• Non ci sono risultati.

CHAPTER VII. Discussion 188

Discussion

CHAPTER VII resumes the main findings of the dissertation, and discusses them in light of the research questions and the theoretical framework that were initially formulated. Across several subchapters, also the limitations of this dissertation, its policy relevance, and perspectives for future research on the topic are discussed.

VII.1. Main findings 189

The strength and significance of the findings varied according to different calculation methods, different types and numbers of control variables introduced in the models, the lagging of the independent variables of interest, and different world regions. Fixed effects models were found to be more appropriate than random effects models which could be confirmed by Hausman tests. The pooled models did not meet the appropriate assumptions for the structure of the dataset, while the between models did not produce significant results.

VII.1.1. Effects of terrorism on homicide

As far as the models on the effects of terrorism on homicide were concerned, introducing either fixed (individual) or random effects bore only marginal differences in terms of the strength of the coefficient, and no effect on their significance. Stronger differences became apparent when comparing the fixed effects models with the pooled and between models. Both the pooled and between models exhibited coefficients for the terrorism mortality rate (log) that were roughly twice as high. While the coefficient in the pooled models were highly significant, the coefficients in the between models were not. The implications of this are twofold. Firstly, there appear to be omitted variables varying at country level that mediate part of the positive effect between terrorism and homicide. Secondly, the terrorism mortality rate can significantly predict differences within countries over time, but not differences between countries. Apart from the control variables and omitted variables varying at the country level, there seem to also be common variables varying over time that influence the relationship between terrorism and homicide. This becomes apparent when comparing the results for the individual and two-ways fixed effects models. The additions of time-fixed effects lead to a drop in the coefficient by roughly a sixth.

In comparing different world regions, the most robust findings were available for Europe. The coefficient for the terrorism mortality rate (log) in Europe was roughly twice as high as the corresponding coefficient in the overall model, indicating that increases by 1 percent in the terrorism mortality rate are associated with increases in the homicide rate by almost 0.5 percent.

Calculated with time-fixed effects and robust standard errors, the coefficient remained insignificant for each of the other world regions. The clustering of all regions other than Europe, on the other hand, led to a coefficient that was significant at 0.1 level and close in strength to the coefficient measured in the overall model. The interpretation of regional differences in the findings remained difficult and may possibly relate to both the limitations of

CHAPTER VII. Discussion 190

the dataset and fundamental aspects that remain to be discussed in the following sections of the discussion.

The lagging of the terrorism mortality rate (log) by one year caused further changes to the coefficient. It increased rather strongly to roughly 0.3. As for regional differences in the lagged model, the coefficients for Europe and for the group of regions other than Europe exhibited roughly the same strength as in the overall model. For Europe, this meant that the coefficient dropped rather sharply due to the lagging, namely from increases of roughly 0.48 percent in the homicide rate for every 1 percent increase in the terrorism mortality rate to roughly 0.3 percent. Also, the significance of the coefficient dropped considerably for Europe, namely from the highest to the lowest level (p<0.1), while the significance of the coefficient for the group of other regions increased (to p<0.05). A possible explanation for this may be that the terrorism mortality rate influences the homicide rate stronger in the short term. A reason why this effect did not become visible for the group of other regions may be caused by covariance between terrorism and warfare in combination with stronger long-term effects of warfare (compare with section VII.1.2 below).

Given that both the foregoing examination of trends in terrorism, warfare and other violence, and the calculation of the correlation between them, pointed to a strong overlap, separate models for terrorism were run for those country-years that did not exhibit any kind of warfare.

While the significance of the coefficient climbed to the highest level, that almost led to a doubling of the effect, namely from 0.24 percent increases in the homicide rate for every 1 percent increase the terrorism mortality rate, to 0.42 percent. It appears that the isolated effects of terrorism on homicide are much stronger in settings where no warfare occurs. But again, the measurements may be biased due to a potentially disruptive effect of war on the underlying data. Comparing the coefficient for the terrorism mortality rate (log) between different world regions for those country-years that did not experience any warfare showed that the increase in the effect is entirely due to countries in regions other than Europe. The coefficient for Europe remained precisely the same as almost no complete observations for European country-years in which warfare occurred were available. For the group of other countries, however, it rose to 0.37 percent increase the homicide rate for every 1 percent increase in the terrorism mortality rate, and gained significance. When also excluding those country-years from the dataset which exhibited major violence other than warfare, the coefficient for the terrorism mortality rate (log) dropped in strength and became insignificant. This confirmed that in fact many

VII.1. Main findings 191

occurrences of terrorism were reflected in the MEPV dataset where they were coded as other violence.

Considering measurements of terrorism other than the terrorism mortality rate (i.e. absolute counts of attacks, lethal attacks and number of deaths) to predict the homicide rate did not lead to consistent and significant results. As it seems, not only the lethality of terror attacks, but also their lethality in relation to the size of the overall population of the country where the attacks occurred, play an important role as to whether terror attacks lead to an increase in the homicide rate. This links to Landau and Pfeffermann’s (1988, 500) conclusions about the effects of prolonged states of warfare on homicide in Israel. They maintained that “what really affects the number of homicides is not just the existence of security-related tension but rather the occurrence of security-related loss of, or injury to, human life.”

The addition/omission of the polity index score (autocracy/democracy) appeared to play a somewhat mysterious role in moderating the association between the terrorism mortality rate (log) and the homicide rate (log). Without ever being significant in itself, its addition to the model led to a rather considerable drop in the terrorism mortality rate (log) coefficient, namely from roughly 0.3 to a 0.24 percent increase in the homicide rate. So far, studies on the relationship between democracy and either homicide rates or terrorism have yielded ambiguous results (cf. LaFree and Tseloni 2006; LaFree and Ackerman 2009, 362).

VII.1.2. Effects of warfare and other major violence on homicide

The models about the effects of warfare on the homicide rate behaved somewhat differently than the models on terrorism. To begin with, again, calculating either fixed (individual) or random effects produced only marginal differences in terms of the strength of the coefficient, in this case the warfare magnitude scores (log). Rather than suggesting stronger effects like in the case of the terrorism mortality rate, the calculation of pooled and between models led to a drop in the strength of the coefficients. This points to additional omitted variables varying between countries that may moderate the effect of warfare on terrorism. The estimates in the between models were, again, insignificant which indicated that warfare may significantly predict changes in the homicide rate within countries over time, but not between different countries. The coefficient was thereby shown to be sensitive to the type of control variables that were introduced. With a 0.19 increase in the homicide rate for every 1 percent increase in the warfare magnitude score, the coefficient was much lower when introducing the economic control variables while their omission suggested a much stronger effect (0.33). This points

CHAPTER VII. Discussion 192

rather clearly to socioeconomic effects of warfare and their indirect effects on homicide. While studying homicide rates after nation-wars (international warfare), Archer and Gartner (1978, 954) discussed this topic using the “economic factors model”, but found no evidence in its support. However, the results presented in the present study, show that a good part of the effects that different types of warfare bear on the homicide rate may be mediated by precisely the socioeconomic indicators that were introduced as control variables to the models. Thus, as for the effects of warfare on homicide, hypothesis h3 (effects of collective violence mediated through socioeconomic impact) is supported by the findings.

A comparison of different world regions did not produce significant results. As for Europe, the calculation was not feasible as there were almost no complete observations in which European countries experienced warfare.

Similar to the models on terrorism, the addition of time-fixed effects (two-ways) to the full model led to a drop in the strength of the warfare magnitude (log) coefficient to roughly a 0.14 percent increase in the homicide rate for every 1 percent increase in warfare magnitude. When calculated with robust standard errors, the coefficient was only weakly significant (p<0.1).

This changed when the independent variable of interest was lagged by one year. Even though the strength of the coefficient remained the same, its significance rose to the highest level.

Eventually, when lagging the warfare score again but using a five-year average for the homicide rate rather than a single year rates, the coefficient more than doubled in strength. This gives rise to the conclusion that the impact of warfare on homicide may lead to a stronger increase in homicides over time. Besides that, however, warfare may lead to disruptions in the underlying data that may make it difficult to measure its effects in contemporaneous models.

The relationship between homicide rates and major violence other than warfare were also assessed. This did not lead to significant results, however, and may be attributable to limitations of the underlying dataset as discussed in section VII.3.1 ahead. However, the fitting of a combined model about the effects of collective violence on homicide, with the combined magnitude scores of major episodes of political violence (warfare and other violence) as the independent variable of interest, eventually led to a very robust result. The coefficient was stronger and more significant than for any of the datasets individually, suggesting that 1 percent increases in collective violence would lead to roughly 0.18 percent increases in the homicide rate.