Why do simple combat models favour the attacker
Why do simple combat models favour the attacker? Paul R. Syms Dstl Land Battlespace Systems Department OR 48, Bath, September 2006 Dstl/CP 21126
Presentation outline • Introduction – modelling complex problems manageably – trade between embedded and data-driven models • Recognizing the effect • Three examples: movement, surveillance and attrition • Inherent effect or user bias? • Concluding thoughts • Questions 2 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Introduction • Battle is one of the most complex processes to model – many different entity types and interactions … – often no obvious boundaries, and no ‘steady state’ … – e. g. battalion-level model may have 0. 8 M Lo. C*, 1. 5 M data items • Trade-off between model and data: – effects can be represented in either way … – complex embedded models require simple, primitive data – simple models require complex, aggregated data *lines of code 3 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Data-model trade-off • Simple model: U: all of ‘system’ Model • Complex model: Selected subset for study U: all of ‘system’ Model Complex data • Complex data example: Simple data • Simple data example: – kill rates by each system – error budgets, penetration 4 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Recognizing the effect • 1991 observation: 93% of bugs in a wargame ‘bug log’ favoured the attacker before they were fixed … – and the defender benefited when they were fixed • Difficulty matching Lanchester coefficients to a wargame – simpler model favoured attacker for same scenario • More examples … appeared to be a universal effect! 5 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Genesis of problem • Model designer’s prerogative to set boundaries – which processes are included and excluded • “Everything should be made as simple as possible, but not one bit simpler. ” – Albert Einstein – but how should we judge where threshold lies? • Practical constraints favour simple models – less time and cost to build and run, less risk … – thus tend to simplify – e. g. assuming systems always function • failure models seen as unnecessary overhead and complexity 6 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
The trap: simple model, simple data • Simple model: U: all of ‘system’ Model ? ? ? Simple data • In reality, ‘combat degradation’ – Clausewitzian ‘friction’ – and complex system interactions • System performance degrades – by 1 or 2 orders of magnitude … • If not modelled …wrong output … – misleading conclusions • What fills the gap? • Is combat degradation symmetric? – across attacker and defender? 7 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Three practical examples: moving, observing, and shooting
Example 1: movement • T-80 MBT* road speed is 70 kph, 48 kph cross-country … • Normally limited to 30 kph to prevent crew injury – and to have any chance of seeing the enemy • Need to halt to observe, plan and check routes – navigation errors are not unknown … • In reality, tank force advance rates are ≈2– 5 kph – but only possible to maintain this for two days in every three … • Mean speeds decline as more factors taken into account *main battle tank 9 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Example 2: observation • What is P(Acquire) for 10 attacking MBTs vs. 3 defending MBTs at (say) 2 km, if P(one-on-one) = 0. 05? • 10× 3 observers × targets, binomial expansion: 0. 79 • But only 3 observers looking in the ‘right’ arc: 0. 37 • Only 2 targets exposed due to ‘micro Lo. S’: 0. 26 • Account for experimental ‘group effect’: 0. 18 • . . . but how many observers actually alert? < 0. 1? • Probabilities decline as more factors taken into account 10 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Example 3: shooting • US TOW ATGW*: manufacturers claim P(hit): 0. 90 • Manufacturers claim P(kill) of MBT – SSKP*: 0. 80 • Poor ‘boresighting’ can reduce P(hit) on range: 0. 45 • Some targets move or break Lo. S before hit: 0. 31 • P(hit) for similar ATGWs in Lebanon, 2006: 0. 10 – of these, 20% resulted in severe or catastrophic damage … • Resultant ATGW SSKP in Lebanon, 2006: 0. 02 • Probabilities decline as more factors taken into account *anti-tank guided weapon; single shot kill probability 11 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Simple and complex data … • Simple: a ‘top speed’; complex: a practical advance rate – complex model adds halts etc. , simple one would not • Simple: 1 -on-1 P(acquire); complex: battlefield P(acquire) – complex model adds combat realism, simple one would not • Simple: SSKP = 0. 8; complex: battlefield SSKP = 0. 02 – complex model degrades SSKPs, simple one does not • Problems arise when simple models are fed simple data 12 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Concluding thoughts
Effects on attack and defence • Combat degradation is not equal for attack and defence: – attacker needs to move, observe and shoot – defenders ‘only’ observe and shoot … – and attacking targets are more easily seen and hit • If attacker allowed to advance too rapidly, defender has unrealistically few engagement opportunities • If attacker’s observation or firepower are over-estimated, defender can be ‘picked off’ at long range 14 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Summary and conclusions • Probably not an inherent mathematical effect … – but a feature of the way in which we use models – and critically, the data we provide • How should we avoid its potentially distorting effects? • Awareness is the first step – of Clausewitzian ‘friction’ and ‘combat degradation’ – of human performance – people are not robots – knowing the difference between simple and complex data … – using data appropriately, and validating model outputs 15 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
Point to ponder … • In the Cold War, UK’s military posture was defensive – and Mo. D OR techniques relied on complex models … – typically, large multi-entity, stochastic simulations • Post-Cold War, UK has undertaken more offensive ops – and Mo. D OR has shifted towards simpler models … – aggregated simulations, simpler performance models • Is this a pragmatic change to increase responsiveness? – or a subconscious need for more palatable answers? – conspiracy theories abound, despite what the experts say! 16 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
References Von CLAUSEWITZ K. (1832) ‘On War’ (Available in various translations and commentaries) ISBY D. C. (1988) ‘Weapons and tactics of the Soviet Army’ Jane’s Publishing Company Ltd. , London: 516 pp. MOORES B. (2006) ‘A Preliminary Military Assessment of the Lebanon Conflict’ http: //www. libertypost. org/cgi-bin/readart. cgi? Art. Num=155688, viewed 23 August 2006 PUGH P. G. (1992) ‘Operational degradation’ DOAE Memorandum 92103 ROWLAND D. (1987) ‘The Use of Historical Data in the Assessment of Combat Degradation’ J. Opl. Res. Soc. 38(2): 149 -162 17 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
QUESTIONS ? ? ? 18 12 September 2006 © Dstl 2006 UNCLASSIFIED Dstl is part of the Ministry of Defence
- Slides: 18