Category DuWar Databases

Analysis of Force Ratios using the Campaign Data Base (CaDB) – fourth and final continuation

This is the fourth and final continuation of our previous four posts: Analysis for Force Ratios using the Campaign Data Base (CaDB) | Mystics & Statistics ( and Analysis for Force Ratios using the Campaign Data Base (CaDB) – continued | Mystics & Statistics ( and Analysis of Force Ratios using the Campaign Data Base (CaDB) – second continuation | Mystics & Statistics ( and Analysis of Force Ratios using the Campaign Data Base (CaDB) – third continuation | Mystics & Statistics (  It is a part of a briefing on forces ratios I will be giving at HADSS in UK: Schedule for HADSS 2024 | Mystics & Statistics ( and at HAAC near DC: Next Revised Schedule for the Third Historical Analysis Annual Conference (HAAC), 8 – 10 October 2024 | Mystics & Statistics (

All of this analysis of the CaDB was for a reason, it was to determine if odds (force ratios) play out difference at higher level of operations (meaning army level). Are they different at the operational level vice the tactical level of warfare. The answer appears to be no. I do not know of anyone who has actually specifically explored this issue before, so I am not sure there is an existing or countervailing opinions out there.

Of course, my real interesting in looking at this (which I did last year) was because of the war in Ukraine and the upcoming Ukranian spring/summer offensive in 2023. I did brief this at the Second HAAC (October 2023) and in Norway (November 2023). The question I had was does a minor advantage in force ratios or combat power ratios lead to a bigger advantage at the operational level of combat. The answer appears to be no, as this was reinforced by limited movement of the front line in Russo-Ukrainian War since the fall of 2022. 

My final slide in the briefing was “Does this relate to the fighting in Ukraine?” I then asked two questions:

  1. What are the odds?
    1. What is the strength of the deployed Ukrainian Army?
    2. What is the strength of the Russian Army deployed in Ukraine?
  2. What other advantages does the Ukrainian attacker have?
    1. Artillery
    2. Air Support? (Drones?)
    3. Observations/Intelligence
    4. Morale
    5. Training

Now, as it appears that Russia will be on the offensive this spring/summer, then I may need to restructure this slide and also add another point “artillery ammunition supply.”


I am probably going to do some more blog posts on this subject, looking at other levels of combat.


Analysis of Force Ratios using the Campaign Data Base (CaDB) – third continuation

This is a continuation of our previous three posts: Analysis for Force Ratios using the Campaign Data Base (CaDB) | Mystics & Statistics ( and Analysis for Force Ratios using the Campaign Data Base (CaDB) – continued | Mystics & Statistics ( and Analysis of Force Ratios using the Campaign Data Base (CaDB) – second continuation | Mystics & Statistics ( It is a part of a briefing on forces ratios I will be giving at HADSS in UK: Schedule for HADSS 2024 | Mystics & Statistics ( and at HAAC near DC: Next Revised Schedule for the Third Historical Analysis Annual Conference (HAAC), 8 – 10 October 2024 | Mystics & Statistics (

This is a continuation of Section IV of the briefing titled “What is necessary to have a good chance of generating a breakthrough?”

Having put together a table in the last post of force ratios and exchange ratios by outcome, I decided to take a moment to look at each of these cases. Each of these 94 cases is a fully mapped out campaign, many that you have heard of.

First looking at the 29 cases that were coded outcome IV (attacker advances). The average force ratios were 2.69-to-1 and the average exchange ratios were 1.51-to-1:

Force Ratio    Notes

0.58                 HUSKY – US Invasion of Sicily (39 days)

1.05                 HUSKY – UK Invasion of Sicily (39 days)

1.15                 Ardennes Allied Counteroffensive South II (15 days)

1.22                SHINGLE – Allied Landing at Anzio (10 days)

1.23                The West Bank 1967 (3 days)

1.34                 Ardennes Allied Counteroffensive South I (9 days)

1.38                 Graziani’s Advance (6 days)

1.44                 Moselle-Metz (6 days)

1.50                 Ardennes Allied Counteroffensive North (15 days)


1.75 to 1.98     3 cases

2.02 to 2.32     4 cases

2.51 to 2.92     6 cases

3.63 to 4.94     5 cases

6.04 to 10.00   2 cases


What I was really looking for is to see if there is any pattern in these low odds cases. Do they represent particularly odd or unusual cases? They really don’t. It does help to look at the cases though.

I then looked at those 21 cases that were coded as outcome five (defender penetrated). The average force ratios were 2.75-to-1 and the average exchange ratios were 0.64-to-1. There did not seem to be any unusual pattern, although there are a number of Arab-Israeli cases in these low odd penetrations. That is because human factors matter (morale, training, experience, leadership, motivation, etc.). In fact, they matter a lot (and are not considered in most U.S. DOD combat models). 

Force Ratio   Notes

0.78                The Cauldron: Battle of Gazala (21 days)

0.80                The Sinai, 1967 (5 days)

0.93                Golan Heights, 1967 (2 days)

1.01                BUFFALO: Anzio Breakout (9 days)

1.50                KADESH: Israeli Attack in the Sinai (8 days)

1.57                PO Valley Breakthrough (UK) (22 days)

1.67                Battle of Normandy, US Army (31 days)


1.82 to 1.93    2 cases

2.10 to 2.49    3 cases

2.52 to 2.92    2 cases

3.47 to 4.54    5 cases

6.58 to 7.01    2 cases


By the way, if someone is looking for some 3-to-1 rule in this data, good luck. Warfare is more complex than that.

One more post to come on this series of force ratios for army-level operations. Debating what I should discuss next.

Analysis of Force Ratios using the Campaign Data Base (CaDB) – second continuation

This is a continuation of our previous two posts: Analysis for Force Ratios using the Campaign Data Base (CaDB) | Mystics & Statistics ( and Analysis for Force Ratios using the Campaign Data Base (CaDB) – continued | Mystics & Statistics ( It is a part of a briefing on forces ratios I will be giving at HADSS in UK: Schedule for HADSS 2024 | Mystics & Statistics ( and at HAAC near DC: Next Revised Schedule for the Third Historical Analysis Annual Conference (HAAC), 8 – 10 October 2024 | Mystics & Statistics (

Section IV of the briefing is titled “What is necessary to have a good chance of generating a breakthrough?”

We coded some (94), but not all, of the 196 Army-level operations as to outcome. The outcomes are defined as (see War by Numbers for a more detailed description):

  • Outcome I is limited action
  • Outcome II is limited attack
  • Outcome III is failed attack
  • Outcome IV is attack advances
  • Outcome V is defender penetrated
  • Outcome VI is defender enveloped
  • Outcome VII is other.

These definitions are used to create the following table:

Outcome             I        II      III        IV       V       VI       VII

Cases                15        9     10        29      21      8          2

Force Ratios   1.88   3.35   1.80    2.69    2.75   1.86   8.50

Loss Ratios    3.77   1.56   1.66    1.51    0.64   0.05   0.01


Now, I put seven of those numbers in bold. They are worth looking at.

For those 10 operations that were coded as “failed attack”, the average force ratio is 1.80-to-1 while the average loss exchange ratio is 1.66-to-1 (i,e. the attacker lost more than the defender).

For those 29 operations that were coded as “attack advances”, the average force ratio is 2.69-to-1 while the average loss exchange ratio is 1.51-to-1.

For those 21 operations that were coded as “defender penetrated”, the average force ratio is 2.75-to-1 while the average loss exchange ratio is 0.64-to-1 (meaning the defender lost almost twice as many people as the attacker. Note that casualties included kill, wounded, missing and captured). 

One notices that the loss exchange ratio gets even more favorable in mop-up operations (defender enveloped). These are often the operation after “defender penetrated.”

A few other observations:

  1. Failed attacks tend to be lower average odds than successful ones (i.e. 1.80 versus 2.69 and 2.75).
  2. Attackers suffer higher losses than defenders until they are penetrated (1.61 and 1.51 versus 0.64)
  3. These are the same patterns as for division-level combat.

This last point is significant. Are operations with bodies of 60 thousand plus people the same as operations with 10-20 thousand people? At least in the patterns of force ratios required, loss exchange ratios, etc., they are very similar.

More to come (my briefings are long). The obvious next work step would be to finish coding the outcome of the other 102 operations in the CaDB. This is several man-weeks of effort. Not going to take that on now (I am trying to finish up another book).

Analysis for Force Ratios using the Campaign Data Base (CaDB) – continued

This is the continuation of our previous post: Analysis for Force Ratios using the Campaign Data Base (CaDB) | Mystics & Statistics (

In that post was a table showing the force and losses differences between battles won by the attacker, the defenders and those that are drawn. Below is a follow-up table, showing the force ratios for all the campaigns:

Force Ratio      Attacker wins   Defender wins *   Draws **   Notes

0.30                    1                                                                  Suomussalmi

0.52 to 0.73        6                         2

0.77 to 1.00        7                         5

1.01 to 1.25      14                         3                            1

1.27 to 1.50        8                         3                            1

1.55 to 1.75        9                         3

1.78 to 2.00       11                        5

2.02 to 2.50       10                        6                             2

2.51 to 2.92         8                                                       1 ****

3.01 to 4.00         8                      4 ***                       1 ****    Loos (3.97) – defender wins

4.02 to 4.94         8

5.79 to 7.33         5

10.00 to 11.21     2




* Removed from this seven engagements coded as “limited action” and “limited attack.” Their ratios were 0.58, 1.51, 2.90, 2.90, 3.58, 6.55, 12.38

** Removed from this 15 engagements coded as “limited action” and “limited attack.”

*** Three World War one engagements (Festubert at 3.01, Chemin des Dames at 3.33 and Loos at 3.97) and First Cassino (US) at 3.12.

**** Gothic Line Stalemate I at 2.58 and Gothic Line Statement II (US) at 3.08


These are slides 19 and 20 of my briefing. Now, I do not make conclusions on this slide in this briefing or even observations, but…. there are a few that could be made looking at this table. First, a three-to-one rule doesn’t really apply. Second, the defender never wins above four-to-one. Third, clearly there are a lot of factors included in these campaigns beyond simple manpower counts, and…. fourth…. you tell me?

The next slide of my briefing goes into the Section III of the briefing:  “Influence of Human Factors on Combat.” This is all drawn from War by Numbers… so… read the book. I will skip that and my next post will pick up at Section IV of the briefing “What is necessary to have a good chance of generating a breakthrough.” Probably do that post next Tuesday.

Analysis for Force Ratios using the Campaign Data Base (CaDB)

We have not made much use of our Campaign Data Base. (See: The History of the DuWar Data Bases | Mystics & Statistics ( We used it as part of the Enemy Prisoner of War (EPW) studies back in 2000-2001 and have not made use it in the last two decades. But, for a presentation I did last year on force ratios, I blew the dust off of it because I wanted to see if force ratios were different for army-level operations than for division-level engagements. I mean, in the ETO data we have (116 cases), in the force ratios ranging between 1.15-to-1 to 1.88-to1 the attacker won 79% of the time (so much for needing 3-to-1). See: The 3-to-1 rule and the War in Ukraine | Mystics & Statistics ( So the question became, is the pattern we see at army-level different than division-level?

The Campaign Data Base consists of 196 campaigns from 1905 to 1991. They from two days in length to 155 days in length. Only three were over 60 days in length. The problem is that the database is not complete. We assembled it, used it once and have not used it again. There are some holes. For example, we only had the starting strength ratios calculated for 163 cases, we only had the total casualty ratios calculated for 162 and only had the winner calculated for 156 cases. In most cases the missing data is available but has not been assembled. The database just needs a little tender loving care. 

The average attacker strength (99 cases) was 188,909. The average defender strength (96 cases) was 95,497. This comes out to a 1.98-to-1 ratio.

The average attacker losses (176 cases) was 36,076. The average defender losses (172 case) was 47,004. This comes out to a 1-to-1.30 ratio.

The average attacker percent losses per day (163 cases) was 0.69%. The average defender percent losses per day (162 cases) was 1.85%. This comes out to a 1-to-2.68 ratio.

The starting strength ratio (163 cases) was 2.24 (2.24-to-1). The total casualty ratio was (164 cases) 1.35-to-1.

Now, the holes in the database become an issue. This are holes that can be filled given time (read: budget). We have 97 cases where the attacker is coded as the winner, and 38 cases where the defender wins. We have draws in 21 other cases. The rest (40 cases) are currently not coded.

Anyhow, this all produces the following table:

                                                   Attacker   Defender   Draw 

Av. Attacker Strength               208,835    156,821     171,312

Av. Defender Strength                91,486    100,729       96,582

       Ratio                                   2.28           1.56           1.77


Av. Attacker Losses                    34,630      69,098       15,232

Av. Defender Losses                   52,466      64,271       12,632

      Ratio                                     0.66           1.08           1.21


Av. Attacker % per day              0.73           0.98           0.32

Av. Defender % per day             2.59           0.98           0.39

      Ratio                                      0.28          1.00            0.82


Starting Strength Ratio              2.42          2.24            1.79

Casualty Ratio                            1.04          2.51            1.22


Contemplate for a moment what this data is telling you. A few observations:

  1. There is a difference in force ratios between winning and losing engagements (2.28-to-1 vice 1.56-to-1).
  2. There is a difference in casualties between winning and losing engagements (0.66-to-1 vice 1.08-to-1).
  3. The data for these army-level operations does not look significant different than for a division-level operation. This is significant.

I will stop here for a moment. This is from slides 12 – 18 for my force ratios briefing. There is more to come (because my briefings, like some of my books, are never short).


The 3-to-1 rule and the War in Ukraine

There is a 3-to-1 rule that some people quote from somewhere. We have discussed this before: Trevor Dupuy and the 3-1 Rule | Mystics & Statistics ( and The 3-to-1 Rule in Histories | Mystics & Statistics ( and The 3-to-1 Rule in Recent History Books | Mystics & Statistics (

Trevor Dupuy’s argument was always that it took a combat power advantage to advance (attack successfully). This combat power calculations considers weapons, terrain, posture, air support, human factors, etc. Because of the current artillery shell shortages for the Ukrainian Army, logistics may also be a factor.

This combat power advantage often happens at 1.5-to-1 or 2-to-1. Usually is happens by around 2-to-1 (my conclusions – see War by Numbers). For example, here is my chart of force ratios for division-level combat in the European Theater of Operation (ETO) in 1944 from page 10 of War by Numbers:


0.55 TO 1.01-TO-1.00…………ATTACK FAILS………………………….100……………………………………5

1.15 TO 1.88-TO-1.00…………ATTACK USUALLY SUCCEEDS………21…………………………………..48

1.95 TO 2.56-TO-1.00…………ATTACK USUALLY SUCCEEDS………10…………………………………..21

2.71 TO 1.00 AND HIGHER….ATTACK ADVANCES……………………..0…………………………………..42


Notice that the attacker succeeds at force ratios between 1.15-to-1 to 1.88-to-1 in 79% of the 48 cases of division-level combat. It gets better from there. The book also has force ratios from other theaters and campaigns. Some of this has been discussed here before: More Combat Results Tables from War by Numbers | Mystics & Statistics ( and Force Ratios at Kharkov and Kursk, 1943 | Mystics & Statistics ( and Force Ratios in the Arab-Israeli Wars (1956-1973) | Mystics & Statistics (

A rigidly defined 3-to-1 rule tends to create an officer corps of McLellan’s. This rule-of-thumb is doing more damage than good as constructed.

What got my attention is that some people are trying to apply some 3-to-1 rule in Ukraine, and then come to the conclusion that one or the other side cannot advance because they don’t have a 3-to-1 force ratio. Yet, people have been advancing. In fall of 2022 Ukraine re-took Kherson and surrounding areas (see: 2022 Kherson counteroffensive – Wikipedia) and achieved a breakthrough at Balakliya that took back a significant portion of Donetsk province (see: Battle of Balakliia – Wikipedia) and conducted a successful offensive around Kharkiv (see: 2022 Kharkiv counteroffensive – Wikipedia). In 2023 Russia did advance on Bakhmut and took it (see: Battle of Bakhmut – Wikipedia) and in 2023/2024 Russia did advance on Avdiivka and took it (see: Battle of Avdiivka (2023–2024) – Wikipedia). I think in three for those five cases the attacker did not have anything approaching a 3-to-1 advantage. Of course, I have no reliable manpower statistics for either side in any of these five battles, so this is sort of a guess, as is most of the analysis and expert opinions on this war. 

I do not know how many troops Ukraine currently has. I am guessing at least 300,000 deployed. Some people throw out figures in the 600-700,000 range. I have no idea if that are total mobilized estimates or total deployed estimates. The same with Russia, where figures of 600-700,000 are also thrown out, but not sure that is what is actually deployed in Ukraine. I am guessing some number closer to 300,000. Don’t really know, and don’t know who does for certain (see the “Force Involved’ section of this post: The Russo-Ukrainian War – Day 699 | Mystics & Statistics (

Anyhow, I gather the two sides are somewhere near parity in force size. They can certainly concentrate forces to get a local advantage. With current modern intelligence gathering capabilities, concentrating forces is often seen while it is happening and opposing side can respond promptly. So not sure where anyone can get their 3-to-1 advantage.

I did do a test recently, comparing the force ratios in a database over 700 division-level combat engagements to the force-ratios in over 100 Army-level operations. The question was whether force ratios and the success from those force ratios was different at division-level vice army-level. My tentative conclusions were that force ratios for army level campaigns had the “Same patterns as for division-level combat.”

Now, I have not written this effort up. I did brief it last year at the Second HAAC and did brief it in Norway. I will be briefing it again on Thursday, July 11 at HADSS in York (see:  Historical Analysis for Defence and Security Symposiums (HADSS), 8 – 11 July in York, England | Mystics & Statistics ( and for one last time at the Third HAAC (see: Revised Schedule for the Third Historical Analysis Annual Conference (HAAC), 8-10 October 2024 | Mystics & Statistics ( After that, I may write it up, either as a blog post or as a chapter in a book called More War By Numbers, which will probably be delayed until 2026 (see: Current book release schedule | Mystics & Statistics (, which I probably need to update).

Anyhow, the point is, anyone doing analysis for the situation in Ukraine based upon some 3-to-1 rule probably needs to reconsider their analysis.

Measuring Unit Effectiveness in Italy

We are in discussion over revisiting the measurement of combat effectiveness of select units in Italy 1943-1945. This was done by Trevor Dupuy in Numbers, Predictions and Wars (1977) by division using the QJM (Quantified Judgment Model) and was done in aggregate by me in War by Numbers (2017) using simply comparative statistics. If you feel lifeless reading blogs like this, you can rest for a bit through sites such as 홈카지노.

For a little background on page 115 of Understanding War is a chart of German, UK and U.S. units in the Italian Campaign and their CEVs (Combat Effectiveness Values). Their values range from 0.60 to 1.49. The German Hermann Goering Division is the highest rated division at 1.49. This is based upon five engagements. The German 3rd PzGrD was rated 1.17 based upon 17 engagements and 15th PzGrD was rated 1.12 based upon 11 engagements. This was done using the QJM.
    For reference, I would recommend reading the following four books:
1. Understanding War
2. War by Numbers
3. Attrition (optional)
4. Numbers, Predictions and War (optional)
There are two ways to measure combat effectiveness. 1) Do a model run and compared the results of the model run to historical data. This requires 1) a historically validated combat model (there are very few), and 2) confidence in the model. 2) The other option is to do a statistical comparison of a large number of engagements. This is what I did in Chapters 5, 6 and 7 of War by Numbers.
One can measure combat effectiveness by three means: 1) Casualty effectiveness, 2) special effectiveness (distance opposed advance) or 3) Mission effectiveness. This is all discussed in Trevor Dupuy’s work and in War by Numbers.
To date, the only people I am aware of who have published their analysis of combat effectiveness is Trevor Dupuy, me (Chris Lawrence) and Niklas Zetterling. See: CEV Calculations in Italy, 1943 | Mystics & Statistics ( and his book Normandy 1944 (recently revised and republished). There is also a six-volume quantitative effort related to Operation Barbarossa by Nigel Askey, which I have never looked at. Everyone else has ignored quantifying this issue, although there are no shortage of people claiming units are good, bad or elite. How they determine this is judgment (and it is often uncertain as to what the basis is for this judgment).
Now, the original work on this was done by Trevor Dupuy in the late 1970s based upon his data collection and the QJM. Since that time the model has been updated to the TNDM. The engagements used for the QJM validation were then simplified (especially in weapons counts) and assembled into the LWDB (Land Warfare Data Base). The LWDB had around 70 engagements from the Italian Campaign. Since that time we have created the DuWar series of databases which includes the DLEDB (Division-Level Engagement Data Base). See: The History of the DuWar Data Bases | Mystics & Statistics ( We have doubled the number of Italian Campaign engagements to around 140.
There are a total of 141 Italian Campaign division-level engagements in the DLEDB. The first 140 engagements cover from September 1943 to early June 1944. There is almost 12 months of war not covered and not all units in the first part of the campaign are covered. With all the various nationalities involved (i.e German, Italian, U.S., UK, Free French, Moroccan, New Zealand, South African, Poland, Indian, Canadian, Brazilian, Greek, etc.), the Italian Campaign is a fertile field for this work. We are looking at stepping back into this. 
Units involved in engagements in the DELDB:
3rd PzGrD: 25 cases
15th PzGrD: 39 cases
16th PzD: 7 cases
26th PzD: 8 cases
29 PzGrD: 6 cases
65th ID: 5 cases
94th ID: 8 cases
305th ID: 4 cases
362nd ID: 3 cases
715th ID: 2 cases
4th Para D: 3 cases
HG PzGrD: 26 cases
LXXVI Pz Corps: 4 cases
12th Para Rgt: 1 case
1st AD: 3 cases
3rd ID: 19 cases
34th ID: 15 cases
36th ID: 12 cases
45th ID: 20 cases
85th ID: 7 cases
88th ID: 4 cases
509th PIB: 1 case
1st SSF: 1 case
7th AD: 6 cases
1st ID: 9 cases
5th ID: 2 cases
46th ID: 18 cases
56th ID: 24 cases

The History of the DuWar Data Bases

The original databases of battles was developed by Trevor Dupuy and HERO (Historical Evaluation and Research Organization) back in the 1980s. They were published in a six volume work in 1983 as the HERO Land Warfare Data Base. This is back in the days when a data base did not have to be computerized (paper database – how quaint) and database was two words. It is report number 95 listed here: TDI – The Dupuy Institute Publications. Descriptive link is here: Analysis of Factors that have Influenced the Outcomes of Battles and Wars ( Of significance, there is a detailed description of each engagement in these paper reports. It was republished in 1984, 1985 and 1986 as report numbers 100, 103 and 111 here: TDI – The Dupuy Institute Publications. The final publication named the database as CHASE. 

This effort was funded by CAA and was before my time. I came to work for HERO in 1987. There was then some back and forth between CAA, where HERO and CAA got to fighting over details of the content. One analyst at CAA sent 16 engagements out for comment. I did analyze that effort, although that file is now buried on an old Word Perfect DOS-era disk. He had four outside independent historians each analyze four engagements. The end result is the comments made corrections/improvements to 25% of the engagements, the comments did really did not change anything in 25% of the engagements, and the comments actually, if implemented, would have added error the engagements in 50% of the cases. This is fairly typical of outside comments, with 1-out-of-3 or 1-out-of-4 being helpful, and half of them would degrade the product. At that point, the project came to a griding halt, with much animosity between the arguing parties.

Then both HERO and CAA decided to independently computerize their databases. HERO added about four new engagements to their database, maybe corrected a few others, and the programmed it in a flat file called Reflex. It was 603 engagements (working off memory here) and called the LWDB (Land Warfare Data Base). CAA decided to computerize its version of 598 or 599 engagements and it was called the CHASE database. This became the CBD-90 that some people are still using. Neither of these versions included the extensive battle narratives as databases at that time could not handle large text files.

The computerized Reflex version of the LWDB was later purchased by Oak Ridge National Laboratories and published in the book by Dr. Dean Harley. It is a better version than the CBD-90. I did review the CBD-90 over twenty years ago. In the original database, there were a series of factors that were coded as to what degree they influenced the battle. In the CBD-90 about one-third of those factors (or one-third of the engagements that had those factors) – they were blanked out or mis-coded. It was a simple coding error, that as far as I know has never been corrected. 

In the meantime, around 1995 I decided we needed to reorganize and reprogram the database. We had a new database created by Jay Karamales in Access. It included text files. We loaded the old Reflex engagements in the database and then Susan Rich and I proofed the entire database back to the paper copies. Susan Rich then entered in all the narratives into the database. So this was now a complete and proofed version of the 1986 paper database. 

I then broke the database up. One of the problems with the original database is that it has engagements from 1600 next to engagements from 1973 next to a series of day-long division-level engagements from WWII next to some six-month long army-level engagements from the Great War next to battalion-level actions. While there are definitely some historical trends across all these, in some cases, depending on what you are analyzing, it is comparing apples to oranges. So, I took at mostly one-day battles from 1600-1900 and put them in a separate database (243 engagements – the  BaDB. I took all the large army-level engagements (like Battle of Verdun, Battle of the Somme) and put them into a Large Action Data Base – LADB. Basically, moved them out of the way. They were later used in part to help create the CaDB (Campaign Data Base). I put the smaller battalion-sized engagements into a separate battalion-level data base (BLODB). They left us with a core of around 300 engagements in a division-level database, mostly of 1-day engagements. All this work was done outside and independent of any contracted effort and therefore became a Dupuy Institute proprietary product. As with any proprietary product, you have to protect it.

We then expanded all these databases. In the case of the division-level database (the DLEDB), we ended up doing a series of studies for CAA on Enemy Prisoner of War capture rates in 1998-2001. We coded the division-level engagements by outcome and then using that to analyze capture rates based upon the outcomes of the battle. This effort included getting counts of the number captured and the number of deserters in each engagement. This is reports E-1 to E-8 here:  TDI – The Dupuy Institute Publications. The data used (but not the complete listing of the engagement) was included in appendices to these reports. CAA and the U.S. Army is still using these new rates.

We also added engagements to it from our urban warfare studies (CAA), reports U-1 to U-3. We used the database to analyze the urban versus non-urban combat. It was during that study we added engagements from the Channel Ports, Aachen and the three battles of Kharkov (1943). This study is discussed in two chapters in my book War by Numbers. We also took the time and put in 192 engagements from the Battle of Kursk (1943) based upon our work on the Kursk Data Base. All these Kursk engagements are listed (abbreviated) in my big Kusk: The Battle of Prokhorovka book. We also did a study on situational awareness for OSD Net Assessment (Andy Marshall’s old office). This is report SA-1 and also two chapters in my book War by Numbers. We ended up coding 295 division-level engagements based upon their knowledge of the enemy (by reviewing their intel reports of the divisions involved). We then reviewed what was the measurable combat advantage of improved situation awareness based upon real-world combat data. So, as in the EPW study, we took our original database and added additional filled-in fields so as to be able to do properly analyze the issue. This last expansion of the database was completed in 2004.

At that point, the division-level database had 752 cases in it. We had done some additional work on the old Italian Campaign engagements to clean them up and revise them. In particular Richard Anderson collected UK records from PRO and we cross-checked and revised all the UK engagements in the database and expanded the number of Italian Campaign engagements from about 70 to around 140. We then stopped work on the database in 2004.

During that time, we also expanded the battalion-level database to around 200 actions. We also had created a Campaign Data Base as part of our work, to examine operations above division-level and that last more than a few days. This was recently used for my presentation on Force Ratios that I gave at the second HAAC and in Norway in early November. See: The Schedule for the Second Historical Analysis Annual Conference (HAAC), 17 – 19 October 2023 | Mystics & Statistics ( In 2010 we created a small draft company-level database under contract with Boeing of 100 cases. A listing of most of these databases is here: TDI – The Dupuy Institute Publications. It does not include the company-level database, the Battle of Britain database nor the Dupuy Insurgency Spread Sheets (DISS) as we have not updated that page.

Obviously, people are going to ask: how can they get access to these databases. The answer is that you cannot until someone is willing to purchase them at a price that I willing to release them for. With the internet any single sale of the database will result in the release of the entire database to the world. So, any price would have to address the fact that these powerful and unique databases, which are proprietary to The Dupuy Institute, would be shared with the world. This includes potential business competitors. We still rely on contracts for our funding and these databases are part of our “product.” So, cost of giving away an exclusive competitive advantage? We would be willing to sell them to an organization if the price is right and they could then be publicly released. So far no one has made a significant concrete offer to us.


So other links:

Some Background on TDI Data Bases | Mystics & Statistics (

Dupuy Institute Data Bases | Mystics & Statistics

Cost of Creating a Data Base | Mystics & Statistics (

The Division Level Engagement Data Base (DLEDB) | Mystics & Statistics (

Battalion and Company Level Data Bases | Mystics & Statistics (

Other TDI Data Bases | Mystics & Statistics (

Using the DLEDB:

Average Losses per Day in Division-level Engagements on the Eastern Front in 1943 | Mystics & Statistics (

Density of Deployment in Ukraine | Mystics & Statistics (

The U.S. Army Three-to-One Rule versus the 752 Case Division-level Data Base 1904-1991 | Mystics & Statistics (

Comparing Force Ratios to Casualty Exchange Ratios | Mystics & Statistics (

Comparing the RAND Version of the 3:1 Rule to Real-World Data | Mystics & Statistics (

Summation of Force Ratio Posts | Mystics & Statistics (

Amphitheater, 9 – 11 September 1943 | Mystics & Statistics (

Amphibious and River Crossing Engagements in the Italian Campaign 1943-44 | Mystics & Statistics (

The World War I Cases from the Division-level Database | Mystics & Statistics (

The World War II Cases from the Division-level Database | Mystics & Statistics (

Post-World War II Cases from the Division-level Database | Mystics & Statistics (

Force Ratios in the Arab-Israeli Wars (1956-1973) | Mystics & Statistics (

Other discussion:

Battles versus Campaigns (for Validation) | Mystics & Statistics (

Validation Data Bases Available (Ardennes) | Mystics & Statistics (

Validation Data Bases Available (Kursk) | Mystics & Statistics (

Other Validation Data Bases | Mystics & Statistics (

The Use of the Two Campaign Data Bases | Mystics & Statistics (

Measuring the Effects of Combat in Cities, Phase II – part 1 | Mystics & Statistics (

Presentations from HAAC – Urban Warfare | Mystics & Statistics (

The Battle of Britain Data Base | Mystics & Statistics (

Presentations from HAAC – Data for Wargames | Mystics & Statistics (

The U.S. Army Three-to-One Rule versus 243 Battles 1600-1900 | Mystics & Statistics (

The U.S. Army Three-to-One Rule versus 49 U.S. Civil War battles | Mystics & Statistics (

Using the CBD:

The Key to Victory: Machine Learning the Lessons of History | Mystics & Statistics (

Presentations from HAAC – Machine Learning the Lessons of History | Mystics & Statistics (

There is more….

Phalanx Article: What We Have Learned from Doing Historical Analysis | Mystics & Statistics (

Advance Rates in Combat


Advance Rates in Combat:

                Units maneuver before and during a battle to achieve a more favorable position. This maneuver is often unopposed and is not the subject of this discussion. Unopposed movement before combat is often quite fast, although often not as fast as people would like to assume. Once engaged with an opposing force, the front line between them also moves, usually moving forwards if the attacker is winning and moving backwards for the defender if he is losing or choosing to withdraw. These are opposed advance rates. This section is focused on discussing opposed advance rates or “advance rates in combat.”

            The operations research and combat modeling community have often taken a short-hand step of predicting advance rates in combat based upon force ratios, so that a force with a three-to-one force ratio advances faster than a force with a two-to-one force ratio. But, there is not a direct relationship between force ratios and advance rates. There is an indirect relationship between them, in that higher forces ratios increased the chances of winning, and winning the combat and the degree of victory helps increase advance rates. There is little analytical work that has been done on this subject.[1]

            Opposed advance rates are very much influenced by 1) terrain, 2) weather and 3) the degree of mechanization and mobilization, in addition to 4) the degree of enemy opposition. These four factors all influence what the rates will be.

            In a study The Dupuy Institute did on enemy prisoner of war capture rates, we ended up coding a series of engagements by outcome. This has proven to a useful coding for the examination of advance rates. Engagements codes as outcomes I and II (limited action and limited attack) are not of concern for this discussion. The engagement coded as attack fails (outcome III) is significant, as these are cases where the attacker is determined to have failed. As such they often do not advance at all, sometimes have a very limited advance and sometimes are even pushed back (have a negative advance). For example, in our work on the subject, of our 271 division-level engagements from Western Europe 1943-45 the average advance rate was 1.81 kilometers per day. For Eastern Europe in 1943 the average advance rate was 4.54 kilometers per day based upon 173 division-level engagements.[2] These advance rates are irrespective of what the force ratios are for an engagement.

            In contrast, in those engagements where the attacker is determined to have won and is coded as attacker advances (outcome IV) the attacker advances an average of 2.00 kilometers in the 142 engagements from Western Europe 1943-45. The average force ratio of these engagements was 2.17. In the case of Eastern Europe in 1943, the average advance rate was 5.80 kilometers based upon 73 engagements. The average force ratio of these engagements was 1.62.

            We also coded engagements where the defender was penetrated (outcome V). These are those cases where the attacker penetrated the main defensive line of the defending unit, forcing them to either withdraw, reposition or counterattack. This penetration is achieved by either overwhelming combat power, the end result of an extended operation that finally pushes through the defenses, or a gap in the defensive line usually as a result of a mistake. Superior mechanization or mobility for the attacker can also make a difference. In those engagements where the defender was determined to have been penetrated the attacker advanced an average of 4.12 kilometers in 34 engagements from Western Europe 1943-45. The average force ratio of these engagements was 2.31. In the case of Eastern Europe in 1943, the average advance rates was 11.28 kilometers based upon 19 engagements. The average force ratio of these engagements was 1.99.

            This clearly shows the difference in advance rate based upon outcome. It is only related to force ratios to the extant the force ratios are related to producing these different outcomes.


            Also of significance is terrain and weather. Needless to say, significant blocking obstacles like bodies of water, can halt an advance and various rivers and creeks often considerably slow them, even with engineering and bridging support. Rugged terrain is more difficult to advance through and easier to defend and delay then smoother terrain. Closed or wooded terrain is more difficult to advance through and easier to defend and delay then open terrain. Urban terrain tends to also slow down advance rates, being effectively “closed terrain.” If it is raining then advance rates are slower than in clear weather. Sometimes considerably slower in heavy rain. The season it is, which does influence the amount of daylight, also affects the advance rate. Units move faster in daylight than in darkness. This is all heavily influenced by the road network and the number of roads in the area of advance.

            No systematic study of advance rates has been done by the operations research community. Probably the most developed discussion of the subject was the material assembled for the combat models developed by Trevor Dupuy. This included addressing the effects of terrain and weather and road network on the advance rates. A combat model is an imperfect theory of combat.

            Even though this combat modeling effort is far from perfect and fundamentally based upon quantifying factors derived by professional judgment, tables derived from this modeling effort have become standard presentations in a couple of U.S. Army and USMC planning and reference manuals. This includes U.S. Army Staff Reference Guide and the Marine Corps’ MAGTF Planner’s Reference Manual.[3]

The original table, from Numbers, Predictions and War, is here:[4]




                                                                                    Rates in km/day

                                                Armored          Mechzd.          Infantry           Horse Cavalry

                                                Division           Division           Division           Division or

                                                                                                or Force           Force

Against Intense Resistance

    (P/P: 1.0-1.1O)

Hasty defense/delay                4.0                   4.0                   4.0                   3.0

Prepared defense                    2.0                   2.0                   2.0                   1.6

Fortified defense                     1.0                   1.0                   1.0                   0.6


 Against Strong/Intense Resistance

    (P/P: 1-11-125)

Hasty defense/delay                5.0                   4.5                   4.5                   3.5

Prepared defense                    2.25                 2.25                 2.25                 1.5

Fortified defense                     1.25                 1.25                 1.25                 0.7


Against Strong Defense

    (P/P: 1.26-1.45)

Hasty defense/delay                6.0                   5.0                   5.0                   4.0

Prepared defense                    2.5                   2.5                   2.5                   2.0

Fortified defense                     1.5                   1.5                   1.5                   0.8


Against Moderate/Strong Resistance

    (P/P: 1.46-1.75)

Hasty defense                         9.0                   7.5                   6.5                   6.0

Prepared defense                    4.0                   3.5                   3.0                   2.5

Fortified defense                     2.0                   2.0                   1.75                 0.9


Against Moderate Resistance

    (P/P: 1.76-225)

Hasty defense/delay                12.0                 10.0                 8.0                   8.0

Prepared defense                    6.0                   5.0                   4.0                   3.0

Fortified defense                     3.0                   2.5                   2.0                   1.0


Against Slight/Moderate Resistance


Hasty defense/delay                16.0                 13.0                 10.0                 12.0

Prepared defense                    8.0                   7.0                   5.0                   6.0

Fortified defense                     4.0                   3.0                   2.5                   2.0


Against Slight Resistance

    (P/P: 3.01-4.25)

Hasty defense/delay                20.0                 16.0                 12.0                 15.0

Prepared defense                    10.0                 8.0                   6.0                   7.0

Fortified defense                     5.0                   4.0                   3.0                   4.0


Against Negligible/Slight Resistance


Hasty defense/delay                40.0                 30.0                 18.0                 28.0

Prepared defense                    20.0                 16.0                 10.0                 14.0

Fortified defense                     10.0                 8.0                   6.0                   7.0


Against Negligible Resistance

    (P/P: 6.00 plus)

Hasty defense /delay               60.0                 48.0                 24.0                 40.0

Prepared/fortified defense      30.0                 24.0                 12.0                 12.0


*Based on HERO studies: ORALFORE, Barrier Effectiveness, and Combat Data Subscription Service.

** For armored and mechanized infantry divisions, these rates can be sustained for 10 days only; for the next 20 days standard rates for armored and mechanized infantry forces cannot exceed half these rates.


                This is a modeling construct built from historical data. These are “unmodified” rates. The modifications include: 1) General Terrain Factors (ranging from 0.4 to 1.05 for Infantry (combined arms) Force and from 0.2 to 1.0 for Cavalry or Armored Force, 2) Road Quality Factors (addressing Road Quality from 0.6 to 1.0 and Road Density from 0.6 to 1.0), 3) Obstacles Factors (ranging from 0.5 to 0.9 for both a River or steam and for Minefields), 4) Day/Night with night advance rate one-half of daytime advance rate and 5) Main Effort Factor (ranging from 1.0 to 1.2). These last five sets of tables are not shown here, but can be found in his writings.[5]



[1] The most significant works we are aware of is Trevor Dupuy’s ORALFORE study in 1972: Opposed Rates of Advance in Large Forces in Europe (ORALFORE), (TNDA, for DCSOPS, 1972); Trevor Dupuy’s 1979 book Numbers, Predictions and War; and a series of three papers by Robert Helmbold (Center for Army Analysis): “Rates of Advance in Land Combat Operations, June 1990,” “Survey of Past Work on Rates of Advance, and “A Compilation of Data on Rates of Advance.”

[2] See paper on the subject by Christopher A. Lawrence, “Advance Rates in Combat based upon Outcome,” posted on the blog Mystics & Statistic, April 2023. In the databases, there were 282 Western Europe engagements from September 1943 to January 1945. There were 256 Eastern Front engagements from February, March, July and August of 1943.

[3] See U.S. Army Staff Reference Guide, Volume I: Unclassified Resources, December 2020, ATP 5-0.2-1, pages xi and 220; and MAGTF Planner’s Reference Manual, MSTF pamphlet 5-0.3, October 2010, page 79. Both manuals include a table for division-level advances which is derived from Trevor Dupuy’s work, and both manuals contain a table for brigade-level and below advances which are calculated per hour that appear to also be derived from Trevor Dupuy’s division-level table. The U.S. Army manual gives the “brigade and below” advance rates in km/hr while the USMC manual, which appears to be the same table, gives the “brigade and below” advance rates in km/day. This appears to be a typo.

[4] Numbers, Predictions and War, pages 213-214. The sixth line of numbers, three numbers were changes from 1.85 to 1.25 as this was obviously a typo in the original.

[5] See Numbers, Predictions and War, pages 214-216.



The actual paper this was drawn from is here: Advance Rates in Combat

Average Losses per Day in Division-level Engagements on the Eastern Front in 1943

Trevor N. Dupuy, among his 56 verities of combat, states that “Average World War II division engagement casualty rates were 1-3% a day.”[1]

This was based primarily on his research on the Western Front during World War II. For example, just to draw from data from real world experience, the average losses per U.S. division in 82 selected engagements was 1.2% per day in 1943-44. The average strength of these divisions was 14,000. The average loss per German division in 82 selected engagements was 1.8% per day. The average strength of these divisions was 12,000. These engagements were all from the Italian Campaign and the European Theater of Operations (primarily France).[2]

Now for Germany versus the Soviet Union, the loss rates in 1943 were higher for both sides. We do have daily unit records and have assembled them into a series of 192 division-level engagements for the southern part of Battle of Kursk in July 1943 and 64 division-level engagements for the battles around Kharkov in February, March and August of 1943. They show the following statistics:[3]

Battle of Kursk:

                                                            Average Losses:            Average           Average

                                             Cases     Mean     Median             Strength          Force Ratio

Germans attacking               124         0.99        0.78                21,487               1.44

Germans defending                68         0.68        0.52               16,945                0.91

Soviets attacking                    68         3.25        1.67               18,631                1.10

Soviets defending                 124         4.31        3.82               14,930                0.69


Battles for Kharkov:

Germans attacking                 35         0.58        0.48                17,326                2.77

Germans defending                29         0.64        0.50                14,834                0.87

Soviets attacking                    29         2.18        1.56                 17,001               1.15

Soviets defending                   35         5.21        3.05                  6,837                0.36


Slightly different figures will be created using differing selection criteria, but out of the 124 cases of the Germans attacking at Kursk, in only two cases were German losses greater than 3% [4]. They were both cross-river attacks done on 5 July 1943 by the 106th and 320th Infantry Divisions. German losses at Kursk while defending never exceeded 3%. German losses in the Kharkov engagements never exceeded 2% a day.

Soviet losses exceeded 3% per day in 24 cases while attacking at Kursk and exceeded 6% in ten of those cases. Soviet losses exceeded 3% per day in 67 cases while defending at Kursk (in over half the cases) and exceeded 6% in 39 of those cases. Soviet losses exceeded 3% a day in only two cases while attacking at Kharkov and in 16 cases while defending, of which in seven of those cases Soviet losses exceeded 6% per day.



[1] See Colonel Trevor N. Dupuy, Understanding War: History and Theory of Combat (Paragon House Publishers, New York, 1987), page 179.

[2] See Dupuy, Understanding War, page 169. Note that all these WWII engagements were tagged with the note that the data was approximate, more research required. The Dupuy Institute has 282 division-level engagements from the Italian Campaign and ETO that are created from the unit records of both sides. We have not done this comparison using our further developed and more extensive data collection, but suspect the results would be similar.

[3] The data used for this calculation is presented for the Battle of Kursk in a series of 192 engagement sheets in the book by Lawrence, Kursk: The Battle of Prokhorova. This work can be cross-checked by others. The data used for the battles around Kharkov have not been published yet. It might be at some point in the future. The data is currently company proprietary of The Dupuy Institute.

[4] More precise would be to remove all the engagements coded as limited action and limited attack, leaving only those coded as failed attack, attack advances, defender penetrated, defender enveloped and other. In the 124 Kursk cases of the German attacking this would remove 15 cases of limited action, 14 cases of limited attack, and 26 cases where the outcome has not been coded yet. The force ratio is now up to 1.56-to-1 and the average German percent losses are 1.25% while the average Soviet percent losses are 5.83%. Conversely, in the 68 cases where the Soviet are attacking, there are 7 cases are limited action, 9 cases are limited attack and 33 cases where the outcome has not been coded yet. The force ratio for these remaining 19 cases is 1.27 and average Soviet percent losses are 4.05 while the average German percent losses are 0.86.

 In all cases, the mean is calculated as a weighted mean, meaning that it is based upon total strengths compared to total losses. The median is calculated, naturally, by finding the midpoint of all 124 or 68 engagements.


The actual paper this was drawn from is here: Average Losses per Day