The economic and human costs of imposing stringent lockdown policies have arguably been one of the most discussed differences between rich countries and low- and middle-income countries, in the context of COVID-19.
Several analyses argue forcefully that the strictest lockdowns will lead to significant hardship and deprivation for households and individuals who have few savings to fall back on for support while economic activity is suspended (Barnett-Howell and Mobarak Y-RISE, Sen Ideas for India, Teachout and Zipfel IGC).
How much of the effect on mobility and economic activity is due to lockdowns, versus the threat of the virus itself? Existing analysis, mostly in high-income countries as well as some low- and middle-income countries, shows that people begin to restrict movement prior to formal lockdowns, an effect that is increasing with the severity of the outbreak (Maloney and Taskin). Nevertheless, a common view is that lockdowns are the main factor affecting these outcomes in poorer countries. For example, Pronab Sen writes regarding India “[...] the real story is that the economic damage actually has very little to do with the pandemic itself, and is mainly the consequence of policy responses [...]” (Ideas for India). In India, it is difficult to disentangle these factors formally, because the lockdown was enacted suddenly in the entire country.
Indonesia offers a useful case study for this question, as it enacted lockdowns in a decentralized and gradual way, and never reached the level of strictness of other countries. Figure 1 shows a lockdown “stringency” index for Indonesia, compared to a group of middle- and low-income countries that imposed strict lockdowns, as well as Sweden, which imposed few restrictions.
In Indonesia, government measures were applied gradually throughout the country. In a nationwide address on March 15, the president recommended that Indonesians work from home if possible, and announced that lockdown decisions would be left to provinces and districts. He also mentioned economic measures and said that the “business world can carry on as usual” (Jakarta Post). On the same date, Jakarta and other regions closed schools. (Schools closed nationwide by April 24.)
In early March, Indonesia had slightly more reported cases and deaths per capita as India and Brazil (Our World In Data), while underreporting was widely suspected as the government tried to limit information about the virus (Human Rights Watch). (Today, these numbers have diverged significantly: there have been around 10 confirmed deaths due to Covid-19 per million in Indonesia, compared to 12 in India and 271 in Brazil.)
Several provinces enacted “large-scale social restrictions” (Pembatasan Sosial Berskala Besar or PSBB) over the following weeks. This process was gradual. The legal basis was established on March 31st, and it stipulated that applications for PSBB from regional governments must first be approved by the relevant ministries. For example, the government of Jakarta applied on April 2 and its PSBB started on April 10, while the application of a few other regional governments were initially rejected. Long distance train and air travel was also shut down at various dates during April. Mudik, the annual tradition when millions travel to home villages after the end of Ramadan, was formally banned on April 23rd.
Measuring Mobility Patterns using Smartphone Location Data
In order to study how Indonesians changed their behavior during the initial phases of the COVID-19 crisis, we studied movement patterns using smartphone location data obtained from the company Veraset. This type of detailed location data is directly informative about mobility patterns at different scales (within-day commuting, inter-regional travel), and indirectly informative about economic activity as mediated by mobility.
Our key results are that mobility patterns of smartphone users changed abruptly around the time of the first wave of recommendations in Indonesia. Within-day movement fell suddenly and stayed low. Inter-district travel saw a one-time adjustment away from urban areas in late March, followed by low and further decreasing levels afterward. By contrast, mobility was already low and changed little when formal PSBB restrictions entered into effect.
The data underlying this analysis includes location information from almost 60 million smartphone users in Indonesia for at least one point in time over the first five months in 2020. We restrict attention to a small set of users who are observed regularly in the data. The data only covers smartphone users, who are likely to have a higher income. In addition, migrants are 15 percentage points more likely to use smartphones than non-migrants for the same level of household expenditure, so they are over-represented in this data. Nevertheless, in urban areas in Indonesia, smartphone penetration is relatively high to begin with. For example, it is 62% in the Jakarta capital region, 56% in predominantly urban districts, and 42% overall (compared to 70% in the US overall).
Daily Movements Change Suddenly After Presidential Announcement
We first measure within-day movement by comparing where users appear during nighttime hours and during daytime hours, to obtain a proxy of commuting and other daily trips.
Figure 2 shows the daily fraction of phone users who stay at home on a given day, defined as no travel beyond 200 meters away from their nighttime location. Before COVID-19, this fraction ranges from around 50% in Jakarta to 60% on average nationwide.
After the president’s speech on Sunday, March 15, and at the same time as many other countries were introducing formal lockdown restrictions, the fraction starts growing and reaches around 80% two weeks later, where it stays until late May at least. As a benchmark, on national holidays mobility decreases further, with almost 90% of users staying at home. (In results not included here, the distance traveled each day shows a similar and opposite signed pattern after March 15.)
By contrast, there is no noticeable jump in staying at home when lockdown policies are officially introduced. This can be seen for Jakarta around the PSBB launch date on April 10. This suggests that changes in economic activity related to individuals traveling to their workplace, shopping in the city, or commuting for other purposes, had already taken place.
Figure 2. Daily fraction of users staying at home
The sample is weekdays. “No movement” is defined as a user who does not travel beyond 200 meters away from their location that night. The sample is 173,911 users who appear at least 5 days per month between January and May 2020. (Results are similar with a sample of ~1 million users seen in the data at least once each month.)
This change in activity took place in every province, and it was more pronounced in cities and their surrounding areas, as shown in Figure 3 for the capital Jakarta and for Palembang, a medium-sized city on the Sumatra island. After March 15, activity levels converged between urban and rural levels. In urban areas, smartphone users were overall more likely to travel away from home in the initial period. This gap disappeared in April and May. Life and activity in dense urban areas, which typically rely on frequent and distant commuting, seem to be the most visibly affected in relative terms by the pandemic.
In results not included here, we study whether the mobility change differed based on district-level average characteristics, obtained from the SUSENAS 2019 national household survey. We find very similar responses in district that differ based on average household consumption, as well as on the fraction of households in the district who reported in 2019 being worried about not having enough food.
Figure 3. The change in fraction staying at home by district (top). The Jakarta region (left) and Palembang region (right).
The change is measured as the average fraction staying at home after April 1st relative to before March 15.
One-time “Adjustment” in Inter-District Movement
While within-day movement captures commuting and other daily economic and social activities, travel between regions is also a key indicator, capturing a combination of business travel, movements of temporary or recent migrants between their home region and the region where they work, as well as social and family travel.
We use the smartphone location data to look at medium- and long-distance movement between 500 districts (kabupaten). Our measure uses the median nighttime location for a user matched to a district, and we define that a user has moved if on a given night they are in a different district than two weeks prior.
Figure 4 shows the changes in inter-district movement for the three largest cities: Jakarta, Surabaya and Bandung. Before March 15, travel flows between regions were balanced, with around 5% of the study sample being in a different district than two weeks earlier, at any point in time. Travel fluctuates closely around this value, with the exception of the start of the year.
After March 15, there is a large net outflow away from these cities, which lasts approximately two weeks. (This pattern also holds more generally for all urban areas in Indonesia.) Travel flows then fall to around 50-75% of the baseline value, and flows in the two directions approximately converge starting in April. The start of Ramadan (and the concurrent ban of most internal passenger air travel) coincides with a small uptick in movement.
These movements happen in the context of “mudik,” the Indonesian yearly return of migrants to their hometown or home village at the end of Ramadan. In anticipation of possible mudik restriction (which indeed later materialized), migrants may have preemptively left the large cities to avoid getting stuck.
Unlike India, where the suddenly announced and stringent lockdown trapped millions of migrants around the country and led to a major crisis, the near-convergence in movement flows in Indonesia and the lack of a nationwide lockdown suggests that most migrants who wanted to leave had the time to return to home regions. Understanding the impact of this large inter-regional movement on the spread of the virus is beyond what we can speak to in this analysis.
Figure 4. Movement to and from Jakarta
The sample only includes weekdays. It includes users observed on a given night, as well as at least on another night 8 to 14 nights prior. The user moved if she was in a different district on the earliest date when observed, 8 to 14 nights prior. We further impose for both dates that the user was present in the same district at least once in the following 7 nights. (Results without this restriction are qualitatively similar.) Movement is measured as a fraction of the number of users in each city.
Implications for Policy
At the onset of the COVID-19 crises, Indonesians with smartphones sharply restricted their daily mobility following recommendations from the president and other government officials, even in the absence of formal lockdown of most economic activity. This suggests that fear of the virus itself played an important role. At the same time, a one-time net outflow away from cities suggests migrants took this opportunity to return to home regions. This is consistent with migrants anticipating formal restrictions that might trap them away from home (as happened in India), as well as a sudden decrease in job prospects in cities.
Behavior at the beginning of the crisis does not automatically tell us how people are reacting to the current situation and how they will act in the future. On one hand, the novelty of the threat may have mobilized people to put safety ahead of economic activity, while by now, depleted economic buffers may change the risk calculus for many Indonesians. This may be especially true for lower-income workers who may be under-represented in our data. Moreover, a false sense of familiarity with the virus may also attenuate the initial fear. On the other hand, Indonesia now adds over 1,000 cases every day, a number on an increasing trajectory. COVID-19 is not going away, which could keep an important share of the population from returning to their previous level of activity.
These results strengthen the case for active government action to support people most likely to be affected by the continued economic disruption, with or without active government lockdowns (Hanna and Olken in EPoD). They also suggest that behavior may quite sensitive to non-binding public recommendations as well as the information available to the public.
Arya Gaduh, Rema Hanna, Gabriel Kreindler and Ben Olken
We are grateful to Paolo Adajar, Aaron Berman, and Colley Winda for exceptional research assistance, and to the company Veraset for sharing the location data.