterinsurgency FieldManual (FM-3-24/MCWP 3-33.5), US Department of the Army, (Chicago: University of Chicago Press, 2007), xiv (hereafter cited as FM 3-24)
2. Sarah Sewall, "A Radical Field Manual," in FM 3-24.
3. David H. Petraeus and James F. Amos, "Forward" and "Preface," in FM 3-24, xlv, xlviii.
4. Ibid., xlvii, xlv. Cf. Barry R. Posen, The Sources of Military Doctrine: France, Britain and Germany between the World Wars (Ithaca: Cornell University Press, 1984), 14, which states, "Military doctrine includes the preferred mode of a group of services, a single service, or a subservice for fighting wars. It reflects the judgments of professional military officers and to a lesser, but important, extent civilian leaders, about what is and is not militarily possible and necessary."
5. Petraeus and Amos, "Preface," in FM 3-24, xlvii.
6. Frederick W. Kagan, "Why We Need More Troops in Afghanistan," Washington Post, 16 August 2009
7. Seth G. Jones et al., Establishing Law and Order After Conflict (Santa Monica: RAND, 2005), xiii
8. Michael E. O'Hanlon, The Science of War (Princeton: Princeton University Press, 2009)
9. Kalev I. Sepp, "Best Practices in Counterinsurgency," Military Review 85, no. 3 (May/June 2005)
10. Alan J. Kuperman, "Rwanda in Retrospect," Foreign Affairs 79, no. 1 (January/February 2000)
11. James F. Dobbins et al., America's Role in Nation-Building (Santa Monica: RAND, 2003)
12. R. Royce Kneece et al., Force Sizing for Stability Operations (Alexandria, VA: IDA, 2010)
13. James Quinlivan, "Force Requirements in Stability Operations," Parameters 25, no. 4 (Winter 1995)
14. James Quinlivan, "Burden of Victory," RAND Review (Summer 2003).
15. T. E. Lawrence, "The Evolution of a Revolt," Army Quarterly and Defense Journal (October 1920): 7-8.
16. Robert Thompson, Defeating Communist Insurgency (London: Chatto and Windus, 1966)
17. David Galula, Counterinsurgency Warfare (New York: Praeger, 1964)
18. and more recently, Anthony James Joes, Resisting Rebellion (Lexington: University Press of Kentucky, 2004), 171.
19. FM 3-24, par. 1-131.
20. See also par. 1-3, "Political power is the central issue in insurgencies and counterinsurgencies; each side aims to get the population to accept its governance or authority as legitimate"
21. par. 1-4, "Long-term succession COIN depends on the people taking charge of their affairs and consenting to the government's rule"
22. par. 1-108, "In almost every case, counterinsurgents face a populace containing an active minority supporting the government and an equally small military faction opposing it. Success requires the government to be accepted as legitimate by most of that uncommitted middle"
23. par. 1-113, "The primary objective of any COIN operation is to foster effective governance by a legitimate government"
24. par. 1-123, "Political factors have primacy in COIN"
25. par. 1-128, "It is easier to separate an insurgency from its resources and let it die than to kill every insurgent"
26. par. 1-129, "Dynamic insurgencies can replace losses quickly. Skillful counterinsurgents must thus cut off the sources of that recuperative power. Some sources can be reduced by addressing the social, political, and economic grievances that fuel the insurgency. Physical support can be cut off by population control or border security"
27. par. 1-149, "Ultimate success in COIN is gained by protecting the populace, not the COIN force."
28. Kneece, et al., Force Sizing for Stability Operations. These data were, in turn, based on the "Irregular Warfare" data set constructed by the Center for Army Analysis (CAA). The case universe for this data set covers 102 cases, but only 57 of them qualify as "insurgencies" under Jason Lyall and Isaiah Wilson's definition, and force size information is only provided for forty-one of these. For a discussion,
29. see ibid., 47ff.
30. Ibid., 48.
31. The importance of a "light footprint" is often associated with the "British school" of counterinsurgency.
32. See, for example, Journal of Strategic Studies 32, no. 3.
33. On the importance of diminishing marginal returns for military planning.
34. see Hans Binnendijk and Stuart Johnson eds., Transforming for Stability and Reconstruction Operations (Washington, DC: Center for Technology and National Security Policy, 2003), 37.
35. Nagl, Institutionalizing Adaptation (Washington, DC: Center for a New American Security, 2007).
36. FM 3-24, par. 1-68, cf. 1-131, 1-129, 1-149.
37. See, for example, Roger Trinquier, Modern Warfare (New York: Praeger, 1961)
38. Stathis N. Kalyvas, The Logic of Violence in Civil War (New York: Cambridge University Press, 2006)
39. FM 3-24, par. 1-124, 1-126.
40. See, for example, Galula, Modern Warfare, 82
41. Joes, Resisting Rebellion, 234
42. FM 3-24, par. 1-134.
43. See, for example, Joes, Resisting Rebellion, 234
44. Dobbins, et al., America's Role in Nation-Building, xxv
45. FM 3-24, par. 1-150, 1-151.
46. For instance, in his 2009 report to the president recommending that additional forces be sent to Afghanistan, General McChrystal writes, "[I]nadequate resources will likely result in failure.... A 'properly-resourced' strategy is imperative. Resourcing coalition forces below this level will leave critical areas of Afghanistan open to insurgent influence... . Failure to provide adequate resources also risks a longer conflict, greater casualties, higher overall costs, and ultimately, a critical loss of political support. Any of these risks, in turn, are likely to result in mission failure." Headquarters International Security Assistance Force, "COMISAF Initial Assessment," (Unclassified), 30 August 2009 (hereafter cited as HQ-ISAF, "COMISAF Initial Assessment"). 19 It is important to point out that the scope of these studies is not necessarily the same-Quinlivan's paper focuses on "stability operations"
47. Dobbins, et al. and Jones, et al. examine "nation-building"
48. Thomson, Joes, and Galula study "counterinsurgency"-and so recommendations may differ in many ways simply as a function of different units of analysis.
49. See citations in notes 5 and 7. The Institute for Defense Analyses (Kneece et al., Force Sizing for Stability Operations) recommends a troop density threshold of roughly forty troops per one thousand inhabitants in the AO. However, this recommendation is not based on the notion that the odds of success markedly increase at this level of force size; it is based on the predicted probabilities generated from a logistic regression which indicate that counterinsurgents with this size deployment win roughly 75 percent of the time.
50. Quinlivan, "Force Requirements in Stability Operations"
51. Dobbins et al., America's Role in Nation-Building. A third, related report is John J. McGrath, Boots on the Ground: Troop Density in Contingency Operations (Ft. Leavenworth, KS: Combat Studies Institute Press, 2006), which examines seven case studies and recommends thirteen troops per one thousand inhabitants in the AO.
52. These are the UK in Malaya and in Ulster, the United States in the Dominican Republic and in Germany, the UN in Cambodia, and India in Punjab.
53. Quinlivan, "Burden of Victory."
54. Dobbins et al., America's Role in Nation-Building, xvii, 150-51.
55. Ibid. The authors write, "It is unlikely that Iraq would get the same per capita level of international troops, police presence, or foreign aid that Bosnia and Kosovo did. Nonetheless, lessons should also be drawn from the British experience ... of seeking to secure Iraq on the cheap" (ibid., 194). And later, "Figure 10.1 ... indicates that, if Kosovo levels of troop commitment are used, some 526,000 foreign troops would need to be deployed through 2005. At Bosnian levels, this figure would be roughly 258,000 by 2005
56. Approximately 145,000 international troops would still be required to ensure security at Bosnia levels through 2008" (ibid., 197).
57. Jason Lyall and Isaiah Wilson III, "Rage against the Machines: Explaining Outcomes in Counterinsurgency Wars," International Organization 63, no. 1 (Winter 2009). There are actually 173 cases in the "modern period" of Lyall and Wilson's data set, but two were dropped from this analysis. For the Belgian defense of Rwanda (1956-62), it is difficult to distinguish between insurgents and counterinsurgents since the Belgians supported the transition to Hutu rule. For China's 1918 campaign in Tibet, it was difficult to find reliable data on the number of forces deployed. The rest of Lyall and Wilson's case universe is included in the data set.
58. Lyall and Wilson provide the following rule for inclusion in their data set: "An insurgency is defined here as a protracted violent struggle by nonstate actors to obtain their political objectives-often independence, greater autonomy, or subversion of existing authorities-against the current political authority (the incumbent). Two rules for defining a case were chosen: First, we imposed a minimum 1,000 battle death inclusion rule, with at least 100 casualties suffered on each side. Second, the nonstate actor must have adopted a guerrilla warfare strategy. Here, guerrilla warfare is defined as a strategy of armed resistance that (1) uses small, mobile groups to inflict punishment on the incumbent through hit-and-run strikes while avoiding direct battle when possible and (2) seeks to win the allegiance of at least some portion of the noncombatant population. An insurgency is therefore not synonymous with 'civil war' since civil wars can be fought conventionally (that is, with direct battles between opposing armies), with guerrilla tactics, or through nonviolence" (ibid., 70).
59. For quantitative analyses, see, for instance, David E. Cunningham et al., "Non-State Actor Data," http://privatewww.essex.ac.uk/ksg/eacd.html (accessed 1 September 2009), which provides data on rebel force size for some conflicts
60. Valentino et al., "Draining the Sea: Mass Killing, Genocide, and Guerilla Warfare," International Organization 50, no. 4 (Spring 2004), which includes data on the population of the AO for some conflicts
61. Patricia L. Sullivan and Michael T. Koch, "Military Intervention by Powerful States," Journal of Peace Research 46, no. 5 (September 2009), which provides information on troop commitments by great powers for some conflicts. For excellent general resources on armed conflict, demography, and counterinsurgency,
62. see, for instance, Micheal Clodfelter, Warfare and Armed Conflicts (Jefferson, NC: McFarland, 2008)
63. D. M. Condit et al., Challenge and Response in Internal Conflict (Washington, DC: American University, 1957-58)
64. Robert B. Asprey, War in the Shadows (London: Little, Brown, 1975). For reliable resources on demographic and geographic information,
65. see Brian R. Mitchell, International Historical Statistics (New York: Palgrave, 2007)
66. Saul B. Cohen, The Columbia Gazetteer of the World (New York: Columbia University Press, 2008).
67. Lyall and Wilson, "Rage against the Machines," 71.
68. This is a fundamental issue in the literature on grand strategy-in particular, whether the United States and its allies can sustainably maintain a policy of militarily confronting their enemies, and if so, whether their goal should be to contain those enemies or to destroy them.
69. See, for example, Thomas P. M. Barnett, The Pentagon's New Map (New York: Penguin, 2005)
70. Stephen Biddle, American Grand Strategy after 9/11 (Carlisle, PA: Army War College, 2006)
71. Philip H. Gordon, Winning the Right War (New York: Henry Holt, 2007)
72. Ian Shapiro, Containment (Princeton: Princeton University Press, 2007)
73. Michele A. Flournoy and Shawn Brimley, eds., Finding Our Way (Washington, DC: Center for a New American Security, 2008)
74. Melvyn P. Leffler and Jeffrey W. Legro, eds., To Lead the World (New York: Oxford University Press, 2008).
75. Yoav Gortzak, "Using Indigenous Forces in Counterinsurgency Operations: The French in Algeria, 1954-1962," Journal of Strategic Studies 32, no. 2 (2009), 308
76. Martin S. Alexander and J. F. V. Keiger, "France and the Algerian War: Strategy, Operations, Diplomacy," Journal of Strategic Studies 25, no. 2 (2002), 14
77. Anthony Clayton, Frontiersmen: Warfare in Africa Since 1950 (London: UCL Press, 1999):
78. Edgar O'Ballance, The Algerian Insurrection, 1954-62 (London: Faber and Faber, 1967)
79. Condit, et al., Challenge and Response in Internal Conflict, vol. 2, The Experience in Europe and the Middle East (Washington, DC: American University, 1968), 189.
80. See the data set for annual figures.
81. Rod Thornton, "The British Army and the Origins of its Minimum Force Philosophy," Small Wars and Insurgencies 15, no. 1 (2007), 75-83
82. Caroline Kennedy-Pipe and Colin McInnes, "The British Army in Northern Ireland," Journal of Strategic Studies 20, no. 2 (2007), 12
83. Sullivan and Koch, "Military Intervention by Powerful States"
84. Quinlivan, "Force Requirements in Stability Operations"
85. Clodfelter, Warfare and Armed Conflicts, 578.
86. See the data set for annual figures.
87. R. Ernest Dupuy and Trevor N. Dupuy, The Harper Encyclopedia of Military History (New York: HarperCollins, 1993)
88. Clodfelter, Warfare and Armed Conflicts.
89. See the data set for annual figures.
90. Seth Jones, In the Graveyard of Empires: America's War in Afghanistan (New York: W. W. Norton, 2009), 26-36
91. Clodfelter, Warfare and Armed Conflicts, 648
92. Anthony James Joes, Modern Guerilla Insurgency (Westport: Praeger, 1992), 174
93. David C. Isby, War in a Distant Country (London: Arms and Armour, 1992), 81
94. Ian F. W. Beckett, Modern Insurgencies and Counter-Insurgencies (New York: Routledge, 2001), 210
95. Scott R. McMichael, Stumbling Bear: Soviet Military Performance in Afghanistan (New York: Brassey's, 1991), 13.
96. See the data set for annual figures.
97. Panel data could be useful for this purpose, but it would face two main challenges. The first is data availability: reliably estimating troop density fluctuations within campaigns would be far harder than determining their maxima, especially if the researcher allowed for variation in force size and the area of operations. The second challenge is determining and coding an appropriate dependent variable that would capture intra-conflict variations in counterinsurgents' effectiveness, especially since this measure would need to be valid across the full range of conflicts in the data set. There is no reason to believe these challenges are insurmountable, but addressing them would make for a much more complicated (and much different) empirical project.
98. The Military Balance 2006 (London: International Institute for Strategic Studies), 166, reported two hundred forty thousand Iraqi Security Forces and one hundred thirty-five thousand US personnel.
99. Control variables include the four variables from Lyall and Wilson's paper that are statistically significant across the modern era. These are: a dummy indicating whether the counterinsurgent was a foreign "occupier," an ordered variable indicating the extent to which the insurgents received support from a third party, an ordered variable capturing the degree to which the counterinsurgents' forces were "mechanized," and each counterinsurgent's Polity score.
100. See the appendix for more detail on these variables, along with Lyall and Wilson, "Rage against the Machines," 88, 89, 93.
101. In univariate regression, here are the McFadden's R2, McKelvey and Zavoina's R2, and Efron's R2 terms for each variable: troops (.01, .03, .01), troops per square kilometer (.00, .00, .00), troops per insurgent (.01, .02, .02), and troops per inhabitant (.06, .31, .08). In multivariate regression, it is appropriate to use the Akaike Information Criterion, which calculates model fit while introducing a penalty for overfitting. Lower AIC scores indicate better model fit. The respective AIC scores for models including each variable with controls are: troops (1.23), troops per square kilometer (1.23), troops per insurgent (1.21), and troops per inhabitant (1.14). Troop density outperforms the alternative specifications on measures of McFadden's R2, McKelvey and Zavoina's R2, and Efron's R2 in these multivariate models as well.
102. Its advantage over tie-down ratios, however, is not as stark as some people claim.
103. See, for example, Richard L. Clutterbuck, The Long, Long War (New York: Praeger, 1966), 42-43. "Much nonsense is heard on the subject of tie-down ratios in guerilla warfare. ... This is a dangerous illusion, arising from a disregard of the facts" (ibid., 42-43).
104. Table 4 will show that this holds true for both foreign and indigenous counterinsurgents.
105. Estimating this relationship using cubic splines, LOWESS curves, or local polynomial smoothing techniques produces a pattern that is essentially the same: none of them indicates that there are any prominent thresholds for sizing forces nor that the odds of success climb dramatically in response to changes in force size. The main difference in employing these techniques is that they emphasize how counterinsurgents with troop densities between forty and seventy-five troops per one thousand inhabitants in the AO (n = 23) underperform in comparison to the rest of the data. This is an issue discussed in Figure 2 and in the section that follows. None of these techniques indicates that there are any prominent thresholds for sizing forces
106. Specifically, there is no point in the distribution where marginal increases lead to marked jumps in the probability of counterinsurgent success.
107. Thirty-three percent (29 of 89) of counterinsurgents who did not meet this threshold were successful. If we halved the threshold recommended in FM 3-24, then there is only a marginal change in these statistics. Thirty percent (19 of 63) of counterinsurgents with fewer than ten troops per thousand inhabitants were successful as opposed to 40 percent (43 of 108) of counterinsurgents who exceeded that number.
108. The x-axis is bounded at eighty troops per one thousand inhabitants because there is only one counterinsurgent in the data set with a troop density between eighty and 115.
109. The right side of the distribution is bounded at 160 troops per 1,000 inhabitants because there are only three observations between that point and 290 troops per 1,000 inhabitants. Extending the range of the plot to this point would mask the drop in performance in the middle of the distribution and make the relationship seem more consistent than it really is.
110. At the same time, this could simply be the result of a small sample size, since there are only twenty-three observations in the range.
111. Another way to test the functional form of this relationship is to use polynomial terms, but neither a second-order nor a third-order term for troop density approaches statistical significance when added (either individually or together) to any of the regressions in Table 1 or in Table 4 below.
112. Moreover, the p < .05 standard is inappropriate for testing so many hypotheses at once. Bonferroni's inequality indicates that this standard should be adjusted to p < .0025 because Table 2 is testing twenty separate hypotheses at once, increasing the chances that one will be supported by spurious variation in the data. When every threshold dummy and every interaction term is combined (along with troop density) in a single multivariate regression, there is one threshold (thirty-five troops per one thousand inhabitants in the AO) with a positive coefficient that meets a Bonferroni-adjusted standard for statistical significance. However, the term for the interaction of this threshold and troop density is negative, and it is so large (close to -2,000) that it drowns out the dummy's substantive impact.
113. Perhaps surprisingly, these variables are not correlated in a very consistent way. Their correlation coefficient is in fact weakly negative (-0.05 under the first coding and -0.09 under the second coding). Even if observations are limited to cases where the number of foreign counterinsurgents is greater than zero, the correlation coefficient is -0.10 for both codings.
114. The same pattern holds when restricting the model to cases where the number of foreign insurgents is greater than zero. In fact, the pattern becomes more pronounced: the gap in coefficients widens, but the coefficient on indigenous troop density becomes less statistically significant. This suggests that foreign states may be more likely to intervene in wars that are less challenging on the whole, a topic discussed in the next section.
115. These values correspond to a moderate level of mechanization (mech = 2), the counterinsurgent not being an occupier (Occupation Dummy = 0), the insurgent receiving either sanctuary or military assistance (Insurgent Support = 1), the insurgents fighting for ethnic or religious issues (Identity War = 1), the insurgents having a force size of twelve thousand, and the counterinsurgents consisting of 100 percent indigenous forces. The substantive results in Figure 3 do not change if the variables are instead set to their means, maxima, or minima: doing so adjusts the baseline probability of success but not its responsiveness to troop density.
116. For example, one suggestion for an instrument could be the counterinsurgent's regime type, especially given that Lyall found this variable to be uncorrelated with counterinsurgency outcomes.
117. Jason Lyall, "Do Democracies Make Inferior Counterinsurgents?" International Organization 64, no. 1 (Winter 2010). There are two problems with using this instrument, however. The first is that regime type is a weak predictor of force size. The correlation coefficient between Polity score and troop density is -0.03, and the relationship is far from being statistically significant (in a univariate ordinary least squares regression, for instance, its p-value is roughly 0.7). Second, even though counterinsurgents' Polity score is not correlated with outcomes in a probit analysis per se, it appears that anocracies (states with Polity scores between -6 and +6) are significantly less likely to win counterinsurgency wars (see Fig. 5d below). A potentially bimodal relationship between regime type and strategic outcomes does not necessarily challenge Lyall's main finding, but it casts doubt on the idea that regime type would be theoretically valid, even if it were not a weak instrument. In general, structuring the analysis here around instruments would lead to a very different paper and one that would need to spend a large amount of space defending the instruments' validity. This would be a valuable task; but given that the goal of this paper is to establish basic empirical foundations to support a wide range of debates, the use of instruments or other econometric techniques is appropriate for future work.
118. Matching techniques that allow for continuous treatments are difficult to design or to implement, and they can be sensitive to assumptions. The theoretical literature on this method is relatively new.
119. See, for example, Kosuke Imai and David A. van Dyk, "Causal Inference with General Treatment Regimes," Journal of the American Statistical Association 99, no. 467 (September 2004) for a basic foundation of this approach, though there are few examples of it being implemented in applied empirical work. This could be another valuable way to expand the analysis here; but as with employing instrumental variables, it is
120. Not suited for a paper designed to establish basic empirical foundations for future work.
121. In theory, it is possible that threshold effects exist but that they vary in an essentially uniform distribution across campaigns. If true, this might conceivably be driving the linear patterns in the previous section. At the same time, it would be quite remarkable if these thresholds were indeed distributed so uniformly. And if this is the case, then it is still an argument against adopting a fixed rule of thumb for force sizing like the one in FM 3-24.
122. Lyall and Wilson, "Rage against the Machines," 88, Model 4. The variables are counterinsurgent regime type, insurgent's material support, per capita energy consumption, whether the counterinsurgent is an occupier, the logged elevation of the AO, the distance of the AO from the incumbent's capital, and whether the conflict took place during the Cold War.
123. Similarly, the correlation coefficients between difficulty and troop density are negative using a logit model with only those variables that Lyall and Wilson find significant (-0.01)
124. using Lyall and Wilson's Model 6, which includes mechanization and drops the Cold War dummy (-0.06)
125. and using only the variables that Lyall and Wilson find significant in Model 6 (-0.06). This is not to say that it would be impossible to construct an index of difficulty that is positively related to troop density, only that there is evidence to suggest that the correlation may actually run in the opposite direction, that this correlation is not particularly strong, and that the findings in this paper are therefore not being undermined by some obvious selection effect.
126. Patrick M. Regan, "Choosing to Intervene," Journal of Politics 60, no. 3 (1998)
127. James D. Fearon and David Laitin, "Ethnicity, Insurgency, and Civil War," American Political Science Review 97, no. 1 (February 2003).
128. Colonial wars and insurgencies during WWII were combined because neither group alone had enough observations to create a meaningful sample.
129. On the impact of conflict type, see, for example, James Fearon, "Why Do Some Civil Wars Last So Much Longer than Others?" Journal of Peace Research 41, no. 3 (May 2004)
130. Stephen Biddle, "Seeing Baghdad, Thinking Saigon," Foreign Affairs 85, no. 2 (March/April 2006)
131. Monica Duffy Toft, Securing the Peace (Princeton: Princeton University Press, 2010).
132. For instance, a prominent debate in the literature on counterinsurgency is the importance of regular soldiers relative to police or militia. Just as Lyall and Wilson code data on the extent to which counterinsurgent forces were mechanized, it would be possible to code data on the extent to which counterinsurgents integrated police or militia forces into their security structure. In general, there are any number of ways to define the "treatment regime" that counterinsurgents are employing: data on force size provide important traction for determining whether or not the causal impact of these factors runs through the channel of force employment. On controversy over the importance of police in counterinsurgency
133. see, for example, G.D.T. Shaw, "Policeman versus Soldiers: the Debate Leading to MAAG Objections and Washington Rejections of the Core of the British Counter Insurgency Advice," Small Wars and Insurgencies 12, vol. 2 (Summer 2001)
134. Ivan Arreguin-Toft, "Unconventional Deterrence" in Complex Deterrence: Strategy in a Global Age, ed. T.V. Paul et al. (Chicago: University of Chicago Press, 2009).
135. This finding narrowly misses being statistically significant at the p < .05 level. In a probit regression with troop density, and African dummy, and an interaction term, the coefficient (standard error) on the interaction term is -28.9 (15.2).
136. In a probit regression examining troop density, a dummy for anocracy, and an interaction term, the coefficient (standard error) for the interaction is -5.9 (2.7), which is significant at the p < .05 level.
137. Here are some other ways to divide the data that do not meaningfully alter the relationship between force size and success: partitioning conflicts by duration, by the per capita income of the host nation, by the per capita energy consumption of the host nation, by the urban population of the host nation, by whether the counterinsurgency was one of the forty "relevant" campaigns in Sepp, "Best Practices in Counterinsurgency", by the size of the insurgent force (in absolute or per capita terms), by rebel lethality, by counterinsurgent lethality, by each counterinsurgent's overall level of military expenditures per troop, and by terrain type.
138. Stephen Walt, "The Renaissance of Security Studies," International Studies Quarterly 35, no. 2 (June 1991): 218
139. cf. Richard K. Betts, "Should Strategic Studies Survive?" World Politics 50, no. 1 (October 1997): 19.
140. John J. Mearsheimer, "Numbers, Strategy, and the European Balance," International Security 12, no. 4 (Spring 1988)
141. John J. Mearsheimer, "Assessing the Conventional Balance," International Security 13, no. 4 (Spring 1989)
142. Joshua M. Epstein, "Dynamic Analysis and the Conventional Balance in Europe," International Security 12, no. 4 (Spring 1988)
143. Joshua M. Epstein, "The 3:1 Rule, the Adaptive Dynamic Model, and the Future of Security Studies," International Security 13, no. 4 (Spring 1989).
144. Stephen Biddle, "The European Conventional Balance," Survival 30, no. 2 (March/April 1988)
145. Eliot A. Cohen, "Toward Better Net Assessment," International Security 13, no. 1 (Summer 1988).
146. Samuel P. Huntington, "Conventional Deterrence and Conventional Retaliation in Europe," International Security 8, no. 3 (Winter 1983/84).
147. Stephen Biddle et al., Defense at Low Force Levels (Alexandria, VA: Institute for Defense Analyses, 1991)
148. Joshua Epstein, Strategy and Force Planning (Washington, DC: Brookings, 1987).
149. Trevor N. Dupuy, "Combat Data and the 3:1 Rule," International Security 14, no. 1 (Summer 1989)
150. Mearsheimer, "Assessing the Conventional Balance."
151. Different views on these factors are expressed, inter alia, in Kimberly Kagan, The Surge: A Military History (New York: Encounter, 2008)
152. Bing West, The Strongest Tribe: War, Politics and the Endgame in Iraq (New York: Random House, 2008)
153. Stephen Biddle, Michael O'Hanlon, and Kenneth Pollack, "The Evolution of Iraq Strategy," in Restoring the Balance: A Middle East Strategy for the Next President (Washington, DC: Brookings, 2008)
154. Nils B. Weidmann and Idean Salehyan, "Violence and Ethnic Segregation: A Computational Model Applied to Baghdad" (paper presented at the Annual Meeting of the American Political Science Association, Washington, DC, September 2010).
155. In this way, the paper differs from recent analyses such as McGrath, Boots on the Ground and Kneece et al., Force Sizing for Stability Operations, which argue for adjusting the recommended troop density to thirteen and forty troops per one thousand inhabitants in the AO, respectively.
156. An important task for future research is to develop more specific and testable ways of defining these factors and studying how they interact with counterinsurgents' material capabilities. A straightforward way to do this would be to define a taxonomy of counterinsurgent strategies, code dummy variables for each type, and then test interaction terms combining strategy with troop density. (Interpreting interaction terms in nonlinear models is more complicated than in linear models, but the basic research design is the same.) Such a study would face many of the same problems this paper confronts: defining terms, finding and coding data, considering sensitivity to within-conflict variation, selection problems, and population subsets. In addition, it would be difficult to code strategy as an ordered variable in a way that is convincing. For these reasons, examining the interactive effect of manpower and strategy is a subject that is beyond the scope of this paper, but it is clearly worthy of future study.
157. HQ-ISAF, "COMISAF Initial Assessment."