2.1. The effect of sanctions on democracy and human rights

Table 1 summarizes canonical quantitative studies on the domestic consequences of economic sanctions and plots the timeframes they examine. This is further discussed in Appendix A.2. Most country-year studies show a general worsening of democracy and human rights in sanctioned countries, while a small handful show improvements. However, expert assessments of democracy- and human rights-related sanctions threats and impositions often paint a more optimistic picture, with success rates well above 50% (see Appendices A.5 and A.6 on threats and impositions).

Table 1. Selected past research findings on the effects of sanctions on democracy and human rights

‘Range’ column: Dotted vertical line indicates 2005 as a rough demarcation of the increased use of targeted sanctions. ‘Effect’ column: ‘+’ indicates an improvement; ‘=’ indicates no change; ‘–’ indicates a deterioration.

The majority of research – and therefore much of the conventional wisdom – mainly covers the years of the Cold War era and 1990s, mostly running through 2000/2005 due to the coverage of the influential HSE and TIES datasets (see 2.3). Therefore, studies focusing on the years after 1990 and especially 2005 onwards will be instructive for probing contemporary sanctions practice and effectiveness, and whether the move toward more sophisticated ‘targeted’ sanctions has delivered on its intentions. Given the significant investment, high political will, and wide-ranging scholarly input in this reform process (see 2.2), it stands to reason that these innovations have indeed resulted in better sanctioning practice. Greater precision in sanctions tools entails greater pressure on targets, which – in the logic of the 1990s/2000s reforms – should lead to greater effectiveness and less unintended democracy- and human rights-related fallout. Though past work finds negative effects via various pathways (see Table 1 and Section 6), I test sanctions effectiveness ‘naively’ as a matter of policy evaluation (Athey and Imbens, Reference Athey and Imbens2017; Liou et al., Reference Liou, Murdie and Peksen2020: 809, 816). I expect that the major policy shifts surrounding targeted sanctions since the 1990s have had the desired effect of bringing about positive changes in sanctioned countries with respect to democracy and human rights.

2.2. Changes to sanctions policy and world politics

The drastic humanitarian consequences of the comprehensive UN trade embargo on Iraq under Saddam Hussein in the 1990s prompted UN, US, and EU decisionmakers to shift their sanctions policy from broad blockades to sanctions targeted at political elites (e.g., asset freezes) or particular economic sectors (e.g., oil) (Biersteker et al., Reference Biersteker, Tourinho, Eckert, Tourinho, Eckert and Biersteker2016: 25–27; Early and Schulzke, Reference Early and Schulzke2018: 61–62).Footnote ³ Major policy consultations included the “Interlaken Process” on financial sanctions (1998–99/2001), the “Bonn–Berlin Process” on arms, aviation, and travel sanctions (1999–2000/01), and the “Stockholm Process” on sanctions enforcement (2001–02/03) (Biersteker, Reference Biersteker, Eckert and Tourinho2016: 3). Giumelli dates the change to ‘targeted sanctions’ at the UN to the mid-1990s (Giumelli, Reference Giumelli2015: 1352), Hawkins and Lloyd find that the normative shift had gained “substantial support” by the early 2000s (Hawkins and Lloyd, Reference Hawkins and Lloyd2003: 441, 451–452), and Drezner summarizes that the UN and the US had fully “internalized” the notion of targeted sanctions by 2010 (Drezner, Reference Drezner2011: 99–101). Thus, if we roughly date the shift toward targeted sanctions to around 2005, the late 2000s, 2010s, and 2020s are critical for assessing the effects of these changes. If they were successful, we would expect to see positive (or at least less negative) domestic outcomes from the mid-2000s onwards.

Panel A in Figure 1 illustrates the increasing precision of UN, EU, and US economic sanctions since the 1990s and the mid-2000s as the starting point for targeted sanctions (Attia and Grauvogel, Reference Attia and Grauvogel2023). Panel B shows an increase in post-Cold War sanctioning, a slight decrease around the late 1990s reconsideration toward targeted sanctions, and a slow increase that has taken up speed since 2015 (Felbermayr et al., Reference Felbermayr, Kirilakha, Syropoulos, Yalcin and Yotov2020; Attia and Grauvogel, Reference Attia and Grauvogel2023). While only around 25% of sanctioned countries in the early to mid-1990s were subject to asset freezes and travel bans against political elites – two particularly prominent types of targeted sanctions – this has risen to about 60% in the 2020s. Conversely, one quarter of all sanctioned countries were sanctioned comprehensively in the early 1990s, but this has halved to 12%, and these are not full blockades, but rather highly restrictive unilateral sanctions.Footnote ⁴

Alongside these changes in sanctions policy, broader shifts in world politics also make this re-examination worthwhile. For instance, the US and the EU have increasingly “weaponized interdependence,” using their central positions in global financial and informational networks for strategic leverage over other states (Farrell and Newman, Reference Farrell and Newman2019). This has changed the coercive capacity and reach of Western sanctions, potentially also affecting their consequences in sanctioned states.

Figure 1. Trends in Western sanctions, 1989–2021.

2.3. New data on sanctions, democracy, and human rights

Many of the canonical quantitative studies on sanctions cover the late Cold War era through 2005, often using the influential TIES dataset (1945–2005, Morgan et al., Reference Morgan, Bapat and Kobayashi2014, see Table 1 and Appendix A.2). Only recently have datasets with more up-to-date coverage been published: EUSANCT (1989–2015, Weber and Schneider, Reference Weber and Schneider2022), IST (1990–2018/2021, Attia and Grauvogel, Reference Attia and Grauvogel2023), and the GSDB (1950–2021, Felbermayr et al., Reference Felbermayr, Kirilakha, Syropoulos, Yalcin and Yotov2020). The former covers sanctions threats and impositions, the latter two cover imposed sanctions. This article uses IST for its main analyses and EUSANCT for supplementary analyses and illustrations. IST has good temporal coverage (Version 2.0; 1990–2021), and focuses on key sanctions senders and aims I am particularly interested in – the UN, US, and EU’s support of democracy and human rights abroad. In addition to coverage, IST’s detailed and transparent sourcing and coding make it the most reliable reference work on Western sanctions in particular.

Significant data improvements have also been made in the study of democracy and human rights through human-coded datasets (Weidmann, Reference Weidmann2024: 923–924). While the classic studies on sanctions largely use Polity IV or Freedom House data to measure democracy and autocracy Freedom House (Freedom House, 2022), V-Dem now provides more detailed measures (Boese, Reference Boese2019). Similarly, sanctions studies most commonly used CIRI or PTS human rights data (Cingranelli and Richards, Reference Cingranelli and Richards2010; Wood and Gibney, Reference Wood and Gibney2010), which can now be compared to Human Rights Scores (HRS) as a “theoretically informed measurement model” with greater coverage (Fariss, Reference Fariss2014: 297, 301).

2.4. New tools for causal inference with country-year data in IR

The imposition of economic sanctions is a clear and easily identifiable policy intervention. Therefore, it can be usefully examined with causal inference methods designed for policy evaluation (Athey and Imbens, Reference Athey and Imbens2017). Given the country-year data at hand, matching and weighting, combined with difference-in-differences (DiD) estimation, are a useful, applicable methodological approach that is here implemented via PanelMatch (Imai et al., Reference Imai, Kim and Wang2023). This ‘selection on observables’ identification strategy draws upon past work on the treatment assignment mechanism of economic sanctions, i.e., the factors determining sanctions imposition (von Soest and Wahman, Reference von Soest and Wahman2015b; Licht, Reference Licht2017). Although a recognized challenge, potential endogeneity and confounding are often not explicitly tackled in sanctions research, as convincingly argued by Licht (Reference Licht2017: 162) and by Gutmann and colleagues (Reference Gutmann, Neuenkirch and Neumeier2020: 160).

For the case of sanctions research, matching has attractive properties compared to country-year regression because it explicitly focuses on and makes claims about the Average Treatment Effect on the Treated (ATT); the effect of sanctions on countries that typically get sanctioned (see Samii, Reference Samii2016). That said, selection on observables remains a weaker identification strategy than designs building upon random or quasi-random treatment assignment (Keele, Reference Keele2015; Doleac, Reference Doleac2019). Within the predominant country-year framework for sanctions research, such designs remain largely out of reach (see Demena et al., Reference Demena, Reta, Jativa, Kimararungu, van Bergeijk and van Bergeijk2021). Nonetheless, matching and weighting are useful alternatives to the prevailing approaches (see Licht, Reference Licht2017; Gutmann et al., Reference Gutmann, Neuenkirch and Neumeier2020). This study therefore presents a faithful application of Imai, Kim, and Wang (Reference Imai, Kim and Wang2023) to the question of economic sanctions aiming to improve democracy and human rights.

2.5. Framework: replication, extension

Examining the causal effects of sanctions on domestic conditions in the sanctioned country is a highly policy relevant research question, and comparatively straightforward conceptually. Nonetheless, there are many plausible ways of answering the same empirical research question, even when using identical data (Silberzahn and Uhlmann, Reference Silberzahn and Uhlmann2015; Breznau et al., Reference Breznau, Rinke, Wuttke and Nguyen2022). Considering the “researcher degrees of freedom” and “garden of forking paths” inherent in social inquiry (Gelman and Loken, Reference Gelman and Loken2013), this article closely builds on and replicates previous work on the subject (seeGleditsch and Janz, Reference Gleditsch and Janz2016). It uses and extends one article in particular, von Soest and Wahman (Reference von Soest and Wahman2015b). Their article examines the ‘treatment assignment mechanism’ for democracy- and human rights-related sanctions and is therefore a well-suited building block for estimating subsequent causal effects.Footnote ⁵

3.1. Outcomes: democracy and human rights measures

This study’s main research interest is whether Western sanctions cause an improvement of human rights and democracy in the countries they target. Appendix A.3 presents summary statistics for these and all other variables used in this article.

The state of democracy is here defined with a narrow focus on its electoral dimension (‘polyarchy’), understood as a “minimalist” definition (Coppedge et al., Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman, Bernhard, Fish, Glynn, Hicken and Lührmann2020: 42–44; Przeworski, Reference Przeworski1991; Dahl, Reference Dahl1998; Kurki, Reference Kurki2010). Electoral democracy reflects “the core value of making rulers responsive to citizens, achieved through electoral competition” under fair conditions, free civil society participation, and sound electoral organization. Democracy-related sanctions are often threatened and imposed following anti-democratic coups to return to constitutional order or after fraudulent elections to bring about fairly-run ones. An emphasis on this electoral dimension examines the shorter-term changes and improvements sanctions senders hope to achieve, as opposed to slower-moving societal aspects of democracy such as egalitarianism and liberal values (Coppedge et al., Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman, Bernhard, Fish, Glynn, Hicken and Lührmann2020: 45; Lindberg et al., Reference Lindberg, Coppedge, Gerring and Teorell2014: 158, 160–161). Although a significant innovation in the measurement of democracy, V-Dem is not widely used in sanctions research.Footnote ⁶

Human rights are here narrowly defined as the degree to which government agents violate or uphold the physical integrity rights of their population. This covers the absence or presence of extreme repressive practices such as arbitrary imprisonment, torture, and execution (Fariss, Reference Fariss2014: 297; see also Poe and Tate, Reference Poe and Tate1994; Richards et al., Reference Richards, Webb and Clay2015). This definition emphasizes the violation or safeguarding of immediate physical safety, reflecting the conditions that human rights-related sanctions most commonly intend to improve. Narrow definitions of this sort focusing on physical integrity rights are commonly used in cross-sectional work on human rights conditions (Cordell et al., Reference Cordell, Clay, Fariss, Wood and Wright2022: 6). It is usually argued that physical integrity rights are the foundation of all other human rights, and measures of both therefore closely correlate (Cope et al., Reference Cope, Crabtree and Lupu2018).

3.2. Treatment: democracy- and human rights-related sanctions imposition

The treatment is the imposition of democracy- or human rights-related sanctions, sometimes called “democratic sanctions” (Cox and Drury, Reference Cox and Drury2006; von Soest and Wahman, Reference von Soest and Wahman2015b: 18; von Soest and Wahman, Reference von Soest and Wahman2015a). The relevant literature often bundles these two objectives of sanctions because they are conceptually related and often imposed simultaneously. Appendix A.4 lays out how these sanctions goals coincide.Footnote ⁷

The binary treatment measure is whether a given country was sanctioned by the UN, US, and/or the EU in a given year. Among sanctions datasets, IST (International Sanctions Termination) is best suited for studying the research questions of this paper (Portela and Charron, Reference Portela and Charron2022; Attia and Grauvogel, Reference Attia and Grauvogel2023): it covers sanctions from 1990 to 2021 and provides detailed information on sanctions objectives. As we are interested in Western sanctions policy surrounding democracy and human rights, the sender coverage is also appropriate. This simplification of sanctions imposition as a binary variable – despite the fact that sanctions in fact vary in intensity and ‘treatment dosage’ – should be revisited in future work.Footnote ⁸

This study focuses on sanctions imposed and supported by Western states (see also von Soest and Wahman, Reference von Soest and Wahman2015b; Pospieszna and Weber, Reference Pospieszna and Weber2020). Western states are the most active sanctioners, data coverage for Western senders is plausibly complete,Footnote ⁹ and sanctions goals and mechanisms are fairly similar across Western states, making generalizations justifiable. Studies of this type typically only cover sanctions threatened and imposed by the US and/or the EU/EEC (Appendix A.2). In contrast, this study adds UN sanctions to the analysis. The common exclusion of UN sanctions from ‘Western’ sanctions overlooks that UN sanctions require the support of three key Western states as members of the Security Council: the US, UK, and France (the “P3”). Indeed, given the considerable normative power of UNSC resolutions and these sanctions’ binding nature in accordance with the UN Charter, the P3 often push for multilateral UN measures while also imposing their own. Therefore, I argue that UN sanctions are by definition “Western-supported” sanctions, and should therefore be included in these types of analyses.Footnote ¹⁰

Given this definition, Appendix A.7 examines treatment variation from 1990 to 2021. Overall, ‘democratic sanctions’ onset is somewhat rare (see also Beiser-McGrath, Reference Beiser-McGrath2022: 432–435): The timeframe contains 86 cases of onset, or about 3 instances per year on average. 68 of 181 country units (38%) had Western democracy- or human rights-related sanctions imposed on them at some point in the examined timeframe. There are no strong temporal clusters. Therefore, there is enough variation in treatment across time and units, strengthening the internal and external validity of the analysis (Imai et al., Reference Imai, Kim and Wang2023: 591–592).

Sanctions threats are often overlooked in the sanctions literature, although their implicit or explicit threat already significantly affects government behavior (Drezner, Reference Drezner2003). This concern is less relevant for the issue at hand, as explicit threats of democracy- and human rights-related sanctions only rarely result in target acquiescence. This is likely because concessions on democracy represent a major risk to an autocrat’s hold on power (von Soest and Wahman, Reference von Soest and Wahman2015a: 961–962). Appendix A.5 illustrates this for threats over the 1989–2015 timeframe.Footnote ¹¹ Therefore, this paper focuses on imposed sanctions only.

3.3. Estimand: ATT

The estimand is the average treatment effect on the treated (ATT): the average effect of democracy- and human rights-related sanctions onset on democracy and human rights in sanctioned countries.

The assumption for estimating the ATT is $Y^0 {\text{$\perp\mkern-10mu\perp$}}D | X$ – i.e., that the distribution of outcomes for untreated units is the same for both (matched and weighted) control units and (counterfactually) treated units. For the case of sanctions, this assumption means that, once appropriately matched and weighted, human rights and democracy would on average have developed similarly in sanctioned countries as they did in unsanctioned countries if the former had not been sanctioned. In other words, in the absence of sanctions, the trajectories of sanctioned countries would have developed in parallel to those of unsanctioned countries similar to them on key observable characteristics. This latter group’s observed outcomes are thus used to interpolate the former’s potential outcomes (see Morgan and Winship, Reference Morgan and Winship2015; Cunningham, Reference Cunningham2023: 37–76; Blackwell and Glynn, Reference Blackwell and Glynn2018: 1068–1069). Following Imai, Kim, and Wang (PanelMatch), the ATT is estimated as follows (Reference Imai, Kim and Wang2023: 592–593), separately for democracy and human rights:

(1)\begin{equation} \begin{aligned} \delta(F, L) = \mathbb{E}[& Y_{i, t+F}(X_{it} = 1, X_{i, t-1} = 0, [X_{i, t-\ell}]_{\ell=2}^L) - \\ & Y_{i, t+F}(X_{it} = 0, X_{i, t-1} = 0, [X_{i, t-\ell}]_{\ell=2}^L) | X_{it} = 1, X_{i, t-1} = 0]. \end{aligned} \end{equation}

The estimated ATT is the observed difference $\delta$ in outcomes between treated and untreated units $X_{it} = 1$ and $X_{it} = 0$, after both had previously been untreated $X_{i, t-1} = 0$. The ATT can be estimated for different specified timeframes. F and L represent leads and lags, respectively, i.e., the timeframe after treatment for which the ATT is estimated (t+0, t+1, t+2…) and the past timeframes based on which matches are created (…t-2, t-1). In Equation (1), the potential outcome of treated units is shown in line one and that of untreated units is shown in the first part of line two (Imai et al., Reference Imai, Kim and Wang2023: 593). This follows common potential outcomes notation (Morgan and Winship, Reference Morgan and Winship2015). This framework can be readily applied to study and re-examine the effects of economic sanctions on democracy and human rights in targeted countries.Footnote ¹²

4.1. Trigger events

Coups – Successful coups often cause violence and disorder. Anti-democratic coups in particular are frequently met with international condemnation, including sanctions imposition (von Soest and Wahman, Reference von Soest and Wahman2015b). Coup data is drawn from the recent update of the Colpus dataset (Chin et al., Reference Chin, Carter and Wright2021; Chin and Kirkpatrick, Reference Chin and Kirkpatrick2023).

Controversial elections – Sanctions are also frequently imposed when Western monitors raise fraud allegations, are denied participation, or refuse to monitor an election (von Soest and Wahman, Reference von Soest and Wahman2015b: 23). This data is taken from V-Dem and NELDA (Hyde and Marinov, Reference Hyde and Marinov2012; Coppedge et al., Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Marquardt, Medzihorsky, Pemstein, Fox, Gastaldi and Pernes2024).

These two critical ‘trigger events’ variables are generally not lagged in the relevant literature because empirically they almost always precede sanctions imposition, ruling out reverse causality (von Soest and Wahman, Reference von Soest and Wahman2015b: 23, fn8).

4.2. Target vulnerability and sender–target relations

Beyond immediate, drastic trigger events, the target’s vulnerability to external coercion and its extant links to Western states also affect the likelihood of Western sanctions imposition (von Soest and Wahman, Reference von Soest and Wahman2015b).

Target vulnerability and instability – Pro-democracy protests may be interpreted by Western senders as target vulnerability, and may encourage them to impose targeted sanctions to support protesters (Grauvogel et al., Reference Grauvogel, Licht and von Soest2017). This information is drawn from V-Dem’s indicators for civil society repression and pro-democracy protest (Coppedge et al., Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Marquardt, Medzihorsky, Pemstein, Fox, Gastaldi and Pernes2024). Meanwhile, economic weakness – e.g., stagnating growth or rising inflation – may imply a vulnerable target and encourage imposition. I use World Bank data on GDP, GDP per capita, economic growth, and inflation (Arel-Bundock, Reference Arel-Bundock2022).

Intrastate conflict – An increase in intrastate conflict may also make the imposition of human rights-related sanctions and arms embargoes more likely, while causing a deterioration in the outcomes of interest. This is measured by whether there is an increase in the number of ongoing intrastate conflicts in a given year (UCDP/PRIO, 2024).

Neo-patrimonialism – The turn toward ‘targeted’ sanctions took up the insight that certain regime types may be more pliable with sanctions than others – namely, personalist systems highly dependent on ‘buying off’ domestic elites and rivals. Neo-patrimonial states offer clear openings for targeted sanctions and asset freezes (Escribà-Folch and Wright, Reference Escribà-Folch and Wright2010; Peksen, Reference Peksen2019b). This is operationalized with V-Dem’s ‘neo-patrimonialism’ variable (Coppedge et al., Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Marquardt, Medzihorsky, Pemstein, Fox, Gastaldi and Pernes2024).

Political proximity – Finally, UN General Assembly voting proximity to the US and EU members (Bailey et al., Reference Bailey, Strezhnev and Voeten2017) also affects Western sanctions imposition (von Soest and Wahman, Reference von Soest and Wahman2015b). Potential sanctions senders consider proximity and vulnerability strategically: sanctions are most likely against allied states when those allies are stable, and against adversaries when those adversaries are unstable. Conversely, sanctions are less likely against politically unstable allies and against stable adversaries (McLean and Radtke, Reference McLean and Radtke2018).

4.3. Predicting ‘democratic sanctions’ imposition

Von Soest and Wahman’s original results could be successfully reproduced moving from their original Stata code to R. The results also hold when replicated for the 1990–2021 timeframe with the newer datasets summarized above. Appendix A.10 presents these replications.Footnote ¹³ While coups are quite regularly met with sanctions imposition, fraudulent elections are today sanctioned more rarely than they once were. Nonetheless, both remain an important predictor of sanctions onset (Appendix A.9).

Based on von Soest and Wahman’s original approach and taking up the variables outlined in Sections 3.1, 3.2, 4.1, and 4.2, Table 2 shows three rare events logit models predicting the onset of democracy- and human rights-related sanctions by the UN, EU, and US from 1990 to 2021. Model 1 contains a larger range of potentially relevant factors derived from the literature. Model 2 drops the variables found to be statistically insignificant to predicting sanctions onset. The effects of the remaining variables are largely retained. Finally, Model 3 combines the two ‘trigger events’ variables. These trigger events are rare individually (Appendix A.9) – e.g., there are only a small handful of successful coups in any given year – making exact annual matching difficult. Combining coups and fraudulent elections assumes that they function similarly as “triggers” for sanctions imposition (von Soest and Wahman, Reference von Soest and Wahman2015b). Model 3 shows that this trigger events count performs similarly to the individual variables. Appendix A.11 considers whether trade might be a relevant confounder, which it does not appear to be.

Table 2. Predicting the imposition of democracy- and human rights-related sanctions, 1990–2021

Note: Cubic polynomials not displayed. Robust SEs. Models are Firth’s bias-reduced logistic regressions. $\dagger$ $p \lt 0.1$, * $p \lt 0.05$, ** $p \lt 0.01$, *** $p \lt 0.001$.

The upshot of this theoretical and empirical examination of the treatment assignment mechanism is that four main characteristics determine the imposition and non-imposition of UN, EU, and US sanctions intended to improve democracy and human rights:

1. (1) major trigger events (coups, fraudulent elections),

2. (2) pro-democracy protests in the target country,

3. (3) political proximity between sender and target, and

4. (4) increases in intrastate war.

This supports previous findings and adds further insights on how senders might consider domestic opposition and protest (von Soest and Wahman, Reference von Soest and Wahman2015b; Grauvogel et al., Reference Grauvogel, Licht and von Soest2017). These four confounders will therefore be used to match and weight sanctioned and unsanctioned country-years in the main analysis.

5.1. Matching

Having mapped all observations and their treatment variation (see above and Appendix A.7), treated and untreated countries in a given year t are matched. Treated observations – countries with ‘democratic sanctions’ onset in a given year – are matched with countries in the same year that have an identical treatment history for a specified lag period $L$, but were not sanctioned at t. Treatment history is a potential confounder for the present analysis (see also Blackwell and Glynn, Reference Blackwell and Glynn2018: 1069): governments previously sanctioned may react to new sanctions imposition differently than those who have not experienced them in the recent past. This includes learning effects regarding issues such as sanctions evasion, ‘sanctions-proofing’ the domestic economy, and strengthening relations with ‘black knight’ actors (Cilizoglu and Bapat, Reference Cilizoglu and Bapat2020). Matching exactly on the time period also alleviates concerns around the effects of world events such as the end of the Cold War or the 2007–09 recession going unrecognized (Imai et al., Reference Imai, Kim and Wang2023: 594–595). Within these “matched sets,” the single sanctioned country acts as the treated unit and all remaining unsanctioned countries as the control group. Applying this procedure with a lag time of $L = 4$ and a lead time of $F = 3$ provides a total of $n = 60$ matched sets from which estimates are derived (out of a total of $N = 86$ cases of ‘democratic sanctions’ onset in the examined timeframe). Appendix A.12 lists the 60 observations included in the main analysis and the 26 that are dropped.Footnote ¹⁴ The estimates are robust to different lead and lag specifications.

5.2. Weighting and balance

As matches are simply made on treatment history, the matched sets contain many control units very different from the treated unit. The next step therefore weights the observations in accordance with theoretical expectations and past empirical work regarding confounders (see Section 4). The aim is to refine the matched sets and construct very similar treatment and control groups. From a methodological and epistemological perspective, this is similar to Mill’s Most Similar Systems Design (Plümper et al., Reference Plümper, Troeger and Neumayer2019: 5, 15–16). It is also preferable to the often idiosyncratic “effective samples” produced by regular time-series regression because it is theory-driven and explicit in its weighting criteria (Samii, Reference Samii2016).

The confounders on which control units will be weighted are drawn from the above analysis of the treatment assignment mechanism (Section 4). Unsanctioned countries that are very similar to sanctioned countries on the four key factors identified above are assigned greater weight: domestic trigger events, pro-democracy protest, political proximity to the EU and US, and civil conflict. Appendix A.13 illustrates the matching and weighting procedure for an example case, the 2015 imposition of EU and US sanctions on Burundi following a power grab by incumbent president Pierre Nkurunziza and subsequent violence against anti-government protesters. It shows that the procedure acts as we expect it to, assigning the most weight to countries similar on the four identified predictors of sanctions imposition (e.g., Ethiopia, Nigeria, Azerbaijan) and very little weight to very different countries (e.g., Norway, Georgia, Iceland).

Summarizing all matched sets contained in the analysis, Table 3 displays the balance between treatment and control groups on the four main confounders, using covariate balancing propensity score weighting (CBPS). The corresponding model is fitted to all observations in the treated and control groups in the entire timeframe, producing robust estimates (Imai et al., Reference Imai, Kim and Wang2023: 595–596).

Table 3. Weighted/refined/balanced control groups: Covariate balance from $t-4$ to $t+0$, both outcomes

Trigger events and conflict increases used for weighting at $t-4$ to $t+0$, all others at $t-4$ to $t-1$. Standardized mean differences, in SDs.

The variables for trigger events (coups and controversial elections) and civil war onset are balanced and weighted from $t_{-4}$ through $t_{0}$ (i.e., not through $t_{-1}$). As von Soest and Wahman show, these events virtually always occur before sanctions imposition (hence ‘trigger events’), and their measurement at $t_{0}$ is therefore generally not post-treatment in practice (Reference von Soest and Wahman2015b: 23 n8). All other variables are examined from $t_{-4}$ through $t_{-1}$ to rule out post-treatment bias. As shown in Table 3, after weighting, sufficient balance between treated and control groups is achieved, using a threshold of $ \lt |0.25|$, $ \lt |0.2|$, or $|0.1|$ standardized mean differences for all variables (Rubin, Reference Rubin2001: 177; Stuart, Reference Stuart2010: 11, 15; Linden and Samuels, Reference Linden and Samuels2013: 969; Greifer, Reference Greifer2021; Imai et al., Reference Imai, Kim and Wang2023).

5.3. Difference-in-differences estimation of the causal effect

Finally, the ATT is estimated using the above matches, weights, and parameters using DiD estimation. DiD estimation is intended to account for unobserved, time-invariant confounding factors and differences between the treatment and control groups. Figures 2 and 3 illustrate the estimated results for the two outcomes of interest. Parallel trends can also be inspected in these plots. The effect of treatment on outcome – of ‘democratic sanctions’ imposition on democracy and human rights – is estimated from $t$ through $t_{+3}$. The pre-treatment trends are estimated with a placebo test.

Figure 2. Estimated effect of ‘democratic sanctions’ on democracy, 1990–2021 (95% CIs).

Figure 3. Estimated effect of ‘democratic sanctions’ on human rights, 1990–2021 (95% CIs).

Footnote ¹⁵

First, for electoral democracy, democratic sanctions imposition is estimated to have a negative contemporaneous ( $t_{0}$) causal effect of about $-0.06$ [95% CIs: $-$0.09; $-$0.03] on a 0–1 scale, further dipping to $-$0.08 [ $-$0.13; $-$0.04] at $t_{+1}$ and $-$0.07 [ $-$0.10; $-$0.03] in $t_{+2}$ (Figure 2). This suggests a notable negative effect of sanctions on democracy in the targeted country. Put into contemporary terms for 2022, these differences roughly amount to those between Norway and the UK (0.899 and 0.843), Ghana and Sri Lanka (0.633 and 0.575), Uzbekistan and Vietnam (0.221 and 0.157), or the year before and after Viktor Orbán was elected president of Hungary (2009: 0.854, 2010: 0.805).Footnote ¹⁶

Second, the imposition of democratic sanctions is also estimated to have a negative contemporaneous effect ( $t_{0}$) on human rights (Figure 3), at $-$0.26 [ $-$0.41; $-$0.12] on a $-$3.5–5.5 scale. The estimates remain negative in $t_{+1}$ ( $-$0.22 [ $-$0.39; $-$0.05]) and $t_{+2}$ ( $-$0.19 [ $-$0.36; $-$0.06]), before the confidence interval cross zero in $t_{+3}$ ( $-$0.18 [ $-$0.37; 0.02]). For 2021, a difference in the range of $-$0.26 roughly amounts to the differences between Germany and Taiwan (3.61 and 3.39), Romania and Hungary (1.60 and 1.40), or China and Mali ( $-$1.71 and $-$1.93). However, pre-treatment trends are not entirely parallel, so these results are less reliable than those for democracy. This uncertainty compared to the results for democracy may result from how the outcome measures are constructed, ultimately being based on aggregated human coding. While coups and fraudulent elections are clearly observable events, deteriorating human rights conditions are often more subtle. This makes them more difficult to measure and subject to greater uncertainty.Footnote ¹⁷

Substantively, these negative estimates are more similar to the “corrosive” effects Peksen and Drury find for sanctions from 1972 to 2000 (Reference Peksen and Drury2010: 255–256) than the positive effects Wahman and von Soest find for democracy-related sanctions from 1990 to 2010 (von Soest and Wahman, Reference von Soest and Wahman2015a). Even when “signing the bias” (Bueno de Mesquita and Fowler, Reference Bueno de Mesquita and Fowler2021: 176–179), the effects are normatively concerning: While the treated group still experiences slightly more trigger events despite the weighting procedure and therefore the effects may be slightly overestimated, better balance would likely not fully remove the negative effects. That said, it is again worth highlighting the effect estimated here. This analysis examines a certain type of sanctioned country: states faced with ‘democratic sanctions,’ but not so extreme and long-term as to have been sanctioned for virtually the entire post-Cold War era. Appendix A.12 discusses this sample and type of country.

5.4. Robustness checks and sub-sample analyses

I test these main results in supplementary analyses. First, I conduct placebo tests to probe the mechanism driving these negative effects. This uses states sanctioned for reasons unrelated to democracy and human rights as the treatment group (e.g., sanctions imposed due to corruption, drug trafficking, weapons proliferation, etc.). Appendix A.14 describes and presents these tests. In these exploratory models, there seems to be no strong effect on democracy and human rights outcomes. Thus, the trajectory I find among states facing ‘democratic sanctions’ can likely be attributed to these ‘democratic sanctions’ and not to any underlying similarities between countries facing sanctions onset in general. In other words, there seems to be something particular about ‘democratic sanctions’ that causes negative effects on democracy and human rights. This cannot be examined in greater detail given the sample sizes under examination here, but may be attributed to the type of domestic pressure ‘democratic sanctions’ entail (Early and Peksen, Reference Early and Peksen2022: 4–5). ‘Democratic sanctions’ often play out under precarious domestic circumstances, with autocrats clinging to power, civil society aiming to weaken that hold on power, and governments in turn cracking down on these dissidents. ‘Democratic sanctions’ pose a greater threat to an autocratic regime’s survival than other types of sanctions do (see Escribà-Folch and Wright, Reference Escribà-Folch and Wright2010; Grauvogel and von Soest, Reference Grauvogel and von Soest2014; Grauvogel et al., Reference Grauvogel, Licht and von Soest2017; Licht, Reference Licht2017; Liou et al., Reference Liou, Murdie and Peksen2020; Mei, Reference Mei2024, on mechanisms) As discussed, this greater risk also explains why autocrats only rarely ‘fold’ in the threat stage of democratic sanctions (von Soest and Wahman, Reference von Soest and Wahman2015a: 961–962; Appendix A.5).Footnote ¹⁸

Second, I ran a replication of the main analysis using temporal sub-samples splitting the 1990–2021 timeframe into pre- and post-2005 eras (1990–2004 and 2005–2021, respectively). Appendix A.15 presents these analyses. This takes up the turn from indiscriminate sanctions to more targeted measures dating to around 2005 (see Section 2.2). The results suggest that, if anything, the negative effects of ‘democratic sanctions’ may have worsened. Although targeted sanctions may have decreased broader humanitarian fallout as intended by the reforms of the 2000s, autocrats may nowadays react more harshly to ‘democratic sanctions’ than they once did, with democracy and human rights deteriorating as a result (e.g., Liou et al., Reference Liou, Murdie and Peksen2020; Mei, Reference Mei2024). While it is not possible to closely probe the reasons for this dynamic here, it is nonetheless normatively concerning that the negative effects of this type of sanctions do not seem to have markedly improved.