By Yousef MunayyerA headline in a recent New York Times blog, “Does Terrorism Work?” caught my attention. It turns out a new paper recently released by Eric Gould and Esteban Klor, economists at the Hebrew University, concluded that terrorism does in fact “work."
There is little doubt that some on the Israeli right and other proponents of perpetual occupation would draw policy implications from this type of research to justify occupation under the guise of security.
Given that the findings defy the logic behind the advent of the first Likud-led Netanyahu administration (which was preceded by a campaign of suicide bombings), the following Labor-led Barak administration (which was preceded by a significant lull in attacks) and the Sharon Administration which followed (preceded by the start of the second Intifada), I found it necessary to read further.
The abstract to “Does Terrorism Work?” reads in part:
This paper examines whether terrorism is an effective tool to achieve political goals. By exploiting variation in terror attacks over time and across locations in Israel from 1988 to 2006, we show that local terror attacks cause Israelis to be: (i) more willing to grant territorial concessions to the Palestinians; (ii) more willing to accept a Palestinian state; (iii) and less likely to identify oneself as being right-wing.Certainly, there has been empirical research on the efficacy of terrorism. Robert Pape’s “Strategic Logic” is one notable work referenced by the authors of this paper. However, Pape’s work had its problems, including its case study on Israel, which I will not go into here.
Given the case selected for study in this paper, and its counter-intuitive (perhaps contradictory) conclusion, I wondered just what methodological path was taken to get to this end. Here are some flaws I found which cast doubt on these findings.
1. The measure of attitudes regarding territorial concessions is flawed – To come up with a quantifiable variable to measure the attitudes of Israelis over time, the authors pieced together the responses to three different polling questions over the period in question. From 1988-92 respondents were asked if they would concede “In return for a peace agreement, a return of most [emphasis mine] of Judea, Samaria and the Gaza Strip”. In other years they were asked for degree of agreement with the following statement: “Israel should give back territories to the Palestinian for peace” or “To what extent to you agree or disagree to exchange territories for peace?”
Obviously, there is ambiguity in the extent of territories to be relinquished in the second and third questions, but decidedly less so in the first. Respondents who desire to keep all settlements as part of Israel could answer the second and third question comfortably in the affirmative, while respondents with the same desires could not do the same while answering the first question. In one set of questions the definition of “territories” is left up to the personal imagination of the respondent, whereas in the first question the definition is far clearer. Not surprisingly, the majority of respondents during the period the first question was asked were not in support of territorial concessions, while the respondents in time periods when the second in third questions were used to code the variable were more likely to agree to territorial concessions.
What is surprising is that the authors did not find a problem with the stark difference in the wording of the first question, and rather focused on the proper measurement of the second and third question due to the rank order nature of the responses.
2. The theory barely explains the phenomena – The majority of the charts referenced in the paper where the authors analyze the data they have collected are OLS regressions. This in itself may be problematic as other analytical tools may have been better suited for this type of data (Not to mention regression analysis may prove correlation, but it falls short of proving causation). However, let’s look at the OLS regressions at face value.
Only two of the tables (5,6) include R2 values. The R2, for those who are not regression analysts (don’t worry, that’s most people), is a statistical expression of the theories explanatory power. In other words, the R2 lets us know what extent of the variation within the sample is explained by a given theory. In OLS regressions, this is a pretty important statistic.
Further, the larger the model (the more relevant variables taken into account), the higher the R2 tends to be. In the most complicated of the models put forward by the authors, which includes more than 15 variables, the R2 barely approaches 14%. While many of the variables including fatalities, religiosity and the respondent’s background were statistically significant, they only explained a sliver of the variation in attitudes.
3. The sample excludes a significant part of the Israeli electorate - By excluding Arab citizens of Israel and settlers in the West Bank and Gaza who participate in the Israeli electoral system as citizens, the study is undoubtedly skewed.
The exclusion of settlers is particularly problematic. This is a fast growing and increasingly important constituency in Israeli elections. While there is no doubt that they have a personal attachment to the land of the West Bank and Gaza as residents, and this may skew their responses in regard to territorial concessions, the corollary is also true. Israelis who do not live in settlements do not have a personal stake because they do not live in the West Bank and Gaza, and this too also certainly affects their response to questions on territorial concessions.
4. How can we measure if “goals” are met, if “goals” are not defined? - The authors’ argument that terrorism is effective because the goals of the groups are being met assumes that shifting Israeli public opinion is the objective. The authors need to clearly define terror group objectives to determine whether goals are met. The authors of this paper never attempted to do this. In fact, they devote one paragraph of the paper to justifying their understanding of the objectives of these groups. Much of the paragraph is a quote from one leader of one of these groups to a media outlet. Many studies within the literature on terrorism agonize over defining the goals and objectives of groups. This study fell very short of doing the minimum necessary for their assumption to hold water.
5. The micro-level analysis doesn’t control for macro-causes - The authors argue that by taking the public opinion analysis down to the district level, they can isolate larger macro causes. I am not convinced. While they go to great lengths to try to control for the various factors that make different districts different -- particularly religiosity and education -- I still don’t think these rule out exogenous factors effecting Israeli public opinion as a whole.
Further, their effort to focus on effects per district fails to account for the reality that news travels across district borders that exist only on paper. Why is a terror attack that happens in Tel Aviv more likely to worry the people of Tel Aviv than the people of Herzeliya only a short distance to the north? Certainly an attack happening anywhere within the country will be on the news throughout the country in every town, every district and in every resident’s morning paper.
There is far more to discuss regarding this paper, but I think some of the more important points have been made. While the paper does note that the platforms of the Israeli right have changed to include references to making “territorial concessions,” it’s probably a stretch to call this a move to the left. The whole world has wised up to the fact that the Israeli occupation of Palestinian land must end. The fact that Israeli political parties are starting to see this is more a reflection of a general awakening across the globe of what is acceptable vis-à-vis the occupation, than it is a reaction to acts of violence in particular Israeli districts.
