The relative effectiveness of the allied and axis armies in the ETO

[warrenpeace]

JasonC-You did a nice job in summerizing Dupuy. I checked the book out of the library a while ago, so I don't remember exactly how he evaluated battle "success", so I will defer to you on that point.

With regards to his methodology, what Dupuy is doing is basically what a lot of scientists and operational research guys do, create models based on empirical data and then test them. My lab is uses a similar modeling approach to develop a diagnostic test to identify people with kidney cancer. The problem with all models is one of "over fitting", where you assess the validity of the model only using the data you used to make it in the first place. The interesting thing about Dupuy's model is that the model has been tested on separate valadation data sets that were not used to make the model in the first place. It looks like the model is right about 80-85% of the time.

Now if I put on my criticism hat, I will point out that if one flips a coin, one will guess the outcome of a binary event (like a battle) 50% of the time, so saying that a model is right 80% of the time is good, but not great.

Micky D-I've got to read that book!

[Chek]

Chek

I agree that logistical understanding is important for Generals, although two of the best, Rommel and Patton, were not known as great logisticians. My point is that there are two types of logistics, strategic and operational. Strategic logistics are determined at the national level, e.g. things like: How many tanks do we need to make?, How many landing craft?, should we engage in the Manhatten project? etc. Operational logistics are how do we take the stuff that we have and distribute it too the troops such that they have what we need. I would argue that strategic logistical questions are not strictly military but are a combination of military and political. Furthermore, I would state that WW2 was won by the Allies primarily on the basis of these decisions and their superior execution.

I believe, overall, the German Military was quite competent with regards to operational logistics given the strategic decisions that were made. There are a few exceptions (winter gear for 1941 comes to mind). Much of their poor strategic logistic decision making stems directly from their political process in which Hitler made many decisions.

An excellent clarification that makes good sense.

[kensal]

That's all fine, plugging numbers into an equation, but there doesn't appear to be anything in those equations that account for most of the real world factors that affected the German army in 1944. He is also, by default, cherry picking engagements to make his calculations since he isn't selecting every division sized engagement. It's a data set, but it isn't very persuasive to me since I'm well read enough to know about the substantial problems faced by the German Army in 1944. Which battles during the Battle of the Bulge did he select? The ones involving the 101st Airborne at Bastogne or the ones involving the green American forces in the St Vith area?

I don't believe Dupuy does this - the studies made by the Dupuy institute looked at hundreds of engagements in Italy, the Ardennes and Kursk for example. Individual divisions are given individual CEVs. His studies recognise very wide qualitative differences between different divisions in the same army - particular the German army for obvious reasons.

[kensal]

JasonC-You did a nice job in summerizing Dupuy. I checked the book out of the library a while ago, so I don't remember exactly how he evaluated battle "success", so I will defer to you on that point.

With regards to his methodology, what Dupuy is doing is basically what a lot of scientists and operational research guys do, create models based on empirical data and then test them. My lab is uses a similar modeling approach to develop a diagnostic test to identify people with kidney cancer. The problem with all models is one of "over fitting", where you assess the validity of the model only using the data you used to make it in the first place. The interesting thing about Dupuy's model is that the model has been tested on separate valadation data sets that were not used to make the model in the first place. It looks like the model is right about 80-85% of the time.

Now if I put on my criticism hat, I will point out that if one flips a coin, one will guess the outcome of a binary event (like a battle) 50% of the time, so saying that a model is right 80% of the time is good, but not great.

Micky D-I've got to read that book!

That's always the problem with forecasting models though, isn't it. It seems to me that a model that is 80-85% correct in respect of a subject which probably has as many variables as global weather systems is pretty impressive

[ASL Veteran]

Yes. Most researchers have limited budgets and limited amounts of time. They have to replace asking everyone in the world, time machines, or god-like omniscience with DUM DUM DUM!.... statistics.

Garbage in Garbage out

The mumbo jumbo tossed in there describes why you'd be justified in trusting those subjective opinions more than you could under other circumstances.

Not really. A subjective opinion is a subjective opinion no matter how much you try to dress it up as a fact or a statistic.

The bottom line here is that Dupuy even says himself that he can't isolate factors like leadership, doctrine, and training in his calculations. As far as he is concerned they either don't matter enough to him or they are just magically included in his number in some indeterminate way that he can't account for.

[JasonC]

"they either don't matter enough to him or they are just magically included in his number in some indeterminate way"

False.

His CEV number is precisely a residual of demonstrated, consistent, empirical combat performance that is *not* explained by number of weapons, their firepower characteristics, or operational variables like combat stance, terrain, weather, etc. He is precisely insisting that such an effect *is* real because it is stable and reproduced across multiple combats, he knows that there are many militarily relevant quality variables that he expects to produce such an effect, he sees that effect and can quantify how large it actually is.

He is not able to determine what portion of the 1.2 times outperformance German to Allies in Italy in 1944, or 3.0 times outperformance Israeli to Arab in 1967 and 1973, is due to leadership rather than doctrine or training, but he does know that combined, they had an actual effect on combat outcomes that large, in those historical cases.

Of course calculated empirical CEVs "matter to him". He found them, he substantiated them as real, and he quantified them, in the aggregate. He did not need to first postulate their causes to find them in the actual combat performance seen in empirical battles.

[warrenpeace]

For the younger players, this thread should reinforce the importance of taking some some statistics courses. They are useful in a variety of fields and will keep you from using the term "mumbo jumbo"

[JasonC]

warrenpeace said in relevant part "Operational logistics are how do we take the stuff that we have and distribute it too the troops such that they have what we need."

No, that is merely supply. Logistics is a much more fundamental aspect of operational planning, and much more tightly integrated into the "G-3" (operational) thinking, not just the "G-4" management.

How many divisions can actual fight along route X? In what combat posture? How important is it to clear the road network from A to B, to sustain an offensive beyond B? What is the through-put of all the transport links behind the army? What is its base of operations? How many days of combat supply are on hand with the units, within the supply chain, at the base of operations? What is the stable "run rate" between supplies entering the army area and their rate of expenditure, when N divisions are in an attack posture? When 2 N divisions are in an attack posture? What portion of calendar time can the army afford to spend in an attack posture, without running its stocks to zero? How long must operational pauses between major attacks be, to give the transport links time to catch up with the planned rate of expenditure? What reachable army objectives lie within the radius of action defined by that "surge" attack time before the next enforced pause? Can the capture of any of them improve the total rate of supply through-put reaching the army?

Now, what is the enemy supply situation in the same terms? What are the critical bottlenecks in his own supply chains? What points must the enemy defend, or run out of ammunition or fuel in place if he loses? What alternative lines of communication might be available to the enemy if he loses location X, or is forced to fall back past railroad junction Y? How rapidly can he "stand up" a supply chain along those alternate linkages? What is his through-put to the far end of that chain? Can he shift existing stores to the new line of communications - how rapidly? If forced to change his LOC within 2 weeks, what portion of his existing stores along the old LOC would be unused and unmoved, and so effectively destroyed or even captured? Can army interdiction efforts reduce enemy through-put along critical portions of his supply chain?

Logistics operational thinking is not just distributing an available supply of beans and bullets to units in the field, nor does it consist in demanding more supply from the national command authorities, nor in hoping the national or coalition economy can increase the total quantity of war material available. It is an operational planning matter. It dictates military possibilities and goals as much as the arrangement of divisions along the front line or schemes of maneuver do. It is capable of decisive results in itself in many cases - indeed, some are so obvious that maneuver thinking barely even registers that the effects are logistic in origin. (Why do surrounded armies evaporate within about 2 weeks? Answer, above all because the through-put of artillery ammunition reaching them drops to zero).

A logistically aware general thinks in all of the above terms, both about his own army's capabilities and about the capabilities and vulnerabilities of his opponent. He thinks of his trucks and ships as military means as potent as his tanks and artillery pieces. He turns an onrushing flood of war material onto his enemy, with his transport linkages the "hoses" delivering that water-power, and his maneuver formations merely nozzles increasing its pressure at the point of delivery. And he views the enemy transportation network as a vital system of the enemy army, as important to its functioning and potentially vulnerable to military attack as the arteries in the bodies of its soldiers.

That is what we mean by military logistics, and what is meant in the adage that amateurs think about tactics, while professionals think about - all of the above.

[Tarquelne]

Garbage In Garbage Out

Statistics is useful in identifying what's garbage and what isn't. When you researching a messy subject with a lot of "garbage" statistics becomes a primary tool.

It works even better than sneering.

Not really.

Yes, really. Note the "more" in my statement. I didn't say all the cross-checking transforms the ratings into Truth, or whatever. But it does allow you... well, others... to attach more confidence to the results. Both in the non-technical and, loosely-speaking, technical sense of "confidence."

A subjective opinion is a subjective opinion no matter how much you try to dress it up as a fact or a statistic.

Duh. But you can get measures of the reliability of a subjective opinion.

The mere fact than a claim is subjective doesn't mean it's useless. You just need to start thinking about things like error bars when you use it.

For the younger players, this thread should reinforce the importance of taking some some statistics courses. They are useful in a variety of fields and will keep you from using the term "mumbo jumbo"

Lets be fair. In fields where being pig-ignorant AND obdurate are helpful, there's nothing a course on statistics can do but dull your competitive edge.

More seriously: Even the hard sciences often have to use the sorts of correlation-searching, or data-massaging that the poor, benighted social science guys are generally stuck with using all the time. (Assuming they don't just drink themselves into becoming post-modern.) Understanding what statistics can and can't-do is a basic part of being scientifically "literate."

I'm done with this thread. I'm going to go meditate on xkcd 386.

[warrenpeace]

JasonC,

Given your lucid description of logistical planning, do you think the Germans were inferior at it? Was it decisive?

[db_zero]

Been following this thread and tempted to throw in my .02 cents as I may have a few ounces of relevant insights as well as some experience and knowledge of some of the areas touched on.

but for now I'll just lurk and be entertained.

[Ithikial_AU]

The problem of using statistics to try and predict warfare is the enormous number of variables as two opposing forces enter the battlefield. Variables that can favour and hinder certain groups in any number of ways that won't be apparent until after the fact.

- Terrain

- Equipment quantity and quality

- Troop quality and morale factors

- Communications

- Weather

That's just off the top of my head. Trying to put some arbitrary value onto one force or another is a waste of time, especially if you are only selecting certain engagements to study. If I did this in my job (economist) I'd be crucified. If you're doing your research right you'd be following a sample of units (probably at the Divisional level) throughout the war, following that force to see how they performed under different conditions and circumstances. Not giving enough time in the study to take account of a changeover of personnel and leadership. For example take a look at the 7th Armoured Division - very different effectiveness levels in North Africa compared to Normandy.

Granted I don't know anything about this Dupuy methodology but if on the qualitative evidence side he only referred to three military historians then it's not a wide enough sample of evidence. It's like trying to work out an unemployment rate for the USA by only asking the employment status of the people in one of the fifty states apply it to the country and then not follow up with those people again assuming their situation will stay the same in the months following.

And this is why proper research and statistical collection is a very expensive process.

[JasonC]

warrenpeace - if I am grading on a curve, the Americans were the best at it and the late war and professional Brits their only equals. (Early war the Brits ran a lot of amateurish shoestring ops they hadn't thought through, trying to "do something" - e.g. the Norway campaign or Greece - hence the qualification). The Germans were better at it than the Russians, especially early to mid war - they had pros on large staffs with the brains for it, etc. Objectively they are in the middle of the pack on the subject, because others have their own stuff ups, is what I am trying to say with the "curve" comment.

But if I am not grading on a curve it was an area of serious weakness in the German operational art. They performed poorly in the area as a whole, compared to what they could have done, and it hurt them badly on pretty critical occasions. Basically they didn't listen to their professional staffs, because the operations guys and line commanders and higher ups were pushing audacity as the one size fits all solution to all problems. With the predictable result that they pushed way beyond any sensible logistical plan, over and over.

Examples are failure to plan for Barbarossa adequately in logistics terms, breakdowns east of Smolensk in 1941, the shoestring nature of the push to the south in the summer of 1942, the craziness of dividing the force in that campaign (twin objectives of the entire Caucasus plus Stalingrad). In North Africa Rommel was a clinical case of it, pressing way too far because he liked the headlines, when it meant a lost battle at the other end of the rubber band. Over commitment to Tunisia and horrible air losses trying to sustain it is another example.

There are lots of cases where the Germans viewed the war as basically a set piece between "stock" existing forces, rather than an ongoing struggle between "flows", and it hurt them badly. That includes underestimating the Russians, it includes vastly underestimating their own losses and replacement stream needs, it includes underestimating the importance of rail work, it means driving trucks to destruction trying to win the war today and not having them next week, etc.

I hope that helps.

[JonS]

the studies made by the Dupuy institute looked at hundreds of engagements in Italy, the Ardennes and Kursk for example.

Are you sure about that? I'm fairly sure the data set is measured in dozens of battles, not hundreds.

The problem of using statistics to try and predict warfare is the enormous number of variables ...

No, that is exactly what statistics is for.

I can't count the number of times I've seen comments on news websites that dismiss some observation because - in the commentors opinion - the sample results can't be extended to the whole population. FFS. That is the whole reason stats was invented, so you *don't* have to ask everyone every time you have a question! Graaah!

What does change, with sample size, is the margin of error increases as the sample size decreases. But that margin of error often isn't reported, and even less often is it understood.

That is part of the problem I have with Dupuy's methodology - the confidence interval/margin of error is, as far as I can recall, never mentioned. The oft touted numbers of 100 Germans being the "equal" of 120 Americans, and equal to some-unspecified-but-implied-to-be-higher-number of Brits. Well, that's fine, except if the margin of error is at all close to 20% (and given the small sample size of of battles investigated and the number of variables involved in each of those battles I would expect that it is in that range) then statistically speaking there is NO difference between the US, the UK, and Germany.

If the confidence interval/margin of error were included, I would not be at all surprised to find that there was more variability within each national army than there was between the US vs Germany vs UK. By which I mean that the best and worst from each nation was more than 20% from the norm, especially the Germans. They had some really good units, but they had a not inconsiderable number of utterly execrable units too. The Allies had some good and bad units too, but units that actually made it to combat (as opposed to training divisions never intended to face the enemy) generally had a reasonably even level of competence. For example; in Normandy the US 90th Infantry Division has a really bad reputation, probably deservedly so, but it's bad in comparison with other US units in Normandy. It wasn't bad in any absolute sense - it's not like the whole division surrendered, or broke and ran, or anything as dramatic as that. It was 'just' slow, and often failed to achieve offensive objectives.

I also have a problem with Dupuy's methodology in that his sample set of battles isn't a random selection of battles. Instead it's a set of battles for which his team could find data. That's fair enough - the kind of data he needed is notoriously difficult to find for one side in any given battle, let alone both. But unless the statistical model was carefully selected and constructed the conclusions could easily be invalid, since many of the common standard statistical tests assume a random sample.

As an aside, I had a look at Numbers, Predictions and War a year or three ago with a view to using it to backstop calculations for an operational level wargame, and came away with the impression that, while he'd done a tremendous amount of number crunching, the stats used wasn't all that convincing, and actually the final 'product' he'd come up with - i.e., the model he developed - was all-but un-usable. At the time I suspected that part of the un-usability was to protect proprietary information - if the only people who can actually use the TNDM and QJM to produce meaningful output are TDI, then TDI gets more coin every time someone has a question. That might make sense from a business point of view, but it isn't very convincing academically.

Rolling the thread back a bit, I thought that Bagration was actually a very good example of Russian operational art. They didn't get 12:1 odds (or whatever it was) by accident. They got 12:1 odds because they totally outclassed the Germans operationally. They baffled the Germans about the timing, location, and scope of a major offensive, so that when it actually struck the Germans were caught flat footed and with their best assets in completely the wrong place. Which is kind of the point of operational art - beat the other guy at operationally some you can thrash him tactically.

Except that all the guys with Wehrmacht ***** envy ignore the operational level, concentrate exclusively on the tactical level, then whine "wah! They just won because they threw more material at the Germans! wah!"

Jon

Edit: AIUI, USians tend to use Math instead of Maths as the contraction for Mathematics. What do you use as a contraction of Statistics - Stat?

[womble]

Rommel, for instance, never grasped how unimportant the North African theater was in the grand scheme of things. This led him to constantly attempt more than his army with its limited logistical support was capable of.

Michael

I'm wondering why denying British access to Suez and potentially all the Middle Eastern oil (if the conquest of Egypt could have been exploited further east) and granting the Axis powers access to those same strategic assets wouldn't have been an important strategic goal. The Brits thought it was pretty important.

I'm not saying Rommel wasn't logistically naive in expecting to be able to take Egypt on his allowed logistical (and force) shoestring, but given the importance of oil and the Med to the prosecution of the war, wasn't he right to try as hard as he could (even though he'd been sent with a holding brief)? If Hitler had turned the Wehrmacht loose in the Levant instead of starting Barbarossa, what took Rommel 18 months with a couple of divisions and some Italians could have been done much faster, and maybe Gibraltar might even have been taken.

[JonS]

It's a negative objective. The Germans would never have gained anything useful out of it. At best they would be denying the British something. Winning means achieving your own objectives, not denying the enemy theirs.

Securing the Med and the Middle East could have been a useful means-to-an-end for the Germans ... assuming they had had some kind of overarching strategy for the conduct of WWII. But they didn't, instead they were fleeing forward from one transitory or illusory opportunity to the next, regardless of where it took them strategically. In that sense the German campaign in the Med from 41-43 made good headlines by seizing opportunities, but it was a waste of their time because it wasn't serving any useful end.

[womble]

It's a negative objective. The Germans would never have gained anything useful out of it. At best they would be denying the British something. Winning means achieving your own objectives, not denying the enemy theirs.

Securing the Med and the Middle East could have been a useful means-to-an-end for the Germans ... assuming they had had some kind of overarching strategy for the conduct of WWII. But they didn't, instead they were fleeing forward from one transitory or illusory opportunity to the next, regardless of where it took them strategically. In that sense the German campaign in the Med from 41-43 made good headlines by seizing opportunities, but it was a waste of their time because it wasn't serving any useful end.

Were the Germans not struggling for petrochem resources? Would denying the British bases like Aden and Gibraltar not have made it easier to get them to the negotiating table? Isolating India could have precipitated the nascent rebellion of that Crown Possession, making Japan's threat to the US significantly greater by effectively clearing the British out of the Far East, and providing more supply to the Nips. Would these not have been useful ends?

Admitted, they were useful ends that were probably out of reach because of the Russian misadventure, but the conjecture is largely dependent on Hilter choosing to finish one thing before he started another (even though his first way of finishing it hadn't worked). With the force Rommel was given, and the importance assigned to the theater, it wasn't ever going to be more than "headlines", but seizing the means to continue to prosecute your war, relieving an ally of one responsibility so they can assist you on a different front, and establishing internal lines of communication with another ally do seem to be potential useful ends that Germay didn't pursue with sufficient vigour.

[JonS]

Were the Germans not struggling for petrochem resources?

Germany would have moved fewer hydrocarbons back to Germany from Iraq that they did from Baku.

Would denying the British bases like Aden and Gibraltar not have made it easier to get them to the negotiating table?

Why? The Med was already unusable, with everything being sent via Cape Town.

Isolating India could have

India was already isolated.

Admitted, they were useful ends that were probably out of reach because of the Russian misadventure, but the conjecture is largely dependent on Hilter choosing to finish one thing before he started another

So. Yeah. That's exactly my point. When I wrote:

Securing the Med and the Middle East could have been a useful means-to-an-end for the Germans ... assuming they had had some kind of overarching strategy for the conduct of WWII.

What I meant to say was that:

Securing the Med and the Middle East could have been a useful means-to-an-end for the Germans.

Assuming they had had some kind of overarching strategy for the conduct of WWII.

But they didn't. So of course it wasn't.

WWII was like a game of chess in which a clever but erratic neophyte who could only plan two moves ahead went up against a slow and grizzled old swami who'd already forgotten more than the neophyte would ever know. The neophyte got a few licks in with his dazzling seizure of random opportunities - taking a pawn there, and a bishop there. But eventually the swami crushed the neophyte because his pieces were scattered and isolated all over the board, and the swami snaffled them all up with ease in a brilliantly conceived, planned and executed endgame.

potential useful ends that Germay didn't pursue with sufficient vigour.

Sure. Except you never, ever, have all the resources to do everything you'd like to. Which is why having a strategy is so important.

[kensal]

Are you sure about that? I'm fairly sure the data set is measured in dozens of battles, not hundreds.

I was referring at the later Capture Rate study, which looked about 180 divisional engagement. I suppose calling that 'hundreds' though might be a slight exaggeration

[Ithikial_AU]

No, that is exactly what statistics is for.

I can't count the number of times I've seen comments on news websites that dismiss some observation because - in the commentors opinion - the sample results can't be extended to the whole population. FFS. That is the whole reason stats was invented, so you *don't* have to ask everyone every time you have a question! Graaah!

But the answers your sample need to return have to be consistent and readily grouped together, in the simplest form a 'yes/no' answer. (Granted most surveys have a scale or multitude of responses). That's my point, when you look at a military battle you're always going to get very different responses that can't be organised and catergorised so neatly because of the high number of possible answers to your quantitative and qualitative questions. Those answers are the problem, there is never going to be a consistent response between battles or reasons why.

I agree with the rest of your post about concerns with the sample etc.

[warrenpeace]

Ithikial-as an economist you should definitely look at Dupuy's modeling. There is no 3 historian subjective part, that was a fallacy propagated by ASL squad leader. You should look at the link I posted and his book to see more precisely what he did. BTW, I've enjoyed watching your videos on utube.

JonS-I kind of agree with you. When looking at the actual data points in Dupuys book, there clearly would be wide confidence intervals in his modeling. It would be interesting to have access to his data and use somewhat more sophisticated modeling techniques which include confidence bounds.

JasonC-I agree that the Germans viewed campaigns in a short term time frame and did not plan for longer attritional warfare. It probably stemmed from their faith in blitzkrieg which worked pretty well in the 1939-1941 time frame. With regards to Barbarossa specifically, I wonder what the German estimate was about how long the campaign would actually last?

[JasonC]

People overestimate the importance of the middle east for oil in the WW II era. Saudi Arabia had started its development, true, but the first oil even found there was in the 1920s and investment there was still in its infancy. The world oil powerhouse was the US, which alone produced about half the world's oil in 1941. Anyone remember John D Rockefeller and Standard Oil?

The places there was other significant oil production in the 1940s were (1) in Russia especially in the south, Baku, (2) in the Dutch East Indies, Java (ever heard of Royal Dutch Petroleum?), (3) the middle east, divided between Saudi Arabia and Iran, but on the Persian gulf regardless. The only area accessible to the Germans was Romania, but it was a tiny production center for the stuff in world terms.

But getting an army to a spot on the earth's surface under which there is oil does not mean getting oil into one's economy. It is again a question of logistics and infrastructure. The Japanese needed a merchant marine to get oil from Java to Japan, and US submarines sank that merchant marine by 1944. Germany getting to Suez would not have given them a drop, it was all way to the east in the Persian gulf, north and south shores.

The Germans could not sustain 3 mobile divisions as far east as Cairo, carting everything by truck across a single paved highway in desert conditions from Tripoli, the first port with any capacity that was protected enough by airpower to send a freighter there without it being sunk enroute (lots were, actually). How are they supposedly going to fight the force the Brits could sustain by sea supply in central Iraq, say, to reach the shores of the Persian gulf where the oil was? Truck everything 2-3 times as far?

If they drove British air out of Egypt and the Levant they might try to ship by sea to the Levant coast - to what ports, with what merchant marine?

The British could live and fight in the middle east because they had the largest navy in the world and the largest merchant marine (at the start of the war - the US passed them on both scores during the war). They had 21 million tons of shipping able to go anywhere in the world in a few weeks, safely. The Germans started the war with 3 million tons, lost half of it in the early war, and the rest was all bottled up in northern European ports.

The Italians started the war with 3.3 million tons of shipping, but 1.2 million of it was outside the Med when Mussolini declared war - on France's collapse, with no notice (literally 3 days to get home from all around the world) - and practically all that shipping was lost, seized by the British. The combined Axis built 3.8 million tons of shipping over the course of the war. For comparison the Allies built 44 million tons and started with about 30 million tons - and shipping was *still* the main Allied strategic bottleneck clear to mid 1944.

There is simply no way the Germans were ever going to compete with the Allies in any major battlefront where the supplies had to come by ship. Germany was a continental power. It depended on its efficient railroad network as its major arterial "lift'. Trucks could act only as "lighters" to distribute beyond that rail network to fill in the space between the major lines, as the small veins. At great cost and limited through-put it could extend a few hundred miles beyond working rail lines with truck transport, but that stressed its narrowest category of strategic supply, which was fuel.

The rail lines could lift any amount and run on coal. If the Germans can get a rail line within 100 miles of the spot, they can fight there at full combat power. If they can get a rail line within 200-300 miles, they can fight there on a shoestring of emergency supply, getting one panzer corps punching hard for short periods to attempt coups de main and operational brilliancies, which had better work and work fast. That's it.

Rommel never had enough of anything and lived off stuff he captured from the British. He never had enough trucks to keep the full German let alone the Italians as well supplied as far as Tobruk, let alone that plus lift for the infantry and artillery of the force, let alone enough to truck fuel forward to El Alamein without burning that fuel up just getting the truck there. They were never getting to Iraq, and if they did they would have no way in heck of getting anything back out to Germany.

If the Germans had conquered Baku, it would have taken them a minimum of six months and probably more like a year to get the fields back in shape and extend rail lines running German gauge trains that far, to get anything back out. It was the only oil in the world within their operational and logistic reach, besides the Romanian stuff they already had, and their domestic synthetics (from coal).

[JasonC]

On the statistics discussion, there are two things going on. One, people are misunderstanding what Dupuy did and found. And two, people are misunderstanding statistics and how much it can do - it can do more than they think.

On the first point, no the 20% edge to the Germans that Dupuy found for his Italian, Normandy, and Bulge battle sample is not within the margin of error. What he actually did is run the analysis with no CEV - no national "fudge factor" - which means the model counted objective combat power and added factors for stance, terrain, weather etc, where the weight given to each of those factors was found empirically as a model fit for the importance of those factors on the observed outcomes. Dupuy did not make up or guess how important those factors were, he let the regression equation do that, trying to minimize the least squared error between an adjusted combat power comparison and the observed outcome.

And he found with only those factors his model got a third of the outcomes wrong. The adjusted combat powers coming out between 0.9 to and 1.1 to 1 were not draws or randomly distributed, even wins, but nearly uniform German wins. He also noticed that these occasions also seemed to involve times and fights in which Allied airpower was less of a factor, for weather or range from base reasons or both. He was trying to understand why a model without CEVs was mis-classifying a third of his sample.

So he let there be another factor for nationality of the force, and let the regression equation find the best fit with that factor added. And the equation upped the importance of the air power variable, and gave the Germans a 1.2 weight. And with that addition, it was now correctly classifying 90% of the cases instead of missing on a third of them.

Moving a model from correctly classifying 67% to correctly classifying 90% of the sample for the addition of one extra degree of freedom is definitely a statistically significant improvement in that model. The BIC or AIC will improve by a lot.

Of course the actual outcomes are still distributed, not inform or deterministic. Of course there are factors contributing to outperformance or underperformance of the model standard in every individual battle. The reason we do statistics is to try to capture main drivers in the presence of such case by case noise, which we expect to be distributed *randomly* as to scale and sign, from case to case, in ways that are *independent* of the main factors we are trying to model. That is all a statistical model needs to be true about the factors it knows it is leaving out.

A statistical model will then tell you, from how well it classifies all the cases, how well it has done at explaining the empirical variation in outcomes. Classifying 90% of cases correctly is very, very good as statistical classifiers go. The fact that the model improved massively in its correct classifications when it "got to" use a national "fudge factor" as an additional degree of freedom, also tells us quite strongly that there is some real, persistent, not-random skew in the average effectiveness of the different national armies.

Of course that involves some cases in which a German formation underperformed its combat power and others in which an Allied formation outperformed its combat power or both. The race is not always to the swift or the battle to the strong - but that *is* the way to bet. Please do notice, however, that the revision also upped the importance of air power, which was normally on the side of the Allies and by a large amount. Given a choice between having German average unit quality or complete command of the air, complete command of the air is a better thing to have.

1.2 times factors are still minor. Recall that Dupuy's range of indeterminate, "draw" outcomes was 0.9 to 1.1 in width. This means he predicts that other things being equal, an Allied force with 4 to 3 odds will deterministically beat a German one. But it also means that he predicts if the forces are truly equal (including "operational" factors like terrain stance weather effects etc), an equal combat power German force would deterministically beat an Allied one. But "other things being equal" also means the Allies don't get a lot of air support. If they have lopsided air support, that may well fully cover the 4 to 3 ratio above.

More broadly on what statistical models can do... If you give me dozens of variables with no correlation with the outcome, and throw them together with dozens more with slight correlations with the outcome, plus 3 say with strong correlations with the outcome (combat power, nation, stance e.g.), and you have deliberately created the outcome data from such a process, a statistical model will detect those relationships. It will be able to tell that the dimensions that are random add nothing to the fit, by a measure like the Bayesian Information Criterion (a penalty for having extra degrees of freedom), and drop those dimensions. It will keep the minor effect dimensions only if they add something, either systematically but weakly across many cases, or strongly but rarely in a few cases where that minor dimension was far from its typical value (outlier weather or air support e.g.), and will give the largest weight to the main drivers.

Statistics are not perfect in that, and if you raise the number of dimensions into the thousands they can be fooled. (At bottom that happens for a purely "geometric" reason - that most of the space in high dimensional volumes is in the corners - that is, most points are outliers in several respects if the number of dimensions is high enough). But if the dimensions you included had and have plausible real relations with the outcome - certainly the case for things like measured combat power from counting weapon systems, or combat stance - and there are 20 or fewer dimensions involved - they are very good at picking out the truly important and giving it proper weight.

People who doubt this need to make some simulated, "fake" data sets where they put in a real casual relationship, by actually deriving that data from the different variables by a known process, and then run that through some statistical modeling code and see directly how well a statistical model "recovers" those "ground truth" drivers. If you haven't done this, I recommend the exercise. A lot of common intuitions about what statistics supposedly "can't do" or what will supposedly mislead such methods are flat-out wrong. Such methods are not perfect, but they are much more robust than the average man in the street believes.

[db_zero]

Statistical and computer modeling are very useful tools, but like any other tool can be misused and provide a false sense of knowledge and security.

The recession we are but slowly emerging out of was caused by a massive housing/credit bubble that was engineered by the widespread use and belief in esoteric statistical anaylsis and computer modeling there was an unshakable belief that we could eliminate risk. That turned out to be terribly flawed.

Don't get me wrong I think statistical and computer modeling is a fascinating and useful tool, but I I personally believe it has limitations. I also have an opinion of the people who do that for a living, but that's about all I have to say on that matter.

As someone mentioned if your young and in school learning statistics is quite useful. Actuaries are quite employable and can make some pretty decent money, but you really have to love the work. It woukd be a miserable life doing that line of work if you hated it.

[womble]

Actuaries are quite employable and can make some pretty decent money, but you really have to love the work. It woukd be a miserable life doing that line of work if you hated it.

Accountants make jokes about how boring actuaries are...