Tag Prediction

Forecasting the 1990-1991 Gulf War

DoD photo by Regina Ali
DoD photo by Regina Ali

In my last post on the subject of prediction in security studies, I mentioned that TDI has a public forecasting track record. The first of these, and possibly the most well know, involves the 1990-1991 Gulf War.

On 13 December 1990, Trevor N. Dupuy, President of Trevor N. Dupuy & Associates (TNDA), testified before the House Armed Services Committee on the topic of the looming military confrontation between the military forces of the United States and United Nations Coalition allies and those of Iraq.[1] He offered TNDA’s assessment of the potential character of the forthcoming conflict, as well as estimates of the likely casualties that both sides would suffer. Dupuy published a refined and expanded version of TNDA’s analysis in January 1991.[2]

Based on a methodology derived from Dupuy’s combat models and synthesized data on historical personnel and material combat attrition, TNDA forecast a successful U.S. and Coalition air/ground offensive campaign into Kuwait.[3] Using publicly available sources, TNDA calculated that Iraqi forces in Iraq numbered 480,000, U.S. forces at 310,000, and Coalition allies at 125,000.

The estimated number of casualties varied based on a campaign anticipated to last from 10 to 40 days depending on five projected alternate operational scenarios:

  • Operation “Colorado Springs.” A 10-day air campaign aimed at achieving air superiority and attacking Iraq’s ground forces and war-making infrastructure. While TNDA believed an air campaign would proceed any ground offensive option, Dupuy suggested that it could potentially force an Iraqi surrender without the need for a land attack.
  • Operation “Bulldozer.” A frontal assault on Iraqi forces in Kuwait, lasting 10-20 days.
  • Operation “Leavenworth.” A double envelopment of Iraqi forces in Kuwait using an armored turning force in the west and a U.S. Marine amphibious landing in the east.
  • Operation “RazzleDazzle.” Similar to “Leavenworth,” but combined with an assault along the Iraq-Kuwait border by airborne/airmobile forces for a triple envelopment to encircle all Iraqi forces in Kuwait.
  • Operation “Siege.” A combination of an extended Operation “Colorado Springs” and ground force raids on all of Iraq’s borders. After 10-20 days, one of the three ground attack options (“Bulldozer,” “Leavenworth,” or “RazzleDazzle”) would be initiated to conclude the war.

Based on these assumptions, TNDA produced a range of casualty predictions for U.S. forces that TNDA asserted would probably be accurate to within +/- 50%. These ranged from a low of 380 for a 10-day “Colorado Springs” air-only campaign, to a top-end calculation of 16,645 for a 10-day “Colorado Springs” followed by a 20-day “Bulldozer” frontal assault.

TNDA’s Projection of Likely U.S. Casualties

Scenario Duration

Killed

Wounded

Total

+/-50%
Colorado Springs

10-40 days

190-315

190-315

380-630

—
Bulldozer*

10-20 days

1,858-2,068

8,332-9,222

10,190-11,290

5,335-16,645
Leavenworth*

10-20 days

1,454-1,664

6,309-7,199

7,763-8,863

4,122-12,995
RazzleDazzle*

10-20 days

1,319-1,529

5,534-6,524

6,853-8,053

3,717-11,790
Siege*

10-30 days

564-1,339

1,858-5,470

2,422-6,809

1,451-10,479

* Figures include air casualties

Based on these calculations, TNDA recommended the following course of action:

If the above figures are close to accurate (and history tells us they should should be), then the proper solution is to begin the war with the air campaign of Operation “Colorado Springs.” If this should result in an Iraqi surrender, so much the better. If not, then after about ten days of “Colorado Springs,“ to continue the air campaign for about ten more days while initiating Operation “Siege.” If this does not bring about an Iraqi surrender, the ground campaign should be concluded with Operation “RazzleDazzle.” If this has not brought about an Iraqi surrender, then an advance should be made through the desert to destroy any resisting Iraqi forces and to occupy Baghdad if necessary.[4]

In my next post, I will assess the accuracy of TNDA’s forecast and how it stacked up against others made at the time.

Notes

[1] Armed Services Committee, U.S. House of Representatives, Testimony of Col. T. N. Dupuy, USA, Ret. (Washington D.C.: 13 December 1990)

[2] Trevor N. Dupuy, Curt Johnson, David L. Bongard, Arnold C. Dupuy, If War Comes, How To Defeat Saddam Hussein (McLean, VA.: HERO Books, 1991); subsequently republished as How To Defeat Saddam Hussein: Scenarios and Strategies for the Gulf War (New York: Warner Books, 1991).

[3] These are the Quantified Judgement Model (QJM) and Tactical Numerical Deterministic Model (TNDM). Dupuy’s methodological approach and his first cut on a Gulf War estimate are described in Chapter 7 of Trevor N. Dupuy, Attrition: Forecasting Battle Casualties and Equipment Losses in Modern War (McLean, VA.: HERO Books, 1990).

[4] Dupuy, et al, How To Defeat Saddam Hussein, 126

Screw Theory! We Need More Prediction in Security Studies!

Johnny Carson as Carnac the Magnificent; taken from the January 24, 2005 broadcast of The Tonight Show.
Johnny Carson as Carnac the Magnificent; taken from the January 24, 2005 broadcast of The Tonight Show.

My previous post touched on the apparent analytical bankruptcy underlying the U.S. government’s approach to counterterrorism policy. While many fingers were pointed at the government for this state of affairs, at least one scholar admitted that “the leading [academic] terrorism research was mostly just political theory and anecdotes” which has “left policy makers to design counterterrorism strategies without the benefit of facts.”

So what can be done about this? Well, Michael D. Ward, a Professor of Political Science at Duke University, has suggested a radical solution: test the theories to see if they can accurately predict real world outcomes. Ward recently published an article in the Journal of Global Security Studies (read it now before it goes behind the paywall) arguing in favor of less theory and more prediction in the fields of international relations and security studies.

[W]e need less theory because most theory is an attempt to rescue or adapt extant theory. We need more predictions in order to keep track of how well we understand the world around us. They will tell us how good our theories are and where we need better explanations.

As Ward explained,

[P]rediction is deeply embedded in the philosophy of science… The argument is that if you can develop models that provide an understanding—without a teleology of why things happen—you should be able to generate predictions that will not only be accurate, but may also be useful in a larger societal context.

Ward argued that “until very recently, most of this thread of work in security studies had been lost, or if not lost, at least abandoned.” The reason for this was the existence of a longstanding epistemological disagreement: “Many social scientists see a sharp distinction between explanation on the one hand and prediction on the other. Indeed, this distinction is often sharp enough that it is argued that doing one of these things cuts you out of doing the other.”

For the most part, Ward asserted, the theorists have won out over the empiricists.

[M]any scholars (but few others) will tell you that we need more theory. Doubtless they are right. Few of them really mean “theory” in the sense that I reserve for the term. Few of them mean “theory” in the sense of analytical narratives. Many of them mean “detailed, plausible stories” about how stuff occurs.

In light of the uncomfortable conclusion that more detailed, plausible stories about how stuff occurs does not actually yield more insight, Ward has adopted a decidedly contrarian stance.

I am here to suggest that less is more. Thus, let me be the first to call for less theory in security studies. We should winnow the many, many such “theories” that occupy the world of security studies.

Instead, we need more predictions.

He went on to detail his argument.

We need these predictions for four reasons. First, we need these predictions to help us make relevant statements about the world around us. We also need these predictions to help us throw out the bad “theories” that continue to flourish. These predictions will help drive our research into new areas, away from moribund approaches that have been followed for many decades. Finally, and perhaps most important, predictions will force us to keep on track.

But making predictions is only part of the process. Tracking them and accounting for their accuracy is the vital corollary to improving both accuracy and theory. As Ward pointed out, “One reason that many hate predictions is that talking heads make many predictions in the media, but few of them ever keep track of how well they are doing.” Most, in fact, are wrong; few are held accountable for it.

Of course, the use of empirical methods to predict the outcomes of future events animated much of Trevor N. Dupuy’s approach to historical analysis and is at the heart of what The Dupuy Institute carries on doing today. Both have made well-documented predictions that have also been remarkably accurate. More about those in the next post.

Are They Channeling Trevor Dupuy?

TrevorCoverShot

Continuing the RAND description of their hex boardgame:

Ground unit combat strengths were based on a systematic scoring of individual weapons, from tanks and artillery down to light machine guns, which were then aggregated according to the tables of organization and equipment for the various classes of NATO and Russian units. Overall unit scores were adjusted to account for differences in training, sustainment, and other factors not otherwise captured. Air unit combat strengths were derived from the results of offline engagement, mission, and campaign-level modeling.

This looks like some kind of firepower or combat power score, or perhaps Trevor Dupuy’s OLIs (Operational Lethality Indexes). As they say “systematic scoring” one wonders what system they used. Know of only one scoring system that is systematic (meaning the OLIs, which are based upon formulae). The subject is probably best summarized in Dr. James Taylor’s article on “Consistent Scoring of Weapons and Aggregation of Forces:” http://www.dupuyinstitute.org/pdf/v2n2.pdf. This is the same James Taylor who wrote the definitive two-volume work on Lanchester equations.

I do note with interest the adjustment for “differences in training, sustainment, and other factors.” That is always good to see.

Also noted:

Full documentation of the gaming platform will be forthcoming in a subsequent report.

Look forward to reading it.

Lanchester equations have been weighed….

a-knights-tale_1

There have been a number of tests of Lanchester equations to historical data over the years. Versions of Lanchester equations were implemented in various ground combat models in the late 1960s and early 1970s without any rigorous testing. As John Stockfish of RAND stated in 1975 in his report: Models, Data, and War: A Critique of the Study of Conventional Forces:

However Lanchester is presently esteemed for his ‘combat model,’ and specifically his ‘N-square law’ of combat, which is nothing more than a mathematical formulation of the age-old military principal of force concentration. That there is no clear empirical verification of this law, or that Lanchester’s model or present versions of it may in fact be incapable of verification, have not detracted from this source of his luster.”

Since John Stockfish’s report in 1975 the tests of Lanchester have included:

(1) Janice B. Fain, “The Lanchester Equations and Historical Warfare: An Analysis of Sixty World War II Land Engagements.” Combat Data Subscription Service (HERO, Arlington, VA, Spring 1977);

(2) D. S. Hartley and R. L. Helmbold, “Validating Lanchester’s Square Law and Other Attrition Models,” in Warfare Modeling, J. Bracken, M. Kress, and R. E. Rosenthal, ed., (New York: John Wiley & Sons, 1995) and originally published in 1993;

(3) Jerome Bracken, “Lanchester Models of the Ardennes Campaign in Warfare Modeling (John Wiley & sons, Danvers, MA, 1995);

(4) R. D. Fricker, “Attrition Models of the Ardennes Campaign,” Naval Research Logistics, vol. 45, no. 1, January 1997;

(5) S. C. Clemens, “The Application of Lanchester Models to the Battle of Kursk” (unpublished manuscript, May 1997);

(6) 1LT Turker Turkes, Turkish Army, “Fitting Lanchester and Other Equations to the Battle of Kursk Data,” Dissertation for MS in Operations Research, March 2000;

(7) Captain John Dinges, U.S. Army, “Exploring the Validation of Lanchester Equations for the Battle of Kursk,” MS in Operations Research, June 2001;

(8) Tom Lucas and Turker Turkes, “Fitting Lanchester Equations to the Battles of Kursk and Ardennes,” Naval Research Logistics, 51, February 2004, pp. 95-116;

(9) Thomas W. Lucas and John A. Dinges, “The Effect of Battle Circumstances on Fitting Lanchester Equations to the Battle of Kursk,” forthcoming in Military Operations Research.

In all cases, it was from different data sets developed by us, with eight of the tests conducted completely independently of us and without our knowledge.

In all cases, they could not establish a Lanchester square law and really could not establish the Lanchester linear law. That is nine separate and independent tests in a row with basically no result. Furthermore, there has never been a test to historical data (meaning real-world combat data) that establishes Lanchester does apply to ground combat. This is added to the fact that Lanchester himself did not think it should. It does not get any clearer than that.

As Morse & Kimball stated in 1951 in Methods of Operations Research

Occasionally, however, it is useful to insert these constants into differential equations, to see what would happen in the long run if conditions were to remain the same, as far as the constants go. These differential equations, in order to be soluble, will have to represent extremely simplified forms of warfare; and therefore their range of applicability will be small.

And later they state:

Indeed an important problem in operations research for any type of warfare is the investigation, both theoretical and statistical, as to how nearly Lanchester’s laws apply.

I think this has now been done for land warfare, at last. Therefore, I conclude: Lanchester equations have been weighed, they have been measured, and they have been found wanting.

Really…..Lanchester?

RAND described the combat system from their hex boardgame as such:

The general game design was similar to that of traditional board wargames, with a hex grid governing movement superimposed on a map. Tactical Pilotage Charts (1:500,000 scale) were used, overlaid with 10-km hexes, as seen in Figure A.1. Land forces were represented at the battalion level and air units as squadrons; movement and combat were governed and adjudicated using rules and combat-result tables that incorporated both traditional gaming principles (e.g., Lanchester exchange rates) and the results of offline modeling….”

Now this catches my attention. Switching from a “series of tubes” to a hexagon boardgame brings back memories, but it is understandable. On the other hand, it is pretty widely known that no one has been able to make Lanchester equations work when tested to historical ground combat. There have been multiple efforts conducted to test this, mostly using the Ardennes and Kursk databases that we developed. In particular, Jerome Braken published his results in Modeling Warfare and Dr. Thomas Lucas out at Naval Post-Graduate School has conducted multiple tests to try to do the same thing. They all point to the same conclusion, which is that Lanchester equations do not really work for ground combat. They might work for air, but it is hard to tell from the RAND write-up whether they restricted the use of “Lanchester exchange rates” to only air combat. I could make the point by referencing many of these studies but this would be a long post. The issue is briefly discussed in Chapter Eighteen of my upcoming book War by Numbers and is discussed in depth in the TDI report “Casualty Estimation Methodologies Study.” Instead I will leave it to Frederick Lanchester himself, writing in 1914, to summarize the problem:

We have already seen that the N-square law applies broadly, if imperfectly, to military operations. On land, however, there sometimes exist special conditions and a multitude of factors extraneous to the hypothesis, whereby its operations may be suspended or masked.