Pandemic backsliding: Violations of democratic standards during Covid-19

The widespread adoption of emergency powers during Covid-19 raises important questions about what constitutes a (un)democratic response to crises. While the institutions and practices of democracy during normal times are well established, democratic standards during emergencies have yet to be conceptualized in the literature. This makes it difficult to systematically answer questions like - How do states' responses to Covid-19 violate democratic standards? Do such violations make states' responses more effective? Drawing on international treaties, norms, and academic scholarship, we propose a novel conceptualization of democratic standards for emergency measures. We then identify which government responses to Covid-19 qualify as a violation of democratic standards within the framework of illiberal and authoritarian practices, introducing a dataset covering 144 countries from March 2020 onward. In this article, we provide an overview of the extent to which states violated democratic standards in their response to Covid-19 during 2020. We find no relationship between violations of democratic standards and reported Covid-19 mortality. Illiberal and authoritarian practices in response to the Covid-19 pandemic do not correlate with better public health outcomes. Rather, such crisis-driven violations should be carefully observed as they could signal autocratization.


Introduction
Nearly every country in the world adopted emergency measures in response to the Covid-19 pandemic. While proportional responses are a necessary part of crisis management, excessive measures and their abusive enforcement have raised concerns that Covid-19 could "infect" democracy itself. Indeed, the pandemic hit the world during a period of global democratic decline (Lührmann and Lindberg, 2019;Maerz et al., 2020). In places where democracy was already eroding -like the Philippines and Hungary -leaders have excessively expanded executive and policing powers under the pretext of protecting human lives (Gregorio, 2020;Varga, 2020). While such extreme cases provide anecdotal evidence of undemocratic emergency responses, we know very little about the general extent to which countries have used Covid-19 to justify undemocratic behavior or the effects of these actions.
What are democratic standards for emergency measures? How have states violated them during the Covid-19 pandemic? And how do these undemocratic behaviors relate to public health outcomes? In this article, we address these three inter-related questions. First, we develop a novel framework for assessing violations of democratic standards for emergency measures, which is based on international treaties, norms, and academic scholarship. Then we use this framework to create a new dataset measuring the extent to which 144 governments violated democratic standards in their response to Covid-19 between March and December 2020. Finally, we use these data to investigate the relationship between democratic violations and Covid-19 outcomes.
As a result, this article makes three core contributions. First, we develop a novel framework of democratic standards for emergency measures by drawing on the International Covenant on Civil and Political Rights (ICCPR) and its interpretation by human rights experts and academic scholars. We apply this framework to assess how states violated democratic standards in their response to the Covid-19 pandemic from March to December 2020. This framework could be applied to other emergency situations, including epidemics, natural disasters, and perhaps even man-made events like economic crises or war. Thus, we engage with the scholarship on international norms for human rights during emergency situations (e.g. Criddle and Fox-Decent, 2012;Franck, 2010;Linz, 1978;Rossiter, 2002), as well as democracy's benefits (e.g. Bollyky et al., 2019;Wang et al., 2019).
Second, we show that illiberal and authoritarian practices, even if temporary, are prevalent during times of crises. Our descriptive findings demonstrate that levels of democracy prior to the pandemic provide an incomplete picture of how states have reacted to . As one would expect, autocracies tend to commit the most violations; however, democracies are far from immune. Since some democracies engage in violations, this complicates analyses of how the political context shapes Covid-19 outcomes like mortality rates (e.g. Bosancianu et al., 2020;Cassan and Van Steenvoort, 2021;Cepaluni et al., 2020). Instead, our focus on practices complements existing approaches by specifically addressing how states respond to emergencies rather than how they behave during "normal" times. Finally, our analysis shows that the pretense of expedience often used to justify undemocratic responses holds little traction when examined empirically. We find little evidence of a systematic association between violations of democratic norms and lowered Covid-19 deaths or cases. This suggests that as countries grappled with incomplete information and difficult choices about how to appropriately respond to a novel virus, those that chose the path of authoritarianism have fared no better than those adhering to democratic norms. This suggests that when leaders present a trade-off between democratic norms and public safety, their rhetoric may only serve to facilitate executive aggrandizement (Petrov, 2020). As such, pandemic-related violations of democratic standards should be closely monitored, otherwise crisis-driven responses could erode democratic norms and lead to long-term backsliding.
The article is structured as follows: After a brief literature review, we conceptualize seven ways of how democratic standards are violated during crises. We then introduce our dataset and analyze the relationship between violations of democratic standards and Covid-19 mortality rates. Subsequently, we discuss our findings and conclude with an outlook on future research.

International human rights law and emergency situations
What does it mean to respect democratic standards during an emergency? As highlighted by United Nations Experts (2020), government responses must be "proportionate, necessary and nondiscriminatory". This builds on international human rights law -most prominently the ICCPR (United Nations, 2020), which has been ratified by 173 states. Article 4 of the ICCPR states that countries may derogate from some rights if necessary during a "time of public emergency which threatens the life of the nation" (United Nations, 2020). A derogation is the suspension (Hafner-Burton et al., 2011), deviation (Richards andClay, 2012), or elimination (McGoldrick, 2004) of a legal obligationin this case commitments to honor political and civil rights.
Thus, international human rights law allows states to partially suspend democratic institutions, rights, and proceedings, albeit only within certain boundaries. Article 4 of the ICCPR places clear limitations on such derogations, including that they must be proportionate ("strictly required by the exigencies of the situation") and non-discriminatory. It also specifies that certain rights cannot be derogated under any circumstance. These include rights such as the freedom from torture and slavery. Furthermore, states must adhere to a formal procedure: the emergency situation must be formally declared and the UN Secretary General needs to be notified of any derogation (Ellena and Shein, 2020;Human Rights Committee, 2001;McGoldrick, 2004;United Nations, 2020).
While the norms for emergency measures are well-established through the ICCPR and other legal instruments, states do not always strictly adhere to these principles in practice. Instead, governments frequently violate both derogable and non-derogable rights during states of emergencies (Richards and Clay, 2012). This is because political pressures in the face of a crisis "generate strong incentives for states to violate their human rights obligations (Criddle and Fox-Decent, 2012, p.46)." Faced with higher judicial constraints, executives in democracies are more likely to derogate from human rights treaties through formal channels because this allows them to "buy time and legal breathing space (Hafner-Burton et al., 2011, p. 673)." This pattern, for example, led Ackerman (2004) to warn against a "destructive cycle" of repressive laws in the wake of the 9/11 attacks. That said, some scholars find that democracies do not substantially increase human rights violations during states of emergency (Neumayer, 2013). By contrast, autocracies tend to produce longer states of emergency (Hafner-Burton et al., 2011), with significantly more human rights violations (Neumayer, 2013).

Current data and literature on responses to Covid-19
The Covid-19 pandemic has generated substantial scholarly interest about how governments respond to crises. This has prompted several data collection efforts. The policy-focused dataset by Cheng et al. (2020) provides information on the level of government intervention, the geographical areas and groups targeted, and the timing of specific policy responses. Similarly, Hale et al. (2020) collect information on containment and closure, economic, and healthcare responses. The Center for Civil and Political Rights (CCPR) collects data on states of emergency and whether countries notified the UN (CCPR, 2020). The International Institute for Democracy and Electoral Assistance (IDEA) monitors Covid-19's impact on democracy and human rights, including elections, checks on the executive, fundamental rights, participation, and impartial administration (IDEA, 2020). Similarly, the International Center for Not-for-profit Law (ICNL) collects data on how government responses to Covid-19 affect civic freedoms and human rights (Page and Ognenovska, 2020; see also Leng and Lemahieu, 2020).
A rapidly growing literature uses these data to investigate the effects of Covid-19 on social and political life. Sebhatu et al. (2020) and Cheibub et al. (2020) find that more democratic countries reacted slower and were more hesitant to initiate lockdowns and school closures. Other studies address the effects of Covid-19 responses on civic spaces (Bethke and Wolff, 2020), social movements (Pleyers, 2020), and public support for the government (Bol et al., 2020). Charron et al. (2020) find that polarization and populism explain why some European regions saw higher excess mortality than others during the early stages of the pandemic. In Africa, Carlitz (2021) shows that more stringent mobility restrictions are effective at reducing Covid-19 infections. Lastly, a growing strand of literature inquires about the relationship between type of political regime and Covid-19 deaths. While early studies suggest better healthcare performance in democracies (Cepaluni et al., 2020), more recent analyses show that death rates do not vary across regime types (Bosancianu et al., 2020;Cassan and Van Steenvoort, 2021).
Thus, the literature and data on government responses to Covid-19 offer valuable insights that inform our understanding of how states respond to crises. However, none of these projects directly measures violations of democratic standards in government responses to the Covid-19 pandemic. For example, the ICNL tracks legal instruments and government practices across fourteen issues related to human rights and civic freedoms (Page and Ognenovska, 2020). While this information is useful for qualitatively assessing certain types of violations or certain cases, it does not provide a way to systematically compare countries or make general assessments of international human rights compliance. Likewise, IDEA (2020) emphasizes democratic institutions and practices that we expect to see during normal times, rather than assessing whether practices during a crisis specifically violate democratic standards for emergency situations. Assessing these violations is crucial to understanding how states respond to global emergencies and whether international human rights obligations inform those decisions.

A practice-based framework for assessing violations of democratic standards for emergency measures
Our conceptualization of what it means to violate democratic standards during times of crises draws on the guiding principles of the ICCPR and its interpretation by human rights experts and academics. This provides a general framework to assess whether limits on human rights during emergency situations adhere to international norms. Based on this and to provide a conservative assessment, we speak of violations of democratic standards for emergency measures only if restrictions occur in an unambiguously disproportionate, non-necessary, or discriminatory way.
For example, while responses to Covid-19 may promote physical distancing by restricting freedom of movement and assembly, they may not infringe on non-derogable rights like the right to life or freedom from torture. For that reason, unlike some recent studies (Cheibub et al., 2020;Sebhatu et al., 2020), we do not consider the limitation of movement or assembly rights (e.g. lockdowns, school closures, and workplace closures) as violations of democratic standards, unless these measures are considered indefinite, disproportionate, discriminatory, or abusively enforced. We also do not count the postponement of elections as a violation of democratic standards, as it is not always clear how elections and campaigns can be safely organized during a pandemic (Landman and Splendore, 2020).
We take a practice-based approach to study how states respond to emergencies (Schatzki, 2001). This deviates from the prevailing approach in comparative politics and international relations that measures qualities of the regime itself as democratic or autocratic (in degrees or as kinds). Instead, we assess the extent to which government practices deviate from democratic norms during a crisis. As illustrated in Fig. 1, we classify seven practices or types of violations using the framework of illiberal and authoritarian practices developed by Glasius (2018). Illiberal practices infringe on personal autonomy and dignity. Authoritarian practices sabotage accountability by disabling access to information and voice. The intersection between the two categories includes all practices that simultaneously undermine accountability and infringe on personal autonomy and dignity. As such, this framework can be used to track violations of democratic standards for emergency measures because any state can engage in illiberal and authoritarian practices, regardless of the regime type.

Type 1 (discriminatory measures)
The ICCPR (Art. 4) explicitly states that emergency measures should not "discriminate solely on the ground of race, colour, sex, language, religion or social origin." Therefore, discriminatory measures are an illiberal practice that infringe on individual autonomy and dignity. Examples of discriminatory measures include Ugandan security forces disproportionately targeting the LGBT + community in their enforcement of Covid-19 restrictions and Sri Lanka's forced cremation of Covid-19 bodies in contravention of Muslim burial practices (Madrigal-Borloz et al., 2020;Shaheed et al., 2021).

Type 2 (derogations from non-derogable rights)
The ICCPR notes several rights that governments may never violate. These include: the right to life (Art. 6); freedom from torture and "cruel, inhuman or degrading treatment or punishment" (Art. 7); freedom from slavery (Art. 8); freedom of thought, conscience, and religion (Art. 18); and three ICCPR clauses pertaining to the rule of law 4 (Art. 11, Art., 15 and Art. 16). Suppression of these non-derogable rights is never considered legitimate, even during times of crisis (Criddle and Fox-Decent, 2012;McGoldrick, 2004). Therefore, we count any breach of these legal obligations as an illiberal practice. For example, the Philippines has violated Article 7 by engaging in torture and inhumane treatment, including public humiliation and forced confinement in dog cages and coffins (HRW 2021).

Type 3 (abusive enforcement)
Proportionate government responses require even-handed enforcement that avoids physical violence and police brutality (Callamard, 2020). Abusive enforcement infringes on autonomy and dignity and runs counter to the notion that states have an obligation to guarantee subjects "secure and equal freedom" (Criddle and Fox-Decent, 2012). For example, according to Amnesty International, Kenyan police were responsible for the deaths of at least 21 individuals while enforcing curfews from March to July 2020 (Reuters Staff, 2020). It is difficult to imagine any circumstance where such excessive force aligns with the principles of necessity and proportionality (Franck, 2010;McGoldrick, 2004), which is why we conceptualize abusive enforcement as a third illiberal practice.

Type 4 (no time limit on emergency measures)
The ICCPR (Art. 4) calls for emergency measures to be "strictly required by the exigencies of the situation". The UN Human Rights Committee in its General Comment on the ICCPR has noted that this requirement refers to the "duration, geographical coverage and material scope" of emergency measures (Human Rights Committee, 2001, emphasis ours). Thus, the comment clarifies that "[m]easures derogating from the provisions of the Covenant must be of an exceptional and temporary nature" (Human Rights Committee, 2001, emphasis ours). The inclusion of an explicit end date in the legal document that initiates the emergency measures is the most effective way of ensuring that they do not persist beyond the necessary period (Ackerman, 2004;Cormacain, 2020). If emergency measures are arbitrarily and unnecessarily prolonged, they disable voice and thereby sabotage accountability. As Linz (1978, p. 62) argues, emergency measures are "hard to distinguish from other types of autocratic rule when [they] last beyond a well-defined situation." Therefore, we conceptualize the failure to include an end date as an authoritarian practice. In December 2020, 43 countries had emergency measures without a time limit in place.

Type 5 (disproportionate limitations on the role of the legislature)
Emergencies are often considered the "hour of the executive" (Stelzenmueller, 2020). However, according to the Office of the High Commissioner for Human Rights, emergency measures "should be subject to periodic and independent review by the legislature" and emergency legislation requires "adequate legislative scrutiny" (OHCHR, 2020). During Covid-19, national legislatures have demonstrated their capability to perform effectively (Democracy Reporting International, 2020), through innovations like remote meetings and remote voting (e. g. Brazil, New Zealand, Romania), lower quorum requirements (e.g. Sweden), and meeting outdoors to ensure social distancing (e.g. Uganda) (Murphy, 2020). Therefore, we consider vaguely formulated laws that allow the executive to rule broadly by emergency decree an authoritarian practice (Ackerman, 2004;Bar-Siman-Tov, 2020;Petrov, 2020). As previously mentioned, Hungary's new emergency legislation would fall into this category; as would more extreme cases like Zambia where the legislature was adjourned sine die for several months in 2020 due to Covid-19.

Type 6 (official disinformation campaigns)
We consider the dissemination of gravely misleading and false information by governments on key facts about Covid-19 -what it is, how it is transmitted, and how it can be treated or cured -as an authoritarian practice. The deliberative principle of democracy implies that political leaders maintain a reason-based and respectful dialogue with the public (Coppedge et al., 2011;Habermas, 1984;Lührmann et al., 2019). Spreading false information on Covid-19 runs counter to this principle. While some (minor) misreporting may be unintentional, broader disinformation campaigns by government officials systematically hinder public access to fact-based information and therefore constitute an authoritarian act of sabotaging accountability. This includes spreading false information about Covid-19 preventatives or treatments, such as the promotion of hydroxychloroquine by former U.S. President Donald Trump, as well as cases of denialism, such as the late President Magufuli of Tanzania.

Type 7 (restrictions on media freedoms)
The media performs a watch-dog function through investigating and reporting about government actions, which may push governments to prioritize public health (Barnes et al., 2008). Furthermore, the press disseminates important information about the virus to the broader population (Chan et al., 2018;Habersaat et al., 2020). A free media counters disinformation (e.g. US or Brazil), whereas in countries with heavily restricted media (e.g. Turkmenistan) official disinformation is left uncontested. Therefore, we count limitations of media freedoms as a violation of democratic standards that infringes on autonomy and dignity, hinders access to information, and disables voice. As illustrated in Fig. 1, restrictions on the media are a combination of illiberal and authoritarian practices. While some governments justify limits on the media with the intent to stop the spread of disinformation, such provisions need to be specific and narrowly formulated to avoid abusive enforcement (OHCHR 2020).

Tracking violations of democratic standards during Covid-19
To track violations of democratic standards during Covid-19, we develop a set of 15 indicators that are then aggregated into seven ordinal indices capturing the severity of the seven types of violations outlined above. Six of the ordinal indices range from 0 (least severe) to 3 (most severe), while the index for Type 4 violations (time limit) ranges from 0 to 2. As shown in Table 1, we measure most types of violations based on one or two indicators. Tracking media limitations requires a more disaggregated approach, resulting in seven separate indicators for this type of violation. From these seven indices, we also construct an overall index capturing the extent and severity of democratic violations during Covid-19. The Pandemic Violations of Democratic Standards (PanDem) Index is the sum of the seven violation indices divided by the maximum possible score. It ranges from 0 (no violations) to 1 (maximum violations).
A team of trained research assistants hardcoded each of the 15 indicators, drawing on a list of reliable government, non-governmental, and media sources. We compiled a set of guidelines with detailed instructions for the coders and worded each indicator to minimize the role of subjective judgement. For most questions, coders were instructed to consider only measures introduced in response to Covid-19; cases where the authorities did not invoke Covid-19 to impose policies that violate democratic standards are not coded as pandemic violations. For instance, derogations from non-derogable rights are only coded if they are Covid-19 related. The only exception concerns some of the media indicators, where during a pilot study we realized that it was impossible to distinguish generalized restrictions on freedom of information from those specifically adopted in response to Covid-19. Therefore, we asked coders to consider the general media freedom environment in their coding decisions.
To ensure transparency, coders provided a written justification and complete record of sources for each data point. Our data including the full codebook, coding justifications, and sources are available online (Edgell et al., 2020). During the first wave of data collection, we recruited country and regional experts to cross-check the answers for 37% percent of countries. The authors of this article cross-checked most cases where country and regional experts were unavailable. In subsequent rounds, a trained supervisor validated all coding and sources by the research assistants, in consultation with the authors.
The resulting dataset measures the 15 indicators, seven indices, and the PanDem Index in three waves roughly corresponding to financial quarters: 11 March to June 30, 2020 (Q2), 01 July to 30 September (Q3), and 01 October to 10 December (Q4). We treat country-quarter as the unit of analysis even though the dates for Q2 and Q4 do not perfectly match the second and fourth quarters. The data covers 144 countries, almost all those with populations over 2 million.

Empirical overview of violations from March to December 2020
Fig. 2 summarizes the extent of violations during the Covid-19 pandemic across the world. It presents, for each country, the maximum value of the PanDem Index attained between 11 March and December 10, 2020. This summary measure reveals that violations of democratic standards in response to Covid-19 are widespread. Only 14 countries register no violations during this period, while 95 of the 144 countries scored at or above the average of 0.20 (out of 1.0), meaning about 20% of the maximum possible extent of violations. As reported in Fig. 3, 66 countries (46%) saw no change in their scores between Q2 and Q4 2020. Another 55 countries (38%) saw a decrease in violations, while 14 countries (10%) had increasing violations over this time span.
We observe violations in dictatorships and democracies alike, with a high degree of heterogeneity within and across regime types. As one might expect, the PanDem Index exhibits a moderate negative correlation with V-Dem's Liberal Democracy Index (LDI) in 2019 (Coppedge et al., 2021), with a rank-ordered correlation coefficient (Spearman's Rho) of − 0.55. Fig. 4 plots this relationship by fitting a curve through the observed values for the LDI in 2019 and the PanDem Index scores for each country-quarter. According to marginal effects estimates, the negative association holds for cases at or above 0.25 on the LDI (illustrated by the dashed line). Yet, a great deal of heterogeneity persists above this threshold, as shown by the plotted observed values, and as Fig. A1 in the Appendix demonstrates, the distributions in PanDem scores for democracies and autocracies overlap, despite the lower median for democracies (based on Lührmann et al., 2018). In short, both democracies and autocracies have engaged in violations, but on average autocracies have done so more. Fig. 5 illustrates patterns in the types of violations observed within the data. Amongst illiberal practices, we observe discriminatory measures in 24 countries and derogations from non-derogable rights in just 11 cases. For instance, Bulgaria's emergency measures discriminated against Roma communities (Amnesty International, 2020), and the government in El Salvador violated the non-derogable right to no conviction for a crime which was not a crime at the time of commitment (Article 15; see Human Rights Watch, 2020) when they detained hundreds of people for Covid-19-related infractions, sometimes before announcing the restrictions and despite a supreme court ruling that such detentions were illegal. Abusive enforcement is also uncommon, being recorded as "widespread" in only 6 countries (Argentina, Kenya, Nigeria, South Africa, Uganda, and Venezuela).
By contrast, authoritarian practices are more frequent. Fifty-one countries imposed emergency measures without specifying a time limit, spanning most geopolitical regions and including democracies and autocracies (e.g. Bosnia and Herzegovina, Mexico, and Russia). In 29 countries, the legislature's ability to hold the executive accountable is severely undermined. For instance, Zambia indefinitely adjourned its parliament in July 2020 (Jere, 2020) and only resumed parliamentary activities in mid-September 2020. In other cases, like Hungary and Ghana, new legislation provides the executive with broad powers to rule by decree, which could be interpreted beyond actions directly related to the pandemic due to vague formulation. Finally, government disinformation campaigns appear in 39 cases -including those spearheaded by presidents like former President Trump in United States, President Bolsonaro in Brazil, as well as President Berdimuhamedov in Turkmenistan and the late President Magufuli in Tanzania.
Restrictions on the media are the most common type of violation of democratic standards. Ninety countries register major violations by  restricting the flow of information or engaging in harassment of journalists. For example, in Belarus a journalist was detained for publishing an article that questioned official Covid-19 statistics (Reporters Without Borders, 2020a). And in Tanzania, the government shut down several news outlets, jailed journalists, and denied access to vital information about the pandemic to both the media and the WHO (Committee to Protect Journalists, 2020).
Violations do not strongly correlate with each other (see in the Appendix, Table A1), suggesting that governments have engaged in a diverse set of illiberal and authoritarian practices during Covid-19. The Pearson-r correlation coefficients are smaller than 0.30 in absolute value across all types (the largest correlation is between media restrictions and official disinformation campaigns). A Kaiser, Meyer and Olkin (KMO) test gives an overall measure of sampling adequacy of 0.56, which is not much above the conventional threshold for factor analysis. This lowcorrelation suggests heterogeneity in states' practices, providing empirical support for our choice of a seven-fold conceptualization of violations.
Overall, the PanDem data provide several descriptive insights about how countries violate democratic standards for emergency measures during Covid-19. Most countries, including several democracies, have engaged in some form of violation. Yet there is also a high degree of heterogeneity in the extent and form of violations. Moreover, our data suggest that authoritarian practices designed to sabotage accountability through indeterminate states of exception, limitations on the legislature, government disinformation, and restricted access to information are more common during Covid-19 than illiberal practices that undermine human dignity through discrimination, derogations from non-derogable rights, and abusive enforcement.

Violations of democratic standards and public health outcomes during Covid-19
During the Covid-19 pandemic, leaders justified illiberal and  authoritarian practices as necessary for curbing the spread of the virus. Do such violations of democratic standards make states' responses more effective? In this section, we explore whether this is the case based on data from March to December 2020. Our main dependent variable is the (logged) number of reported Covid-19-related deaths per million individuals (from Bosancianu et al., 2021). The independent variable of interest is the extent to which democratic standards are violated during a quarter (i.e. PanDem Index and its component types). As summarized below, we use a lagged response model to account for the autoregressive nature of reported Covid-19 death rates: where ln(Deaths/Million) i,t is the (logged) death rate per million for country i during quarter t, Pandem i,t is country i's PanDem score during quarter t, X i is a vector of country-specific controls, δ t are quarterspecific intercepts, and u t is the error term. Data for the dependent variable are available for all four quarters of 2020, thus including the lagged dependent variable does not reduce the sample size. We estimate country-clustered robust standard errors and include quarter-specific fixed effects to account for contemporaneous shocks. We also report results using country and quarter fixed effects (without a lagged dependent variable) to account for unobserved country-level confounders. Because of the short time-series (T = 3), the country fixedeffects model reflects the estimated within-country association based on 77 countries (56% of the sample) where changes on the PanDem index are observed. Thus, estimates from the country fixed-effects regressions should be interpreted with caution.
To avoid potential confirmation bias from self-selecting the model specifications, we include a set of relevant control variables identified by another team of researchers through a Lasso procedure (Bosancianu et al., 2020;. Specifically, we control for the state's healthcare system capacity (health expenditures per capita and an index of health data quality) and overall health of the population (share of the population over 65 years, respiratory disease prevalence, and life expectancy), which are the most likely confounders for Covid-19 mortality.
We also control for level of democracy in 2019 using the LDI (Coppedge et al., 2021) to ensure our models capture correlations independent of existing democratic conditions prior to the pandemic. A squared term addresses previous findings that hybrid regimes often exhibit different relationships to outcomes as compared to regimes approximating the autocratic and democratic ideals (Dragu and Lupu, 2018;Knutsen and Nygård, 2015). Our results are also robust when we drop this variable from model (see Table A2).
We present our main results in Table 2. Model 1 assesses the overall relationship between democratic violations, as captured by the PanDem Index, and reported Covid-19 deaths. The coefficient is negative but far from significance levels (p = 0.524). As shown in Model 2, this null Coefficients and country-clustered robust standard errors. Models 1 and 3-5 use a lagged response model with quarter fixed effects. Model 2 uses a country-and quarter-fixed effects model. The outcome variable is reported Covid-19 deaths per million (logged) observed within the financial quarter, from Q2 to Q4. *p < 0.1, **p < 0.05, ***p < 0.01.
finding holds when we estimate a country-fixed effects model. This finding provides suggestive evidence that illiberal and authoritarian practices do not play a significant role in reducing reported Covid-19 deaths, contrary to claims made by actors engaging in such practices. Models 3-5 test how individual types of violations correlate with Covid-19 deaths. For three of the seven types of violationsabusive enforcement, discriminatory measures, and media limitationswe find a significant relationship. Abusive enforcement is associated with significantly higher reported Covid-19 deaths within the quarter, particularly, during Q3 of 2020 (see Fig. A4). This correlation could suggest that violent repression to enforce emergency measures exacerbates Covid-19 fatalities, perhaps through a backlash effect whereby violence encourages people to violate restrictions even more, thus increasing transmission of the virus. However, one alternative interpretation is that states are more likely to resort to repression when noncompliance with Covid-19 restrictions is high. If this were the case, the relationship between abusive enforcement and Covid-19 deaths would be spurious and the correlation should vanish after controlling for population compliance with Covid-19 directives. Unfortunately, we lack sufficient data on population compliance with sufficient cross-national coverage to address this issue.
In addition, we observe a significant negative relationship for discriminatory measures and media limitations. Discriminatory measures are relatively uncommonwith only 14 countries committing moderate or major violations within our sample (n = 36). When we reestimate the models removing one country with moderate or major violations at a time, we find no evidence of any single influential cases (see Fig. A10). However, this finding is relatively fragile; the statistical significance for discriminatory measures drops below conventional thresholds in all other robustness tests (see Appendix) and appears to be driven primarily by observations in Q2 of 2020 (see Fig. A4).
Meanwhile, we suspect that the negative correlation between reported Covid-19 deaths and media limitations is driven by a curtailment of freedoms that allow decentralized and accurate reporting of Covid-19 deaths. Such limitations allow governments to monopolize access to information about Covid-19 and under-report mortality rates. This can have reputational benefits, making their response appear more effective domestically and internationally. For example, while the WHO initially praised China's ability to contain the virus (He et al., 2020;Rauhala, 2020), there is growing evidence that Chinese officials underreported Covid-19 deaths. At the very extreme, however, this strategy can backfire, as cases like Tanzania and Turkmenistan illustrate. In both countries severe limits on press freedom allowed the government to report few or no Covid-19 deaths for long periods, resulting in negative international attention (BBC, 2021;Yaylymova, 2020).
Excess mortality statistics offer a potential alternative to official death rates (Weinberger et al., 2020), despite some methodological limitations (Aron et al., 2020;Beaney et al., 2020). Excess mortality is estimated as the number of deaths per million minus the average deaths per million for the previous 5 years. Alternatively, p-scores are calculated as the percentage change in deaths over the previous 5-year average. Oxford's Our World in Data provides these numbers for 63 countries within our sample (Aron et al., 2020;Giattino et al., 2021aGiattino et al., , 2021b. Models using these data tend to over-represent Europe, Central Asia, and the Americas while underrepresenting countries in the MENA, Asia-Pacific and omitting sub-Saharan Africa altogether (Fig. A5). In the Appendix, we report additional diagnostics (Table A7 and Fig. A6) and re-estimate our main models with deaths per million logged using this smaller sample (Table A6).
As illustrated in Fig. 6, our results generally hold if we use excess mortality as the outcome despite the data limitations (also see Table A4). The PanDem index is not significantly correlated with excess deaths per million (logged). However, we continue to see a positive correlation for abusive enforcement and a negative correlation for media limitations. We also observe a significant correlation between the PanDem Index and derogations from non-derogable rights, although this result should be taken with caution as it is driven by just 5 countries committing this type of violation. Limitations on the legislature also has a weak negative correlation with excess deaths, but only at the p < 0.1 level, and therefore, the relationship cannot be considered robust. In the appendix, we report additional results for p-scores (see Table A5), which exhibit a weak statistical correlation with PanDem at the 0.1 level. Further, Fig. 6 shows no significant correlation between reported Covid-19 cases per million (logged) and the PanDem index (also see Table A8). Our prior observations regarding abusive enforcement and media limitations remain when using this outcome measure.
In the Appendix, we report additional robustness checks. First, states with higher death rates might simply respond with greater violations of democratic standards. However, as shown in Fig. A3, we see little evidence of such reverse causality. Deaths reported in the previous quarter are not significantly associated with scores on the PanDem Index and show no significant relationship with five of the seven types of violations. Similar results hold when accounting for cumulative reported deaths. Only media limitations show some indication of reverse causality, giving some support for the above discussion about the effects of information manipulation on reported Covid-19 deaths. Second, we also check for whether the results are heterogeneous across quarters. As shown in Fig. A4, greater violations correlate with significantly higher Fig. 6. Comparing results for logged deaths per million, logged excess deaths per million (p-scores), and logged cases per million as the dependent variable. Estimated coefficients and 90% confidence intervals from lagged response models. See Table 2, A4, A8 for full results. PanDem coefficients come from Model 1 and coefficients for disaggregated types come from Model 4 of these tables. death rates in Q3 but with significantly lower death rates in Q4. These opposed trends result in the overall null estimates found in the main results. Finally, the main findings are also robust when we re-estimate the models without the LDI as a control variable (Table A2) and when estimating a simple cross-sectional pooled OLS model with reported Covid-19 deaths as of December 28, 2020 and the maximum scores on PanDem and each violation type (Table A3).
The short time-series allows us to draw only partial conclusions. The results provide suggestive evidence against the claim that authoritarian tactics have helped to curb the pandemic. Proponents of these tactics tend to point to low mortality rates in a few carefully chosen countries. Yet, such simple correlations fail to hold once we systematically account for average trends across a large sample of cases. In short, although these models cannot demonstrate a null causal effect, they challenge the common argument advanced by proponents of authoritarian tactics who argue that there exists a clear correlation between such tactics and public health.

Conclusion
This article provides a new approach for studying how states respond to emergencies like the Covid-19 pandemic. We contribute to the rapidly growing body of literature on pandemic-related government policies by offering a novel conceptualization of democratic standards for emergency measures using a practice-based approach. This allows us to measure the extent and severity of violations committed in 144 countries from March to December 2020.
The PanDem dataset shows that most countries have engaged in at least some violations of democratic standards since the beginning of the pandemic. While more common in autocracies, such violations are also prevalent in democratic regimes. We also find that a great deal of heterogeneity in violations, with authoritarian practices being more common than illiberal ones and the overlapping practice of infringements on the media being the most common.
Our findings also suggest that we should be skeptical of claims that democratic standards must be violated to protect the population from the virus. Empirically, we do not find any systematic correlation between the severity of violations and reported Covid-19 deaths. We should be especially vigilant when such claims of prioritizing life over democracy are made by leaders in already autocratizing countries. In weak democracies and hybrid regimes, violations of democratic standards could be harbingers of autocratization, as leaders take advantage of the pandemic's exceptional context to consolidate power, sideline opposition, and silence critics.
There are important limitations to testing causal arguments from observational data, and therefore the findings of this article should be taken as suggestive rather than definitive. Moreover, official death rates may not reflect true mortality rates from Covid-19. These numbers might be biased downwards in countries that lack sufficient monitoring or testing capacity or that actively manipulate information. Leaders seeking to downplay the risks of the virus or to appear more capable at containing it may manipulate official death statistics. This would suggest a negative correlation between reported deaths and access to information. And indeed, we find some support for this correlation in our models.
We see several avenues for future research. Variations within single types of violations such as official disinformation campaigns could be further explored with in-depth analyses. For example, some governments such as Madagascar (World Health Organisation, 2020) have "only" disseminated disinformation on treatments whereas regimes such as Turkmenistan (Reporters Without Borders, 2020b) fully deny the existence of Covid-19 in their country. Analyzing public communication on the Covid-19 pandemic more generally -beyond disinformation campaigns -would provide insights into how different governments justify violations of democratic standards. Apart from this, we still know very little about which socio-economic, political, and structural factors drive high death rates. So far, the Covid-19 pandemic appears to be non-discriminatory, spreading to countries regardless of level of economic development or democracy. While this article mainly focuses on violations of democratic standards and included several socio-political aspects as control variables, the latter need to be extended and further tested.

Declaration of competing interest
None.