How Societal Approval Impacts Violence

Our theory presumes an Islamic society experiencing a large-scale, sustained campaign of violence carried out by organized militant groups motivated by some set of political goals. However, motivation alone is insufficient to achieve these goals. As Shapiro and Fair (Reference Shapiro and Fair2010, 84) argue, “militant organizations cannot engage in meaningful violence without some measure of popular support,” which, as we explain below, is in large part conditional upon community approval of the means used to achieve desired political ends.

Militant groups depend on societal support for manpower, materiel, sanctuary, and information to maintain their activities (Paul Reference Paul, Davis and Cragin2009), and court an audience of current supporters, the indifferent, and those that actively resist the group that can move across these categories in both helpful and harmful directions (Petersen Reference Petersen2001). Militant group survival requires that active support outweighs active resistance and that the group avoid actions that might push individuals from support to indifference, or worse, from either support or indifference to active resistance. Among supporters, “true believers” who share militant political goals, or “opportunists” who join a winning side, are most likely to withdraw support due to normative disapproval. Coerced support alone will likely be insufficient to offset losses due to normative disapproval. Since coercion is itself costly, it has the potential to push people toward resistance, creating a spiral requiring ever increasing coercion which may be beyond militant capabilities to sustain (e.g., Kalyvas Reference Kalyvas2006; Condra and Shapiro Reference Condra and Shapiro2012). It follows that for groups to balance tactical objectives and survival, maintaining voluntary support and, more importantly, avoiding active resistance, is crucial (e.g., Arjona, Kasfir, and Mampilly Reference Arjona, Kasfir and Mampilly2015). Indeed, the larger and the more ambitious a group, the more it depends on sustainable support and approval from the community in which it operates, for two reasons.

First, community support enables a militant group to grow and replenish its membership. While small terrorist groups numbering in the dozens can husband their resources by hiding from government forces or society for long periods of time, large militant organizations actively engaged in protracted campaigns of political violence cannot operate without losses, particularly if they are carrying out suicide attacks (Pape Reference Pape2005). The major militant groups operating in Afghanistan, Pakistan, and Iraq fall into this latter category, having sustained strengths estimated in the tens of thousands of fighters. Although foreign fighters represent one potential source of new recruits, they are rarely sufficient to replace the substantial losses incurred in intense campaigns of political violence. Indeed, an analysis of conflicts across the Islamic world since 1945 found that foreign fighters never comprised more than 5% of total combatants in any of them, including Iraq (Hegghammer Reference Hegghammer2010/11).

Second, community support is essential to enable large militant groups to avoid detection, surveillance, and elimination by the security forces of the government (Pape Reference Pape2005, 80–82). The fighters for the groups in our study are overwhelmingly volunteers, which requires these groups to keep a relatively high profile; therefore, it is impossible to prevent members of the local community in which the group operates from gaining basic information that would be useful to the government, such as the identity of recruiters, common locations for recruitment, and means of communication (Berman and Matanock Reference Berman and Matanock2015). Accordingly, without broad sympathy from the local population, these militant groups would be especially vulnerable to penetration, defection, and informants (Berman, Shapiro, and Felter Reference Berman, Shapiro and Felter2011). As active resistance increases, it becomes more difficult for the group to operate at all, let alone achieve its political goals. The threat to militant groups is very real: the Sunni militant group Al Qaeda in Iraq (AQI) was nearly destroyed when Iraqi Sunnis, its core constituency, turned against them and ultimately joined anti-Al Qaeda Awakening militias once the opportunity presented itself (Biddle, Friedman, and Shapiro Reference Biddle, Friedman and Shapiro2012).

Even militant groups that benefit from foreign support must manage the problem of backlash resulting from attacks that violate fundamental community norms in their areas of operation. While some argue that external support insulates militant groups from popular opposition (e.g., Salehyan, Siroky, and Wood Reference Salehyan, Siroky and Wood2014; Toft and Zhukov Reference Toft and Zhukov2015), foreign resources are unlikely to completely substitute for local resources for all but the smallest and most socially isolated groups. Local support is critical to the success of large and protracted insurgent campaigns that contest and control territory, as is the case for many Islamist organizations. The loss of local support matters not just by reducing material benefits that could be replaced by external sources, but also by pushing otherwise indifferent segments of society into more active resistance, from basic noncooperation to forming antigroup militias (Jentzsch, Kalyvas, and Schubinger Reference Jentzsch, Kalyvas and Schubinger2015). Moreover, this threat applies to small and isolated groups sustained heavily by external support, because a single betrayal can destroy them.

Put simply, militant groups will avoid violence when it risks the kind of societal disapproval that puts critical support at risk.

How Islamic Holiday Norms Affects Societal Approval

We argue that the primary way in which the Islamic calendar affects militant group decisions about violence is by creating anticipated costs for attacks on particular days. This mechanism is normative: on the most important holidays, there exists a deeply felt sense within Islamic societies that violence disrupting observance is especially inappropriate, or taboo—a tradition stretching back to the religion’s contentious formative years (Firestone Reference Firestone1999; Jackson Reference Jackson2002). Current and potential supporters judge militant groups not only by tactical success and failure but also the degree to which their violent activity conforms to predominant cultural—and in this case religious—norms. Normative violations thus risk turning current and potential supporters into bystanders or adversaries. Groups therefore anticipate punishment for violence that violates these norms, inducing restraint.

To be sure, societal approval is a complex phenomenon with variegated sources operating individually as well as interactively at multiple levels of society. For example, violence against innocent civilians, forcible recruitment of child soldiers, human trafficking, sex slavery, ethnic cleansing, use of human shields, and public executions all undoubtedly violate societal norms (e.g., Wood Reference Wood2009), that will force groups to consider the costs to support they might suffer for violations. Our argument is not that the Islamic calendar supersedes these other important norms, but that it works in conjunction with them. As a structural moderator, the Islamic calendar magnifies these existing general norms while also adding an additional time-delimited one. For example, an attack against civilians may generally cause outrage, but the same attack on a major Islamic holiday more so because it adds an additional source of anger. In short, the religious calendar interacts with other sources of societal disapproval but also produces a distinct independent effect.

In the case of the religious calendar, the mechanism of societal disapproval operates through the shared meaning holidays create in a societies with a common deeply held religious tradition. How does this work? According to Benedict Anderson and other scholars, understandings of time are fundamental to the self-identity and consciousness of national communities. For a religious community, its understanding of time creates an association among members that links earthly chains of events to the Devine in a manner they can all experience (Anderson Reference Anderson1983). As Elie Podeh (Reference Podeh2011, 16) explains, collective identity is fostered when “[h]olidays help establish an annual cycle of remembrance designed to ensure that several times every year members will recall certain ‘sacred’ moments from their collective past.”

In our cases, the most important holidays are Islamic religious holidays for which societies are given days off from their normal everyday obligations to dedicate themselves to observance. Islam is the official religion in Afghanistan, Pakistan, and Iraq, with overwhelming majorities in each state identifying as Muslim. Important Islamic holidays are thus observed by the entire national community generating widespread solidarity and empathy on these days. Public days off facilitate the dedicated observation of rituals and practices and thus signify important normative focal points in a given national context. Our theory does not classify holidays without days off—such as most days during the sacred month of Ramadan—as “important holidays,” not because they are insignificant to believers, but because they are not exclusively set aside for dedicated religious observance. Violence on important Islamic holidays is most likely to generate societal outrage because, in contrast with violence on Islamic holidays without days off, this violence mars the sanctity of days entirely reserved for practices central to the reaffirmation of religious identity.

Of course, Islamic societies are also divided along sectarian lines, and this is evident in the celebration of holidays, with some more significant for one sect than another, and others taking place sometimes days apart. Ashura, the day Shia Muslims commemorate their founding, is perhaps the best known example and serves as a flashpoint for sectarian violence in Iraq and Pakistan, and more recently in Afghanistan. However, important Islamic holidays are observed by Sunni and Shia alike and thus have the potential to reduce violence between sectarian rivals that share the same underlying religious norms (Hoffman Reference Hoffman and McAuliffe2016). When important religious holidays are observed with a national day off, it underscores their significance across any sectarian divides that may be present. Attacks on such religious holidays are thus likely to be met with society-wide outrage and disapproval, creating fear in the minds of militant groups contemplating violence on these days.

The central mechanism of societal disapproval operates irrespective of a given militant group’s ideology (whether religious or secular) because they are subject to a local normative context they do not control. It is society’s understanding of religion that constrains militant violence and not the particular understandings held by the militants themselves. Militants, like other potential norm entrepreneurs, cannot easily ignore or reframe a preexisting religious cultural landscape because these understandings are rooted in the diffuse semiotic practices of individuals and are thus challenging to transform (Wedeen Reference Wedeen2002). Therefore, it is the preexisting core beliefs of their local supporters that matters to militant groups contemplating violence. Nevertheless, even foreign donors may well share these norms of restraint on Islamic holidays and punish groups that defy them (e.g., Stanton Reference Stanton2016). The violation of holiday norms thus poses a significant risk to any militant group operating in an Islamic society, even to those whose primary support base is non-local (contra Toft and Zhukov Reference Toft and Zhukov2015).

Alternative Mechanisms

Two alternative mechanisms could also explain why violence would decline on important religious holidays. Both of them operate by increasing the practical challenges posed to militant groups contemplating violence, rather than through militant fears of lost societal support for attacking on these days.

First, important Islamic holidays might affect violence through the mere fact that they are also public holidays, and that it is the disruption entailed in any day off that is important, not the religious meaning of the holiday per se. Days off are associated with significant changes in the pattern of societal activity, with individuals potentially gathering in their homes or other dispersed locations leaving militant groups with fewer vulnerable targets for their violence (Hassner Reference Hassner2011, 501–503). Yet, in addition to religious holidays, societies also tend to observe a number of secular days off (not least being the weekend) that would similarly reduce available targets. If it is the day off that explains systematic reductions in violence on important religious holidays, it should do so on secular days off as well.

The second alternative mechanism for reduced violence on major religious holidays in Islamic societies is the heightened security measures taken in anticipation of militant violence on these days. Since these focal-point opportunities for violence are easy to foresee, governments routinely deploy additional security forces in vulnerable areas.Footnote ³ Accordingly, violence may go down on these days not because society disapproves of it, but because executing attacks is more difficult. However, if this mechanism explained the reduction of violence, it would do so any time counterinsurgent forces were deployed, including on any society-wide day off when they are also routinely brought out in force to anticipate possible militant disruption.

By including both secular days off and recent counterinsurgent military activity in our analysis, our research design allows us to assess both alternatives.

Expected Islamic Calendar Effects

Our theory predicts that the moderating effect of the Islamic calendar on violence is conditioned on public holidays, resulting in the 2 x 2 representation provided by Table 1.

TABLE 1.

Predicted Effects of Islamic Calendar and Public Holidays on Violence

Based on our theory, we expect important Islamic holidays to have regular and predictable restraining effects on violence in our cases. We identify important Islamic holidays as those with official nationally observed days off.

However, our theory also confronts three further alternative hypotheses on the relationship between holidays and violence. First, it is possible that our theory is incorrect on the suppressive effect of holidays, religious or otherwise, and that the enhanced opportunity to conduct attacks for tactical or symbolic reasons would lead to an increase of violence on such days (e.g., Toft and Zhukov Reference Toft and Zhukov2015; Hassner Reference Hassner2011). Second, it may be Islamic religious meaning alone driving a suppressive moderating effect. In other words, even Islamic holidays without days off would experience the same violence-restraining effect. Finally, societal disruption or added security precautions may act to suppress violence on any public holiday, rather than just religious ones.

Dependent Variables

We test for the effects of the Islamic calendar on incidences of political violence in our three cases, in Pakistan and Iraq at the province-day, and in Afghanistan at the Regional Command (RC)-day.Footnote ⁴ We disaggregate our country cases to subnational geographic units to account for local variation in weather conditions. We define political violence broadly to include (1) both fatal and nonfatal violent attacks; (2) initiated by organized militant groups and unaffiliated individuals; (3) using a wide range of methods from kidnapping to assaults to suicide attacks; (4) against a wide range of government, military, and civilian targets; (5) in support of political objectives, broadly defined. We include violence by all militant groups within the geographic boundaries of our cases, both “Islamist” and otherwise. Pooling violence irrespective of group motivation is a strong test of our claim that it is societal norms, not a particular group’s ideology, that generates the suppressive effect of important Islamic holidays. If non-Islamist groups operated according to different rules, including their violence would lower the chances of confirming our theory.

We use five different sources of political violence events data, two that span across our cases and three more specific to each case. Because each dataset is measuring the same underlying construct—political violence–we expect Islamic calendar effects to be visible across them. If this is true, we can be confident that our findings reflect real effects in militant decision making on the timing of attacks.Footnote ⁵

We use two main datasets—the U.S. National Counter Terrorism Center’s World Incident Tracking System (WITS) database and the National Consortium for the Study of Terrorism and Responses to Terrorism’s Global Terrorism Database (GTD) —to measure political violence across all three cases in our study. Both the WITS and GTD databases use media reports to collect global political violence initiated by nonstate actors (militant groups and individuals) against a wide range of government, security, and civilian targets while excluding ordinary crime and militant violence resulting from ground battles initiated by the state (LaFree and Dugan Reference LaFree and Dugan2007; Wigle Reference Wigle2010; Sheehan Reference Sheehan, Lum and Kennedy2012). They are not identical, but both capture a broad spectrum of both terrorist and insurgent violence while excluding potentially confounding events. For each dataset, we aggregate attacks to the province/RC-day for the WITS from 2004 to 2011 and for the GTD from 2004 to 2014.

In addition to the main two datasets, we supplement each case with a third, case-specific dataset of political violence events. For the Iraq case, we use the Iraq Body Count (IBC) database provided by the Empirical Studies of Conflict Project (ESOC) to create a province-day dataset of fatal events attributed to nonstate actors from 2004 to 2009 (Condra and Shapiro Reference Condra and Shapiro2012). For the Afghanistan case, we use insurgent violence reported in unclassified Significant Activities data (SIGACTs) from 2008 to 2014 released to the public via the U.S. Central Command (CENTCOM).Footnote ⁶ SIGACTs are individual “significant” events involving U.S. and NATO military forces in Afghanistan, and include a wide variety of enemy and coalition activity. Each event is timestamped, geolocated, and categorized based on event type. To construct our RC-day dataset, we use event classifications to filter the SIGACT data for enemy actions geolocated within the borders of Afghanistan and indicate intentional insurgent actions. Finally, for our third dataset in the Pakistan case, we use geolocated data on nonstate political violence from 2004 to 2014 collected in the the Pak Institute for Peace and Security’s (PIPS) Digital Database on Conflict & Security.

Figure 2 charts our dependent variables in each case, aggregated to the month for ease of comparison. Given the variation in levels violence captured by these datasets, any consistent findings across them should be taken seriously, while inconsistent findings should be interpreted cautiously. Summary statistics for our nine dependent variables are presented in Table 2.

FIGURE 2.

Distribution of Attack Dependent Variables in Afghanistan, Iraq, & Pakistan, 2004–2014

TABLE 2.

Summary Statistics of Dependent Variable Event-Counts in Afghanistan, Iraq, and Pakistan, 2004–2014

The Islamic Calendar

To model the Islamic calendar, we constructed variables mapping Islamic holidays to their corresponding Gregorian dates.Footnote ⁷ The two largest religious holidays on the Islamic calendar are the 3-day festival of Eid al-Fitr, which begins on the first day of Shawwal and marks the end of the fasting month of Ramadan, and the 4-day festival of Eid al-Adha, which takes place on the 10th day of Dhu al-Hijja and coincides with the Hajj, the annual pilgrimage to Mecca in Saudi Arabia that all Muslims are obligated by the Koran to complete at least once in their lives. A third important holiday, Mawlid an Nabi, commemorates the prophet Mohammad’s birthday and is celebrated during the month of Rabi al-thani, by Sunnis on the 12th and Shia on the 17th. Throughout the Islamic world, Muslims are given official days off to celebrate these holidays, as they are in our three cases of Iraq, Afghanistan, and Pakistan. Two Islamic holidays in the month of Muharram, the Islamic New Year, celebrated on the 1st, and Ashura, a holiday most central to Shia and celebrated on the 10th, are also public holidays in our three cases. Other potentially relevant Islamic holidays include Laylat al-Bara‘at and Lyalat al-Qadar during Ramadan, along with Isra‘ and Mi‘raj during the month of Rajab, which are not public holidays, and the first day of Ramadan and the Day of Arafa, which are only public holidays in Afghanistan.

We also include five sacred months as “religious holidays without days off” in our study. Ramadan is perhaps best known. It is Islam’s holiest month, the only month mentioned by name in the Koran, during which Muslims worldwide are obligated to fast during daylight hours (Plessner Reference Plessner, Bearman, Bianquis, Bosworth, van Donzel and Henrichs2016). Four additional months are the so-called Al Ashur al Haram or peace months—identified in Hadiths as the months of Rajab, Dhu al-Qidda, Dhu al-Hijja and Muharram—during which warfare is expressly forbidden (Firestone Reference Firestone1999). None of these five months are fully public holidays, though Rajab, Dhu al-Hijja, and Muharram have Islamic public holidays that take place during them.Footnote ⁸ Finally, we also include Fridays, Islam’s weekly day of prayer (Jumma), which is part of the weekend and therefore a day off in Afghanistan and Iraq, though not in Pakistan. The full set of holidays is presented in Table 3.

TABLE 3.

Holidays on the Islamic Calendar

The start of any Islamic month is governed by the sighting of the new moon at sunset, which can lead to variation among cases on the specific Gregorian date on which Islamic holidays are celebrated (Karamati et al. Reference Karamati, Negahban, Pakatchi, Poor, Madelung and Daftary2016). To capture the correct dates on which Islamic holidays were celebrated in our cases, we constructed case-specific Hijri calendars based on official moon sightings marking the beginning of each of the 136 Hijri months in our study. For Pakistan, we used official declarations by the Central Ruet-i-Hillal Committee, while for Iraq and Afghanistan, which follow Saudi Arabia, we used official declarations by the Saudi Supreme Court.Footnote ⁹

Controls

Perhaps the biggest threat to our ability to infer the effects of holidays on violence is the possible omitted-variable biases exerted by alternative structural moderating factors. Accordingly, in addition to religious holidays, we also control for secular holidays, economic, weather, and political cycles, as well as counterinsurgent military activity.

To test whether the relationship between Islamic holidays was an the result of having a day off, not the religious significance of the Islamic holiday per se, we identified secular public holidays to include as controls in our model.Footnote ¹⁰ In our cases, secular holidays such as Independence Day and Labor Day are not linked to the Islamic calendar. Pakistan and Iraq use the Gregorian calendar, whereas in Afghanistan, with the exception of Labor Day, secular holidays are marked on the Shamsi Hijri calendar, the same solar calendar used in Iran. In addition, we included secular weekend days (i.e., weekend days other than Fridays: for Iraq, Thursdays, and for Pakistan, Sundays). Afghanistan has only one official weekend day, Fridays. The full list of secular national days off for each case is presented in Table 4.

TABLE 4.

Secular Holidays by Country-Case

The existing literature has particularly emphasized the importance of food prices (e.g., Bellemare Reference Bellemare2015) and oil prices (e.g., Fearon Reference Fearon2005) on domestic political unrest and violence. To control for broad factors resulting from these international economic cycles, we include measures of daily global prices of two critical commodities, grain (the International Grains Council composite Grain and Oilseed Index) and oil (Brent crude prices from the U.S. Energy Information Administration).

Four meteorological variables—temperature, cloud cover, precipitation, and snow cover—have featured prominently in the literature on weather as a structural moderator of violence (e.g. Winters Reference Winters1998; O’Loughlin, Linke, and Witmer Reference O’Loughlin, Linke and Witmer2014; Carter and Veale Reference Carter and Veale2013). We measured these variables at the level of our geographic units to maximize fidelity to local weather conditions. We use global gridded daily averages estimated from a mix of weather station and remote sensing data and aggregated to the province/RC-day, derived from NCEP Reanalysis data provided by the National Oceanographic and Atmospheric Administration’s (NOAA) Earth Systems Research Laboratory for temperature, cloud cover, and snow cover (Kalnay et al. Reference Kalnay, Kanamitsu, Kistler, Collins, Deaven, Gandin, Iredell, Saha, White, Woollen, Zhu, Leetmaa, Reynolds, Chelliah, Ebisuzaki, Higgins, Janowiak, Mo, Ropelewski, Wang, Jenne and Joseph1996), and the Global Precipitation One-Degree Daily Data Set for precipitation produced by NASA’s Goddard Space Flight Center (Huffman et al. Reference Huffman, Adler, Bolvin and Gu2009).

As the literature on electoral violence suggests, pre-election periods are particularly dangerous because they act as focal points that combine an opportunity for symbolic political contestation with the potential to have direct influence on electoral results (e.g., Staniland Reference Staniland2015). In addition, because the United States plays a prominent role in each of our cases, we also control for the possibility that American election cycles influence the timing of militant violence. To control for political cycles, we account for periods immediately preceding these important national elections (executive and legislative) using election data from the National Elections Across Democracy and Autocracy Dataset (Hyde and Marinov Reference Hyde and Marinov2012) through 2010, and updated through 2014 using popular sources. We control for the presence of a local national election, and the presence of an American election, within the 30 days proceeding the observation in question.

Finally, to control for the suppressive effect of recent counterinsurgent military activity, we include a 1-day lagged measure of militants killed in clashes with government forces in a given province/RC-day from 2004 to 2014 based on data in the UCDP-GED version 5 dataset (Sundberg and Melander Reference Sundberg and Melander2013; Croicu and Sundberg Reference Croicu and Sundberg2015). We combined militant fatalities with uncategorized fatalities for the included events to capture the greatest possible range of government counterinsurgent activity.

Model Specification

The overdispersion of our count dependent variables makes negative binomial regression the appropriate statistical estimator for our purposes. The negative binomial is given by:

(1) $$\begin{equation} P(Y_t)={\displaystyle \frac{\Gamma (\theta +Y_t)}{Y_t!\cdot \Gamma (\theta )}}\cdot u^{\theta }_t(1-u_t)^{Y_t}, \end{equation}$$

where P(Y _t ) is the probability of Y _t events at time t, Γ is the gamma function, u _t is θ(θ + λ _t ), θ is $\frac{1}{\alpha }$ , α is the dispersion parameter, and λ _t specifies the negative-binomial model parameters (Hilbe Reference Hilbe2007). Our estimations employ a λ _t of the general form:

(2) $$\begin{eqnarray} ln(\lambda _t)&=&\beta _{0} +\beta _{1}LocalElection_{t} +\beta _{2}USElection_{t} \nonumber\\ &&+\,\beta _{3}OilPrice_{t} +\beta _{4}GrainIndex_{t}\nonumber\\ && +\, \beta _{5}Temperature_{t} +\beta _{6}SnowCover_{t}\nonumber\\ && +\,\beta _{7}Precipitation_{t} +\beta _{8}CloudCover_{t}\nonumber\\ && +\, \beta _{9}COINActivity_{t-1}\nonumber\\ && +\,\beta _{10}DaysFromOnset_{t} +\beta _{11}Y_{t-1}\nonumber\\ &&+\, {\bf GregorianYear_{t}\phi } +\,{\bf Province_{t}\phi }\nonumber\\ && +\,{\bf HolidayPredictors_{t}\beta } +\epsilon _t, \end{eqnarray}$$

where the exponential function of ε _t is a gamma-distributed error term with a mean of 1 and a variance of α. HolidayPredictors_t represents a vector of holiday indicators, which we use to perform our central hypothesis tests. GregorianYear_t and Province_t represent vectors of indicators capturing the year and location of each observation creating an unconditional fixed-effects negative binomial model, which is preferable to the conditional given the manageable number of panels in our data (Hilbe Reference Hilbe2007, 199–208). Since our data takes the form of a time-series panel, we include a lagged value of the dependent variable (Y _{t − 1}) to account for possible autocorrelation across consecutive cases, and the count DaysFromOnset _t to capture trend effects.Footnote ¹¹

Primary Analysis

To assess the effect of the Islamic calendar on violence, we performed nine negative binomial regressions. Here, the key holiday predictor variables are operationalized according to the typology presented in Tables 3 and 4 above as the most straightforward and direct test of our hypotheses. These regressions are grouped into three clusters of three, with each model estimating the effect of our predictors separately on each dependent variable in each case. Table 5 reports the results.

TABLE 5.

Negative Binomial Regression Analysis of Daily Violence in Afghanistan, Iraq, and Pakistan, 2004–2014

Notes: Standard errors in parentheses; p-values in brackets; *Significant at .05, **Significant at .01, ***Significant at .001 (Two-Tailed)

Our primary analysis provides a strikingly strong confirmation of H1 across all nine models, thus confirming our core theoretical expectations. Controlling for numerous most-likely confounding factors, we find consistently negative and significant coefficients for days that are “Both Religious and a Public Holiday,” which represent the effect of such days on insurgent violence relative to a “normal” day in the life of each conflict. Indeed, this religious public holiday indicator is the only variable to achieve such a consistent statistically significant effect. Notably, H2 was not supported, as religious days that were not also public holidays—which includes most days in the month of Ramadan—did not significantly and consistently reduce violence. Similarly, public holidays in general also failed to consistently predict systematic increases (H3) or decreases (H4) in levels of violence, all else being equal. Taken together, these findings not only provide a 9 for 9 confirmation of our core hypothesis but also suggest that this suppressive effect does not generalize to just any holiday, religious or secular.

To identify the substantive suppressive effect on violence of important religious holidays, we conducted simulations of our results using the Clarify software (King, Tomz, and Wittenberg Reference King, Tomz and Wittenberg2000). Table 6 provides the percentage shift in the mean expected values yielded by our 1,000-per-model Clarify simulations, holding all else constant.Footnote ¹²

TABLE 6.

Substantive Shift in Significant Clarify Mean Expected Violent Attacks (Source: Negative-Binomial Analysis of Afghanistan, Iraq, and Pakistan, 2004–2014, Table 5)

Note: Significant substantive declines bolded for emphasis.

Cast in terms of percentage-shifts, Table 6 presents the relative impact of our religious public holiday predictors versus the other significant explanatory variables. Several important conclusions follow. The impact of religious public holidays is not only consistently negative, but sizable in effect: a decline in violence of roughly 25% on average compared to a “normal” day in the life of each conflict. This result needs to be understood in the context of other structural-moderating causal forces operating on daily violence levels. Indeed, none of our control variables garnered suppressive effects of a substantially different magnitude, with virtually all of them well below the 25% average decline found with religious public holidays. Not only do religious public holidays exhibit a significant negative relationship with observed violence levels, but also their influence is likely unique in terms of both size and consistency across our cases. Theoretically, this result makes sense given that—unlike the effects of politics, weather, economics, and counterinsurgent activity which might vary a great deal across different circumstances—Islamic holidays are shared across our three otherwise very different country-cases.

To explore the sensitivity of our models to specification issues, including the effect of the large number of independent variables required to capture our various controls, we reran all nine of our primary models including only the three holiday predictor variables, with substantively similar results (see Supplemental Table 1).Footnote ¹³ Second, there is a risk that our findings were an artifact of our choice to use a negative binomial estimator for our count response dependent variables. To test whether a binary estimator would have produced different results given the large number of zeros in our data, we reran all nine of our estimations using logit regression, and again found substantively similar results (see Supplemental Table 2).Footnote ¹⁴ This initial set of checks gives us confidence that our findings are sufficiently robust to alternative model specifications.

Exploration of Individual Holidays

Did our decision to “group” holidays together into categories for our primary analysis obscure the effects of Ramadan, or other specific holidays? To explore this possibility, we assessed the effect of each holiday or “holiday period” individually. This extension allows us to evaluate which holidays are having possibly disproportionate, or hidden, effects on violence levels that cannot be detected when they are grouped together for the sake of estimation.

To facilitate our discussion, we only present the summary results of the Clarify simulations derived from these models, again reporting the percentage shift in the mean expected-values for statistically significant explanatory variables in Table 7, holding all else constant (see Supplemental Table 4 for the full estimation results).

TABLE 7.

Substantive Shift in Significant Clarify Mean Expected-Violent Attacks By Holiday (Source: Negative-Binomial Analysis of Afghanistan, Iraq, and Pakistan, 2004–2014, Supplementary Table 4)

Note: Significant substantive declines bolded for emphasis.

From Table 7, at least two important conclusions are evident. First, among the results on the 18 different Islamic holidays we examine, there are almost no examples of a holiday exerting a significant positive effect on violence, and none that do so consistently. With a scattered few exceptions, Islamic holidays are almost always associated with declines in violence. Second, the dearth of significant results on secular holidays confirm they have no analogous systematic effect on violence (either positive or negative). These results challenge the common presumption of a systematic Ramadan surge in violence: we found a significant increase in only one of our nine models, a significant decrease in another, and no significant effect in the other seven.

Significantly, while few Islamic holidays exert consistent effects across our cases, the exceptions are precisely those “public holiday” religious days that serve as the focal point of our theory. In particular, Eid al-Fitr proved to be the only Islamic holiday associated with significant declines in violence across all nine models. We also found substantial suppressive effects on Eid al-Adha, though they were statistically significant in only six of our nine models. Friday was found to have a significant negative effect on violence in all six of our models associated with Afghanistan and Iraq, but none in Pakistan—the sole country-case among our three which does not observe a “public holiday” that day of the week. Taken together, these results demonstrate the strengths and weaknesses of Hassner’s (Reference Hassner2011) analysis. While they replicate Hassner’s findings on these three holidays in Iraq, they nevertheless reveal that not all of these holidays have clearly analogous effects across Islamic societies. Moreover, our nonsignificant findings on Fridays in Pakistan suggests a pattern our theory is better equipped to explain.