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.
=

# Happily Under-Insured Ever After? The Role of Beliefs in Household Specialization

Authors: Ana Costa-Ramón, Ursina Schaede, Michaela Slotwinski, Johannes Stupperich  
Discussant: None  
Video: https://www.youtube.com/watch?v=1kb3a99sA-E&t=9865s


## Talk (02:44:25 – 03:26:49)

[02:44:25] [applause] Great. Thank you so much, Katarina. Next up, we'll have uh Ursa, who will be telling us about happily underinsured  
[02:44:39] ever after. >> Okay,  
[02:44:58] great. Thanks so much uh to the organizer for putting us on the program.  
[02:45:02] Um this is a project with a fantastic team of co-authors. Two of them are in the room today. So Anna Costa from Zurich is sitting right here and Johannes who's our former RA and now a PhD student at Stanford is hiding in the back of the room which he shouldn't be  
[02:45:17] doing. [laughter] And um Mika is uh um also in Zoric and Nhatelin is joining us uh on Zoom today.  
[02:45:26] All right. Um in many industrialized countries uh households specialize after children get born uh with mothers uh typically reducing their labor force attachment.  
[02:45:39] Uh and on the one hand this kind of specialization may in fact uh maximize joint household income especially if we think that there are kind of convex returns to investments in the labor market. But on the other hand, the  
[02:45:53] specialization might also imply a financial risk for the lower earner in the household if it ends up that the couple ever separates. And this financial risk arises because in most countries typically after a couple gets  
[02:46:06] divorced, the lower earner no longer partakes in the career returns that acrew to the higher earner post separation. or at least they don't um sort of partake in those returns to a similar extent as if the couple had  
[02:46:19] stayed married. Of course, um if we live in a world with perfectly rational agents, um couples who decide to have a kid and how to specialize would sort of correctly know what their separation likelihood is and  
[02:46:34] then sort of make the specialization decision uh based on on that risk and sort of correctly ensure that risk.  
[02:46:41] However, there's sort of two um empirical pieces that kind of hint that this might not be the world that we live in. Um on the one hand, psychologists have shown um in studies already in the 1990s that people tend to have relatively overly optimistic beliefs  
[02:46:56] when it comes to assessing their own separation likelihood relative to say a national level benchmark. [snorts] And then second, um, across many different countries, economists and other social scientists have shown that women relative to men tend to face much more  
[02:47:11] pronounced financial challenges after divorce, which at least to some extent suggests that the couple hasn't managed to smooth consumption and income equally for both sides of the couple if the state were to occur. And so in this  
[02:47:24] paper, we're trying to kind of get at this question headon and we're trying to ask um whether couples specialization choices properly account for the risks of separation or whether it might be the case that sort of overly optimistic  
[02:47:38] beliefs about happily staying together ever after leads women to kind of oversp specialcialize in home relative to market.  
[02:47:46] The way that we try to operationalize this question is that we ask whether beliefs about the risks of separation affect the household specialization and insurance decision. And what we're trying to do here is we're trying to shock uh these beliefs about the risk of  
[02:48:01] separation in a large scale field experiment.  
[02:48:05] We run this RCT. Um, let [laughter] let me just conclude with like three slides of intro and then I'll I'm I'm happy to wave the 10-minute rule. Great.  
[02:48:17] But if things are unclear about the setting, you should ask me now. Um, all right. We're running this experiment in Switzerland with about 1,500 female teachers. They're all in a relationship and they all have specialized in the sense that they have kids and they all work part-time. So the  
[02:48:32] average woman uh in our sample, this is actually quite similar to the Netherlands, uh works two days uh per week. Our treatment is going to consist in a documentary style video with real protagonists. And the goal of the video  
[02:48:45] is to make uh the probability and the consequences of divorce relatable while at the same time offering but in a relatively subtle way some pointers on mitigation of financial risk. And what's nice about our setting is that we can  
[02:48:59] run our our RCT and collect the survey data as we want and then we can merge it to uh personnel records of the employer to observe these women's work hours after the treatment uh one year later.  
[02:49:12] I'm going to show you that our treatment increases insurance against the risk of separation through two main channels that should both ensure that the wife has um has a better consumption or income smoothing around the time of  
[02:49:27] divorce if that should occur. The first one is a financial channel where couples could save more today or set up compensation payments within the couple uh to make sure that the wife is better financially insured in the ca in the  
[02:49:40] case of divorce. And we see that um uh treated women score.3 standard deviations higher on a savings index.  
[02:49:48] That's across some survey measures of plans to save and compensate. And we also see that the treatment um is 55% more likely relative to the control group mean to sign up for a compensation tool. Um this is incentivized and this  
[02:50:01] tool sort of lets them calculate the magnitude of compensation payments based on prior work history and future plans.  
[02:50:09] The second channel to um ensure uh against uh the risks of divorce is if the wife already today has a higher labor supply and therefore a higher income if divorce were to occur in the future. And we do also see some movement on this channel. Um we see that the  
[02:50:23] treatment group is 2.5% over the control mean uh sorry increases their work hours by 2.5% over the control mean.  
[02:50:33] One dimension of heterogeneity that we were particularly interested in was looking at whether how well insured against the risk of separation you are at baseline matters for your response to our treatment. And we do find that women  
[02:50:46] with lower levels of insurance, if divorce were to occur to them tomorrow, um, move more strongly on this labor supply margin, they increase their hours by 6.5%.  
[02:50:58] What's the implications of our findings? We think that um these findings kind of show that the currently low levels of maternal labor supply that we observe in the real world are at least in part the  
[02:51:11] result of women underinsuring against the potential risk of separation.  
[02:51:17] All right, I'm going to wave the 10-minute rule now. Uh [laughter] so Sonia, did you want to ask a question?  
[02:51:33] the the risk of um separation is also a function of how much you invest in domestic life. So it's a function of household specialization.  
[02:51:44] >> It could be. Yes. >> So can you look so in in your outcomes I guess what's relevant is can you look at whether they're more likely to separate?  
[02:51:53] >> Yes. So, we can't look at separation in administrative data, but we ask them um in a follow-up a large question battery about their happiness with their relationship in many different ways. And if anything, the treatment actually makes the relationship more stable. Uh  
[02:52:08] and as you'll see in the result, the treatment also makes them the treatment group much more likely to seek actually a conversation with their partner um about this topic. But yeah, thanks. All right.  
[02:52:21] Um we contribute to three main strands of literature with this paper. Um first um a lot of amazing scholars in this room have studied the legal and economic uh environment and how that affects the marital allocations of couples. Um in this paper we try to contribute to this  
[02:52:36] by emphasizing that beliefs also seem to play a role in how uh households specialize.  
[02:52:41] Second um a large literature is centered around explaining why it is that mothers work less when children arrive. Um in this paper we're trying to look at maternal labor supply within this household level arrangement and we're trying to document that sort of over  
[02:52:55] optimism about how long this arrangement will last may lead women to over specialize in home. And then third a large literature in behavioral economics is on non-standard uh beliefs and we apply this to family economics. What's  
[02:53:09] nice about our setting is that in a real world context, we can kind of shift these beliefs and then measure uh a high stakes uh real world outcome and show that that can move.  
[02:53:20] All right, this is a road map um of the talk. I'm going to start by giving you a little bit more context information about uh female labor supply in Switzerland and the legal framework of divorce.  
[02:53:34] Similarly actually to the Netherlands. So it was great to have the prior presentation. Um a lot of mothers work a lot of moms with kid kids participate in the labor force but most of these mothers work uh part-time. So part-time  
[02:53:47] is very prevalent in the Swiss context. The other thing that's sort of useful to keep in the back of your mind as we think about the implications of divorce is that Switzerland is a country in which um pension is uh the large part that ensures your standard of living  
[02:54:02] during retirement is directly proportional to salary. So what you're going to pay in during your time in the labor market is sort of very closely tied to what you're going to get out.  
[02:54:13] We're working um with female teachers and um which is sort of a a useful sample for us. Um but I just want to highlight that they look very similar to average mothers especially to average mothers with tertiary education. So  
[02:54:27] similarly to the uh average mom in Switzerland, female teachers experience large decreases in their work hours. So in their employment level around uh the childbearing ages. One thing that's nice um uh about uh working with teachers is  
[02:54:41] that they have relatively low hurdles to increase their work hours. Um so we conduct the experiment to during a time of teacher shortages meaning there should be low demand site hurdles to adjust labor supply and we time the  
[02:54:55] intervention at the beginning of the yearly planning period where teachers try to figure out with their principles how much they're going to work uh in the next school year. I think there was a question. Yeah. Is there a particular age group for these teachers? Are they  
[02:55:07] still reproductive ages? >> The fertility decision. Yes. So unlikely. These uh the average age in our sample is 41 and the age of the  
[02:55:20] youngest child is I think 7.5. So we think by and large these teachers have likely completed fertility. Yeah. Um Anelie is a setting where there's a  
[02:55:32] >> great question. Yes. Um, so there's no official statistics. The best thing that we have is estimates from lawyers who write these contracts. And those estimates say that less than 2% of marriages have a prenup agreement. Yeah.  
[02:55:45] And then one in the back. >> Sorry, I was gonna ask given a%.  
[02:55:56] >> Thank you. No, great segue. Perfect. [laughter] This is really important and it's also going to matter from country to country.  
[02:56:04] So I'm I'm also happy to to talk about how that uh validates against other settings. So in Switzerland the divorce rate is similar to the US at 40%. Uh the divorce patterns among teachers are similar to those uh with tertiary um  
[02:56:17] education and the legal framework is such that any assets including any pension savings that you accumulate during the years in which the couple is married are going to get split in half at the time of the divorce. Okay.  
[02:56:30] However, Switzerland is also a context um even more strongly than the Netherlands where there's as good as no spousal alimmony. So what that means is that while there are child support payments for which actually both parents  
[02:56:43] are financially responsible, there's no payment like no spousal maintenance that would go from the former higher earner to the lower earner for their own financial uh stability. So you're just  
[02:56:56] responsible economically for yourself and prenups and notoriized contracts. Of course, you could make such a a contract uh and exempt yourself from those overall rules, but they're very rare in  
[02:57:08] practice. Um, not for nothing, in Switzerland, similar to other countries, we do observe relatively large gender gaps among divorced parents and those come from there being relatively little catch-up to overcome these sticky  
[02:57:23] specialization choices that happened during marriage. Um so while we do see that uh divorced mothers work more relative to the mothers who would who stay married, they still tend to work a lot less than divorced fathers. Part of that is because the children tend to  
[02:57:38] stay much more with the mother than with the father.  
[02:57:41] Um the households of divorced mothers about a quarter of those are classified as having low financial resources. Uh just in contrast, 8% of household of divorce stats or 8% of households with kids in general are classified as low  
[02:57:55] financial resources. And despite the fact that the gender g sorry that the pension gets equalized at the time of divorce because often divorce happens quite some time before people retire, we still observe pretty large gender gaps  
[02:58:08] in in terms of pension receipt of about 30%.  
[02:58:15] All right. Yeah. >> So the division of the pension is administrative.  
[02:58:24] So there's no enforcement issue. >> There shouldn't be an enforcement issue.  
[02:58:30] But that's a good question. Yes. Because this is um these are accounts that are held by the government or by pension funds. Yeah.  
[02:58:37] >> And so it just so so there's a decision in the court and then the pension gets split.  
[02:58:43] >> Correct. Yeah. Yeah. or the couple can decide outside of the court, but this is sort of what gets applied. Yes. Uh there there could be if there's sort of child maintenance payments there, there might be delays and and people sort of not  
[02:58:57] paying. Um yeah. Okay. Great. >> Yeah.  
[02:59:04] >> Is there a um an age requirement that I don't know about duration of marriage to have a right to your husband's pension even if you divorce in? Um so for the pension it's really just the number of  
[02:59:16] years that you're married. Um so we're doing this study after like five years where the federal um the federal court has significantly lowered the spousal alimony claims. It used to be the case  
[02:59:30] earlier that they would the court should take into account the duration of the marriage to decide if there was claim for spousal alimony. But even in the early 2000s um I think less than 15% of  
[02:59:43] what of mothers of kids got awarded spousal alimony. Yeah.  
[02:59:48] Okay. All right. Before diving into the results of the RCT, I just want to spend one slide um going through a conceptual framework to look at what happens to household specialization and savings  
[03:00:02] when couples revise their beliefs about the separation risk.  
[03:00:07] For that, we set up a very simple toy model. It's a two period household model that features a wife who's the lower earner and a husband who is the higher earnner. And the household in period 1  
[03:00:21] makes a choice over two main variables. They choose how much to save in period 1 to transfer over to period 2. And they're going to decide what share of the wife's um time goes into the labor market. with the rest of her time, the  
[03:00:35] wife's going to produce a household public good that benefits both spouses.  
[03:00:39] And we think and model tow think of tow and model tow as an adjustment cost. So it's going to determine the wife's income in both periods. What that means is that you can think of this as um uh human capital depreciation for taking  
[03:00:53] time out of the labor market or uh habit formation. For example, with probability five, uh the couple gets divorced in period 2. Um in which case savings are going to be split equally, but each and each spouse keeps  
[03:01:08] their own income kind of reflecting the Swiss uh setting. So what happens in this framework if the household shifts upwards their belief about the likelihood of getting separated?  
[03:01:20] What that means is that the divorce state in period one becomes more likely and gets more weight in the household's maximization problem. And this is uh the state in which the wife has lower consumption. So the household will react by wanting to better ensure the ri the  
[03:01:35] wife against this consumption drop. And as I mentioned they can do that through labor market investment and savings. The prediction of the model is that when FI goes up, the wife's labor market investment always weekly goes up.  
[03:01:47] Whereas the savings response can be ambiguous. And the reason that that's the case is that labor supply provides a very targeted form of insurance because um any income in period 2 that the wife makes will acrue directly to her own.  
[03:02:01] Whereas savings are going to get split at the time of divorce and are therefore essentially tax taxed. So it can be the case that labor market investments is such an effective insurance channel that  
[03:02:13] it can crowd out a savings response. In terms of welfare, um what we show is that raising overly optimistic separation beliefs in a couple should increase that or increases the couple's  
[03:02:26] x anti-expected utility. Of course, exposed uh it could lower it. Um, but there's going to be a distributional impact in the sense that it's going to shift expected consumption utility from the higher to the lower earner within  
[03:02:39] the couple. Yeah, Emily, >> maybe I got a little lost, but I I thought you said that the experiment is actually going to make the couples more stable. So, kind of lowers it's it's kind of this I'm having to trying to think of like maybe see what I forget Greek letters goes up, but then it  
[03:02:54] actually goes down in the actual results. And why why not also model the knowledge of how insured I am as opposed to the divorce probability.  
[03:03:02] >> Sorry, the last point I just >> so you're you're modeling right now an increase in probability of divorce.  
[03:03:07] Whereas you could also model >> belief about the probability of >> but you could also model as a change in the belief about how insured I am.  
[03:03:13] >> Yes. >> Which then if the relationship becomes more stable, all the results would still go through.  
[03:03:18] >> Yes. So in the paper we actually have um uh we have a parameter so we can uh vary how much income each spouse keeps in the case of divorce. We show that if you shock the belief about that the predictions uh go exactly in the same  
[03:03:32] direction but that's a great question. Thank you. Yeah. Yeah.  
[03:03:39] Um is there any room for heterogeneating beliefs? uh and do you have a sense of uh before your experiment uh maybe uh women are more or less optimistic than than husbands and if that matters for the predictions?  
[03:03:53] >> Yeah, thanks for bringing that up. Um so there's sort of maybe two things I I want to mention that your question sort of alludes to. So the first thing is that we don't know the correct what  
[03:04:07] should be your correct separation belief, right? That's inherently unobservable. it's going to be unobservable to the individual as well.  
[03:04:14] What we can show in this paper is that on average we have good reason to believe that these women are too optimistic relative to a benchmark um about their separation likelihood. And I can show you a directional effect in the  
[03:04:28] RCT that when I shift these belief upwards um then people rearrange their specialization decision. Um, but I'm not going to be able to sort of talk about  
[03:04:40] optimality more generally. Yeah. >> You know, I appreciate what you do for simplicity. Yeah.  
[03:04:48] >> But in a bit more complicated model, that probability of divorce will go up if people expect there to be more divorce the following period. Because if  
[03:05:02] they work more, women are more willing to walk away because they have greater insurance.  
[03:05:08] >> So that would say in a world in which you could observe the couples and say who's the one who's walking away, it would say that women would be more likely now to be the ones who walk away.  
[03:05:18] >> Yeah. >> And Okay. So >> I'm just saying >> this is very simplifi this is very simplified and it it's mostly just to kind of fix ideas about the parameters that we're talking about. point well noted and if you have suggestions on how to extend that we're up very early.  
[03:05:31] >> It'd be interesting to see just heterogeneity in terms of you know >> the prior divorce belief.  
[03:05:35] >> Yeah. Transfer [clears throat] TV has she has money and even a model of bargaining some utility as long as you  
[03:05:49] don't transfer. >> Okay.  
[03:05:53] >> Thank you. >> Yeah. No no thanks. Point well taken.  
[03:05:56] And sorry to go back actually to your question, one thing to mention here as well is that we were incredibly concerned about contaminating the control group. So we didn't actually elicit baseline beliefs. Uh we also found that in pilots when you ask people straight out what they think their  
[03:06:10] divorce likelihood is, everybody's just going to tell you zero. So I'm going to show you the ways [laughter] in which we try to get at that. All right. Um let me talk a little bit uh about the data and the design. Um, we  
[03:06:24] recruit teachers in two cantons in Switzerland through their employer.  
[03:06:28] We're going to do an individual uh level randomization that's stratified and we end up with a main analysis sample of,438 female teachers. They're each between 25 and 50. They all have a partner, but they're not necessarily all married and  
[03:06:42] they all have kids. They all work part-time. You can see the timeline here. So we have a baseline intervention and endline at the end of November 2023 which is just at the start when teachers start to bargain with their principal about how much they want to work next  
[03:06:57] year. And then 3 months later we do a follow-up uh that has about a 64% response rate. Um and one year later we can observe these teachers employment level in the administrative data.  
[03:07:10] I'm going to show you the treatment material in the next uh slide. It consists of a documentary style video of about five minutes. We hired a journalist from Swiss public television to produce this video for us. And the  
[03:07:23] video just shows these protagonists narrating their personal story. There's no additional commentary um or something like that. And our goal was to make both the possibility and the potential consequences of divorce more relatable  
[03:07:36] um and offer some pointers on how to mitigate risk. We have an active control group that watches public television videos of similar duration. We have one control that looks at mental health helpseeking for adults because we wanted  
[03:07:50] to have a bit more of a sensitive topic and one topic that's completely unrelated. So how kids process images on screen when we do treatment effects against either one of those control groups, there's no much difference.  
[03:08:03] Okay. Um, for privacy reasons, I switched the faces and changed the names of our protagonist. Um, the video features a woman uh named Sandra who recently got divorced. Um, she has two  
[03:08:17] kids in her early teens. And when she became a mother, her and her ex-husband um decided that they wanted the kids to grow up at home with a lot of nurturing and love. And so they decided that she would work two days a week. Sorry, did you have a question, Laura?  
[03:08:32] >> I have a clarifying question. Sorry. Yeah.  
[03:08:35] >> Sorry. So, this one mentions finances and it sounds like maybe your control videos don't and I I that's my clarification here.  
[03:08:42] >> Yeah. Yes. >> So, part of your treatment could just be the mentioning of finances in general.  
[03:08:47] >> That matters. Let me think about that uh a little more. Yeah. Um I think that that could be a concern. Um thanks.  
[03:09:00] Um, all right. So, I I think maybe a better answer to your question is that I'm going to try to show you that their beliefs also  
[03:09:12] move. Um, yes. So, in that sense, yeah, thanks. Okay. So, uh, this woman says that about divorce, she always thought that it only happens to other people.  
[03:09:24] Although one has of course the stat the statistics in mind and in the end it was actually her who decided the relationship was no longer working for her. She mentions the financial implications for her and that there's a clear difference between her and her ex-husband where he essentially  
[03:09:38] continues on his path in the labor market but for her the financial burden remains for years to come. And she also says that she was absolutely not aware that when the couple jointly made the specialization choice that she would pay  
[03:09:51] the financial price of it. We wanted to contrast that with a [laughter] a couple who's still happily together, but who's actively trying to ensure this  
[03:10:03] risk. Um, so Claudia and Thomas um also have two kids and they actually also specialized when the kids arise arrived, but they set up a notorized compensation agreement and they talk at length about  
[03:10:17] this agreement. So Thomas compensates half of Claudia's lost salary and all of her lost pension income. And sort of in the video they together emphasize the importance to proactively discuss this topic in a relationship when things are still good, not at the point where  
[03:10:32] things are getting sour. All right, we're going to run very simple um regressions where we look at an outcome with a treatment indicator.  
[03:10:40] We're going to use PDS lasso to select controls um and have straight a fixed effects in there. we're uh balanced at baseline in the follow-up and the administrative data. And we can also show that there's no selective attrition uh by treatment status interacted with  
[03:10:55] baseline uh controls for the follow-up and for the admin data. All right. Now I can't see the timer anymore. All right.  
[03:11:03] Um motivational evidence. Um before showing you the results, let me just show you why we think that our sample has relatively optimistic beliefs about uh their own relationship. And as I mentioned, we're going to try to do that  
[03:11:16] in two relatively indirect ways. Um, the first way we try to get at that is with a list experiment where we randomize participants into seeing one of two questions.  
[03:11:29] The first group of participants um sees a question that asks them to think about the four closest couples that they know and what how many out of those they think will divorce over their lifetime or separate. and the other group sees  
[03:11:43] the same question but is asks about the four closest couple plus their own relationship. By comparing the counts between those groups, we can back out an indirect measure of own separation likelihood.  
[03:11:54] Let me first show you the national average. So um our participants are around age 40 on average. So what I'm showing you as the national benchmark is if you're sorry conditional on still being married at age 40, what's your  
[03:12:09] divorce likelihood? that is 27% and we see that participants think that that's exactly actually the rate at which their friends will divorce. When it comes to their own divorce likelihood, we sort of back that out from the group wise  
[03:12:22] comparison. We see that that's at 16% and significantly lower than the national level benchmark. The second way we try to get at the divorce likelihood because we also wanted an individual level measure is that we give respondents after the treatment. So this  
[03:12:37] is only using the the control group a question battery about different scenarios in retirement and we asked them sort of what do you think is the likelihood that one of your kids will still live with you but in that we kind of hide some questions that ask them about the probability of still living  
[03:12:51] with their current partner. If we use that question we see that about 14% of the sample uh sorry that the probability of not living with their current partner at the time of retirement uh is 14%. So relatively close to what we get out of  
[03:13:04] the list experiment. >> All right, diving into results. Um yeah, >> very quickly um maybe from administrative data you could get the incidents among teachers mean more educ I mean the friends help but unless the  
[03:13:18] friends are their colleagues. I mean is it a selected group in many ways is educated and pro-social and so >> oh you want me to compare the teachers to the 27 >> national average doesn't tell me very much.  
[03:13:30] >> Fair enough. Yeah, we can do that. Thanks.  
[03:13:35] All right. Okay. I'm first going to show you insurance along financial channel channels. Um these are all elicited in the survey. Um we have a sign up for a compensation tool that's incentivized by participants entering a lottery and then  
[03:13:49] deciding um sort of if they win, if they want to take the full amount or pay some amount to get access to this tool. We see that 23% of the control group sign up for this compensation tool and that increases by about 13 percentage point  
[03:14:03] for the treatment group which is a 55% increase over the control group mean.  
[03:14:10] We also ask participants to provide both their own and their partners' email address to facilitate access to that tool. And we actually see similarly sized treatment effects for people putting in a valid email address both  
[03:14:23] for themselves and their partner. We also ask in the survey whether they plan to set up compensation payments within the couple or whether they plan to save more. We see a treatment effect for compensation payments, but we actually don't see a significant  
[03:14:38] treatment effect for plans to save more. Um we combine the bold printed outcomes as prespecified into a savings index and we see that the treatment moves uh this by.3 standard deviations going to labor supply. What I'm going to  
[03:14:55] show you here as an outcome is the change in the employment level in the academic year after the intervention minus uh the employment level in the year of the intervention. So before the treatment what we see is the control  
[03:15:10] group slightly increases their hours by uh less than 1 percentage points and we see that the treat treated teachers increase their work hours by 1 percentage point which doesn't uh seem like a very large number but relative to  
[03:15:24] the control group mean so these people work around 40% um amounts to about 2.5%.  
[03:15:32] All right. In the survey, we also see uh we also ask about their employment level plans uh into the future. Of course, those are not verified in the admin data, but we also see that the treatment  
[03:15:45] um sort of shifts uh plans to work more in the future um by about 7% over the control group mean. Yeah.  
[03:15:54] >> Um in practice, how did these teachers actually increase their hours? Do they have to like take on another day or like how lumpy are hours in the setting?  
[03:16:02] >> Yeah, they're relatively non-lumpy. So, they're relatively flexible to increase by sort of just two hours or four hours.  
[03:16:11] Um, in our prior paper, we did see I think here we have fewer observations.  
[03:16:16] In our prior paper where teachers increase their labor supply, we saw kind of a bump in the distribution of teachers sort of shifting by a half day.  
[03:16:25] Um, here we have fewer observations. So this is harder to see. Um >> just one one thing. Uh there are children involved.  
[03:16:33] >> Yeah. >> And so does >> they might be lost. [laughter] >> So does this mean that children in Switzerland have like two teachers a week?  
[03:16:41] >> Yes. Yeah. >> Oh, okay.  
[03:16:43] >> Yeah. >> I just couldn't figure out how we were meshing the children who after all are full-time.  
[03:16:49] >> Yes, I know. [laughter] >> With teachers who are part-time. Okay.  
[03:16:56] >> Yeah. I I I think one of our co-authors on the prior paper, her kid went to a school with 400 kids and 100 teachers.  
[03:17:04] So that's how Yeah. [laughter] Denise, >> um did you have a chance to ask in the future whether they uh have a decision of like uh intend to switch from part-time to full-time or is it a hard  
[03:17:19] decision to make to begin with? >> I think we could get that from the data.  
[03:17:23] Um, I think as you can see here, very few people actually want to work full-time. Teaching is also an occupation where the men tend to not work full-time in Switzerland because it is so flexible. Yeah.  
[03:17:36] >> Do you see uh any changes in the child care arrangements that they make after afterwards? I mean, I don't know if it's the case in Switzerland. In Germany, for example, school ends really early.  
[03:17:46] >> Yeah. So when you talk to women, they actually say like, "Look, this is the reason why I'm working part-time because I just cannot afford to do full-time.  
[03:17:54] It's just very it's not >> I think sort of the treatment effects that we measure here on labor supply are not super huge. So my guess, but that's just an educated guess because we don't actually ask that in the survey, that  
[03:18:08] there's some capacity to increase that they haven't fully um gotten to and they're doing that thanks to the thanks to the treatment. Yeah. Okay. Yeah.  
[03:18:23] [sighs] >> Yeah. Unfortunately, we didn't measure that. That would have been nice. Yeah.  
[03:18:27] if the if the husbands um take take over more um in prior work we have found that the men really don't move even when the women increase their labor supply. Yeah.  
[03:18:37] All right. Good. Okay.  
[03:18:42] Um let me just mention um this heterogeneity dimension that we do. So we try to um capture the ability of women to be better able to consumption smooth in the case that divorce would occur. And we built this thing that is  
[03:18:55] called uh separation insurance index where we put in labor market attachment variables and savings variables that are both kind of meant to capture how well a woman would be insured if divorce were to occur tomorrow. When we do  
[03:19:08] heterogeneity by uh the separation insurance index which I'm plotting on the x-axis and look at treatment effects on the labor market outcomes. You can see in the left panel um that all of the movement, all of the treatment effect comes from women with relatively low  
[03:19:23] levels of insurance at baseline. Um and that's similarly the case for the long-term employment plans. Um we think or and we see that this group sort of increases their work hours by 6.5% if  
[03:19:37] we're doing a a median split of the sample. Um in the paper we have an extension of our model and we think this is sort of roughly consistent with the low insurance group having um more demand to fill additional insurance and therefore moving both on the financial  
[03:19:52] and the labor market uh channel. All right. Um let me talk a little bit about mechanisms before um concluding.  
[03:20:02] Yeah, sorry Jess. Do you look at all heterogeneity by the experience to friends who have divorced? Yes. I'm still trying to wrap my mind around why  
[03:20:14] a relatively short video has a bigger effect on people's beliefs than having observed the experiences of their own social network which presumably modular the comments earlier  
[03:20:29] national. >> Yeah. So we we actually thought that that would matter a lot. It's possible that we're just underpowered. This is a split that we could do and we asked baseline variables to elicit that.  
[03:20:41] Concretely, we asked them if in the last 10 years they accompanied a person close to them through divorce and some other things. And we don't really find meaningful differences in treatment effect by prior experience.  
[03:20:55] Prior experience at least.  
[03:21:00] That's a good question. From the top of my head, I think it's around 30% that say yes to this question of uh in the last 10 years. Uh but we we should check again. Yeah.  
[03:21:13] Thank you. All right. Um let me show you what we can measure in terms of movements of beliefs and attitudes around separation.  
[03:21:22] Um in the end we have sort of a bundle treatment that might move many different things. Um, we build a perceived separation likelihood index across this question battery in retirement where we ask respondents about the probability of  
[03:21:36] living with their current partner and having income from their current partner's pension and savings um contributing to their own household income both absolute and relative to their own. When we combine these uh questions into an index, we see that  
[03:21:50] treated women update their beliefs about separation likelihood by.13 standard deviations.  
[03:21:58] Another thing that women uh could learn from the treatment is that the financial consequences of divorce are more impactful than what they thought they were. So we try to get at that by asking  
[03:22:11] them some attitude questions about whether they think the lower earner, sorry, the partner specializing in child care shoulders a financial risk, whether they approve uh of compensation within a couple and whether they approve of a  
[03:22:26] savings vehicle that is uh separate property for the wife uh and should ensure her in the case uh of of separation. And we also see a movement on this index by point4 standard deviations from the second um um protag set of  
[03:22:44] protagonists. Um these women could also have learned that they should start to have a conversation in their household and we actually see a lot of movement on these variables. So we see that the treatment group is 20 percentage points  
[03:22:58] more likely to want to talk to their partner after the treatment and they're less likely to want to discuss this topic with their friends or their their colleagues. Um when we ask them and but this is conditional so not necessarily  
[03:23:12] causal conditional on talk wanting to talk with their partner the topic that they plan to talk about. we see that they want to talk about how the uh their video that they saw relates to their own situation or the situation of their  
[03:23:25] household rather than the situation of people close to them or society overall.  
[03:23:30] In the follow-up, we do see that these women have followed up on those plans and about twothirds of the treatment group have had a conversation with their partner.  
[03:23:40] In terms of how easy it is to have this conversation, um we see some evidence that this is not without difficulty and perhaps not the most comfortable topic to raise at home. So we see that the treatment group is more likely to rate  
[03:23:55] uh how satisfied they are with how this conversation went as neutral sorry more likely to rate it as neutral rather than satisfied compared to the control group.  
[03:24:04] >> Clarification the control group is having a conversation about like mental health.  
[03:24:07] >> Correct. Yes. Yes. And this is and there's different rates that these conversations happen.  
[03:24:12] >> Yes. So these are not causal. Sorry, I should be more clear here. These could just reflect compositional differences of on the margin who's more likely to have a conversation, who we moved in the treatment group. We think that they're kind of suggestively showing that this  
[03:24:25] might not be such a comfortable topic. Yeah. Yeah. But take those with a grain of salt. And I'm I don't want to make any causal claims here. Yes. And we also see that they're a little less likely to uh rate this conversation as positive  
[03:24:38] and and rather uh use negative or or neutral um attributes.  
[03:24:44] What happens to relationship stability? Um let me just show you that when we ask respondents on a liquor scale how happy they are in their relationship, 80% in the control group say they're either happy or very happy. That doesn't change  
[03:24:57] for the treatment group. We have a relationship satisfaction index where we let respondents rate their relationship on bipolar adjectives. Uh we don't see any movement on that index. And when we ask respondents directly if they  
[03:25:10] actively thought about uh their own separation, their own divorce or talked about this topic with anybody, if anything, we actually see that uh the treatment group is less likely to have thought actively thought or brought up  
[03:25:24] that uh that topic. All right.  
[03:25:29] With that, let me just uh conclude. Um so what we try to show here is that making the possibility and consequences of separation relatable increases insurance against the risk of divorce via financial insurance um an increase  
[03:25:43] in financial insurance and a decrease in household specialization. So our treatment group works 2.5% more in terms of work hours. And I think sort of the key contribution that we're trying to make with this paper is we try to show  
[03:25:56] that overoptimistic beliefs about the risk of separation lead to under insurance through excess household specialization. And we also think this kind of highlights an important role for policies that would facilitate more effective insurance instruments uh  
[03:26:11] within couples. So for example, Sweden has introduced uh a few years ago a pension account that spouses can transfer from one to the other and that sort of remains their own property and doesn't get uh allocated to the joint  
[03:26:25] marital property. So having easily access accessible savings vehicles such like that could help couples uh specialize as much as they want but sort of ensure that that specialization  
[03:26:39] risk. Thank you VERY MUCH. >> [applause] >> ALL RIGHT. THANK YOU SO MUCH, URSA. All right, we will have a 15 minute break.