46 and 56 British Infantry Division at Salerno

[Richard Anderson]

The Pentagon never did any "analysis" of the data. What they did, in various iterations of ever growing silliness, is nitpick the numbers.

Well they did produce this:

To be fair, yes they did produce some nice graphs...and did the same with the ACSDB and KDB, but they never did any analysis with it. Instead, they found "anomalies" and so eventually convinced themselves the data was flawed, which gave them an excellent excuse to sit on their hands and do nothing. Which they have done successfully to his day.

(And yes, I am being particularly mean to Walt Bauman's team, which did the KOSAVE II study using the KDB. It was a worthy effort, but like all such Army studies, eventually went nowhere. Perhaps its telling that the Army renamed the U.S. Army Concepts Analysis Agency as the U.S. Army Center for Army Analysis...they could have just gone ahead and called it the Redundant Redundancy Agency.)

[Tom from Cornwall]

Rich,

I'm not sure "disdain" is the correct word? The report simply described the assessment of the two divisions (albeit that of the 46th was not really correct, see below).

The 56th Division assessment was built partly on incorrect data. Since at least three of the six engagements nearly doubled British casualties and that is a strong component of score effectiveness, which helps determine CEV, the average weighted CEV score for the division should probably be higher than the 0.60 for nine engagements as recorded in NPW. Unfortunately, we were never funded to rework the data base.

Nor was the report correct in stating that the 46th Division was assessed as performing "particularly poorly"...it scored 0.96, better than the other four divisions, better than six of the seven U.S. divisions, and better than three of the 12 German divisions, while nearly matching two of the other German divisions.

So essentially, anyone using the Dupuy numbers as a basis for any analysis of relative combat performance (British, US and German) in Italy is simply building a house of cards (or "on a very sticky wicket" as we would say in the home of terrible cricket!).

BTW I always knew that you were one of a large well-funded team!

Regards

Tom

[Richard Anderson]

So essentially, anyone using the Dupuy numbers as a basis for any analysis of relative combat performance (British, US and German) in Italy is simply building a house of cards (or "on a very sticky wicket" as we would say in the home of terrible cricket!).

Um, no, anyone using the figures from NPW for the British 56th Division must realize they are flawed, 41 years old, and were part of a much smaller database. If you want an update with greater fidelity I suggest you consult War by Numbers...and argue the results with Chris rather than me. I have not worked for TDI for ten years now (well, come this November) and did not know he had re-run the engagements in the TNDM.

Meanwhile, while the old weighted CEV comparison might shift for the 56th Division relative to its nine engagements, I doubt its relative position - or the general conclusions - will shift much in what is now a 752-engagment data base rather than an 80-odd engagement data base.

BTW I always knew that you were one of a large well-funded team!

No, I am a large team, but not well-funded.

[Tom from Cornwall]

Hi Rich,

War by Numbers

Thanks - I'll see if I can pick that up at a reasonable price.

...argue the results with Chris rather than me

But that wouldn't be half as much fun...

I doubt its [56th Division] relative position - or the general conclusions - will shift much

Which, to be honest, is what I'm trying to understand. If 56th Division's performance was so relatively poor, firstly why was that and secondly why was that despite the strength of it's Divisional Commanders (Graham and Templar)?

It is clear, though, that when they were sent piece-meal to reinforce the Anzio bridgehead there was some discontent at being thrown once more into battle without rest.

Regards

Tom

[Aber]

I doubt its [56th Division] relative position - or the general conclusions - will shift much

Which, to be honest, is what I'm trying to understand. If 56th Division's performance was so relatively poor, firstly why was that and secondly why was that despite the strength of it's Divisional Commanders (Graham and Templar)?

See Rich's earlier comment about the database being not truly random.

There is likely to be some element of survivorship bias in the database ie if there was not an extended "battle" but just a simple and quick "bite and hold" operation, would it still be included in the database? Similarly, battles are more likely to be at critical points for the defence - does this build in bias? 56th Divisions' value in the database is mostly dependent on Salerno and Mount Camino - are these truly representative of the division's performance?

There are of course all the other points which could be discussed - how you define the battle, getting data on units and casualties, estimation of combat power of weapon systems, and the model used to evaluate results.

[Richard Anderson]

There is likely to be some element of survivorship bias in the database ie if there was not an extended "battle" but just a simple and quick "bite and hold" operation, would it still be included in the database? Similarly, battles are more likely to be at critical points for the defence - does this build in bias? 56th Divisions' value in the database is mostly dependent on Salerno and Mount Camino - are these truly representative of the division's performance?

Exactly. Too many get wrapped up in the "performance ranking" table in NPW. That is simple a snapshot of those divisions in those engagements. Expand the engagements and that can shift. For example, the U.S. 1st Armored Division "ranking" is built off of three engagements, one of which - Velletri - was also imperfectly recorded...in fact, it may have never happened at all, at least in the way it is described in the data base.

So correcting the three Salerno engagements for the 56th Division, plus adding in a Garigliano engagement - where the 56th and 46th were highly successful, especially in relation to the ill-fated U.S. 36th ID - and the "ranking" changes. But the "ranking" is nonsense anyway. Trevor understood that, but I think included it in order to make a point...he had long realized that with his audience he needed a hammer to make a point and have it sink into their brains.

At one time I had hoped to use the additional engagement data I had accrued and created a new data base that would have tracked a division or divisions for each side on a nearly daily basis for an extended period of the Italian campaign, so both the exciting days of battle and the boring days of tension in between would have been included. No funding though and no time to do it on my own.

There are of course all the other points which could be discussed - how you define the battle, getting data on units and casualties, estimation of combat power of weapon systems, and the model used to evaluate results.

All very true, which is why Chris's work should be interesting. One problem with the original data bases - both the QJM and LWDB - was that they were somewhat amorphous in terms of definitions. In the QJM "divisional" data base, "division" manpower ranged from 4,500 to nearly 100,000 men and from one to eighteen days. The LWDB was worse and included campaigns, such as the Somme. The more modern data bases are more carefully defined, as Chris described in his blog, so can be examined at divisional-level over the course of one or two days, or at corps-level or larger over the course of campaign periods, or at a smaller level in periods sometimes shorter than a day. They are also much larger, especially at the divisional level. All that improves the fidelity, comparability, and randomness of the data bases...although except for the campaign data they still tend to focus on the exciting bits rather than the boring bits in between.

[Tom from Cornwall]

Rich,

But the "ranking" is nonsense anyway...I think [Trevor] included it in order to make a point

You'll have to excuse my stupidity, but could you take up his hammer and let me know what was the point?

Aber,

56th Divisions' value in the database is mostly dependent on Salerno and Mount Camino - are these truly representative of the division's performance?

I guess my doubt on the original work would be the fact that when I look at 56 Division at Salerno and Mount Camino, I see a division (or bits of a division) given very difficult tasks. So not just are those engagements truly representative of 56 Division's time in Italy, but is the analysis truly representative of those engagements.

Regards

Tom

[Richard Anderson]

You'll have to excuse my stupidity, but could you take up his hammer and let me know what was the point?

That combat is more than counting up weapons systems and laying out shooter-target SSPK matrices. That differences in doctrine, training, leadership, and motivation result in varying performance on the battlefield. That just because the Western powers were democracies and their hearts were pure that did not necessarily translate into combat power.

Yes, I think the chart is somewhat nonsensical, mostly because people blow it way out of proportion, but it does have validity for those specific cases (well, the ones involved in the original Italian data at least) since the set of British/American units versus the set of German units is pretty small.

I guess my doubt on the original work would be the fact that when I look at 56 Division at Salerno and Mount Camino, I see a division (or bits of a division) given very difficult tasks. So not just are those engagements truly representative of 56 Division's time in Italy, but is the analysis truly representative of those engagements.

Well, since the analysis of the 56th Division is of the results in those nine engagements, then the analysis by default is truly representative of those engagements. And it is truly comparable to those divisions adjacent to them, such as the 46th Division and the American 3d, 36th, and 45th ID that were given similarly very difficult tasks, so I don't think the 56th Division was being picked on. OTOH, those engagements are only truly representative of those engagements in Italy, but are only representative of a very small part of their total time in Italy...maybe only 20 or so days in total of say just the 160-odd days from Salerno to the end of February at Anzio. That is why I say that "disparaging" the 56th Division - or really any of the other "poor" performers - is unfair...and pretty meaningless. I could happily disparage one of the "good" German performers - 3. Panzergrenadier Division - because the evidence is that it was actually a very problematic division initially in Italy. From September 1943 through the end of the year (and I suspect well into 1944) it had issues with desertion and motivation...too many of its enlisted men and apparently a good number of NCO were Volksliste III, which caused lots of heartache for Graser...but that simply doesn't come out in the data base.

[Aber]

I guess my doubt on the original work would be the fact that when I look at 56 Division at Salerno and Mount Camino, I see a division (or bits of a division) given very difficult tasks. So not just are those engagements truly representative of 56 Division's time in Italy, but is the analysis truly representative of those engagements.

I obviously don't have Rich's in-depth knowledge of the database and models and how they evolved over time, but I had similar doubts when I looked at Numbers, Predictions and War and tried to drill below the surface.

In theory as Rich said:

That combat is more than counting up weapons systems and laying out shooter-target SSPK matrices. That differences in doctrine, training, leadership, and motivation result in varying performance on the battlefield.

and the model should be able to provide useful analysis about the importance of say concentration at the point of attack, or artillery support; and the sensitivity of the results to key assumptions.

However, when I looked at reproducing some of the calculations I found:
- defining the battle (start, end, width, depth) is challenging eg a battalion attacking an entrenched company becomes a division vs brigade
- the opposing force strengths (and associated weapon systems) is dependent on the above, and the quality of records available
- the calculation of combat power from the number of weapon systems uses a large number of modifiers which IIRC can change the value for say tanks by a factor of 10 or more
- the model for totalling the different elements of combat power is opaque, non-linear and was derived essentially by trial and error
- estimation of results includes distance advanced and casualties which depend on the quality of data, and the judgemental mission success factor which you have commented on before

Therefore looking at the database as a whole there are a number of issues:
- non-randomness in selection of engagements to include
- potential inherent bias in the model especially in the calculation of combat power eg does it under or overvalue light machine guns; relative power of mortars vs artillery etc
- definition of forces involved eg do you count the whole division as engaged, if it was a brigade attack, as they were potentially available?
- basic data issues eg what was the actual strength of the units on the dates, what were actual casualties
- judgements on misssion success, especially when it would have been a challenging mission

Many of these issues will be less important if they are treated consistently across the whole database, which is difficult to demonstrate without effectively replicating the whole exercise. There are some known data issues with the original NPW data eg British casualties, German casualties treated below divisional level, German Corps artillery. I'm interested in seeing Christopher Lawrence's updated book to see how he deals with these points.

[Richard Anderson]

Many of these issues will be less important if they are treated consistently across the whole database, which is difficult to demonstrate without effectively replicating the whole exercise. There are some known data issues with the original NPW data eg British casualties, German casualties treated below divisional level, German Corps artillery. I'm interested in seeing Christopher Lawrence's updated book to see how he deals with these points.

I don't disagree, but I also do not agree with the "combat is unknowable" school, which thus insists on a know-nothing analysis based upon their gut...exemplified by Mansour, Hart, and others.

The problem with the naysayers is that all the analysis we did - much of it without using the QJM/TNDM - after Trevor died - twenty-three more years of work and refinement - have not significantly changed his original conclusions. And I suspect that Chris's analysis will show the same.

BTW, when I re-examined the Italian cases, not only did I find that the 56th Division casualties were overstated in many cases, I also found the same was true for German artillery strengths. Too often the original analysis used a cookie-cutter German "corps artillery" that simply did not exist.

[Tom from Cornwall]

Hi,

War by Numbers just arrived. With a whole chapter based on the Italian engagements. Not sure that any of our concerns have been addressed but not sure whether the detail is sufficient to allow a realistic debate. The bottom line is that the analysis suggests that the British performance was even worse than originally calculated. Hmmm, I wonder how willing Mr Lawrence would be to open up his analysis to serious debate. I'd love to understand the "judgement" used to calculate some of the less mathematical factors.

Tom

[Richard Anderson]

Hi,

War by Numbers just arrived. With a whole chapter based on the Italian engagements. Not sure that any of our concerns have been addressed but not sure whether the detail is sufficient to allow a realistic debate. The bottom line is that the analysis suggests that the British performance was even worse than originally calculated. Hmmm, I wonder how willing Mr Lawrence would be to open up his analysis to serious debate. I'd love to understand the "judgement" used to calculate some of the less mathematical factors.

Tom

I suggest you contact him via the TDI forum or through one of his blog posts on Mystics and Statistics.

[Tom from Cornwall]

Rich,

Thanks, will do.

Regards

Tom

[Tom from Cornwall]

One of the interesting statistics that I am beginning to identify is the number of sick that were being suffered by 10 Corps at Salerno. For example, after most of the fighting had died down the 10 Corps DDMS reported that:

Battle casualties for 24 hrs ending 1600 hrs today. Wounded offrs 3 O.Rs 39. Sick Offrs 1 ORs 117.
Evacuated by sea 181 incl 21 Rangers.

I'll see if I can dig down into the Fd Ambulance, etc, war diaries and see if they identify how many of these are due to diseases picked up in North Africa (especially Malaria I guess) and how many can be ascribed to the fighting (battle exhaustion, I think they termed it at the time).

I'll post up the previous days as well.

Regards

Tom

[Tom from Cornwall]

As discussed, direct from the entries in 10 Corps Signal Log from the Corps DDMS:

12 Sep 43
Battle casualties figures as follows:
Period ending 9/10 – offrs 25 ORs 424.
Period ending 10/11 – offrs 23 ORs 280.
Figures for period ending 11/12 NOT yet received.
Evacuated by hospital ship 11 Sep – 297.

14 Sep 43
Battle casualty figs for 24 hrs period ending 1600 hrs.
Offrs 8, O.R’s 282.
Evacuation by hospital ship today 248.

17 Sep 43
Battle casualties for 24 hrs ending 1600 hrs today.
Wounded offrs – 7, ORs – 275.
Sick offrs – 6, ORs – 109.
Evacuated by sea – NIL.

19 Sep 43
Battle casualties for 24 hrs ending 1600 hrs today.
Wounded: Offrs 7, ORs 127.
Sick: Offrs 4, ORs 106.
Evacuated by sea 140.

20 Sep 43
Battle casualties for 24 hrs ending 1600 hrs today.
Wounded offrs 3 O.Rs 39.
Sick Offrs 1 ORs 117.
Evacuated by sea 181 incl 21 Rangers.

21 Sep 43
Battle casualties for 24 hrs ending 1600 hrs today.
Wounded Offrs 5, ORs 93.
Sick Offrs 12, ORs 78.
Evacuated by sea 188.

Regards

Tom