Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=

# Economic and Psychological Returns to Social Relationships: Alleviating Constraints to Network Formation in Malawi

Authors:   
Discussant: None  
Video: https://www.youtube.com/watch?v=mC4ESywpIDc&t=8286s


## Talk (02:18:06 – 03:08:40)

[02:18:06] >> Yeah, I'll I'll have your time. >> All right, everyone, let's get started.  
[02:18:16] [clears throat] Okay, we are uh delighted to have uh Gabriella Fleshman. Gabriella, please take it away.  
[02:18:23] >> All right. Hi everybody. Thank you for inviting me to speak. Uh my name is Gabriella Fleshman and today I'll be talking about the economic and psychological returns to social relationships in Malawi.  
[02:18:34] So social connections play a fundamental role in economies. In economics, we associate them with instrumental returns such as financial and labor market inclusion serving as a conduit for the diffusion of information and technologies as well as risk sharing in  
[02:18:47] low-income contexts. But also note that because networks are segmented and heterogeneous, they can reinforce inequities keeping resources among the resource rich.  
[02:18:57] At the same time, social relationships may provide intrinsic value. They are a very predict powerful predictor of well-being and these correlations have largely been established by other disciplines.  
[02:19:08] But networks form indogenously. So much of this literature is correlational and we know even less about how networks form or what constrains them. So in this paper I ask what is the value of a social tie and what prevents them from  
[02:19:22] forming. So within this big motivating question I'm going to ask three specific things.  
[02:19:29] First, what are the causal, economic, and psychological returns to social connection? And I excuse me, identifying causality matters both in terms of magnitude and the direction. Um, since social relationships are potentially the largest predictor of things like upward  
[02:19:44] mobility and reverse causality arguments are plausible but would require a different policy instrument.  
[02:19:51] Secondly, I'm going to ask how these intrinsic and instrumental benefits of relationships interact. They could be complimentary to one another. We know that income and psychological well-being have a causal relationship with each other. So if relationships provide both these things, they can mutually  
[02:20:05] reinforce one another. Or there could be trade-offs where different types of relationships offer different benefits that are independent of one another.  
[02:20:13] And then finally, I'll ask what constrains link information and for whom?  
[02:20:18] I ask these questions among a sample of women who have moved villages at the time that they get married within rural central Malawi. So they've been uprooted from their families and existing networks creating the conditions for social isolation and vulnerability.  
[02:20:32] But this is a very widespread phenomenon both looking traditionally at the eth ethnographic atlas as well as in modern data from the DHS where 50% of women who are married and living in rural areas in subsaharan Africa are not living in their home village. And in central  
[02:20:46] Malawi specifically in the DHS I find that these women are disadvantaged relative to women who are in their home villages.  
[02:20:53] So what do I do? I run an RCT where I provide exogenous linking opportunities between these women by resolving constraints to sharing a meal. This is a social setting that allows for personal conversations and these women report that they would like to do it more often  
[02:21:08] than they currently do. Specifically, I facilitate low SCEs women in inviting others by showing them a list of the names of women who have said they are willing to receive an invitation and then sending these  
[02:21:20] invitations on their behalf. I also create exogenous variation in who they can interact with through the experiment. Some women are randomized so that they only see the names of and can only invite low SCES women like themselves. Other only see the names of  
[02:21:35] highs women and the rest see a random mix of both.  
[02:21:39] Finally, I cross randomize a meal subsidy across all three arms.  
[02:21:44] Okay, so what do I find? Pooling all of these arms, having the opportunity to send these facilitated invitations leads to an increase in non-aggricultural self-employment within just one month.  
[02:21:56] One year later, these women experience large improvements in their food security and this is uh the largest effects uh on this index are driven by the lean season. So this is well for relevant because it has implications for consumption smoothing and they also experience large  
[02:22:11] reductions in mild to severe depressive symptoms. So any linking is extremely valuable for both economic and psychological well-being.  
[02:22:20] But now looking heterogeneously at the socioeconomic status of the guests who they invite, we see that the access to these high SCES guests are what drive these economic benefits while access to low SCEs guests drive the treatment  
[02:22:34] effects on uh depress depressive symptom reduction. So there is a trade-off between the low SCEES and the high SCES relationships since women cannot get both sets of benefits from one type of person.  
[02:22:47] So if these relationships are so valuable, why don't they form? Well, 81% of women send any invitation across all of the arms. So who they can invite and uh what kind of meal they can serve does not affect their propensity to invite on  
[02:23:01] the extensive margin. But I do find that the meal subsidy does change the composition when they have choice towards high SCES guests. So the most binding constraint to taking any initiative is that women don't know who  
[02:23:14] to invite and it takes a margin of effort which this invitation sending list resolves for everybody.  
[02:23:22] Okay. So this paper speaks to the literature on the economic value of social ties in development. We have a large and canonical literature on how networks can facilitate risk and consumption smoothing, but we actually have very little causal evidence on the role of a new network link because  
[02:23:37] networks form indogenously. And so I provide evidence that an exogenous network shock does improve ex consumption smoothing.  
[02:23:45] There's also a literature in high-income countries about the value of cross social class ties specifically for low-income people. And I provide uh quite a bit of external validity to the literature in this evidence base that's causal. um since most of this literature tends to be within education contexts in  
[02:24:00] high-income countries and non-experimental.  
[02:24:03] I also contribute to the literature on the psychological value of social ties providing what to my knowledge is the first causal evidence that real life social relationships do improve mental health in the long run.  
[02:24:17] But novel to both these literatures, I show that these uh effects do not come from the same people, implying that there is a trade-off when social uh network formation is constrained, which uh you know is a nice segue into this final contribution, which is that I  
[02:24:32] do identify some causal frictions in social network formation. First, there are social frictions between individual people that even when they want to link and they have the opportunity, social frictions are still inhibiting them from  
[02:24:45] uh from interacting with one another. I also causly identify price as a friction in crossocial class linking um implying that it serves as a determinant of income based.  
[02:24:58] All right, so I'll go through the experimental design and the data. I'm going to use the first stage to answer this question about what constrains network formation and then use the second stage to ask what are the impacts of these social relationships.  
[02:25:12] So first I recruit women who are adults who move to the village after the age of 14 and have been there for fewer than 20 years. So this is how I define migrant status. So you can think about this as women who did not grow up in the village. At this stage, I give them the  
[02:25:27] opportunity to publicly signal their interest in later stages of the experiment in receiving an invitation for a shared meal. And at this point, 99% of women agree. So, women are already kind of signaling their interest in this type of intervention.  
[02:25:41] I then use the data from this recruitment to construct an index of socioeconomic status. I assign the top 20% to be high SCEs and the bottom 80% to be low SCEs. And throughout I'm mostly going to be focusing on this low  
[02:25:55] SCES subsample. So now I'll just provide you a few stylized facts about um their characteristics.  
[02:26:02] So first income is an important social cleavage.  
[02:26:06] Women are 12% more likely to link with someone who has the same type of roof as them if they were to link at random within the village.  
[02:26:14] Secondly, networks are not sparse. Women are connected with six others on average, but their networks are dominated by their husband's extended family and by weak ties. So only 36% of them have any strong tie, which for the  
[02:26:27] purposes of this talk you can just think about as a very trusted best friend.  
[02:26:32] And to help explain why their networks look this way, I want to emphasize that these are young and vulnerable women.  
[02:26:38] The median woman arrived in the village four years ago, so they're not brand new to the village, but they haven't been there very long in the grand scheme of things. at 20 years old. So half of these women were married as teenagers.  
[02:26:50] And lastly, meal sharing is common. Women say on average that they shared three meals in the past week, but they say they'd like to share meals more often and that it's rarely with their own friends. So when I ask them to name everyone who they shared meals with in  
[02:27:04] the past year, only about half of them are naming anyone who is not their relative.  
[02:27:10] So from this low SCES subsample, I randomized 1,600 women for my intervention. And I'll describe exactly what I do with them in a couple slides.  
[02:27:20] And then I randomize another 500 low SCES women to be guests. Guests are the women who will have the opportunity to receive invitations. Um, and so for the purposes of today, I'm not going to be doing analysis with them. And so just think about them as uh people I'm implementing as a part of my  
[02:27:35] experimental arms. I also randomized 500 high SCES women to be guests.  
[02:27:43] So everything going forward is going to be focused in this low SCES intervention sample. So I'm always going to be comparing low SCES women with one another. So I'll describe what I do in that sample next.  
[02:27:55] So before I go into the experimental arms, I want to just briefly um describe the intuition of my conceptual framework to help motivate the intervention.  
[02:28:04] So social relationships provide utility in two ways. First is intrinsically there's some non-market value of companionship such as a psychological return. The benefits that we feel from just being in the company of others or knowing we have friends.  
[02:28:18] Secondly, there is an instrumental utility and these are the economic returns. Friends hook us up with jobs.  
[02:28:24] They risk share with us which allows us to consume and from that consumption we get utility.  
[02:28:29] Now, women may be inhibited from building relationships by a few things.  
[02:28:33] First, they may face information asymmetries. They may not know if other people are willing to link or who specifically is willing to link with them. Secondly, even if they know who's willing to link, there may be real costs. There could be effort costs such as time or shyness. And in my sample,  
[02:28:48] what people describe most anecdotally is a fear of rejection.  
[02:28:52] There could also be financial costs. Social activities have prices. And these women are budget constraint.  
[02:29:00] So within Yeah, >> just a quick question. So is there is there any situation where it's not appropriate to actually have these kinds of friendships? Is there anyone who would want to prevent these kinds of friendships or these kinds of linkages?  
[02:29:13] Like you know in the community like is there Yeah. How appropriate is it always to just link up with random people or to have gatherings and are there people who would actually not want this?  
[02:29:24] >> Yeah. So within the setting uh linking across gender would be inappropriate which is why I do everything with within women. Um other than that not in any ways that people describe to me. Um and you know people always have the  
[02:29:37] opportunity to say no. Um and there there are some cases where you know people might say like the reason that I didn't actually follow up on this is because she never came and you know did followed up on the initial invitation which might signal that that's happening  
[02:29:51] but that's actually very rare. So I'll show you that actual meal sharing conditional and invitation sending is very high and invitation sending is very high. So there's no evidence that you know within this set setting there's kind of like systematic reasons why some people shouldn't interact.  
[02:30:14] So, both happen. Um, I think more I I don't remember the numbers off the top of my head, but I'm fairly certain that it is more common for women to just share them either with themselves or with other women. So, oftentimes these meals happen at lunch when men are out  
[02:30:28] in the fields working. Um, and so, you know, women might be at home doing chores. I mean, these women also go out into the fields. Um, but you know, anecdotally, these are usually lunchtime meals when what men are out of the house.  
[02:30:56] >> Yeah. So, I do not hear anybody talk about combining resources. So what people talk about is that they um so outside of the experiment the way that this happens is you might just kind of be walking by you go say hi to your friend and she's happens to be cooking  
[02:31:10] so she says oh why don't you join me so it's generally you know whatever they would cook for their household anyways um I don't see any evidence that kind of aside from with the the meal subsidy which I'll get into that the contents of the meals are changing because they're  
[02:31:23] combining resources. Okay.  
[02:31:27] So the first intervention arm is what I call the inviter treatment and in this arm I facilitate invitation sending for women and then I randomize another 400 women to control. So what this treatment does is it provides information which is the names of six neighboring women who  
[02:31:42] have been randomized to be guests who women understand to have expressed interest in receiving an invitation for a shared meal. and then also reduce the effort costs of sending these invitations by having a numerator's uh offer to send invitations on the  
[02:31:55] inviter's behalf to up to five women. So what this does is exogenously reduce the believed probability of rejection and the actual effort costs of initiating these relationships.  
[02:32:07] Now within the inviter arm I want to understand the effects of interacting with different types of people. But if I let people just ex indogenously select in different type types of relationships, I won't be able to differentiate selection from the actual effects of a different type of  
[02:32:21] relationship. So to shut down indogenous sorting, I assign some women to what I call the high SCES guest list where every name is a high SCES neighbor generating exogenous cross SCES interactions only.  
[02:32:35] In another arm, every name is a low SCES neighbor which generates exogenous within SCES interactions.  
[02:32:42] But now to understand women's own preferences when they have choice over both low and high SCES women, I have a third arm which I call inviters with the random guest list where the guest names are a random mix of both low and high SCES neighbors generating exogenous  
[02:32:55] interactions with indogenous sorting across all three arms. I cross randomize a voucher. This is a coupon that's redeemable at a local butcher. So, this is exogenously reducing the financial  
[02:33:08] cost of serving a nice meal with meat and making it the same cost as serving a simple everyday meal without meat.  
[02:33:17] I conduct four surveys. The first is my recruitment survey where I elicit that interest in being a guest and almost everyone says yes. I then return one month later to allocate the treatment, distribute the vouchers, and actually send invitations.  
[02:33:30] return one month later at the time that the voucher expires to collect short-term outcomes and then finally return one year later for an online survey to get my longerterm outcomes.  
[02:33:42] Okay, so to understand the first stage, we're going to look at three sets of results. First is just who do they send invitations to within the context of the experiment. Do they take up the treatment? But to actually understand, you know, what this means for downstream  
[02:33:56] outcomes, we're going to look at the follow-on effects of these first invitations. Do they actually share meals? Um, and just to clarify, after the invitation sending, we don't do anything experimentally to follow up on this. Everything after this is on their own valition  
[02:34:11] and then one year later, have their networks changed in ways that are different from the control group and what might have happened just over the course of the year.  
[02:34:20] >> I was just getting >> getting Okay.  
[02:34:25] All right. >> Okay. No, I'm done. Yeah.  
[02:34:29] >> Sorry. >> Uh, so in any classes society, I'm from one, I live in another one. It it might be costly for the high SCES person to affiliate with interact with a low SCES  
[02:34:42] person, right? So, and and they might have to pay some social costs, right?  
[02:34:46] Like we often do in the US, right? And um and in that case, you know, like for a high SCES person, it they have an excuse in this particular experimental setup that oh, I'm I'm not willingly  
[02:35:00] interacting with low SCES people is because Gabriella told me to and that's how it's coming up. Whereas in a natural setting, it may not happen. So I'm I'm just curious about your thoughts on the naturalness of your setting. So how do we think about the implications of it for policy?  
[02:35:15] >> Yeah. Um let's see. Okay. No, I don't. the button is later but I'll just say in words um so I yes this does not happen normally it is within the context of the experiment but that said um you know  
[02:35:29] this meal sharing itself is something that is pretty common and also women have been warned basically that this is happening and they've basically been given the opportunity to opt out at this first recruitment stage they could have said I do not want to receive  
[02:35:42] invitations right they they say that they want to receive invitations the rates of meal sharing and then also So reciprocating those first shared meals are non-ifferial by the SCES of the guests. So you know they might follow up on that first shared meal but they don't  
[02:35:56] have to then reciprocate on their own. Yes, there might be norms that suggest they should but we don't see any evidence that they're they're you know kind of stingy or unwilling relative to the low SCES guests. Now, something else I do is I compare women whose  
[02:36:10] invitations are rejected um within the experiment with those who say they've had, you know, they've invited someone and then no one's followed up on it um in the control arm. So, you can think of that as outside the experiment rejections. And those two groups are balanced. So, we don't really see  
[02:36:25] evidence that who is is sharing meals within the experiment is different than those who are sharing meals outside the experiment. Yeah.  
[02:36:33] kind of context later on what about like how big the villages are there different tribes pass things like that what is normal  
[02:36:48] >> yeah so there are kind of two predominant tribes in this area but 90% of women in the sample belong to one tribe so we shouldn't be thinking about that as a particularly uh salient cleavage here relative to income um the villages are big so each of these  
[02:37:02] villages uh go from span from like 600 to a,000 people. Um and so they're pretty dense. Um and so I'm actually thinking of I I create um I just basically assign a neighbor as someone within 800 meters. So they're always  
[02:37:15] inviting someone from kind of within a certain radius of them, but not not respecting village boundaries since these villages kind of tend to bleed into one another in some circumstances and are very big. And even within that 800 meter uh meter radius, I do see that  
[02:37:30] women are more likely to say they know the people closer to them. So there is, you know, people don't know everyone in the village. Yeah.  
[02:37:46] So naturally your intervention >> Yes, I I think that's right. I would  
[02:38:00] just argue that that is true for everybody, not just the cross social class ties. It's removing some amount of, you know, it's just costly to go and make yourself vulnerable and say, "I want to hang out with you," when you don't know how they're going to respond.  
[02:38:13] Okay. So, do they take up the treatment? This graph is showing you the percent of women who invite at least one person across the three subgroups when they do not have the voucher. So, first takeup  
[02:38:26] is very high. 81% of women send any invitation. Almost all of these are going to links who are outside their baseline network. So, these are not shared meals that would happen otherwise. And we don't see any differences based on whether they have  
[02:38:39] high or low SCEs guests. So norms or extra effort that we might think would inhibit inviting high SCES guests actually are not differentially binding.  
[02:38:49] Now when we compare these groups with the women that do have the voucher, we don't see any voucher treatment effects.  
[02:38:56] So inviting someone isn't better than inviting no one regardless of their social class or the food that can be served. suggesting that the binding constraint is what the inviter treatment resolves for everybody which are these information asymmetries and effort  
[02:39:09] costs. But now who do they prefer to invite when they have choice? So now we're looking at inviters who have the random guest list and the percent of them that send any invitation to a high CS woman  
[02:39:23] and the percent that send an any invitation to a low SCES woman and they are exactly the same. So inviters seem to be very split in who they prefer.  
[02:39:32] So what kinds? >> So uh no not on the guest list. So the the way it's constructed is that I randomize 500 low CES women to be guests and 500 high CES women to be guests. So yes, high CES women are over represented  
[02:39:46] on the guest lists. Yeah. invitation.  
[02:40:02] >> So I um so okay anecdotally I asked women like how do you know who is well better to do than than others and they list things that are kind you know that are within my SCES index and observable.  
[02:40:14] I also show women photos of each woman as they so actually the photos I showed back uh here. These are photos that these are the photos that an inviter would see when choosing their guests. So they have some information which is  
[02:40:28] visual. These are also their neighbors and they never invite someone who they said they've uh never seen before. So they know who these women are. Most of them say they've talked to one another.  
[02:40:37] Um they just aren't engaging with each other in in you know more meaningful activities. Yeah.  
[02:40:52] or they just own them. >> Uh, okay. So, yes, something I I forgot to mention is that, um, most women are just sending one or two invitations even though they can send up to five. Um, and so I can just show you here, um, that a  
[02:41:06] good portion of them are in are sending invitations to both, but um, but it's like less than 25%. So the rest are just sending one invitation to a high or low SCS woman.  
[02:41:24] >> Yeah. So we just go door todoor and we say that you know in future like we're going to come back and then we're going to offer some people the opportunity to send invitations to other people. Um you know do you want to have the opportunity to receive that invitation. So because  
[02:41:38] we've come and said it's not normal but we've prepared them for it. So when people actually receive the invitations they know the context of what this is about is it isn't totally out of the blue. They signed up for this in advance or for this opportunity.  
[02:41:54] Oh >> um no not that I mean not that the enumerators told me about. They kind of just accepted it. And and recall like women say they would like to share meals more often than they are. They they are sharing meals. So it's not like a  
[02:42:08] totally weird thing for them to kind It's not like I'm asking them to to participate in some activity they aren't used to. Uh we're just basically saying like hey do you want the opportunity to do this thing with some of your neighbors? Um and you know they all they  
[02:42:23] say that they would like to do it. Um they just think that other people don't.  
[02:42:31] >> Yes. >> Yeah. Another question. Yeah.  
[02:42:44] And so maybe this gapes would be larger connected.  
[02:42:58] >> Yeah. So I mean I've looked heterogeneously by baseline networks and don't really see much. Um but yes I mean there are differences in in baseline networks. um and baseline networks are homophilic with respect to so the way that I measure homophyle with people  
[02:43:11] outside my sample is by uh roof roof material since they can give me the roof material of every network connection um so I do see homophy u but I don't see u heterogeneity >> just you don't have to re answer but I think Laura was saying if you like the very specific people on the list if they  
[02:43:26] knew them already then they weren't going to use this and that might differ between the high and low so not their general connectedness but the like specific connectedness with the low my know when to ask this but like I'm  
[02:43:39] just curious how what if you think of this as like specific to development in this context like I know I don't not hear a lot about like women in social norms and I'm thinking of manis in the US  
[02:43:52] >> yeah how is the low middle low poor country relevant here >> well I think that the the implications for what they're missing out on right um so you know I'm going to find very large effects on food security and these women are very close to subsistence Right? So  
[02:44:06] having the opportunity to link with someone that provides this very large benefit has huge implications for their welfare. And the fact that they're not forming these links because they're just kind of shy and embarrassed to to go ask  
[02:44:18] someone to to hang out is quite I mean it's quite striking. They're leaving a lot of welfare benefits on the table. Um and now that's not untrue in low high income countries as well, right? Okay. I mean, we have evidence that connectedness with higher income peers  
[02:44:33] does improve um well-being, but not generally among just neighbors. Right?  
[02:44:38] These are women who live right next to each other, but still somehow they're not able to access these benefits of of having one another in their lives.  
[02:44:49] >> Yeah. >> Sorry, one more time it No, no. I think that the uh I think the  
[02:45:13] food security benefit is more like kind of localized to the local, you know, to this context because people are very food insecure like very food insecure.  
[02:45:22] Um no, I mean I think depression benefits are good for everybody.  
[02:45:35] I'll get there. Yeah, a few slides. Yeah, one more.  
[02:45:55] >> That that's Um, so you're saying that like from the guest's perspective that there's something like trustworthy because this is happening through an organization and not just um  
[02:46:09] perhaps I mean but you know you're still telling like they still have to follow up on this on their own accord and you know perhaps it gives them some level of confidence uh because it's happening through uh an intermediary but  
[02:46:23] ultimately whatever happens after the you know after the enumerator ers leave is is up to them and whether or not you know they don't even if they share the meal they don't have to do anything which is then going to provide these downstream benefits right um so that comes from the information they choose  
[02:46:37] to share within these meals and so I think there are a lot of steps at which people have to make active choices to kind of be willing to engage here that would make it a little bit hard to conceive that this is entirely experimental artifact yeah  
[02:46:52] >> okay Sorry, I couldn't quite hear you.  
[02:47:07] >> Uh yeah, that's that's coming. Okay, so now when we look at the voucher treatment effects among women with the random guest list, we see that the voucher does have treatment effects on the composition. It leads to an increase in invitations to high SCEES guests  
[02:47:21] without any changes in invitations to low SCES guests. So what's happening is that there's an increase in sending invitations to both low and high SCES guests. So this tells us that when interaction with low SCES people is an option which is the best approximation  
[02:47:36] of you know some the real world outside of this experiment prices do constrain crosses linking.  
[02:47:43] Okay. So inviters do take up the treatment. They are constrained on the extensive margin by information and effort. And their preferences are mixed but with a stronger preference for high SCES guests when prices of meat are low.  
[02:47:57] Do they actually share meals? Yes, they do. 63% of women share meals within one year and the invitation sending pattern sorry the meal sharing patterns largely uh follow the invitation sending patterns. So there are no differences between low and highs guests in actual  
[02:48:12] meal sharing. So what happens to long-term networks?  
[02:48:17] Going to look at three uh sets of outcomes. The first is the network degree. So this is the number of people who they report interacting with across all of their network activities. The second is churn. So this is going to give us a measure of how the networks are changing even if they're not  
[02:48:31] expanding or contracting. So it's just the number of people who are new to the network who they did not list at baseline plus the number of people who have dropped from the network.  
[02:48:40] Finally, I'm going to look at network composition, particularly with respect to strong ties, which recall just thinking about as like a best friend and the origin of ties. So, are these your own friends or are they your husband's relatives, which make up the majority of  
[02:48:53] the networks for these women? Okay, so networks do not expand. There's absolutely no treatment effect on the size of the network. But getting to the questions that we've had uh in in this last question session, uh who is in them  
[02:49:07] does change quite a lot. So network churn increases. People are being added to the network and being dropped from the network at equal rates.  
[02:49:16] This is true regardless of who they can invite.  
[02:49:20] So now what are different about these people who are new to the networks. So we find some suggestive evidence that the composition is shifting away from their husband's relatives and towards their own friends. These are small magnitude. They're not, you know,  
[02:49:33] statistically significant, but these are the two main components of the network.  
[02:49:37] And the direction of effects is going towards their own friends.  
[02:49:41] >> Yeah. >> Help us understand how you elicit what a network is. I mean network means something to us but it doesn't mean the same thing to them by stretch imagination and elicitation.  
[02:49:51] Very very related to how the question they're asked and how persistent the interview are and so forth. So it's really hard to know what to think of this.  
[02:49:58] >> Yeah. So I asked them um you know for this activity who do you engage who did you engage with over the past year? And so I asked this for >> this being a meal share.  
[02:50:07] >> No, for a number of activities. Those activities are meal sharing. So who did you host uh and who did you visit? Um uh economic relationships like who hired you for peacework? Um who are your business associates if you have a business? Um and then things like who do  
[02:50:21] you go to for advice? Who do you borrow from and lend to? Um who do you share secrets with? And then just a catchall which is anyone else we haven't mentioned.  
[02:50:32] you ask about are they included in this whole?  
[02:50:40] >> So they would be they're included here if they list them. So I don't I don't ask about them specifically and I also don't kind of have them pre-listed or anything like that because I don't want to elicit networks differently between control and treatment.  
[02:50:54] >> Yeah. So I have not linked all the names in line yet.  
[02:50:59] Okay. So when we look at a few other better powered outcomes, we do see some some more evidence about own friends becoming more important in the network.  
[02:51:08] So there is an increase in the probability that the network is majority one's own friends and an increase in the probability that there is at least one strong tie who is your own friend. When I put these three measures together, we get a precise 0.1 standard deviation  
[02:51:21] increase in this uh presence of one's own friends in the network.  
[02:51:26] And again, this is no different across any of the treatment arms on any of these measures. So regardless of the social class of who you have the opportunity to interact with, it leads to more of your own friendships.  
[02:51:38] And there's some evidence that this is coming from a few specific activities.  
[02:51:41] So women are more likely to seek advice from their own friends. Uh some of the the subgroups are more likely to uh share secrets with their own friends and all groups are less likely to lend to their husband's relatives.  
[02:51:55] Okay, so women take up the treatment, they actually share meals, and then their networks do change in ways that are different from the control group with some evidence that this is primarily driven by new friendships. Um, with some evidence it's displacing relatives.  
[02:52:09] So now, what are the impacts of these relationships?  
[02:52:16] >> A bit more about you say a bit more about why you make this distinction between husbands,  
[02:52:30] relatives, friends. I think I can why why this would be important but you know one can imagine having close relationship on one's in-laws.  
[02:52:45] [laughter] >> Uh well okay so first if you just look so you know they have like about six and a half friends on average and the two largest components of their networks are their own friends and their husband's relatives. The other people who aren't included here are like their own relatives. You know some people are you know even if they live in different  
[02:52:59] villages the person who they seek advice from they'll call up their mom for example but you know their most important people in just in quantities are their husband's relatives and their own friends. Now if they are for example having marital issues going to your  
[02:53:13] husband's mom might be a challenge. So we might think that for social support that could be important. Also in terms of like the kinship taxation literature there is some evidence that you know family ties and kinship ties can be extractive. Um not necessarily. We don't know like it could be that the  
[02:53:28] friendship ties are also extractive. We don't know but there are we do have some evidence that that those types of ties tend to you know place uh expectations upon people. So again it's not it's an empirical question like what is changing in the network and how does that affect  
[02:53:43] people downstream but there are some reasons to think they could be different.  
[02:53:48] why they would get different effects but additional way to answer my question about the context like with passion locality these young women are mostly getting thrown into their husband's families this is an attempt so there's  
[02:54:02] like a reason they don't have their own life or friendships uh yeah >> yes one just one clarifying point on that is that in this context there aren't norms prohibiting them from forming their own networks right it's  
[02:54:15] it's to do with these like social frictions and >> seems relevant for why a lot of their friends are.  
[02:54:25] >> Yeah. Yeah. Absolutely. Absolutely. Yeah.  
[02:54:31] >> Yes. Yeah. Uh they do not change over the course of the experiment. So we do not see um any sort of evidence of backlash. Actually, if anything, divorce rates go down.  
[02:54:42] Okay. Okay. So I'm going to look at three main outcomes. The first is a food security index. And this was a year of food shortages. So um this is a you know a relevant outcome but particularly relevant in this year where more than half of the control group said that  
[02:54:57] during the lean season they were eating one meal per day that lasted three months. And this is also a particularly sensitive sample with half of them being pregnant or breastfeeding at some point during this trial. Next I'm going to look at farm meals. So this is a subsistance agricultural society. So  
[02:55:12] we're thinking about this as the main source of income and I collect the endline data right after the harvest. So the agricultural season falls between the intervention and the endline.  
[02:55:23] >> I guess what you're telling us now combined with the voucher if I if I understand correctly voucher could also just be used for the family. So the fact that they ch presumably it affects who invite so it means they're using it  
[02:55:36] possibly socially means they value these relationships a fair amount otherwise they could eat the meat breasts. So I so I do find that um only about half of women do share uh with half of women with the voucher share meat in the shared meal. Um so you know a fair  
[02:55:51] portion of them are just consuming it for the household. No one shares meat in any shared meal without the voucher. So I mean it still is a very large treatment effect there. Um, and then one point on on the food security index when I measure it at the one-mon mark, I drop meat consumption so that it's not going  
[02:56:05] to be mechanically related related to the voucher.  
[02:56:09] And then my measure of psychological well-being is mild to severe depressive symptoms. And I'm just going to use a binary cutoff using a standard cutoff on the CESD depression scale. So you should think about this as indicating people who are at risk of depression.  
[02:56:22] And to give you a sense of the prevalence, 36% of the control group um you know had symptoms of depression at endline. So this is fairly high.  
[02:56:32] Okay. So pooling all of the treatment groups, there are no effects at the one-mon mark, but one year later there is a.13 standard deviation increase in food security.  
[02:56:43] There are no effects on farm yields, but notably this is very noisy, but I'm going to take it as a null.  
[02:56:49] And again, no effects at the one month mark, but at the one-year mark, there are large reductions in depressive symptoms.  
[02:56:57] Now, when I compare across the three different uh arms, I find that it is these crosses interactions that drive these food security benefits with the effect among inviters with the high SCES guest list being three times as large as the effect among inviters with the low  
[02:57:11] SCES guest list. On farm meals, there are no effects across any subarm, but it is notable that the relative uh levels do follow the relative levels on the food security index.  
[02:57:25] But now when we look heterogeneously at depressive symptoms, things go in the opposite direction. So it is inviters with the low SCES guest list whose treatment effects are more than twice as large as the treatment effects among  
[02:57:37] inviters with the high SCES guest list. interpretable things across social groups in  
[02:57:50] how we should think about this scale. >> Yes.  
[02:57:59] >> So that's a follow on project. So I did randomize the treatment saturation. Um and I don't find any evidence of spillovers on these main outcomes but I'm looking at much more on the network stuff in in the following project here  
[02:58:12] >> related to Rob's question. Um so when I look through all look at all the columns right so for example let's start with farm meals right in the social networks literature in development there's a lot on social networks and technology adoption for example in agriculture  
[02:58:25] right and then for food security there's a lot on like risk sharing and gifts and gift exchanges and so forth right so uh like I would find it uh even more credible if we could look at the sort of  
[02:58:39] the outcomes the intermediate outcomes that would ultimately lead to >> that's Okay, so all of these relationships are extremely valuable, but the benefit depends on who they are with. So where are they coming from? So first we're  
[02:58:54] going to look at the effects of the highest yeses relationships, and that's what we'll focus on with the remaining time. So in the paper, I go through a whole bunch of different potential explanations. I'm going to focus on the three today that you might think follow the most directly from from the  
[02:59:06] experiment. So first, maybe they're just sharing more meals, right? when they are sparse on food themselves, they can go to this new friend's house and have a shared meal, and that should smooth their consumption. But we actually don't see any effects on long-term differences in meal sharing. So, they're not sharing  
[02:59:21] meals more often, either as a host or as a guest.  
[02:59:26] Okay? So, maybe they're borrowing more often, maybe uh from these women directly, or these women are kind of hooking them up to formal financial resources. we don't see any evidence of changes in borrowing um either in terms of the amount or the formality or the  
[02:59:40] frequency I should mention as well. So finally we're going to look at employment. So maybe they're being referred for jobs, they're ear learning information about income earning opportunities and here we do see positive evidence. So the main forms of  
[02:59:54] income earning besides agriculture are peacework and self-employment.  
[02:59:59] So first we'll look at peace work and we're just looking here at the number of times they say they worked per month both during the growing season and the harvest season. You don't see any effects across any of the arms. So they're not being for example referred  
[03:00:11] to more peacework jobs. Now when we look at self-employment and so you should be thinking about this as very very smallcale businesses. So the most common is just selling prepared foods around the village.  
[03:00:23] Here we do see treatment effects at that one month mark. These are actually very large um and they're still positive later but no longer significant pulled.  
[03:00:32] When we look at the different inviter arms, we see that these are entirely driven by inviters with the highest CS guest list where these effects are very large. This is a 50% increase in the probability of earning income from self-employment within just one month  
[03:00:46] and the effects persist. Now the treatment effects do get smaller but this is really explained more by this increase in the control mean. So women are starting new self-employment activities within a month and then able to continue earning income through these  
[03:01:00] activities through the lean season up until we come uh to survey them one year later.  
[03:01:07] So how are these women able to start these new self-employment activities within just one month? Going to argue that it is primarily driven by information. So in their self-reported interactions with these highs guests first they're more likely to say that they are choosing them specifically  
[03:01:21] because they are knowledgeable. they're more likely to say that they are uh discussing business. And a lot of inviters in an open-ended uh uh discussions about what they're doing with with these experiments and what the what the intervention did for them, they  
[03:01:36] describe information sharing. I also observe that women are directly copying the self-employment activity of the guests who they share a meal with.  
[03:01:46] And then when I ask women how they were able to start a new self-employment activity, the most common reason is that someone told me it is profitable. So that just don't know what to do basically and these women give them useful advice.  
[03:01:58] Now we would think credit would be important. You know these are very small scale businesses but there are some fixed costs and indeed 31% of women who start new businesses say that it is because someone helped them find capital. But we don't have any detectable treatment effects on  
[03:02:12] borrowing. not at that one month mark when the businesses start up or later.  
[03:02:16] So, it's relevant but not the binding constraint that these highs women are helping to resolve. Market linkages is another potential uh explanation, but women on like are not engaged in these relationships. They don't say they have regular customers or suppliers very  
[03:02:30] often. So, it's just not prevalent enough to explain the large treatment effects.  
[03:02:36] Okay. So, I go through a number of other um explanations in the paper. Overall, I do find some evidence of other things going on. Women with the highest CES guest list are more likely to be cash cropping. They say they trust their network links with borrowing and  
[03:02:50] lending, even though I don't see any effects on borrowing and lending. So, there could be some risk mitigation going on. So, there are a number of other things that could be happening, but the largest and most striking treatment effects are on that self-employment.  
[03:03:12] potential guests have self-employment test whe >> Yeah, I could do that. Um, since  
[03:03:27] conditional on being high SCS, they're randomized to your list. So I I would have random variation on on that but I haven't looked at it but yeah I could.  
[03:03:36] Okay. So I argue that these food security gains are driven by these high SCS women sharing information that leads to self-employment income streams and some degree of risk mitigation as well.  
[03:03:50] So then what is explaining these depression reductions?  
[03:03:54] Well, I argue that these relationships are functioning as a mental health resource in and of themselves. It's not that these lows women are kind of referring women to resources outside, you know, that improve their mental health. And that the benefit of a  
[03:04:09] relationship for mental health is increasing in proximity and social class. My evidence for this is that I find equally large depressive symptom reductions among the richest of the low SCES women when they are randomized to  
[03:04:22] have a high SCES guest. So if you are sufficiently close in social class to anyone whether it's from above or below they experience depressive symptom reductions. So it can't be driven by social comparison for example or by some  
[03:04:36] specific information or resource that the low SCES women are knowledgeable about and are able to share. So then what is it about these relationships that makes them so useful for mental health? Well, I have some suggestive evidence that inviters with the lowest  
[03:04:50] EES guest list are experiencing less loneliness and they're more likely to share secrets with their own friends rather than their relatives, which gets back to this point about what might be different about relationships with friends or with relatives. Um, and I do  
[03:05:04] find that the the main difference in conversation topics between inviters with the low SCES guest list um, and inviters with high SCES guest list besides the the business uh, discussions which I mentioned is that women with low SCES guests are more likely to say that  
[03:05:19] they discuss romantic relationships. >> This name on the end seems really helpful because like the first mechanism is just like a oneoff thing and then presumably the depression they had continued with that person.  
[03:05:32] >> Yes. Although it is also possible that for example uh you know you become a friend of a friend right so that the person that you share a meal with invites you you know introduces you to one of their friends and then that's the friendship that actually sticks so it's not dispositive if I don't find that but  
[03:05:46] yes it could be it would be interesting no okay so with my last few minutes I'll conclude so I find that information asymmetries and effort costs of initiating social interactions are  
[03:06:00] inhibiting these relationships that provide very large economic and psychological returns, suggesting that there may be a significant market failure going on here, but that it's very easy to overcome with this simple intervention.  
[03:06:12] I also find that prices play a role in income based off which has theoretical relevance for social capital poverty traps since it is precisely these crosses interactions that provide these large economic benefits.  
[03:06:25] This is some of the first causal evidence of the impacts of social relationships on consumption smoothing and psychological well-being. But notably, I find that these are not coming from the same people, implying that there is this trade-off when relationship building is constrained.  
[03:06:39] And finally, I want to emphasize that I am generating meaningful social connections between existing neighbors, people who already know one another. So, the barriers to accessing the benefits of relationship likely go beyond having the opportunity to meet, which might  
[03:06:52] explain why we don't always see positive impacts um in the added value of group based interventions because oftentimes they don't do more than bring people together. There are these other frictions to engaging in meaningful conversation, including price. So, building specific efforts to build  
[03:07:07] social capital among marginalized groups could make policy more inclusive and could also uh induce virtuous cycles if they're able to then make better use of their resources. Couple questions. Yeah, >> I wonder if the phenomenon that Brady  
[03:07:21] talked about his talk may be relevant here as well. So element from a new social network connection that in your mind one reason why there's less  
[03:07:34] um you know there's less of this going on without without your intervention should be perhaps.  
[03:07:40] >> Yes. So, so one thing I find is that women are quite pessimistic about other people's willingness to interact. Um, and so I mean yes, there could be some stochcastic element. Um, that said, I mean, the average effects are very large. Um, I I haven't thought much  
[03:07:55] about that. But I I do think that this, you know, baseline information asymmetries are, you know, enough combined with this large cost of the possibility of being rejection being rejected to to explain this. I do see  
[03:08:08] beliefs also updating on the basis of observing uh more of these relationships happening uh in in the village but then they revert once the the meal sharing through the intervention has died down suggesting that people aren't able it's you know it's not enough just to change  
[03:08:23] beliefs because there are still large costs >> oh sorry >> okay thank So now just a reminder the next