What explains the second wave of COVID-19 cases in India?
Part 2: Changes in Behavior
India is experiencing a second wave of SARS-CoV-2 cases. In the previous post in this series, we explored the idea of a specific geographic origin of a new variant causing a new wave of infections. In this post, we explore some of the ways behavioral changes intersect with infection dynamics and public policy to shape how a pandemic progresses.
II. Behavioral response
A second explanation for the March spike is individuals have begun to take fewer precautions to stop the spread of COVID. There may be several reasons for the change in behavior.
One is lockdown or COVID fatigue. Perhaps people are slower to respond to increasing risk because they are tired of voluntary or mandatory COVID precautions such as staying home and wearing masks. Under this explanation, over time random small outbreaks become bigger outbreaks because people are slower to adjust to risk. This sort of behavior would waves of steadily growing amplitude. The problem with this explanation is that we do not observe rising waves; instead we observe a singular large wave starting in February or March.
A second reason is that individuals started taking more risk for some reason just before the start of the second wave. For this explanation to work, we have to have some triggering event. We can think of two, but there may be others we have not thought of.
One possible trigger is that the expansion of vaccination made people believe that it was safe to let their guard down. While India’s vaccination campaign started on January 15, it restricted eligibility to frontline workers until March 1. In addition, the government has announced ambitious plans to vaccinate 300 million persons by July. This explanation is a version of what is sometimes called the Peltzman effect or risk compensation.
Another possibility trigger is that there was a jump in migration around February, perhaps due to the change in elections or change in agricultural seasons. Maharashtra conducted village level elections in 50% of the villages in the state across all districts other than Mumbai. The elections saw a turn out of 79% (reference). Pradeep Awate, Maharashtra’s chief epidemiologist, attributes the current surge in the state to three factors, one of them being people from cities returning to their villages for elections (reference). While this explanation might plausibly explain the outbreak in Amravati, an open question is how it spread nationally so quickly.
In the remainder of this post, we’ll break down the behavioral explanation into two parts. One, could behavioral change be responsible for the second wave? Two, if behavior is responsible, which specific behavior is responsible?
A. Is behavior responsible?
Before we critique the behavioral-change explanation, it is useful to note that behavioral response to risk must have played a critical role in the COVID-19 pandemic. First, lockdowns are perhaps the best evidence. Although people argue whether the lockdown or voluntary social distancing was effective in reducing activity and infections, the fact is that lockdowns are actions taken by people: people in government choose to implement them, and members of society choose to comply or not. They are a species of collective response.
Second, look at a plot of Rt for any country or state, including India, and you will see an extended period of time where the value of Rt fluctuates around 1. Contrast that to the usual compartmental models (e.g., SIR models) that assume contact rates are constant over time. The SIR models have a susceptible curve that looks like a backwards S, i.e., susceptibles start at a high level but gradually decline and then plateau ; the implied reproductive rate, which is proportional to the fraction of the population that is susceptible (Rt = (/) S/N), should monotonically fall, hitting 1 only briefly when S/N=/. The only way to reconcile this is changing contact or recovery rate or the possibility of reinfection, i.e., behavioral change, medical innovation or a different disease pattern.
There is, however, a behavioral explanation for why the epidemic hovered at around Rt = 1 for a long time. Economics suggest that individuals increase their risk-taking activity when infection risk is falling (~Rt < 1) and decrease their risk-taking when infections are rising (~Rt > 1). This theory, attributable to Josh Gans, can explain a good deal of the data across many countries.
That said, behavior alone is unlikely to be a full explanation for the epidemic. The “Rt tends towards 1” behavioral model has difficulty explaining the second wave in India. In several places, we observe Rt > 3 and for multiple weeks. Behavior should not deviate so much from Rt = 1 under the behavioral model, right?
One possibility is that we are looking at the wrong outcome: infections. People are not concerned about infection per se; they are concerned about death. Perhaps infections are rising, but deaths not as much. The reason is that we are using the wrong epidemiological model. The standard compartmental models (SIR, SEIR) assume that recovery is permanent. But immunologists will tell you that the immune system is inside the body, not outside it. It is not a shield against infection; rather it helps you clear infection faster. This does ultimately reduce infection, but indirectly because it reduces the probability of infecting others by shortening the infectious period. The solution is to use a modified compartmental model that allows for reinfection, but with a lower death rate. The bottom line is that perhaps we can rectify the behavioral model by using a model with reinfection and suppose that behavior adjusts to keep the change in death rate hovering around 0 - equivalent to a reproductive rate for COVID death hovering around 1 during an epidemic.
All this is fine in theory, but what does the data say? Is this new behavioral story borne out in the mobility data? Is it borne out in the death data in India? We tackle mobility here, then mortality in a separate section.
Unfortunately for the behavioral theory, there is no direct evidence to support it in the mobility data. Figure 2 plots changes in Google mobility in India around the time of the outbreak. We don’t see a spike in mobility around the beginning of March. Indeed, mobility is declining during February and March. Remarkably, mobility peaked before vaccination -- around the beginning of January. This does not rule out behavioral change: it is possible that the drop in March is people being extra cautious starting at the introduction of vaccination because it was clear when the need for distancing would end. But that sort of behavioral story does not explain India’s March spike.
Figure 2. Mobility trends in India and Maharashtra.
One might interject that perhaps Google mobility data doesn’t measure behavioral change well. Perhaps it only picks up behavior by high socioeconomic status persons, or the sort of person that uses Google maps a lot. It certainly does not pick up mask-wearing. Perhaps Google mobility doesn’t measure behavioral response and thus cannot be used to reject behavior change as an explanation. Perhaps these complaints are true. But then we are stuck with little direct evidence to support the behavioral explanation of the second wave.
One important piece of evidence would be mortality rates. Were mortality rates declining, justifying greater risk taking the eyes of the population? On this we are still waiting for an answer. The second wave is still fairly recent; it started in Maharashtra in mid-February and the rest of India in mid-March. Deaths are a lagging indicator of cases. And death data have their own issues.
It turns out that whether the behavioral theory or the new variants story is correct both depend on the death data. So we will return to the issue of the case fatality rate when we discuss the new variants explanation for the second wave.
B. What caused the change in behavior?
Recall that there are 2 main triggers that have been proposed. One is vaccination policy; the other is migration, perhaps for elections or agricultural work.
The argument for why vaccination policy may be the trigger for greater risk-taking is threefold. First, in other epidemics, such as HIV, there is evidence that reducing risk increases risk-taking. In economic terms, activity has benefits and costs. The benefits are that one may enjoy the company of others and that interaction is necessary to earn income or consume. The cost is the risk of dying from COVID. Individuals engage in activity until the marginal benefit equals the marginal cost. If vaccination reduces the marginal mortality risk, then people will engage in more activity.
Second, the timing of the second wave is telling. It did not occur in December (after the first wave, but before vaccination) or even January (when only health care workers were eligible for vaccination). It occurred at the beginning of March, when India announced it would expand vaccination to anyone over 65 and those over 45 who had certain comorbidities.
Third, several countries (in Europe and Southeast Asia) started experiencing new waves around the time that India did. This coincidence suggests that vaccination rather than migration as the triggers, as other countries do not have migrations similar to those in India. Moreover, the timing rules out seasonality, i.e., people huddle indoors when it is cold in the North and hot in the South, as the timing of seasons in India differ from the timing of seasons in other countries.
One problem with the vaccination story is the post hoc ergo propter hoc fallacy. Just because one thing comes after another does not mean the second was caused by the first. To validate that vaccination was the cause of increased risk taking, we would like some surveys tracking, for example, consumer confidence and showing that confidence was correlated with vaccination. Even stronger would be some evidence of greater risk taking in those places with greater number of vaccine-eligible populations or places where vaccination rollout was faster. Even though India’s campaign is national, there is regional variation in vaccination.
But we don’t have such evidence. Instead we have surveys that suggest little demand for vaccination. These don’t necessarily rule out the vaccination story: it could be that demand fell because of free-riding on other people getting vaccinated. However, they do not provide direct evidence for moral hazard either.
A second problem with the vaccination story is that it doesn’t show up in the mobility data.
Perhaps migration was responsible for at least the start of the second wave. Before we dive into this explanation, it should be noted that the lack of evidence in the Google mobility data is not a strike against the migration theory because Google’s data picks up movement within a district, not migration across districts.
There is also some direct but informal evidence in favor of migration. While the Indian Railways said there would be no special Shramik trains to take migrants home in April, the central government-run railway did increase the number of trains it was operating on a regular basis. (The difference is that the Shramik trains run in 2021 were free, while regular trains are not.) In addition, it does seem there has been an increase in return migration to Northern belt states. (Our information comes from discussions with state government officials.) Unfortunately, there is no good quantitative data (we know of) on train and bus travel across district and state borders.
Even if we suppose that migration is a cause, a natural question is why there is greater migration just now.
Agricultural migration. One possible trigger for migration just before the second wave is seasonal migration for agricultural work. India has two agricultural seasons: kharif and rabi. Rabi crops are planted in mid-November and harvested in March-May. Perhaps there was an increase in labor migration from cities or towns to rural areas for the harvest. This movement increased crowding on trains and buses and perhaps even spread infection nationally. (There is also migration back to cities after, a possible future concern.)
While we do not have direct evidence on the amount of movement, we have some evidence that the national lockdown in 2020, which started March 24, had blocked labor migration for the 2020 Rabi harvest. Moreover, we can look at the timing of wheat output, one of the major crops of Rabi, to determine timing of production. Unfortunately, it is difficult to get month-wise wheat (or other Rabi crop) production. So we rely on two indirect measures: wheat prices and emergency wheat stock held by the government. You can see the timing of price dips (since equilibrium price falls with increased supply) in March and surges in government stock in May or so in the figures below from a US Department of Agriculture paper on India’s wheat production. Since harvesting occurs before production hits markets and go-downs, the timing is consistent with agricultural migration just before the second wave.
Election. A second possible trigger for the second wave is elections. Maharashtra had gram panchayat (village level) elections on January 15 with 79% turnout. There was a concern, voiced by the state’s chief epidemiologist that many people had return from cities to their villages to vote in that election. This was followed by an outbreak in Amravati in the second week of February.
But we have some questions about this explanation. First, the amount of movement for election is likely to be smaller than for seasonal agricultural migration. Second, the elections in Maharashtra took place all over the state, not just in Amravati district, which is the northeast part of one of India’s largest states. Third, other states had elections around this time and did not see outbreaks around the same time. Himachal Pradesh had village and town council elections on January 17-18, Rajasthan has municipal and town council elections on January 28, Andhra Pradesh had village council elections on February 9, 13, 17 and 21, and Gujarat had municipal elections on February 21. Why did Amravati start first? Finally, is it possible that elections triggered an outbreak 3 weeks later?
It would be difficult to demonstrate one way or the other that elections are responsible, perhaps through the mechanism of migration, for the February-March start of the second wave. Therefore, we do not put tremendous stock in it.
Reverse causation. Before we conclude our discussion of migration it is worth noting that, instead of migration causing outbreaks which require lockdown, it could be that lockdown causes migration that leads to outbreaks. I.e., it is possible that the cure is worse than the disease.
Back in 2020, a concern that was raised was that the March lockdown interrupted the seasonal migration. Then, when the lockdown was eased on May 4, 2020, there was massive return migration as people sought, in part, to villages where they might otherwise have been in a typical year. Some worried that this may have spread the epidemic to rural areas in May and June. In effect, the lockdown changed the timing of migration and thus of the epidemic. To be clear, we do not blame the lockdown for the fact of an outbreak. If the lockdown had not happened, maybe cases would have peaked in India earlier. But it is reasonable to suppose that lockdowns (and their easing) affected the timing of the outbreak.
Moving forward to 2021, this logic suggests that, even if migration did not cause the second wave, it is possible that the fear of a lockdown triggered early migration in March that compounded the second wave that had already started. This may have hastened the peak of the second wave. We do not have data to know for sure. But it is important to consider the causes of migration, and outbreaks may be both a cause and effect of migration.
There remains a great deal of uncertainty about what role behavior played in the second wave. However, its relative important depends on the role of new variants and data on deaths, which we will consider in our next two posts.