What We (Do Not) Know about the Diffusion of Democracy Protests

What Terms Are Useful for Thinking about Democracy Protests?

Our analysis focuses on democracy protests, which are defined as “public demonstrations in which the participants’ primary demand is that countries adopt or uphold open and competitive elections” (2019, p. 2364; for further details, see Brancati 2016, 5-6). We understand that some readers may perceive our focus on democracy protests and their diffusion among neighbors as somewhat narrow. We believe this is not the case for several reasons. Democracy protests, we maintain, merit an analysis of their own because they are quite different from other forms of regime contention. Their demands pose a much greater threat to regimes than protests only seeking changes in public policy. Yet, their demands are more likely to resonate with activists in neighboring countries than other maximalist demands such as autonomy or independence. Moreover, democracy protests are at the heart of the debate regarding protest diffusion (Bunce and Wolchik 2006; della Porta 2014; Weyland 2014).

Although we agree that “pro-democratic contention and regime change cascades are set in motion by a mixture and combination of economic, social, and political demands,” we do not believe that this makes our definition unusually narrow (Weyland, X). Democracy protests, like other types of protests, are often catalyzed by multiple factors and can make demands on governments that extend beyond democracy (Beissinger 2013; Brancati 2016). We nevertheless maintain that it is important to think of protests in terms of their primary demands. Thinking of democracy protests in these terms also provides a hard test of our argument since protests that have coherent messages, according to diffusion theories, are more likely to be successful and to spread to other countries as a result (Benford and Snow 2000; Soborski 2016; Hatab 2018).

Our definition could nonetheless be problematic, as Kurt Weyland claims, because it excludes from the analysis a few protests that have been associated with well-known waves of protests such as the Arab Spring protests in Libya (2011), Tunisia (2010–2011), and Syria (2011). We anticipated this concern in our original article, replicating our analysis including these protests, and found that the results remain the same if these protests are included, even if the sample is restricted to 2011 and/or to only countries located in the Middle East and North Africa (see Online Appendix Tables A55–A57). As a further robustness test, we also dropped ambiguous cases for which it is debatable whether or not democracy was their primary demand (see Table 4 of Why Democracy Protests Do Not Diffuse) and repeated the analysis using three alternative measures of protests that included, but were not limited, to democracy protests (see Online Appendix Tables A46–A48). As in the original analysis, we found that a country is not more likely to experience any kind of protest if its neighbors recently experienced a similar protest.

We also do not believe that our focus on relatively close neighbors (i.e., within 50 km, 800 km, and UN-designated world regions) is unwarranted, either from a theoretical or empirical perspective. Although it is possible for protests to diffuse across greater distances due to technological innovations, we think this is unlikely. Nonneighboring countries are less likely to be similar to other countries and to provide relevant models for mobilization. Media coverage of protests in nonneighboring countries is also likely to be less extensive than coverage of neighboring countries since publics are likely to be less interested in distant protests. Our contention is consistent with Gleditsch and Rivera (2017), who find evidence of diffusion for nonviolent campaigns among neighbors but not globally. Furthermore, we believe that protests are unlikely to diffuse via demonstration effects over greater distances and time periods if they do not diffuse over shorter ones when the excitement and attention around protests are greatest. In any case, it is also worth noting that our analysis does not require protests to spread sequentially from closer to more distant neighbors as has been suggested.

How Should We Be Thinking about Diffusion?

In our article, we consider protests to have diffused “if, and only if, protests in one country made protests in the other more likely to occur” (2019, p. 2359). This is consistent with the standard notion of diffusion, which describes a situation in which “prior adoption of a trait or practice in a population alters the probability of adoption for remaining non-adopters” (Stang 1991, 325; Elkins and Simmons 2005, 36-38). However, a few comments to this forum raise questions as to whether this is a useful way in which to think about diffusion, so here we explain our choice more fully.

Weyland (2014) objects to our conceptualization of diffusion on the grounds that “[s]ince external stimuli do not always lead to successful emulation, diffusion should not be defined by increased chances of emulation” (p. 32). Valerie Bunce and Sharon Wolchik seem to share this view, arguing that diffusion occurs even if protests do not occur in other countries as long as the latter adopt “innovative demands for democratic change in the context of authoritarian regimes” (p. 2418). These innovations, which Bunce and Wolchik detail in their extensive research, include roundtables, legal challenges mounted by citizens, legal reforms enacted by reform-minded parliaments, and voter registration and turnout drives, among others.

We see value in this alternative conceptualization of diffusion in a number of ways. Most obviously, it captures the possibility that individuals are inspired by protests in other countries and seek to emulate them but do not because they anticipate that they will not be successful in this regard. In other words, democracy protests in neighboring countries may change the preferences of citizens in neighboring countries for protests without altering their behavior. This is an important point but one that is also consistent with what we argue and find in the empirical analysis.

It also allows for the possibility that protests in neighboring countries influence actors in other ways. We certainly recognize this possibility: as we noted in our article, “[o]ur analysis does not exclude the possibility that other aspects of protests, including their strategies, tactics, and techniques, diffuse across countries, nor does it rule out the possibility that democracy protests inspire other forms of mobilization” (p. 2380). Undoubtedly, these innovations may be very consequential in some instances. However, we do not think this means that a focus on protest-to-protest diffusion is undue. Innovations such as roundtable negotiations, preemptive legal reforms, online blogs, and so forth are very different from physical demonstrations. They are often initiated by different actors, pose very different risks for participants, and raise different challenges to regimes, and, therefore, we propose that they should be analyzed apart from protests. Such an approach is especially important since certain innovations may only diffuse where protests are already ongoing.

Our conceptualization of diffusion does not exclude the possibility that protests may diffuse in certain circumstances, or that any international factors matter, as has been asserted (Weyland, 2392). In fact, we explicitly state in the paper that

[o]ur analysis does not deny the possibility that democracy protests inspire protests in other countries in particular cases or that certain individuals are inspired to participate in democracy protests by protests in other countries, only that these cases are not part of a general trend. Lastly, our analysis does not rule out the possibility that other international factors, such as exogenous economic shocks, influence the likelihood of democracy protests to occur in countries, only that democracy protests in other countries do not. (pp. XX)

None of this ought to detract from the importance of understanding particular cases and the motivations of individuals. At least a thousand people died in Egypt, and many more were imprisoned, tortured, and sentenced to death, for protesting against the 2013 coup d’état deposing the country’s first democratically elected president (Roth 2014). Meanwhile, more than half a million people have thus far died in Syria as a result of a civil war that emerged from democracy protests in 2011 (Specia 2018). Much excellent work by the participants in this forum and others have provided rich empirical studies of such cases, which we could hardly do justice to in a single article (Bunce and Wolchik 2006; Beissinger 2007; Weyland 2009; Kienle and Sika 2015).

However, our objective in this project was not to identify individual cases or persons who may have been inspired by protests in neighboring countries, but to understand whether or not democracy protests diffuse in general. While we think it is important to understand particular cases, we it is equally important to understand general trends. As Houle and Kayser point out, “the diffusion literature has reached the point at which a null result constitutes an undeniable contribution to knowledge. The finding of Brancati and Lucardi (2019) makes precisely such a contribution by finding no evidence that a critical mechanism for the diffusion of democratization—the diffusion of democracy protests—occurs” (2019, p. 2434). It is exactly the reluctance to publish null findings that has resulted in “publication bias” in several academic disciplines (Gerber, Green, and Nickerson 2001; Franco, Malhotra, and Simonovits 2014).

Kurt Weyland suggests that our null findings lack importance because diffusion arguments would not have predicted that small and unsuccessful protests would spread to other countries (p. 2397). However, without the exhaustive analysis undertaken in this article and the massive data collection it required, we would not know that most democracy protests are typically too small, and the concessions they extracted from regimes too minor, for emulation to occur. Nor would we know how small is too small, or how minor is too minor, for diffusion to occur.

We do not believe this discussion about the importance of null findings ought to be cast in terms of a battle between quantitative versus qualitative research. We do not subscribe to the belief that quantitative approaches are inherently superior to qualitative approaches or that any methodology within either approach is inherently superior to another, but that each approach has advantages and disadvantages and that questions should be tackled from multiple angles using different methods. In order to understand specific cases, we believe that both qualitative and quantitative research is needed. The former offers the important flexibility that Kurt Weyland notes, and the ability to understand processes that are hard to quantify, while the latter allows one to compare large numbers of individuals within and across cases and, thus, to detect general trends.

How Much Confidence Can We Have in Our Null Findings?

When searching for a needle in a haystack, you can never be completely sure if you did not find one because there is no needle in the haystack, or that you just have not found it, and that if you looked longer, you would. Democracy protests are not exactly needles in haystacks: 310 democracy protests occurred between 1989 and 2011, but they occurred infrequently enough that it is reasonable to question whether the null results we find in our analysis are due to the infrequency of democracy protests or to a lack of diffusion. While it is possible that our null results are driven by the coding choices that we made or the small number of protests included in our analysis, we think it is unlikely for a number of reasons.

We do not think our coding decisions have driven the results because our choices are reasonable and because we use a number of different alternative coding schemes that produce consistent results. We analyzed the potential for diffusion over five different time intervals (i.e., 45 days, 90 days, 120 days, 360 days, and elections periods), three different historical periods (i.e., the fall of communism, the Internet era, and the Arab Spring), and three different definitions of neighbors (i.e., 50 km, 800 km, and UN-designated world regions). Our choices allow us to detect diffusion within East Central Europe in 1989 and the early 2000s, and within the Middle East and North Africa in 2011, because we analyze diffusion processes over short and long time periods, including across election periods, and because we analyzed each of these periods individually using the different time intervals specified above.

We also do not think the relatively small number of protests in our analysis has driven the results either because, as indicated in the paper, the huge number of observations included in the analysis makes it more likely that whatever results we find will be statistically significant. ² Moreover, the effect of elections is consistently significant across all our models. Yet, there are fewer observations involving elections in our analysis than there are protests in neighboring countries. ³ We also undertook a number of tests in our original article to determine the sensitivity of our results to the number of democracy protests in the analysis. All of these suggest that our results represent a true null effect. We collapsed the data, using weeks, months, or years as our analysis instead of days, so that the fraction of observations in which protests occurred is higher and found equally weak evidence of diffusion effects (see Online Appendix Tables A19–A44). We also analyzed other types of anti-regime protests that include many more instances of protests, and the results are also insignificant (see Online Appendix Tables A46–A48).

That said, our analysis only pertains to the post–Cold War period. It is conceivable that democracy protests may diffuse more commonly in the future. Weyland’s (2014) research suggests that this may be the case if, in the future, protests are organized less often by mass organizations, and more often by the masses, because those outside these organizations are more likely to rely on cognitive shortcuts that lead them to overestimate the odds of protests being successful in their countries. Thus far, the trend has not been in this direction. Throughout the 1989 to 2011 period, less than one-third of all democracy protests were organized by the masses, but the future may be different due to the widespread adoption of social media.

What Can Our Null Findings Contribute to Theory Development?

We argue in our article that democracy protests are generally not likely to diffuse across countries because the motivation for and the outcome of democracy protests results from domestic processes that are typically unaffected or undermined by the occurrence of democracy protests in other countries. We do not simply argue, as has been suggested, that democracy protests are driven by domestic factors, but that protests in neighboring countries do not significantly affect the domestic factors that motivate democracy protests in the first place. Democracy protests, as one of us has argued elsewhere, are more likely to arise when economic crises occur because crises raise discontent within society for governments in general and authoritarianism in particular (Brancati 2014, 2016). This elevates support for opposition candidates, who are more likely to win elections (Lucardi 2019; Reuter and Gandhi 2011) or, failing that, to organize protests when they believe that support for them is high (Brancati 2014, 2016). In contrast, democracy protests in neighboring countries do not significantly increase people’s awareness of their own or other’s discontent, nor do they significantly alter the economic conditions in other countries.

Our argument and findings give rise to a number of new questions as to why diffusion processes generally fail and under which conditions, if any, they succeed. Do protests, for example, fail to diffuse across countries because activists are not inspired by protests in other countries or because governments undertake measures to preempt protests? In our article, we find that countries whose neighbors experienced a democracy protest in the last 360 days were not more likely to restrict civil liberties or reduce media freedom within the year (see Online Appendix Table A15), which suggests that it is not the latter. However, further analysis is needed to systematically measure the responses of governments to protests in neighboring countries in order to answer this question and to determine why some authoritarian governments adopt preemptory actions while others do not. Are these autocrats those most capable and, thus, least likely to experience democracy protests in the first place?

Even if democracy protests are not generally likely to diffuse, are other types of protest likely to diffuse and if so, why? Protests organized by nongovernmental organizations, such as those against climate change, or those motivated by cross-border issues, may be more likely to diffuse because they have strong transnational networks, whereas those who organize democracy protests do not. Almost two-thirds of democracy protests are organized by opposition parties and/or their supporters (Brancati 2016, 18-23).

If democracy protests diffuse in very narrow circumstances, what are these circumstances? Our analysis does not find evidence that certain factors commonly thought to make protests more likely to diffuse, such as the influentialness of the neighboring countries, the similarity of states, the size or frequency of the neighboring protests, or the responses of governments to neighboring protests, matter. However, there may be other factors that are not common, but that might make protests more likely to diffuse. Henry Hale hypothesizes that whether countries share a common ideology, such as communism, might matter in this regard. Kurt Weyland suggests that a combination of factors might be decisive. Specifically, he argues that diffusion should occur after a long period of quiescence followed by large-scale protests that result in the expulsion of the autocrat from power, and that “[b]ecause all three conditions have to coincide to light a contentious wildfire, these outbreaks are rare” (p. 2398). To investigate this possibility, we looked at how often democracy protests broke out between 1989 and 2011 when large protests in which the chief executive was removed from office occurred in neighboring countries.

Between 1989 and 2011, there were eighteen democracy protests in which the chief executive was deposed—at sixteen of which turnout at the largest rally reached ten thousand or more people. ⁴ Eight of these events correspond to either the 1989 Revolutions (Czechoslovakia, Eastern Germany, and Romania), the Color Revolutions (Serbia [2000], Georgia [2003], and Kyrgyzstan [2005]), or the Arab Spring (Egypt and Yemen [2011]). ⁵ Depending on the neighbor definition employed, between seven and twelve (38.9–66.7 percent) of these cases were followed by a democracy protest in at least one neighboring country within a year. When we normalize the data to account for the number of neighbors each country has, only 9.8 percent to 14.4 percent of these protests were followed by a democracy protest in at least one neighboring country within a year. ⁶

If we also account for the baseline probability that a country will experience a protest in a given year in the absence of diffusion, these figures appear even smaller. To calculate this probability, we randomly sampled thirteen countries that would “experience” a democracy protest ⁷ and then calculated what proportion of these countries’ neighbors would also experience a democracy “protest.” We repeated this process 1,000 times. Even in the absence of diffusion—indeed, assuming that protests occur entirely by chance—the probability that a country experiencing a democracy protest will have at least one neighbor that also experiences a protest ranges between 22.7 percent and 52.4 percent, depending on the neighbor definition employed. When these figures are normalized by the number of neighbors, on average, 6.7 percent to 7.0 percent of a protesting country’s neighbors will also experience a bogus democracy “protest.” While lower than for the case in which the chief executive was deposed, the difference is not large, and the values increase quickly as the average number of protests per year goes up.

Even if democracy protests diffuse under very narrow circumstances whatever they are, certain individuals within countries may be inspired by protests in other countries to protest. If so, how much importance do they ascribe to these protests in their decision to protest? Existing surveys, however flawed, indicate that protests in other countries were not the main reasons why people joined protests in the Orange Revolution in 2005 and the Egyptian uprising in 2011. ⁸ However, more specific questions are needed to know whether or not people in these countries, or in future others, would have organized or participated in the protests were it not for the inspiration they received, indirectly or directly, from protests elsewhere.

Lastly, how do these findings help to explain the spatial and temporal clustering of democracy across the world? Christian Houle and Mark A. Kayser offer one such explanation. They hypothesize that democracy protests may help explain the collapse of authoritarian regimes, while diffusion processes explain whether states transition to democracy or not once regimes have broken down. From Brancati (2016), we know that very few democracy protests between 1989 and 2011 resulted in regime collapse, as they define it (i.e., the irregular removal of the national leader from office). All but seven of the seventeen authoritarian regimes in this period in which the national leader was deposed transitioned to a democracy within one year of the protests. The countries that did not were Mongolia (1989), Tajikistan (1991), Nigeria (1993), Ivory Coast (2000), Kyrgyzstan (2005), Togo (2005), and Yemen (2011). ⁹ None of these countries were in very democratic neighborhoods. Their hypothesis is, thus, certainly plausible, but further analysis is needed to test their hypothesis more rigorously.

Hale, meanwhile, suggests that protests may catalyze regime-ending splits among political elites, or follow from them, and urges that any research going forward on the relationship between protests and democratization ought to recognize the enmeshed nature of states and societies. Using data from Reuter and Gandhi (2011) on defections within hegemonic parties, we find that democracy protests do not occur around elite splits very often, but where they do, they occur more often after splits than before them—offering more support to the second scenario that Hales describes than the first. ¹⁰ A democracy protest occurred within one year of 29 percent of the defections (i.e., eight of twenty-eight) that occurred between 1989 and 2011. A defection occurs when a member of the ruling party leaves the party to run as an independent or a candidate for another party in elections determining the chief executive (i.e., presidential elections or legislative elections in the case of parliamentary systems). A democracy protest occurred the year before the defection in three of these eight instances. In the remaining cases, the protest occurred within one year after the defection, and in all five of these cases, the protests were organized by opposition supporters against the outcome of the elections.

The Way Forward

Rather than closing the case on diffusion processes, the discussion in this forum has raised new questions and identified several fruitful avenues for future research. These questions pertain to the specific circumstances under which democracy protests might diffuse and what aspects of democracy protests, if any, are generally likely to diffuse, even if democracy protests themselves do not. They also relate to the implications of our finding for future research on democratization. Each new question requires a different set of tools. There are no universally best methods. Only different methods for different questions that hopefully bring about new answers, both theoretically and empirically.