Introduction

COVID-19 has disrupted medical, social, and economic systems worldwide. Rampant infections and high death counts have led governments to implement several policies aimed at containing and combating the disease. These efforts have ranged from the relatively benign to the oppressive, from public awareness campaigns, curfews, and social distancing orders to limits on public gatherings, travel restrictions, and school and business closures. In many cases, governments have resorted to stay-at-home orders (SAHOs), which have restricted people to only essential activities such as meeting medical and sustenance needs [1]. Public health policies such as SAHOs that tightly restrict social interaction have been shown to effectively limit the spread of the disease [2], but the consequences in non-health domains have been severe [3]. School closures have impacted over 60 percent of all students globally, with localized closures affecting millions more [4]. Business closures have caused major economic ripples across a wide range of countries [5], leaving millions furloughed [6] and predictions that the pandemic may result in an additional 70 to 100 million people living in extreme poverty into the foreseeable future [7]. Because of disruptive effects like these, decisions to issue SAHOs are not taken lightly by government leaders.

The impacts on Africa have been substantial. The continent surpassed more than 100,000 deaths from COVID-19 within the first year of the pandemic [8]. Public opinion surveys of African Union member states conducted about six months into the pandemic indicate that seven in 10 respondents say they had experienced difficulties getting food and a similar proportion reported lower income than the previous year [9]. Further, the effects of the pandemic are expected to lead to a drop in GDP across African countries of between 1.4% and 7.8% [10]. Despite substantial variation in economic and social conditions, African countries are exclusively categorized as “developing economies” in the UN’s 2020 World Economic Situation and Prospects report [11], and many struggle to provide basic social services. The pandemic exacerbates this already challenging situation, particularly in Sub-Saharan Africa.

The public health situation in Africa and elsewhere created by the coronavirus is dire, which makes SAHOs, one of public health policy makers’ most effective tools for fighting infectious disease, an obvious policy option. But the serious and far-reaching social and economic consequences of SAHOs are highly predictable and far from trivial, which also make SAHOs a very costly policy option for governments to take. The objective of this study is to assess the relative roles of these competing considerations. In particular, it is to test hypotheses derived from the public health policymaking literature regarding the roles played by medical and political factors, which have been identified as key forces in public health policy [12, 13], as well as social, economic, and external factors in motivating governments in Africa to issue a SAHO in response to COVID-19 (N = 41) or not (N = 13) in the first five months of the pandemic. To the authors’ knowledge, this is the first study to comprehensively evaluate the role that the five common factors related to public health policy played in the issuance of SAHOs in response to COVID-19 in Africa.

Given this objective, the following section describes the policy context of Africa, and the next presents previous research that provides theoretical explanations for African government’s responses to the pandemic along with nine associated hypotheses to be tested. This review focuses on common medical, political, social, economic, and external factors that often influence public health policy creation [14]. The next section details the method of analysis used, event history analysis, and the data collected, which come from sources such as the World Bank, United Nations, World Health Organization (WHO), and Oxford University. The results section offers several analyses and robustness checks of the findings, which, ultimately, suggest that African governments’ decisions were significantly influenced by medical factors and factors external to the countries, only questionably influenced by political factors, and not influenced at all by social or economic factors. Finally, the study concludes by summarizing the results, addressing this study’s limitations, and placing its findings in a larger context about pandemic-motivated decision making.

The African policy context

With more than 50 countries, over 3,000 ethnicities, and between 1,250 and 2,100 spoken languages, Africa is one of the most culturally and politically complex regions on Earth (e.g., [15]). Societies across the continent function with widely varying institutional structures and governing ideologies that are influenced by colonial legacies, tribal heritage, and long-standing ethnic fractures (e.g., [16, 17]). Due to later democratization, political parties have been constrained by limited experience and, as a result, political institutions have remained under-developed in many cases [18]. Many African governments have been crippled for decades by corruption, ineffective policy creation, and terroristic armed militias. Wars, coups, and ethnic massacres have characterized the volatile political atmosphere and highlighted the problems caused by poorly considered borders inherited from previous colonial powers [16, 17].

However, not all of Africa confronts the same problems or struggles equally. The northern region is frequently studied in conjunction with the Middle East due to their greater cultural, political, and economic similarities [19–21]. Northern Africa is comprised of predominantly Arab states including Algeria, Egypt, Libya, Morocco, South Sudan, Sudan, and Tunisia. Tunisia is an outlier in the region as the only state to successfully transition to a democratic regime following the Arab Spring in 2011 [22]. The rest of Northern Africa’s authoritarian government structures, pro-Middle Eastern foreign policies, and political and economic alliances, like OPEC, reflect a deep connection with the Middle East [23–25]. In particular, oil production gives North African countries greater economic resources than many others on the continent.

The rest of the continent, referred to commonly as Sub-Saharan Africa, can be divided into four sub-regions according to the United Nations Statistical Division: Eastern, Middle, Western, and Southern. Each of these sub-regions has distinct cultures and policy contexts, drawing both from their own diverse histories and the influence of former colonial powers such as Britain and France [18]. Large states such as Nigeria possess significant developmental potential but have been subject to disconnects between abundant resources and policy direction [26]. In several African states, the aftermath of independence saw shifts away from political pluralism and the consolidation of power under one political party [18]. This has resulted in the retention of authoritarian rule for states like Libya while some states, such as Ghana, have seen eventual democratic reform attempts with varying levels of success. Other states like the Central African Republic have suffered a series of coups and civil wars that installed rulers with varying interests in democracy [18, 27, 28]. As these situations demonstrate, Africa is far from a single conglomerate of culturally or politically homogenous countries and, according to some, even dividing Africa into Northern and Sub-Saharan regions fails to capture its true diversity [26].

The vulnerable political nature of Africa is of even greater concern in the context of a pandemic. Failure to mitigate a disease outbreak can cause lost confidence in the state, resulting in the loss of domestic and even global legitimacy in extreme cases [29]. This can trigger internal and external political instability, degrading national security even in developed states [29, 30]. While this danger may make African states more likely to act to mitigate the pandemic, failure may result in a spike in regional instability and conflict. The resulting refugee crises, for instance, would place even greater strain on a region troubled by multiple long-standing refugee situations while also facilitating more rapid transmission of the virus across borders as infected people flee to safety [29]. Failure against a pandemic can exacerbate long-term humanitarian crises.

Public health and development remain salient topics in African politics and economics. The continent carries a disproportionate share of the global disease burden [31] and of epidemics [32] relative to its population. Health crises like AIDS, malaria, cholera, influenza, tuberculosis, and Ebola have ravaged the continent and demonstrated the consequences of inadequately structured healthcare policies (e.g., [33]). The United Nations’ International Health Regulations (IHR) offer a benchmark for evaluating states’ preparedness to mitigate pandemics. The IHR identifies “core capacities,” or baseline requirements, for factors deemed essential to combating pandemic diseases [34]. These capacities include “national legislation, policy and financing, coordination and National Focal Point communications, surveillance, response, preparedness, risk communication, and human resources and laboratories” [34]. In 2019, Africa averaged 46% prepared for pandemics based on the IHR core capacities scores, while countries globally averaged 63% [35]. As Fig 1 shows, though, the average masks a great deal of variation in Africa. Scores ranged from lows of 25% or less in the Central African Republic, Equatorial Guinea, Guinea-Bissau, and Lesotho to highs of 70% or more in Algeria, Egypt, Rwanda, South Africa, and Tunisia.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Distribution of UN International Health Regulations core capacities in Africa.

https://doi.org/10.1371/journal.pgph.0000112.g001

While AIDS had and continues to have a profound effect in Africa, the response to Ebola seems more pertinent to the response to COVID-19 due to its recency and means of transmission. Ebola impacted three West African nations most significantly: Guinea, Liberia, and Sierra Leone. These states suffered significant loss of life throughout the two-year outbreak, culminating in 11,000 deaths and inflicting significant impacts on the countries’ healthcare systems and economies [36]. Being three of the poorest countries in the world, under-funding and under-preparedness of their health systems contributed to Ebola’s ability to manifest as a pandemic [37]. For instance, the three countries lacked many of the conventional programs of strong public health systems such as effective prevention, vaccination, monitoring, isolation, and health promotion strategies. Moreover, they lacked enough healthcare workers in general and specialist healthcare workers in particular [38]. Recent measures show their ongoing public health predicament. They are at or slightly below the African mean in IHR total scores with Sierra Leon having attained 40% of core capacities for surveillance and response to public health events of international concern, Guinea having attained 44%, and Liberia 46%. These scores, though, put the countries well below the global mean of 63%. But during the Ebola crisis the countries deployed what resources they had to acquire tests, isolate people who tested positive, and trace and quarantine others who may have come in contact with the disease [38]. Neighboring states such as Mali and Kenya began their response strategies with border closures and bans on air travel to and from the epicenters [37]. In all, though, the Ebola crisis highlighted the weaknesses of African public health systems [37]. The hospital-based curative care structure existing in the post-colonial era was insufficient and under-equipped to control the outbreak efficiently, particularly in countries and on a continent already bearing a disproportionately high communicable disease burden compared to the rest of the world [37]. Eventually, foreign intervention helped end the crisis despite an initially lethargic response on the part of the WHO [36, 37]. Western countries, particularly France, Great Britain, and the United States, sent thousands of troops and healthcare workers and pushed aggressively to develop treatments and potential vaccines.

Method and data

Policy researchers often use panel data and event history analysis (EHA) to identify factors related to the probability that a policy will be adopted during a given time period [39, 40]. The panel data constitute the “event history,” which in this case captures every day from January 31 to June 7, 2020, for each of the 54 African countries with observations in the form of country-days. It is important that the timeframe under consideration is theoretically significant [72]. January 31 is the day after the WHO declared an international emergency related to COVID-19, and June 7 is 30 days after the last African country issued a SAHO during the first wave of the pandemic. The decision to terminate the event history 30 days after the last issuance is based on a number of considerations. None of the countries that refrained from issuing a SAHO to that point later issued a SAHO in 2020. One month is a discrete and conventional timeframe. It balances the need to give reasonable time to capture additional issuances (the mean [median] time between issuances was 2.7 [1] days and the maximum was 18 days) with the need to avoid confounding the effects of time by selecting an overextended timeframe. For the dependent variable, SAHO, each country-day is coded 0 when the country has not issued a SAHO and 1 when it has. Ultimately, 41 of the 54 countries issued a SAHO during the event history. Following EHA practices [70], a country is removed from the event history starting the day after it issues a SAHO. Based on this coding scheme, Burundi, the first country to issue a SAHO (March 5), contributes 35 observations, and the 13 countries that never issued a SAHO each contribute 129 observations. As a result, the event history includes 4,138 country-day observations to be analyzed.

Because the dependent variable is dichotomous, the relationships are estimated using probit regression. Technically, then, the relationships are estimated using pooled cross-sectional time series probit, which produces maximum likelihood estimates. The probability that a country will issue a SAHO on a specified day given that it has not already issued a SAHO (i.e., its hazard rate) can be estimated from the maximum likelihood estimates. Table 1 presents the life table for these data, including the hazard rate for the pooled countries.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Life table for pooled countries.

https://doi.org/10.1371/journal.pgph.0000112.t001

Besides the dependent variable, SAHO, the event history includes conventional internal determinants of policy adoption–political, social, and economic factors [39]–as well as internal determinants related to medicine that are specific to public health policy adoption [14]. The dataset also includes two external determinants of policy adoption–geographic and temporal diffusion [39]. For more details on the measures, see Box 1 and Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Measures by country.

https://doi.org/10.1371/journal.pgph.0000112.t002

Medical factors are tested with a measure representing threat to health system (Hypothesis 1a), previous experience with serious infectious disease (Hypothesis 1b), and healthcare system preparation for the coronavirus pandemic (Hypothesis 1c). The first measure is threat of the disease. Given concerns about “flattening the curve,” this is represented by the ratio of the cumulative number of COVID-19 cases in the country per medical doctor. This ratio ranges from 0 in the early days of the event history to almost 46 (Sao Tome and Principe) as time progressed. Because the ratio is highly positively skewed, for the regression analyses it was recoded into a series of four indicator variables with 0 cases serving as the comparison group and the remaining three groups of approximately the same size indicating progressively higher ratios. Hypothesis 1a suggests that as this ratio increases the probability that a country will issue a SAHO will also increase. This study tests robustness of this concept by replacing this ratio with cumulative COVID-19 cases per 100,000 population. This number ranges from 0 in the early days of the event history to 417 (Djibouti) as time progressed. It, too, was recoded into a series of indicator variables due to positive skew for the regression analyses. The expectation is that as this number increases the probability that a country will issue a SAHO will also increase.

The second medical measure is experience with a serious infectious disease as represented by experience with an Ebola outbreak between 2011 and 2020. This is coded as an indicator variable such that the eight countries with Ebola outbreaks (i.e., Democratic Republic of the Congo, Guinea, Liberia, Mali, Nigeria, Senegal, Sierra Leone, and Uganda) are coded 1 and the others 0. Hypothesis 1b suggests that countries with Ebola outbreak experience will be more likely to issue a SAHO.

The third medical measure is healthcare system preparation status for the COVID-19 pandemic as represented by the WHO’s Country Preparedness and Response Status report [53]. This measure is based on the self-assessed operational readiness capacity of each country and its current location on a continuum of response scenarios regarding preparedness, imported cases, and local transmission. The measure includes five categories ranging from the lowest level of country preparedness capacity, which is “no capacity” (coded 1), to the highest level, which is “sustainable” (coded 5). As shown in Table 2, the countries range from the lowest level of capacity in the Central African Republic and Comoros (coded 1) to the next-to-highest level in Algeria, Egypt, and Mauritius (coded 4). Hypothesis 1c suggests that countries with less capacity will be more likely to issue a SAHO.

The political measures include regime type (Hypothesis 2a), state capacity (Hypothesis 2b), and colonial legacy (Hypothesis 2c). Regime type is conceptualized in terms of limits on citizen liberties. Freedom House classifies countries as “not free,” “partly free,” and “free” based on their scores on indices of political and civil liberties [79]. This measure offers a direct look at the argument regarding regime practices of limiting citizen freedoms. As indicated in Table 2, Freedom House categorizes most states in Africa as not free or partly free, with a modal category of partly free. Hypothesis 2a suggests that partly free and free countries will be less likely to issue a SAHO than not free countries. This study also evaluates an alternative measure of regime type to test the robustness of results. The Center for Systemic Peace’s Polity measure scores countries in terms of autocracy versus democracy [75]. African Polity scores range from a highly authoritarian -9 in Eswatini (Swaziland) to a highly democratic 10 in Mauritius and Cape Verde. Table 2 indicates the mean on the continent is almost 3. The expectation is that authoritarian countries will be more likely to issue a SAHO.

State capacity is represented by the Center for Systemic Peace’s State Fragility Index [76]. This measure assesses countries’ ability to manage conflict, make policy, and deliver essential services. The scores reported in Table 2 range from 0 in Mauritius, indicating a condition of “no fragility,” to a nearly maximum score of 24 in the Democratic Republic of Congo, indicating “extreme fragility.” While the mean score in the region is about 13, eight of the 11 most fragile countries in the world are located in Africa. Hypothesis 2b suggests that less fragile countries will be more likely to issue SAHOs. This study tests robustness of state capacity by substituting in the model the World Bank Government Effectiveness Index. This index captures perceptions of the quality of public services and policies in countries. Scores range from -2.5 to 2.5 with higher scores indicating greater administrative capacity [80]. In Africa, the scores range from -2.4 in South Sudan to 0.9 in Mauritius with a regional mean of -0.8. The expectation is that more effective countries will be more likely to issue SAHOs.

Colonial legacy is captured by a pair of indicator variables with one variable coded 1 for countries with a French colonial history and 0 otherwise, and a second variable coded 1 for countries with neither a French nor British colonial history and 0 otherwise. The comparison group, then, is composed of the countries with a British colonial history. Table 2 shows that there are 20 African countries with a predominately British colonial history, 19 with a predominately French history, and 15 with neither. Hypothesis 2c suggests that former French colonies will be more likely to issue a SAHO than former British colonies.

The social measures include individualist versus collectivist culture (Hypothesis 3a) and population vulnerability due to social arrangements (Hypothesis 3b). Culture is captured by type of healthcare system as indicated by government expenditure on health as a percentage of total health expenditures. This reflects countries’ placement of responsibility for healthcare in regard to public responsibility, indicated by higher government spending as a percentage, versus individual responsibility, indicated by lower government spending as a percentage. Currently in Africa, the average government health expenditure as a percentage of total health expenditures is about 35% as reported in Table 2, with a wide range from a minimum of about 8% in Guinea-Bissau and South Sudan to a maximum of about 76% in Botswana. Hypothesis 3a suggests that countries where government contributes more as a percentage will be more likely to issue a SAHO. This study tests robustness of culture by substituting in the WHO’s Universal Health Coverage Service Coverage Index, which captures the availability in a country of essential health services without financial hardship on a scale of 0–100. In Africa, the scores range from 25 in Somalia to 78 in Algeria with a regional mean of about 48. The expectation is that countries with greater healthcare coverage will be more likely to issue SAHOs.

Population vulnerability due to social arrangements is captured by the UN’s measure of percent urban population. According to UN data, the mean urban population percentage across Africa is around 45%, but that varies widely between countries. More than 70% of the populations in Algeria, Djibouti, Gabon, and Kenya live in urban areas, while less than 15% in Burundi and the Democratic Republic of Congo do the same. Hypothesis 3b suggests that the probability of a SAHO increases as the percentage of urban population increases. This study tests robustness of this concept by substituting in the UN’s measure of population density per square kilometer. According to this measure, population density ranges from 3.1 people per km² in Namibia to 623.0 in Mauritius with a mean among the 54 countries of about 102. The expectation is that the probability of a SAHO increases as the population density increases. This study further explores sociodemographic effects by testing for effects related to age, ethnic fractionalization, and education.

The effect of economic forces is assessed using the UN’s measure of GDP per capita (logged). Table 2 shows that the mean GDP per capita for 2019 in Africa was about US$2,374, with Somalia on the low end at US$104 and Seychelles on the high end with US$15,692. Hypothesis 4 suggests that countries with higher GDP per capita will be less likely to issue a SAHO. This study tests robustness of this concept by substituting in the UN’s measure of unemployment rate. Unemployment ranges from 0.4% in Niger to 28.2% in Lesotho with a mean among the 54 countries of 8.8%. The expectation is that the probability of a SAHO increases as the unemployment rate increases.

For completeness related to regional variation [26] and to produce more precise estimates, the analyses also attempt to account for the political, social, and economic heterogeneity of the continent by including a series of five indicator variables for region with Northern Africa serving as the comparison group for Eastern, Middle, Southern, and Western Africa. These are the regions identified by the United Nations Statistics Division for use in its publications and databases. This study also assesses regional effects at a higher level with a single indicator variable in which the Sub-Saharan countries are compared to their Northern Africa counterparts.

The last set of measures captures external factors related to geographic (Hypothesis 5a) and temporal diffusion (Hypothesis 5b). Geographic diffusion is the percentage of a country’s bordering states that have previously issued a SAHO. Hypothesis 5a suggests that as this percentage increases, the probability of a SAHO will also increase. The measure of temporal diffusion captures the S-shaped social learning curve of acceleration then deceleration with the inflection point–the day on which diffusion turns from acceleration to deceleration–set to the mathematical midpoint of the event history on day 66. Hypothesis 5b suggests that as the day number gets closer to the inflection point, the probability of a SAHO will increase. This study tests the robustness of temporal diffusion using a linear learning process based on a count of days, starting with 1 on the first country-day and increasingly sequentially to 129 on the last country-day. The expectation is that the probability of a SAHO increases as the linear count increases.

Results

As a baseline, Table 3 presents the results of 12 bivariate probit models designed to capture the zero-order correlation of each of the independent variables with the dependent variable. The models include 4,138 country-days. Model fit ranges from BICs of 402 to 497 and pseudo R²s from 0.00 to 0.16. The estimates are based on robust standard errors, and all p-values are based on two-tailed tests. The results indicate that seven of the independent variables are at least marginally statistically associated with the issuance of SAHOs when the effects of the other independent variables are not controlled for. Model 1 generally suggests that greater threat as represented by more cases per medical doctor is associated with a greater probability of the issuance of a SAHO, but the effect is not monotonically positive. Model 7 suggests that as collectivism in the form of government spending on health increases so does the probability a country will issue a SAHO. Model 9 indicates that wealthier countries, not poorer ones, are more likely to issue a SAHO. Finally, Models 11 and 12 indicate that SAHOs diffused geographically (p = 0.08) and temporally. The remaining models indicate that the bivariate relationships are not statistically meaningful.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Bivariate probit models with average marginal effects.

https://doi.org/10.1371/journal.pgph.0000112.t003

Table 4 presents the results of nine multiple variate probit regression models. Model 1 is offered as the theoretically best model, which includes the variables that most effectively capture the factors expected to affect the issuance of a SAHO in an African country. Models A1 to A8 provide robustness checks for several of the hypothesized relationships by respecifying Model 1 using alternative measures. The models include 4,138 country-days. Each model is statistically significant. Model fit ranges from BICs of 485 to 529 and pseudo R²s from 0.21 to 0.26. The estimates are based on robust standard errors due to evidence of heteroskedasticity, and all p-values are based on two-tailed tests. Tests suggest that collinearity is not unduly affecting the regression coefficients or standard errors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Multivariate probit models with average marginal effects.

https://doi.org/10.1371/journal.pgph.0000112.t004

Fig 2 presents an overview of the results in terms of the average marginal effects (AMEs) of each measure derived from Model 1. This reporting focuses on AMEs because probit estimates provide limited substantive information other than direction and statistical significance, although all probit estimates are reported in Table 4. AMEs indicate the mean marginal effect of each variable for all observations. In this case the AMEs show the change in probability of a country issuing a SAHO on any given day assuming it has not already issued a SAHO based on a one-unit change in each continuous measure and a one-unit change from the comparison group for each indicator variable. The figure and results reported in Table 4 indicate that six of the 19 individual measures evaluated in this study yield a statistically significant relationship at conventional levels with countries’ decisions to issue SAHOs (note that temporal diffusion is statistically significant, but the change in probability is small and not discernible on this figure), and two yield marginally significant relationships (p < 0.10).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Average marginal effects derived from Model 1.

NOTE: Temporal diffusion is statistically significant (p < 0.001), but the effect is not discernible in this figure.

https://doi.org/10.1371/journal.pgph.0000112.g002

The models include three medical measures related to Hypotheses 1a to 1c. Offering considerable support for H1a, the results suggest that increased threat of disease was associated with increased probability of issuing a SAHO, which is consistent with efforts to flatten their pandemic curves to avoid threat to their healthcare systems. This is represented by a series of indicator variables that represent the ratio of COVID-19 cases to medical doctors by country and day with the comparison group being 0 cases per medical doctor. The AMEs derived from Model 1 indicate that countries in the lowest group of cases per doctor (> 0.0000–0.0068) were 1.4 percentage points more likely to issue a SAHO than when there are no cases per doctor and 1.7 percentage points more likely when in the middle group (0.0069–0.1080). The highest group (0.1081–15.8182) was 1.8 percentage points more likely, but the large standard error reduced the effect to statistical insignificance. Further analyses indicate there is no statistical difference between any of the three groups with cases. Table 4 shows that all eight of the pertinent models yielded a statistically significant effect in this regard as did Model A1, which substituted cases per 100,000 population for cases per MD. Further, previous experience with deadly infectious disease, the second medical factor, also likely played a role as suggested by H1b. The eight countries with substantial Ebola cases since 2011 were statistically more likely to issue a SAHO in all eight models. In particular, Model 1 suggests these countries were 1.9 percentage points more likely than the other countries. Finally, preparation status also played a role as suggested by H1c. Model 1 implies that less prepared countries were more likely to issue a SAHO, and the effect is almost one percentage point for each level of drop in preparedness. Overall, given the support for H1a, H1b, and H1c, the results indicate that medical considerations played nontrivial roles in countries’ decisions to issue SAHOs in response to COVID-19.

The models include three political factors, which are related to Hypotheses 2a to 2c. In terms of Hypothesis 2a, the regression coefficients estimated in Model 1 suggest that Freedom House “partly free” and “free” regimes were statistically less likely to issue a SAHO than “not free” regimes (comparison group). In terms of AMEs, partly free regimes were 2.2 percentage points less likely to issue a SAHO than not free regimes, and free regimes were 2.3 percentage points less likely. This relationship is robust across all but one of the remaining pertinent models (A1 and A3-A8). Additional analyses using Wald tests indicate there is no statistical difference between the “partly free” and “free” regimes in any of the models (e.g., Model 1: χ²(1) = 0.08, p = 0.78). Model A2, which replaces Freedom House category with Polity score to test the robustness of the effect, fails to confirm the statistically significant relationship, although the signs on the coefficient and AME are in the expected negative direction. Overall, while Model 1 supports Hypothesis 2a as do the other models using the Freedom House measure, Model A1 with the alternative measure of regime does not. Together, this suggests somewhat mixed support for Hypothesis 2a.

The second political factor, state capacity, appears to have played an unclear role in the countries’ decisions to issue SAHOs. Consistent with H2b, Model 1 indicates that as state fragility increased, a country’s likelihood of issuing a SAHO decreased, but the effect was marginally statistically significant (p < 0.06). Six of the eight pertinent models yield a similar relationship. On the other hand, the Government Effectiveness Index in Model A3 is statistically unrelated to the dependent variable. Further, and contrary to H2c, there is no evidence that colonial legacies played a role, including comparing countries with French colonial legacies to countries with no British or French colonial legacy. Overall, given the mixed support for H2a, limited support for H2b, and lack of support for H2c, it appears that political forces played a limited role at most in the countries’ decisions to issue SAHOs.

The models also account for social factors related to the perception of health as a public/collective versus private/individual responsibility and living arrangements. Contrary to H3a, countries in which healthcare is provided in a more collective manner, through a greater government contribution to healthcare expenditures, were not statistically more likely to issue a SAHO according to any of the models, including Model A4, which replaced government healthcare expenditures with Service Coverage Index score. Moreover, and contrary to H3b, the same can be said for the models addressing living conditions in terms of greater urban population and its alternative, population density. Although not detailed here, additional analyses of other sociodemographic measures also failed to yield statistically significant effects including measures of ethnic fractionalization [81], religious fractionalization [81], net migration, years of compulsory education, and percent population 65 years old and older. Overall, the results suggest that social factors played no role in the countries’ decisions to issue SAHOs.

Contrary to H4, GDP per capita failed to achieve statistical significance in any of the models in which it was specified as was the case for unemployment rate, which served as an alternative specification in Model A6. Overall, the results suggest that economic considerations played no role in the countries’ decisions to issue SAHOs.

Finally, the models also capture factors external to the countries that may have played a role, in particular geographic and temporal diffusion. Contrary to H5a, only one of the models suggests that the percentage of bordering countries previously issuing a SAHO had a statistically discernible effect on the probability a country would issue its own SAHO. On the other hand, the results are consistent with H5b. Fig 3 presents the cumulative distribution of issued SAHOs over time and confirms the expected S-shaped curve. Further, all the models indicate SAHOs diffused temporally following an S-shaped curve, though Model A7 suggests that diffusion did not fit a linear pattern. In some way, though, the failed robustness check suggests that it was not the simple passage of time that played a role, it was the distinct social learning curve that accounted for the effect. Overall, these analyses indicate that a specific pattern of temporal diffusion played a significant role, but geographic diffusion did not.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Cumulative distribution of SAHOs by date.

https://doi.org/10.1371/journal.pgph.0000112.g003

Further, Model A8 indicates that Sub-Saharan countries were 1.2 percentage points less likely to issue a SAHO than their Northern Africa counterparts, but the results in Model 1 and the remaining models suggest the effect is primarily driven by countries in Eastern and Middle Africa.

Relative effects of the five broad factors

It is also useful to assess the relative effects of the five broad factors in public health policymaking to gain an understanding of which considerations leaders prioritized in their decision-making regarding coronavirus SAHOs. Table 5 reports a number of analyses that offer some guidance on the relative effects of the medical, political, social, economic, and external factors. The first analyses present postestimation tests of Model 1 of the joint significance of the variables that compose each factor. These Wald tests suggest that the variables that compose the medical factor as a whole are statistically significant as are the variables that compose the external factor, while the variables composing the political, social, and economic factors are not.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Relative effects of the broad factors.

https://doi.org/10.1371/journal.pgph.0000112.t005

The remaining analyses in Table 5 are based on models that include only the variables composing each factor. For instance, the analysis of the political factor includes a model using only the regime type, fragility, and colonial legacy measures. The first of these analyses reports Wald χ² tests, which indicate, again, that the medical and external variables are statistically significant but also that the economic variable is statistically significant (as in Table 3). While being a less-than-ideal measure of model fit, the next column reports McFadden pseudo R² for the models using each group of variables. Again, the models using the medical and external factors stand out in their explanatory power from the other factors. The final analyses use the Bayesian Information Criterion (BIC), which scores models based on the uncertainty of their estimates [82]. Estimate uncertainty decreases as BIC values decrease, therefore, models with lower BIC values are more desirable. The table shows that the best model by far is generated by the group of external measures. The BIC for this model is almost 100 points less/better than the worst model (political) and more than 35 points less/better than the next best model (medical). To put this in context, Raftery [82] states that BIC score differences of 0 to 2 points represent “weak” evidence for preferring one model over another, differences of 2 to 6 points represent “positive” evidence, differences of 6 to 10 points represent “strong” evidence, and differences of 10 points or more represent “very strong” evidence. By this analysis, then, the external factor played the largest relative role by far in the decision to issue SAHOs followed by the medical factor, which was followed at a large distance by the economic then social and political factors.

Conclusion

The objective of this study is to identify factors related to African countries’ issuance of SAHOs in response to the COVID-19 pandemic. The analyses capture conventional measures of public health policy adoption related to medical, political, social, economic, and external factors. To the authors’ knowledge, this is the first study to comprehensively assess how the broad range of public health policymaking factors affected the issuance of SAHOs in Africa. More broadly, it sheds light on how leaders in Africa prioritized competing factors in their response to the coronavirus pandemic. Table 6 summarizes the results by hypothesis. It suggests that medical considerations played consistent roles, while external and political considerations played inconsistent roles. It also suggests that social and economic considerations played little to no role.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Summary of results by hypothesis.

https://doi.org/10.1371/journal.pgph.0000112.t006

The findings in Table 4 consistently indicate that in terms of immediate medical conditions, increased disease threat–the strain imposed on the healthcare system as indicated by the number of cases per medical doctor and, as a robustness check, the number of cases per capita–was associated with a greater probability that a country would be locked down in support of H1a. In particular, the effect of cases per doctor is consistent with public health experts’ calls to “flatten the curve” to reduce the probability that healthcare systems would be overwhelmed by patients [48]. They also show that experience with deadly infectious disease was consistently related to countries’ responses as asserted by H1b. Countries with publicly reported Ebola cases and outbreaks since 2011 were more likely to issue a SAHO than countries without, which supports the assertion that countries learn from their experiences (e.g., [49]). Further, there is consistent statistical evidence that as a country’s pandemic preparation status increased, the probability it would issue a SAHO decreased as suggested by H1c, which is consistent with the proposition that more prepared countries had other viable response options than issuing a SAHO (e.g., [51]). These results and those reported in Table 5, which showed robust combined effects for the medical variables across several tests, suggest that medical considerations played a substantial role in decisions to issue SAHOs.

The results presented in Table 4 also suggest that political factors played something of a role in SAHO decisions. The evidence from several models shows that more authoritarian regimes were more likely to issue SAHOs than more democratic regimes as proposed by H2a, but the robustness check did not confirm the effect. It seems concentration of power may have made it easier to issue disruptive policies like SAHOs (e.g., [54]), but the effect is still open to further investigation. There is also evidence in some of the models that state administrative capacity may have mattered as countries with greater capacity were more likely to issue a SAHO as suggested by H2b, but, again, the effect was not confirmed by the robustness check. Concerns about undermining state legitimacy [57], then, remain open for more research. There is no evidence to support H2c that colonial legacy played a role, as most of the related AMEs were very close to 0 and statistically insignificant. This may be consistent with recent findings that the influence of colonial institutions in Africa is waning over time [83]; but see [58]). In all, these results hint at a role for political considerations but, combined with the results in Table 5, which showed little to no joint effect for the political variables, suggest that the hypothesized political effects did not play a significant role in the decision to issue SAHOs despite arguments about the primacy of political considerations in public health policy [12, 13].

Further, there is no evidence in Table 4 that social factors mattered, whether in the form of cultural orientation toward shared responsibility for public health (H3a) or population vulnerability (H3b). Additional analyses using other socio-demographic measures also failed to detect a meaningful effect. Contrary to H4, economic considerations appear to have played no role, despite the bivariate results presented in Table 3 and previous research suggesting economic conditions play a key role in public health policy due to the extraordinary costs associated with healthcare (e.g., [36, 63]). Finally, the external factor, particularly temporal diffusion, was associated with substantial effects as suggested by H5b. Specifically, the results suggest the probability of a SAHO increased following a common S-curve pattern often associated with a social learning process (e.g., [45]). While the effect was not robust to the alternative linear specification of temporal diffusion proposed in H5a, it may be that the lack of confirmation underscores the importance of the social learning curve and not the simple passage of time. Significantly, many of the analyses, particularly in Table 5, imply that this form of temporal diffusion was a major factor, if not the major factor, in decisions to issue SAHOs.

Overall, the results, particularly from Table 5 regarding relative effects, suggest that medical considerations were a priority for decision makers, but that temporal diffusion, an external factor, played the most significant role. On the other hand, the evidence is mixed at best regarding the effect of political considerations and clear that social and economic factors played little to no role. The lack of effects for some factors, particularly the social and economic factors, may be related to the speed with which the SAHOs were issued. The 41 countries that locked down issued their SAHOs within a 64-day period. The quickly mounting worldwide cases and deaths drew significant public attention, which may have pushed the pandemic into the realm of Christensen et al.’s [46] “wicked” crises composed of transboundary, unique, and uncertain threat, which made an accelerated response rational [45]. Over such an abbreviated and anxiety provoking period of time, it is likely that longer-term factors, such as social and economic issues, would be uncertain and incompletely understood relative to the medical considerations and, therefore, devalued relative to the medical considerations.

When evaluating this study’s findings, one must also account for its limitations. Among them, the full event history record only covers about five months, which is an extremely short period of time in policy adoption research. The urgent nature of the issue drove this timeframe [45], though, and there is no reason to believe that the EHA methodology is unable to accommodate such a brief period [70]. Further, while the results address 54 nations and effects on more than 1 billion people and account for multiple factors, the nations are quite heterogenous and such a broad perspective may mask more important effects that are unique to specific nations or regions. The results do not account for the effects of subnational policies that may impact national policy [84], nor do the data capture the effects of non-public data such as leaders’ personal incentives and fears or even the limitations on the information they had to make their decisions in the early phase of the COVID-19 pandemic. Moreover, much of the data come from authoritarian regimes and/or governments with very limited state capacity. As such, the validity and integrity of the data should be rigorously considered [85]. Finally, previous research indicates that the media can have an important effect on policymaking (e.g., [86]), but resource limitations, financial as well as temporal, prevented the collection of media data.

In conclusion, this study gives researchers and policy makers an early view of a wide range of government responses to the COVID-19 pandemic. In particular, it gives a view of the policy responses from a region of the world facing a disproportionate share of the global disease burden [31]. It indicates that medical factors played consequential roles, which is consistent with calls to “follow the science.” But it also suggests that governments learned from each other or, given the short time frame, were possibly increasingly pressured as time passed to issue SAHOs to comply with normative pressures [40]. While there is some evidence that regimes granting greater freedoms to their citizens were less likely to issue a SAHO, the evidence is far from conclusive. Finally, there is no evidence that social or economic factors were consequential despite the important roles often attributed to them in public health research (e.g., [14]). In all, the results suggest that those favoring a public health-centric response to the COVID-19 pandemic should be pleased with the approach of African leaders.