Retaliating against Terrorists: Erratum, Reanalysis, and Update
Nevin, John A., Behavior and Social Issues
Several commentators have raised questions about my article, "Retaliating against terrorists" (Nevin, 2003). In that article, I examined sequences of terrorist attacks and government responses in seven terror campaigns and concluded that there was no reliable evidence that violent retaliation either increased or decreased the average intensity of terrorist attacks. The conclusion was based on a quantitative treatment of accounts taken from The New York Times, and as such may depend on the faithfulness of my transcription and on the particulars of the data analysis. Two difficulties have been brought to my attention.
First, one reader noted that there are errors in Table 2, which was intended to exemplify the way in which I treated the ordinal-scaled data on the severity and intensity of terror attacks and government responses.1 I apologize for any confusion that readers experienced, and here present a corrected version of Table 2.
Second (as I acknowledged) it is not proper to calculate correlation coefficients based on ordinal data. Although correlation coefficients were used only to describe relations between sequences of events, and not to evaluate statistical significance, some readers have found them to be misleading. In the interest of quantitative propriety and accuracy, therefore, I reanalyzed all seven data sets using reported deaths (a ratio scale) rather than scaled severity. The result is that fewer incidents are available for analysis, but the conclusions based on my reanalysis are unchanged. Figure 1 is based on calculation of deaths per month caused by terror attacks and by government responses aggregated over five successive incidents involving activity by terrorists or authorities. The terrorists' victims included government officials, soldiers, police, and civilians; the governments' victims included civilians as well as the targeted terrorists. The y-axis displays the rate of killing by terrorists after a governmental response relative to the rate of killing before the governmental response. The scale is logarithmic so that (for example) two-fold increases and two-fold decreases are equal in their distance from 1.0 (no change). The x-axis displays the death rate inflicted by governmental authorities over the five-episode block that intervened between the two samples of terrorist killing rates, also on a logarithmic scale.
Inspection of Figure 1 suggests that the points are about equally likely to be above or below 1.0 on the y-axis, even for cases where there were no deaths attributable to government action (the vertical stack of points above 0.2 on the x-axis). The slope for all data pooled is 0.0002-essentially zero. This analysis, and several others reported elsewhere,2 confirm the original conclusion: There is no evidence that retaliating against terrorists either decreases or increases the average severity of subsequent terror attacks.
Several people have asked whether the severity of governmental responses is related to the severity of preceding terror attacks. To address the question, I reversed the analysis of Figure 1, and obtained a slope of 0.0234, which does not differ from zero. At least on average, then, there is no evidence that authorities either increase or decrease the severity of their responses to terror attacks as a function of the severity of those attacks.
The form of Figure 1 is the same as that used in research on resistance to change, also known as behavioral momentum (for review see Nevin & Grace, 2000). In these analyses, based largely on research with pigeons and replicated with humans and rats, the slope of the function relating relative changes in behavior to the intensity of a disrupter measures the persistence of that behavior. Laboratory research has found repeatedly that the slope of the function depends inversely on the rate and magnitude of reinforcement: The larger or more frequent the reinforcer, the shallower the slope. By analogy, then, it appears that terror campaigns generate substantial reinforcement for their leaders and participants. …