Predicting the Duration of the 2003 Gulf War

On April 1, 2003, we posted a research report predicting the duration of the war between the United States and Iraq. The original webpage and research report appear below. The report and follow-on analysis have now been published in the journal Foreign Policy Analysis. A .pdf file of that publication is available here. This article includes predictions about the likely duration of a US-Iraq war if the Iraqis employed a punishment (guerrilla) strategy (83 months), and estimates the duration of counterfactual U.S.-North Korean and U.S.-Syrian wars (12-24 months).

April 1, 2003

In recent days, the media has made much of the ups and downs of the war between the United States and Iraq. Such a focus on the earliest days of the war is unwise, because since 1816, half of all wars have lasted more than 5 months, with the average war between 1816 and 1992 lasting approximately 17 months. In this report, we apply our previously-developed statistical model of war duration to the current war situation involving the U.S. and Iraq to forecast the duration of the war. While it is too early to tell how the war will progress, no one should have expected the war to be over in a matter of days.

Overview

In 1996, in the American Political Science Review, we analyzed the duration of interstate wars between 1816-1985. [Bennett, D. Scott, and Allan Stam. 1996. “The Duration of Interstate Wars, 1816-1985.” American Political Science Review 90 (June):239-257.] The primary focus of our analysis lay in determining how several theoretical factors have influenced the duration of war, and whether their influence matches our theoretical understanding of war. The model may also be used to forecast the duration of other out-of-sample wars. In this report, we use the model presented in Bennett and Stam (1996) to predict the likely duration of the current (2003) war between the United States and Iraq.

Our model includes military, political, demographic, industrial, and geographic factors. It is not based on the type of intense wargaming that has been conducted by the U.S. military, but rather is based on a statistical model that quantifies the effect of multiple factors on the duration of wars since 1816. By focusing on the factors that have affected war duration historically, and comparing the duration of many wars over a long time period, we can make projections about current and future conflict. While knowing that the median war lasted 5 months tells us something about the likely duration of this war, our model can do better than just predicting the mean or median. Ultimately, we present our estimates of the likely duration of the 2003 U.S.-Iraq war under four scenarios. Our best estimate of the likely duration of the war (given the evolution of the war thus far, and assuming that the United States is able to maintain its maneuver-based strategy) is approximately 2½ months. If the U.S. is forced to turn to a pure attrition-based strategy in which it is forced to defeat most or all Iraqi units through direct combat, our estimate of the war’s possible duration stretches to nearly a year.

Available Materials

The research report contains a discussion of the statistical model, the factors included in it, and provides the coefficient estimates from the model used to generate these forecasts. The spreadsheet used to make the forecasts is also available, and may be used to alter the assumptions used to forecast the current war's duration. For those interested in better understanding the construction of the variables in the model and their interpretation, we refer you to the article in the American Political Science Review. Finally, to facilitate re-analysis and extensions of the statistical model, we provide the data and command file necessary to replicate the analysis. The data set we analyze draws heavily on the Correlates of War project, as well as Allan C. Stam III, Win, Lose or Draw (Ann Arbor: University of Michigan Press, 1996).

Obtain the full report (Adobe Acrobat .pdf format) by clicking here.

Obtain the forcasting spreadsheet (Microsoft Excel .xls format) by clicking here.

The data used for the analysis (Stata and Excel format), and the command file used to analyze it, are accessible by clicking here.

Summary of Forecasts

We make a set of contingent predictions about the 2003 war. The full research report (v1.1, April 1, 2003) discusses the statistical model and data underlying these estimates, and the accompanying forecasting spreadsheet shows the values of the variables and coefficients employed.