War in the Pacific as Planned--Allied

[herwin]

I saw a lot of flak about my ideas about ground combat, combined with snide comments about who would develop the software. Actually, I did design such a model for the USMC and it was implemented and used, although not as a simulation tool--it was used as a scenario generator for testing. There are a few problems with the ground combat model:
1. Casualties are badly off. This can be fixed by tweaking the engagement model to incorporate target acquisition. It would encourage right-sizing your combat teams instead of the use of mass.
2. Tempo is off. This can be fixed by eliminating the AV comparison at the end of the battle. Instead, keep rough track of who owns the hex in percentage terms, and place bases and MLR at the halfway point in the hex. A successful attack (your casualty percentage is less than the defender's) adds to your hex ownership. Counterattacks then move things in the reverse direction.
3. Resistance is off. This can be fixed by having the amount of change in hex ownership reflect the relative casualty percentages of the last day's battle. The fortifications in the hex need to be concentrated in the MLR, so that advance is slowed as you approach the position and speeds up as you advance past it. Armour plays a role by increasing the advance rate. Engineers reduce the fortification level.

I'm trying in this to make the ground combat model more abstract.

Please note, I'm not just concerned about the ground game engine because it hurts the Allies. I'm envisioning the likely outcome of my later offensive operations, too. I've been doing things in China and Mindanao (in emulation of my Japanese opponent's tactics) that are effective but quite ahistorical.

[JWE]

ORIGINAL: herwin
I saw a lot of flak about my ideas about ground combat, combined with snide comments about who would develop the software. Actually, I did design such a model for the USMC and it was implemented and used, although not as a simulation tool--it was used as a scenario generator for testing.

I really don't think you should be telling people that you designed simulator software for the USMC; at least not in any place where I can be found.

Thank you. J
[ps] I am a principal in a very highly regarded military R&D firm and know many of the Industry and Service folks personally. You really don't want to get my dander up.

[herwin]

February 5/6 1942

Lost a DD in ASW ops near Johnston Island.

My opponent has noticed the small merchies moving between Palembang and Singapore. No losses yet.

Betties showed up in the Makassar Strait. I guess I need to pull back my patrol line.

Bombardment attack at Singapore. 29 troops lost.

Japanese capture Wuchow, but the Chinese move to control the supply routes there.

[herwin]

February 6/7 1942

Two attacks away from Singapore falling:

Japanese Deliberate attack

Attacking force 92231 troops, 752 guns, 495 vehicles, Assault Value = 2906

Defending force 46731 troops, 568 guns, 268 vehicles, Assault Value = 688

Japanese engineers reduce fortifications to 1

Japanese adjusted assault: 3026

Allied adjusted defense: 1319

Japanese assault odds: 2 to 1 (fort level 1)

Japanese Assault reduces fortifications to 1

Combat modifiers
Defender: terrain(+), leaders(+)
Attacker:

Japanese ground losses:
     4413 casualties reported
        Squads: 9 destroyed, 166 disabled
        Non Combat: 8 destroyed, 169 disabled
        Engineers: 15 destroyed, 63 disabled
     Vehicles lost 57 (8 destroyed, 49 disabled)


Allied ground losses:
     4849 casualties reported
        Squads: 132 destroyed, 32 disabled
        Non Combat: 103 destroyed, 213 disabled
        Engineers: 4 destroyed, 22 disabled
     Vehicles lost 65 (6 destroyed, 59 disabled)


Assaulting units:
   5th Engineer Regiment
   14th Tank Regiment
   114th Infantry Regiment
   23rd Ind Engineer Regiment
   11th Infantry Regiment
   56th Infantry Regiment
   21st Division
   4th Ind Engineer Regiment
   16th Infantry Regiment
   38th Division
   1st Tank Regiment
   24th Infantry Regiment
   56th Recon Regiment
   113th Infantry Regiment
   56th Engineer Regiment
   148th Infantry Regiment
   6th Tank Regiment
   Imperial Guards Division
   42nd Infantry Regiment
   55th Infantry Regiment
   5th Recon Regiment
   41st Infantry Regiment
   21st Infantry Regiment
   91st Naval Guard Unit
   25th Army
   5th Field Artillery Regiment
   32nd Field AA Battalion
   7th Field Construction Battalion
   3rd Medium Field Artillery Regiment
   3rd Ind. Mountain Gun Regiment
   20th AA Regiment
   56th Field Artillery Regiment
   18th Medium Field Artillery Regiment
   34th Field AA Battalion
   5th Field Construction Battalion

Defending units:
   SSVF Brigade
   3rd Cavalry Regiment
   2nd Loyal Battalion
   1st Malay Battalion
   22nd Australian Brigade
   2/17 Dogra Battalion
   Singapore Fortress
   1st Manchester Battalion
   2nd Malay Battalion
   27th Australian Brigade
   2nd Gordons Battalion
   9th Indian Division
   AHQ Far East
   2nd HK&S Heavy AA Regiment
   III Indian Corps
   1st HK&S Heavy AA Regiment
   224 Group RAF
   3rd Heavy AA Regiment
   Malaya Army
   223 Group RAF
   112th RAF Adv Base Force
   1st Indian Heavy AA Regiment
   2nd ISF Base Force
   Malayan Air Wing
   24th NZ Pioneer Coy
   22nd Indian Mountain Gun Regiment
   113th RAF Adv Base Force
   110th RAF Adv Base Force
   111th RAF Adv Base Force
   Singapore Base Force
   109th RN Base Force

Betties have moved up to Balikpapan. Time to evacuate Soerabaya.

Note added: I told my opponent he needed eight divisions to take Singapore. I think he believed me.

[herwin]

I guess I need to lay my cards on the table.

In 1972, I took a leave of absence from a PhD in algebraic topology and joined TRW Systems as a Member of Technical Staff (MTS) in a systems engineering organisation. Basically, I was a trainee systems engineer. At that point, I had some military training but no military experience. Over the next six years, I learned and practiced military operational analysis, modelling and simulation, and large scale systems engineering, mostly in ballistic missile defense and satellite systems, but with a significant sideline in combat modelling and intel analysis. My major interest area has always been complexity and non-linear systems, and since what happens on the battlefield is one of the most complex human activities, I was particularly interested in the effects of technology and human behavior on combat.

In 1978, Norden Systems was bidding on the MIFASS Program for the USMC, and asked TRW to be a subcontractor for the operating system, database management system, and the test and training simulation. The Government-required programming language (CMS-2) was obscure and the required computing environment slow and with very limited memory, so the program was facing significant technical risks. MIFASS was the first phase of deploying MTACCS, with a later phase being the TCO command and control system. My interests and background made me a natural candidate to write the simulation proposal, so I joined the proposal team. Norden won the bid, and TRW received a subcontract for its parts of the proposal. The development organisation within TRW offered me a position as the development lead for the simulation team, but I declined as it required a transfer between departments.

Norden was aware that their systems engineering expertise was weak and had head-hunters out. I was contacted and interviewed. I expressed my particular interest in working on TCO, and eventually I accepted a position at Norden as a systems engineer on the MIFASS and TCO programs. About six months into the MIFASS project, Norden realised it was in serious trouble, and reorganised the program. The chief systems engineer for TCO moved to MIFASS, and I found myself replacing him. I was in that position for six years, working for Bob Ritts. During that period, I was also the MIFASS system engineer for the operating system, database management system, and the test and training simulation and the assistant systems engineer for communications and air support. I also ran a couple of IR&Ds and was the system architect for other proposals, including AFATDS and SORAK.

By 1986, it was clear that the hardware problems that afflicted MIFASS were unsolvable and it would fail its operational test, so that TCO would not be happening in the form originally planned. I returned to TRW and worked on a series of programs as an systems engineer, systems modeller, and eventually as a security architect, mostly for submarine and US Navy intel systems. I stayed in touch with my USMC contacts, and that led to TRW work on a series of USMC programs. With the end of the Cold War, I branched into civilian programs, and I was, for example, the security architect for the Treasury Communications System. About 1992, I decided to return to school (GMU) part-time to do a PhD in computational neuroscience—yet another interesting complex non-linear system. I became an adjunct (part-time) lecturer (and later assistant professor) of computer science there—which gave me a faculty parking place [:)]. I continued at TRW, working in communications systems engineering, military systems analysis and modelling, and security.

My dissertation research was a computational sensorimotor model of bat biosonar—requiring me to model both biosonar and bat flight. I had the appropriate background for the former, but had to learn aerodynamics for the latter. I defended my PhD at the end of 1999. Sunderland offered me a senior lectureship of computing—equivalent to an assistant professorship in America—and I started there in January 2001. My current research is in computational neuroscience of the auditory system, and I also maintain an interest in how the mammalian brain models the world and generates plans.

WitP-AE is a lovely game--in turns it gives me joy and aggravation. I would love to see it perfected, but I'm old enough to realise there's little hope of that happening. It embodies two problems I hope someday to help solve--what are the neural mechanisms underlying morale, and how do people plan? I believe the former is needed for a realistic model of WWII ground combat, and I believe solving the latter is the only viable path to true AI.

If you see me grumpy, please remember I'm a perfectionist. In peace,

[jrcar]

Herwin,

while no ATO expert the ones I've been involved with usually tell you:

the target (or expected type of target)
The effect
the number of aircraft in each package
the weapons (weaponeering happens at a very high level)
key routes (may even define Hi-LO-HI or similar.

Actually when planning these the team usually goes into a lot of detail, even if not fully in the ATO (which is the summery). Ingress, egress, airdefence, deconfliction are all considered in the ATO planning process.

So there is plenty of "How" in an ATO. Actually as an Army guy I was very surprised at how prescriptive they are... but as these drive the logistics cycle, the aircrew cycle and maintenance cycle it does make sense.

In that regard AE is fairly similar (if not as prescriptive as an ATO) you give a unit the type of mission/missions, (by setting percentages of REST, TRAINING, CAP etc you can also control to a degree the sortie size). You don't get to direct the weaponeering, and you don't get to set the routes (which you can in an ATO).

On another post the aims of AE were to fix air and fix naval issues of WITP. Land combat improvement was out of scope of AE originally. But as we had a number of land guys, and we had a degree of "programmer" time we did manage to do some work on the land model.

I agree with you that the land model is the worst of the three primary combat models in the game. It is gernerally "good enough" in 60-80% of the combats. I think its biggest problem is that there are lots of fairly binary rolls, you pass a morale check or not, you pass a leadership test or not. So two good die rolls give you one end of the spectrum, two bad ones the other with not a lot of in between (but it does happen). Combat isn't as gradual as one would expect.

OTOH the Pacific war is full of combats that were like that. Battles that were expected to take ages, over in a day or two. Ones that were expected to be walkovers, bogged down. Leadership, planning, training and morale (as usual) were the main factors in these, not the weaponeering.

And I wouldn't trust staff planning tables, they are just a start point to start planning from.

Cheers

Rob

ORIGINAL: herwin

I mentioned above I'd like to have an air tasking order capability. That comment was posted elsewhere and drew a lot of flak that made me suspect some of the people commenting didn't know what an ATO is. So go here for a definition. There are two underlying ideas that need emphasis:
1. The ATO is what not how. People complain about the amount of micromanagement in the game. That's because the player needs to define both what and how. I'd like to get the player out of the business worrying about how.
2. To implement the ATO, commanders think in terms of sorties, not aircraft. That is, an air task requires a certain number of sorties armed and equipped in a certain way. Squadrons and groups are sortie generators. I'd like to get the player out of the business of micromanaging sortie generation.

Combining those two ideas would produce a simpler and more abstract user interface, one better matched to how commanders operated in WWII.

Note, the game already does this fairly well in naval ops.

[herwin]

ORIGINAL: jrcar

Herwin,

while no ATO expert the ones I've been involved with usually tell you:

the target (or expected type of target)
The effect
the number of aircraft in each package
the weapons (weaponeering happens at a very high level)
key routes (may even define Hi-LO-HI or similar.

Actually when planning these the team usually goes into a lot of detail, even if not fully in the ATO (which is the summery). Ingress, egress, airdefence, deconfliction are all considered in the ATO planning process.

So there is plenty of "How" in an ATO. Actually as an Army guy I was very surprised at how prescriptive they are... but as these drive the logistics cycle, the aircrew cycle and maintenance cycle it does make sense.

In that regard AE is fairly similar (if not as prescriptive as an ATO) you give a unit the type of mission/missions, (by setting percentages of REST, TRAINING, CAP etc you can also control to a degree the sortie size). You don't get to direct the weaponeering, and you don't get to set the routes (which you can in an ATO).

On another post the aims of AE were to fix air and fix naval issues of WITP. Land combat improvement was out of scope of AE originally. But as we had a number of land guys, and we had a degree of "programmer" time we did manage to do some work on the land model.

I agree with you that the land model is the worst of the three primary combat models in the game. It is gernerally "good enough" in 60-80% of the combats. I think its biggest problem is that there are lots of fairly binary rolls, you pass a morale check or not, you pass a leadership test or not. So two good die rolls give you one end of the spectrum, two bad ones the other with not a lot of in between (but it does happen). Combat isn't as gradual as one would expect.

OTOH the Pacific war is full of combats that were like that. Battles that were expected to take ages, over in a day or two. Ones that were expected to be walkovers, bogged down. Leadership, planning, training and morale (as usual) were the main factors in these, not the weaponeering.

And I wouldn't trust staff planning tables, they are just a start point to start planning from.

Cheers

Rob

ORIGINAL: herwin

I mentioned above I'd like to have an air tasking order capability. That comment was posted elsewhere and drew a lot of flak that made me suspect some of the people commenting didn't know what an ATO is. So go here for a definition. There are two underlying ideas that need emphasis:
1. The ATO is what not how. People complain about the amount of micromanagement in the game. That's because the player needs to define both what and how. I'd like to get the player out of the business worrying about how.
2. To implement the ATO, commanders think in terms of sorties, not aircraft. That is, an air task requires a certain number of sorties armed and equipped in a certain way. Squadrons and groups are sortie generators. I'd like to get the player out of the business of micromanaging sortie generation.

Combining those two ideas would produce a simpler and more abstract user interface, one better matched to how commanders operated in WWII.

Note, the game already does this fairly well in naval ops.

OK, thanks.

ATO: my goal would be to get the player out of defining 'how'. I would envision a game version as containing something like target locations (handling task forces somehow intelligently), target type and grossly specified target effect, grossly specified strike size (sorties), and commander's priority. Also, the commander would be defining just the high priority targets; routine targets would be selected by the game engine as it does now. The game engine would then treat the available air units as sortie generators contributing to the solution.

Ground combat: I accept the problems here; I just get frustrated with the advantage mass has over over right-sizing. Here's a figure from Helmbold, R. L. (1971) Decision in Battle, Rand Technical Report R-772-PR, page 43, that shows the expected pattern.

[herwin]

Feb 7/8 1942

No significant activity.

[herwin]

I received a coast watcher report of a fast CVTF at Rabaul. I have two CVTFs 18 hexes away, and I have three search bases closer (Port Moresby with Hudsons, Shortlands and Tulagi with PBYs) to give me coverage. No submarine detections in the Coral Sea. No sign of Betties at Rabaul, despite bait being thrown in the water. This could get interesting.

[herwin]

February 8/9 1942

Singapore falls--I need to start pulling out of the DEI. When Japanese tank units appear in Burma, I will also need to get my speed bumps deployed. At least this time, the KB will not be providing CAS.

Japanese Deliberate attack
 
Attacking force 89911 troops, 752 guns, 492 vehicles, Assault Value = 2791
 
Defending force 43880 troops, 571 guns, 262 vehicles, Assault Value = 630
 
Japanese engineers reduce fortifications to 0
 
Japanese adjusted assault: 3062
 
Allied adjusted defense: 1353
 
Japanese assault odds: 2 to 1 (fort level 0)
 
Japanese forces CAPTURE Singapore !!!
 
Combat modifiers
Defender: terrain(+), leaders(+), experience(-)
Attacker:
 
Japanese ground losses:
      6187 casualties reported
         Squads: 46 destroyed, 108 disabled
         Non Combat: 45 destroyed, 314 disabled
         Engineers: 26 destroyed, 26 disabled
      Vehicles lost 58 (6 destroyed, 52 disabled)
 
 
Allied ground losses:
      49128 casualties reported
         Squads: 884 destroyed, 0 disabled
         Non Combat: 3108 destroyed, 0 disabled
         Engineers: 204 destroyed, 0 disabled
      Guns lost 805 (805 destroyed, 0 disabled)
      Vehicles lost 255 (255 destroyed, 0 disabled)
      Units destroyed 31
 
 
Assaulting units:
    24th Infantry Regiment
    23rd Ind Engineer Regiment
    1st Tank Regiment
    5th Engineer Regiment
    5th Recon Regiment
    113th Infantry Regiment
    56th Infantry Regiment
    148th Infantry Regiment
    16th Infantry Regiment
    38th Division
    55th Infantry Regiment
    Imperial Guards Division
    21st Infantry Regiment
    41st Infantry Regiment
    56th Engineer Regiment
    6th Tank Regiment
    4th Ind Engineer Regiment
    14th Tank Regiment
    114th Infantry Regiment
    56th Recon Regiment
    11th Infantry Regiment
    21st Division
    42nd Infantry Regiment
    91st Naval Guard Unit
    7th Field Construction Battalion
    20th AA Regiment
    3rd Ind. Mountain Gun Regiment
    18th Medium Field Artillery Regiment
    5th Field Artillery Regiment
    32nd Field AA Battalion
    34th Field AA Battalion
    25th Army
    3rd Medium Field Artillery Regiment
    56th Field Artillery Regiment
    5th Field Construction Battalion
 
Defending units:
    2nd Loyal Battalion
    27th Australian Brigade
    22nd Australian Brigade
    2nd Gordons Battalion
    Singapore Fortress
    3rd Cavalry Regiment
    1st Malay Battalion
    SSVF Brigade
    2/17 Dogra Battalion
    1st Manchester Battalion
    2nd Malay Battalion
    9th Indian Division
    111th RAF Adv Base Force
    2nd HK&S Heavy AA Regiment
    24th NZ Pioneer Coy
    110th RAF Adv Base Force
    2nd ISF Base Force
    Malayan Air Wing
    113th RAF Adv Base Force
    3rd Heavy AA Regiment
    1st HK&S Heavy AA Regiment
    1st Indian Heavy AA Regiment
    223 Group RAF
    Singapore Base Force
    Malaya Army
    22nd Indian Mountain Gun Regiment
    III Indian Corps
    AHQ Far East
    112th RAF Adv Base Force
    224 Group RAF
    109th RN Base Force

[herwin]

Just some random thoughts that might trigger someone else's intuition: I've been trying to come up with an approach to understanding decision in combat so I can incorporate it in a statistical model. One approach is a competing risk model--that is, there are two different risks and either may result in a decision on the battlefield, one corresponding to the attacker losing and the other to the defender losing. I then need something to serve as a time dimension--I think defender's percentage casualties, y, are a possibility as they can be taken to represent the maximum penetration of the attack. The independent variable, x, is then the attacker's percentage casualties. There's a symmetry requirement--if you swap x and y, you swap the two risks. OK, what kind of statistical model does that lead to?

OK, treat the primary parameter as the 'duration' of the battle (defender's percentage casualties), while the secondary parameter is the log of the (ratio of the attacker's casualty percentage to the defender's casualty percentage), since that's probably independent of the duration. The log transformation deals with the symmetry of the casualty percentage ratio. The winner at battle termination is probably predicted by the secondary parameter, but that would have to be tested. It may be dependent on both.

To use this in a game, run the firepower duel (with target acquisition in addition to target engagement) until the survival model predicts that the battle is over (or the day's engagement is finished or one side has 100% casualties). National morale perhaps comes in here, since the Japanese were willing to take 100% casualties. At that point use the secondary parameter to determine the winner of the battle. Take into account the battle possibly continuing ('surviving') until the next day. The main thing for the model to predict is when the battle terminates. In terms of WitP-AE, there would be three changes to the game engine:
[ol][*]Incorporate target acquisition in the firepower duel[*]Terminate the firepower duel earlier to reflect a reduced tempo of battle[*]Determine battle termination based on relative percentage casualties during the firepower duel, not an AV comparison. Take into account the high chance the battle will continue into the next day.
[/ol]The mstate package in R sounds like it gives me what I need.

It's easier than that. Since there are no survivals past 100% defender's casualties, I can simply do a glm for a gamma distribution. If that shows duration is insensitive to who's winning, I can use the above model. I suspect, however, that duration is sensitive to whether it's a serious mismatch. A fit of duration to the absolute value of the mismatch parameter can be tested to see if it influences duration. If it does, I'll revise the logic a bit to end the battle earlier if one side is being slaughtered. (Routs start in the rear area of the side being routed, so losing side casualties are actually lowered.)

[herwin]

February 9/10 1942

With the fall of Singapore, Japanese forces become more aggressive. No significant engagements, but more minor engagements and ASW operations.

[herwin]

I received a copy of the Helmbold data base as it currently stands (about 650 battles). If you use the the defender's casualty percentage as a proxy for the battle duration, battle durations are log-normally distributed with mean duration about 10% casualties and standard deviation about a factor of two. You get the same pattern if you use attacker's casualty percentage or the average of the two. More importantly, the duration of a battle is relatively insensitive to the ratio of the percentage casualties on the two sides. A model consistent with this is that most battles continue until the soldiers on the two sides have a consensus as to who's losing (who's taking higher percentage casualties), at which point the loser retreats if their culture allows them to. Of course, Japanese ground units wouldn't, causing the battle to continue until they were wiped out or their commander ordered a retreat.

In game terms, choose a decision point from a random log-normal distribution with mean 0.1 and standard deviation 2x. Run the firepower duel until the average of the percentage casualties on the two sides equals that decision point. Then see who's losing. If it's the allies and they're defending, they retreat or surrender. If it's the allies and they're attacking, the attack fails. If it's the Japanese, roll the land skill of the commander to see if he does the sensible thing. If the commander fails his roll, the Japanese forces in the hex continue to defend/attack for that day. Japanese troops do not surrender. Commands carry over day to day until countermanded.

I should add that the probability of the consensus being objectively wrong is overall less than 10%, and reflects the size of the difference of the log10 of the percentage casualties for each side. I.e., sometimes in indecisive battles, the loser wins. The easiest way to handle this is to calculate the difference of the log10 of the percentage casualties for each side and then add a small normally distributed error (with mean zero) before deciding who won.

[herwin]

February 10/11 1942

The Allies don't expect to hold the Dutch East Indies or central China. There is hope for Port Moresby, Timor, and Burma, but only if Japan treats them as sideshows. Small garrisons are being deployed at ports and bases where a continuous Allied presence would speed up the counter-offensive.

Betties seem to be operating out of Balikpapan and perhaps Kuching.

[witpqs]

With respect to the AE map, what do you consider central China in this context?

[herwin]

ORIGINAL: witpqs

With respect to the AE map, what do you consider central China in this context?

The region centred on the route from Sinyang to Hanoi.

[witpqs]

Including Changsha and surroundings?

[herwin]

ORIGINAL: witpqs

Including Changsha and surroundings?

Yes. That provides an overland alternative route for resources and oil.

[herwin]

February 11/12 1942

Netties now based in Singapore in addition to Balikpapan!

The III Philippine Corps just showed up in Davao with two full-strength divisions. The Japanese garrison in Davao is a brigade...

[herwin]

I think battle termination involves an information collection process. If it were a one-sided non-zero-sum continuous game with information collection, the fight would continue as long as the attacker thinks it has a high enough chance of winning to justify continuing. Think of a continuously updated estimate of the chance of winning that is compared to a constant threshold. If the estimate drops below the threshold, the attacker abandons the battle. I don't think we see that. To prevent bluffing from being a viable strategy, real casualties have to be taken, and we do seem to see that. Also, if you graph the log of the fractional casualties of the two sides at battle termination, it looks like a gaussian distribution, and the degree of mismatch does not seem to affect the choice of quitting point, so a model of battle termination can be quite simple--the battle appears to last until both sides have reliable data on who's likely to win (whose taking lower fractional casualties), and the losing side then quits the field.

Or if Japanese, maybe not. Only the Japanese in WWII were idiotic enough to hang around for a done deal, and even then an intelligent commander would see the writing on the wall.