Can Lockdown Reduce Infection Growth in Developing Countries? Evidence from COVID-19 Cases

Background: When crippled with COVID-19 infection, a substantial number of countries have adopted ‘lockdown’ or similar measures to suppress the spread. This instrument is often considered as the only viable option for curbing infection spread both in developed and developing countries; however, some experts have a cynical view on its effectiveness. Exploiting cross-country lockdown information, effectiveness of lockdown on slowing the pace of COVID-19 can be elicited. Methods: The study intends to understand whether lockdown or similar measures can suppress infection growth in developing countries. In this pursuit, the study uses panel regression-based difference in difference and GMM estimation method. Results: This study finds that lockdown type measures are not as effective in developing countries as in developed nations. Nevertheless, staying at home order, income support programs, and other social distancing measures are found to be effective for both developed and developing countries. Also, the timing of the lockdown is found to be vital. One the one hand, enforcing a nationwide lockdown too early, i.e., when cases are very low, may not yield expected outcome; on the other hand, enforcing lockdown too late is also ineffective. Conclusion: Even though this study does not find strong evidence of the effectiveness of lockdown in curbing infection growth in developing countries, these findings do not necessarily suggest that lockdown should not be enforced in developing countries. Rather it indicates that lockdown should be combined with other complementary measures such as contact tracing, extensive testing, income support for the poor, effective management of informal and migrant workers to make the lockdown effective. Merely declaring lockdown, without accompanying other must-have measures, will hurt the economy without contributing much to reducing the growth of infection.


Background
To contain COVID-19, a significant number of countries have adopted 'lockdown' or similar measures. One view sees this type of instrument as the only viable option of curbing infection spread both in developed and developing countries. The countervailing view is that this will not work well in developing countries for several reasons. First, the cultural and institutional practice is much differing in developing countries whenever it comes to the tendency to comply with laws and regulations. Second, a vast majority of developing world has lower literacy rates, which may be positively correlated with superstitious beliefs that a contagious disease like COVID-19 will not target the pious people rather the sinners. Third, millions of impoverished people, who live hand to mouth, are hosted in the developing countries. They cannot afford to confine themselves at home for a long time as not going out of home is tantamount to no income and no food. Therefore, maintaining social distancing-key objective of a lockdown--is not a choice for millions of people in those countries. In addition, those countries cannot afford to provide financial and food assistance at a scale that can reach most of the poorest segment of the population, let alone to the ones who are relatively well off. Be it for relapse of regulation or lack of affordability, or economic structure; these countries cannot afford lockdown for longer period. As a result, developing countries may not be able to suppress the spread of the outbreak successfully with 'lockdown'--a widely prescribed playbook of curbing the spread of infection. This strategy, however, has been touted as the most, if not the only, practical strategies to restraint the spread of infectious diseases. This is even more true for COVID-19 since it is enormously contagious.
By May 2020, more than one-third of the countries of the world have adopted lockdown type measures with closings school colleges, restricting the movement of the transports, and other draconian measures, partly influenced by the prescriptions of many international institutions or maybe just due to 'following the crowd' principle. For instance, India and Bangladesh have also adopted this policy within few days of first COVID-19; however, their cases, especially in India, have been skyrocketing since then. This rocky rise has influenced many people to believe that lockdown type measures are not an effective instrument in those areas even though it has been showing promises in the western world.
While lockdown itself is an essential tool, the timing of the lockdown turns out to be crucial as well. For instance, Fiji is one of the countries which adopted lockdown very early; the same is true for Australia, and both were able to reduce the infection and keep total tally very low. New Zealand, Sri Lanka have also shown some illuminating success. On the other hand, many countries, including Italy and the US, have adopted lockdown (or some form of it) in such a stage that it was difficult to see any impact. Because it is the nature of the epidemic that follows logistic type distribution, so when they are in an exponential growth stage, it is incredibly difficult, if not impossible, to reduce the infection significantly. The characteristics of having a relatively long incubation period, being extremely contagious, having the feature of asymptotic cases makes the fight against COVID-19 an uphill battle. Though much debates as to the effectiveness of lockdown are seen in the media talks and in political spheres, any rigorous study in this issue is lacking which has left further debates to grow. Some empirical exercises on the impact of non-pharmaceutical intervention (NPI) is available for China (Lau et al., 2020) or for some specific regions especially for Europe (Flaxman et al., 2020); and a few more theoretical work are also evident that explains the impact of misinformation (Bursztyn, Rao, Roth, & Yanagizawa-Drott, 2020) and optimal timing of lockdown (Alvarez, Argente, & Lippi, 2020). In a recent paper, Moghbelli et al., 2020 argued that an RCT can provide better answer since countries are taking several intervention altogether which makes it difficult to isolate the impact of any specific intervention. While RCT can provide a better evidence on what works and what does not, it is difficult or costly to implement such experiments in pandemic condition.
While one group takes effectiveness of lockdown as granted, another group is flooding the claims of its ineffectiveness-with both groups completely without a strong evidence. In theory, it should work since lockdown means an increase in social distancing, and so the virus will not find its hosts, and its effort to grow will be thwarted. Nevertheless, in practice, a lockdown--unless it can be enforced the way draconian governments have recently executed--may not ensure a slower growth in infections. Instead, it may create a public outcry if it persists for long time, and as a result, people may violate the lockdowns limiting its prowess to curb infection. Furthermore, in most of the developing countries, many migrant workers live in the big cities; closure means returning home and at a large number, and flocking together. It may increase the spread of the virus, increasing the chance of community transmission.
Using COVID-19 case data from all countries of the world, this study attempted to answer this question-whether lockdown can tame the spread of the growth of infection. To do so, this study employs a sophisticated modeling approach---it uses both a panel version of difference-indifference (DiD) and a dynamic panel regression (GMM) modeling approach.
The study finds that lockdown does not necessarily reduce the infection in developing countries even though it does a pretty good job in curbing infections in the developed nations. Most importantly, the study finds that it is the timing of lockdown that matters most rather than the lockdown itself. The study does not necessarily imply that developing countries should not enforce lockdown. Rather, this study deduces that lockdown should be enforced along with other must-have measures to reap the fruit of it.
The paper is organized as follows. After the introduction, the paper provides an account of methods. Then it presents the findings with discussions followed by a conclusion.

Data
The data includes the daily infection rates from December 29, 2019 until May 9, 2020. This data is collected from the European Center for Disease Control and Prevention (European Centre for Disease Prevention and Control, 2020). Along with case data, this study also uses dataset from three other sources. One is the google mobility data which provides change in the mobility compared to base period  in six key areas: grocery and pharmacies, parks, transit stations, retail & recreation, residential and workplace (Google, 2000). This dataset offers a reasonable proxy measures of compliance to the lockdown which has already been used in other COVID-19 related studies (Sampi, 2020;Yilmazkuday, 2020b). And mobility data has been used in other studies to understand the impact social distancing (Yilmazkuday, 2020a) . The second data source is the stringency of lockdown collected from the Blavatnik School of Government, University of Oxford (Hale, Petherick, Phillips, & Webster, 2020). This dataset provides various levels of closure and containment measures and economic responses of government towards COVID-19 as well as a stringency index. In this study, I use eight measures of closures and containments, stringency index and income support measures. All are ordinal scale meaning the higher number, the higher the extent, except stringency index which is measured as index and normalized to 100. Third data set is from a Wikipedia page (wikipedia, 2020) which lists the dates of when lockdown was enforced in each countries. This dataset is cross-checked with other sources including newspapers of the respective countries. Figures 1-8 provide some visual explanation of the data and their relationship among the variables used in this study.

Empirical strategy
Along with traditional difference in difference estimation method, two broad empirical strategies are applied to understand the impact of lockdown on the growth of COVID-19 Cases: Panel Regression-Based Difference-in-Difference (DiD) estimation and GMM estimation.

Regression Based DID estimates
Instead of a straight difference in difference estimates, the panel regression-based estimation method is used. There are a few advantages of regression difference in difference estimation in comparison with straight DiD estimates: a) it is easy to calculate standard errors b) other related control or confounding variable can be applied c) inclusion of multiple period or trend is possible, d) it can capture the impact under different level of intensities (Shimul, 2017), and in this study, I used both lockdown and its intensity. A similar approach has been applied to understand the impact of social mobility on COVID-19 spread (Yilmazkuday, 2020b). The following regression technique is used to estimate the impact of lockdown: Here, = daily growth of infection, Treatment=a dummy if the observation is in the treatment group i.e. the country has adopted lockdown; Post=post treatment dummy, = other controls or confounding variables, =country/panel, and =time(day)

GMM Estimates
The estimation technique (1) can provide a good estimate under certain assumptions: no serial autocorrelation, no unobserved heterogeneity, and the regression equation does not include dynamic component (i.e., lag of dependent variable). However, this estimation technique will not be appropriate if any of these assumptions are violated. While some of the problems can easily be tacked by using robust standard errors and fixed effects (if there are time-invariant components), if any dynamic components are used or if there is an endogeneity (reverse-causality), then OLS or Panel Regression with fixed effects will not be ideal choice of estimation. In that case, the coefficients will be biased; and either Instrumental variables or GMM approach would be preferred. Since the growth of infection will likely to be largely determined by the growth of infection of previous periods, the inclusion of autoregressive variables is necessary. Also, the inclusion of lag dependent variable can capture many unobservable components. In this case, the specification will be: However, the inclusion of lag dependent variables introduces other problems in the estimation. Now, if OLS or Panel Regression will be biased and inconsistent since the error term will be correlated with due to the presence to lag dependent variable in the right-hand side of the equation. Arellano and Bond (Arellano & Bond, 1991) offer an estimation technique--difference GMM--to deal with the problem associated with heterogeneities. Taking a first difference of the equation (2), we get- The problem is in (3), ∆ , −1 is still endogenous. Arellano and Bond show that if this variable is instrumented with lags of their differences under some sets of moment conditions, this estimation will be valid, and this called difference GMM. However, Blundell and Bond (Blundell & Bond, 1998) show that if T is a small or dependent variable is highly persistent, then difference GMM will pick weak instruments leading to invalid results. Even though in this study, T is large for most countries, time-persistency (i.e. this period's value depend on its past values) is extremely likely as this is the nature of infection. However, Blundell and Bond (Blundell & Bond, 1998;Blundell, Bond, & Windmeijer, 2000) show the way to tackle the estimation problem posed by the inclusion of lag dependent variables when there is time-persistency.
They show that instruments from both differences and lags can be used to make the instruments stronger under some set of moment conditions, and this method is called system GMM. It performs better than the difference GMM in a condition that matches with this study (potential time-persistency). Therefore, this study uses system GMM. Table 1 reports traditional difference in difference estimates to understand the effect of lockdown in the growth of infection. The DiD coefficient (coefficient of lockdown-post) is statistically significant for developed countries whereas this coefficient is statistically insignificant for the panel of developing countries. Once we include stringency of lockdown then the same coefficient becomes insignificant for both developing and developed countries. The variable 'lockdown' is a dummy of whether the country adopted lockdown or not whereas the stringency is more about the extent of it. While stringency denotes the government response to curb the spread of infection; other important aspect--the citizen's response to the governments' call--is also pertinent. The google mobility data can capture that attribute.

Effects of lockdown and its stringency
[ Table 1 will be about here]

Effects of lockdown with mobility and testing
[ Table 2 will be about here] As in Table 2, it is evident that when mobility data (here increase in the 'staying at home') is included in the regression, then the coefficient of post-lockdown remains insignificant for low income countries but it is still significant for developed countries (in one model). However, an increase in compliance factor i.e. staying at home appears to be effective in curbing the infection growth. Inclusion of tests level does not affect the coefficients of the regression drastically. As explained in the methodology section, the growth of infection in the current period might be largely explained by the previous periods and so it is important to use a dynamic panel data model to understand the true impacts. The following section illustrates the results obtained from the GMM estimates. Table 3 demonstrates the impact of lockdown along with other controls on the growth of infection. As expected, the previous periods' growth appears to be a significant predictor of the disease growth.

Effects of lockdown along with stringency
[ Table 3 will be about here] The after-lockdown variable, which is included to understand the impact of lockdown, assumes an unexpected sign for developing countries. The same coefficient, however, is statistically highly significant (p-value <0.01) for developed nations. Mobility has an expected sign but not statistically discernable. The trend coefficient is negative for both developed and developing countries though it is significant for developing countries only. Instruments used in this regression appear to be valid, as demonstrated through Sargan and Arellano-Bond, and Hansen tests.

Effects of income support
Although lock-down seems to have a little effect on curbing the spread of infections in developing countries, income support shows a strong impact. Table 4 shows that income support is statistically significant (p value<0.01) in curbing the spread of infection. Interestingly, the same variable is not statistically significant for developed countries.
[ Table 4 will be about here]

Timing of lockdown
The timing of lockdown can play a significant role in curbing the spread of infection. For instance, if a strict lockdown is enforced in the beginning, it might be easier to stop the growth of infection. On the other hand, enforcing lockdown too early may backfire if people do not take it seriously as there could be a behavioral issue related to the compliance.
[ Table 5 will be about here] For example, if cases are very low when lockdown is enforced, people may underestimate the actual risk which may affect the compliance as well the effectiveness of these non-pharmaceutical interventions. Moreover, people in developing countries--where the massive level of income support is not possible to render--will not be able to comply with stay home orders as a vast majority of them live hand to mouth, and they have to come out of home for some income opportunities just to feed their family. To understand whether the timing of lockdown is crucial, the timing of lockdown (lockdown after how many days since first cases) is categorized as lockdown within 15 days, within 30 days, within 45 days, and after 45 days or never. Regression Table 5 shows that the timing of lockdown does not seem to be important for developing countries as lockdown does not affect the growth of infection at all; however, the developed countries that enforced lockdown within 30 days of first case detection were able to contain the spread of virus significantly.

Type of closure and mobility
Various forms of closures have been adopted in multiple countries ranging from school closure to public event closure and almost complete shutdown of the international passenger traveling. While the stringency index provides an aggregate of all these measures, it is quite likely that all measures would not be equally effective. For instance, when there is no recorded infection, school closure may not be an optimal response; rather, screening international traveler, identifying and isolating them would be. Likewise, the types of mobility can also determine the impact of lockdown. For instance, a substantial increase in staying at home and a considerable reduction in public transports can help reduce the infection and slow down its pace. While stringency index measures government's response, mobility can be treated as the citizens' compliance in response to the government's measures.

Discussion
This study has shown that lockdown has been successful in reducing the speed of infection spread in developed countries but not in developing countries. In addition, not all government measures are equally effective. While stringency has a little to do in curbing infection growth, citizens' response appears to be extremely useful in this regard.
In both developed and developing countries , staying at home is found to be an effective measure to slow down the infection growth, and a similar conclusion is drawn from other studies related to COVID-19 (Flaxman et al., 2020;Yilmazkuday, 2020b). Most interestingly, the extent of income support has a substantial effect on reducing transmission of diseases in the developing world.
Several key issues of discussion have spawned from this study. First, government response teaming up with citizens' responsibility turned out to be instrumental in succeeding 'flattering of the curve'. While most of the draconian measures appear to have little or no effect in the developing countries, staying at home help reduce the spread. Developing countries cannot afford to provide continuous income support to the poor for an extended period; if it can be done; a reduction of infection's growth can be possible.
Second, while the lockdown is essential, the timing is also equally important. Surprisingly, developing countries that adopted lockdown policies within 15 days of first case detection were not notably successful in reducing infection compared to the countries that took lockdown after 15 days. Though it may sound counterintuitive, it is quite plausible. For instance, in Bangladesh, it was observed that people have already reduced the mobility even before lockdown is enforced, and so lockdown did not bring anything new in the table as far as the citizens' response is concerned. On the other hand, many people may not fully perceive the severity and the spreading capability of COVID-19 when the number is too low. If the lockdown is enforced 30 days after the first case, by the time, the population may see 50 to few hundred cases, and this may help them under the true risk of spread which again can reduce mobility and influence their behavior. And as the most of the countries did not experience an explosion of cases within 30 days, contract tracing, cluster lockdown instead nationwide lockdown and other instruments could have effectively been applied. This strategy would help keep economic activities alive without jeopardizing the population's health. Also, this can help reduce 'compliance fatigue'.
Third, although many developing countries were quick in enforcing lockdown, there was a significant mismanagement and lock of proper planning to implement it. For instance, declaring lockdown a few days early and allowing a lot of migrant workers to leave the city in congested settings have increased the risk of higher spread. Moreover, income support to the poor would be much less than 'enough' in the developing countries (Biswas, n.d.).
Forth, seeing the successful containment of COVID-19 in China, and getting recommendations from various international organizations, many governments of the developing countries went on enforcing lockdown without giving much-needed attention to other parts of the recommendations such as increase the capacity of massive testing, introducing contact tracing at scale, preparing the health system especially increase in capacity of isolation. Without all these measures, lockdown will not bring much health benefit, though the economic loss expected to be massive (BD economy loses Tk 33b every day during shutdown: Study, n.d.; Fraser, n.d.).
Some limitations of this study are worth mentioning. First, COVID-19 spread is still in place, so the current study uses the interim data, and so inference made here may not persist in the long run. Second, most of the dependent variables especially stringency related variables are not continuous, rather ordinal, which might have some effects on the results. Third, the inclusion of too many instruments may under-reject the null, and so that might have resulted in more insignificant coefficients for developing countries. Moreover, the presence multicollinearity among few variables are very likely that might have caused under-rejection of null. Forth, instead of estimating R0 or Rtthe traditional measure infection growth--the current study uses the growth of infection directly. However, interestingly the findings of this study match with studies where R0 or Rt used for developed countries (Flaxman et al., 2020;Yilmazkuday, 2020b), so this variable not necessarily a weak one.

Conclusions
The study shows that lockdown type measures are not highly effective for developing countries even though these types of means are very effective in developed nations. Nevertheless, staying at home order and income support programs and other social distancing measures are found to be effective for both developed and developing countries. In addition, the timing of the lockdown also appears to matter. This study does not suggest that lockdown should be enforced in the developing countries. Rather, it suggests that lockdown should be combined with contact tracing, extensive testing, income support for the poor, management of informal or migrant workers-to make the lockdown effective. Merely declaring lockdown, without other required measures, will hurt the economy without contributing much to reducing the growth of infection.

RCT-Randomized Control Trial
DiD-Difference -in -Difference

Declarations
Ethics approval and consent to participate: Only secondary and public use data is used and so ethical approval and consent to participate was required Consent for publication: Author fully agree to publish it in the journal. It is not funded project and single authored, no other consent was required Availability of data and material: All data, programme codes, materials will be available on request    Standard errors in parentheses * p < .10, ** p < .05, *** p < .01 Standard errors in parentheses * p < .10, ** p < .05, *** p < .01   Standard errors in parentheses * p < .10, ** p < .05, *** p < .01      Standard errors in parentheses * p < .10, ** p < .05, *** p < .01    Standard errors in parentheses * p < .10, ** p < .05, *** p < .01