Prevention- Versus Promotion-Focus Regulatory Efforts on the Disease Incidence and Mortality of COVID-19: A Multinational Diffusion Study Using Functional Data Analysis

Abstract

International marketing has rarely explored the diffusion patterns of the spread of a disease or analyzed the factors explaining the differences in the disease incidence patterns. The rapid diffusion of the novel coronavirus has engulfed the entire world in a very short time. Many countries experienced different levels of disease incidence and mortality despite implementing similar nonpharmaceutical interventions (NPIs). Drawing on the regulatory focus theory, the authors propose a framework to conceptualize and investigate the comparative efficacy of diverse NPIs that countries could adopt to prevent or curtail the diffusion of the disease incidence and mortality. They categorize these NPIs as prevention focused (containment and closures) or promotion focused (relief measures and public health infrastructure) and discuss the moderating factors that enhance or impede their effectiveness. Employing functional data analysis, the authors examine a comprehensive data set across 70 countries. They find that prevention-focused interventions inhibit disease incidence, while promotion-focused interventions enhance the nation’s ability to respond to medical emergencies and augment people’s ability to isolate themselves and slow the spread. The authors also generate insights on how a reallocation of resources between prevention- and promotion-focused efforts influence the evolution of disease incidence and mortality, with various countries falling in different clusters.

Keywords

COVID-19 functional data analysis international diffusion pandemic regulatory focus theory

International diffusion typically refers to studies focusing on the diffusion of a commercial innovation across at least two countries. There are over 50 studies on international diffusion in just the top academic journals in marketing, strategy, and economics. There are two streams of research in international diffusion. One stream analyzes the differences between diffusion processes in any two countries and determines if the differences between these countries are social and cultural. Examples of this type of research are found in Takada and Jain (1991), Gatignon, Anderson, and Helsen (1989), Helsen, Jedidi, and DeSarbo (1993), Sood and Kumar (2018), Sood and Van den Bulte (2016), and Kumar et al. (1998). The other stream of research studies spatial diffusion models, focusing on the interaction between the diffusion processes in two or more countries. The interaction is typically captured through lead-lag effect (Eliashberg and Helsen 1996; Kalish, Mahajan, and Muller 1995), where the sales process in the lead country (i.e., the country where the product was first introduced) is modeled to affect the sales process in the lag country (i.e., the country where the product was introduced a few time periods later). Another method to study the interaction among the diffusion processes in two countries is Putsis et al.’s (1997) “mixing model,” which explores the existence of such interactions. Kumar and Krishnan (2002) use a codiffusion model to study the interdependencies among the diffusion patterns in various countries. Sood and Kumar (2017) analyze the impact of diffusion of multiple generations of innovations across multiple countries. These studies observed that, when a new product is introduced early in one country and with a time lag in subsequent countries, the consumers in the lag countries learn about the product from the lead country adopters, resulting in a faster diffusion rate in the lag countries. Kumar (2013) describes the important findings of major contributions to the field and identifies major issues for further investigation.

However, international marketing has rarely explored the diffusion patterns of the spread of a disease or analyzed the factors explaining the differences in the disease incidence patterns (Dekimpe, Parker, and Sarvary 2000; Lipsitch, Swerdlow, and Finelli 2020; Van Bavel et al 2020). The context for this study is the prevailing outbreak due to COVID-19. COVID-19 is a severe acute respiratory syndrome (Li et al. 2020) that originated within Wuhan, China in late 2019 and expanded to a growing number of countries globally (Coccia 2020). On March 11, 2020, the World Health Organization (WHO) officially labeled the novel coronavirus (i.e., COVID-19) outbreak as a pandemic. The timing, severity, and impact of the pandemic varied substantially across countries and also across regions within a country (Verity et al. 2020). While most countries suffered widespread health problems, psychological anxiety, hospitalizations, deaths, and massive economic losses, some countries struggled to contain the pandemic more than others (Anderson et al 2020). The sudden and global spread of COVID-19 elicited a wide range of government interventions. In the absence of a clear and proven medical treatment to treat infections, countries adopted various nonpharmaceutical interventions (NPIs) to contain disease incidence and mortality (Anderson et al 2020; Flaxman et al 2020; Oh et al 2020). Governments initiated such NPIs as restriction of gatherings, school and business closures, temporary economic and health relief, and public information campaigns. Countries varied significantly in both the extent and timing of NPI implementation, including social distancing, contact tracing, and testing norms (Koo et al. 2020). Some countries assimilated and implemented the emerging recommendations and insights on possible best practices to prevent infection and the behavioral changes to mitigate the spread of infection faster than other countries (Sharma, Borah, and Moses 2021; Verity et al. 2020; Zandvoort et al. 2020).

Despite these interventions, the COVID-19 pandemic has affected some economies more severely than others (Zandvoort et al. 2020). For example, South Korea and Taiwan experienced a decline in new infections much sooner than Brazil and the United States. This raises the question of why some countries suffered a higher number of COVID-19 infections and mortalities than others did despite sharing similar characteristics. We propose that countries varied in their approach toward encouraging the public to follow norms imposed/suggested in a country through effective marketing of the NPIs. The regulatory focus theory can act as the key link in explaining these differences in health outcomes across countries (Higgins 1997; Keller 2006). Aaker and Lee (2006) find that framing a message based on an outcome that is consistent with a person’s regulatory orientation can help achieve a desirable behavior. In other words, they suggest that “prompting promotion-focused people to think about gains and non-gains (versus losses and non-losses) and prompting prevention-focused people to think about losses and nonlosses (vs. gains and nongains) should bring about the ‘just-right feeling’” (p. 16). Ramanathan and Dhar (2010) show that sales promotions framed as gains (vs. nonlosses) appeal to promotion-focused (vs. prevention-focused) customers, thus leading to larger basket size. Similarly, Zhao and Pechmann (2007) demonstrate that messages framed with a prevention (promotion) focus when targeted toward adolescents with a prevention (promotion) focus are most effective in convincing them to quit smoking. Petersen, Kushwaha, and Kumar (2014) use this prevention- and promotion-focused interventions to persuade customers from all over the world to buy financial products. A meta-analysis by Keller and Lehmann (2008) suggests that message tactics influence the people’s responses to health-related recommendations. In line with these findings, the regulatory focus theory can help explain the differences in outcomes of government interventions on disease incidence. Countries that were better able to motivate the public to alter their behavior in line with findings from epidemiologists and medical experts could possibly lower both the medical costs of hospitalizations, treatments, and deaths and the economic costs of lockdowns, bankruptcies, and closures. In this study, we examine differences in the efficacy of prevention- and promotion-focused COVID-19 NPIs across countries and build on the insights from the theoretical and modeling literature on new product diffusion in multinational markets to answer the following research questions:

Could effective NPIs have prevented or slowed the spread of the disease?

If NPIs in one country help curtail the spread, can other countries follow the same strategies with similar effectiveness?

If not, what are the moderating factors that either enhance or impede the effect of the NPIs on the spread of the disease?

To answer these questions, we propose a conceptual framework based on the regulatory focus theory to investigate the efficacy of various NPIs on disease incidence and mortality across countries. The regulatory focus theory outlines two types of self-regulation driving motivation-promotion focus, which is related to aspirations and accomplishments, and prevention focus, which is related to obligations and responsibilities. Both prevention and promotion orientations vary across individuals but can be induced through external messaging. Some individuals may respond better to certain NPIs if they are prevention focused, as those connect with people’s innate desire to avoid infections. Other individuals may respond better to certain NPIs if they are promotion focused, as those connect with people’s social responsibility toward others. We examine a comprehensive set of prevention- and promotion-focused NPIs adopted by various countries.

We employ functional data analysis (FDA), a collection of statistical techniques especially suited for the analysis of curves or functions in this present context. Functional data analysis provides numerous benefits to the investigation of disease incidence across countries. First, FDA overcomes the known limitations of the Bass model, the popular model on multinational new product diffusion, and can be used for stable estimates and accurate predictions even when data past the peak sales are not available. Second, FDA can predict disease incidence in a country by integrating information from (1) past patterns of disease incidence in that country, (2) past patterns of disease incidence in other countries, and (3) knowledge of the promotion-/prevention-based interventions and other unique characteristics of that country. We collect a unique and comprehensive data set on disease incidence and mortality across 70 countries. We then use FDA to study the dynamics of disease incidence and mortality. We use functional clustering to identify groups of countries that are similar in the patterns of evolution and then employ functional regression to investigate the impact of NPIs on disease incidence and mortality.

Our findings cluster countries across diverse geographic, demographic, or economic characteristics with similar patterns of disease incidence and mortality and help predict the evolution of disease incidence and mortality. The efficacy of prevention-focused closures activities was higher than prevention-focused containment activities in mitigating disease incidence. Promotion-focused relief measures were more effective than prevention-focused containment activities. However, the levels of susceptibility toward the disease moderated the impact of prevention versus promotion-focused regulatory efforts on disease mortality of COVID-19. Our simulations of prevention- and promotion-focused activities yield insights into the relative efficacy of these NPIs and help formulate pandemic response strategies.

In the next section, we develop a conceptual framework to investigate the impact of NPIs on disease incidence and mortality. Subsequently, we discuss the research methodology, model results, public policy implications, and the limitations and suggestions for future research.

Research Motivation and Conceptual Framework

Drawing on the literature review and our interviews with physicians and government officials in select countries, we identify the factors influencing disease incidence and mortality. We propose a conceptual framework to investigate the efficacy of various NPIs on the evolution of disease incidence and mortality across countries.

Factors Affecting Disease Incidence and Mortality

The global threat of COVID-19 spurred numerous research organizations to initiate investigations into the various factors influencing the disease incidence and mortality. Physicians and health care workers collaborated to predict the dynamics of transmission, diagnosis, and prognosis of disease infection and the impact of the pandemic. The WHO initiated the Solidarity Trial for Vaccines, a coordinated multinational effort to run global clinical trials and test promising treatments (Kupferschmidt and Cohen 2020). Researchers and public health agencies from more than 100 countries collaborated on multiple projects to repurpose existing drugs, test unapproved drugs that showed positive results in animal studies in the prior coronaviruses (severe acute respiratory syndrome [SARS] and Middle East respiratory syndrome [MERS]), and develop vaccines (Branswell 2020; Zhang et al 2020). Researchers developed and estimated statistical, economic, and epidemiological models on swaths of medical, sociological, and environmental data (Chowell et al. 2016; Ferguson et al. 2020; Rhodes et al. 2020). Academic journals, popular press, and media regularly shared these findings across countries and facilitated the dissemination of new findings (Jun et al. 2020).

Researchers also explored the role of environmental factors in the transmission of the novel coronavirus using collated geographical and meteorological data. Extant literature suggests some factors that may influence disease incidence include human mobility rate (Kraemer et al. 2020), medical systems and infrastructural response (Jinjark et al. 2020), humidity and temperature (Liu et al. 2020), and intrinsic transmissibility (Lipstitch, Swerdlow, and Finelli 2020). Subsequent studies show a weak correlation between temperature, humidity, and COVID-19 viral persistence (Yuan, Jiang, and Li 2020). A nation’s level of democracy, size of vulnerable populations, citizens’ perceptions, and behavioral responses toward stringent policies are likely to define its mortality growth rate (Van Bavel et al. 2020). The proportion of the urban population is considered a risk factor for the pandemic’s diffusion as the urban population faces more exposure in all aspects (Jinjark et al. 2020). At the same time, age disparities and economic conditions are likely to affect the number of observed cases across the globe. A study by Davies et al. (2020) suggests that low-income countries with younger population structures will have a low number of expected per capita incidence of clinical cases than countries with older population structures. In addition, Liu and Xiao (2013, p. 6) emphasize that “decreasing population migration is not as effective as improving the recovery rate for controlling an epidemic diffusion.” Therefore, countries with a higher level of mobility at the initial level saw a higher peak of mortality rates in the early stage of the pandemic, and countries with a larger elderly population, a greater share of employees in vulnerable occupations, and a higher level of democracy took longer to reach their peak mortalities (Jinjark et al. 2020). Ongoing research continues to yield new insights on the transmission and clinical manifestations of COVID-19.

NPIs

Because there was no medical cure or vaccine available to treat or prevent infections, countries responded to a higher disease incidence and mortality by implementing various NPIs to help slow the spread of COVID-19. These NPIs included actions to limit the transmission of infection across people such as school closures, restriction of gatherings, business closures, and stay-at-home orders and actions to provide temporary economic and health relief to the public. Several studies have reported initial estimates of the efficacy of NPIs (Flaxman et al. 2020; Lai et al. 2020).

We draw from the regulatory focus theory to propose a conceptual framework and organize these NPIs as well as to investigate the impact of these NPIs on disease incidence and mortality. The regulatory focus theory (Higgins 1997) suggests that human motivation is driven by the seeking (promotion) of pleasure and the avoidance (prevention) of pain. According to this theory, the effectiveness of a message is influenced by the mode of presentation: messages presented as gains are more influential under a promotion focus, and messages presented as losses are more influential under a prevention focus. People differ in their pursuit of goals by balancing their promotion and prevention orientations. We propose that the NPIs differ in their prevention versus promotion focus, and a country’s choice of the specific types of NPIs influence the disease incidence and mortality (see Figure 1). For example, a public information campaign to educate people on the importance of social distancing in lowering the disease incidence is an example of a promotion-focused NPI. In contrast, issuing stay-at-home orders and limiting the mobility of the public to lower disease incidence is an example of prevention-focused NPI. Social distancing was found to be strongly correlated with slower disease incidence (Lai et al. 2020; Tellis, Sood, and Sood 2020a). The manner in which a message is delivered influences its impact on behavioral change. Promotion-focused NPIs focus on gains, with sensitivity to the presence of positive outcomes, and prevention-focused NPIs focus on avoiding negative outcomes.

Figure 1.

Conceptual framework.

Prevention- Versus Promotion-Focused NPIs

We interviewed a total of 28 physicians and 22 government officials across the United States, United Kingdom, India, Middle Eastern countries, Brazil, Germany, and Australia to obtain information on the policies perceived to be related to prevention-focused and promotion-focused NPIs and the susceptibility of the population to COVID-19. Given the consensus opinion shared by the physicians and government officials, we converge on variables that define each of these proposed measures, which are consistent with the literature and popular press.

Drawing on the collective evidence and our observations, we identify two levels of prevention-focused regulatory responses. On the one hand, some governments’ interventions were directed toward hindering public interactions to contain the disease incidence. Some countries imposed limits on private gatherings. Some imposed restrictions on very large, moderate-sized, and small gatherings. A few countries imposed restrictions on internal movement between cities/regions, whereas others only issued recommendations against optional travel. Some countries issued shelter-in-place orders confining people to their homes, whereas others allowed travel exceptions for daily exercise, grocery shopping, and “essential” trips. Some countries required the closing of nonessential businesses for some sectors or categories of workers, while others left the decisions to work from home to individual business owners.

On the other hand, some governments’ interventions were directed toward legally curtailing movement through mandated closures. Some countries enforced the closing of all schools and universities, while others limited the closings to only some schools or colleges or only to some levels of education. Some countries shut down all public transportation, thereby significantly reducing the means of transportation available to most citizens. Other countries allowed internal travel but closed or severely limited all international travel. In some countries, all passengers arriving into the country were both screened for infection and quarantined, whereas other countries banned the arrival of any passenger from certain locations.

The collective evidence and our observations suggest that many countries also released additional resources to support the public health infrastructure. Additional funding was provided in some countries to support the current health expenditure per capita. Some countries responded to the higher demand for physicians by increasing the hours of duty or fast-tracking medical students’ graduation. Many countries responded by increasing the number of hospital beds by opening new hospitals and expanding existing ones. Many governments initiated interventions to assuage the health and economic consequences of the pandemic and enhanced the social welfare provisions. Some countries developed and implemented public information campaigns on COVID-19 (Djalante et al. 2020; Oh et al. 2020). New policies were developed and implemented to either control COVID-19 testing to targeted segments on the basis of specific criteria (e.g., essential workers, key personnel, those admitted to hospital or those that came into contact with a known case, people who returned from overseas) or expand testing to all (e.g., “drive-through” testing available to asymptomatic people) (Gilbert et al. 2020; Prem et al. 2020). Some countries employed digital tools such as location-based proximity tracing and data integration to support surveillance, testing, contact tracing, and strict quarantine (Keeling, Hollingsworth, and Read 2020; Zandvoort et al. 2020). Governments in many countries provided different levels of direct cash payments to people who lost their jobs or could not work. Some countries provided other cash incentives such as freezing financial obligations, stopping loan repayments, preventing services (e.g., water) from being disconnected, or banning evictions.

Countries also vary in their degree of susceptibility to disease incidence. Drawing on the collective evidence, we identify various factors that may influence the overall susceptibility to the disease and increase the risk of incidence and mortality. These factors include higher exposure from international tourist arrivals and departures, high population density, high mobility, and general health of the population. Higher disease susceptibility increases the risk of getting infected by a disease.

There is also a large diversity of the manner, timing, and stringency levels in which these NPIs were implemented. Political ideology and polarization dictated the selection of suitable pandemic response strategy often at odds with findings from research and science (Carothers and O’Donohue 2019). Governments in many countries struggled to reach a consensus on the selection and implementation of NPIs. Partisan differences influenced the government measures and messaging of information on mitigation of COVID-19 (Tellis, Sood, and Sood 2020b). Moreover, while one may have a predisposition to one orientation, both prevention- and promotion-focused actions can influence behavior depending on situational demands (Higgins 1997). An empirical analysis can help assess the efficacy of the various prevention- and promotion-focused NPIs on disease incidence and mortality.

Research Methodology

In this section, we describe our approach to uncover the dynamics of the disease incidence and the impact of prevention and promotion interventions on the disease incidence and mortality. Any proposed model needs to do the following:

Quantify the evolution of disease incidence: we achieve this objective by quantifying the underlying basic dynamics in the evolution of disease incidence using FDA.

Identify common patterns in the dynamics of evolution to assess the drivers of disease incidence: we achieve this objective by identifying common patterns in the evolution using functional principal component analysis (fPCA). We then use these common patterns to predict the evolution for individual countries while controlling for external influences such as prevention and promotion interventions and underlying demographics.

Ensure parsimony and address possible endogeneity of disease incidence.

Data

We collect data on the COVID-19 daily incidence and deaths from the Johns Hopkins Coronavirus Resource Center, U.S. Centers for Disease Control and Prevention, and Our World in Data on 70 countries from January 1, 2020, to July 18, 2020. In line with our review of extant literature, press releases, and interviews with physicians and government officials, we supplement these data with information on government responses and demographic variables from the Oxford COVID-19 Government Response Tracker, WHO, World Tourism Organization, European Centre for Disease Prevention and Control, and other sources. The sample includes countries across six continents and with varying levels of population, economic prosperity, health care development, timing of first incidence, and levels of international travel. To illustrate the sample representativeness of the population, we compare and show that the mean values of the both the sample and the country populations on the variables used in the study are similar (for details, see the notes to Table 1). We aggregate the data on a weekly basis and organize our data in a standard panel data setup. The standard cross-national measures enable us to conduct a systematic comparison across countries. We include several measures for both prevention- and promotion-based strategies to account for the large diversity of environmental, health care, political, and demographic factors across countries. As stated previously, we also interviewed physicians and government officials to categorize the NPIs into prevention- or promotion-focused interventions.

Table 1.

Constructs and Operational Measures of Variables (Study Sample of Countries).

Construct (Variable)		Source	N	Mean	SD
COVID-19 spread	Total COVID-19 cases per million	ECDC	1,558	1,357.0	2,953.7
	New COVID-19 cases per million/week	ECDC	1,558	23.0	54.3
	Total COVID-19 deaths per million	ECDC	1,558	61.0	138.0
	New COVID-19 deaths per million/week	ECDC	1,558	.8	2.2
Prevention regulatory focus	Limits on private gatherings	ECDC	1,467	2.6	1.5
	Curtailment on internal movement between cities/regions	ECDC	1,467	1.2	.8
	Orders to “shelter-in-place” or otherwise confine to the home	ECDC	1,467	1.2	1.0
	Partial closing of workplaces	ECDC	1,467	1.7	1.1
	Closing of public transport	ECDC	1,467	.8	.8
	Complete school/university closures	ECDC	1,467	2.2	1.1
	Ban/severely limit international travel	ECDC	1,467	2.9	1.2
Promotion regulatory focus	Presence of public info campaigns	ECDC	1,466	1.9	.4
	Government policy on contact tracing after a positive diagnosis	ECDC	1,466	1.4	.7
	Direct cash payments to people who lose jobs or cannot work	ECDC	1,465	.9	.8
	Government policy on access to testing	ECDC	1,529	2.4	2.1
	Freezing financial obligations for households	ECDC	1,466	1.7	.8
Infrastructure	Current health expenditure per capita, PPP (international $)	WHO Global Health Expenditure	1,466	1.0	.8
	Physicians (‘000) per 1,000 people	WHO Global Health Workforce Statistics	1,528	2.3	1.3
	Hospital beds (‘000) per 1,000 people	OECD, Eurostat, World Bank	1,527	3.3	2.6
Susceptibility	International tourism: number of arrivals per capita^a	WTO	1,564	.7	.8
	International tourism: number of departures per capita^a	WTO	1,564	.5	.6
	Urban population (%)	UN Population Division	1,564	68.4	22.2
	GDP per capita (‘000)	World Bank	1,546	30.1	22.9
	Median age of population (years)	UN Population Division	1,546	35.3	8.2
	Life expectancy at birth in 2019 (years)	UN Population Division	1,564	76.9	6.1
Test^b	Total tests (per 1,000 people)	National reports	998	32.8	51.9
Test^b	New tests (per 1,000 people)	National reports	928	.6	1.1

^a The mean (SD) of these variables for the sample of 70 countries is not significantly different using two-sided two-sample t-test of unequal variances from that of the larger population of 200 countries.

^b We did not use data on total and new tests as this information is available for less than 50% of all countries.

Notes: Source: Oxford COVID-19 Government Response Tracker (Hale et al. 2020). ECDC = European Centre for Disease Prevention and Control; WTO = World Tourism Organization; GDP = gross domestic product; PPP = purchasing power parity.

Modeling of individual curves

Functional data analysis is a fast-growing field in statistics but relatively new in marketing (Foutz and Jank 2010; Sood, James, and Tellis 2009; Stremersch and Lemmens 2009). The central paradigm in FDA involves treating the observed curve as the unit of observation (in traditional statistics, each observed point is the unit of analysis). We describe the underlying evolution of the disease incidence for each country using spline-based smooth functional curves.

The first step is to translate a set of observations on disease incidence or mortality into an underlying function. Suppose that a curve, X(t), has been measured at times t = 1, 2,…, T. Then, the smoothing spline estimate is defined as the function h(t) that minimizes for a given value of λ > 0 (Hastie, Tibshirani, and Friedman 2001).

\sum_{t = 1}^{T} {[X (t) - h (t)]}^{2} + λ \int {h^{″} (s)}^{2} d s .

The first squared error term in Equation 1 forces h(t) to provide a close fit to the observed data while the second integrated second derivative term penalizes the curvature in h(t). The tuning parameter λ acts as a roughness penalty function to measure the degree of departure from the straight line. Smaller values of λ produce more flexible estimates and vice versa. We select λ to provide the smallest cross-validated residual sum of squared errors (Hastie, Tibshirani, and Friedman 2001). Green and Silverman (1994) show that a finite dimensional natural cubic spline provides a unique solution to Equation 1. The cubic spline is divided into L regions over the entire period, where larger values of L generate a more flexible spline (Ramsay and Silverman 1997). Within the lth region, h(t) is cubic polynomial of the form

h (t) = ρ_{l} + ρ_{1 l} t + ρ_{2 l} t^{2} + ρ_{3 l} t^{3} .

There are many benefits to modeling each curve with the aforementioned approach. First, it recovers the underlying functions for the observed data. Second, the smoothening helps in eliminating noise in the data (Foutz and Jank 2010). Third, the functions and its higher-order derivatives enable insights into the dynamics of evolution of the two metrics of the pandemic: disease incidence and mortality. Fourth, the underlying functions can be used for additional analyses: functional principal components (fCPs) to generate a parsimonious representation of the curves and functional clustering to identify groups of similar curves and relate them to observed characteristics of countries and the adopted NPIs. We discuss each of these methods next.

Functional principal components

We use fPCA to find the dominant modes of variation in the evolution of disease incidence and mortality across all countries in our sample (see Figure 2) (Dass and Shropshire 2012; Sood, James, and Tellis 2009). The process of computing fPCs shares similarities with the traditional principal component analysis, with one exception: the unit of analysis is now the function computed in the first step. We retain only those fPCs that explain a large combined proportion (>95%) of the total variation in the data. As the resulting fPC curves form the basic building blocks of each curve, some curves “score higher” on the first principal component curve that captures the linear trend, while others score higher on the second component curve that captures the curvature. The resulting principal component scores of the metrics tell us the exact “mix” of principal components that enter a specific curve, as well as which shapes are more prevalent in one curve and how they differ from the shapes of another curve.

Figure 2.

FDA approach toward predictive functional clustering.

Functional clustering

Next, we use the functional clustering to segment the curves to identify common patterns (see Figure 2). Here, we segment curves using their principal component scores. In particular, we perform functional cluster analysis (Sugar and James 2003) by applying the k-means algorithm to the principal component scores of each curve. For a set of n response curves $y_{1} (s), y_{2} (s), ..., y_{n} (s)$ , the k-means clustering algorithm partitions the curves into k clusters $H = {h_{1}, h_{2}, ..., h_{K}}$ so that it minimizes the within-cluster sum of squares of each partition.

Isolating latent factors

We use exploratory factor analysis to isolate latent factors and reduce redundancy from all the operational measures. We use PROC FACTOR in SAS with Varimax rotation and retain the appropriate number of factors using the minimum eigenvalue criterion of 1.0. The results are largely supportive of the loadings of variables with the hypothesized constructs (see Table 2). All the variables related to prevention-focused government actions measuring the government responses toward hindering public movement load as hypothesized on a single factor. This factor, which we call Containment_Policies, includes prevention-focused activities such as confinement orders to “shelter-in-place” and/or stay at home, levels of restrictions on internal movement between cities/regions of a country, workplace closings, and limits on private gatherings. These policies help contain the spread of the disease incidence and subsequently, mortality. Each measure has a loading in excess of .5. This factor explains >90% of total variance. Similarly, all the variables related to prevention-focused government actions measuring the government responses toward mandated closures load as hypothesized on a single factor. This factor, which we call Closures_Policies, includes even more restrictive prevention-focused activities such as closure of all schools and universities including child-care facilities, restrictions on international travel, and the cancellation of all public transportation. Each measure has a loading in excess of .5. This factor explains >90% of total variance.

Table 2.

Analysis to Identify Latent Factors Influencing New Infections.

Factors	F1	F2	F3	F4	F5
Factor 1: Containment Policies (Prevention Focused)
Orders to “shelter-in-place” and otherwise confine to the home	.80
Curtailment on internal movement between cities/regions	.78
Partial closing of workplaces	.77
Limits on private gatherings	.69
Factor 2: Closures Policies (Prevention Focused)
Complete school/university closures		.68
Ban/severely limit international travel		.61
Closing of public transport		.56
Factor 3: Public Health Infrastructure (Promotion Focused)
Physicians (‘000) per 1,000 people			.81
Current health expenditure per capita, PPP (international $)			.68
Hospital beds (‘000) per 1,000 people			.63
Factor 4: Relief Support (Promotion Focused)
Government policy on access to testing				.56
Providing direct cash payments to people who lose jobs or cannot work.				.55
Freezing financial obligations for households				.55
Presence of public info campaigns				.47
Government policy on contact tracing after a positive diagnosis				.41
Factor 5: Country Susceptibility (Moderator)
Life expectancy at birth in 2019					.88
Median age of the population					.85
GDP per capita (‘000)					.82
Urban population (%)					.70
International tourism: number of departures per capita					.61
International tourism: number of arrivals per capita					.49

Notes: PPP = purchasing power parity; GDP = gross domestic product.

All the variables related to promotion-focused government responses and measuring the public health infrastructure and social welfare provisions load as hypothesized on two factors. The first factor, which we call PH_Infrastructure, includes promotion-focused resources such as the number of physicians to treat patients, current health expenditure per capita, and the number of hospital beds that help in addressing the challenges during a pandemic. Higher medical expenditures in the times of pandemic ensure that there are sufficient physicians and hospital beds to keep both confirmed and suspected patients under surveillance and thereby avoid the spread of self-treatment or ignoring the symptoms. The second factor, which we call Relief_Support, includes more direct promotion-focused resources to increase overall awareness about the risks of the pandemic such as public information campaigns, governments’ policy on initiating contact tracing after a positive diagnosis, and the presence of clear policies for access to testing. It also includes resources to support the population, enhance its ability to withstand the loss of jobs and earnings and the ability to travel and work. Relief policies such as freezing financial obligations for households and providing direct cash payments to people who lost their jobs or cannot work helped slow the disease spread. Each measure has a loading in the excess of .5. These two factors collectively explain >80% of total variance.

Collectively, the four factors characterize the NPIs as having either a prevention or promotion focus in line with the feedback from the frontline medical professionals. The prevention-focused interventions (containment and closures) inhibit disease incidence. The promotion-focused interventions (relief measures and public-health infrastructure) enhance a country’s ability to respond to medical emergencies and augment an individual’s ability to isolate themselves and slow the spread.

Finally, all the variables related to a country’s susceptibility to disease incidence load as expected on a single factor. This factor, which we call Country_Susceptibility, includes numerous country-level attributes and characteristics which have the potential to increase the disease incidence. This factor includes measures of general health (e.g., life expectancy at birth and median age of the population), environment, (e.g., average temperature), economic prosperity (e.g., gross domestic product per capita), demographics (e.g., urban population density) and population mobility (e.g., number of international departures and arrivals). Each measure has a loading in the excess of .5. This factor explains > 90% of total variance.

Functional regression

We next explore the drivers of disease incidence and mortality respectively and examine the efficacy of various prevention- versus promotion-focused strategies. The incidence (number of new cases) in a country within a specified time depends on the extent of prevalence (number of total cases) of the disease in the prior period and the various measures initiated to control the spread. We include lagged values of cumulative incidence ( $C I N C I D E N C E_{j t - 1}$ ) to account for prevalence. Similarly, the new cases of mortality in a country within a specified time depends on the prevalence of the disease in the prior period, the country’s susceptibility, and various relief support measures initiated to mitigate the deaths. Governments across countries adjusted the levels of prevention- and promotion-focused interventions by curtailing movement, mandating closures, and educating and supporting the public.

Addressing endogeneity issues

We account for the possible endogeneity bias of governments’ actions and disease incidence and mortality in several ways. First, the models are specified using only lagged variables. Second, we employ the two-stage predictor substitution (2SPS) method (Hausman 1978; Terza, Basu, and Rathouz 2008). Similar to the linear two-stage least squares estimator, the 2SPS method to correct endogeneity bias generates predicted values for the endogenous variables by estimating an auxiliary (reduced-form) regression in the first stage of 2SPS. The equation of interest is then estimated after replacing the endogenous variables with their predicted values in the second-stage regression. Thus, we estimate Equations 3 –5 to estimate predicted values of prevention-focused interventions toward containment and closures and promotion-focused relief support interventions to assuage the health and economic consequences of the pandemic.

C N T N_{j t} = δ_{j} + δ_{1} I N C I D E N C E_{j t - 1} + δ_{2} C I N C I D E N C E_{j t - 1} + δ_{3} M O R T A L I T Y_{j t - 1} + ϵ_{1 j t},

C L S R_{j t} = κ_{j} + κ_{1} I N C I D E N C E_{j t - 1} + κ_{2} C I N C I D E N C E_{j t - 1} + κ_{3} C M O R T A L I T Y_{j t - 1} + ϵ_{2 j t},

R E L I E F_{j t} = ϱ_{j} + ϱ_{1} I N C I D E N C E_{j t - 1} + ϱ_{2} C I N C I D E N C E_{j t - 1} + ϵ_{3 j t},

where

$I N C I D E N C E_{j t}$ = Number of new COVID-19 cases,

$C I N C I D E N C E_{j t}$ = Number of cumulative cases,

$M O R T A L I T Y_{j t}$ = Number of new deaths attributed to COVID-19,

$C M O R T A L I T Y_{j t}$ = Total number of new deaths attributed to COVID-19,

$C N T N_{j t}$ = Prevention-focused interventions toward hindering public interactions,

$C L S R_{j t}$ = Prevention-focused interventions toward mandated closures,

$R E L I E F_{j t}$ = Promotion-focused interventions to assuage the health and economic consequences of the pandemic, and

j and t represent the country and week, respectively. δ, κ, and $ϱ$ are parameters to be estimated. $ϵ_{1 j t}$ , $ϵ_{2 j t},$ and $ϵ_{3 j t}$ represent the error terms.

Moreover, the impact of both types of interventions is observed in the number of new incidences over time. We expect the impact of mandated closures and the relief support provided to the people to be effective within a shorter time period (two weeks) than the interventions to hinder public interactions and mobility (four weeks). Thus, we include a four-week lagged predicted value of the factor on containment policies ( $\hat{C N T N_{j t - 4}}$ ) but a two-week lagged predicted value of the factor on mandated closures ( $\hat{C L S R_{j t - 2}}$ ) and the factor on relief measures ( $\hat{R E L I E F_{j t - 2}}$ ). We do not use lagged values of the factor on public health infrastructure ( $P H I N F R A_{j t}$ ), as levels of the public health infrastructure did not change significantly in a short period for a substantial fraction of the countries in our sample.

We account for observed heterogeneity across countries by including dummies of functional clusters of countries with similar patterns of evolution of disease incidence ( $F C L U S T I_{k}$ ) and mortality ( $F C L U S T M_{k}$ ) over time and the interactions with the government intervention variables. As we have discussed, the positive effect of government interventions may be influenced by the intrinsic susceptibility of a country toward the disease. We include the identified factor reflecting the variables influencing the overall susceptibility ( $S S C P T B L T Y_{j t}$ ) and its interactions with the relief support activities to account for moderating the effect of higher susceptibility on the impact of government interventions on disease incidence across countries.

\begin{array}{l} I N C I D E N C E_{j k t} = \propto_{j t} + β_{1} C I N C I D E N C E_{j t - 2} + β_{2} \hat{C N T N_{j t - 4}} + β_{3} \hat{C L S R_{j t - 2}} + β_{4} {\hat{R E L I E F}}_{j t - 2} + β_{5} P H I N F R A_{j t} + β_{6} S S C P T B L T Y_{j t} + β_{7} S S C P T B L T Y_{j t} \times \hat{R E L I E F_{j t - 2}} + \sum_{k = 1}^{K - 1} γ_{k} F C L U S T I_{k} + \sum_{k = 1}^{K - 1} υ_{k} F C L U S T I_{k} \times \hat{{\hat{R E L I E F}}_{j t - 2} + \sum_{k = 1}^{K - 1} ρ_{k} F C L U S T I_{k} \times \hat{C L S R_{j t - 2}}} + ϵ_{j k t} . \end{array}

We model disease mortality by including dummies of functional clusters of countries with similar patterns of evolution of disease mortality ( $F C L U S T M_{k}$ ) over time, country susceptibility, and the interactions with the public health infrastructure.

M O R T A L I T Y_{j k t} = ϕ_{j t} + μ_{1} C I N C I D E N C E_{j t - 2} + \sum_{k = 1}^{K - 1} π_{k} F C L U S T M_{k} + μ_{2} S S C P T B L T Y_{j t} \times C I N C I D E N C E_{j t - 2} + μ_{3} S S C P T B L T Y_{j t} \times P H I N F R A_{j t} + ζ_{j k t},

where

$\hat{C N T N_{j t - 4}}$ = Endogeneity correction for prevention-focused interventions toward hindering public interactions,

$\hat{C L S R_{j t - 2}}$ = Endogeneity correction for prevention-focused interventions toward mandated closures,

$P H I N F R A_{j t}$ = Promotion-focused interventions to support public health infrastructure,

$\hat{R E L I E F_{j t - 2}}$ = Endogeneity correction for promotion-focused interventions to assuage the health and economic consequences of the pandemic,

$F C L U S T I_{k}$ = Individual dummies for functional cluster of COVID-19 incidence,

$F C L U S T M_{k}$ = Individual dummies for functional cluster of COVID-19 mortality,

$S S C P T B L T Y_{j t}$ = Intrinsic susceptibility of a country toward the disease, and

k represents the functional. α, β, γ, λ, η, ϕ, π, υ, and μ are parameters to be estimated. We assume that $ε_{i j k}$ and $ζ_{i j k}$ are correlated and estimate the two equations as a system of equations using PROC SYSLIN in SAS. Other variables are as defined previously.

Results

We first present the descriptive statistics of the key constructs and the model-free evidence. We then present results of the functional data analyses on the underlying dynamics of disease incidence and mortality. We finally present the drivers of the disease incidence and mortality, respectively.

Model-Free Results

Because of the incremental patterns of confirmed coronavirus cases on an everyday basis, countries are at various stages of the pandemic curve. Many countries, including the United States, United Kingdom, Russia, Italy, Spain, and Germany, are severely affected, with an exponential rise in cases due to lack of strict measures and seriousness toward the pandemic diffusion restriction (Anderson et al. 2020; Coccia 2020). In contrast, Japan, Singapore, Hong Kong, and Thailand have maintained a relatively flat curve with their timely deployment of well-thought-out steps designed to contain the spread of coronavirus. South Asian regions including India, Bangladesh, Pakistan, and Sri Lanka exhibited a lower degree of positive number of cases compared with North America and European states in the initial weeks of the pandemic.

Figure 3, Panel A, plots the evolution of total cases per million across selected countries and reveals numerous patterns. Some countries exhibit a high control over the disease incidence, and the number of new cases are few throughout the observed period. Other countries experience high control in the initial stages, with low levels of disease incidence in the beginning, but seem to lose control in later weeks. There are some countries with an almost linear growth pattern, whereas others exhibit S-shaped evolution patterns. Finally, many countries exhibit high levels of disease incidence from the beginning, with no signs of abatement. Figure 3, Panel B, plots the average number of total cases per million across the sample and shows that the number of cases continued to increase after an initial short period of incubation. Figure 3, Panel C, plots the first order differential of the average curve and reveals the rate of change in the average number of cases per million. The plot shows that the average rate of growth in the total number of cases started to decline in later periods.

Figure 3.

Model-free results of incidence and mortality.

Figures 3, Panels D–F, plot the corresponding curves for the evolution of total deaths per million across selected countries and reveal similar patterns. Figure 3, Panel E, reveals that the average number of deaths per million was low in the very initial stages before the numbers started to increase. Figure 3, Panel F, shows that the decline in the average rate of number of deaths per million declined much faster than the average number of total cases per million.

To explore the factors that may have resulted in this variation across countries, we compare the prevention- and promotion-based interventions discussed in the previous section. A combination of interventions such as mask mandates, citywide sanitization, partial lockdowns, time-based curfews, social distancing practices, and aggressive screening and testing is employed to battle the rising tide of positive cases. Figure 4, Panels A–D, plot these differences. The plots reveal that even though there are some similarities in the approach toward intervention across countries, there are substantial differences even within each continent. Therefore, there is a need to explore the relative efficacy of each type of intervention after controlling for unique characteristics within each country and its individual evolution pattern.

Figure 4.

Model-free results of prevention- and promotion-focused policies.

Functional principal components of disease incidence and mortality

Figure 5, Panel A, shows the fPCs of disease incidence. We see that the first principal component fPC1 (black line) is flat and monotonically declines over the observed time period. The first principal component represents the amount by which a country’s disease incidence is above or below the average disease incidence across all countries in the sample. Countries with a positive fPC score on fPC1 experience higher-than-average levels of disease incidence, and those with negative scores have below-average levels of disease incidence. Moreover, countries with a positive score decline in disease incidence over the observed time period, and those with a negative score monotonically increase over time. The second principal component fPC2 (red line) represents the rate of change of the disease incidence. The second principal component represents the amount by which the rate of change of a country’s disease incidence is above or below the average rate of change of disease incidence across all countries in the sample. Categories with a positive fPC score on fPC2 grow slowly in the early weeks but rapidly increase in later weeks, whereas those with a negative score are associated with a fast initial growth and a slowdown in later weeks.

Figure 5.

FDA approach toward predictive functional clustering of COVID-19 incidence.

We retain the first two fPCs, as they collectively explain more than 99% of the total variance. The fPC1 is the most prevalent component and explains at least 90% of the total variation across all countries. The fPC2 is a less common component and explains not more than 9% of the variation across all countries. The combination of the two principal components also reveal differences in the evolution of disease incidence across countries. For instance, countries that show an S-shaped evolution in disease incidence have high scores on both fPCs, whereas those with mainly linear trends have high scores only for the first fPC.

Figure 5, Panel B, shows the fPCs of disease mortality. The first principal component fPC1 (black line) is flat and monotonically increases over the observed time period. The interpretation is similar to that of a disease incidence, and it represents the number by which a country’s disease mortality is above or below the average disease mortality across all countries in the sample. Countries with a positive score on the first principal component experience higher-than-average levels of disease mortality, and those with negative scores have below-average levels of disease mortality. Similarly, countries with a positive score decline in disease mortality over the observed time period. The second fPC represents the way the disease mortality evolves. fPC2 initially declines in the early weeks and then increases sharply in later weeks. Countries with a positive score may experience declining mortality rates in the beginning but suffer higher mortality rates in later weeks.

We retain the first two fPCs because they collectively explain more than 99% of the total variance. The first fPC is the most prevalent component and explains at least 96% of the total variation across all countries. The second fPC is a less common element and explains not more than 3% of the variation across all countries. The combination of the two principal components also reveals differences between the evolutions of disease mortality across countries. For instance, countries that show an S-shaped evolution in disease mortality have high scores on both fPCs, whereas those with mainly linear trends have high scores only for the first principal component.

Functional clustering of disease incidence and mortality

Figure 5, Panel C, shows the results of functional clustering of disease incidence. We used the first fPC as the basis of clustering due to its high explanatory power of the variation across countries. The “jump” approach of Sugar and James (2003) suggests between five and seven clusters. We opt for five to provide the most parsimonious representation. Each curve illustrates the pattern of disease incidence of a typical country in that cluster and differs significantly from the pattern in other countries. For example, while all the countries in both Clusters 2 and 3 show an almost-linear growth in disease incidence after an initial threshold, the countries in Cluster 2 exhibit a comparatively longer period of low disease incidence that may deceptively allay fears and lead countries to lower their caution. One advantage of this FDA-based analysis is that we can identify clusters of countries that exhibit similar patterns of disease evolution even if they may differ significantly in other geographic, demographic, or economic characteristics. Countries from one cluster can learn from the pandemic response strategies of countries in other clusters on how to contain the pandemic. Figure 6, Panel A, plots the different clusters in a map to illustrate how countries from the same cluster of disease incidence may belong to diverse geographical areas.

Figure 6.

Functional clustering of countries.

Figure 5, Panel D, shows the functional clustering of disease mortality. We used the first fPC as the basis of clustering due to its high explanatory power of the variation across countries. The “jump” approach of Sugar and James (2003) suggests five clusters. Figure 6, Panel B, plots the different clusters in a map to illustrate how countries from the same cluster of disease mortality may belong to diverse geographical areas.

Drivers of disease incidence and mortality

Table 3 presents the results from estimating Equation 6. A lower disease incidence is driven by both factors measuring prevention-focused activities—containment policies (−8.27, p < .05) and closure policies (−68.18, p < .001)—controlling for total disease incidence (1.12, p < .001). The results suggest that mandated closures are more effective than containment efforts in controlling the disease incidence. Promotion-focused activities in the form of relief support also lower the disease incidence (−36.48, p < .001), as does better public health infrastructure (−5.08, p < .01). Both promotion-focused strategies (relief support) and prevention-focused strategies (closure policies) had a differential impact on each of the functional clusters of countries. We do not find support for susceptibility to disease incidence as a moderator between the relationship between a government’s NPIs and disease incidence with either the main effect (.06, p > .05) or the interaction effect with relief support (−7.88, p > .05) being significantly different from zero. The explanatory power of the model is reasonably high, especially for cross-sectional data (adj. R² = .69).

Table 3.

Drivers of COVID-19 Incidence.

Variable	Est.	SE	p-Value
Intercept	4.81	3.836	.211
Total disease incidence ( $C I N C I D E N C E_{j t - 2}$ )	1.12	.055	<.0001
Containment policies ( $\hat{C N T N_{j t - 4}}$ )	−8.27	4.473	.045
Closure policies ( $\hat{C L S R_{j t - 2}}$ )	−68.18	12.907	<.0001
Relief support ( ${\hat{R E L I E F}}_{j t - 2}$ )	−36.48	7.870	<.0001
Public health infrastructure ( $P H I N F R A_{j t})$	−5.08	1.766	.004
Incidence Cluster 2	−15.93	5.750	.006
Incidence Cluster 3	−12.53	3.839	.001
Incidence Cluster 4	−.56	3.548	.875
Incidence Cluster 5	6.50	3.111	.037
Country susceptibility ( $S S C P T B L T Y_{j t})$	.06	1.891	.976
Country susceptibility × Relief support	–7.88	4.482	.079
Relief measures × Incidence Cluster 2	–13.26	6.838	.045
Relief measures × Incidence Cluster 3	157.75	37.705	<.0001
Relief measures × Incidence Cluster 4	16.71	9.641	.083
Relief measures × Incidence Cluster 5	2.19	4.705	.642
Closure policies × Incidence Cluster 2	–44.18	9.702	<.0001
Closure policies × Incidence Cluster 3	–19.92	24.959	.425
Closure policies × Incidence Cluster 4	80.78	16.476	<.0001
Closure policies × Incidence Cluster 5	25.34	7.273	.001
N (adj. R²)	1,021 (.69)

Table 4 presents the corresponding results from estimating Equation 7. Lower disease mortality is driven by total disease mortality (−.01, p < .001). Higher country susceptibility further increases disease mortality (−.01, p < .001). The results suggest that countries with better public health infrastructure record higher disease mortality (38.46, p < .001). We believe that this seemingly counterintuitive finding is reflective of better hospital records and better access to medical care for patients suffering from COVID-19. Countries with better existing health infrastructure will also likely have older populations, which is a known risk factor for COVID-19. These patients would also likely have more comorbidities, compounding their risk.

Table 4.

Drivers of COVID-19 Deaths.

Variable	Est.	SE	p-Value
Intercept	4.64	3.264	.155
Total disease incidence ( $C I N C I D E N C E_{j t - 2}$ )	.01	.002	<.0001
Mortality Cluster 2	44.70	6.629	<.0001
Mortality Cluster 3	86.88	11.322	<.0001
Mortality Cluster 4	241.89	9.378	<.0001
Mortality Cluster 5	363.32	11.521	<.0001
Country susceptibility × Total disease incidence( $S S C P T B L T Y_{j t} \times C I N C I D E N C E_{j t - 2}$ )	.01	.002	<.0001
Country susceptibility × Public health infrastructure ( $S S C P T B L T Y_{j t} \times P H I N F R A_{j t}$ )	38.46	3.369	<.0001
N (adj. R²)	1,021 (.64)

Next, we explore how the findings in one country could have helped other countries in formulating their pandemic response strategies. Countries where the first case of COVID-19 has not been observed or where the first case was observed later can learn from other countries in the same cluster that experienced the disease spread sooner about the future trajectory of disease incidence and mortality in their cluster. This is possible, as FDA accommodates curves of unequal lengths and can make predictions on disease spread with missing data based on the principal components of similar countries. Similarly, all countries can learn from countries in other clusters about the relative effectiveness of various NPIs. In the next section, we present results of a simulation on how a reallocation of resources between prevention- and promotion-focused efforts using our estimates could have either enhanced or impeded the effect of the NPIs on the spread of the disease in two countries.

Formulating pandemic response strategies

This study can help public policy officials explore how a reallocation of resources between prevention- and promotion-focused efforts might have influenced disease incidence. We simulate the impact of alternative prevention- and promotion-focused strategies for two countries that were among the highest in total cases in July 2020: Brazil and India. Both Brazil and India have been struggling with mitigating disease incidence. India registered the first case of COVID-19 on January 30, 2020, almost a month before Brazil did on February 25, 2020. Both countries share similar levels of susceptibility to the pandemic because of their high population size, geographical diversity, human mobility, poverty, and underdeveloped medical infrastructure. India implemented more stringent prevention-focused strategies including mandated stay-at-home orders, restrictions on public gatherings and international travel, and contact tracing and testing policies. The government also responded promptly to provide relief support and ran numerous public information campaigns. However, Brazil has significantly better public health infrastructure than India, with more hospital beds and physicians per capita and larger health expenditures per capita, and thus it did not opt for the stringent prevention-focused strategies. New cases in India grew comparatively more slowly before increasing a faster pace in later weeks (Cluster 3). New cases in Brazil, in contrast, grew at a fast pace from the very beginning and at a higher rate than that in India before leveling off in later weeks (Cluster 4). By mid-July 2020, Brazil had more than 190 new cases per million, per week, whereas India had only about 25 new cases per million, per week.

Our model can provide insights on the drivers of the pandemic and how India might have limited its levels of disease incidence to a fraction of the levels in Brazil in spite of its significantly higher population density (∼18 times higher), higher percentage of population classified in extreme poverty levels, and earlier onset of the pandemic. Following the variation in the implementation on each factor across our sample, we simulated the evolution of disease incidence in both countries for the following scenarios based on the most common levels of implementations of the prevention and promotion factors across countries: lowest levels of each factor (e.g., least stringent containment or closures, least exhaustive relief measures) and 5% and 10% improvement in the factor.

The results of the simulation based on the FDAs yield at least two insights: First, the results of functional clustering suggest that the severity of the pattern of disease incidence based on initial weeks was worse in India compared with Brazil. While the pattern of a typical country in Cluster 4 (e.g., Brazil) exhibits a slowing down in later weeks, the countervailing forces are much weaker in Cluster 3 (e.g., India). Second, the results of functional regression provide insights into the relative efficacy of alternative response strategies. For example, Brazil could have achieved a higher reduction in new cases per week by strengthening more mandated closures and increasing relief support than by directing these resources toward containment strategies or enhancing public health infrastructure (see Figure 7). In contrast, the potential for mitigating disease incidence in India was higher with more resources directed toward relief efforts and public health infrastructure (see Figure 8). With strict curfew, surveillance, and lockdown measures when the disease incidence was low, India kept the disease incidence lower than Brazil (Asad, Srivastava, and Verma 2020). However, low relief and stimulus initiatives affected the capitalization on these containment and closures measures due to economic reasons ( The Lancet 2020). Thus, whereas Brazil would have benefited from more stringent containment interventions, strengthening the relief efforts and the public health infrastructure would have helped India capitalize on the implemented strict containment and closures interventions earlier. In absence of such measures, our results suggest that Brazil may exhibit a decline in weekly disease incidence much earlier than India.

Figure 7.

Simulation of influence of factors on evolution of disease incidence in Brazil.

Figure 8.

Simulation of influence of factors on evolution of disease incidence in India.

Discussion

The goal of the present study is to investigate the relative effectiveness of diverse NPIs in mitigating disease incidence and mortality and determine whether learnings from one country (cluster) could help the formulation of pandemic response strategies in other countries (in both the same cluster and other clusters). We first present a summary of our findings on the relative effectiveness of diverse NPIs. We then discuss the contributions to public policy and the application of the results. Finally, we offer some thoughts on how the new methodology of FDA can be used to explore other research topics in international marketing and discuss some limitations and ideas for future research.

Effectiveness of NPIs on Mitigating Disease Spread

Following the outbreak of the novel coronavirus in China in December 2019, cases were reported worldwide, in at least 180 countries across six continents. Governments announced millions in emergency funding and released vast resources to mobilize public health management systems to counter COVID-19. In the initial months, there was little knowledge of the symptoms, prognosis, or treatment of COVID-19. In absence of known pharmaceutical interventions, governments adopted a large variety of NPIs to educate the public and minimize the rate and extent of disease incidence and mortality.

Some of these NPIs were promotion-focused, targeting individuals who are more sensitive to the presence of positive outcomes, and others were prevention-focused, targeting individuals who are more sensitive to avoiding negative outcomes. Some examples of prevention-focused NPIs include limits on private gatherings, closing of nonessential businesses, and closing of schools and universities. Some examples of promotion-focused NPIs include strengthening the public health infrastructure, increasing the number of hospital beds, and running public health campaigns.

The efficacy of these various types of NPIs on mitigating disease incidence and mortality varied across countries for several reasons. An analysis of the dynamics of the evolution of disease incidence using FDA revealed five clusters.

Countries in Cluster 1, with low levels of disease incidence throughout the observed period, had the highest levels of both prevention- and promotion-focused NPIs and low levels of susceptibility.

In contrast, countries in Cluster 5, with high levels of disease incidence from the start of the pandemic, had moderate levels of prevention- and promotion-focused NPIs and susceptibility.

Countries from Clusters 2, 3, and 4 demonstrated a delayed onset of disease incidence and varying levels of growth over time; these countries had either low or moderate prevention- and promotion-focused NPIs but high levels of susceptibility.

On the whole, prevention-focused closure activities were more effective than prevention-focused containment activities in controlling disease incidence. However, promotion-focused relief measures were more effective than prevention-focused containment activities in controlling disease incidence. Moreover, the impact of prevention versus promotion-focused regulatory efforts on disease mortality of COVID-19 were moderated by the levels of susceptibility toward the disease.

Contribution to public policy

The findings from this global study serve to guide the development of an evidence-based countrywide pandemic response. For example, countries with a steep increase in COVID-19 incidences can evaluate what would have happened if they had adopted more effective containment measures and what sort of sources and policies would have been required to implement high-level containment measures at a country level. Similarly, countries with higher incidences of mortality could take precautions to contain vulnerable populations by providing them with all the necessary assistance to survive during a pandemic. The countries with poor medical and public health infrastructure must contemplate their future readiness to deal with such crises. Policy makers may need to think innovatively to establish the required health infrastructure in a cost-effective way.

The study provides insights on how various countries in the different clusters dealt with the pandemic in the given context and the consequences they faced. Policy makers can learn from other cluster countries in terms of implementing prevention and promotion measures to lower disease incidence and mortality in the given context. Furthermore, countries from the same cluster could cocreate the required policies and measures, as this sort of coordination would attest to the validity of the efforts. Going forward, this study could provide directions to the policy makers to use a combination of preventive and promotion measures to sustain at a country level. For example, drawing on the susceptibility level, the number of disease incidences, and mortality rate, Cluster 1 countries could categorize the regions within their country in different zones. Therefore, rather than implementing a high level of blanket prevention and promotion measures, the government could strategically define the containment zones and formulate relevant closure activities and relief measures. Furthermore, policy makers could benchmark all the factors from the framework and assess a country’s pandemic situation. If necessary, countries could seek aid and support from other nations and international bodies (e.g., the WHO).

Following the decision made on the prevention- and promotion-focused measures, governments can be prepared to implement any public policy decisions. For example, if a country has to implement strict containment measures and closures, adequate resources may be needed to support the millions of self-employed and daily wage earners. Likewise, the government may have to support affected industries (e.g., small and medium-scale enterprises, hospitality, aviation) so that firms do not shut down, creating irreversible job losses. If the pandemic continues for long, or if there is any recurrence, the findings of this study may provide guidance for any country to take preemptive actions such as creating funds and provisions to provide relief support in such times.

Expanding the use of nonparametric models in global marketing

This is the first study, to our knowledge, to expand the insights from extant literature on new product diffusion in international marketing to the context of diffusion of disease spread and mortality. We showcase the advantages of FDA, a nonparametric statistical approach, in developing insights on the diffusion patterns of the spread of a disease. Despite the rich insights from decades of research in international marketing, researchers have not explored the spatial-temporal diffusion of the spread of a disease. Functional data analysis is well-suited to examining many applications in international marketing, where the availability of time-series data poses challenges, is subject to outliers, or is measured imprecisely. The underlying approach of FDA to express discrete observations in the form of a function allows one to infer underlying dynamics across single or multiple time series in spite of missing or noisy data. Nonparametric statistical methods such as FDA require fewer or less stringent assumptions about the underlying distribution of the key variables. For example, in the present application, the jury is still out on the impact of geographic, demographic, environmental, climatic, biological, behavioral, and other factors on disease incidence (Coccia 2020; Van Bavel et al. 2020). In the absence of strong theory and comprehensive data, FDA helps researchers examine the penetration of disease across countries as a net effect of these multiple variables and enables them to develop and test theory. Similarly, researchers may employ FDA to investigate topics such as the impact of sociopolitical and psychopolitical variables on global brands or how political, technological, and cultural forces shape consumers’ preferences of global and local brands (Kumar 2015; Steenkamp 2019).

Limitations and future research

We acknowledge that this study has several limitations. First, we are limited by the availability and quality of data about this dynamic phenomenon. We limited the data sources to only those with high levels of credibility and perceived neutrality (e.g., academic institutions such as the Johns Hopkins Coronavirus research center, prominent research bodies such as the European Centre for Disease Prevention and Control and World Bank). Second, we have not yet explicitly controlled for differences in the scope, type, and sample of testing across states or regions within a country. Future researchers could examine the validity of these findings after incorporating the correction of varying levels of testing across countries. Third, we do not account for the economic costs of the prevention- and promotion-focused strategies. Public policy officials have to balance the health and economic costs in their decision making, and future research might account for such considerations. Finally, our approach could also be extended to analyze what measures might be required for a typical country to switch from the patterns of one cluster to another (e.g., for India to switch from Cluster 3 to 4).

Footnotes

Acknowledgment

The authors thank the JIM review team and their colleagues for their valuable feedback on the previous versions of the manuscript. They thank Renu for copy editing.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Associate Editor

Kelly Hewett

References

Aaker

Jennifer L.

Lee

Angela Y.

(2006), “Understanding Regulatory Fit,” Journal of Marketing Research, 43 (1), 15–19.

Anderson

Roy M.

Heesterbeek

Hans

Klinkenberg

Don

Hollingsworth

T. Déirdre

(2020), “How Will Country-Based Mitigation Measures Influence the Course of The COVID-19 Epidemic?” The Lancet, 395 (10228), 931–34.

Asad

Ali

Srivastava

Siddharth

Verma

Mahendra K.

(2020), “Evolution of COVID-19 Pandemic in India,” Transactions of the Indian National Academy of Engineering (published online September 5), DOI:10.1007/s41403-020-00166-y.

Branswell

Helen

(2020), “WHO to Launch Multinational Trial to Jumpstart Search for Coronavirus Drugs,” STAT (March 18), https://www.statnews.com/2020/03/18/who-to-launch-multinational-trial-to-jumpstart-search-for-coronavirus-drugs/#:∼:text=The%20World%20Health%20Organization,drugs%20to%20treat%20Covid%2D19.

Carothers

Thomas

O’Donohue

Andrew

(2019), Democracies Divided: The Global Challenge of Political Polarization. Washington, DC: Brookings Institution Press.

Chowell

Gerardo

Sattenspiel

Lisa

Bansal

Shweta

Viboud

Cécile

(2016), “Mathematical Models to Characterize Early Epidemic Growth: A Review,” Physics of Life Reviews, 18, 66–97.

Coccia

Mario

(2020), “Factors Determining the Diffusion of COVID-19 and Suggested Strategy to Prevent Future Accelerated Viral Infectivity Similar to COVID,” Science of The Total Environment, 729, 138474.

Dass

Mayukh

Shropshire

Christine

(2012), “Introducing Functional Data Analysis to Managerial Science,” Organizational Research Methods, 15 (4), 693–721.

Davies

Nicholas G.

Klepac

Petra

Liu

Yang

Prem

Kiesha

Jit

Mark

, CMMID COVID-19 working group, et al. (2020), “Age-Dependent Effects in the Transmission and Control of COVID-19 Epidemics,” Nature Medicine, 26, 1205–11.

10.

Dekimpe

Marnik

Parker

Philip M.

Sarvary

Miklos

(2000), “Multimarket and Global Diffusion,” in New-Product Diffusion Models, Mahajan

Vijay

Muller

Eitan

Wind

Yoram

, eds. New York: Springer, 49–73.

11.

Djalante

Riyanti

Lassa

Jonatan

Setiamarga

Davin

Sudjatma

Aruminingsih

Indrawan

Mochamad.

Haryanto

Budi

(2020), “Review and Analysis of Current Responses to COVID-19 in Indonesia: Period of January to March 2020,” Progress in Disaster Science, 6, 100091.

12.

Eliashberg

Jehoshua

Helsen

Kristiaan

(1996), “Modeling Lead/Lag Phenomena in Global Marketing: the Case of VCRs,” working paper, University of Pennsylvania.

13.

Ferguson

Neil M

Laydon

Daniel

Nedjati-Gilani

Gemma

Imai

Natsuko

Ainslie

Kylie

Baguelin

Marc

, et al. (2020), “Impact of Non-Pharmaceutical Interventions (NPIs) to Reduce COVID-19 Mortality and Healthcare Demand,” Imperial College COVID-19 Response Team (March 16), https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-COVID19-NPI-modelling-16-03-2020.pdf.

14.

Flaxman

Seth

Mishra

Swapnil

Gandy

Axel

Unwin

Juliette

Mellan

Thomas A.

Coupland

Helen

, et al. (2020), “Estimating the Effects of Non-Pharmaceutical Interventions on COVID-19 in Europe,” Nature, 584 (7820), 257–61.

15.

Foutz

Natasha

Jank

Wolfgang

(2010), “Research Note: Prerelease Demand Forecasting for Motion Pictures Using Functional Shape Analysis of Virtual Stock Markets,” Marketing Science, 29 (3), 568–79.

16.

Gatignon

Hubert

Anderson

Erin

Helsen

Kristiaan

(1989), “Competitive Reactions to Market Entry: Explaining Interfirm,” Journal of Marketing Research, 26 (1), 44.

17.

Gilbert

Marius

Pullano

Giulia

Pinotti

Francesco

Valdano

Eugenio

Poletto

Chiara

Boëlle

Pierre-Yves

(2020), “Preparedness and Vulnerability of African Countries Against Importations of COVID-19: A Modelling Study,” The Lancet, 395 (10227), 871–77.

18.

Green

Peter J.

Silverman

Bernard W.

(1994), Nonparametric Regression and Generalized Linear Models. New York: Chapman and Hall.

19.

Hale

Thomas

Angrist

Noam

Cameron-Blake

Emily

Hallas

Laura

Kira

Beatriz

Majumdar

Saptarshi

(2020), “Oxford COVID-19 Government Response Tracker,” Blavatnik School of Government, https://www.bsg.ox.ac.uk/research/research-projects/coronavirus-government-response-tracker.

20.

Hastie

Trevor J.

Tibshirani

Robert J.

Friedman

Jerome

(2001), The Elements of Statistical Learning Data Mining, Inference and Prediction. New York: Springer-Verlag.

21.

Hausman

J.A.

(1978), “Specification Tests in Econometrics,” Econometrica, 46 (6), 1251–71.

22.

Helsen

Kamel

Jedidi

Kristiaan

DeSarbo

Wayne S.

(1993), “A New Approach to Country Segmentation Utilizing Multinational Diffusion Patterns,” Journal of Marketing, 57 (4), 60–71.

23.

Higgins

E. Tory

(1997), “Beyond Pleasure and Pain,” American Psychologist, 52 (12), 1280–1300.

24.

Jinjarak

Yothin

Ahmed

Rashad

Nair-Desai

Sameer

Xin

Weining.

Aizenman

Joshua

(2020), “Accounting for Global COVID-19 Diffusion Patterns, January–April 2020,” Economics of Disasters and Climate Change, 4 (October), 515–59.

25.

Jun

Wen

Aston

Joshua

Liu

Xinyi

Ying

Tianyu

(2020), “Effects of Misleading Media Coverage on Public Health Crisis: A Case of the 2019 Novel Coronavirus Outbreak in China,” Anatolia, 31 (2), 331–36.

26.

Kalish

Shlomo

Mahajan

Vijay

Muller

Eitan

(1995), “Waterfall and Sprinkler New-Product Strategies in Competitive Global Markets,” International Journal of Research in Marketing, 12 (2), 105–19.

27.

Keeling

Matt J.

Hollingsworth

T. Deirdre

Read

Jonathan M.

(2020), “The Efficacy of Contact Tracing for the Containment of the 2019 Novel Coronavirus (COVID19),” Journal of Epidemiology and Community Health, 74 (10), 861–66.

28.

Keller

Punam A.

(2006), “Regulatory Focus and Efficacy of Health Messages,” Journal of Consumer Research, 33 (1), 109–14.

29.

Keller

Punam A.

Lehmann

Donald R.

(2008), “Designing Effective Health Communications: A Meta-Analysis,” Journal of Public Policy & Marketing, 27 (2), 117–30.

30.

Koo

Joel R.

Cook

Alex R.

Park

Minah

Sun

Yinxiaohe

Sun

Haoyang

Lim

Tao Jue

, et al. (2020), “Interventions to Mitigate Early Spread of SARS-CoV-2 in Singapore,” Lancet Infectious Diseases, 20 (6), 678–88.

31.

Kraemer

Moritz U.G.

Yang

Chia-Hung

Gutierrez

Bernardo

Chieh-Hsi

Klein

Brennan

Pigott

David M.

, et al. (2020), “The Effect of Human Mobility and Control Measures on the COVID-19 Epidemic in China,” Science, 368 (6490), 493–97.

32.

Kumar

(2013), Profitable Customer Engagement: Concept, Metrics, and Strategies. Thousand Oaks, CA: SAGE Publications.

33.

Kumar

(2014), “Understanding Cultural Differences in Innovation: A Conceptual Framework and Future Research Directions,” Journal of International Marketing, 22 (3), 1–29.

34.

Kumar

(2015), Global Marketing Research. Thousand Oaks, CA: SAGE Publications.

35.

Kumar

Ganesh

Jaishankar

Echambadi

Raj

(1998), “Cross-National Diffusion Research: What Do We Know and How Certain Are We?” Journal of Product Innovation Management, 15 (3), 255–68.

36.

Kumar

Krishnan

Trichy V.

(2002), “Multinational Diffusion Models: An Alternative Framework,” Marketing Science, 21 (3), 318–32.

37.

Kupferschmidt

Kai

Cohen

Jon

(2020), “WHO Launches Global Megatrial of the Four Most Promising Coronavirus Treatments,” Science (March 22), https://doi.org/10.1126/science.abb8497.

38.

Lai

Shengjie

Ruktanonchai

Nick W.

Zhou

Liangcai

Prosper

Olivia

Luo

Wei

Floyd

Jessica R.

, et al. (2020), “Effect of Non-Pharmaceutical Interventions to Contain COVID-19 in China,” Nature, 585, 410–13.

39.

The Lancet (2020), “India Under COVID-19 Lockdown,” The Lancet , 395 (10233), 1315. https://doi.org/10.1016/S0140-6736(20)30938-7.

40.

Ruiyun

Pei

Sen

Chen

Bin

Song

Yimeng

Zhang

Tao

Yang

Wan

(2020), “Substantial Undocumented Infection Facilitates the Rapid Dissemination of Novel Coronavirus (SARS-CoV-2),” Science, 368 (6490), 489–93.

41.

Lipsitch

Marc

Swerdlow

David L.

Finelli

Lyn

(2020), “Defining the Epidemiology of COVID-19—Studies Needed,” New England Journal of Medicine, 382 (13), 1194–96.

42.

Liu

Jiangtao

Zhou

Yao

Jinxi

Zhang

Xiuxia

Lanyu

Xiaocheng

, et al. (2020), “Impact of Meteorological Factors on the COVID-19 Transmission: A Multi-City Study in China,” Science of the Total Environment, 726, 138513.

43.

Liu

Ming

Xiao

Yihong

(2013), “Modeling and Analysis of Epidemic Diffusion with Population Migration,” Journal of Applied Mathematics, 2013, 583648.

44.

Juhwan

Lee

Jong-Koo

Schwarz

Dan

Ratcliffe

Hannah L.

Markuns

Jeffrey F.

Hirschhorn

Lisa R.

(2020), “National Response to COVID-19 in the Republic of Korea and Lessons Learned for Other Countries,” Health Systems and Reform, 6 (1), e–1753464.

45.

Petersen

J. Andrew

Kushwaha

Tarun

Kumar

(2015), “Marketing Communication Strategies and Consumer Financial Decision Making: The Role of National Culture,” Journal of Marketing, 79 (1), 44–63.

46.

Prem

Kiesha

Liu

Yang

Russell

Timothy W.

Kucharski

Adam J.

Eggo

Rosalind M.

Davies

Nicholas

, et al. (2020), “The Effect of Control Strategies to Reduce Social Mixing on Outcomes of the COVID-19 Epidemic in Wuhan, China: a Modelling Study,” The Lancet Public Health, 5 (5), e261–70.

47.

Putsis

William P.

Balasubramanian

Sridhar

Kaplan

Edward H.

Sen

Subrata K.

(1997), “Mixing Behavior in Cross-Country Diffusion,” Marketing Science, 16 (4), 354−70.

48.

Ramanathan

Suresh

Dhar

Sanjay K.

(2010), “The Effect of Sales Promotions on the Size and Composition of the Shopping Basket: Regulatory Compatibility from Framing and Temporal Restrictions,” Journal of Marketing Research, 47 (3), 542–52.

49.

Ramsay

James O.

Silverman

Bernard W.

(1997), Functional Data Analysis. New York: Springer-Verlag.

50.

Rhodes

Tim

Lancaster

Kari

Lees

Shelley

Parker

Melissa

(2020), “Modelling the Pandemic: Attuning Models to Their Contexts,” BMJ Global Health, 5 (6), e002914.

51.

Sharma

Amalesh

Borah

Sourav Bikash

Moses

Aditya C.

(2021), “Responses to COVID-19: The Role of Governance, Healthcare Infrastructure, and Learning from Past Pandemics,” Journal of Business Research, 122, 597–607.

52.

Sood

Ashish

James

Gareth M.

Tellis

Gerard J.

(2009), “Functional Regression: A New Model for Predicting Market Penetration,” Marketing Science, 28 (1), 36–51.

53.

Sood

Ashish

Kumar

(2017), “Analyzing Client Profitability Across Diffusion Segments for a Continuous Innovation,” Journal of Marketing Research, 54 (6), 932–51.

54.

Sood

Ashish

Kumar

(2018), “Client Profitability of Diffusion Segments Across Countries for Multi-Generational Innovations: The Influence of Firm, Market, and Cross-National Differences,” Journal of International Business Studies, 49, 1237–62.

55.

Sood

Ashish

Van den Bulte

Christophe

(2016), “Wider Gaps in a Flatter World? The Speed of New Product Diffusion in Rich Versus Poor Countries,” MSI Working Paper Series Report No. 16-113, https://escholarship.org/content/qt1jv6g2sq/qt1jv6g2sq.pdf.

56.

Steenkamp

Jan-Benedict E.M

. (2019), “Global Versus Local Consumer Culture: Theory, Measurement, and Future Research Directions,” Journal of International Marketing, 27 (1), 1–19.

57.

Stremersch

Stefan

Lemmens

Aurélie

(2009), “Sales Growth of New Pharmaceuticals Across the Globe: The Role of Regulatory Regimes,” Marketing Science, 28 (4), 690–708.

58.

Sugar

Catherine A

James

Gareth M.

(2003), “Finding the Number of Clusters in a Dataset,” Journal of the American Statistical Association, 98 (463), 750–63.

59.

Takada

Hirokazu

Jain

Dipak

(1991), “Cross-National Analysis of Diffusion of Consumer Durable Goods in Pacific Rim Countries,” Journal of Marketing, 55 (2), 48–54.

60.

Tellis

Gerard J.

Sood

Ashish

Sood

Nitish

(2020a), “How Long Should Social Distancing Last? Predicting Time to Moderation, Control, and Containment of COVID-19,” USC Marshall School of Business Research Paper, https://doi.org/10.2139/ssrn.3562996.

61.

Tellis

Gerard J.

Sood

Ashish

Sood

Nitish

(2020b), “Why Did US Governors Delay Lockdowns Against COVID-19? Disease Science vs Learning, Cascades, and Political Polarization,” USC Marshall School of Business Research Paper, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3575004.

62.

Terza

Joseph V.

Basu

Anirban

Rathouz

Paul J.

(2008), “Two-Stage Residual Inclusion Estimation: Addressing Endogeneity in Health Econometric Modeling,” Journal of Health Economics, 27 (3), 531–43.

63.

Van Bavel

Jay J.

Baicker

Katherine

Boggio

Paulo S.

Capraro

Valerio

Cichocka

Aleksandra

Cikara

Mina

, et al. (2020), “Using Social and Behavioural Science to Support COVID-19 Pandemic Response,” Nature Human Behaviour, 4 (5), 460–71.

64.

Verity

Robert

Okell

Lucy C.

Dorigatti

Ilaria

Winskill

Peter

Whittaker

Charles

Imai

Natsuko

, et al. (2020), “Estimates of the Severity of Coronavirus Disease 2019: A Model-Based Analysis,” Lancet Infectious Diseases, 20 (6), 669–77.

65.

Yuan

Shu

Jiang

Si-Cong

Zi-Lin

(2020), “Do Humidity and Temperature Impact the Spread of the Novel Coronavirus?” Frontiers in Public Health, 8, 240.

66.

Zandvoort

Kevin V.

Jarvis

Christopher I.

Pearson

Carl A.B.

Davies

Nicholas G.

, CMMID COVID-19 working group, Ruwan

Ratnayake

, et al. (2020), “Response Strategies for COVID-19 Epidemics in African Settings: A Mathematical Modelling Study,” BMC Medicine, 18, 324.

67.

Zhang

Tengyue

Yudi

Wenshuai

Aiping

Yang

Yanli

Kai-Feng

(2020), “Clinical Trials for the Treatment of Coronavirus Disease 2019 (COVID-19): A Rapid Response to Urgent Need,” Science China Life Sciences, 63 (5), 774–76.

68.

Zhao

Guangzhi

Pechmann

Cornelia

(2007), “The Impact of Regulatory Focus on Adolescents’ Response to Antismoking Advertising Campaigns,” Journal of Marketing Research, 44 (4), 671–87.