CORDA A BAeSEMA Company.

Transcription

1 CORDA A BAeSEMA Company. Casualty Estimating or Whose Data Do You Want To Bel i eve? M J Young, R C Goodman and C McLeod The views expressed in this document are those of the authors only, and should not be construed to represent oflcial views of eirher CORDA or the MOD. September 1997 T171 Apex Tower, 7 High Sueer, New Malden. Surrey KT3 4LH Telephone: Fax: OBAeSEMA Limited 1997 CORDA is P trading n m of BAeSEMA Limited ALL RIGFITS RESERVED The copyight in lhis document. whish coniains information of a pmpnerary nnturc. IS vested in BAeSEhiA Limited. The content of this docvmcnt my not bo used for pur~oses other than that for wbch U has becn supplied and may not be reproduced. either wholly or in pan in my way whatsoever. nor may it be used by, or ie content divulged to. any other perron whomsoever wilhoul the pior WeKen pcmission of BAeSEMA Limited.

2

3 T 17 1 Introduction I, Before the start of a campaign both politicians and the military will want an estimate of the casualties. The analyst can use many tools to make this estimate. These can be based on the results of computer models or historical analysis of past campaigns. However the many different systems for generating casualty rates make differing assumptions and produce different predictions for casualties that would be suffered in the campaign. 2. To alleviate this problem, DACOS 58 (OA) commissioned CORDA to produce a prototype casualty estimation tool. This is not yet another new model making another different set of assumptions, but rather a quick way of presenting and comparing the results of as many existing casualty estimating methods as possible. The user inputs a particular scenariq and decides which of the possible models to use, the casualty estimation tool will then genqate the casualties predicted for that scenario and method. 3. The casualty estimation tool is tri-service allowing differing methods to be used to predict land, sea and air casualties, and can present the results in differing formats for the different potential audiences. 4. The tool uses two different models to generate naval casualties, four for land casualties and one for air casualtiesc- In this paper the casualty numbers from a number of historical engagements have been compared with: b The casualties predicted by the calculator. b The casualties predicted using a method for calculating land casualties specified by Trevor Dupuy. No method is yet capable of predicting the casualties with less than a 5% error in all circumstances. The Need 5. Before the start of a military campaign, many people try to predict its result, both in terms of campaign outcome and in terms of the casualties that will be taken. They are almost universally wrong. 6. Politicians (ought to) want to know the magnitude of the human sacrifice that needs to be made to gain the ends sought by the military operation. Military staff need to know the casualty rates for more mundane reasons. They need to know how many wounded they will need to treat, and how many battle casualty replacements (and of what kind) they need to have available to maintain unit strengths. 7. This begs the question of What is a casualty?. Each of the three groups above needs to use a different definition of casualty. For example, medical services need to know the number of casualties not requiring hospitalisation due to minor injuries and disease, as they place strain on the medical system. However, these casualties will have very little effect politically and as they are returned to duty quickly do not require battle casualty replacements.

4

5 T17 1 CORDA 8. For the purposes of our calculator, each casualty is assumed to fall into one of nine tvdes. as sdecified in the table below. d Died before receiving medical treatment Died of wounds Hospitalised wounded battle casualties (in-patients) Interest shown by customer. Politicians Medical Battle casualty Services replacements High No Yes, by unit role. High Yes Yes, by unit role. Medium Yes Yes, by unit role. Wounded battle casualties(out patients) DNBI died before receiving medical attention DNBI died of injury / died of disease DNBI hospitalised DNBI out-patients I Low Medium Medium Low VeryLow Yes No Yes Yes Yes No Yes, by unit role. Yes, by unit role. Yes, by unit role. No Prisoners of War/ Missing High No Yes, by unit role. 9. Unfortunately data on casualties is rarely available broken down in such detail either historically or as model output. Therefore it has been necessary to assume that the ratios between the different casualty types remain constant by branch of service and these ratios are assumed where data is unavailable. This can be dangerous as the ratios may easily. vary for a number of reasons. An army that is losing badly, for example, will probably have a higher proportion of casualties as prisoners than an army that is winning. This particular effect was taken into account by the calculator, but others were not. IO. As an important aside we also have to remember that our data sources may use different definitions of casualty. One source may include DNBI, another source may only include DNBI casualties that die, yet another may not say at all. There is also no universally accepted definition of all the categories above. For example, during World War I1 the British Army counted frostbite casualties as DNBI, but the US counted them as Wounded Battle Casualties (thus entitling frostbite casualties to a Purple Heart). - Our Method Our method was to construct an ORBAT and then calculate a loss rate and apply it to the ORBAT. The user can specify what method is used to calculate losses for each branch of service. The different methods used and their results are described in the sections below: Naval Casualties: Method 12. The naval casualty rate is critically dependent on the number of ships sunk and damaged. The user may either input the number of damaged or sunk ships directly to the model or automatically generate it from a method given by the David Taylor Research Center. This method uses the size of the force and the number of days of combat at different intensities to calculate the number of ships sunk or damaged. 13. Once the total number of damaged and sunk ships is known then the casualty rate on board ships of the particular type is multiplied by the size of the crew to give the total casualties. The user can select a casualty rate from one of two sources: Those of the Naval Historical Research Center in San Diego, derived from World War I1 US Pacific data.

6

7 T171 CORDA Those of the David Taylor Research Center. 14. These data sources do not give figures for casualties to submarines. We therefore took a sample of data for crew losses for 21 of the 52 US submarine sinkings in World War II. From this an average loss per sinlung was calculated which was applied to submarines no matter which casualty estimation method was used for surface ships. 15. Ships can sink in many different ways due to differing levels of damage, and this obviously affects theaumber of casualties on board the ships. A ship that explodes when alone will have very few survivors. One that sinks slowly close to other friendly ships or a friendly port will have very many. This effect is well illustrated by the submarine data. Of the 21 submarines lost in our data sample, 17 were sunk with no survivors, 3 with some survivors and one where all the crew survived. (The ship was scuttled after being damaged. All the crew survived the sinking.) 16. The model does not consider casualties to personnel in shore installations, or captured or missing naval personnel. Naval Casualties: Results 17. There are two aspects to verifying the naval casualty model. Firstly there is the ship damage generator that calculates the numbers of ships damaged'or missing from the intensities of the action. The second is the calculation of the numbers of casualties from the number of sunk and damaged ships. We tested both these sub-models. In general we found that: The ship damage generator was not particularly accurate The mechanisms for predicting the numbers of casualties from the numbers of sunk and damaged ships worked reasonably well, given the inherent variability of the statistics. 18. We studied two historical campaigns to verify the calculator. The first was the action around Crete in When the German airborne forces attempted to capture Crete, they also attempted some sea-borne landings. These sea-borne landings were successfully opposed by the Royal Navy, despite continual air attacks by the Luftwaffe. No Germans managed to land from a boat onto Crete. However the capture of airfields by the German paratroopers enabled reinforcements to be airlifted in, so the Germans eventually captured the island. 19. The second campaign was the Indo-Pakistan War of In this war Pakistan lost control of what is now Bangladesh to a combination of an internal revolt by the Bengalis and an Indian invasion. There were a few naval engagements. In the east Indian Forces defeated the Pakistani navy, but in the west the battle was less decisive, with India losing a frigate and Pakistan a destroyer. 2. Figure 1 compares the historical ship losses with the results generated using the ship loss generation methods available. There is only a weak correlation. The ship loss generation method should be improved Figures 2 to 5 illustrate the predicted number of battle casualties given the number of ships sunk and damaged. For these figures the historical number of ships sunk or damaged has been used to calculate the number of personnel casualties. The predictive power of this part of the calculator seems quite good. Of the two methods of calculating casualties the method using data from the NHRC seems better. Army Casualties 22. For the army casualties a daily loss rate is calculated. This is a function of the unit's activity. By summing these losses over the number of days duration of the expected battle, we produce a total casualty rate for each unit involved. This rate is then multiplied by the number of men in the unit to get the total number of casualties. 23. The user may select which set of loss rates will be used by the calculator to derive the casualties. These loss rates may be radically different from one another. At the moment the model holds data on four sets of m y casualty methods comprising:

8

9 TI71 CORDA The War Office Method of the 194s and 195s. Ths method allows the user to select one from 12 possible loss rates for combat generally based on World War II experience. Each day of a conflict for each division is described as being Intense, Normal or Quiet and the historical casualty rate is then applied for each arm of service. The US Casualty Assessment method (obtained from notes given to us by Maj. Chris Best). This was used by the US Army medical services to calculate their medical support needs. This gives casualties for Normal or Intense combat for five different types of engagement. It has a modification for the force ratio. A method assumed from the result of Exercise May Fly (obtained from notes given to us by Maj. Chris Best). In this method, operations are described as Severe, Moderate, Light or Minimal, and definitions given for what constitutes each type of operation (e.g. Covering force operations are assumed to be light ). A method given in the book How to make war by JF Dunnigan. This specifies casualty rates for Attacking or Defending in five different operation types. There is also a modification for the force ratio. Casualty rates are appiied for each arm of service. Army Casualties. 24. Figure 6 shows the losses to an armoured division involved in 3 days intense combat and LO days light combat, according to the different methods. The division was assumed to win and to outnumber the enemy 4 to 1 in combat power. This shows the casualties predicted by the different models broken down by casualty type. This figure shows that there is a huge difference between the results of these different models. 25. To validate the results of these casualty predictive tools, a series of historical land battles were entered into the calculator and their results compared with the outputs from the model. In general there was a huge variation in the results obtained from the different models in the calculator. 26. As an example historical comparison Figure 7 shows an attempt to predict the ground casualties suffered by the allies in the Gulf Conflict in The other comparisons are out by varying degrees, often grossly so. 27. According to the strict definitions used by the model the Gulf Conflict should consist of four days of Intense combat. However, it could be argued that the conflict, although of a high tempo was not really Intense. Figure 7 therefore also shows the casualty rate if the Gulf Conflict is considered to be 4 days of Normal combat. This shows that the War Office method (using the AORG recommended rates) now comes quite close to predicting the correct numbers of casualties, but still predicts about double the casualties actually suffered. 28. Besides the methods implemented in the calculator, the method suggested by Trevor Dupuy of the Historical Evaluation and Research Organisation (HERO) in the book Attrition: Forecasting Battle Casualties and Equipment Losses in Modem War by was also investigated. This method has not been implemented in the model. This book implies good predictions of casualties can be made before the battle starts. It was published just before the Gulf War. Even though HERO predicted considerably lower casualty rates than most other sources, the rates were still much higher than reality. The War Office method comes the closest to the actual numbers, with the other methods being wildly out. In defence of his method in this case, there is a factor in the calculation that works out a divisor for the casualties depending on the amount by which one side outnumbers the other in terms of fighting power. This varies between 1 to I against and 4 to 1 in favour. How to calculate fighting power is also given in his page 4

10

11 T17L CORDA 29. The models were not used to predict Iraqi casualties, mainly because it is not definitely known how many Iraqi casualties there were. 3. Figures 8, 9 and 1 show attempts to predict the casualty rates in three other campaigns, the Indian invasion of Hyderabad in 1948 (Figure 8)* the Japanese offensive at Kohima in India in 1944 (Figure 9), and Operation Battleaxe in Libya in 1941 (Figure 1) 3 1. The main reason for the difference between these results divergence is not only the base casualty rates assumed but also: Some models apply the same casualty rates to all arms of service, others have,different rates for each ann of service. The differing values of some multiplicative factors given in the model; for example some models take into account how much a side outnumbers the enemy, others do not It is even possible to find historical battles where the casualty rate exceeds those predicted by even the most bloodthirsty of our predictors. Figure 11 shows the results of the Canadian raid on Dieppe in In this raid 6% of the men who landed became casualties at the end of one day s fighting. This rate is not reproduced by any of our models. Air Force Casualties 33. There are two sources of casualties to air force personnel. Firstly, there is the casualties to air crew in lost or damaged aircraft. Secondly, there are the casualties on air bases caused by enemy air attack. 34. The numbers of air crew lost depends on the number of aircraft lost. The user can either specify the number of aircraft lost directly, or have the computer calculate this for him from sortie rates and loss rates. The model uses loss rates per sortie taken from the Gulf and from Normandy in Once the number of lost aircraft has been calculated, the numbers of casualties resulting from this is obtained by multiplying it by the number of crew and dividing by the probability of a crew member of a lost aircraft becoming a casualty. Default probabilities use data from the Gulf. 36. Attacks on friendly air bases are dealt with by specifying the number of enemy aircraft involved in the attack and their type of attack. Attacks on air bases can kill air crew on the ground as well as ground crew. 37. The air force model is not as well developed as the army and navy models. No historical validation has yet taken place. We hope this will be a prionty in the next phase of the development of the calculator. Split of casualties into categories 38. After the total number of casualties has been derived, the calculator then splits them into the nine different casualty type categories using a series of assumptions and rules about the ratios between different casualty types (killed, wounded in-patient, wounded out-patient, etc). These are different for each branch of service, but are applied independently of the method used to calculate the casualty numbers. t book. Using his method with data now available about actual force strengths. it appears that we outnumbered the Iraqis about 16 to 1 in fighting power according to his definition. It is not surprising therefore that using a divisor for 4 to 1 + would seriously overestimate our own casualties at 16 to I. A similar thing happens with the numbers of troops. The larger the force the lower the casualty rate. Col Dupuy s method has 1.~ as the largest army size, whereas there were over 3, allies, so this could again argue toor lower casualties.

12

13 T171 CORDA Discussion of res u I ts 39. There seems to be no method that is capable of perfectly predicting the loss rates on both sides in a series of land or naval engagements in all circumstances. (See Figures 7-1 1). The values given in the War Office method often give the best land casualty rates, 4. The methods often do not agree with one another even on which side will receive the largest number of casualties. A major reason for this is that the War Office method and the Operation Mayfly method do not take into account the size of the opposition whereas the others do. The US casualty assessment method and to a lesser extent Col Dupuy s method make a very large correction for this, often multiplying one side s casualties by a significant factor Even if the casualty predictor model were to give perfect answers, it does not follow that we will be able to verify this because the casualties reported in the history books are often themselves inaccurate. Often you only get one side s (usually the victor s) history sf what happened. On the rare occasions when you get both side s accounts then there can be quite a difference in opinion as the table below shows. Note that both sides claim that their losses were between a half anda sixth of the enemy s. Personnel Casualties (killed) Personnel Casualties (k+w+m) Tank Casualties Aircraft Casualties Indian Losses Pakistani Losses Indian Version Pakistani Indian Version Pakistani Version I Version 2,2 12-2,92 9,5 5, ,3 11,348-22,2 Notknown 1 3, ,75 I i It is easy to think of factors that could affect casualty rates that have not been included in any of the land models. The points below comprise a list of factors that are not taken into account in any model, but we believe could easily at least double battle casualty rates. The relative amount of artillery ammunition available. Artillery accounts for the majority of casualties on the modem battlefield and if one side had a shortage of artillery it would probably inflict many fewer casualties than it receives. This may explain the disparity in the Gulf (where the Iraqis could not fire their guns) and at Kohima (where the Japanese had few guns and a great shortage of ammunition). Air superiority. If one side has air superiority it may suffer fewer casualties. Ths may not be only due to the casualties caused directly be air attack, but also due to the indirect effects of air superiority (for example better intelligence on enemy locations due to air reconnaissance). Winning or losing. None of the models take into account that a losing army should take more casualties than a winning one. We have added an optional switch to the War Office method so this can be done, but this is not in the original method. 44. In some cases the outcome of a campaign, and hence the casualties may be genuinely unpredictable before the start of the campaign. History is full of incredibly minor events that have influenced the outcome of a battle. It is literally impossible to model every event that could influence battle outcome when it could go either way. A small event may make one side start to lose the battle. Once a side has started to lose it becomes harder and harder to turn

14

15 T171 CORDA things round. This may mean that it may be easier to predict casualties in wars where there is little doubt as to the outcome, rather than ones where the outcome is closer. 45. The David Taylor method of assessing the number of ships damaged or sunk gives poor results. It may be better to use an empirical method based on historical analysis. Unfortunately if the ship loss rate is low, then any method will appear less reliable because we are dealing with integer events and small numbers. Conclusions 46. We conclude the following: a A tri-service calculator has been built that presents a database of methods and supporting data. This has been tested for war; though the calculator is designed to be used for war and operations other than war.. a From the simple analysis presented here, it is clear that no reliable method has been found to date that will consistently predict the losses over all the spectrum of operations. Of those we have looked at the War Office method seems the best land model, and that using the NHRC data seems to be the best naval model a War is chaotic and so we genuinely may never be able to accurately predict the outcome and hence the casualties before the start. a People provide wrong information about casualty figures, often in good faith. Even if we know one side s casualty figures we rarely know the others. This makes historical validation difficult. a Comparisons using different methods are not consistent and the use of different assumptions means that one method may predict much higher casualties than another in some scenarios and much lower in others. a All battles are unique and can be claimed as exceptions to almost any rule (especially the Gulf Conflict and indeed the next conflict). pnge 7

16

17 Y-

18

19 8 W 8 5: 8 8 R,

20

21

22

23 z: N 8 N z:

24

25 i I ln N 8 N

26

27 I ndna 'f,t -; i 2 1 i I,,' L b i a 8 m h (.

28

29 .- E c i,$, 'i! U E (P i *.. b i i i P I I 1: I 8 m E T 5: 8 N

30

31 r'. Y i ka 33 E 8 z E N r 8, m ( 8 N

32

33 I- O! j I.- S CI m I j ZI U - m" f I-.. Q, i I i I i I

34

35 ~ 7! I! I!! il.-, C 5 1 i I j i I i i!!! 1!!,,!,!! 8 gg $a Ij, I! ~ i!! i, 8 U-J Z n 8 Lo 8 v * m z N, L n 8 c 5: r

36

37 .- E U ) - > >.- E I!. p i, t I U I!: c I i!i m o Ln m m F 5: N : 8,

38

39

40