Our estimation strategy already captures many potential confounding factors with the country-year and region fixed effects. Nevertheless, regional time-varying factors can potentially bias our results. In the main analysis, we control for two kinds of local economic shocks and for population growth. In table 3.9, we provide three additional robustness checks. First of all, conflict and political decisions likely depend on past
Chapter 3. Political favoritism and conflict
conflict events. To check whether our result is driven by pre-trends, we control for the share of past conflict years within the last three years. The results in column 1 of table 3.9 show that conflict events are positively correlated over time but the inclusion of past conflicts in the regression only marginally changes the estimates of the leader region indicators. Second, we include additional natural-resource shocks other than oil and gas as controls. For this purpose, we use the natural-resource information provided by Berman et al. (2017). The inclusion of 10 major minerals changes our regional favoritism effects on conflict intensity only marginally. Third, political trends like provincial independence efforts can influence election outcomes and potentially result in political violence. If this is the case, our estimation results so far are biased.
We address this potential endogeneity with the inclusion of provincial time trends that control for average political developments and conflict dynamics in a province.
Our results are robust to the inclusion of provincial time trends and very similar in magnitude compared to the baseline regression.
TABLE3.9: Further controls
IHS(casualties)
(1) (2) (3) (4)
Leader autoct−1 -0.086* -0.096** -0.101** -0.101**
(0.044) (0.040) (0.043) (0.044) Leader non-autoct−1 0.007 0.011 -0.002 -0.007
(0.017) (0.015) (0.018) (0.017)
Past conflict 0.323***
(0.056)
Observations 1,053,512 1,053,512 1,089,755 1,177,805
R-squared 0.413 0.440 0.384 0.429
Country-year FE Yes Yes Yes Yes
Region FE Yes Yes Yes Yes
Controls Yes Yes Yes Yes
Resource controls Yes
Provincial time trends Yes
Note:The table reports OLS regression estimates of regressing the inverse hyperbolic sine of the number of casualties on dummies of (non-)autocratic leader regions. Regressions include fixed effects and control variables as indicated in the table. Standard errors are clustered at the country level. *** p<0.01, ** p<0.05, * p<0.1.
Attitudes and behavior towards the government may be more benevolent in the home region of the political leader compared to other regions in the country as they are in-group members with the leader. Coming from the same region and potentially sharing the same values and tradition, citizens of the home regions may feel more sympathetic towards the leader and have more trust in his/her government, reducing the likelihood of attacks or revolts against the leader in the home region.
If this is the case, we would have falsely attributed the reduction in the intensity of conflict in the birth regions of autocratic leaders during the leaders’ time in office
to political favoritism. We analyze this potential bias by comparing the reported attitudes and behavior towards the leaders in their home regions during their time in office and prior or after that time. More precisely, we compare whether respondents of Afrobarometer surveys report to have more trust in the leader, are more likely to approve the performance of the leader and to be politically active when the region is the birth region of the current leader compared to other times. The results in table 3.10 show that persons residing in the birth regions of autocratic leaders do not report and evaluate the leaders differently than other citizens, once we control for region and country-year fixed effects. Only citizens in the home regions of non-autocratic leaders report to have more trust in their leaders. Hence, we find no difference in attitudes and behavior towards the government between regions in autocratic countries, validating our main results.
TABLE3.10: Robustness check: change in perceptions
(1) (2) (3)
Trust Performance Activism leader leader
Leader autoct−1 -0.098 -0.195 0.071 (0.166) (0.192) (0.044) Leader non-autoct−1 0.160*** 0.139 0.053
Observations 7,036 7,097 7,170
R-squared 0.746 0.728 0.817
Note:The table reports OLS regression estimates of regressing index variables on the lagged leader region dummies. Outcome variables are based on the Afrobarometer rounds 1 to 6. Regression models as specified in table 4.2. ***
p<0.01, ** p<0.05, * p<0.1.
To examine the geographic scope of regional favoritism, we re-estimate our baseline result on a higher (the first) administrative level, which we refer to as provinces.
The results are shown in table A.2. We estimate two distinct regression models. In column 1, we control for country-year and province fixed effects, whereas column 2 adds the control variables accounting for economic shocks at the provincial level.
Throughout all regression specifications, we find no statistically significant coefficient.
Yet, the magnitudes of the coefficients are comparable to our baseline regressions.
This emphasizes that favoritism with respect to conflict has a rather local scope.
Lastly, we check whether the effect is driven by regime changes or irregular entries into office. Political transitions like the concentration of power to the leader can be accompanied by or achieved with political violence. In order to test whether the effect is driven by countries that recently switched from an autocratic or to an autocratic regime, we classify countries as switchers if there has been a change in the political system in the last five years and estimate a differential effect for these countries. The results are presented in column 1 of table A.3. We find no significant
Chapter 3. Political favoritism and conflict
difference between well-established autocracies and countries that recently switched to an autocratic system in the regional favoritism effect on conflict.
Leaders, in particular autocratic leaders, may take over a country through violent means, potentially leading to an increase in conflict in the capital region. This rise in violence at the end or beginning of the time in office, may provoke a difference in conflict intensity between the home region of a leader and other regions in the country (including the capital region) that we would falsely interpret as a reduction of conflict in the home region. To check whether the effect is driven by these kinds of irregular entries into offices, we estimate heterogeneous effects between leaders that regularly or irregularly entered office as defined in the Archigos database. The results presented in column 2 of table A.3 show no differential effect of home regions of autocratic leaders that entered office irregularly compared to home regions of autocratic leaders with regular entrance.
3.8 Conclusion
In this study, we have investigated the effect of political favoritism on the occurrence and intensity of conflict in a global dataset of 172 countries. Using a region-year panel with information on 836 national leaders, our analysis extends the political favoritism literature by providing empirical evidence on a further dimension of favoritism namely with respect to conflict and security precaution. We combine geo-coded conflict information from the UCDP with data on the birthplaces and ethnic affiliations of political national leaders over the years 1989 to 2015. We include fixed effects along two dimensions (region and country-year) and controls for regional economic shocks and population density to estimate the causal effect of political favoritism on conflict. As politicians may shape the distribution of violence in various ways, we investigate the mechanisms of the effect. More specifically, we analyze three channels: the ‘welfare channel’, the ‘in-group favoritism channel’, and the
‘coup-proofing channel’.
Our results show that political favoritism is not linked to the incidence of conflict but reduces the intensity of violence in the birth regions of autocratic leaders while they are in office. These regions experience around 10% fewer casualties during that time.
We find no evidence for a reduction in conflict intensity in leaders’ ethnic homelands, but ethnic groups that belong to the same ethnicity as the current leader are less involved in conflict. We identify two channels through which the effect occurs: the
‘in-group favoritism channel’ and the ‘coup-proofing channel’. In leader regions fewer state-based casualties occur, potentially due to an increase in military presence.
Moreover, in the leader regions, there is a higher extent of perceived public-sector corruption. This supports the anecdotal evidence showing that patronage in the allocation of officer positions leads to higher corruption and a change in behavior
of the armed forces. Thus, this study also shows how state capacity and past coups determine conflict.