"Hit or Miss? The Effect of Assassinations on Institutions and War", Jones & Olken 2007; excerpts:

"Assassinations are a persistent feature of the political landscape. Using a new data set of assassination attempts on all world leaders from 1875 to 2004, we exploit inherent randomness in the success or failure of assassination attempts to identify assassination’s effects. We find that, on average, successful assassinations of autocrats produce sustained moves toward democracy. We also find that assassinations affect the intensity of small-scale conflicts. The results document a contemporary source of institutional change, inform theories of conflict, and show that small sources of randomness can have a pronounced effect on history.

To the best of our knowledge, the only related paper along these lines is Zussman and Zussman (2006), who find evidence that assassinations of senior members of Palestinian organizations affect Israeli stock returns.

Analyzing the effects of assassination is difficult. While some assassinations may be associated with historical turning points, the direction of causation is difficult to establish, especially since assassination attempts often occur (as we will show) in times of crisis, such as during war. To overcome this problem, we employ a large set of assassination attempts and use the “failures” as controls for the “successes”. To focus on the cases where the success or failure of the attempt was most likely determined by chance, we consider only those attempts in which the weapon was actually used – the gun fired, the bomb exploded, etc. The identification assumption is that, although attempts on leaders’ lives may be driven by historical circumstances, conditional on trying to kill a leader the success or failure of the attempt can be treated as plausibly exogenous. For example, Hitler’s early departure from the beer hall in 1939, which may have saved his life, came only because bad weather prevented him from flying back to Berlin, forcing him to leave early for a train.

We find that assassinations of autocrats produce substantial changes in the country’s institutions, while assassinations of democrats do not. In particular, transitions to democracy, as measured using the Polity IV dataset (Marshall and Jaggers 2004), are 13 percentage points more 2likely following the assassination of an autocrat than following a failed attempt on an autocrat. Similarly, using data on leadership transitions from the Archigos dataset (Goemans et al., 2006), we find that the probability that subsequent leadership transitions occur through institutional means is 19 percentage points higher following the assassination of an autocrat than following the failed assassination of an autocrat. The effects on institutions extend over significant periods, with evidence that the impacts are sustained at least 10 years later.
Looking at military conflict, the results show that assassinations affect conflict, but only in limited contexts. We examine two data sources: the Gleditsch-Correlates of War dataset (Sarkees, 2000; Gleditsch 2004) and the PRIO/Uppsala Armed Conflict Database (Gleditsch et al. 2002). We find that successful assassination lead to an intensification of small-scale conflicts relative to failed assassination attempts. For high-intensity conflicts, we find somewhat weaker evidence that successful assassinations may have the opposite effect, hastening the end of large-scale conflicts already in progress. These results suggest heterogeneous effects of assassinations that depend on conflict status.
All of these results tell us about the difference in outcomes following success and following failure. Our approach does not distinguish whether the effects are driven by successful assassination (e.g., killing an autocrat leads to more democracy), failed assassination (e.g., trying but failing to kill an autocrat leads to increased suppression), or both.

To establish a baseline list of leaders, we use the Archigos dataset, v2.5 (Goemans et al., 2006), which identifies the primary leader for each country at each point in time from 1875 to 2004. Archigos provides a data set of 2,440 leaders from 187 different countries.
To collect the assassinations data, we consulted the archives of three major newspapers: The New York Times, The Washington Post, and The Wall Street Journal. We used a large set of keyword searches (detailed in the Appendix) and placed several limitations on the returned results. First, we excluded coup d’etats – cases in which the murder or attempted murder of the leader was conducted by an individual or group in an attempt to seize power for themselves. Second, we excluded “uncovered plots” to assassinate leaders, limiting ourselves to cases in which the would-be assassins actually undertook the attempt. For the main specifications in the paper, we further restrict our attention to “serious attempts,” which we define as those cases in which the weapon (the gun, bomb, etc.) was actually discharged, as opposed to cases where the attempt was thwarted prior to the weapon being used. As shown below, our results are broadly robust to different restrictions on the nature of failed attempts.
For each assassination or attempted assassination found, we recorded the date and location of the attack, the weapons used, and the result for the leader, as well as information when available on other casualties and whether the attack was carried about by a group or solo actor. The data includes 298 assassination attempts, of which 251 are “serious attempts” and 59 result in the leader’s death. A list of the successful assassinations is presented in Table 1. To ensure that the data collection methodology captured all relevant assassinations, once the newspaper searches were complete we cross-referenced the assassinations found by the searches with all assassinations listed in da Graca (2000), Jones and Olken (2005), and the Archigos data. This exercise showed that our keyword searches produced all relevant assassinations.4,5...It is more difficult to conclusively assess our effectiveness in capturing assassination attempts; however, there are several reasons to believe that our method was effective. First, we ran the keyword searches sequentially, first with the New York Times, which produced 263 attempts, then the Washington Post, which produced an additional 33 attempts, and then the Wall Street Journal, which produced only 2 additional attempts. The rapidly diminishing returns to further searches suggest that we are accurately capturing publicly-known assassination attempts.

With regard to weapons, guns have been the most common instrument, used in 55% of attempts, and explosive devices the second most common, used in 31% of attempts. Guns have kill rates of about 30%, while explosive devices are much less likely to kill the leader, with success in only 7% of cases where the device was actually engaged. At the same time, explosive devices produce the greatest number of casualties among bystanders, with the mean number of dead and wounded six and eight times larger than for gun attacks. Explosive devices thus appear to be both a particularly violent and particularly ineffective tool.6
Table 2 further shows that the vast majority of assassination attempts occur in the leader’s home country, with only 4% occurring outside the national borders. Attempts are slightly more likely to be carried out by solo assassins than by groups of assassins (59% to 41%). Both solo and group attacks show a similar propensity to kill the leader, although group attacks tend to be far bloodier for bystanders.

Panel A indicates that the annual rate of assassinations increased in the late 19th and early 20th century, decreased substantially during the 1940s (perhaps as a result of heightened security during World War II), and has been at relatively high levels since 1950. Currently, the world witnesses the assassination of a national leader in one of every two years. Interestingly, the frequencies of attempts and successes closely track one another. In fact, the conditional probability of killing a leader given a serious attempt is not trending, remaining at about 25% through time.

To identify this effect we employ the inherent randomness in whether an attack is successful or not. For example, John F. Kennedy did not escape the bullet that killed him, even though it was fired from 265 feet away and the president was in a moving car (Warren et al., 1964). But Idi Amin did survive an attack in 1976, when a thrown grenade bounced of his chest and killed several bystanders.

Table 5 breaks down these effects by the tenure of the leader at the time of the attempt and by the duration of the effect. Each cell reports the coefficient on SUCCESS from a separate regression, where the sample is shown in the column and the duration of the change used to calculate the dependent variable is shown in the row. The top panel indicates that the short-run move to democracy is particularly large following the assassination of long-tenured autocrats, for whom a successful assassination increases the probability of democratic transition in the next year by 20 percentage points relative to a failed assassination. The distinction between tenure is less clear with time however. The most interesting result in this table appears in Panel A column (4), which shows that democratic transitions following assassinations of autocrats appear to be sustained 10 years later. The point estimate suggests that initially autocratic regimes are 19 percentage points more likely to be democracies 10 years after the attempt if the assassination succeeded rather than failed. Twenty years into the future, however, the results are substantially attenuated using the binary Polity IV measure.

Looking at Table 6, we see three primary results. First, there is weak evidence that successful assassination attempts, compared to failed assassination attempts, tend to hasten the end of intense wars (i.e. wars with greater than 1000 battle deaths). This effect appears in Panel B, column (1), and suggests that successful assassination lowers the probability of continued, intense conflict by 25 percentage points. Although the effect is quite large in magnitude, it is only marginally significant (P-value of .08 parametrically and .13 non-parametrically) and is not significant when we restrict to the post World War II period. The post-war results are difficult to interpret, however, because there are few observations of intense wars after 1946. Overall we conclude that there is some evidence, but only weak evidence, for an effect on intense wars. Second, there is evidence that successful assassination attempts, compared to failed attempts, lead to increased intensity of existing moderate-level conflicts. This is seen in Panel B, column (3), where we see a 33 percentage point increased probability that a war intensifies when the leader is killed. This large point estimate shows some significance (P-value of .05 parametrically and .13 non-parametrically) even though the sample size is substantially smaller due to the fact that the PRIO data exists only for the post-1945 era.

[backfire effects?]

We start by considering whether assassination attempts are predictable and find that they are – and in interesting ways. Table 8 shows the results of estimating (4). The annual rate of assassination attempts is 0.7 percentage points higher in autocracies than in democracies. The baseline probability of an attempt in a given country-year is 2.4%, so this implies that autocrats are approximately 30% more likely to be the target of attacks in a given year. Attacks are also 2.8 percentage points more likely during wartime - more than doubling the background probability – which makes war a particularly powerful predictor of assassination attempts. Interestingly, these results are consistent with the results in Section 3, which showed that assassinations of autocrats had an impact on institutional change, and that assassinations had an impact on wars that were in progress. The results here suggest that potential assassins may understand that assassinations against autocrats or wartime leaders are more likely to have an effect, and hence are more likely to attempt to kill precisely those leaders where it would make a difference.22
These results are broadly consistent with the findings of Feierabend et al. (1971) and Iqbal and Zorn (2003). Feierabend et al. consider the correlates of assassination attempts from 1948-1967, and, consistent with our findings, find that assassination attempts are more common in poorer countries, more autocratic (or, in their terminology, more coercive) countries, and in countries involved in war. Iqbal and Zorn consider predictors of successful assassinations since World War II and find, as we do, that political institutions and war predict assassination. Both studies are limited to the question of predicting assassinations, rather than assessing the consequences of assassination.

Another interesting result that emerges in Table 8 is that attempts are more common in countries with larger populations; doubling the population increases the probability of an assassination attempt each year by 0.35 percentage points. Though this may seem like a small effect, this implies that the leader of a country the size of the United States (population 300 million) is 1.8 percentage points, or about 75 percent, more likely to be assassinated each year than the leader of a country the size of Switzerland (population 7.5 million). This population effect is sustained in a multivariate context, so that it does not appear to proxy for per-capita income, institutions, or war status. One natural interpretation is that the number of would-be assassins rises with a country’s population, whereas there is only one leader in each country. The ratio of would-be assassins to leaders, and hence the probability of an attempt, therefore increases with population.

We find several interesting results. Keeping in mind the caveats about identification in this section discussed above, we see that most of the effects identified in Section 3 appear to be driven by successful assassinations, though there are some cases in which failures may have effects. The first three columns on Table 9 investigate the absolute value of changes in the POLITY2 dummy. The results here suggest that it is successful assassinations that are driving the results. Similar insight is provided by the second set of columns, which consider moves toward democracy. Examining autocrats, successful assassination increases the probability of democratic transition in the next year by 13 percentage points compared to years without assassination events, while failed assassinations suggest a modest and statistically insignificant 1 percentage point fall in the probability of democratic transition. The effects of failure are amplified when we consider the percentage of “regular” leader changes in the ensuing 20 years, where successful assassinations of autocrats are associated with 16-21 percentage point increases in the percentage of regular future leader transitions while failure is associated with 6-7 percentage point declines in the percentage of such leader transitions. If these estimates of failure actually represent the true causal effect of a failed assassination (as opposed, perhaps, to selection effects not controlled for perfectly with the propensity score methodology), then this would suggest that autocrats may slightly tighten their grip on power after failed assassination attempts.

Given that only 25% of assassination attempts are successful, if we take the point estimates in Table 9 literally, they imply that the average effect of assassination attempts on democracy is only modestly positive ex-ante, if positive at all. Overall, the results imply that one would expect a 6-7 percentage point move toward democracy if the assassination succeeds (approximately 25% of the time), and a 2 percentage point move towards autocracy with failure (approximately 75% of the time), suggesting an approximately zero net effect on average. Focusing on autocrats, meanwhile, suggests a modest, positive move to democracy in expectation, with the point estimates implying a 3% ex-ante increased probability of democratization from assassination using the POLITY2 measure and essentially no mean shift ex-ante using the Archigos measure of future leader transitions – far smaller than the 15-20% average move to democracy comparing success with failure. Thus, a policy of assassination attempts creates risk – it increases the probability that there will be a change in a country’s institutions – but if the probability of an attempt succeeding is 25%, there are at most modest gains in democracy on average."

#terrorism #assassination #politics