Skip to main content
SpringerLink
Log in

Does Repression of Campaigns Trigger Coups d’État?

  • Chapter
  • First Online:
The Politics of Repression Under Authoritarian Rule

Part of the book series: Contributions to Political Science ((CPS))

  • 759 Accesses

Abstract

Does complementarity between restrictions and violence stabilize authoritarian power-sharing in the face of popular rebellion? Scholars widely concur that the central political conflict in authoritarian regimes plays out between people on the inside of the regime. This chapter adds to the debate and studies coup attempts in light of two interconnected hypotheses. First, violence against campaigns destabilizes power-sharing because it exposes a weak leadership. Second, this adverse effect of violence declines as the routine level of restrictions increases, because restrictions act as a sorting mechanism for uncompromising political opposition. Both hypotheses are tested using Bayesian multilevel statistical analysis on a data set of 253 coup attempts in 198 authoritarian regimes between 1949 and 2007. This study design allows separation of repression’s time-dependent effects from its context effects, and it demonstrates the value of Bayesian methods for studying rare political phenomena such as coups d’état. The chapter’s conclusion, however, is straightforward: Once citizens form campaigns, repression can only deteriorate the situation because it opens a frontline right at the center of authoritarian rule.

This is a preview of subscription content, log in via an institution to check access.

Access this chapter

Log in via an institution
Chapter
USD 29.95
Price excludes VAT (USA)
  • Available as PDF
  • Read on any device
  • Instant download
  • Own it forever
eBook
USD 99.00
Price excludes VAT (USA)
  • Available as EPUB and PDF
  • Read on any device
  • Instant download
  • Own it forever
Hardcover Book
USD 129.99
Price excludes VAT (USA)
  • Durable hardcover edition
  • Dispatched in 3 to 5 business days
  • Free shipping worldwide - see info

Tax calculation will be finalised at checkout

Purchases are for personal use only

Institutional subscriptions

Notes

  1. 1.

    In the aftermath of the 1988 coup, the State Law and Order Restoration Council (SLORC), Burma’s new and until recently current military junta, changed the country’s name to Myanmar.

  2. 2.

    Allegedly, the military played a decisive role in that development. See Ferrara (2003) and Lintner (1990).

  3. 3.

    This analysis disregards the question of success because every single coup attempt expresses important divisions among authoritarian elites. On the difficult definition and measurement of successful coups see Kebschull (1994), Powell and Thyne (2011).

  4. 4.

    See Tullock (1987, Chap. 2), for an early adaptation to authoritarian rule.

  5. 5.

    Alternatives to the language and theory of rational choice exist. More traditional studies of the coup d’état prefer classificatory approaches. Accordingly, a coup d’état involves different narratives depending on whether it is a veto, reform, or guardian coup (see Ezrow and Frantz 2011, 98). However, those classifications constitute ex post attributions. They provide helpful summaries of political outcomes once a coup has been attempted. This chapter, however, focuses on the prelude the coup d’état.

  6. 6.

    For a related account that discusses the merits of hereditary succession in non-monarchies see Brownlee (2007).

  7. 7.

    On the importance of disciplined nonviolence see Martin (2007), Chenoweth and Stephan (2011).

  8. 8.

    It is an events-based measure that includes acts of politically motivated assassinations, purges, guerilla warfare, protest, riots, and general strikes. See Powell (2012, 1028).

  9. 9.

    See Nepstad (2013) for a related account.

  10. 10.

    This observation is not new. Actor-centric theories of regime transitions often distinguish between regime soft- and hardliners who play “games of transition” (Przeworski 1992). According to this classical literature, the outcomes of transitions from authoritarian rule depend on whether soft-liners can strike bargains with moderate opposition representatives while keeping hardliners under control (Przeworski 1991; O’Donnell and Schmitter 1986). The result is an optimistic picture of democratization that ascribes “independent and decisive significance to the agents” (Møller and Skaaning 2013a, 128), but deemphasizes structural requisites of democracy (Merkel 2010, 488). For the study of the coup d’état distinctions such as soft- and hardliner are of little use because they require post hoc judgments of coup outcomes. Moreover, those labels strongly relate to the study of regime transitions, but this chapter focuses on their prelude. Therefore, transitology offers little leverage on the fundamental theoretical question of this chapter even though it starts from similar considerations.

  11. 11.

    Note that the hypothesis does not generate expectations on the effect of violence under the alternative condition. Some may regard this as an undertheorized conditional relationship (Berry et al. 2012). Notwithstanding, extant research shows that this effect is tough to anticipate. On the one hand, totalitarian regimes have routinely combined massive, ideologically motivated, and violent suppression without suffering coup attempts. On the other hand, coups were routine in Middle Eastern and North-African authoritarian regimes at much lower levels of violence. More importantly, H1’s veracity does not require any such statement. The hypothesis merely requires the effect of violence given campaigns is positive and larger in magnitude than the estimate under the alternative condition.

  12. 12.

    See Chap. 3.

  13. 13.

    See Kruschke (2014) and McElreath (2016) for applied introductions to Bayesian inference and data analysis. Jackman (2009) and Gelman et al. (2013) offer mathematically more rigorous, but readable introductions. For a discussion of multilevel analysis from a Bayesian perspective see Gelman and Hill (2006).

  14. 14.

    See Sect. 5.4.2.

  15. 15.

    The reader should deemphasize relative frequencies of 0.5, which summarize only four authoritarian regimes.

  16. 16.

    Refer to the appendix for further information.

  17. 17.

    This measure divides the total annual value of the national oil and gas production in 2000 US Dollars by the size of the population (Gleditsch 2002; Ross and Mahdavi 2015).

  18. 18.

    Figures are based on version 5.0 of the National Material Capabilities dataset.

  19. 19.

    The ratio of total imports and exports to GDP per capita measures trade openness. All figures are in current-year US Dollars.

  20. 20.

    The indicator lparty in the Democracy & Dictatorship dataset (Cheibub et al. 2010) gives this information on a three-point scale. It is 0 if no legislature exists or all members of parliament are nonpartisan, 1 if only the regime party is represented in parliament, and 2 if multiple parties hold seats in parliament.

  21. 21.

    The indicator is a Herfindahl index and ranges between 0 and 1. As its value increases, so does the probability that two randomly drawn individuals come from different ethnolinguistic groups.

  22. 22.

    The analysis takes advantage of R (R Core Team 2016), brms (Bürkner 2017), and STAN (STAN Development Team 2016).

  23. 23.

    Assume the simplified regression model \(logit(Pr(Y = 1)) = \alpha _{0[j]} + \beta _{01} c + \beta _{02} v + \beta _{03} c v + \beta _{11} r + \beta _{12} c v r\). Here \(\alpha _{0[j]}\) is a short-hand notation for the varying intercept of authoritarian regime j, and \(\beta _{0\cdot }\) through \(\beta _{1\cdot }\) indicate regression coefficients at the regime-year respectively regime level of the model. Let c, v, and r represent ongoing campaigns, within-differences in violence, and between-differences in restrictions. The interaction terms in question are \(\beta _{03} c v\) and \(\beta _{12} c v r\). The partial derivative with respect to v is \(\frac{\partial logit(Pr(Y = 1))}{\partial v} = \beta _{02} + \beta _{03} c + \beta _{12} c r\). Since c is binary, the equation simplifies to either \(\frac{\partial logit(Pr(Y = 1 | c = 0))}{\partial v} = \beta _{02}\) or \(\frac{\partial logit(Pr(Y = 1 | c = 1))}{\partial v} = \beta _{02} + \beta _{03} + \beta _{12} r\). This pair of equations exactly matches hypothesis H2.

  24. 24.

    The average predictive comparison of a focal predictor u is defined as (Gelman and Hill 2006, 466):

    $$\begin{aligned} B_u(u^{(lo)}, u^{(hi)})&= \frac{1}{n}\sum _{i=1}^{n} b_u(u^{(lo)}, u^{(hi)}, v_i, \theta )\\ b_u(u^{(lo)}, u^{(hi)}, v_i, \theta )&= \frac{E(y|u^{(hi)}, v, \theta ) - E(y|u^{(lo)}, v, \theta )}{u^{(hi)} - u^{(lo)}}. \end{aligned}$$

    It is a difference quotient comparable to a partial derivative. However, this procedure has two important advantages over partial derivatives. First, for any given change of the focal predictor u, i.e., for any \(u^{(hi)} - u^{(lo)}\), v includes the joint distribution of all other regression inputs. In effect, average predictive comparisons summarize how a defined change in u plays out under all empirically observed circumstances. In effect, average predictive comparisons account for the typically non-constant effect of u in generalized linear models. Second, \(\theta \) includes all estimated model parameters, i.e., the entire posterior distribution of the regression coefficients. Therefore, average predictive comparisons also average over the uncertainty in model estimates.

    Each credible interval about the average marginal effects in Fig. 5.2 covers 90% of the posterior distribution. Violence is fixed at 0 and 0.46, which approximates an increase by 1 standard deviation. The figures evaluate 18,000 random draws from the posterior in order to keep the computational burden manageable.

  25. 25.

    The average predictive comparisons fix violence at 0 and 0.46, which approximates an increase by 1 standard deviation. Both figures are based on 18,000 randomly sampled draws from the joint posterior. Rug plots show the distribution of between-differences in restrictions. Each credible interval covers 90% of the posterior distribution.

  26. 26.

    See Appendix 5.8.3 for the Model IV equivalent of Fig. 5.3.

  27. 27.

    This refutation translates into powerful evidence against the hypothesis because the prior lends it an unfair advantage over the data (see Gelman et al. 2013, 56).

  28. 28.

    Successful coup coalitions exercise political power for at least one month. Otherwise, the power grab counts as failed.

  29. 29.

    The NAVCO 2.0 data provide rich information on campaigns. However, that information does not easily lend itself to measuring resolve. Does it take more resolve to shower soldiers with flowers or to point rifles at them? Do secessionist demands require more resolve than calls for democracy? Do parallel institutions such as opposition media, schools, or even courts reflect higher resolve or do they merely correlate with demands for secession? Empirical answers to those questions require much theoretical elaboration on resolve.

  30. 30.

    Both plots report confidence intervals at the 95% level for these estimates.

References

Download references

Author information

Authors and Affiliations

  1. University of Potsdam, Potsdam, Brandenburg, Germany

    Dag Tanneberg

Authors
  1. Dag Tanneberg
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Dag Tanneberg .

5.8 Appendix

5.8 Appendix

5.1.1 5.8.1 Summary Statistics

5.1.2 5.8.2 Summary of Within- and Between-Differences

The upper row in Fig. 5.5 plots the bivariate distribution of within- and between-differences in violence and restrictions. The figure considers complete cases only. The left panel plots between-differences for 179 authoritarian regimes, the right panel shows within-differences for all corresponding 3,941 regime-years. Both scatter plots inform on the empirical picture of between- and within-differences in political repression. Together, these plots counter a central, returning objection to the separability of violence and restrictions.

Table 5.6 Summary statistics
Full size table
Fig. 5.5
figure 5

Distribution of within- and between-differences in repression

Full size image

Between- and within-differences in political repression might correlate so strongly that they blend into one. Consequently, separating violence from restrictions would be impossible. Both scatter plots in Fig. 5.5 refute that objection. They show positive correlations between short-term variations and long-term levels in political repression. Those correlations are statistically significant at the 95% confidence level.Footnote 30 However, at 0.3 respectively 0.23, these correlations are negligibly small. Therefore, the between- and within-differences in violence and restrictions are not mere scholastic terms. Rather, they reflect observed differences.

Table 5.7 Correlation of restrictions and violence given campaigns
Full size table

The information paradox in human rights measurement constitutes another returning concern. Accordingly, the least restrictive authoritarian regimes should be the most violent ones. Moreover, the information paradox implies that the variance in short-term fluctuations in violence should increase as a function of restrictions. The lower row of Fig. 5.5 refutes both implications of the information paradox. It plots the mean and standard deviation in within-differences of violence for every decile of between-differences in restrictions. Since neither the central tendency nor the spread of violence increases systematically from left to right, the information paradox finds scant support in the data.

Ongoing campaigns might confound the clear empirical separability of within- and between-differences in violence and restrictions. Therefore, Table 5.7 repeats the exercise for every combination of the five variables. It reports Pearson’s \(\rho \), its confidence intervals at the 95% level, and t-statistics. Empirical associations between both forms of political repression are present, but once again, they are negligibly small. In conclusion, restrictions can be separated from violence, no matter the level of analysis or the presence of campaigns.

5.1.3 5.8.3 Results for a Fully Specified Interaction Term

Figure 5.6 follows up on Model IV in Table 5.3. The upper row shows the conditional mean effect of violence. The lower row reports average predictive comparisons. Each credible interval covers 90% of the posterior distribution. The average predictive comparisons fix violence at 0 and 0.46, which approximates an increase by 1 standard deviation. The average predictive comparisons sample 18,000 posterior draws. Finally, rug plots show the distribution of between-differences in restrictions.

Fig. 5.6
figure 6

Results on the moderating effect of restrictions

Full size image

The figure supports the same substantive conclusions summarized earlier. Short-term increases in violence always destabilize authoritarian power-sharing, and this adverse effect of violence increases when campaigns challenge highly restrictive authoritarian regimes.

5.1.4 5.8.4 Results for an Alternative Coding of the Coup d’État

Table 5.8 reports posterior mean effects and standard errors for a Bayesian multilevel analysis with weakly informative priors. It replicates the preceding analysis, but exploits data provided by Marshall and Marshall (2016). The substantive conclusions drawn earlier are unaffected by the choice of data.

Table 5.8 Bayesian multilevel analysis with weakly informative priors
Full size table
Table 5.9 Fixed effects logistic regression results
Full size table
Table 5.10 Cross-classified Bayesian multilevel analysis
Full size table

5.1.5 5.8.5 Fixed Effects Estimation Results

5.1.6 5.8.6 Cross-Classified Bayesian Multilevel Analysis

Rights and permissions

Reprints and permissions

Copyright information

© 2020 Springer Nature Switzerland AG

About this chapter

Check for updates. Verify currency and authenticity via CrossMark

Cite this chapter

Tanneberg, D. (2020). Does Repression of Campaigns Trigger Coups d’État?. In: The Politics of Repression Under Authoritarian Rule. Contributions to Political Science. Springer, Cham. https://doi.org/10.1007/978-3-030-35477-0_5

Download citation

Publish with us

Policies and ethics

Search

Navigation