The figure below is the 2-core of the full network. As you might imagine, the full network is quite large, but the vast majority of nodes are pendants; therefore, visualizing the 2-core is much more useful. Those that twittered with #rstats are in red and labeled, while all of the intervening nodes are in blue.

I am a bit of a wallflower in this crowd, but I am a relative newcomer to both Twitter and R so this makes sense. You can see that powerhouse gurus like @CMastication, @revodavid, @dataspora, @hamiltonulmer, etc. are very central. Also, this visualization nicely depicts the size and connectedness of the R community on Twitter. For example, the network contains only 181 nodes and has a diameter of 7. On the other hand, there are several disparate cliques around the central actors (e.g., @gappy3000 and myself form a small clique).

I look forward to digging further into the Twitter data on Tuesday, and if you are attending and have any suggestion or specific questions please let me know and I will try to work them in.

Introduction

Over the past several years a growing body of literature has focused on rational choice and terrorism—with a particular focus on suicide terrorism. Early attempts focused on empirical analysis of various factors that influence an individual’s decision to take their own life in the name of a terrorist cause. More recently, several formal models have been proposed to generate predictions about the rational condition under which such a decision is a best response. All of these models, however, begin by assuming an individual exists that is motivated to kill themselves, and from there attempt to predict the outcome of that decision. This is not an unreasonable assumption, given that the existence of these individuals is self-evident; however, there is clearly a preliminary step. The body of work on suicide terrorism has yet to examine the process by which these individuals were persuaded to think that committing an act of suicide terrorisms was—in fact—a rational decision. Furthermore, it is reasonable to assume that in non-extreme cases (e.g., mental illness, etc.) individuals are not naturally motivated to suicide; therefore, a model that examines the dynamics of persuasion to self-destructive means may be very useful in understanding how rational human beings can be convinced to kill themselves for a cause.

Using the basic framework of deliberation proposed by Hafer and Landa, in the following research proposal I present a modified version of their framework to explore the dynamics of persuasion to extreme ends. For this preliminary piece I first present the modified model, then conclude by briefly discussing a lab experiment of the basic modeling framework.

A model of self-destructive persuasion

Without reviewing it in its entirety, the Hafer and Landa (H&L) model of deliberation is predicated on two simple yet profound assumptions. First, that individuals have both active and latent elements of their persuasive consciousness, and current opinion or motivation is controlled by the active elements. In the act of deliberation, however, various messages can activate the latent elements, which in turn can persuade an individual to a new position. This model is very useful because it does not impose a complicated set of assumption, while at the same time captures the fundamentals of deliberation in an intuitive way. To describe the process of self-destructive persuasion, focusing on the selection and motivation of a particular suicide terrorist, it will be necessary to impose further restrictions on the model.

Terrorist activity is inherently a highly selective act. At ever step in the process choices about who, what, where, when and how are made under circumstances of incredible scrutiny and risk. This model will focus on the `who’ decision, and through which will attempt to describe the dynamics of persuasion to suicide terrorism. To begin, I first consider the process of used in selecting a suicide terrorist. For expository purposes, suppose there is a terrorist organizer seeking to recruit a suicide bomber. The organizer wishes to find the individual that will be the most effective suicide bomber, but in order to do that he must receive signals from potential recruits about their willingness and ability. In the H&L model there is no cost associated with deliberation, but in this environment saying the wrong thing to the wrong person can result in sever consequences; therefore, I propose a series of modifications that encompass this reality.

In the standard H&L model the active and latent persuasive elements of each actor are simply members of a set. In this game, however, these elements must have a natural ordering. For example, the latent or active element that corresponds to suicide terrorism is considerably more `extreme’ than the others, and should be ordered as such. Rather than a set, I model the persuasive elements as members of an ordered n-tuple, wherein the elements are ordered by their relative severity. This change allows actors to form beliefs about the attitudes of their counterparts. For example, if the total set of possible persuasive elements is {A,B,C,D,E} and some actor sends a D, than the receiver of that message has a strong indication that this person is closer to the extreme. The game, therefore, proceed as follows.

First, there are a set of players N={1,2,…,n}, where player n is the sole recruiter in the game, and all other players are possible recruits. For each players the number of elements is equal and drawn with uniform probability from set of possible elements. Each player, however, varies in the elements that are active and latent; with one exception, for any player with the extreme element, e.g, the desire to commit an act of suicide terrorist, that element will always begin as latent and must be activated by the recruiter. The recruiter wishes to activate the latent extreme element of those actors that have it, but he cannot simply send that message at all times in an attempt to activate the other players.

The recruiter incurs a large cost, denoted as C, whenever he sends the extreme signal, and this cost can only be recuperated as a function of a successfully recruiting a suicide bomber. This is an attempt to model the cautious conditions under which terrorist recruiters operate. Likewise, only players with a latent extreme element have a utility function that benefits from executing a suicide attack, and thus they `want’ to be found by the recruiter even though they do not consciously know this. Other players are indifferent to the presence of a recruiter, while some are inclined to expel the recruiter from the game. Again, this further compounds the difficult scenario facing the recruiter. For each of these player types the distribution of persuasive elements will roughly correspond to their utility, e.g., on average players that wish to expel recruiters will have the least extreme elements, indifferent players will be somewhat more extreme, and so on.

Given this dynamic the game proceeds as something akin to a game of Battleship. The recruiter must attempt to solicit information about likelihood that a given player has the extreme latent element by sending progressively extreme messages in deliberation. At the same time, however, the recruiter must not send the extreme signal to players that do not have that element. Concurrently, players that benefit from exposing a recruiter are attempting to use the deliberative process to find these players and remove them. Finally, as this is an iterative process of message sending, it will be useful to add a simple discount factor to all utility functions, denoted , where all players are risk averse.

The game ends when the recruiter sends the extreme message, which can result in several different outcomes. First, the recruiter can successfully find a player with a latent extreme element and activate it (best outcome for recruiter). On the other hand, if the recruiter is unsuccessful he will either be expelled from the game by a player whose utility is increased by this (worst outcome form recruiter), or he will misestimate am indifferent player’s likelihood to have an extreme latent element and incur the cost C (intermediate outcome for recruiter).

Clearly, actual specifications of utility functions for all players are needed to infer outcomes from this game; however, in this brief I will not be defining these functions explicitly. Instead, this piece is meant to simply establish a basic framework for creating a model of self-destructive deliberation. In the final section I describe how this framework might fit into a lab experiment for testing.

Lab experiment

As was the case with H&L model, a lab experiment may be an excellent way of testing the theories generated from the model described above. For a lab experiment of this game, one player will be assigned as the recruiter and another as the expeller (a player that benefits from exposing and removing a recruiter from the game). The remaining players are computer agents, consisting of a randomly assigned set of indifferent players or potential recruits (those that benefit from executing an attack). As stated above, the game proceeds as something similar to a game of Battleship; therefore, the operationalization of the lab experiment will mimic this in form.

Players compete in an online environment where they see all of the players in the game, which at the start are represented by empty bins. In each round players can broadcast a message from their set of persuasive elements to some finite set of other players in the game. If that message is an active element of one of those players’ set it is highlighted in that player’s bin in yellow. If the message activated a latent element of that set for some player it is highlighted in that player’s bin in green, and if the element had no effect (e.g., was not in their set of persuasive elements) it is highlighted in red. In any round the recruiter player can direct the extreme message at a single player. When this occurs the game ends and player utilities are calculated based on the function associated with their type and discounted by the number of rounds played.

The game played in the lab generates an interesting dynamic between the two human players. Both players want the game to end quickly, but are highly incentivized for different outcomes. Again, a major limitation to this model in its current form are the lack of actual utility functions. That said, it is likely that even very simple utility functions based on and C would generate the desired outcomes.

Photo: Wikimedia

In their section reviewing predictive models broadl the authors state at the outset the difficulty in predicting human behavior, even going so far as to say that current social science models are ill-equipped to answer such questions. While it is clearly the case that there does not exist a black box (magic ball?) to predict the exact time, location and magnitude of a given terrorist attack, the authors begin by saying these models are useless, but then go on to suggest as a solution exactly the types of social science models used in practice as a solution. For example, the authors state:

There may be one way forward for predictive modeling of rare events. Suppose we assume rare events are drawn from a distribution that includes other events that are sufficiently common that we can observe many of them, enough to evaluate a model. We could, for instance, assume that small observable events and large rare events are sufficiently related in their causes that we are willing to assume large rare events are just the high-magnitude tail of some underlying distribution of events.

The notion of underlying distributions and prior belief are at the foundation of all Bayesian statistical analysis, of which there are several examples examining terrorist events. I find it a bit intellectually dishonest to propose this framework as a novel “way forward” given the growing literature on the subject. In addition, a running theme through the paper is the importance of the power-laws in examining the distribution of terrorist events. This recycled observation has become a bit of a nagging trend in terrorism studies; therefore, I will not go any further but to say in both of these cases a cursory review of the literature would have benefitted this work greatly.

To their credit, in their section on game theory the authors propose an interesting model for terrorist target selection; unfortunately, their conclusions seem to run in contradiction to observed selection behavior. That is, rather than follow an equilibrium path to a target that provides a best response to the counter-terrorism tactics of the target country, the MITRE model predicts a terrorist will chose randomly among a finite set “defended targets.” The game (pg. 67-69) is very straightforward, and I recommend those interested review it; however, for comparison I also suggest the model of target selection proposed here for a somewhat more intuitive model.

Finally, the authors make one final and fatal mistake at the conclusion of the paper, wherein they actually attempt to predict the probability of a high magnitude terrorist event (pg. 89). What is troubling is that for nearly 90 pages the authors point out the difficulty of these predictions, then in a single calculation of expectation using the power-law distribution and a mean of 2,400 terrorist events worldwide peg the probability at ~7%. To the econometrically inclined this is somewhat laughable, given the numerous issues: over/under reporting of terrorist events based on geography (how often are terrorist attacks in China or DPRK reported?), autocorrelation of attack success over time (how are terrorists becoming more effective over time?), country fixed effects (variation in CT policies in different countries), the list could continue. I understand that JASON was most likely directed by DoD to provide an actual number, but I point these issues out only to highlight to difficulty and danger of doing this.

Despite my misgivings, the paper will likely be influential in the U.S. defense community, and as such should be at the top of everyone’s reading list for this week.

Photo: IBM

The question is a very good one, and the authors investigate it with a unique data set:

Our analysis overcomes these difficulties by relying on a detailed data set of suicide terror attacks and local economic conditions together with a unique empirical strategy. The available data set covers the universe of suicide Palestinian terrorists during the second Palestinian uprising, combined with quarterly data from the Palestinian Labor Force Survey on districts’ economic and demographic characteristics, and Israeli security measures (curfews and Israeli induced Palestinian fatalities).

The punchline…

…a successful attack causes an immediate increase of 5.3 percent in the unemployment rate of an average Palestinian district (relative to the average unemployment rate), and causes an increase of more than 20 percent in the likelihood that the district’s average wage falls in the quarter following an attack. Finally, a successful attack reduces the number of Palestinians working in Israel by 6.7 percent relative to its mean. Importantly, these economic effects persist for at least two quarters after the attack.

While I think this paper introduces a very important research paradigm, I have a concerns with some of the technical assumptions built into their analysis, and the overarching reliability of research focusing exclusively on terrorism in the Israel/Palestine conflict. With respect to the technical assumptions there is one line in the paper that struck me as very problematic: “Our empirical strategy exploits the inherit randomness in the success or failure of suicide terror attacks as a source of exogenous variation to investigate the effects of terrorism on the perpetrators economic conditions.”

I find it very difficult to accept the notion that success and failure is random across suicide attacks—especially within this particular conflict. There is clearly no support for a theory that selection of suicide attack sites is random; therefore, it follows that the success of an attack would also be a function of both the selected target as well as the learning process occurring by both the attackers and defenders. There is, therefore, an expectation of high autocorrelation across success for attacks happening within a relatively small geographic area. Such difficulties highlight the general problem of external validity for terrorism studies that focus solely on the Israel/Palestine conflict.

It is not surprising that researchers often default to data on terrorist attacks from this conflict. Given the relative openness of Israel’s democratic government, the media attention on Palestine, and the—unfortunate—frequency of attacks there exists are large amount of data from this conflict. As I have mentioned before, however, it is very difficult to infer causality from this data given the natural interconnectedness of the conflict dynamics. As I mentioned, there any large-N study of terrorism in this context has enormous selection problems, as terrorists learn innovate to evade the defensive tactics of the ISF, and the Israelis create new policies that may provoke and dissuade the terrorist activities. There are no other ongoing low-intensity conflicts that have issues at this level, making it difficult to draw parallels between findings from research focusing and Israel and Palestine and another other conflict.

I am curious as to others’ thoughts on this issue of external validity, and welcome your comments.

Photo: Norman G. Finkelstein

The title of the talk is the name of the modeling technique I have been developing, which I am calling the “Structurally Induced Random Graph Model.”

There is an accompanying paper that will (hopefully) be published as part of an IEEE conference proceedings. Once the paper has been edited and submitted I will upload a copy of it here.

This research is extremely preliminary, and as such I welcome all questions and comments. I am hoping to get a lot of constructive feedback from the audience at West Point tomorrow, but if you are not able to attend I welcome your thoughts in the comments or email as well.

The code used to generate the networks in the slides is available in the ZIA Code Repository at Python/Networks/SIRG/SIRG_dev1.py. Happy testing!

UPDATE: I forgot to mention that I will also be live-tweeting the conference to some extent under the hash-tag #USMANets09. If people are already live-tweeting under a different has-tag let me know and I will switch, but I could not find anything.

The current foreign policy rift over the future of the U.S. military’s engagement in Afghanistan is being framed as a philosophical battle between the counterinsurgency school on one side; represented by Gen. Stanley McChrystal and the omnipresent CNAS, and the counterterrorism school; primarily championed by Vice President Biden. Is this, however, a constructive framework for analyzing the situation in Afghanistan? Consider, for example, the recent tragic helicopter crashes that claimed the lives of 14 Americans. On the topic, Noah Schactman reports at Danger Room:

In short, helicopters are the irreplaceable connective tissue of the Afghanistan war effort — and its potential Achilles’ heel. “It’s our strategic weak point,” a defense official told Danger Room.

The mission in Afghanistan is thus contingent on the availability of certain resources, e.g., helicopters to move people and supplies. Leaving questions of what constitutes victory in Afghanistan aside, at a minimum it is reasonable to posit that any achievement in Afghanistan begins with the Afghan people beginning to govern themselves. All things being equal then, we would rather send city planners, teachers, doctors, etc., to Afghanistan in order to achieve this minimum goal, but are constrained by the violent environment and unforgiving terrain. We must, therefore, forgo sending these people in order to send soldiers, guns and helicopters. The current debate would be much more productive if framed as a budget constraint, between guns (soldiers and helicopters) versus butter (doctors and teachers), rather than a philosophical one.

If considered from this perspective the question then becomes: What is the minimum number of troops needed to promote Afghan sovereignty, and given this budget constraint how can the remaining surplus be used to expedite this process. In all likelihood, what has been construed as a philosophical debate is actually a difference of opinion on how to maximize this budget constraint. McChystal may believe the state of the world is of the type that requires far more guns than butter, while Biden is of the other opinion.

If this is the case, then what is needed is better information about the state of the world, meaning that if both parties are simply maximizing this constraint they would come up with the same answer given the same data. Of course, some may argue that given McChrystal’s close proximity to the situation and Biden’s relatively far one he has the better information and is thus making the better calculation; however, given the tone of the current debate it does not seem that this is what motivates his policy recommendations.

Photo: Danger Room

In that vein, here is a preview of that research in the form of my abstract, which will be updated in the coming weeks with the slides from my talk and eventually the paper.

Title: Structurally Induced Random Graph Model of Social Networks

The research and development of random graph models of social networks has provided great insight into the dynamics of complex network systems. The primary shortcoming of these models; however, is their treatment of the atomistic component of a network—the vertex. These models assume vertices exist in a vacuum, bringing no exogenous structure to the network system and only form endogenous structure once inside a network. This assumption, however, is entirely contrary to the fundamental dynamics of social interaction observed in nearly all settings. The following research attempts to bridge the gap between current random graph models of social networks and the process by which individuals form new social structure in the real world. This paper presents a new random graph model deemed the “structurally induced random graph model”, or SIRG, which is derived from two key assumption. First, actors do not enter networks as isolates; and second, new structure in a network will resemble previously observed structure in that network. The paper proceeds as follows; first, a brief review of the literature on random graph models is provided. Then, the basic framework of the SIRG model is then presented, which is followed by a description of a single algorithmic implementation. Next, results from this implementation are presented, including several example networks generate by the model, and concluding remarks focus on the benefits of this model over current family of models and future research.

If you are attending the conference let me know, as an undergrad I participated in a conference on terrorism at West Point and had an outstanding time in the Officers’ Club one evening. I look forward to meeting all the Army net-heads!

Videos available for your viewing pleasure after the jump…

The highlight for me was the depth to which Sen. Coburn will sink by implying that funding political science research is “water-boarding the next generation.” It should be noted that political scientist were among the first to note the logical disconnect between torture and intelligence gathering.

I welcome everyone’s thoughts on both speeches, particularly if you feel more compelled to support Sen. Coburn’s amendment after hearing his speech. Finally, it appears that a vote on this amendment will not take place until later this week, as such I encourage anyone who has not yet signed my online petition to do so immediately!

A recent working paper entitled, “The Political Economy of Counterinsurgency Violence,” seeks to fill this void by offering a simple formalization of counterinsurgency strategy. In fact, the author ask an extremely important question in the opening paragraph:

Why are counterinsurgents often so brutal toward civilians if classical counterinsurgency theory is correct in suggesting that successful counterinsurgents must win—not destroy—the hearts and minds of the population?

To understand this dynamic by modeling counterinsurgency as a game with three players. First, a coalition of a rebel group and their popular support within a community, and second the counterinsurgent. To achieve its goals, the counterinsurgent seeks to divide this coalition through a mixture of violence and concession, which tempts some side in the coalition to defect on its partner for short-term gains and forgo long-term goals. Formally, the game is played as a public goods game, where each player has some level of “profit” it extracts from the insurgency, which is offset by the cost of participation. Thus the counterinsurgent seeks to short-circuit the profit chain through the threat or execution of violence.

What falls out of this model is an very interesting observation about how insurgency are a function of the active micro-economies where they take place. As the author states:

The rebels’ profit from insurgency increases due to windfall and black market revenue, external aid, natural resources, taxation, remittances, looted property and labor, and the availability and attractiveness of the rebels’ sanctuaries. An increase in the rebels’ accountability to the population and a decrease rebel profit results from restrictive geography, vulnerability to the population’s disloyalty caused by the nature of the rebel group’s organization, and the presence and strength of quasi-judicial institutions with which to sanction rebels’ abusive behavior…Factors negatively and positively affecting the actors’ relative profit during insurgency ought to correlate with the government’s use of indiscriminate violence.

The model is clever, and the author’s keen attention to the work of key counterinsurgency scholars comes through in his incorporation of critical elements of insurgency in the model. What is interesting, however, is how the model does not do a good job of predicting the kind of indiscriminate violence observed by Lyall in his research. The author here uses case studies from Guatemala and Turkey to support his theory, but given the high profile of Lyall’s work it would have been much more satisfactory to get a model that explained those observations. Of course, it is not the job of a modeler to fit data, and it may simply be the case that Lyall’s natural experiment is flawed, and this model requires better data for testing; either way, the article is very engaging and I highly recommend it.

Photo: New Internationalist