Appendices/The Ethno-geography of Oil and the Onset of Ethnic War
Appendix A: Countries within our Dataset and Summary Statistics
A1. Countries within Our Dataset
Afghanistan, Albania, Algeria, Angola,Argentina, Armenia, Austria, Australia, Azerbaijan, Bangladesh, Belarus, Belgium, Benin, Bolivia, Bosnia and Herzegovina, Botswana, Brazil, Bulgaria, Burma, Burundi, Cambodia, Cameroon, Canada, Central African Republic, Chad, Chile, China, Colombia, Costa Rica, Cote d'Ivoire, Croatia, Cuba, Czech Republic,Democratic Republic of the Congo, Dominican Republic, Ecuador, Egypt, Eritrea, Estonia, Ethiopia, Finland, France, Gabon, Gambia, Georgia, Ghana, Greece, Guinea, Guatemala,Guinea-Bissau, Haiti, Honduras, Hungary, India, Indonesia, Iran, Iraq, Israel, Japan, Jordan, Kazakhstan, Kenya, Kuwait, Kyrgyzstan, Laos, Latvia, Lebanon, Liberia, Lithuania, Macedonia, Madagascar, Malawi, Malaysia, Mali, Mauritania, Mexico, Moldova, Mongolia,Morocco,Mozambique, Namibia, New Zealand, Nepal, Netherlands, Nicaragua, Niger, Nigeria, Panama, Paraguay, Pakistan,Peru, Philippines, Poland,Republic of the Congo, Romania,Russia, Rwanda, Saudi Arabia, Senegal,Slovakia, Slovenia, Spain, Sri Lanka,Switzerland, South Africa, Sudan (formerly United), Syria, Taiwan, Thailand,Tajikistan, Trinidad and Tobago, Togo, Tunisia, Turkey, Turkmenistan, Uganda, Ukraine, United States, Uzbekistan, United Kingdom, Venezuela, Vietnam, Yemen, Zambia, Zimbabwe
A2. Control Variables
Following standard procedures, our control variables are also divided into two levels: group and country.At the group level, we control for the same set of variables as inCederman et al (2010) and Wucherpfennig et al (2011) because we wish to present our results as comparabletotheir results as possible: doing so makes our results easier to interpret. At the group level, we thus control for the size of the population of the minority group, GDP per capita (country), size of the territory occupied by a group, war history (i.e., previous conflict), and peace years.We also control two additional geographical variables that have been shown to be significantly associated with ethnic war: a minority group’s distance to the state capital (positive), a minority group’s distance to the nearest border (negative). We also control for “excluded” (a group is excluded from the central government)and “downgraded” (i.e., a group that used to enjoy a more advantageous political status is now downgraded), two political indicators from the EPR dataset that have been shown to be strong predicators of ethnic conflict (Cederman et al 2010; Weidmann et al 2010; Wucherpfennig et al 2011).
At the country level, wecontrol for the same set of variables as in Wimmer et al (2009), again because we wish to present our results as comparable to their results as possible: doing so makes our results easier to interpret. At the country level, we thus control for the ratio of the excluded population, center segmentation, imperial past, GDP per capita, size of the population, ethno-linguistic fractionalization (ELF), regime change, and anocracy etc. Finally, we also control for oil production per capita (Ross 2012).
We do exclude control variables that are almost certainly endogenous to the onset and duration of ethnic and non-ethnic civil war such as Polity score and GDP growth rate, as pointed out by Sambanis (2004b) and Hegre and Sambanis (2006).
Table AA-1: Summary Statistics
- Group level
N / mean / Standard deviation / min / max
Ethnic conflict (EPR) / 30754 / 0.0049 / 0.07 / 0 / 1
Ethnic conflict(ACD) / 30754 / 0.01 / 0.11 / 0 / 2
Oillocation (USGS) / 30754 / 0.32 / 0.47 / 0 / 1
Oillocation (Lujala) / 17591 / 0.42 / 0.49 / 0 / 1
Distance to capital / 24450 / 4.86 / 2.00 / 0 / 8.70
Distance to border / 24450 / 2.39 / 1.96 / 0 / 8.70
excluded / 29179 / 0.61 / 0.49 / 0 / 1
downgraded2 / 30754 / 0.01 / 0.11 / 0 / 1
Size of core territory / 29198 / -3.11 / 2.24 / -10.8 / 0
War history / 30754 / 0.11 / 0.36 / 0 / 3
GDP per capita (ln) / 30090 / 7.87 / 1.12 / 3.27 / 11.61
Size of population (ln) / 30273 / 10.07 / 1.85 / 5.58 / 14.08
Oil price / 30673 / 15.12 / 12.81 / 1.12 / 54.52
- Country level
N / mean / Standard deviation / min / max
Oil location (USGS) / 7843 / 0.48 / 0.50 / 0 / 1
Oil location (Lujiala) / 4871 / 0.66 / 0.47 / 0 / 1
Ethnic conflict (EPR) / 7843 / 0.02 / 0.12 / 0 / 1
Ethnic conflict(ACD) / 7843 / 0.03 / 0.23 / 0 / 2
Ratio of excluded population / 7826 / 1.86 / 1.57 / 0 / 4.60
segmentation / 7826 / 1.67 / 1.86 / 0 / 14
Imperial past / 7843 / 0.47 / 0.31 / 0 / 1
ELF / 7843 / 0.39 / 0.28 / 0.001 / 0.925
GDP per capita (ln) / 7712 / 8.10 / 53.36 / 0.031 / 3302.924
Size of population (ln) / 7746 / 9.22 / 1.40 / 5.605 / 14.099
Area of mountainousterrain(ln) / 7843 / 2.22 / 1.38 / 0 / 4.421
Regime change / 7843 / 0.12 / 0.32 / 0 / 1
Anocracy / 7648 / 0.24 / 0.43 / 0 / 1
Oil production per capita (ln) / 7746 / 2.03 / 12.55 / 0 / 270.931
Ongoing war / 7687 / 0.17 / 0.38 / 0 / 1
Oil price / 7843 / 20.72 / 21.81 / 1.12 / 97.26
Source of Oil Location Data: this paper (for details, see main text)
For detailed descriptions of the dependent and control variables at the group and country level, see Wucherpfennig et al (2011); and Nils-Christian Bormann (2011), “Codebook forGeoEPR-ETH,” Nov. 11, 2013.
Appendix B: Additional Robustness Checks with USGS Data
In this Appendix, we report results from two sets of further robustness checks.
The first set of robustness check is using the product of oil-location and oil price (variable name, oilloc_price) as the key explanatory variable. As explained in the main text (section IV), we reason that oil price is mostly exogenous to any particular ethnic or non-ethnic civil war within a country because oil price has been mostly driven by consumption and overall production in the world rather than by any particular ethnic or non-ethnic civil war. Oil price is from Ross and Mahdavi (2015). At the country level, we limit the sample to only oil-producing countries because anything occurred within countries without any oil (production) should not have anything to do with the oil within them (they have none).
What should be noted here is that the product of oil location and oil price is not a conventional interactive term because oil price does not vary across groups or countries. As such, in regressions, we do not include oil location and oil price when using the product of oil location and oil price as the key independent variable.
In table AB-1, we show that using the product of oil-location and oil price performs as the key explanatoryvariable leads to almost identical results: the product of oil-location and oil price is significantly and robustly associated with the onset of secessionist ethnic war, but not governmental ethnic war.Also note that when regressed alone with the onset of civil war as the dependent variable, oil price is significantly associated with the onset of governmental civil war (model 3). When the product of oil location and oil price is the explanatory variable, however, it is only significantly associated with the onset of territorial civil war (model 4). The same results hold at the country level when the dependent variable of the onset of ethnic war coded by EPR (table AB-2), although the product oil location and oil price can no longer differentiate territorial civil war from governmental civil war within the ACD dataset.
In second set of robustness checks, we use frithlogit, or penalized maximum likelihood logit regression that check rare events biases (Firth 1993),to make sure that our results are not driven by some specific (rare) events. Again, our results hold: results from frithlogit models are essentially identical to those obtained from normal logit models (tables AB-3 and AB-4).Note that because frithlogit regression cannot handle multinomial dependent variable, we only perform frithlogit regressions with the onset of ethnic war coded by EPR, without further differentiating ethnic war into ethnic infighting and ethnic rebellion.
Table AB-1:Location of Oil*Oil Price and the onset of ethnic conflict, group-level, with USGS dataset (Dependent variable is onset of ethnic conflict experienced by a group, 1946-2005)
(1) / (2) / (3) / (4)Ethonset (EPR) / Ethonset (EPR) / Ethonset (ACD) / Ethonset (ACD)
territorial / governmental / territorial / governmental
Oilprice / 0.022*** / 0.0075 / 0.038***
(0.0078) / (0.011) / (0.0099)
Oillocation* / 0.023*** / 0.032*** / 0.017
oil price / (0.0077) / (0.010) / (0.014)
Distance to / 0.15** / 0.27*** / 0.089
capital / (0.067) / (0.089) / (0.095)
Distance to / -0.15** / -0.26** / -0.035
border / (0.068) / (0.10) / (0.076)
Excluded / 1.30*** / 1.12*** / 1.42***
(0.28) / (0.42) / (0.36)
Downgraded2 / 1.74*** / 1.22** / 2.03***
(0.37) / (0.52) / (0.46)
Size of core / 0.27*** / 0.11 / 0.60***
territory / (0.087) / (0.10) / (0.12)
Warhistory / 0.59*** / 0.35 / 0.93***
(0.17) / (0.31) / (0.24)
GDP per capita / -0.37*** / -0.31 / -0.38***
(ln) / (0.13) / (0.22) / (0.14)
Population size / -0.10 / -0.0084 / -0.30**
(ln) / (0.11) / (0.14) / (0.14)
No. of peace / -0.23*** / -0.10 / -0.37*** / -0.079 / -0.17* / -0.054
Years (group) / (0.074) / (0.074) / (0.096) / (0.12) / (0.086) / (0.13)
constant / -4.19*** / -1.75 / -4.26*** / -5.73*** / -4.18** / -0.18
(0.32) / (1.30) / (0.40) / (0.57) / (1.94) / (1.66)
Pseudo R2 / 0.026 / 0.11 / 0.033 / 0.13
N / 24791 / 19749 / 24791 / 19749
Wald chi2 / 41.1*** / 150.4*** / 58.2*** / 397.5***
Group FE / YES / YES / YES / YES
Note: models 1 and 2 are logit model and model 3 and 4 are multi-nominal logit model.
Robust standard errors clustered according to group in parentheses;* p<.10, ** p<.05, *** p<.01.
Splines of peace years are controlled but not shown.
Table AB-2. Oil Location*Oil Price and the onset of ethnic conflict, country level, with USGS dataset.Sample includesonly oil producing countries.
(Dependent variable is onset of ethnic conflict experienced by a country, 1946-2010)
(1) / (2)Ethonset(EPR) / Ethonset(EPR)
Oil location *oil price / 0.022*** / 0.015**
(0.0061) / (0.0078)
year / -0.021** / -0.021*
(0.0097) / (0.012)
No. of peace years / -0.20 / 0.056
(0.16) / (0.36)
Ratio of excluded population / 0.15
(0.12)
Segmentation / 0.051
(0.033)
Imperial past / 1.30*
(0.67)
ELF / 2.25***
(0.65)
GDP per capita(ln) / -0.084**
(0.043)
Size ofpopulation (ln) / 0.19*
(0.11)
Area of mountainousterrain (ln) / 0.18
(0.12)
Regime change / 0.21
(0.38)
Anocracy / 0.27
(0.39)
Oil production percapita (ln) / 0.019
(0.023)
Ongoing war / 0.28
(0.92)
Constant / 38.6** / 32.4
(19.2) / (23.6)
Pseudo R2 / 0.057 / 0.17
N / 3794 / 3733
Wald chi2 / 29.3** / 272.9***
Country FE / YES / YES
Note: All are logit models.Robust standard errors clustered according to country in parentheses;
* p<.10, ** p<.05, *** p<.01.Splines of peace years are controlled but not shown.
Table AB-3. Firthlogit models at the group level with USGS dataset
(Dependent variable is onset of ethnic conflict experienced by a group, 1946-2005)
(1) / (2)Ethonset(EPR) / Ethonset(EPR)
Oillocation / 0.48**
(0.19)
Oillocation * Oil price / 0.023***
(0.0073)
Distance tocapital (ln) / 0.14** / 0.15**
(0.069) / (0.069)
Distance toborder (ln) / -0.14** / -0.15**
(0.061) / (0.062)
excluded / 1.26*** / 1.28***
(0.25) / (0.25)
downgraded2 / 1.76*** / 1.76***
(0.28) / (0.28)
Size of core territory (ln) / 0.27*** / 0.26***
(0.076) / (0.075)
Warhistory / 0.71*** / 0.60***
(0.18) / (0.19)
GDP per capita (ln) / -0.37*** / -0.36***
(0.10) / (0.10)
Size of population (ln) / -0.11 / -0.099
(0.069) / (0.069)
No. of peaceyears / -0.10 / -0.10*
(0.064) / (0.063)
constant / -1.71* / -1.71*
(1.02) / (1.02)
N / 19749 / 19749
Wald chi2 / 176.9*** / 179.2***
Group FE / YES / YES
Note: All are firthlogitregression models.
Robust standard errors clustered according to group in parentheses;* p<.10, ** p<.05, *** p<.01.
Splines of peace years are controlled but not shown.
Table AB-4. Firthlogit models at the country level with USGS dataset
(Dependent variable is onset of ethnic conflict experienced by a country, 1946-2010)
Samplesin model 2 and model 3 includesonly oil producing countries.
Model(1)Ethonset (EPR) / Model (2)
Ethonset (EPR) / Model (3)
Ethonset (EPR)
Oil location / 0.66** / 1.49**
(0.28) / (0.70)
Oil location * oil price / 0.015*
(0.0091)
Ratio of excluded population / 0.27*** / 0.083 / 0.15
(0.089) / (0.14) / (0.13)
segmentation / 0.11** / 0.056 / 0.050
(0.044) / (0.058) / (0.057)
Imperial past / 0.73** / 1.39** / 1.22**
(0.34) / (0.57) / (0.57)
ELF / 1.74*** / 1.99*** / 2.26***
(0.47) / (0.71) / (0.73)
GDP per capita (ln) / 0.0012** / -0.059 / -0.072**
(0.00052) / (0.039) / (0.036)
Size of population (ln) / 0.12 / 0.14 / 0.21*
(0.077) / (0.12) / (0.11)
Area of mountainous terrain (ln) / 0.19** / 0.19 / 0.17
(0.090) / (0.15) / (0.15)
Regime change / 0.20 / 0.19 / 0.24
(0.25) / (0.37) / (0.37)
Anocracy / 0.34 / 0.35 / 0.29
(0.21) / (0.32) / (0.32)
Oil production per capita (ln) / 0.0081 / 0.042** / 0.044***
(0.0097) / (0.019) / (0.017)
Ongoing war / 0.12 / 0.052 / 0.14
(0.56) / (0.87) / (0.84)
year / 0.0057 / -0.0091 / -0.021*
(0.0059) / (0.0089) / (0.012)
No. of peace years / 0.16 / -0.0010 / 0.028
(0.21) / (0.34) / (0.32)
Constant / -19.8* / 8.78 / 32.9
(11.7) / (17.6) / (24.4)
N / 7566 / 3733 / 3733
Wald chi2 / 122.5*** / 70.1*** / 75.6***
Country FE / YES / YES / YES
Note: All are firthlogitregression models.Robust standard errors clustered according to country in parentheses;* p<.10, ** p<.05, *** p<.01.Splines of peace years are controlled but not shown.
Appendix C: Results with the PETRODATA dataset
In this appendix, we present regression results based on oil-location information derived from the PETRODATA dataset (Lujala et al 2007). Overall, although the number of observations constructed from the PETRODATA dataset is much smaller than the number observations constructed from the USGS data, we obtain essentially the same results.
Table AC-1 reproduces the models at the group level as in Table 1 of the main text. Again, oil location (Lujala, group)is positively and significantly associated the onset of ethnic war, as recoded by the EPR dataset, and this result is robust to the inclusion of a battery of control variables (models 1 and 2). Moreover, oil location is positively and significantly associated the onset of territorial (i.e., ethnic) civil war but not governmental civil war (models 3 and 4).
Table AC-2 reproduces the models at the country level as in Table 2 and Table 3 of the main text. Model 1 is our baseline model, and oil location (Lujala, country) is positively and significantly associated with onset of ethnic war. Again, this result holds as we add more and more control variables progressively (Model 2).In model 3, we move to more fine-grained analyses of onset of ethnic war, with ethnic warbeing divided into ethnic infighting among power-holders (infighting) and rebellion (i.e., an excluded ethnic group rebels against the state), according to the EPR dataset. Again, the results strongly support our theory and hypotheses:oil location (Lujala, country) remains positively and significantly associated with the onset of ethnic rebellion but not with the onset of ethnic infighting. Again, the overall result holds as we add more and more control variables progressively in model 4.
Table AC-1. The Ethno-geographical Location of Oil and the onset of ethnic conflict, group-level, with the PETRODATA dataset by Lujala et al (2007)
(Dependent variable is onset of ethnic conflict experienced by a group, 1946-2005)
(1) / (2) / (3) / (4)Ethonset(EPR) / Ethonset(EPR) / Ethonset (ACD) / Ethonset (ACD)
territorial / governmental / territorial / governmental
Oillocation / 0.52* / 0.96*** / 0.88*** / -0.11 / 1.36*** / 0.46
(Lujala,group) / (0.27) / (0.22) / (0.31) / (0.45) / (0.27) / (0.39)
Distance to / 0.13 / 0.18 / 0.11
Capital (ln) / (0.11) / (0.15) / (0.17)
Distance to / -0.15* / -0.25** / 0.019
Border (ln) / (0.080) / (0.10) / (0.086)
Excluded / 1.35*** / 0.98** / 2.02***
(0.34) / (0.39) / (0.49)
Downgraded2 / 2.11*** / 1.46*** / 2.53***
(0.38) / (0.52) / (0.47)
Size of core / 0.17 / -0.019 / 0.72***
territory / (0.12) / (0.13) / (0.17)
Warhistory / 0.75*** / 0.61** / 1.06***
(0.17) / (0.26) / (0.23)
GDP per capita / -0.46*** / -0.52** / -0.46**
(ln) / (0.16) / (0.21) / (0.20)
Population size / -0.27* / -0.27* / -0.42**
(ln) / (0.14) / (0.16) / (0.21)
No. of / -0.26*** / -0.12 / -0.36*** / -0.13 / -0.13 / -0.11
Peaceyears / (0.078) / (0.078) / (0.10) / (0.13) / (0.087) / (0.14)
Constant / -3.83*** / 0.10 / -4.31*** / -4.88*** / -0.66 / 1.21
(0.34) / (1.62) / (0.48) / (0.61) / (2.06) / (2.51)
Pseudo R2 / 0.034 / 0.13 / 0.040 / 0.15
N / 15892 / 14060 / 15892 / 14060
Wald chi2 / 42.5*** / 163.9*** / 56.3*** / 628.5***
Group FE / YES / YES / YES / YES
Note: models 1 and 2arelogit model and model 3 and 4are multi-nominal logit model.
Robust standard errors clustered according to group in parentheses;* p<.10, ** p<.05, *** p<.01.
Splines of peace years are controlled but not shown.
Table AC-2.The Ethno-geographical Location of Oil and the onset of ethnic conflict, country-level,with the PETRODATA dataset by Lujala et al (2007)
(Dependent variable is onset of ethnic conflict experienced by a country, 1946-2010)
(1) / (2) / (3) / (4)Ethonset (EPR) / Ethonset (EPR) / Infighting / rebellion / infighting / rebellion
Oillocation / 0.73*** / 0.68** / 0.13 / 0.89*** / 0.75 / 0.67**
(Lujala, country) / (0.25) / (0.27) / (0.44) / (0.28) / (0.60) / (0.26)
Ratio of excluded / 0.23*** / -0.23 / 0.42***
population / (0.088) / (0.19) / (0.096)
Segmentation / 0.085** / 0.34*** / 0.037
(0.043) / (0.12) / (0.044)
Imperial past / 0.70 / 4.13** / 0.22
(0.42) / (1.80) / (0.45)
ELF / 1.72*** / 1.98 / 1.76***
(0.54) / (1.46) / (0.54)
GDP per capita (ln) / -0.083** / -0.10 / -0.080**
(0.036) / (0.063) / (0.038)
Population size (ln) / 0.079 / -0.72*** / 0.25***
(0.067) / (0.22) / (0.076)
Ongoingwar / 0.39 / 1.56 / 0.21
(0.64) / (1.39) / (0.72)
year / 0.0048 / 0.0035 / 0.023* / 0.00067 / 0.028* / -0.0011
(0.0056) / (0.0064) / (0.012) / (0.0056) / (0.016) / (0.0071)
Area of mountainous / 0.11 / 0.26 / 0.055
terrain (ln) / (0.089) / (0.16) / (0.10)
Regime change / 0.0071 / -0.23 / 0.027
(0.26) / (0.74) / (0.29)
Anocracy / 0.35 / 0.073 / 0.44*
(0.24) / (0.49) / (0.24)
Oil production per / -0.039 / -0.082 / -0.027
capita (ln) / (0.062) / (0.10) / (0.068)
No. of peaceyears / -0.0095 / 0.22 / 0.23 / -0.051 / 0.59 / 0.17
(0.11) / (0.24) / (0.26) / (0.12) / (0.42) / (0.28)
Constant / -13.2 / -14.5 / -51.8** / -5.41 / -61.0* / -7.26
(11.1) / (12.7) / (24.7) / (11.2) / (31.9) / (13.8)
Pseudo R2 / 0.039 / 0.11 / 0.041 / 0.14
N / 4871 / 4694 / 4871 / 4694
Wald chi2 / 26.6** / 155.3*** / 44.8** / 380.9***
Country FE / YES / YES / YES / YES
Note: models 1 and 2 are logit model and model 3 and 4 are multi-nominal logit model.
Robust standard errors clustered according to country in parentheses;* p<.10, ** p<.05, *** p<.01.
Splines of peace years are controlled but not shown.
Appendix D: Replication of Morelli and Rohner (2015)
In this appendix, we how that the statistical results reported by Morelli and Rohner (M&R 2015) are not robust at all, contrary to their claims. As we have argued in detail in section III of the main text, this is mainly due to the flawed logic of their key explanatory variables, oil concentration at the country level (i.e., Oil Gini) and at the group level (i.e., R1/R1+R2).
To begin with, M&R’s results at the country level are extremely fragile, even with the onset of all civil war as the dependent variable. As shown in model 2 in Table AD1 in Appendix D, at the country level, even with OLS models, simply clustering standard errors according to country eliminates all the statistical significances associated with “oil concentration” (i.e., Oil Gini). And with logit model, although oil concentration is significant with clustering standard error according to country (model 3), it becomes insignificant after controlling merely three control variables (population, GDP per capita, democracy score). Notably, these three control variables have the right signs and are (in-)significant as reported in other studies (i.e., a larger population size and a lower GDP per capita are associated with higher risk of civil war whereas democracy score is negative but generally insignificant).
M&R’s (2015) results at the group level are equally fragile. As shown in Table AD2, oil concentration at the group level (i.e., R1/R1+R2) has no robust relationship with the onset of ethnic war, even without adding any control variables, whether the sample includes all groups (Models 1 to 3) or only the non-governing groups (Models 4 to 6). Oil concentration has robust relationship only with the onset of civil war (table AD3). These results suggest that M&R’s (2015) results at the group level are mostly driven by using the onset of all civil war as their dependent variable even though their key explanatory variable is an ethnic group variable, exactly as we suspected.
Table AD-1. Oil Ginihas no robust association with the Onset of Civil War, Country Level
Dependent variable is onset of civil war at the country level (Prio_onset_high).
(1) / (2) / (3) / (4) / (5) / (6)OLS, FE / OLS, FE / logit / logit / logit / logit
Oil Gini / 0.0432*** / 0.0432 / 1.874*** / 0.735 / 0.717 / 0.245
(0.0154) / (0.0267) / (0.425) / (0.616) / (0.609) / (0.534)
Population size / 0.424*** / 0.389*** / 0.292***
(WDI, ln, lag 1 yr) / (0.103) / (0.107) / (0.102)
GDP per capita / -0.407** / -0.437*** / -0.323**
(PWT, ln, lag 1 yr) / (0.164) / (0.165) / (0.141)
Democracy score / -0.0378 / -0.0367 / -0.0253
(Polity, lag 1 yr) / (0.0231) / (0.0231) / (0.0199)
New State / -1.256* / -1.487*
(0.738) / (0.780)
Peace duration / -0.0529***
(country level, lag 1 yr) / (0.00977)
Constant / 0.00904*** / 0.00904** / -4.556*** / -8.235*** / -7.354*** / -5.113**
(0.00275) / (0.00413) / (0.201) / (2.134) / (2.264) / (2.027)
Observations / 7,569 / 7,569 / 7,569 / 5,372 / 5,372 / 5,372
Wald chi2 / - / - / 19.40*** / 63.74*** / 65.78*** / 108.14***
R2 or Pseudo R2 / 0.001 / 0.001 / 0.031 / 0.086 / 0.092 / 0.145
S.E. clustering / No / Country / Country / Country / Country / Country
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Table AD2: Oil concentration (R1/R2+R2) has no robust association with the onset of ethnic war even when the sample is non-governing groups.
Dependent variable is the onset of ethnic war at the group level. All models are logit model.
(1) / (2) / (3) / (4) / (5) / (6)Oil concentration / 1.871*** / 1.871* / 1.871* / 0.735* / 0.735 / 0.735
(i.e., R1/R1+R2) / (0.187) / (1.091) / (1.094) / (0.411) / (0.449) / (0.447)
Constant / -5.136*** / -5.136*** / -5.136*** / -4.673*** / -4.673*** / -4.673***
(0.121) / (0.175) / (0.178) / (0.127) / (0.134) / (0.140)
Observations / 14,206 / 14,206 / 14,206 / 7,466 / 7,466 / 7,466
Wald Chi2 / 2.94* / 2.93* / 2.71* / 2.68 / 2.70
Pseudo R2 / 0.055 / 0.055 / 0.055 / 0.003 / 0.003 / 0.003
S.E. clustering / No / Dyad / Group / No / Dyad / Group
Note: Samples in Models 1-3 are all groups; whereas samples in Models 4 to 6 are non-governing groups.
Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
Table AD3: Oil concentration is only associated with the onset of all civil war when the sample is non-governing groups.
Dependent variable is the onset of civil war (group level). All models are logit models.
(1) / (2) / (3) / (4) / (5) / (6) / (7)Oil concentration / 0.154 / 1.058*** / 1.058*** / 1.058*** / 1.411*** / 1.488*** / 1.579***
(i.e., R1/R1+R2) / (0.302) / (0.322) / (0.386) / (0.387) / (0.381) / (0.391) / (0.341)
Mountainous terrain / 0.543** / 0.766***
(group level) / (0.247) / (0.264)
distance to capital / 0.147* / 0.0676
(group level, ln) / (0.0785) / (0.0972)
Peace years / -0.0500*** / -0.0769***
(group level) / (0.00667) / (0.00875)
Population size / 0.139** / 0.270***
(WDI, ln, lag 1 yr) / (0.0619) / (0.0630)
GDP per capita / -0.349*** / -0.0760
(PWT, ln, lag 1 yr) / (0.114) / (0.105)
Democracy score / 0.00163 / -0.000155
(0.0169) / (0.0153)
Constant / -5.664*** / -5.636*** / -5.636*** / -5.636*** / -5.804*** / -5.267*** / -8.489***
(0.0704) / (0.0743) / (0.101) / (0.106) / (0.519) / (1.285) / (1.290)
Observations / 63,869 / 53,053 / 53,053 / 53,053 / 53,053 / 36,542 / 36,542
Wald chi2 / 0.25 / 8.25*** / 7.5*** / 7.46*** / 65.92*** / 25.06*** / 120.46***
Pseudo R2 / 0.000 / 0.004 / 0.006 / 0.003 / 0.045 / 0.016 / 0.093
S.E. clustering / None / None / Dyad / Group / Dyad / Dyad / Dyad
Note: The sample in Model 1 includes all groups whereas those in Models 2 to 7 include onlyno-governing groups.
Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
Appendix E: Additional Comments on Several Related Studies
In this Appendix, we provide a more extensive survey of related studies. By so doing, we seek to show that our study really betters existing studies.
We acknowledge that two unpublished conference papers with a similar focus exist (e.g., Hunziker and Cederman 2012; Condra 2013). Because they remain unpublished, we do not address them specifically other than insisting that our theoretical development and empirical results are superior to what those two papers have reported.For instance, Hunziker and Cederman (2012) focused only on “giant oil fields”, thus leaving aside the possibility that an ethnic group does not have to have a giant oil field within its core territory to rebel. Meanwhile, Condra (2013) deals with Africa alone and his results are not rigorously tested.
Among unpublished works, onlyHunziker’s (2014) unpublished dissertationcame close to what we report here, andhis quantitative results are very similar to our results. His project and our project have different strengths. Whereas Hunziker (2014, esp. chaps. 4 & 5) uses oil basin to instrument for oil production, our key explanatory variable (i.e., “oil location”, see section IV below) is a straightforwardly exogenous independent variable. Paine (2016) also reported some similar results but his focus is the development of a formal bargaining model.
Before we proceed further, a key caveat is in order. Many earlier studies on oil and conflict (ethnic or non-ethnic) deployaggregate data at the country level thus do not explicitly link the subnational properties of oil with the subnationallocation of ethnic groups. Most of these studiesconclude that oil is generally a curse (for reviews, see Ross 2006; 2010; 2014; Blattman and Miguel 2010; Van de Pleogg 2011). These results, however, have been questioned from time to time (e.g., Basedau and Lay 2009; Brunnschweiler and Bulet 2009; Cotet and Tsui 2013; cf. Lei and Michaels 2014; Wegenast and Basedau 2014; Bell and Wolford 2015). Because out study deploys subnational data and explicitly links the subnational location of oil with the subnationallocation of ethnic groups, our study differs fundamentally from these earlier studies and are more akin to studies with sub-national data. We thus do not engage with this literature with aggregate data extensively, although we do cite some of them when appropriate. We concur with Smith’s (2014) call that the two levels (i.e., national and sub-national) should be synthesized. Indeed, our results below, which cover both thegroup-level and the state-level, provide a possible explanation why results from earlier studies with aggregate national data have been inconsistent.