Legislative INGO Restrictions and Their Intended Effects

The worldwide proliferation of INGOs during the 1980s and 1990s occurred in a particularly favourable political environment. During this period, the rise of a pro-NGO norm among state donors of aid and intergovernmental organizations (IGOs) actively promoted the spread of NGOs to non-Western countries (Reimann Reference Reimann2006). However, an increasing number of governments came to perceive the activities of NGOs and especially INGOs as misaligned with their own goals and priorities. One of the reasons for this is that INGOs play a key role in what Keck and Sikkink (Reference Keck and Sikkink1998) describe as a boomerang dynamic, where local groups collaborate with international activists to persuade powerful states and IGOs to put pressure on governments in such areas as human rights and environmental protection. INGOs regularly use their ability to gather information locally to highlight gaps between international commitments and the actual behaviour of governments, and to shame them vis-à-vis international and domestic audiences (Peterson, Murdie and Asal Reference Peterson, Murdie and Asal2018; Risse et al., Reference Risse, Ropp and Sikkink1999). Moreover, transnationally connected non-state groups are sometimes perceived as a direct threat to regime survival, especially after the ‘colour revolutions’ in post-Soviet countries in the 2000s (Gilbert and Mohseni Reference Gilbert and Mohseni2018).

In principle, governments that aim to reduce INGOs' ability to pursue undesired activities could exclusively use informal means. These include intimidation and extrajudicial detention of INGO staff and activists, as well as the threat and use of violence against people and property. While this option typically remains available to governments, an increasing number of them have chosen to limit the operation and activities of INGOs using formal legislation. Legal restrictions have several advantages over informal repression (Chaudhry Reference Chaudhry2022). First, governments may hope that the mere passing of laws will have a chilling effect, leading NGOs to self-censor without the need for active enforcement measures. Second, the adoption of a formal law turns members of the judiciary from potential prosecutors of abuses into helpers of the executive in the application of restrictions. Third, informal violence can decrease the domestic public legitimacy of government action (Lupu and Wallace Reference Lupu and Wallace2019). Fourth, restrictions can be justified by invoking the general right of sovereign countries to regulate private activity to defend public interests (Poppe and Wolff Reference Poppe and Wolff2017). This right can sometimes be presented as an international obligation, as in the case of restrictions on financial transactions that are justified as measures to prevent the financing of terrorism (Koo and Murdie Reference Koo and Murdie2022; Njoku Reference Njoku2022). Asserting that restrictions are in line with international norms and rules can protect governments from the punishments that powerful states and IGOs may impose for informal repression.

These potential benefits have led many governments to adopt laws constraining NGOs and specifically their ability to establish and maintain transnational connections (Christensen and Weinstein Reference Christensen and Weinstein2013; Glasius, Schalk and De Lange Reference Glasius, Schalk and De Lange2020). Domestic political conditions play an important role in triggering such measures. Restrictions are more likely when there is a gap between what states committed themselves to under international human rights law and their actual human rights practices (Bakke, Mitchell and Smidt Reference Bakke, Mitchell and Smidt2020). They are also more likely when authoritarian leaders face competitive elections (Dupuy, Ron and Prakash Reference Dupuy, Ron and Prakash2016). What other governments do matters as well: states learn from other states and are more likely to adopt restrictive laws when peers adopt the same type of laws, especially if they are in the same world region or are members of illiberal international organizations (Adolph and Prakash Reference Adolph and Prakash2022; Bromley, Schofer and Longhofer Reference Bromley, Schofer and Longhofer2020; Gilbert and Mohseni Reference Gilbert and Mohseni2018; Glasius, Schalk and De Lange Reference Glasius, Schalk and De Lange2020).

There is evidence that INGO restrictions have the intended effect. Dupuy, Ron and Prakash (Reference Dupuy, Ron and Prakash2015) find that most local human rights NGOs disappeared from Ethiopia after the 2009 Charities and Societies Proclamation. In Bangladesh and Zambia, restrictions made cross-border NGO collaborations on human rights and other causes increasingly fragile and their advocacy more tempered (Fransen et al. Reference Fransen2021). Based on a large cross-national sample, Smidt et al. (Reference Smidt2021) find that if governments impose multiple types of restrictions, the ability of domestic NGOs to monitor local human rights abuses declines, which, in turn, reduces international naming-and-shaming campaigns that depend on information from those local NGOs. They also find that multiple restrictions are associated with a decline in the number of human rights INGOs and anti-government protests.

Unwanted Side Effects of Legislative INGO Restrictions

While restrictions seem to have the desired effect of hampering criticism of a government's human rights practices, they are also likely to have effects that the government did not intend and does not want. We argue that this is, to a significant extent, due to a distinct drawback of legal instruments compared to informal intimidation and repression: it is more difficult to target them with a high degree of precision against those organizations that the government perceives as hostile. The application of laws can be selective, but the laws themselves need to be formulated in sufficiently general terms and be applicable to a relatively broad class of organizations. Legal theorists often identify ‘generality’ as a constitutive feature of law (Fuller Reference Fuller1964; Hart Reference Hart1961). Experimental evidence from different world regions shows that this understanding is shared by the general public (Hannikainen et al. Reference Hannikainen2021). Hence, the benefits described in the previous section would be unlikely to materialize if the laws defined their addressees narrowly. Political leaders recognize this and draft new legislation accordingly. Based on their extensive compilation of NGO laws, Dupuy, Ron and Prakash (Reference Dupuy, Ron and Prakash2016, 300) note: ‘[w]ith very few exceptions, these laws do not impose restrictions on specific categories of NGOs, such as human rights organisations or health groups’. In the few cases where the laws single out specific types of organizations, the criterion for special treatment is involvement in ‘political’ activities (Christensen and Weinstein Reference Christensen and Weinstein2013). This is an (often intentionally) fuzzy criterion and can be stretched to include some activities of organizations that would not regard themselves as political. To be sure, once the laws are passed, executive actors and their allies in the judiciary can interpret and apply them in a very targeted way. However, the general nature of legislation means that it is likely to affect a broader range of organizations and (potential) individual members than the set of intended targets, for two reasons.

First, some requirements imposed on NGOs, such as complex, time-consuming and costly registration and reporting requirements, must be fulfilled regardless of their specific mission. Indeed, the authorities might not know what issues an organization is concerned with (and thus how threatening it may become) until it obtains the relevant information through the registration process. For instance, the NGO law adopted in Russia in 2006 made registration and annual reporting more cumbersome for the entire non-profit sector and not only those NGOs directly involved in ‘political’ activities (Skokova, Pape and Krasnopolskaya Reference Skokova, Pape and Krasnopolskaya2018, 541).

Second, restrictive laws typically leave significant uncertainty as to what is expected of organizations and whether they risk coming under the scrutiny of state agencies. A notable example is the ‘foreign agent’ law adopted in Russia in 2012. Before its scope was extended in 2019, the law applied to NGOs receiving funds from abroad while engaging in ‘political activity’. Based on an extensive set of interviews with a diverse range of NGO staff in Russia, Skokova, Pape and Krasnopolskaya (Reference Skokova, Pape and Krasnopolskaya2018, 555, references omitted) document the vast impact of the law:

Given the administrative burden imposed by restrictive legislation and the possibility of intrusive checks, restrictive legislation can potentially affect the vitality of INGOs in sectors beyond those that governments are keen to control. One of these sectors is healthcare, which will be the focus of the rest of our discussion.

Governments do not usually have health INGOs in their sights when they change legislation, unless they work in areas such as reproductive and LGBTQ+ health. Occasionally, governments deny health INGOs permission to remain in a country, as Pakistan's government did with several INGOs in 2017 (Stacey and Bokhari Reference Stacey and Bokhari2017). However, on the whole, governments do not aim to reduce the number and activities of INGOs working in the field of health (Toepler et al. Reference Toepler2020). Hence, some INGOs did not withdraw from Ethiopia after the legislative change of 2009, but simply removed ‘rights’ from their re-registration applications and continued to work on health-related issues (Dupuy, Ron and Prakash Reference Dupuy, Ron and Prakash2015, 435). However, even if health-focused INGOs may be less affected than those promoting civil rights, they too are likely to suffer from the higher administrative hurdles, constraints and uncertainty that are imposed on the non-governmental sector as a whole. The hampering of INGO activities can occur due to active government intervention, as in the case of Russian authorities classifying certain public health organizations as ‘foreign agents’ against their will (Brechenmacher Reference Brechenmacher2017, 24). However, health INGOs can also be affected more indirectly, for instance, by delays in government approvals when officials are particularly cautious. For example, during the political turmoil in Egypt following the fall of President Mubarak's regime in 2011, ‘nongovernmental groups struggled to register with the Ministry of Social Solidarity … and groups waited for months to get the green light even for relatively uncontroversial health and education projects’ (Brechenmacher Reference Brechenmacher2017, 42). Increased regulation also exposes health INGOs to uncertainty about how authorities interpret specific legislative provisions. For instance, Ethiopia's Charities and Societies Proclamation prohibits any organization from spending more than 30 per cent of their budget on administrative overheads. Government authorities adopted a broad definition of such overheads, which meant that ‘health organisations providing mobile outreach services, trainings for health extensions workers, and clinical mentorship suddenly had to classify all of their core activities as administrative expenses’ (Brechenmacher Reference Brechenmacher2017, 79), which substantially limited the geographic scope and substantive range of their activities.

Restrictions on INGOs can have adverse consequences for human health in countries where they are implemented. Such negative effects can materialize because health INGOs impact health outcomes in at least four ways. First, INGOs diffuse medical knowledge and best practices among health professionals across borders. By providing their members with information on clinical protocols, management practices and other aspects of healthcare, and by stimulating debate on the relative merits of different treatments, INGOs can improve the quality of healthcare provided locally. For instance, the International Council on Women's Health Issues has held twenty-one international conferences since 1983, which spread knowledge on the health and well-being of women, and facilitated the establishment of training programmes in low-income countries (Noerager Stern Reference Noerager Stern2003). Numerous INGOs contributed to producing the Integrated Management of Childhood Illnesses (IMCI) framework and provided a forum to share best practices (Dalglish et al. Reference Dalglish2015). Building on the IMCI framework, a collaboration between the US and Armenian Red Cross societies succeeded in increasing breastfeeding, maternal knowledge of child illness signs and physician-attended deliveries in rural Armenia (Thompson and Harutyunyan Reference Thompson and Harutyunyan2009).

Second, INGOs can help government policymakers identify and implement more effective health policies. For instance, the International Confederation of Midwives played a major role in assessing the benefits and risks of the gynaecological drug misoprostol (Millard, Brhlikova and Pollock Reference Millard, Brhlikova and Pollock2015).

Third, INGOs can help improve population health by promoting norms about healthcare and policy interventions that are aimed at protecting groups like stigmatized communities, women and children, whose welfare is at risk of being neglected in the health policy process (Harris Reference Harris2019; Shiffman Reference Shiffman2016; Storeng et al. Reference Storeng2019). For instance, Save the Children helped generate political priority for newborn survival in Nepal, leading to a national strategy for neonatal health and increased resources from 2000 onwards (Smith and Neupane Reference Smith and Neupane2011). The World Federation for Mental Health was founded in 1948 and supported numerous initiatives to raise the policy and political profile of mental health, combat the stigma attached to it, and promote the human rights of psychiatric patients (Millar and Abou-Saleh Reference Millar and Abou-Saleh2011). Health INGOs can also improve health outcomes by pressuring government to spend more on the health of their population (Murdie and Hicks Reference Murdie and Hicks2013).

Fourth, INGOs can improve the provision of health services by channelling resources and implementing programmes on the ground, especially in low-income countries. Some resources originate from INGO members themselves, as in the case of the Norwegian branch of Caritas Internationalis providing funding for tuberculosis programmes run by the East Timor branch during and after the violent conflict in that country (Martins et al. Reference Martins2006). INGOs may also increase the amount of official development aid for health. Dietrich (Reference Dietrich2013, 701) notes that donors:

Given that restrictions may forestall or dissolve links between local actors and health INGOs, and that health INGOs can have a positive effect on human health through agenda setting, implementation and knowledge transfer, we hypothesize the following:

The hypothesized effect is not only unintended, but also unwanted, by governments. Under nearly all political regimes, political leaders' legitimacy depends partly on their perceived success in improving the well-being of their population, including their health (Duckett and Munro Reference Duckett and Munro2022; Rockers, Kruk and Laugesen Reference Rockers, Kruk and Laugesen2012). A survey of forty-seven countries from across the world showed that health ranked among the top two priorities of respondents in most of them (Kaiser Family Foundation and Pew Global Attitudes Project, 2007). Opponents of restrictions often use the expected harmful impact on health services to argue against restrictive legislation. Berger-Kern et al. (Reference Berger-Kern2021) show how such arguments helped campaigners and parliamentarians thwart the adoption of such laws in Kenya (in 2013) and Kyrgyzstan (during 2013–16). Nevertheless, the global trend has been towards the increasing adoption of INGO restrictions.

By describing specific effects as unintended and unwanted, we do not assume that political leaders would have abstained from legislative restrictions if they had possessed clear knowledge of the existence and magnitude of such effects. As noted in the previous section, from their point of view, restrictive laws have advantages over both informal repression and a liberal policy. It is plausible that at least some leaders regard the diminished access of their citizens to international health expertise and resources as a price worth paying to insulate them from destabilizing foreign influences. Even doctors who have only the best interest of patients at heart sometimes continue to prescribe a medicine despite its side effects, and a fortiori this applies to political leaders whose motives are not always so pure. We would expect to see substantial heterogeneity among governments in how they weigh political gains against (expected) health costs, but in the remainder of this article we limit ourselves to examining whether such health costs arise.

Research Design

Our argument on the unintended consequences of INGO restrictions has two observable implications that we will probe in the empirical section of this article. We analyze the impact of INGO restrictions on the presence of health INGOs and the disease burdens in 21 disease focus areas in these countries. This section discusses the main indicators used to measure our dependent variables, independent variables and control variables.

Dependent Variables

Our first dependent variable is the number of health INGOs active in a given country. The data were collected from the Yearbook of International Organizations (YIO). The YIO is the standard source for membership information of INGOs in the literature (Brass et al. Reference Brass2018; Murdie Reference Murdie2014; Murdie and Hicks Reference Murdie and Hicks2013). It should not be seen as a census of all INGOs active in a given country. Instead, it is generally seen as a reliable relative indicator that can sort countries according to the number of INGOs active within their borders and measure change over time (Longhofer et al. Reference Longhofer2016). Membership is a binary variable coded as 1 if a health INGO has members in a given country and as 0 otherwise. We coded it for the years 1988, 1992, 1996, 2000, 2004, 2008 and 2011. Since membership does not fluctuate rapidly by year, we filled in intervening years using the last available value.

In addition to the country-level measures of health INGOs, we also utilize more fine-grained measures of INGOs working on particular disease focus areas. INGOs are not active universally across health focus areas. Instead, many INGOs work on tackling particular diseases. Therefore, we disaggregated the data on INGO memberships to the twenty-two disease focus areas included in the Global Burden of Disease (GBD) study. Organizations were coded to disease areas using YIO subject codes. We reviewed this coding based on the name of organizations and descriptions on their website. Organizations that could not be attributed to one of the twenty-two GBD disease categories were included in a residual category. Given such issues as food safety, zoonoses and antimicrobial resistance, we also included certain veterinary and plant health INGOs that are relevant for human health. Our dataset includes 1,638 INGOs, with members in 193 countries. A total of 932 focus on one or multiple specific disease areas, while 706 of them fall into the residual health category. We include a full list of INGOs and their disease focus areas in the Online Appendix (see Table A23).

Our second dependent variable is changes in DALYs lost due to each group of diseases (GBD 2019 Diseases and Injuries Collaborators 2020). We use the first difference of DALYs lost per 100,000 people (DALY rate) as a dependent variable because we are interested in changes in the health burden over time, rather than overall disease burdens. The measure of DALYs has been developed and promoted by the World Health Organization (WHO) and the World Bank since the early 1990s to account for both deaths and disability in measuring the burden of disease. DALYs are calculated by combining the years of life lost (YLL) due to a particular disease and the years that affected people have to live with a disability (YLD). The GBD study reports disaggregated DALYs lost due to different sources of mortality and disability. We use level-two cause data, which is disaggregated into twenty-two disease categories. Our main estimations exclude the ‘interpersonal violence and self-harm’ category as a cause of mortality and disability because addressing interpersonal violence is closer to the concerns of human rights or security-related INGOs than of health INGOs, which are the focus of this article, but this category is included in a robustness check. Our dataset covers the remaining twenty-one disease categories. The categories and the average DALYs lost to these disease areas across all countries are listed in Table 1, sorted from the highest to the lowest average disease burden.

Table 1. DALYs lost globally per 100,000 people in twenty-one disease categories per year (average 1993–2017)

The methodology for calculating disease burdens in the GBD studies is not uncontroversial. For instance, critics of the early versions of the GBD study questioned decisions relating to disability weights, age weighting, discounting and life-expectancy estimates, which led to major changes in how these factors were addressed in later editions (Chen et al. Reference Chen2015). There is also a debate on how total disease burdens should guide the allocation of scarce resources to alternative health uses (Baccini, Heinzel and Koenig-Archibugi Reference Baccini, Heinzel and Koenig-Archibugi2022; Bendavid et al. Reference Bendavid2015). While important in many respects, such debates are less relevant for our specific purpose because a year-to-year reduction of DALYs lost to a disease is to be welcomed regardless of the specific weighting and discounting criteria applied. However, we also tested Hypothesis 2 using alternative health indicators and report the results in a later section.

Estimating the Impact of INGO Restrictions on the Presence of Health INGOs

We probe the extent to which restrictions suppress the growth of the INGO population. To this end, we estimate a two-way fixed-effects (TWFE) model regressing the total number of health INGOs on the existence of restrictions placed on them. Some have argued that coefficients in TWFE models can be interpreted as causal effects because the TWFE model is equivalent to a weighted average of difference-in-difference estimators. However, recent methodological advances call this interpretation into question and argue that it depends on strong assumptions (Baker, Larcker and Wang Reference Baker, Larcker and Wang2022; Goodman-Bacon Reference Goodman-Bacon2021; Imai and Kim Reference Imai and Kim2021). Therefore, our models can only be interpreted as well-specified observational regressions. Nevertheless, we include tests of common threats to inference in TWFE models in the ‘Robustness tests’ section. The estimator compares changes in the health INGO population in countries over time. All covariates are lagged by one year to avoid simultaneity bias. The standard errors are clustered by country to avoid biased standard errors due to serial correlation. Table 2 displays the results from five models. Model 1 only includes the variable measuring INGO restrictions. In Model 2, we also incorporate our control variables. In the subsequent regressions, we disaggregate restrictions: Model 3 includes operational barriers; Model 4 includes advocacy barriers; and Model 5 simultaneously includes both types of restrictions.

Table 2. Restrictions and health INGOs

Notes: Clustered standard errors at the country level in parentheses. ⁺ p < 0.10; * p < 0.05; ** p < 0.01; *** p < 0.001. The full table with control variables is displayed in the Online Appendix (see Table A1).

The results clearly show that restrictions on INGO activity are associated with a decrease in the INGO population. The coefficient is negative and statistically significant at conventionally accepted levels in all four models. For example, each additional year of INGO restrictions is associated with a decrease in the total number of health INGOs by around 3.6 in Model 2. The average number of INGOs is approximately forty-nine. Hence, if the average country had INGO restrictions for five years, our model would estimate that the INGO population would be around 36 per cent lower than if it did not (eighteen fewer INGOs).

However, the influence of INGO restrictions on the INGO population appears to be heterogeneous. Operational barriers consistently decrease the number of INGOs, while advocacy barriers do not. The finding is consistent in all models presented in Table 2. The coefficients for operational barriers are negative and statistically significant (p < 0.05). The estimated yearly decrease in the number of INGOs active in a given country due to operational barriers is, on average, between three and four. At the same time, the coefficient for advocacy barriers fails to attain statistical significance at conventional thresholds and is inconsistent across model specifications. These results are consistent with the argument that health INGOs can weather measures aimed at political activities, notably by adjusting their priorities (Fransen et al. Reference Fransen2021), but they find it more challenging to avoid the impact of legislation covering their operations in general.

Estimating the Impact of INGO Restrictions on Changes in the Burden of Disease

We now examine whether restrictive laws affect the ability of countries to prevent the loss of human life and health due to different diseases. To this end, we focus on our primary dependent variable: the change in DALY rates lost to the main twenty-one disease focus areas in the GBD study. We control for all disease-specific yearly variation by including disease-year fixed effects. We also include country fixed effects to account for time-invariant heterogeneity at the country level.

Model 6 in Table 3 displays the association of INGO restrictions and the number of health INGOs with changes in the disease burden. Model 7 controls for the number of INGOs present in a given country-disease area to account for the possibility that countries with more INGOs are also more likely to introduce INGO restrictions.

Table 3. Restrictions and DALY rates

Notes: Clustered standard errors at the country-disease level in parentheses. ⁺ p < 0.10; * p < 0.05; ** p < 0.01; *** p < 0.001. The full table with control variables is displayed in the Online Appendix (see Table A2).

The models show an association between INGO restrictions and increases in the DALYs lost due to the twenty-one disease areas. The coefficient for INGO restrictions is statistically significant (p < 0.05) and positive in Model 6. Each year with restrictions increases DALY rates by around 15.7 (see Model 6). The average change in DALY rates in the sample is −40.92. Hence, if the average country introduced INGO restrictions, we would expect these restrictions to undo around 38 per cent of the yearly progress in improving DALY rates across the twenty-one disease areas that we observe between 1992 and 2017. Furthermore, we find some evidence that health INGOs improve human health, as the coefficient for INGOs is negative and statistically significant (p < 0.001).

In subsequent regressions, we disaggregate different types of restrictions. We focus on operational barriers in Model 8, focus on advocacy barriers in Model 9, include both kinds of restrictions simultaneously in Model 10 and control for INGO presence in Model 11. In line with our theoretical expectation, the coefficient for operational barriers is positive (p < 0.05) and statistically significant across models. This result supports the argument that operational INGO barriers have unintended consequences and hamper the ability of health INGOs to perform important functions for national healthcare systems. In contrast, the coefficient for advocacy barriers fails to attain statistical significance at conventional thresholds. This finding could suggest that among the ways in which INGOs can contribute to reductions in disease burden, advocacy is the least important. However, it could also reflect that health INGOs promote norms and policies more through service design and delivery, sometimes in collaboration with public authorities (Brass Reference Brass2016; Noakes and Teets Reference Noakes and Teets2020). When confronted with new laws specifically aimed at curbing political activities, even advocacy-oriented INGOs may find ways of continuing norm and policy promotion, for instance, by reframing their demands in a more technical and less confrontational way (Fransen et al. Reference Fransen2021). We highlight this as a relevant question for further research.

Since we focused on one-year lags in INGO restrictions, previously estimated models cannot tell us much about the long-run association between INGO restrictions and changes in DALY rates. To better understand restrictions' long-run consequences, we re-estimated Model 6 using two-year, four-year, six-year, eight-year and ten-year lags in INGO restrictions as the primary independent variable. Figure 1 displays the coefficients from these five models. The coefficients are all positive but become weaker over time. The coefficients for two-year, four-year and eight-year lags are statistically significant (p < 0.1). Even eight years after restrictions have been put in place, the impact of INGO restrictions still appears to be present, worsening DALY rates by around 8 and thereby undoing around 19 per cent of the mean yearly progress in improving such rates. The findings presented in Figure 1 suggest that INGO restrictions negatively contribute to population health in the short and long run.

Fig. 1. Long-run consequences of INGO restrictions for human health (90 per cent confidence level).

Note: Models 12–16 replicate Model 6, but the variable for INGO restrictions is lagged by two, four, six, eight or ten years.

Robustness Tests

Our robustness tests aim to address four main concerns: the validity of our model specifications; endogeneity; sample restrictions; and robustness to alternative specification choices. First, we aim to verify our model specifications. We need to ensure that our results are not driven by governments becoming more illiberal and undermining public service provision at the same time. Our models would violate the common-trend assumption – which supposes that outcomes in the treatment and control groups would have been the same if countries never implemented INGO restrictions. We include three leads and lags (pre- and post-treatment dummies) in the model and show that pre-treatment dummies are not statistically significant at conventional thresholds (see Figures A1 and A2 in the Online Appendix). We further include a country-specific time trend in our models that adjusts for linear pre-treatment trends within countries (see Table A3 in the Online Appendix). The coefficients remain positive and marginally significant. Additionally, it is important to understand which differences between groups drive our results (see Tables A4 and A5 in the Online Appendix). TWFE models are weighted averages of four difference-in-difference comparisons: (1) country-disease areas that were treated in different years; (2) country-disease areas that were always treated versus those that were treated in different years; (3) country-disease areas that were never treated versus those that were treated in different years; and (4) country-disease areas that are always treated versus those that are never treated (Goodman-Bacon Reference Goodman-Bacon2021). Biases are particularly problematic when treated groups are used as effective comparisons in the first of the four comparisons. Therefore, we decompose the results into the four groups and ensure that results are not substantially driven by greater weight for this group (Baker, Larcker and Wang Reference Baker, Larcker and Wang2022; Goodman-Bacon Reference Goodman-Bacon2021). We perform a decomposition following Goodman-Bacon (Reference Goodman-Bacon2021) for the TWFE models focusing on INGOs and DALY rates (see Tables A4 and A5, as well as Figures A3 and A4, in the Online Appendix). The decomposition shows that, if anything, the timing groups lead to a more conservative estimation of the impact of INGO restrictions on INGO membership (see Table A4 in the Online Appendix). Additionally, the findings for the impact on health outcomes are consistent and very similar across the first three groups, which are responsible for a combined weight of around 98 per cent (see Table A5 in the Online Appendix). These findings strongly indicate that our estimator makes the appropriate comparisons between treatment and control groups for the conclusions we draw from the data.

The second group of robustness tests address concerns around endogeneity. We believe that the main threat to inference would be low government responsiveness to the needs of citizens, which could translate into both restrictions on their civil society activities and neglect of healthcare. As discussed, our models controlled for important covariates that capture drivers and indicators of (non-)responsiveness (notably, democracy, civil society participation, universal public services and changes in DALY rates in t – 1). We go one step further in testing robustness by employing a matching strategy to achieve covariate balance between our treatment and control groups using entropy balancing (Hainmueller Reference Hainmueller2012). We reweight units according to our main control variables in Table A6 in the Online Appendix (see Model 34) and further include the mean DALY rate change value by each country-disease and disease-year (see Model 35).

Moreover, we conduct a placebo check to further minimize the possibility that our results on health INGO memberships are driven by low government responsiveness. If the association between restrictions and health INGO memberships was fully due to differences in government responsiveness, we should expect to find a similar association between restrictions and a country's health IGO memberships (the latter being even more driven by government choices than INGO memberships). We use data on membership in health IGOs from the same source as our data on membership in INGOs, that is, the YIO. Despite the correlation between IGO and INGO memberships (r = 0.49), there are no statistically significant differences between countries with restrictions and without restrictions in the number of their health IGO memberships (see Table A7 in the Online Appendix). This increases our confidence in the finding on the effect of restrictions on INGO membership.

So far, our strategies to address endogeneity have relied on the assumption that observable differences can account for low government responsiveness. We aim to overcome this limitation in a subsequent robustness check using instrumental-variable regression. This modelling technique utilizes a variable – the instrument – that needs to fulfil two main criteria: it needs to be a predictor of our primary independent variable of interest (relevance); but it can only affect the dependent variable through its association with the independent variable of interest (excludability). We identify a variable that predicts changes in the disease burden only through the introduction of INGO restrictions. Glasius, Schalk and De Lange (Reference Glasius, Schalk and De Lange2020) provide strong evidence that INGO restrictions diffuse between regional neighbours. Therefore, the number of regional peers that have introduced INGO restrictions is a relevant predictor of INGO restrictions. We believe that the variable is excludable, conditional on country-disease fixed effects. In other words, increases in the number of peers that have introduced INGO restrictions increase the likelihood that a given country introduces INGO restrictions. At the same time, there is no other clear causal pathway through which they should affect changes in the disease burdens of that country. Hence, we use the number of countries in the same region that have introduced restrictions as an instrument. Model diagnostics show that the instrument is sufficiently strong, and the results remain marginally significant when re-estimating the models using two-stage least-squares regression (see Table A8 in the Online Appendix).

The third group of robustness checks shows that alternative choices regarding our samples do not substantially alter our results. We estimate the impact of INGO restrictions on the number of INGOs active in each disease area and on changes in DALY rates at the country level (see Table A9 in the Online Appendix). The results are substantially similar to the findings presented earlier. In our main models, we excluded interpersonal violence as a disease category. To ensure that our results do not rely on excluding interpersonal violence, we re-estimate models using the complete set of twenty-two disease areas (see Table A10 in the Online Appendix). Furthermore, we re-estimate models excluding high-income countries to address concerns about inappropriate comparisons (see Table A11 in the Online Appendix). We also re-estimate models for each of the twenty-two disease areas separately for both dependent variables. The coefficients predicting changes in INGOs are negative in twenty-one out of twenty-two models and are statistically significant at conventional thresholds in six disease-specific regressions (see Tables A12–14 in the Online Appendix). The coefficients predicting DALY rates are positive and statistically significant in eight disease areas (HIV/AIDS and sexually transmitted infections; respiratory diseases and tuberculosis; other infectious diseases; chronic respiratory diseases; diabetes and kidney diseases; musculoskeletal diseases; sense-organ diseases; and mental health). Our primary variable of interest has the expected sign but fails to attain statistical significance at conventional thresholds in seven disease areas.Footnote ² The only disease area where INGO restrictions are associated with a statistically significant decrease in the disease burden is substance-use disorders (see Tables A15–17 in the Online Appendix). Overall, the disaggregated models show that INGO restrictions affect human health in many disease areas.

In the fourth group of robustness checks, we implement alternative specification choices. We employed country and disease-year fixed effects in the main specifications presented earlier. In robustness checks, we include country, disease and year fixed effects separately. We also incorporate country-disease and year fixed effects, as well as country-disease and disease-year fixed effects in Table A18 in the Online Appendix. Additionally, we control for further lags of trends in disease burdens by including three-, five- and ten-year lags of our dependent variable (see Table A19 in the Online Appendix).

We also consider alternative dependent variables. DALYs are preferable as a measure of health burden because they systematically combine information on disability and death. Nevertheless, in Table A20 in the Online Appendix we re-estimate models using three alternative dependent variables from the GBD study: YLLs, death and disease prevalence. The results show that INGO restrictions are associated with statistically significant increases in YLLs (see Model 101) and death rates (see Model 102) but do not predict changes in disease prevalence (see Model 103). These findings suggest that INGO restrictions have a harmful effect because they undermine access to, and the quality of, healthcare and not because they make diseases more common.

Finally, we also include a host of additional control variables (see Table A21 in the Online Appendix) and exclude control variables to ensure that findings are not driven by overadjustment bias (see Table A22 in the Online Appendix). We control for the overall DALY rate to account for the disease-specific burden in a given country. A further source of omitted variable bias could be economic growth. Economic growth could decrease disease burdens and give health professionals resources to participate in international associations (World Bank 2021). Another possible confounder may be the overall engagement of countries in the international community working on specific diseases. This involvement would serve as an alternative means of policy transfer for best practices on disease alleviation. To account for this argument, we control for the number of disease-specific and general health IGO memberships of a given country, based on data we collected from the YIO. We also control for the overall number of health INGOs (disease-specific and residual categories) to ensure that we do not pick up any other changes in the health INGO population unrelated to INGO restrictions. To address the concern that our control variables cannot fully capture differences in access to healthcare, we further control for an expert rating of the degree to which poor-quality healthcare undermines citizens' ability to exercise their rights in the domestic political process from V-Dem (Coppedge et al. Reference Coppedge2021). Finally, we verify that overadjustment bias does not drive our results and exclude the control variables of DAH and civil society participation. The results are robust to these additional specification choices.