How Cases Performed?

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

index. As to “primary commodities dependency”, with non-liner relation with civil war, a case has a CWP index when the case in the “middle” group.²⁵

It is natural to assume that CWP index is accumulative. In other words, a case tends to encounter civil war, when the case has higher number of CWP index. Based on the assumption, my standard for a theory or a combination of theories with better explanatory ability is: within the theory or the combination, most cases with high number of CWP index certainly encounter civil war.

3.4. How Cases Performed?

In this part, I would like to assess the explanatory ability of the theories. There are mainly two parts: (1) deciding which way is better to apply theories, that is, I will make a choice by taking certain combination of theories or just choosing one of them into further discussion; (2) judging which independent variable might be the necessary conditions behind the difference on the level of political violence/movement during the Arab Spring.

3.4.1. How to Apply Theories?

As mentioned in Section 3.3, my standard to assess the explanatory ability is: (1) how successfully certain one or combination of theories predicts that cases with more CWP indexes will be struck by civil war; (2) parsimoniousness, that is, explaining by less variables.

Firstly, I will evaluate the predictive ability of the combination of those three theories in two situations: including and excluding independent variables with missing value.²⁶ As Appendix III show, the tendency presented is roughly fit the

25 According to the theoretical base, there is non-monotonic correlation between primary commodities dependency and civil war. In other words, when a case’s value on the variable is around 33%, the possibility of civil war will be the largest. The cases fell within “middle” group include Yemen with around 25.9%, Iraq with around 33.9%, Algeria with 34.6 %, Saudi Arabia with 42.3 % and Qatar with 46.1%, and they are close to typical case with 33%. As a result, I choose “middle” on primary commodities dependency as CWP index.

26 By excluding independent variables with missing value, we can make a comparison based on relatively consistent criteria. Specifically speaking, we can avoid the extreme situation that: a case, in fact, might have a lot of CWP indexes, but, due to the unavailability of relevant data, most of its performance on independent variables are set as missing value.

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

prediction of civil theories, that is, cases with more CWP indexes tend to have civil war. The four cases with the highest index number include Algeria (9), Iraq (9), Yemen (9), and Syria (7), and almost all of them went through civil war in the Arab Spring, with Algeria as the only exception. Similarly, most cases out of the top four did not experience civil war in 2011, and the only exception was Libya (2). Excluding independent variables with missing value from discussion, we can even get a similar result. Cases experiencing civil war have larger number of CWP indexes, for example Iraq has 8, and both Syria and Yemen have 6. Two exceptions are: Algeria has 6 indexes but did not encounter civil war during 2011; Libya has only 2 indexes but went through a harsh civil war. To some extent, the combination of the three theories has predictive power for the difference among the level of political violence/mobilization during the Arab Spring, though with two outliers: Algeria and Libya.

Secondly, I will evaluate if it remain enough ability to explain the differential level of political violence/mobilization by using only two of three theories; similarly, under both situations including and excluding variables with missing value. Under the framework of Boix plus F-L, as Appendix IV show, in terms of CWP index number, the top four are Algeria (6), Yemen (6), Iraq (5), and Egypt (4), and other cases has less than 4 indexes. It causes more outliers, that is, Syria, with only 3 indexes, drops out from the group predicted to have civil war by theories, and Egypt, with 4 indexes, seems to participate in the group with high risk of civil war. The two abnormal cases in Appendix III, Algeria and Libya, still exist. Excluding variables with missing value, it leads to a more problematic result: both Algeria and Iraq has 4 indexes, so it is natural to rank them as the top two. However, it is hard to decide the third and forth based on the prediction of the two theories, because Egypt, Syria, and Yemen all have 3 indexes. In short, the combination of Boix and F-L does not work better at explaining than the three theories combination.

By the withdrawal of F-L, under the combination of Boix and C-H, the top four among fifteen cases are Yemen (7), Algeria (6), Iraq (6), and Syria (6) in Appendix V.

There is no increase or decrease on the number of outlier, which means no evident change in predictive power. Yemen, Syria, and Iraq, with, real happening of civil war, are predicted to have civil war. Still, Algeria and Libya are out of expectation of theories. Given taking away variables with missing value, the top four are Iraq (5), Syria (5), Yemen (5), and Algeria (4). To sum up, the combination of Boix and C-H works not worse than the three theories combination, and it is more parsimonious.

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

Within the combination of C-H and F-L by pulling out Boix, the group predicted to experience civil war, the top four, includes Algeria (8), Iraq (8), Syria (7), and Yemen (7) in Appendix VI. Most cases with 5 CWP indexes and below did not encounter civil war in 2011, except for Libya (1). In other words, there is also no change of the abnormal cases’ number. Excluding the variable with missing value, Iraq (8), Algeria (6), Syria (6), and Yemen (6) still fall into the CWP group. As a result, the combination of C-H and F-L works not worse than the combination of all theories but is more parsimonious.

Basically, the combination of all theories, both the combinations of Boix and C-H and of Boix and F-L have great explanatory power for the difference in the level of political violence/mobilization. However, for the pursuit of parsimoniousness, the last two combinations prevail against the former. Now, the question is: between the other two combinations, which one is better?

Then, we need to evaluate the predictive power of two combinations’

components, that is, test Boix, C-H, and F-L respectively. About F-L, as in Appendix VII, the boundary between high CWP group and low CWP group is ambiguous.

According to my criterion selecting the top four among fifteen cases, The top three are clearly: Algeria (5), Iraq (4), and Yemen (4), but the forth is indeterminate, because there are three cases with 3 indexes, that is, Egypt, Syria, and Morocco. The problem still exists after deleting variable with missing value. The top two are cases with 4 indexes: Algeria and Iraq. But there are three cases with 3 indexes competing for the two vacancies left. When we only take F-L into account, the predictive power is not better than the above two combination.

In Appendix VIII, under the framework of Boix, there is no evident trend. The case with most CWP indexes is Yemen (2), and there are four cases with 1 index for the three vacancies left: Algeria, Egypt, Iraq, and UAE. In general, there are five outliers: Libya and Syria, with the onset of civil war in 2011, fall out of the group expected to have civil war; Algeria, Egypt, and UAE, without experiencing civil war during the Arab Spring, are predicted to have civil war. In short, the predictive power of Boix is worse.

Under C-H, as Appendix IX show, Syria (6), Yemen (6), Iraq (5), and Algeria (5) are expected to have civil war, and after withdrawing variable with missing value, the top four includes Yemen (6), Iraq (5), Syria (5), and Algeria (4). Similar to the two theories’ combination, in the prediction of only C-H the outlier are Algeria, with more CWP indexes but without truly happening civil war, and Libya, with less indexes but

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

suffering from civil war during the Arab Spring. In sum, C-H, alone, has good predictive power and is more parsimonious than the two above-mentioned combinations: Boix plus C-H and C-H plus F-L.

After evaluation of the predictive power of each theories or combination of theories, I decide to exclude Boix and to keep C-H and the combination of F-L and C-H for further discussion. Boix has not good explanatory ability for the situation during the Arab Spring, so that we can withdraw it. With the same reason, we can also exclude the combination of Boix and C-H. In contrast, C-H has good predictive power and is parsimonious, that is, C-H might be the best choice to interpreting the differential level of political violence/mobilization during the Arab Spring. However, I still keep the combination of C-H and F-L, though less parsimonious, because it has some variables deserving deeper discussion.

3.4.2. Which Independent Variables Might be Necessary?

In this stage, I will decide which independent variables are closer to necessary conditions for the happening of civil war during the Arab Spring. To judge those variables necessary or not, I use two criteria as below: (1) if the CWP index frequently appears among cases experiencing civil war in 2011; (2) if the CWP index only appear in cases encountering civil war. Given an independent variable fits these two standards, it might be the necessary condition.

The object for judging includes variables within the combination of C-H and F-L, and I only take variables without missing value into account here.²⁷ That is, the independent variables under scrutiny are listed as below: “anocracy”, “ethnic domination”, “GDP growth rate”, “GDP per capita”, “middle secondary schooling”,

“oil exporter”, “population”, “primary commodities dependency”, “peace duration”, and “political instability”. All of them are within the combination of C-H plus F-L and without missing value.