CELL PHONES AND COLLECTIVE ACTION

Given the breathtaking spread of cell phone technology worldwide and the particularly fast expansion on the African continent, citizens across many regimes have vastly improved means for private, direct, and immediate long-distance communication. The availability of cell phone technology and networks to citizens in some of the poorest regions in the world has been lauded as an important transformative force for economic development. In particular, the decrease in communication costs associated with the rising availability of cell phones has been linked to a boost in labor and consumer market efficiency (Abraham Reference Abraham2007; Aker Reference Aker2010; Aker and Mbiti Reference Aker and Mbiti2010). This research emphasizes the diminishing effect of cell phone technology on information asymmetries between market participants. For example, in the case of Indian fishers, cell phone technology allowed for the monitoring of prices in nearby markets without the need to personally attend the market, while also giving fishers access to sell goods to markets at further distances (Abraham 2007). Similar developments have been noted for agricultural markets in Africa (Aker Reference Aker2010; Aker and Mbiti Reference Aker and Mbiti2010). In addition, African entrepreneurs are developing ways in which cell phones can be used to increase market efficiencies and deliver services to customers.Footnote ⁴ The increasing availability of cell phone coverage has gone hand in hand with the use of mobile money and mobile banking (Donner and Tellez Reference Donner and Tellez2008). In fact, the development to make payments and transfers via mobile money instead of cash or credit cards has proliferated widely in Africa and lowers transaction costs for many market participants and citizens.Footnote ⁵

Interestingly, in Africa, this digital revolution is largely driven by private entrepreneurs, which have built up an extensive wireless infrastructure in a matter of years, often independent of governments or government-funded infrastructure.Footnote ⁶ Private cell phone providers have increased coverage at a vastly faster rate than landline providers. Today many areas that had never been connected to landline communication networks are covered by cell phone networks (Africa Partnership Program 2008; The World Bank Reference Foster and Garmendia2010). One of the advantages of cell phone networks is that the expansion is much costly in terms of infrastructure investments and thus a more decentralized expansion is possible.

Existing economics research provides evidence of lower transaction costs through the provision of cell phones. While much of this work has emphasized the positive effects of this new technology on economic outcomes, research on the direct effects of cell phones in the political sphere is not quite as common. However, Aker, Collier, and Vincente (Reference Aker, Collier and Vincente2011) show that in the case of Mozambique, cell phones can be used for voter education and can increase political participation in elections, as well as demands for accountability. The major takeaway is that cell phone usage, the availability of hotlines to voters, and text messaging can have positive effects on the political information available to voters as well as their political participation. In a similar vein Bailard (Reference Bailard2009), using country level analysis as well as provincial data for Namibia, finds that the use of cell phones by citizens can decrease corruption. She argues that cell phones change the information environment, as they decentralize and increase the spread of information. In addition, the proliferation of cell phones increases the probability of detection of corrupt officials and thus alters “the cost-benefit calculus of corrupt behavior by strengthening oversight and punishment mechanisms” (Bailard Reference Bailard2009, 337). Evidence further suggests that, through text messaging services, cell phones have been used to inform citizens of government wrongdoings, monitor elections, or report violence in many African states (Diamond Reference Diamond, Diamond and Plattner2012, 11).

More generally, observers of current events have linked cell phone technology to collective action, in particular peaceful protest, producing a new “protest culture” (Lapper Reference Lapper2010). In the context of authoritarian regimes, examples of cell phones aiding the organization of protests around the world are abundant, ranging from China in 1999, where Falun Gong was able to stage a large protest in a secure government complex, to Manila, Philippines in 2001 (Philippine Daily Inquirer 2001), or Kiev, Ukraine during the Orange Revolution (Diamond Reference Diamond, Diamond and Plattner2012, 12).Footnote ⁷

Yet, cell phone technology does not only affect collective action in authoritarian governments. Protesters in Madrid, Spain in 2004 were able to organize quickly using text messaging (Shirky Reference Shirky2008, 180). The increased organization capabilities of protesters have been noted by the police in the riots in London in the summer of 2011, as well as protests over G20 summits (Bradshaw Reference Bradshaw2009; Sherwood Reference Sherwood2011).

The link between political behavior and cell phone usage is also borne out in survey data. The 2008 wave of the Afrobarometer public opinion survey includes a question on the usage of mobile phone technology and protest behavior (Mattes et al. Reference Mattes, Bratton, Davids and Africa2010). A simple regression of the protest item on cell phone usage, controlling for a number of socioeconomic factors, reveals a positive and highly statistically significant effect, i.e., cell phone users are more likely to participate in protests.Footnote ⁸

While these observations and emerging scholarship highlight the positive effects of cell phone technology for peaceful forms of collective action, we argue that cell phones have another important effect: improved communication through cell phones can facilitate organization and coordination of groups for the purpose of violent collective action.

In a recent working paper, Shapiro and Weidmann (Reference Shapiro and Weidmann2012) pose a similar question about the spread of cell phone coverage and political violence in Iraq. The authors start from a theoretically ambiguous point. On the one hand, they emphasize that the availability of cell phones could lead to increased violence as it strengthens the position of insurgents against the coalition forces. On the other hand, cell phones could allow for better insurgent surveillance by U.S. and Iraqi forces, as well as lower the cost of whistle blowing on terrorists for the local population. Using district level data and a difference-in-difference design, the authors find that the expansion of the cell phone network in Iraq is associated with decreases in successful violent attacks by insurgent forces. Shapiro and Weidmann (Reference Shapiro and Weidmann2012) contend that this is due to the extensive use of cell phone surveillance by U.S. and Iraqi anti-insurgent forces as well as successful whistle-blower programs. Similarly, in the African context, Livingston (Reference Livingston2011) argues, that while cell phones might empower rebel groups and produce more violence, there also exists the potential for a reduction in violence through improved monitoring for international peacekeeping or governmental forces, although such efforts have been rare so far.

While improved monitoring and well-organized counterinsurgency activities can leverage cell phone coverage to increase the capacity of the state to uphold the monopoly of violence, it is unclear how easily this can be achieved in the African context. Furthermore, there exist strong theoretical considerations that suggest the marginal benefits of improved communication technology are substantial for insurgent groups.

Organizing violence is fraught with challenges. Successful insurgent activity requires solving various collective action and coordination problems (Kalyvas and Kocher Reference Kalyvas and Kocher2007; Wood Reference Wood2003), such as the free-riding problem (Olson Reference Olson1965). This is particularly true when it comes to the organization of political violence, where participation is risky and benefits are often unclear (Shadmehr and Bernhardt Reference Shadmehr and Bernhardt2011). Free riding within groups arises because members of insurgent groups have to endure the high costs of engaging in violence, but the potential payoffs for toppling the government will accrue to the wider population. Hence, rebel leaders have to ensure that group members actively contribute continuously throughout the conflict. Collective action problems also arise in the support network of rebel groups. Effective insurgencies rely strongly on the tacit support of the local population (Kalyvas Reference Kalyvas2006). Here, insurgents have to convince supporters to offer material support or valuable information from local residents, who themselves have an incentive to free ride.

In addition, insurgent groups suffer from strong coordination problems. Even if rebel groups can convince members to actively fight and the local population offers tacit support, military action needs to be carefully coordinated to be successful. Warfare against state forces with superior military technology, firepower, and training relies on careful plotting of attacks, appropriate timing, coordination of group movements in target areas, and managing the retreat to safe havens. While organizing protests is often about getting the right people together at the right time and place, insurgent violence requires the coordinated interplay of independent groups across distant geographic locations and time.

Recent work on mass media and violence has shown preliminary evidence on how radio and television can facilitate or block civil violence (Warren Reference Warren2013; Yanagizawa-Drott Reference Yanagizawa-Drott2012). Warren (Reference Warren2013) shows a reduction in militarized challenges to the state, if mass media access is widespread across its territory. The “soft power” of mass media enables the government to dissuade insurgent collective action through dissemination of progovernment propaganda. Observers of propaganda radio in Africa have highlighted the potential dangers for ethnic strife and violence (Livingston Reference Livingston2011). Going beyond qualitative accounts, Yanagizawa-Drott (Reference Yanagizawa-Drott2012) uses data on radio access in Rwandan villages to document the effects of “hate radio” on killings between Hutu and Tutsi during the genocide. Here, the use of mass media by one conflict faction shifted public perception and facilitated violent collective action. Both arguments emphasize the role of mass media in creating shared beliefs about the enemy and the convergence of privately held information. This can facilitate or hinder collective action and coordination. The formal literature on information and coordination problems in collective action against the government has also emphasized the importance of public (potentially government controlled) and private signals (Edmond Reference Edmond2012; Shadmehr and Bernhardt Reference Shadmehr and Bernhardt2011).

We contend that, in contrast to mass media, access to individual communication technology like cell phones can undermine the effects of government propaganda and, more importantly, play an integral part in overcoming other specific collective action and coordination problems inherent in insurgent violence. Through improved communication and monitoring, cell phone technology aids overcoming internal collective action problems, allows the distribution of information to tacit supporters in the wider population, and, on an operational level, allows for real-time coordination of insurgent activity.

Several organizational technologies can be used to improve cooperation among group members when dealing with the free-riding dilemma. Selective incentives and external punishment can be used effectively by rebel leaders to elicit support from rank-and-file insurgents and civilian supporters. At the same time, free-riding behavior can also be curbed through repeated interaction, increased communication, and improvement in the monitoring of group member's actions. The cheap availability of cell phones naturally improves and increases the communication between group members and allows for the tightening of group networks. The interaction between group members becomes more likely as the provision of cell phones makes long distance communication easier, especially in the context of rural insurgencies in which factions operate apart from each other for longer periods of time. The reduction of transaction costs resulting from the access to cell phone technology is especially valuable in many infrastructure-poor African regions, where this development makes personal long-distance communication possible for the first time. It is important to note that this does not require each individual to own a cell phone device, as cell phones can be shared collectively between group members or villagers.

Enlarging the communication network of rebel groups as well as increasing the rate of communication by group members should raise in-group trust between individual participants. The possibility for fast and easy communication boosts the propensity and rate of information sharing within groups, creating a shared awareness among group members. As Shirky (Reference Shirky2008, 51) writes, collective action is critically dependent on group cohesion. The expansion of within-group communication is likely to foster shared beliefs and awareness of groups, thus providing one channel of easing collective action. The higher rate of communication between individual group members also makes the transmission of messages and instructions from group leaders through the decentralized network more likely and efficient. Furthermore, the increase in two-way communication vastly raises opportunities for monitoring each other's behavior. Rebel leaders can exert better control over their rank and file and their wider support network, thus limiting free-riding behavior.

On a more general level, the spread of personal communication technology to the general population aids the flow of information and the coordination of beliefs not only within the particular groups, but also in the population. In instances when public or corporate private news sources are unavailable or pro-regime, the increased possibility of cell phone communication can aid the distribution of news. Anecdotal evidence suggests that cell phone communication can be useful as a substitute to traditional media, where the press is suppressed.Footnote ⁹ Indeed, tipping-point models of protest and popular support (Kuran Reference Kuran1991; Lohmann Reference Lohmann1994) suggest that if citizens are able to communicate their privately held beliefs about the regime, without fear of reprisal, public support for the regime can quickly transform into widespread opposition. The spread of cell phones makes the transmission of news to citizens throughout the country more likely. The support for insurgent activity can increase in the general population when news about government wrongdoings are communicated through citizen communication. For example, when news about indiscriminate killings by the government are more likely to travel through the population via cell phones, the general population may adjust the calculus of participation in nonviolent protests or even insurgent groups (Kalyvas and Kocher Reference Kalyvas and Kocher2007). Reportedly, cell phones have been used effectively by Syrian rebels to spread information on government atrocities and rebel victories, greatly aiding insurgency efforts (Peterson Reference Peterson2012). The ability to spread information about government violence against civilians or other forms of repression through private communication networks should thus improve the position of the insurgents within the population.

Apart from affecting a group's ability to address collective action problems, the distribution of cell phones aids the coordination of actions, especially during asymmetric insurgent warfare. On a basic level, it allows insurgent commanders to better plan and implement operations. As noted above, successful insurgent warfare against the state requires high levels of coordination. The availability of cell phones can aid violent groups in the planning and execution of operations. Reportedly, Charles Taylor successfully utilized mobile phone technology to coordinate and control his rebel commanders in Liberia's civil conflict (Reno Reference Reno2011, 4). Similarly, while Shapiro and Weidmann (Reference Shapiro and Weidmann2012) find a negative effect of cell phone availability on violence in Iraq, other research suggests that insurgents were aware and made use of the advantages of cell phones. One simple indication is that cell phone towers, in contrast to other infrastructure, were spared from insurgent attacks (Brand Reference Brand2007). In addition, the use of cell phones to communicate enemy movements, scouting, and other intelligence has been emphasized (Cordesman Reference Cordesman2005; Leahy Reference Leahy2005). Stroher (Reference Stroher2007) highlights the use of cell phones by Iraqi insurgents as an organizational tool, for the spread of information, as well as to provide propaganda to group members and the population.

This also indicates that the gain of cell phone technology by rebels can possibly close the technological gap between government troops and the rebel movement. Prior to the availability of cell phone communication to private citizens, it is likely that the government had a significant advantage when it comes to in-group communication and group coordination. This likely affects combat strategies as well as, indirectly, the probability of winning for each side. The availability of cell phones may thus decrease or close the size of this gap. Common conflict models assume technology as an important factor in determining the probability of winning of the fighting parties (Blattman and Miguel Reference Blattman and Miguel2010; Grossman Reference Grossman1991). Increasing the probability of winning by insurgents or rebels in turn should make the onset of conflict more likely.

While modern communication technology can play an important role for peaceful collective action in the form of protests, the marginal benefit of coordination is likely to be larger for organized violence. Protest in dense urban environments already enjoys several advantages for information sharing, monitoring, and coordination. Urban environments often offer other tools and opportunities to spread information and long-distance communication is less important in cities. Rural insurgents, on the other hand, can derive large benefits from private, mobile long-distance communication outside of major population settlements.

DATA

Testing the above specified argument requires a sample of cases in which violent collective action can conceivably be influenced by cell phone technology, as well as spatially disaggregated data on conflict and cell phone coverage. Given these requirements, focusing on the African continent offers several advantages over other world regions. The African continent has been, and still is, a major hotspot for organized violent conflict (The World Bank 2011), yet also exhibits strong temporal and spatial variation thereof. At the same time, cell phone technology has proliferated at a rapid pace across the continent in the last 15 to 20 years (Buys, Dasgupta, and Thomas Reference Buys, Dasgupta and Thomas2009), including to regions with characteristics that make them more prone to hosting violent events (e.g., aggrieved populations, poverty, difficult terrain, etc.). Often cell phones are the first long-distance communication device available in those areas. This confluence of factors creates an ideal environment to assess the impact of modern communication technology on facilitating violent collective action. Most other world regions lack such a high level of variance in conflict and access to cell phone technology. In addition, high-quality georeferenced data on conflict events is scarce for most regions of the world. Fortunately, recent years have seen an increase in the number of available conflict datasets that provide this type of information, in particular, for Africa.

For our primary analysis, we rely on the recently updated conflict data provided in the UCDP Georeferenced Event Dataset (UCDP GED) (Melander and Sundberg Reference Melander and Sundberg2011; Sundberg, Lindgren, and Padskocimaite Reference Sundberg, Lindgren and Padskocimaite2011). The UCDP GED includes yearly event data on organized violence in Africa from 1989 up to 2010. Violent events are included in the data if the conflict with which the event is associated has totaled 25 or more deaths and the event itself led to at least one death.Footnote ¹⁰ We use data on organized forms of violent collective action, instead of data on protests, for two reasons: First, our theoretical argument is geared specifically to the effects of cell phone communication technology on organized and violent forms of collective action. Second, quality and coverage of georeferenced data on organized violent collective action in Africa is higher than for other, more spontaneous and nonviolent forms of collective action.

Each event in the conflict dataset is specified to a location through longitude and latitude coordinates and by a date. We can use these data to map violent events across Africa for a number of years.Footnote ¹¹

Importantly, since the event data are based on news reports, one might expect the danger of measurement bias, as cell phone coverage may affect the probability of reporting of events. This is a valid concern, but we believe it is mitigated through several factors. For one, the UCDP coding team relies on a large number of print, radio, and television news reports from regional newswires, major and local newspapers, secondary sources, and expert knowledge, attempting to cover events even in remote locations without access to cell phone coverage. Furthermore, the focus on events with at least one death increases the likelihood of better event coverage in comparison to more low intensity events like peaceful protests or strikes.Footnote ¹² A quality comparison of the UCDP-GED and ACLED data by Eck (Reference Eck2012) concludes that the UCDP data have higher quality and report often dramatically more events in rural or remote areas compared to ACLED. In addition, we also control for a number of other factors that would account for measurement bias in the event count in our empirical models, such as distance to the capital, local GDP per capita, or population size. Conditional on these factors, it is unlikely that cell phone coverage will be associated with any further over-reporting of events.Footnote ¹³

Data on cell phone coverage are provided through Collins Coverage by Harper Collins Publishers. The data are made available by cell phone companies via the GSMA or Collins Bartholomew.Footnote ¹⁴ The availability and extent of coverage is represented via spatial polygons. We received data on GSM 2G network coverage for the first quarter of 2007, 2008, and 2009.Footnote ¹⁵ Our data only indicate the availability of cell phone services, not network traffic and usage by citizens. Information on usage is simply not available and without further information on the number of subscribers might also be misleading with regard to the role of cell phone communication for collective action in the wider population. As noted above, the argument does not require the ownership of cell phones by each individual, as phones can be shared within groups and villages. More importantly, we believe assessing the effect of coverage is more relevant from a policy perspective. While individual use of cell phones is hard to measure and control, coverage is the first and most important step in extending access of cell phone technology to the wider population.

Figure 1 shows the distribution of conflictual events in Africa in 2008, as well as areas with available GSM 2G coverage in 2007. As one can easily see, cell phone coverage is most widely spread in South Africa, Namibia, Kenya, as well as in northern Africa (specifically Morocco, Tunisia, and Egypt). However, coverage has expanded massively in the past years and has become more and more available in other areas of the continent. While coverage is more likely in coastal areas, the map clearly shows that it has been expanded further into the continent and away from population centers. Areas with a clear overlap in cell coverage and conflict events are in Algeria, the DRC, Kenya, Nigeria, Uganda, and Zimbabwe.

FIGURE 1.

Cell Phone Coverage 2007 (black) and Conflict Locations 2008 (gray) in Africa

To analyze the relationship between the local availability of cell phone coverage and the occurrence of violent events we follow Buhaug and Rød (Reference Buhaug and Rød2006) in relying on spatially disaggregated grid cells as our units of analysis. Our grid is partitioned into 0.5×0.5 decimal degree resolution cells, i.e., each grid cell is approximately 55 km × 55 km large. Using such high-resolution spatial units of analysis allows us to avoid problems of data aggregation common in cross-national studies of violence. The grid was created by Tollefsen, Strand, and Buhaug (Reference Tollefsen, Strand and Buhaug2012) at the Peace Research Institute Oslo (PRIO). The PRIO-Grid dataset provides grid cells and data for the whole world on a yearly basis from 1946 to 2008. Given our particular interest, we only use the data concerning Africa.

Using the grid provided in the PRIO-Grid dataset we create our dependent variables based on conflict locations in the UCDP GED dataset (Melander and Sundberg Reference Melander and Sundberg2011; Sundberg, Lindgren, and Padskocimaite Reference Sundberg, Lindgren and Padskocimaite2011). First, we generate a conflict indicator, a binary variable that takes the value of 1 in cases where one or more conflictual events were registered by UCDP GED in the given grid cell in 2008, and 0 otherwise. Our dataset consists of 10,674 cells, of which 3.3% experience violent conflict in 2008, thus conflict is quite rare. Second, for additional robustness checks we create a conflict count variable that counts the number of conflictual events in 2008 according to the UCDP GED data for each grid cell. Despite the increase in variation between grid cells, we use the count measure only as a secondary variable, because multiple counts within each grid-cell year are likely to be realizations of the same conflict process.

Our main independent variable of interest is generated in a similar manner. For each grid cell an indicator for cell phone coverage is created that takes the value of 1 if cell phone coverage existed in 2007 and 0 otherwise.Footnote ¹⁶ The distributions of cell phone coverage in 2007 and conflict locations in 2008 are presented in Figure 1. In 2007, cell phone coverage was available in 37% of grid cells; in 2008 this increased to 38%.

To identify a potential causal effect of cell phone coverage on violent conflict events we rely on three complementary strategies: First, we use a series of standard models on the cross-sectional data for 2008 and control for a number of potential confounding factors to approximate the potential causal effect of cell phone coverage. Second, we take the same cross-sectional data and implement an instrumental variable strategy that leverages exogenous variation in our main independent variable. Third, we use conflict data and lagged cell phone coverage for 2008, 2009, and 2010 to construct a short panel for African grid cells and implement a set of panel data approaches that exploit over-time variation.

CROSS-SECTIONAL ANALYSIS

Using the data described in the prior section, we estimate a series of cross-sectional statistical models to evaluate our hypothesis. The main measure of conflict we use is the simple binary conflict indicator for each grid cell. Naturally, we utilize the generalized linear model framework to formulate our probability models. The dependent variable y_i for each grid cell i in 2008 is binary,

and is modeled as a binomial process. We link the response variable to observed covariates via a standard link function (i.e., logit) to the linear predictor η: P(Y = y|X) = μ = g(η). The linear predictor in turn is a function of control variables and our cell phone coverage indicator: where x _i is a vector of control variables and the intercept and c_i is the indicator of cell phone coverage. The parameter γ measures the impact of improved communication technology on violent collective action.

We consider five alternative estimation approaches. Each of these models has certain advantages and disadvantages. They address distinct issues present in our data and differ in the severity of assumptions needed to attribute causal effects to the estimated cell phone coverage parameter. Table 1 shows parameter estimates and z statistics for all models. The first column shows a baseline specification of only control variables, estimated with a standard logit model and robust standard errors to address issues of heteroskedasticity. The second column simply adds our cell phone coverage indicator as a covariate. The third column presents the results of a rare-events logistic regression to account for rare events bias (Tomz, King, and Zeng Reference Tomz, King and Zeng2003). The fourth logit model includes country-level random effects to vary baseline levels of conflict across countries.Footnote ²² The standard random intercept model assumes zero correlation between the random effects and other covariates. To further control for potential omitted variable bias, we present in column 5 the results for a mixed effects logit model that also includes the country-level means of the cell phone coverage indicator. Including the country-level mean of the variable of interest allows for a correlation between the country random effect and the mean level of cell phone coverage (Gelman and Hill Reference Gelman and Hill2008, 506), removing the effects of country-level unobservables that affect cell phone coverage (Bell and Jones Reference Bell and Jones2012). Last, we also include the estimates of a linear probability model estimated via OLS that allows the inclusion of country-level fixed effects to control for any unobserved time-invariant country characteristics that might bias our findings.Footnote ²³

TABLE 1.

Binary DV Models

^† p = 0.1. *p = 0.05. **p = 0.01. ***p = 0.001.

The baseline specification in column 1 shows that a number of our control variables perform as expected. Prior levels of conflict have a statistically significant and positive effect on experiencing a conflict event in 2008. Similarly, population counts in the grid cell and mountainous terrain also increase the probability of conflict. On the other hand, in line with theoretical expectations and prior empirical findings the percentage of land with irrigation technology and GDP per capita reduce conflict (Buhaug and Rød Reference Buhaug and Rød2006; Lujala, Buhaug, and Gates Reference Lujala, Buhaug and Gates2009; Buhaug et al. Reference Buhaug, Gleditsch, Holtermann, Østby and Forø2011; Fearon and Laitin Reference Fearon and Laitin2003).

Across all models which include our measure of cell phone coverage, the cell phone coverage indicator is estimated to increase the probability of conflict and is precisely estimated—statistically significant below the 1% or even the 0.1% level. Even when controlling for the country level of cell phone coverage and only exploiting within country variation, as in the mixed effects logit model, or including country fixed effects, we always find a clear positive effect. Given that we control for a sizable number of confounding variables, as well as unobserved country-level factors, the results in Table 1 offer a good first approximation of the effect of cell phone coverage on political violence. In addition, including the cell phone coverage variable improves model fit statistics. A likelihood ratio test between a model including cell phone coverage and the nested model results in a significant test statistic at the 1% level in favor of the model including our variable of interest. We implement further analyses of the model fit, amongst others using separation plots (Greenhill, Ward, and Sacks Reference Greenhill, Ward and Sacks2011); these are displayed in the online Appendix.

INSTRUMENTAL VARIABLES

One important concern is the potential endogeneity between conflict events and cell phone coverage in each grid cell. It is plausible that conflict destroys cell towers and reduces coverage, suggesting a potentially negative relationship in the data. Although we lag cell phone coverage by one year in our models to address causal ordering and the estimated positive effect suggests we might actually underestimate the true relationship, we aim to address fundamental endogeneity concerns with an explicit identification strategy relying on an instrumental variable.

Prior work on the spread of cell towers in Africa has identified a number of geographic characteristics that predict cell phone tower locations (Buys, Dasgupta, and Thomas Reference Buys, Dasgupta and Thomas2009). However, in our case, for most of these variables the exclusion restriction of the instrumental variable estimator is likely to be violated since remoteness, difficult terrain, or population density have been found to predict conflict as well. Yet, in addition, Buys, Dasgupta, and Thomas (Reference Buys, Dasgupta and Thomas2009) identify the regulatory environment of African countries as an important predictor of cell phone coverage. A series of studies has found that healthy private competition in the cell phone market leads to better coverage and provision of cell phone services, in comparison to single-provider state-run systems (for references see Buys, Dasgupta, and Thomas (Reference Buys, Dasgupta and Thomas2009, 1495)). A World Bank study on telecommunications policy reform in 24 African economies finds important policy changes pertaining to privatization, increased competition, the formalization of regulations, and the creation of regulatory agencies, all contribute to improved coverage (The World Bank Reference Foster and Garmendia2010). Buys, Dasgupta, and Thomas (Reference Buys, Dasgupta and Thomas2009) find the World Bank's Country Policy and Institutional Assessment (CPIA) measure of regulatory quality is an important variable that affects cell phone coverage, even for highly spatially disaggregated data. We argue that the CPIA regulatory coverage measure is a good instrument for our purposes, since it is a robust predictor of cell phone coverage, i.e., avoids “weak instrument” criticisms, and has a strong claim to justifying the exclusion restriction. While regulatory quality affects private market competitiveness and the extent of cell phone coverage, we believe it is unlikely that an alternative, unmeasured causal link to violence exists.Footnote ²⁶

A potential issue could be that poorer African countries were forced to introduce regulatory reform in light of budgetary pressures and demands by outside actors. However, we find no aggregate link between the level of development and a country's regulatory score, as measured by the World Bank data. Furthermore, the estimated models include a control for GDP per capita levels and hence any effects regulatory quality might exert on violent collective action through poverty is accounted for.

We use the average CPIA regulatory quality score from 2005 to 2007 as our instrument for cell phone coverage. Initially we estimate a simple linear probability model via two-stage least squares (2SLS) and robust standard errors, which generally does well in identifying marginal effects, even with binary dependent variables (Angrist and Pischke Reference Angrist and Pischke2009, 197–205). We estimate models with and without the spatial conflict lag. For both models in the first stage regression of cell phone coverage on regulatory quality, our instrument, is highly statistically significant. The Kleibergen-Paap rk LM statistic and the Wald F statistic are very high and we are able to reject the null hypothesis of under- and weak identification. The results of the second stage are reported in Table 2, showing coefficient estimates and z statistics for the binary dependent variable models. Controls largely perform as expected, but more importantly the instrumented cell phone coverage variable is positive and highly statistically significant for both models. The exogenous variation in cell phone coverage induced by the regulatory quality increases the probability of local conflict events.Footnote ²⁷

TABLE 2.

Instrumental Variable Models

^† p = 0.1. *p = 0.05. **p = 0.01. ***p = 0.001.

PANEL DATA

Our last approach to estimate the effect of cell coverage on violent collective action exploits variation over time. Data on cell phone coverage is available for 2007, 2008, and 2009. We utilize these data to construct a short, three-year panel of our grid cells in 2008, 2009, and 2010. This offers us an additional opportunity to not only use geographic variation in coverage, but also changes over time. While in 2007 cell phone coverage existed in 36.9% of all grid cells, that value increased to 37.8% in 2008 and 42.5% in 2009. The expansion of cell phone coverage allows us to compare grid cells before and after the expansion. In a first step, we estimate OLS models for the binary and count dependent variable with country and year fixed effects and our standard set of controls, clustering standard errors at the country level. Again, we find a highly statistically significant and positive effect of cell phone coverage on violent conflict events (detailed results are presented in the online Appendix). More importantly, the panel structure allows us to now include grid-cell fixed effects. Here, we control for all observed and unobserved factors for each grid cell in the three-year period from 2008 to 2010. Since our standard control variables at the grid level are constant over time, we only include the cell phone coverage indicator as a predictor, apart from the grid cell and year effects. Table 3 shows the estimated coefficient for the cell phone coverage indicator for the binary and count measure.

TABLE 3.

Panel Data

*p < 0.05. **p < 0.01.

Clustered standard errors in parentheses.

For both models, the cell phone coverage variable is estimated to be positive and is statistically significant below the 5% and 1% level, respectively. The coefficient size suggests an increase of one percentage point for the linear probability model and 0.05 events with the count dependent variable, very much in line with our prior estimates of the effect size. The results are substantively unchanged if we log-transform the count to make the distribution appear more normally distributed or alternatively estimate a Poisson fixed effects model. Again, this confirms our previous findings that the expansion of cell phone coverage in Africa facilitates violent conflict events.

Overall, our quantitative models demonstrate a clear positive association between cell phone coverage and the occurrence of violent organized collective action. This effect persists when controlling for a series of standard explanations of violence, as well as unobserved, time-invariant factors at the country and even grid level. Plainly, our results suggest that local cell phone coverage facilitates violent collective action on the African continent.