The Effects of Exposure to Conflict, Insecurity, and Violence on Popular Attitudes Towards Democracy in the Sahel
In this paper we look at the effect of exposure to conflict, insecurity, and violence on popular attitudes towards democracy in four Sahel countries: Burkina Faso, Mali, Niger and Nigeria. Much attention has been given to the consequences of conflict, insecurity, and violence on state formation and democratization (Tilly, 1992; Centeno, 2002; Wantchekon & Neeman, 2002; Bermeo, 2003). However, the study of such consequences has largely focused on cases where regime change occurred. What has been less studied in the empirical literature are cases where conflict, insecurity and violence did not result in regime change, but may have nonetheless weakened the quality of democracy (Cheeseman et al., 2018).
As noted by Cheeseman et al. (2018: 38) it is common in the Africanist literature to argue that “variations in the quality of democracy are shaped by the strength and independence of political institutions”. Here we focus on a ‘bottom-up’ approach to understanding the strength of such institutions by focussing on the popular support that such institutions garner. As is widely accepted in the literature, institutions require such support – at least in the long run – to function smoothly, efficiently and sustainably without reliance on coercion (Lipset, 1960; Easton, 1975, Mishler & Rose, 1999; Dalton, 2000; Norris, 2011). Our research therefore asks whether exposure to conflict, insecurity and violence has consequences for democracy through its effect on public attitudes towards democracy and possible alternative regime forms.
Specifically, we query two causal mechanisms. First, we explore a retrospective and instrumental mechanism: could exposure to conflict, insecurity, and violence influence attitudes towards democracy through its effect on public evaluations of government and state performance? Here, a decline in positive attitudes towards democracy may occur as the result of negative perceptions and evaluations of government performance in protecting people from such events. Second, we explore a prospective and instrumental mechanism: could exposure to conflict, insecurity and violence influences influence attitudes towards democracy through improving evaluations of the military and non-democratic actors? Here, a change in attitudes towards democracy would occur as the result of improving perceptions and evaluations of military and non-democratic actors who may be perceived as stronger or more capable in protecting from future exposure to such events.
To test these two mechanisms, we use nationally representative Afrobarometer (AB) data (round 8) to measure attitudes towards democracy, trust in the president and the military and support for non-democratic rule. In addition, we use Armed Conflict Location & Event Data (ACLED) to capture incidents and intensity of conflict, insecurity and violence. We match the Afrobarometer and ACLED data using geo-spatial information which both sources provide.
Our models suggest that people who live in areas with less events of violence, conflict and insecurity have more trust in the military, more trust in the president and more positive evaluations of how government is handling violent conflict. We find that more trust in the army and more positive evaluations of government handling violent conflict – in turn - is significantly correlated with more support for military rule. This means that living in relatively peaceful areas in the studied countries is significantly linked with more support for military rule, while living in more violent and insecure areas is significantly linked with less support for military rule. We do not find the same results when predicting demand for democracy into the model instead of support for military rule. In this case we find that trust in the military and trust in the president are uncorrelated with demand for democracy and more positive evaluations of how government is handling violence is significantly associated with less demand for democracy.
Burkina Faso, Mali, Niger and Nigeria are selected for several reasons. All four countries are characterized by multiple ongoing conflicts. These conflicts are varied in type and intensity and have generally changed in type and intensity over time. At the same time, the countries have seen a general worsening of their political freedoms and democracy ratings (see table 1).
Dependent Variable: Demand for democracy
In survey studies, some uncertainty surrounds what exactly ordinary people mean by ‘democracy’ when they voice support for it. Critics such as Schedler and Sarsfield (2007) have argued that the commonly used measures of ‘support for democracy’ often lack reference to concrete attributes of democracy. Such “vacuous conceptions of democracy” (Schedler & Sarsfield, 2007: 639) are problematic as it is unclear what, if anything, ‘democracy’ means to the respondents, and if such understandings are consistent across different groups and contexts. This is further compounded by the normative nature of democratic rule. Widely held support for democracy in survey studies around the world may lack reliability due to interviewer effects which make it unclear whether reported support for democracy reflects actual views and attitudes of the respondent, or simply reflects the “‘almost universal’ practice of […] ‘paying lip service to democracy’” (Inglehart, 2003:51). Lastly, it is also unclear from measuring only support for democracy whether respondents may support democracy and hold “conflicting values” (Schedler & Sarsfield, 2007: 639). Rather, as Bratton and Mattes (2001: 457) state, support for democracy is best queried in “concrete terms and in the form of comparisons with plausible alternatives”.
In survey studies, including ‘plausible alternatives’ is referred to as measuring ‘authentic democratic support’. The premise of this measure is that democratic and non-democratic norms and ideals are inherently incompatible. Thus, someone who prefers democracy but can still accept or see merit in non-democratic forms of governance may display normative and practical support for democracy, but not authentic support. Only if a person supports democracy practically and rejects non-democratic alternatives does someone report authentic support. Afrobarometer uses a constructed index called ‘demand for democracy’ to tap authentic support for democracy. Demand for democracy captures whether someone voices support for democracy as a regime type and rejects nondemocratic alternatives, such as one-man, military or one-party rule. The index is made up of ‘support for democracy’ and as three variables which probe support for alternative regimes forms.
To measure support for democracy, respondents were asked:
Which of these three statements is closest to your own opinion?
Statement 1: Democracy is preferable to any other kind of government.
Statement 2: In some circumstances, a nondemocratic government can be preferable.
Statement 3: For someone like me, it doesn’t matter what kind of government we have.
To measure rejection of non-democratic alternatives, respondents are asked:
There are many ways to govern a country. Would you disapprove or approve of the following alternatives?
Only one political party is allowed to stand for election and hold office?
The army comes in to govern the country?
Elections and Parliament are abolished so that the president can decide everything?
An index of ‘demand for democracy’ is computed by combining the four variables with a scale running from 0 to 4. Someone who is indifferent to or supportive of non-democratic alternatives score lower on this scale, while someone who supports a democratic regime and rejects other alternatives scores higher. Accordingly, a score of 0 is labelled as ‘no demand for democracy’, while a score of 4 is coded as ‘full demand for democracy’. Demand for democracy is highest in Nigeria, with 52% of respondents having ‘full’ demand, followed by Mali (40%) and Niger (38%) (see figure 1). Conversely, only around a quarter of Burkinabe appear to have full demand (27%). In general, majorities of respondents agree with at least 3 of 4 components of demand for democracy, while only less than 5% in each country have no demand for democracy.
Disaggregating the index suggests that respondents most widely rejected one-party rule and oneman rule across the four countries (see figure 2). Respondents were consistently less willing to reject the idea of military rule and were generally less frequent to say that democracy – as a regime type – was preferable to any other regime form.
Exogenous variables: exposure to conflict, insecurity, and violence
To measure a respondent’s exposure to conflict, insecurity, and violence, we use ACLED to compute several scores based on the distance of a respondents’ enumeration area to events of conflict, insecurity, and violence in the 5 years prior to the interview.
For this time period, the ACLED data includes 13604 events, of which 60.6% are in Nigeria (n=8238), 18.9% in Mali (n=2567), 12.2% in Burkina Faso (n=1659) and 8.4% in Niger (n=1140). Across all four countries, events of conflict, insecurity, and violence increased year on year leading up to the Afrobarometer fieldwork (see Appendix 1). The ACLED disaggregates events of conflict, insecurity, and violence by type of event (as well as sub-type). Across all four countries, violence against civilians, battles and protests are the most common event type (see figure 4). Other forms – such as explosions/ remote violence, riots and strategic developments are less frequent.
To compute scores of exposure to conflict, insecurity, and violence I combine Afrobarometer and ACLED data using GIS. GPS information is collected by Afrobarometer at the level of the Primary Sampling Unit (PSU). The Afrobarometer sample is stratified using the main subnational units of government and urban or rural location. Following this stratification, PSUs are randomly selected. Afrobarometer limits the number of interviews per PSU to 8. As such, the assumption we make is that people within a PSU experience the same level of exposure based on their geographic proximity.
To compute exposure scores, ACLED events are limited to 2 degrees latitude and longitude from the respondent’s PSU. This cut-off equates to a radius of approximately 250km in the Sahel region. Two considerations informed this cut-off. First, for practical reasons, a cut-off was needed to keep the data manageable. Second, many of ACLED’s event type are likely limited in their impact on people. Applying this cut-off produces 4.069,056 reports across the four countries. These reports reflect the distance of each Afrobarometer respondent to any ACLED event within a 250km radius which occurred in the 5 years prior to the interview in the respondent’s respective country. Figure 4, below, displays the respective locations of AB and ACLED data in the four countries.
Several strategies are employed to compute the exposure scores.
First, we computed a basic count score. This score reflects the total number of events recorded by ACLED which occurred within the radius 2 degrees longitude and latitude (~250km) of the respondent’s PSU. This score however does not consider proximity of events. Therefore, we computed a second score which reflects the mean distance of all events within the determined radius. Third, we computed a score which reflects the closest event to the respondent.
Fourth, we computed an intensity score of events. It is highly plausible that more violent events influence people’s attitudes and evaluations differently than less violent events. We use the number of recorded fatalities as a proxy for the intensity of an event. An overview of descriptive statistics of the four scores is given by country in table 2, below.
Endogenous variables: popular government evaluations, attitudes towards the military and nondemocratic leaders as well as security perceptions
Several endogenous variables are considered using Afrobarometer data. First, we include people’s evaluation of how government is handling preventing violent conflict. The 4-point response scale runs from negative assessments (“very badly”) to positive ones (“very well”). Second, we include people’s trust in the army and the head of the executive (president/ prime minister). Both variables are measured on 4-point scales running from no trust (trust “not at all”) to high trust (trust “a lot”).
To test the mechanisms outlined in the introduction, we construct a series of structural equation models which include the 4 exposure scores as exogenous variables. Popular performance assessment of government handling conflict, trust in the army, trust in the president, and demand for democracy are included as endogenous variables. As noted above, Afrobarometer data across the four countries suggests that people are least opposed to military rule, compared to one-man and one-party rule. Moreover, if events of conflict, violence and insecurity are likely going to change people’s perceptions of an actor, it is the military. We therefore run a second series of models in which we enter the support for military rule variable instead of the demand for democracy variable. For both dependent variables we first consider all events – irrespective of event type -- before disaggregating by event type. A conceptual path model for the models run in this section is displayed in figure 6.
Standardized regression weights for all models are given in tables 3 and 4, below. Regression weights for non-significant effects are not listed for ease of reading the table. All significant results are significant at either the p< 0.05 or p<0.01 level (an overview of parameter estimates is given in appendix 3 and 4).
People who live in areas with more events are less positive in their evaluations of how government is handling violent conflicts, have less trust in the military and less trust in the president. Comparing across the different exposure scores, the additive sum of events is most strongly correlated with government evaluations, trust in the military and trust in the president, while proximity (closest event) is least strongly correlated. This suggests that people take a broader assessment into consideration rather than only their immediate surroundings.
Whether an event is violent or not matters in regard to how exposure to such events correlate with evaluations of government handling violent conflict, trust in the president and trust in the military, respectively. As might be expected, violent events have a stronger effect on government evaluations than protests and non-violent events. Here the greater the number of events that the respondent is exposed to, the more negative the evaluations of government handling conflict is. Although the question of government performance specifically asks about violent conflict, we find that a greater number of protests and non-violent events are also correlated with more negative evaluations of government, although the correlations are considerably smaller in magnitude.
Counter to what we might expect, we find a positive correlation of fatalities with evaluations of government as well as trust in the president and the military, respectively. This means that greater numbers of fatalities are significantly correlated with more positive evaluations and more trust. This might be an indication of a ‘rallying around the flag’ effect during violent events. However, disaggregated by events type, our results suggest that the ‘rallying around the flag’ effect of fatalities during violent events applies only to government and the president, but not the military.
Looking at different measures of exposure, trust in the military is most strongly correlated with lower overall number events, followed by a higher number of fatalities. However, the correlation looks very different depending on what type of event is considered. The results show that fatalities at violent events and non-violent events are not correlated with trust in the military but are negatively correlated with trust in the military for protests.
What is more, we find no significant correlation between fatalities at protests with trust in the president or evaluations of government performance. This may suggest that fatalities at protests are associated with military conduct rather than a failure of government or the President.
How government is perceived to be doing in handling violent conflict is significantly correlated with more trust in the President and more trust in the military. Across different event types the strength of correlation between government evaluations and trust in the president and military, respectively, remains comparable in magnitude.
But does this matter in regard to people’s attitudes towards forms of government? The results suggest that neither trust in nor trust in the president are significantly correlated with how much people want democracy (measured through demand for democracy). However, we do find that people’s evaluation of government is significantly correlated with how much they want democracy. The results suggest that people who are more positive of how their government is handling violence conflict are less in favour of democracy, versus non-democratic alternatives.
Indeed, we only find a significant – albeit weak – significant effect of government performance evaluations on demand for democracy. Interestingly, this effect is negative, suggesting that more positive government evaluations significantly predicts less demand for democracy. Here it is possible that the perception of whether the government is in fact democratic at the time is important.
Our models suggest that neither trust in the army nor trust in the President significant predict demand for democracy. But does exposure to conflict, violence and insecurity shape people’s attitudes towards military rule? Unlike in the initial path models for demand for democracy, we find a significant path linking exposure to conflict, violence and insecurity to support of military rule though trust in the army. Analogous to the models discussed in the previous section, people who experience a lower number of events, have a greater mean event distance and more fatalities (again paradoxically) are more trusting in the army. This higher trust translates into more support for military rule.
Here the results suggest that more positive evaluations of government handling violent conflict is significantly correlated with more approval of military rule in the respondent’s country. In addition, we find a significant – albeit weak – correlation of more trust in the military and approval of military rule.
Faced with ongoing conflict, violence, and insecurity in their countries, what does exposure to such insecurity do to Sahelians attitudes towards democracy and military rule? The models in this paper suggest that greater exposure is significantly correlated with more negative government evaluations as well as lower trust in the military and the president. While lower levels of trust in the president and military, respectively, are not significantly correlated with demand for democracy, more negative evaluations of government performance are significantly associated with more demand for democracy. In other words, greater exposure to violence, conflict and insecurity appears to be significantly linked with more demand for democracy through performance evaluations of government handling said conflict. Conversely, less exposure to conflict, violence and insecurity is appears to be linked to less demand for democracy in these countries, again through more positive performance evaluations. More specifically we find that more positive evaluations of how government is handling violent conflict is significantly associated with more approval for military rule.
Faced with ongoing conflict, violence and insecurity in their countries, Burkinabe, Malians, Nigerien and Nigerians who are less exposed to such events report more approval of military rule than those who experience greater exposure to such events. Containing or reducing the occurrence of conflict, violence and insecurity in these countries may thus not lead to more democratic rule but a publicly backed and supported military rule.