In the 400+ years since the birth of modern statistics, data has been collected on everything from life expectancy and planetary motion to little league batting averages and micro-loans in rural Bangladesh. As technology catches up with the world’s desire to better predict the future and understand the past, applications have expanded to include dynamic models of the global economy and more recently the probability of a terrorist attack. The danger with relying on this methodology, of course, is that the same statistical biases that contributed to the recent financial chaos may cause more harm in the real world than they ever did on Wall Street…
Calculated Terror
How a computer model predicts the future in some of the world’s most volatile hotspots.
BY AARON MANNES, V.S. SUBRAHMANIAN in Foreign Policy
An Israeli tourist feigns surrender in front of a metallic cut-out of a soldier at the Golan Heights outpost of Mount Bental.
On Oct. 27, a Katyusha rocket was fired from Lebanon and struck down in an open area outside the northern Israeli town of Kiryat Shmone. This was the ninth such rocket strike since the end of the 2006 war between Israel and Hezbollah. No group claimed responsibility for the attack, but smaller Palestinian groups hoping to spark another round of fighting are the most likely suspect. Hezbollah, despite its extreme anti-Israel politics, did not join the fight, even after Israeli counterstrikes.
The “Blue Line” separating Israel from Lebanon is one of the most volatile borders in the world. But predicting when this area, and other tense regions throughout the world, will erupt into violence often appears to be little more than guesswork. How can policymakers overcome their own biases and limited information to anticipate if an incident like the recent rocket strike on Israel will spark a larger conflict, like the 2006 war, or fizzle out?
Increasingly, the answer is: Develop a computer model from historical data. The University of Maryland’s Laboratory for Computational Cultural Dynamics (LCCD) constructed one such model that predicted this period of quiet along the Israeli-Lebanese border, and also provides insight into Hezbollah’s priorities. LCCD developed a framework, known as Stochastic Opponent Modeling Agents (SOMA), that examines historical data and automatically generates rules assessing the probability that a group will take certain actions under certain conditions.
SOMA examines historical data about the group’s behavior and tries to find conditions such that, when the condition is true, the group takes a given action with high probability and, when the condition is false, the group takes the action with very low probability. A human analyst could make these connections when there are relatively few variables being tracked. But when there are dozens of variables there are millions of such possible rules — far more than an analyst can process. This is often the case in the interconnected world of Middle East politics, where events are shaped by the actions of many actors working in a diverse array of countries.
SOMA rules have also been extracted on the behavior of other Middle Eastern groups. Hamas, for example, is twice as likely to commit kidnappings during periods of conflict with other Palestinian organizations (the probability increases from approximately 33 percent to 67 percent). If another round of Fatah-Hamas fighting erupts in the West Bank, this may present a new challenge for Israeli security. While the rules had not been extracted in 2006, it is worth noting that the Israeli soldier Galid Shalit was kidnapped as the conflict between Hamas and Fatah expanded after the 2006 Palestinian elections.
SOMA is not specifically designed to model the behavior or Hezbollah or even of terrorist organizations — it has also examined the behaviors of various actors in the Afghan drug trade under different circumstances. This model was built on a hypothetical situation, not systematically gathered data, but demonstrates the way in which SOMA can be applied to a broad range of conflicts and scenarios. The analysis showed that frequently used strategies such as burning poppy fields and destroying drug labs in Afghanistan are unlikely to lead to a long-term decline in the Afghan drug trade.
Models require data, and limitations of that data can limit the accuracy of a system such as SOMA. For the analysis of Hezbollah (and several other groups) SOMA used the Minorities at Risk Organization Behavior (MAROB) data set created at the University of Maryland’s Center for International Development and Conflict Management. MAROB identifies factors that motivate members of ethnic minorities to form activist organizations and move from conventional politics to terrorism. MAROB has systematically collected information on more than 150 variables from over 100 organizations across the Middle East during the last several decades. Hezbollah is one of the organizations profiled; the data collected covers Hezbollah from its 1982 founding through 2004.
In examining the rules generated by SOMA about Hezbollah’s behavior, the most striking finding was the correlation between Hezbollah attacks on Israeli citizens and Lebanese elections. Since the re-establishment of Lebanon’s parliamentary democracy in 1992, there was a 62 percent chance Hezbollah would target Israeli civilians (primarily through rocket attacks) in any given year through 2004. In off-election years the likelihood jumped to 78 percent, while in election years the probability was negligible. The one election year in which Hezbollah fired rockets at Israel was 1996. Though Hezbollah won a propaganda victory when Israel’s response caused heavy Lebanese civilian casualties, the organization lost parliament seats in the 1996 elections. Hezbollah has since sought to keep its fighting with Israel within certain boundaries, avoided major escalations during election years, and re-emphasized its provision of social services within Lebanon.
The test for any model is whether or not its predictions hold. During Israel’s Operation Cast Lead against Hamas in Gaza, there was concern that Hezbollah would initiate a second front to aid its ally. But Hezbollah offered only rhetorical support to Hamas. During the Gaza operation, a few rockets were fired from Lebanon into Israel and Hezbollah quickly and credibly denied responsibility. With an election later in the year, Hezbollah determined that it could not risk renewed violence with Israel — particularly in the wake of the 2006 war, which many Lebanese felt was brought on by Hezbollah and that left much of south Lebanon in ruin.
Beyond its predictive value, these findings provide insight into Hezbollah’s behavior and priorities. The SOMA results highlight how Hezbollah needs to maintain its position within Lebanon’s political system, even if that restricts its ability to wage war on Israel. Though the conflict with Israel is the organization’s raison d’être, Hezbollah’s leadership has taken steps — such as participating in the Lebanese parliament — to prevent the party from being politically isolated.
However, this balancing act is becoming more difficult to maintain, as these two aspects of Hezbollah are coming into increasing conflict. The 2006 war inflicted massive costs on the Lebanese people, which has made them less likely to tolerate Hezbollah’s foreign adventurism. Earlier in October an explosion at a private home revealed the presence of a Hezbollah arms cache (Hezbollah disputes this). This incident reminded the Lebanese that Hezbollah remains capable of launching another round of fighting with Israel, and raised the specter of the conflict being sparked by accident. The recent Israeli seizure of a cargo ship carrying nearly 400 tons of weapons apparently intended for Hezbollah raised further concerns that Lebanon could again become a battlefield between foreign powers. If one appreciates that Lebanese popular opinion exerts a strong influence on Hezbollah’s actions, it should be clearer that another Israel-Hezbollah war remains unlikely.
SOMA’s analysis of Hezbollah’s behavior serves both analysts and policymakers by making a specific prediction about the group’s likely actions, and also by highlighting this important underlying dynamic. As the data collected expands in breadth and depth, it may become possible to make specific predictions about how, when, and under what circumstances regional changes will occur. While the Oct. 27 rocket strike on northern Israel seems to be just the sort of incident which could cause an unpredictable chain reaction in the region, in the future its repercussions may largely be known before the rocket leaves the ground.