 So, welcome to the last session of the conference. In this session, the question that presenters will try to answer is, how do we associate with COVID-19 affect people's behavior in a global South? So, you have wonderful papers from Indonesia, for example, and then Peru. And then you have two different sets of presenters. The first one is other ones doing experiments, which measure actual and observed behavior. And then you have the standard surveys, where respondents state their preference and how they react to risk associated to COVID-19. So, I hope to start with Magat, who will present precarity and redistribution. It's a survey, I believe, so. Hi, my name is Janaka Magat, and today I'm going to present shared work on precarity and preferences for redistribution in weak states. In particular, we provide evidence from the Philippines. There's always been a puzzle in the political economy of redistribution with regard to the different implications and results in studies in the developed world and in the developing world. In particular, while studies in advanced industrialized nations show that labor market risk is related to social policy preferences divide, in particular, low-risk workers tend to favor employment-based social insurance, like pensions, whereas high-risk workers prefer non-contributory and needs-based social assistance. This link is not present in the developing context. Similarly, while risk is predicted to increase support for redistribution in the developed context, we don't see this relationship in the developing world as well. So, part of the reason for that is precarity is proxied by labor market status. In particular, the formal sector is considered low-risk and therefore not precarious because of its high levels of employment protection and secure benefits, and the informal sector is high-risk and precarious because they lack social protection and they lack secure social benefits. But informal is not always precarious, especially in the developing country context. So think about independent contractors. They are formal because they pay their taxes, but they're also precarious because they have short-term contracts and they have to look for contracts every so often. Meanwhile, your stores, like this is what we call a sorry, sorry store, it's informal, they're not registered with the state, they do not pay taxes, but they're also not precarious because they have a loyal customer base, they have relatively high incomes or high revenue, and they even have large connections with formal firms that deliver goods and services to them for distribution. So, we show in this paper that precarity is actually distinct from formality because formality is essentially defined by state rules and regulation, whether work is registered with the state or not, whereas precarity is based on employment conditions, whether work is uncertain, unstable, and insecure. So now that we fix the definition of precarity, we want to understand what really is the effect of this precarious condition on preferences for redistribution. So first, we want to see whether precarity actually does lead to social policy divide in particular, whether precarity leads to less support for social insurance or more support for social assistance, and if precarity leads to preference for more redistribution. So in order to answer this, we did an online survey of 1500 individuals in Manila, Philippines, which is 17 contiguous cities and accounts for 40% of the country's GDP, it also has almost 20 million people. Respondents were recruited through Facebook ads and they were offered some compensation upon the completion of the survey. What is interesting is that since we cannot experimentally randomize precarity, we leverage the current pandemic as a shock to precarity with the idea that people who are more aware or are affected by COVID are more likely to say that they are precarious. So because there's imperfect compliance in the sense that people who receive the prime may not necessarily say they're precarious and people who have not received the prime can also be precarious, we use a two-stage least squared approach with COVID prime as the instrument. So we estimate these models and our estimate parameter of interest rather is data one. So we find that that precarity actually leads to preference divide. The more precarious you are, the less likely you are to support social insurance. But it does not have an effect on preferences for redistribution. And that's partly because there's an expectations gap. In developing countries with truncated welfare state, people do not actually expect to benefit from redistributive policies. So even if they prefer one policy over the other, they're not likely to demand policy changes. And this effect is moderated by informality in the sense that because informal workers are excluded from social insurance because they do not make traditional contributions, actually does not really have an effect on their demand for such policy because regardless of precarity, they should just always or at least be indifferent with social assistance. So we show the precarity is distinct from formality and that preferences do change in response to shifts and risks in the developing country context as long as we measure precarity properly. Contra expectations formal workers do prefer non-contributory social assistance as when they are precarious and that informality dampens the effect of precarity on redistributive preferences. Thank you. Thank you for your presentation. Now I turn to Belinda. OK, thank you very much Clarice and everyone. Let me share my slides very quickly and then I will set my timer for five minutes. Belinda will present something on interventions and communication tools and OK, thank you. Can everyone see my slides, please? Yes. OK, awesome. All right. So five minutes. Let me go. So this is John work with my co-author Francis and Anna, Georgia State. And we really start out with this puzzle saying, look, you know, we're in a pandemic. During the pandemic, there's been a lot of interventions regarding lockdown. And kind of imagine when you are in lockdown, especially when you're in these emergency situations, you need to communicate with friends and family. You can't do it. You want to make a phone call. You want to use the web. You want to access social media. You know, you have these sudden needs to communicate and you have these barriers to communication, especially if you are in a situation where you don't income and you cannot easily buy, for example, phone credits. So we said, OK, what is the benefit of communication interventions like giving people phone credits for a number of economic outcomes, especially the game of mental health? Should communication interventions come as kind of one time large transfers or in many small trenches? And it's a fact that, again, as a result of COVID, you know, throughout the world, many countries have been passing these communication interventions where, for example, AT&T gave free time gigabytes of data for 60 days, the government of Ghana reduced the communication service tax from 9 percent to 5 percent. So all these programs, lots of them have been kind of proliferating throughout the world. And yet there's very poor evidence on the impacts of these programs, these kind of new communication interventions on people's economic outcomes, especially during the pandemic. So what we noticed in Ghana where we had work on going was that in 2020, right, so when the lockdown hit, if you look at people's purchases, every other household purchase kind of declines during the lockdown in around March 2020, except for these are low income consumers. So except for mobile credits, right? So people were purchasing more of these phone credits, mobile credits. And that was going on, again, among very, very poor consumers, while every other expenditure category was going down. So we said, okay, let's look at this in detail. It seems like people are really valuing these kind of communication credits, these phone credits, a lot. And especially, you know, this is a context, our context is in Ghana. We have very high mobile cell phone subscription, about 134 percent in Ghana. So we went into this nationally representative sample, and we essentially said, let's do an RCT where we assign people phone credits for free, these low costs, very cheap $7 per person, phone credits for free, and see how that impacts, especially mental health outcomes. So this is the intervention. So we give people 40 Ghana CDs. This is around $7 per person. We have two treatment arms. So one is a lump sum treatment arm, where we say we give you a one-time transfer of the $7 in mobile phone credits. So again, these are not cash grants. These are phone credits directly. The other treatment arm is installments. So we say we give you two installments of about $3.50 worth of phone credits each. And of course, in the control, nobody receives any phone credit. And we partner with a major telecom company to do this. So one way the outcomes were interesting, and we wanted to understand, again, people are unexpectedly confronted with the need to call, especially during these emergency situations like lockdown, during COVID, or in emergencies more generally, how does that affect their mental health? So we use this a very well-known psychological measure in the psych literature called the Kessler Psychological Distress Scale, which is a measure of mental distress. It asks people things like, in the past seven days, have you felt depressed? Have you felt unable to get up from bed, et cetera? It's a value that goes between 0 and 50. And if you have a value greater than 30, that is a sign of mental distress. We also look at domestic violence outcomes. So we basically ask subjects whether or not they have threatened or hit their partner in the past seven days. So our treatment, again, very straightforward. We are doing an RCT. So this is a very straightforward OLS framework. And we are basically controlling for some individual controls using LASSO to make sure that we are not doing key hacking. We do a bunch of no-bus checks for attrition, et cetera. And our results are very robust. So these are the results. So what we found was that in the treatment arms, where people are getting these phone credits, again, very low costs. This is $7 per person. We start significant decreases in incidences of mental distress by about 9.8% relative to the control. Significant decreases in the incidences of severe mental distress by about 26%. And also very interestingly, a decrease in the incidences of domestic violence where partners are working to hit their partner, or threatening their partner more generally, fell by about 6.3%. The effects were much stronger in this installment arm where people were getting these phone credits over time and in the lump sum arm as well. And we also saw that the effects were concentrating among the extremely poor households, individuals in the informal sector, and women also experienced slightly better improvements in mental health as well. We also saw, maybe not surprisingly, that people were able to say, hey, now I can make calls. They kind of loosened their communication constraints. They were reported being more able to make calls, being able to borrow SOS air time, seek digital loans, et cetera, as a result of these transfers. So in conclusion, low-cost communication interventions, very, very low-cost $7 per person can have these very significant effects on mental well-being. And we think that this is something that we should be looking at as a policy perspective, going forward, especially since, you know, thinking about the value of ICT for mental health and its economic outcomes. Thank you. Thank you, Belinda. It's a very interesting paper. I like the design. Now we go to Hillamy where he's going to present the sort of a field experiment on beneficiary characteristics in the COVID-19 Mutual Aid Platform in Indonesia. So... Hi, my name is Machur Hillamy, and I'm going to present my joint project with Regent Linn from the University of Hong Kong and UNICEF Corrianto at Nanyang Technological University. This is a project where we embed an experiment on an online giving platform built to provide mutual aid for people affected by the COVID-19 pandemic. So, as with many countries, Indonesia was hit hard by COVID. Economy slowed down. A lot of people lost their jobs, lost their incomes and their government, while they have been stepping up their social protection responses, but a lot of gaps remain. And it is this gap that individual donors are trying to provide and step in. So, the main question that we're interested in is that we're interested in what does the algorithm decision process look like? And specifically, in this project, we're asking two more specific questions. One is very specific on the impact of the choice site. We want to see what is its impact on the donor decision making. And we also want to see what kind of characteristics from the beneficiary that attracts the mission from individual donors. So, as I've mentioned, we embed an experiment in the platform where we randomly vary how many beneficiaries are displayed with the potential donors at a time, but other than that, we also, the beneficiaries are also randomly selected from the database and we also run an online survey. However, due to the time constraint, I will be focusing on the first experiment. So, the platform that we're partnering with is called Vajirata. On this platform, workers who are losing their income or losing their jobs can sign up to receive money. And at the same time, individual donors can come to see the, can come to the website and see randomly selected beneficiaries and they can donate as they like. They can directly send money through ePayment platform. So, this might remind you of other websites like Govanme or DevDirectly, but in our study, the donors actually give directly to the beneficiary for other beneficiary like Kiva. But for Kiva, that's a loan, whereas in our study, it is pure charitable transport. So, we randomized the donors to see either three recipients at a time, eight or 10. And at the bottom, if they can either refresh to get a fresh set of beneficiaries of the same number, or they can donate to any one of, or several of the beneficiaries that they like. So, this is one of the cards that they see. So, each beneficiary will get a single card. On this card, they will see the beneficiary's name, what are their occupation, their location, some narratives that relate to their social media, the details of their ask, and also the payment channel through which the donor can send the money to the beneficiary. So, what do we think we will see? Broadly speaking, we think that the donors may face some choice overall. They would also be considering certain groups more variably, such as people with closed social systems or people in their group. Or there might be other considerations like the amount of money that they ask or whether they've been laid off or whether they've received some beneficiaries' needs or deserve more of their aid more than the others. So, what we are estimating is pretty straightforward because it's a randomized experiment. We will regress the donation outcome on treatment group indicators. So, this is the sub-site where we set the PEN as the PEN beneficiary at a time as the control group because that's what the website has been running before we introduced this experiment and then the results are presented here. So, what we see is that there are more donors. So, when we show fewer donation targets like three at a time compared to 10 at a time, we see more people are donating. And because more people are donating, then we have a lot more than we have a higher average donation. It's nearly doubled, actually. And for each of the donations conditional on the donors themselves donating, we don't see any significant change in terms of the amount that they're donated. So, this is roughly just a little bit more than $10. Why do we see this? We see that the donors hit more refresh. The donors who see the viewers, they hit more refresh, which we think is indicative of them having a more fine tune control of their information set so that they don't get overwhelmed. Especially, we see that compared to the donors who see three beneficiaries at a time, compared to 10 beneficiaries at a time. But overall, we don't see the total of the beneficiaries that they see still far fewer than people who have done the massive incentive. So, that's all that I have time for now. I'm looking forward to your comments. Thank you for your presentation. Now, we turn to Salome Mokwa, who will present her paper on stockpiling in COVID-19 and CITM of African countries. Salome, you may start. Hi, my name is Salome Mokwa-Mensa, and I am a second year PhD student at the Lidio University of Technology in Sweden. So, I'm presenting a paper I'm writing with some colleagues entitled, Stockpiling and Food Worries, Changing Habits and Choices in the Mist of COVID-19 Pandemic. In summary, this paper investigates the effects of the concern about the lookout spread and economic impact of COVID-19 on the change in the amount of food and necessary supports in 12 sub-Saharan African countries. So, we used two waves of survey data by Geopool for these countries, and we used the multi-neutral logits in Mystimix and Amodo. We find significant effects of the concern of COVID-19 on the change in packet size and food boards. We also find heterogeneous effects across gender group and rural urban divide. The contents of COVID-19 might be promoting stockpiling behavior among those with no food worries. COVID-19 has the potential of affecting all the pillars of food security proposed by the FAO. The control measures put in place to restrict the virus, that is the restriction in movements, causes farm labor shortage, which in tenders traps harvesting, and hence the amount of food that enters the markets. The closure of food markets, like schools and restaurants, causes food waste, which is a threat to food worries of vulnerable people, and it also affects the size of food packages that individuals buy. There is also the likelihood of stockpiling. So risk-averse people stockpiled to make themselves believe that they can hedge against the risk of shortage, which causes shortage, increased prices, waste on given food distribution, the break in food supply chain. However, it reduces the frequency of shopping, and hence limits the COVID-19 exposure rates. So here is the empirical model that we estimate, where the dependent variable is the amount of food and necessities, but X captures the control variables, and the units of analysis is the individual and not households. The variable COVID-19 is defined by the concern about the local spread, and concern about the economic impact of COVID-19. This table presents the heterogeneous effects that we find. So the COVID spread concern and the economic concern of COVID would have significant impact on female than male. So the probability of buying smaller packet size and bigger packet size in relative to buying the same size would increase when the concern of the local spread of COVID increases. However, among the females, the probability of buying bigger packet size only increases when the concern of economic impact of COVID-19 increases. Among the rural and urban cities, the effects of COVID-19 for both the local spread and economic concern is more significant among rural cities, and they increase the probability of buying smaller packet size in relation to buying the same packet size. This table also presents the relationship between food worries and the concern of COVID-19. So here we see that the concern of COVID-19, which is the spread and economic impact of the virus has significant impact on those who are not worried about food. And here, the probability of buying more or buying bigger packet size here increases significantly among those who are not worried about food. In conclusion, we look at the implications of food supply, welfare of food, deprived individuals and future research. So for developed countries, food waste is mostly low at the consumption stage relative to developed countries, but this is usually high at the production level, processing and transport stages. In the short-term, stockpiling might increase the availability of food for households. However, in the medium to long-term, it might cause perpetual food shortage in economy. This can lead to high prices and increase food worries for deprived homes. Government can introduce some interventions that can free up resources for especially the vulnerable people. Thank you. Thank you very much, Salome. Now we have the last presenter, Alan Sanchez on domestic violence and lockdowns in Peru. So hi everyone, thank you for the opportunity. So this is a study that we did with colleagues from Oxford University in Peru. So as part of a longitudinal study that we have been doing for a while now. So the context is that we are collecting data as part of the young lives longitudinal study. And then this study collects data in Peru, Ethiopia, India and Vietnam. And in 2020, we're planning to go to a field for a face-to-face survey. Well, obviously we couldn't do it, but we had everything ready, including up-to-date contact information of the cohort. And we are basically tracking two cohorts, the younger cohort that was born in 2000, between 2001 and 2002 and the older cohort that was born seven years before that. So it's been 20 years that we have been following these two cohorts with low attrition rates. And so we were ready for a new visit to the cohort. And so COVID happened. We changed and we moved to a phone survey approach. We did actually a new survey based on the focus on the COVID. And one of our, I mean, the survey is actually quite long but one of the modules, the idea was to measure whether there had been an increase in domestic violence during the lockdowns in Peru. And one thing to say is that we were actually able to find about 90% of the sample that had a phone number. And actually, which actually represents 80% of the full sample. So there is some attrition, but yeah, it's unavoidable. And the other aspect to mention is that since this is a longitudinal study, we had data from before, including exposure to domestic violence in the past. So we kind of have a baseline. And also another thing to say is that the sample, although it's not a representative sample, it's a sample that is a random sample of districts in Peru. It has districts in urban and rural areas in all the regions, well, in all the climatic regions. So it contains a lot of information, but it excludes the 5% wealthiest districts in the country. So it's a proper sample. You can see here as a context, the evolution of the COVID cases last year. So we had started from April, from April to end of June, a national lockdown. It was a very strict lockdown with the state of home requirement. And then after that, we moved to a regional lockdown, but it was quite many regions that were actually in lockdown during this period. And it was a quite a tough period. People were allowed to go out to buy food, medicines, and to work for those economic sectors that were actually had the permission. So we collected the data for this study between August and mid-October, kind of in the middle of the regional lockdowns. We asked retrospective questions about things that had happened during the lockdown, including domestic violence. But bear in mind that it was actually a long survey. So we were not able to ask too many questions about violence. But also in addition to that, it's a sensitive thing to ask during a phone call because if something happens, you are not able to actually activate a protocol because you are not even able to move out of your house. So for this reason, for other ethical reasons, we tried to find a different approach to measure the potential increase in domestic violence. And the way we did this is using something called least randomization, which is something that has been used before obviously there is a long literature. And in this case, even in this case, we applied to this specific context. We did it as follow. So you have the sample of people that you are aiming to contact. You randomly choose a control group and a treatment group. For the control group, you basically mention four statements, such as for instance, during the lockdown, I was able to spend more free time than previously doing exercise and so on. So out of the four statements, you ask with how many statements you agree with, not with which specifically, but with only with how many. That's it. And then for the treatment group, you ask the same four statements plus a new one. And that one is I was physically hurt more often by someone in my household during the lockdown. So this is actually the key aspect for us. And because the two groups are balanced, we expect that the difference actually gives you the actually the prevalence of the increase in domestic violence during the lockdown. We did actually a double list design meaning that the control group, there was another listing which the control group was a treatment group and vice versa. We did a lot of piloting to make sure that there were no floor and ceiling effects. And as a result of that, we found that about 8% of the sample reported that there was an increase in domestic violence during the lockdown. We actually were not able to find any difference by gender. The key difference we found was that those that had reported being victims of domestic violence in the past four years ago for them, the increasing domestic violence during the lockdown was about 20%. That's one of the main findings. So that's it. That's those are the main findings. But also we think that this provides, I mean, our contribution is more on two aspects. On the policy side for Peru, it was actually important to measure the increase in domestic violence during the lockdown because it was quite a very long lockdown and people didn't have too many ways to actually look for help. But also on the methodological side, this is also a contribution because you can use this tool to ask about many things that are sensitive to us during a foreign survey. The negative, the limitation is that you can only ask about a few things, right? Because of time consideration. The last thing to say is that these studies since we submitted has been actually published in the social science as a sample population health journals so you can actually find more information there. Thanks a lot. Thank you, Alan, for your very nice presentation and well-designed paper. So I will start the Q&A and if you have questions, either you type it in the Q&A box in the session tab or you open your video in the audience as the presenters themselves. So we have Stephen Milo. Stephen, do you want to address your question to Salome and Belinda? Or do you want me to read it, Stephen? Okay, I'm going to read it. So let's start with Belinda. Stephen Milo asks, do you think that the effect you measured is short-term or is it a long-lived? So thank you for the question, Stephen. The way we organized the experiment, the first one, the first treatment was a lump sum. So what we were finding is that in the lump sum treatment, the effects are smaller, right? So the kind of short answer to your question is that it depends on the nature of the intervention. When you have the repeated treatment, so the treatment where we had two installments over time, so this was between September and December of 2020, then you saw that the effects were much stronger and you think last much longer. So the policy response there would be to have repeated installments of these phone credits and not just the lump sum transfer and then you get these longer-term effects. And then we also have from Stephen again to Salome sample size. How did you sample? And why individuals and not households? He said that food is a joint household expense in most Sub-Saharan, Sub-Saharan African countries. So he didn't understand how you measured worry about food, how it's measured. So you see us two questions. First, I would like you to address the sample size. One, and then the second question, is it individual? Why is it individual, not household? And the third question, how do you measure worry about food? So take your time with those three questions. Okay, thank you so much. And thank you, Stephen. So concerning the sample size, we focus on individuals and not household because we sample by phone, the text message. So Geopool was the one that sent out the questions. So the individuals will have to answer and they are answering individually and not the household. And of course, food is a joint household expense, right? But we are looking at how, like what the individuals think about it. And also food worries. So whether you worried about food or not, we measured by your resources, whether you have money or resources to buy food. So if you have the resources to afford food or to buy food, then you are not worried about food. But if you don't have the resources, then you are worried about food. I hope I answered your questions enough. So any other questions, please? Hey, Clarice, I think there is a question for me on the chat. Oh, okay. Yes, not on the Q&A, but yeah. Well, that if you're alone. To talent paper. Okay, rich and timely data, good design and highly relevant. We probably agree I read your paper in half. But she asked, is whether your interviews give insights on what are the factors that drive domestic violence, like factors like unemployment and whether the intensity of violence is more severe for a given group. And then third question, like Salome, what is the proposed policy intervention for this? So, Alan? Thank you, Clarice. And thank you for the questions. So we did actually explore potential heterogeneous effects. To be honest, we didn't actually pre-register the data analysis, but I can tell you that we were not able to actually find a specific driver. But I need to really mention to stress that this was actually a national lockdown. The entire country was in lockdown. There was also this period of freedom of lockdowns, but for almost three months, the entire country was in lockdown. And so it means many people were actually unable to go to work. And actually there were actually, the level of depression and anxiety reached 40% for depression, 30% for anxiety in the entire sample. So everyone was affected in a way. But the main heterogeneous effect is the one that we show that those that were affected by violence in the past were the most affected. That was actually the key finding. Because of what I mentioned, we could, because the lockdown was very extended, we were not able to actually measure differences because difference were, that you were in lockdown either three months or four months. So the marginal month didn't actually make a difference. Maybe there might have been a difference for those affected by, let's say 15 days versus four months. But comparing those affected three months to four months, we couldn't find any difference. And about the proposed police interventions, well, actually, I have to be honest that our contribution was mainly on the methodological side but obviously for us it was the key thing was, how can these people ask for help? Because at that time we had a system for a helpline but many people were actually saying that it doesn't work when you call no one answers. So actually more than a specific policy intervention that we proposed, it was like more evidence to push on the idea that these helplines need to scale up and become available for everyone and through there we also tried to influence policy. Thank you and very nice answers. So Salome, I have another question for you from Felix Dade. He asks, how does inequality play out in stockpiling and to what extent do households stockpile, do poor households stockpile? Okay, so I realized that I forgot to answer about the sample size on the previous question. So I would answer that if that's okay before I answer Felix's question. Okay, so for the sample size, we sample about 400 individuals for each of the 12 countries but for both round, but for the second round we had that Rwanda and, sorry, Uganda and Tanzania. We got a sample size lower than that for them in the second round. So now to Felix's question, inequality in stockpiling. So we see that females stockpile more from the results that we had and we can explain it in this way that usually it's the women who go out to the market to buy foodstuffs and necessities for their homes. So if anything at all, they would exhibit most of these behaviors because the other ones actually buying their food and the other necessities. And for poor households, we see that they do not stockpile because they have food waste. They don't have the resources to buy more. So the results show that they were buying smaller packages or like, and more than the rural, sorry, more than the urban households. So it's because for his first question is because it's women who go out to buy the food and also for poor households because they don't have the resources, okay? So they are worried more about food. So they don't have the ability to stockpile. Actually, I have several questions for each one of you because I read all of your papers. What I will do is I will just send them by email and then you can get some of my comments because I still have several questions here. Okay, Belinda in one minute, how do you deal with the possible hot turn effects? It's from Stephen again. Okay, thank you Stephen again. So what we do is that the service are phone service. So thank you for that question. I forgot to mention that. So these are over the phone with the enumerators and so we're asking them questions about their mental health. For domestic violence, we actually can do two things. So not only is it over the phone, most of the sample are head of household. So they're men, about 90%, I think 80 something percent of them are men. And so what we do is we ask them, have you threatened or threatened to hit your partner or threatened your partner in any way in the past seven days? So we're asking them to elicit their domestic violence responses. And so that way we think we can actually put a lower bound effect to say, look, you're less likely to admit that you are a perpetrator of domestic violence than you are. Maybe to admit that you're a victim because of fear, et cetera, there's a literature on this. So those are the two ways that we do this. Thank you. And we have one last question. Maybe I can answer this because I read your papers. Can we say that there is an exaggeration in many studies and research related to risk associated with pandemic? I think the papers in this session has addressed that. So you have, for example, Allen's paper and using a randomized lease experiment, he was able to properly address potential measurement errors and bias when asking respondents. The problem in surveys, for example, is people sell the report. Yes, they may exaggerate, but you have seen that, for example, John's paper, she used a survey experiment so she can measure the property, the effect of her target variable because she used an experimental setting in her survey. So my answer to this question is not necessarily. Yeah, the results of surveys and experiments in this session are not exaggerated though. Do you want to add one more? Actually, I just have an additional question. So when we did our survey experiment, it was during the lockdown in Metro Manila. So you would expect that everyone is aware of COVID and is affected by COVID because they cannot go out and they cannot go to work. And yet we see that our treatment, which was the COVID frame, was valid in the sense that the control group actually did report lower levels of risk and vulnerability compared to the treatment group, which kind of shows that even if COVID was very salient, just priming individuals have actually increased their percentage of risk. And then I have to add, because we weren't able to ask Hilmi on his nice field experiment. Actually, it's a very well-designed work. I suggest you read it when you have time, but what is not in that paper is potentially you can see how the donor's identity, say his ethnicity in Indonesia affects his preferences in terms of the characteristics of the recipient. In his paper, with his data, he can check whether, for example, donors from outside Jakarta are more likely to donate to recipients outside Jakarta too. So he is able to see whether there is in-group or out-group preferences in terms of donations related to COVID. It's a wonderful paper, so just you check it for yourself because we don't have time anymore. I thank you for this session. So I hope the five of you got the comments that you like and maybe I will send something when I'm able to clean up my notes. And then I hope you enjoyed the session. We have to go to the very last event at the main stage. Bye-bye.