 research that we did some years ago now in Liberia, well, which I'll talk about in a minute. So, but by way of a little motivation and connection to this conference, we often find development organizations and also state agencies choosing some, well, sometimes to implement development projects at the community level through either, they make a decision about whether to do it through same gender groups, typically all women groups or mixed gender, sometimes more official in terms of connected to formal institutional structures. In there are some areas, in particular microcredit, where at least according to Esther Duflo, they've been directed almost exclusively at groups of women. The logic for the choice of these decisions is not always completely spelled out, although there are usually references to it in RFPs and the kind of project documents. It may have to do with just a goal of promoting gender equality, equity. In some cases, it may be based on a notion that women will be better stewards of resources to be used for public goods than men would be. In cases of kind of health and education, there's often maybe a notion that women would have greater motivation to use the resources on behalf of children than men would. Now, there's some previous work that finds that in experimental settings, and possibly also, there's a big review, I can't remember that I think we'll refer to in the paper that talks about both experimental settings and also, let's say real world settings, that tend or sometimes find that women contribute more when they're interacting in all women groups. This is a paper that has results that are quite similar to what we find, although even though it was implemented in Nairobi slum and it's quite different in many ways, nonetheless, Grig and Bonet find that women, they were playing simple kind of four person collective action contribution games and they mixed whether they were four women, four men or two men and two women, and they found that women contributed more in the all women condition than in the mixed condition. It's similar to what we find. So in our case, what we're talking about was done in the context of a randomized control evaluation of a project implemented by the International Rescue Committee and funded by DFID of a community driven reconstruction program that took place in two districts of Northern Liberia, of Wajima and Zorzor, and the main project was about evaluating this community driven reconstruction project and whether it was effective in various ways. But when we went into the project, we had McCartin, myself and Jeremy Weinstein who was the other main collaborator on that part of the main project. We had doubts about whether the CDR program would work at all and because we were going to a lot of effort and it was gonna be a big deal, we embedded a couple other variations or interventions and one of them was kind of orthogonal to the random assignment of the CDR program. We randomly assigned whether in part of our evaluation involved like giving people in the treated and control villages what we call a real life collective action problem to see if the CDR treatment actually improved a behavioral, made for behavioral change to improve ability to undertake or generate resources in a collective action problem. We implemented orthogonal to this that half of the villages, the participants in the contribution game were all were, it was known we would only choose women to play and in the other half, it would be 12 men and 12 women. We didn't have enough power to choose a set that would be all men, unfortunately. So the 24 people would play this real life collective action problem. Why didn't this happen to anybody else? Sorry. Sorry. Where we gave them a small endowment, well it was a decent for the place about $5 in three 100 Liberian dollar notes and they made a private decision about how much of if any to put in an envelope, which they knew would then at the end of the day would be we'd come together in a community meeting and open the envelopes and count them out and that we the implementers would add our matching contributions at two different interest rates. Some people knew that their contribution would be doubled and other people knew that it would be increased by a factor of five. People knew their interest rate when they were choosing and so to summarize the main results, what we found was that in the places where it was 24 women chosen to be game players, they contributed markedly more than when they were playing, then in the mixed groups where it was 12 men and 12 women. In terms of the total money raised for the village in public good that they could spend however they wanted, they raised 84% of the total possible in the all women groups and about 75% of the total in the mixed communities and interestingly this was not because, I mean one possible explanation would be that women just give or just more community spirited unconditionally or irrespective of the composition of the groups but that was not the case. They in the mixed communities, the women contributed about the same amount as the men did. In fact, marginally less. It was they only contributed more when they knew they were playing with other women and here's at the individual level, the percentage of the 300 contributed on average was about 82% for the women in the all women communities. It was about 74% for women in the mixed which is slightly less than what the men contributed which was about 75% in the mixed community. So the goal of the paper is to try to explain this pattern and what we do is we use surveys of game players that we gave them after they made their private contribution decision and a very simple structural model to try to estimate the different weights or motivations we'll call them that participants put on different considerations in what I'll call conditions, three conditions. Women players in the all women communities, women players in the mixed communities and men players, male players in the mixed communities. And one of the things that's been interesting for McCartan and I is been, in general, there are great advantages to randomized controlled revolution or treatment assignments but one often comes up like, okay, we know with high confidence that X caused Y but then the question you get is like, why did it do that? And this is much harder to answer and this is one kind of approach that's maybe, I'll be interested in your reactivist of how convincing or successful you think it is. So to summarize what we think we found and the paper that's distributed is we'll change some, I'm sorry for it being late, we really got wrapped up and stuck on some things with the estimation and learning how to do this but women in the all women groups seem to have had a higher value for contributing independent of their value for the public good, their concerns about matching what others were doing and independent of any fears about that their contributions might not be anonymous and perhaps social or other punishment might be involved. We think that the best explanation is that many participants in all the communities thought that what we were doing was providing, it was giving them some kind of test of community spiritedness and that this game was a mechanism for figuring out whether how good was this community at raising money and that perhaps they probably expected that, well, if we do well on this test, maybe they'll bring us more money. And in line, that was pretty clearly, I'd say a broadly held understanding of what was going on here and what I'll argue is based on the estimation, what we'll show you is that it looks like the women in the all women condition put more weight on signaling to us that they were good, that this was like, we're community-minded. And we speculate with some evidence, we think that this may be the result of a social identity effect, where that when told it's gonna be women chosen to play, it got people thinking themselves as part of what you could call a team women of the village, the women representatives of the village, as opposed to in the mixed treatment, it was maybe more that they just thought, well, we're just like random village members. And so there'd be a motivation to perform well on behalf of the village perhaps, but perhaps the identification as the women of the village was more motivating. Okay, so to give a little bit of background, way back in the early 2000s, Jeremy and I actually started talking to International Rescue Committee, which was interested in doing where they had hired some people they wanted to do a better job on rigorous evaluation. We worked with them to put together this randomized controlled trial for their CDR type programming, community-driven reconstruction, and randomly got eventually this worked for the stiffed funded project in Liberia. And we implemented this random assignment of their program to 43 of 82 possible villages. The goal of the CDR program was post-conflict democratic institution building at a local level to try to, with a goal of increasing social cohesion and collective action capacity, as is kind of typical with CDR programs, the premise was that civil war had destroyed local institutions and made for lots of bad blood and hatreds and so on, and that it was important at the village level to have programs that would lead people to work together and also there was something of what's also typical in these programs, something of an anti-chief or a premise that the local institutions are bad authoritarian and so you're introducing these democratically elected community development communities and so on. But that's for, that was another part of the project. We evaluated the impact of the, or tried to evaluate the CDR program with surveys, baseline and after the implementation and also this more behavioral real life collective action problem that was intended to test whether they actually did improve their ability to raise funds for a community project. And the CDR stuff was published in two papers, one in the papers and proceedings of the AEA early on, it was just kind of like, here's the main effects and a little bit of speculation about that and then a much longer paper that finally we got out in APSR in 2015 that went into the mechanisms and what might have explained the main result, which was that it actually did work, which was surprising to us. Or what we found was people raised more money in the CDR treated communities than in the control communities. Okay, so as I mentioned, but out of interest and because of doubts about whether the CDR program would do anything, we had built in a couple other treatments and the one I'm talking about here is gender composition treatment. And this came from an interest in whether how gender composition might affect collective action capacity. As I mentioned, we didn't have the resources for an all men treatment, which is unfortunate. To go over the way the game worked, I'll try to do this quickly to get to the results. We first would go to the community and have a community meeting where we explained that they could receive up to around $420 for a development project that they could choose, the community could choose what to do with it. They were pretty much totally free. We didn't constrain them there. What they would need to do was, we were gonna come back in a week, randomly sample 24 adults, give them an endowment. They would make these contributions privately. And what communities also had to do was when we came back to play the game, they had to fill out, have a form filled out that would indicate two projects that they might use the money on, although there was no constraints as to whether they actually do that. And also indicate three people who we called community representatives, which would be who we would give the money to at the end of the day. And we did interviews on game day with the community representatives and the chief, if possible, and the game players. So one week later, we would come back, collect the form, sample the household players, play the game, go have another meeting where we count out the money and provide it. So an important thing to note here is that the village had a week to spread the news, organize, discuss the game, how they would approach that, I mean, if they wanted to. But during that week, they didn't know who was gonna play, who would be picked to play, so it wasn't like they could lobby or pressure or course the players because they hadn't picked yet. So a little more detail on, as I mentioned, the main thing is 42 villages where they knew that 12 men and 12 women would play. The other ones, it was gonna be, they knew that it would only women would be chosen. Mentioned, everybody got this, the players would get 300 Liberian dollars and half of the players knew that their contribution would be multiplied by two and the other by five, the other half by five and we would conduct the surveys. Okay, so main results, so what did we find? So here we've got, this is what happened in all women groups, this is what happened in the mixed groups divided by the women players and the 12 men players. And what you can see is that there was high levels of, this is zero, contributed zero, 100, 200, 300. Most people contributed the whole 300 and the rate was higher in the women's groups, the average for everybody being 247, which is more and statistically very clearly significantly differently than what you had in the mixed groups. And you can see the men and women are pretty similar in the mixed groups in terms of what they were doing. So this is just showing you the statistical significance and so on. So the treatment effect among women while comparing women in mixed to women in all women groups was like 27 Liberian dollars on average and can reject that it was, that's just by accident. Some other interesting patterns here. What about the interest rate effect? Well, it turns out that the women responded to the interest rate like quite significantly, whether they, regardless of which group there and the men paid no attention to the interest rates. So the women contributed in the all women groups, they contributed on average about 24 more Liberian dollars if they had the higher interest rate, pretty similar or slightly less for women in mixed group, the men just don't pay any attention to the interest rates. Another interesting pattern, we asked them after in the survey lots of questions, but some of the questions were about their expectations about what other people would do in their community. And we asked a couple of questions from which we can produce an estimate of what the person, the expected average contributions of others in the playing the game in the village were. And what we see is generally speaking, we'll call it over optimism. There's other ways of interpreting it, but people generally said people would give more than what the actual community average was. The women in mixed on average expected average contribution in the community of about 273, the actual average was less than that. There's slightly greater over guessing or optimism in the women in the mixed communities, similar a little bit less, but similar for the men in the mixed communities. So everyone was a little bit over optimistic. Further, the expectations correlate with the actual. So even though everyone's a little optimistic, there was to some degree rational expectations communities that were communities just to an extent, correctly predicted whether they would give more or less. So let's see. And then this is a little bit different. So we also asked a question that said, who do you think will give more? Will men give more than women? Women give more than men or will it be about the same? And interestingly, everyone thought women would be more generous pretty much. So, and with the highest being women in all women, said 83% said women would give strictly more. 73% of women in mixed said women would give strictly more and almost half of the men said women would give strictly more. So with the rest of them, the other 50% divided between saying it would be the same and less. Okay, so how to explain these patterns? What we try to do and what's been very interesting for us is we try to use a simple structural model of we're starting with the game player's decision problem where we're gonna think about contributions being zero, one, two, three and hundreds of Liberian dollars. And we're gonna separate out in this representation for considerations or motivations, which I'll talk about in a minute, but I wanna stress first, some of these terms refer to data, things that we have measured in one case with the interest rates randomly assigned and some of them are parameters that we're gonna wanna estimate. So this is I's utility for the contribution XI, it's gonna depend, this is like the total amount raised in the community and you can see this is I's part of it, the interest rate times what they contribute. So this is like a linear value for the public good or whatever the money gets spent on. We're gonna have a consideration that bears on like wanting to match what other people are doing in the community with GANLA being the weight put on a parameter indexing like weight put on matching what other's doing and row being kind of a target, E sub I is the I's reported, the survey based measure of expectations, so that's data. Where we have a measure of, we have asked a couple of questions about are you concerned or do you think that other people will be able to find out what you gave and we're gonna use that as, which is correlated with giving in general. So like the more people said, I think other people might find out, they tend to give more. So we're including that as like in an attempt to estimate weight put on fears of punishment. And then finally, like value for own use, like disutility for not being able to use the money for your own purposes. That's this last term with the parameter to estimate being alpha, which could be positive or negative. This is gonna combine both the negative aspect of not having the money for your own use, but also could be just like, well, I wanna signal, contribute to signaling that this is a good community independent of the money raised for the public good or matching what anyone else is doing or worries about anyone punishing me or disapproving of the results. Okay, so I basically just went through this, a couple of other things to note, for identification, we need to fix something here. And what we've done is basically fix the value of the public good at one relative to the other considerations. So in the estimates for like alpha or gamma or so on, you should think of that as multiples, like two times the value that's being put on public good or three or whatever, let's see. And we have some kind of a priori constraints from the model that gamma needs to be positive, rho needs to be positive, yeah. Okay, so that's the basic decision problem at the individual level. So what are we gonna do with this now? Well, I'll refer to these parameters, the matching gamma, the target, what do you wanna match exactly? Would be one if you wanted to try to be exactly where everyone else was or maybe you wanna get a little bit less than what others are doing or a little bit more, that's rho. There's a fee is the concern about punishment, alpha, your own use or signaling value. What we wanna do is estimate the average motivations across these three conditions, women in all women, women in mixed, men in mixed. And we can then interpret the average parameters in all women groups versus the women in mixed as a treatment effect. And we can compare and think of these comparisons to the men as interesting comparisons. You can't really think of them literally as treatment effects because you can't assign men to an all women's group and have them be women. So okay, so what's our approach for estimating this? What we're gonna do is assume some of these are plausibly constant within the village, although they may vary across villages and we're gonna set gamma, rho, and fee as the ones that are constant and we'll let individuals value for own use slash signaling variation have heterogeneity within the people selected to play within the village. And so we're gonna assume that's a random variable that varies across individuals within communities and has a mean that's village specific but also condition specific. So it can be different for men and women in the mixed groups for example and also has a condition specific standard deviation that also can be specific to the village. Well, given that you can derive a likelihood very, it's very straightforward that I chooses 0, 1, 2, or 3 for a given parameter set and given data under the assumption that I is maximizing this utility function. That's easy enough to do but then estimating this is not so easy and what we've done is try to learn how to do this with evasion model using this R, well another platform thing called Stan that Andrew Gellman and collaborators put together. So that's what we said we've been, spent a lot of time figuring out how to make this work but we're more or less there and what we find is here these are, so there's actually two, well we've been arguing back and forth and trying to figure out what the, there's several different ways you could do this. I'm gonna show you what we have. I'm gonna show the one I marginally prefer although I'm not sure I have a strong view as yet as to what's best but what we're doing here is to estimate the model separately each village by village. So there's only 24 people and there's only 12 people in the mixed conditions so one of the reasons for using evasion approach rather than just like maximum likelihood is there's not much data and with evasion we're using a likelihood function but we recover a whole estimated posterior distribution. It doesn't really matter whether you have 12 or one for that matter but we're 24 for being able to come up with estimates. So anyway, what do we find? So these are the alpha remember is the own use value and what we find is that in the all women communities they had lower, their value for, as compared to women in the mixed communities they put less value on own use and more value on other considerations possibly or we think including kind of signaling that this is, we're a good community. The women in the all women communities also put slightly more weight on concern about non anonymity. It's, they're pretty similar on to the mix both men and women in mixed on the way put on matching and the target what they want to match. And the height of these bars says that matching is at least in this estimation comes out as a pretty significant consideration relative to the public good. Three times their value for the public good. Three or four times by these estimates. So I think I just summarized what I said here. It is a bit odd, one of the odd things here and this can vary depending on exactly how we do this but the row is saying that individuals on average want to give more than what they think other people are giving which is arguably kind of odd. And maybe soaking up in the other way which I won't really present but a more pooling random effects type approach in that estimation this comes out smaller and more let's say reasonable slightly less than one but that is associated with this estimating higher in fact positive alpha weighted on value for signaling or minus own use which is a small relevant public good and in that in those destinations pretty much regardless how we do it the comparisons of all women to mixed and actually the ordering for each of these things comes out the same and the only thing that varies is kind of how much they are relative to the public good and each other. Okay, let's see. So yeah, so statistical significance. The alphas are significantly bigger in the all women treatment as are the fees that equals 0.08 where there's not much difference in the gammas and rows. So okay, so to finish up at the community meetings in the all women when we explained that only women could be chosen to play I saw this myself and you'd see the women at the meeting sometimes kind of sit up in their seats and appear like noticeably like pleased and in a way like, okay, we're gonna be the ones who play men would sometimes kind of wander off at that point and as I mentioned people I think pretty clearly interpreted what we were doing I mean, they were confused. This was pretty bizarre. I mean, not entirely bizarre because they're used to the idea or many of them were used to the idea of like matching contributions and so on but the game aspect was a little unusual. Anyway, they interpreted as a test of community spiritedness and possibly probably thought that if they did well, we'd bring more money. So based on the estimation results and those kind of observations my best guess is that the women felt like in the all women condition felt like they were like the representatives of the women of the village and that was more motivating than thinking of themselves as just like any members of the village. It could also be that they have stronger network connections with each other so that in the all women condition they had more concern that a poor overall performance would lead to more social punishment or disapproval whereas in the mixed, the women would not be seen as specifically responsible as women. And that's consistent with us getting tending to get a higher weight on the fee in the estimations. So to conclude, we don't find evidence that women are unconditionally more inclined to contribute, participate. They're not just like women are, it's not like they're just better in terms of community-mindedness. Rather, it's conditional on working with, knowing they're working with other women and not men which is intriguing. It's not totally clear what it means. We're proposing, it looks like, our guess is that there's some kind of a social identity effect going on here. Does this have any implications for policy or development practice? I don't know. Maybe you could imagine designing interventions where we make more, where the development organizations make more explicit appeals to the pride of identity groups within the communities being addressed. To specific sub-identity groups, in particular women, could be used to try to motivate participation and or contributions or behavioral change. But it's clearly, this is tricky. You don't want to worsen or create divisions and there's a small literature which finds that interventions that might do that can lead to backlash. I'm mentioning one here, a couple have come up in previous presentations, but there's a nice paper, I think by Jessica Gottlieb, where evaluating a program which was designed to provide information about basic local government functions and ability to provide funding for local public goods that required that women be present in these meetings. And what she found was that after that, she had various measures of what happened in villages after this that there did appear to be something women appeared to voluntarily do less and participate less. And her best guess as to what was going on was that their presence or their involvement in the project had created a certain amount because it was in violation or different from the traditional norms had kicked up some resistance that they were concerned about. Okay, so I'll stop there.