IFPRI Discussion Paper 02271 September 2024 To Defer or Differ Experimental Evidence on the Role of Cash Transfers on Nigerian Couples’ Decision-Making M. Mehrab Bakhtiar Marcel Fafchamps Markus Goldstein Kenneth L. Leonard Sreelakshmi Papineni Poverty, Gender, and Inclusion Unit INTERNATIONAL FOOD POLICY RESEARCH INSTITUTE The International Food Policy Research Institute (IFPRI), a CGIAR Research Center established in 1975, provides research-based policy solutions to sustainably reduce poverty and end hunger and malnutrition. IFPRI’s strategic research aims to foster a climate-resilient and sustainable food supply; promote healthy diets and nutrition for all; build inclusive and efficient markets, trade systems, and food industries; transform agricultural and rural economies; and strengthen institutions and governance. Gender is integrated in all the Institute’s work. Partnerships, communications, capacity strengthening, and data and knowledge management are essential components to translate IFPRI’s research from action to impact. The Institute’s regional and country programs play a critical role in responding to demand for food policy research and in delivering holistic support for country-led development. IFPRI collaborates with partners around the world. AUTHORS M. Mehrab Bakhtiar (m.bakhtiar@cgiar.org) is a Research Fellow in the Poverty, Gender, and Inclusion Unit of the International Food Policy Research Institute, Washington, DC. Michael Fafchamps (fafchamp@standford.edu) is the Satre Family Senior Fellow at the Center on Democracy, Development, and the Rule of Law at Stanford University and Courtesy Professor at the Department of Economics, Stanford University, Standford, CA. Markus Goldstein (markusgold@gmail.com) is a Lead Economist at the Africa Gender Innovation Lab, The World Bank Group (on leave). Kenneth L. Leonard (kenneth@umd.edu) is a Professor at the Department of Agricultural and Resource Economics, University of Maryland – College Park, College Pardk, Maryland. Sreelakshmi Papineni (spapineni@worldbank.org) is an Economist at the Africa Gender Innovation Lab, The World Bank Group. Notices 1 M. Mehrab Bakhtiar is the corresponding author on this paper. 2 IFPRI Discussion Papers contain preliminary material and research results and are circulated in order to stimulate discussion and critical comment. They have not been subject to a formal external review via IFPRI’s Publications Review Committee. Any opinions stated herein are those of the author(s) and are not necessarily representative of or endorsed by IFPRI. 3 The boundaries and names shown and the designations used on the map(s) herein do not imply official endorsement or acceptance by the International Food Policy Research Institute (IFPRI) or its partners and contributors. 4 Copyright remains with the authors. The authors are free to proceed, without further IFPRI permission, to publish this paper, or any revised version of it, in outlets such as journals, books, and other publications. iii Abstract We conduct an original lab-in-the-field experiment on the decision–making process of married couples over the allocation of rival and non-rival household goods. The experiment measures individual preferences over allocations and traces the process of deferral, consultation, communication and accommodation by which couples implement these preferences. We find few differences in individual preferences over allocations of goods. However, wives and husbands have strong preferences over process: women prefer to defer decisions to their husbands even when deferral is costly and is not observed by the husband; men rarely defer under any condition. Our study follows a randomized controlled trial that ended a year earlier and gave large cash transfers over eighteen months to half of the women in the study. We estimate the effect of treatment on the demand for agency among women and find that the receipt of cash transfers does not change women’s bargaining process except in a secret condition when the decision to defer is shrouded from her husband. This suggests that the cash transfer to women increases their demand for agency but does not change the intra-household balance of power enough to allow them to express it publicly. Keywords: intra-household decision-making, bargaining, cash transfers iv Acknowledgements We thank Pam Jakiela, Rosella Calvi, Eliana Carranza, Mushfiq Mobarak, Gautam Bastian, Neslihan Uler, Joao Montalvao, Marco Castillo, Michael O’Sullivan, Tihitina Andarge, Tonima Tasnim Ananna, Sk Md Bakhtiar Hossain, conference participants at Midwestern International Economic Development Confer-ence (University of Chicago) 2024; NEUDC/Yale University 2022; CSAE/Uni. of Oxford 2022; Population Association of America 2022 Annual Meeting; Advances in Field Experiments (University of Chicago), 2020; and seminar participants at Dartmouth College, National University of Singapore and George Mason University for their helpful comments. Marietou Sanogo, Lawal Ishaq Simiyat, Oti Iheanyi, Ayuba Umar, Aaron David and Christy Shanding gave extraordinary assistance in carrying out the fieldwork. We thank the Africa Gender Innovation Lab and Bruce and Mary Ann Gardner Dissertation Enhancement Fellowship for funding the fieldwork of the lab-in-the-field experiments. The pre- analysis plan for this paper can be found at https://www.socialscienceregistry.org/trials/3360. The lab-in-the-field experiment was approved by the IRB at the University of Maryland: 1352047-1 1 Introduction Policymakers focus on progress in women’s empowerment and equality because women con- tinue to experience unequal outcomes, even within households (see Duflo, 2012, for a review). Intrahousehold inequality in consumption and health outcomes has been documented world- wide (e.g., Altonji and Blank, 1999; Azmat et al., 2006; Beaman et al., 2017; Bertrand and Hallock, 2001). In addition, large gains in measurable outcomes have been documented for women, children, and even the whole household when women in poor households are pro- vided additional resources such as cash transfers (e.g., Bandiera et al., 2017; Carneiro et al., 2021). Money given to women is more likely to be used for investments in education, children’s nutrition, and housing than money in the hands of their husbands (e.g., Adato et al., 2000; Duflo, 2003; Fiszbein et al., 2009; Hoddinott and Haddad, 1995; Thomas, 1994) and increasing women’s say in family finances raises savings and investment (Armendáriz and Morduch, 2010; Ashraf, 2009; Duflo, 2012). Much of the literature assumes that the pattern of unequal consumption is a result of lower empowerment, in line with the dominant economic model of intrahousehold bargaining power in which the utility function of individual members is solely defined over material outcomes (e.g., Browning and Chiappori, 1998; Chiappori, 1988, 1992, 1997; Chiappori et al., 2002; McElroy, 1990; McElroy and Horney, 1981). Indeed, when this assumption is combined with a symmetry assumption about the other-regarding preferences of the spouses,1 equality in bargaining power implies equality in outcomes. In this model, the welfare weights or bargaining power of spouses is subsumed in a sharing rule which can be inferred from the intrahousehold division of rival consumption (Brown et al., 2021; Cherchye et al., 2017; Dunbar et al., 2013). Implicit in this modeling framework is the assumption that individuals do not care about the way these outcomes are achieved. Perhaps as a result, economists have paid less attention to the process of decision–making within the household. Even when more attention is paid to process, such as whether women have executive agency and whether external interventions can improve female agency within the household (e.g., Dhar et al., 2022; Karimli et al., 2021; Pitt et al., 2006; Riley, 2022), the emphasis remains on the instrumental value of agency, that is, on its capacity to affect material outcomes, rather than on agency’s intrinsic value. With a few exceptions (e.g., Afzal et al., 2022; Fernandez et al., 2015),2 the simple question of whether empowerment per se is valued by women (similar to the ‘capabilities’ view of Nussbaum, 2001; Sen, 1999) has been mostly 1E.g, spouses only derive utility from their own (rival) consumption, or have symmetrical altruistic preferences with identical welfare weights. 2Fernandez et al. (2015) examine the correlation between subjective well-being and the right to make particular decisions. Although provocative, the results are not identified. 1 ignored in the literature on intrahousehold resource allocation. To examine the link between consumption and empowerment, we present results from an original lab-in-the-field experiment conducted with married couples in Northern Nigeria, a setting within a lower middle-income country where women’s agency and rights are severely constrained. We combine traditional elements of economic laboratory design (experimen- tally assigned controls and separation of individuals) with field elements (a random sample of married women received significant unconditional cash transfers over eighteen months be- fore the study took place) and unique lab–in–the–field elements designed by the team (three laboratory shopping stalls with separate categories of female, male and household items). Finally, the experiment includes a randomly assigned secret treatment in which most deci- sions are shrouded so neither spouse can tell what their spouse chose or what processes they followed to reach that decision. We find that the preferences of both male and female participants are egalitarian on average. Despite what seems to be a lack of overt disagreement over budget allocations (as in Almås et al., 2020), spouses have strong preferences over the way decisions are made. Women are far more likely than men to consult their spouse, defer to them, and accommodate their wishes. In contrast, male participants tend to make a decision on their own without consulting or accommodating their wife’s wishes. Furthermore, we find that women over- defer and men under-defer compared to optimal decisions. The experimental design allows us to conclude that spouses’ preferences over the way decisions are made are consequential: they impose an allocative efficiency cost on already very poor households. The experiment also enables us to estimate the effect of an unconditional cash trans- fer (UCT) program on both material outcomes and the process by which these outcomes are reached. We can compare the joint decision–making processes of control and treated households, including consultation, communication (honest or not), deferral, revision and accommodation of spouse’s preferences and a measure of the willingness to pay for agency over own consumption. From these measures, we can examine several important features of the decision–making process. First, we test, in a controlled environment, whether the randomized cash transfer treatment increases the desire for private consumption and leads to higher material welfare. Second, we test whether treatment has an impact on the willing- ness to exert agency or to relinquish control to a spouse, either secretly or openly. We do so in a way that enables us to test whether subjects’ willingness to pay for agency is driven primarily by instrumental value, or whether subjects deviate from taking the action that would maximize their own material utility. Third, we test whether the treatment increased or decreased efficiency in decision–making. The UCT intervention had a small effect on the social preferences of men and women; 2 women allocate more goods or money to themselves if they received the UCT. Treated women are less likely to defer their choice to their husbands, but only when their deferral decision is shrouded and therefore kept secret from their husband. UCT treated women are also slightly less willing to consult. Taken together, these suggest a slight improvement in procedural empowerment but only when decisions are shrouded. Lab-in-the-field experiments have been used to measure demand and willingness to pay for agency within a household (e.g., Abbink et al., 2020; Afzal et al., 2022; Almås et al., 2018; Iversen et al., 2011; Jakiela and Ozier, 2016; Mani, 2020; Schaner, 2016), but there are only a few studies that look into the effect of a cash transfer on experimental measures of female empowerment. Alm̊as et al. (2018), report results from a lab-in-the-field experiment in North Macedonia where a CCT (for children staying in school) was randomly allocated to either the head of household (usually a man) or a woman. They find that, compared to wives of male recipients, female recipients of this long-term CCT exhibit a lower willingness to pay (WTP) to appropriate an additional windfall for themselves instead of their spouses. This, they argue, shows an improved bargaining power of the CCT-receiving women: they are willing to pay less than others because they have more say over the allocation of the windfall even if it is received by their husband. However, other channels could explain this finding3 and our study, by including a secret condition treatment, helps unwrap some of the issues raised in that paper. In the following section, we develop a conceptual framework for analyzing the behavior of our participants. Section 3 explores the setting and details of our lab experiment and explain how we use it for examining the behavior of households. Our empirical work is divided into two sections. First, in Section 4 we examine the behavior of the households in the control group, a baseline of behavior. Second, in Section 5 we examine how this behavior was changed by the UCT experiment. Finally, in Section 6, we conclude. 2 Conceptual framework In the canonical model of intrahousehold consumption allocation (e.g., Becker, 1964, 1981; Chiappori, 1988; McElroy and Horney, 1981), the decision that a couple makes is represented 3For example, fairness norms could explain why women who were beneficiaries of a CCT program are less likely to object to a new transfer to be given to their spouse instead of receiving it themselves again. Moreover, former recipients of a CCT might want to avoid ‘taxation’ from family and relatives (see Jakiela and Ozier, 2016) after experiencing it first-hand and may rather have their husbands deal with such attempts. This issue is salient in Alm̊as et al. (2018) because the information on being a recipient of the CCT program was common knowledge. 3 as the solution to: Maxx ωuh(ch) + (1− ω)uw(cw) subject to ch + cw = y (1) where w and h represent the wife and husband, respectively, y is the combined income of the couple and ch and cw are the consumption expenditures of the husband and wife, respectively. The relative welfare weights of the husband and wife, ω and 1−ω, represent bargaining power over consumption: the larger ω is, the larger ch is relative to cw. It follows that the relative bargaining power of the two spouses can be inferred from their consumption: an empowered wife consumes more (Cherchye et al., 2017).4 A simple theory of intrahousehold agency Empowerment, however, is also about having a say in decisions that affect you. The call for female empowerment is partly because of its anticipated material benefits, as it allows women to better defend and promote their interests. But it is also about the process by which decisions are made: just as a country is not considered democratic simply because it fulfills the material needs of its population, the empowerment of women cannot be solely judged by their material welfare. Which spouse believes they have the right to decide how to allocate the household’s consumption budget? Who consults and who is consulted on that decision? Does the consulted spouse distort his or her reported preferences? Are reported preferences taken into account by the other spouse? Our experiment is designed to capture each of these important dimensions of the decision–making process within the couple while controlling for the consumption preferences of each spouse. Afzal et al. (2022) offer a straightforward formalization of preferences over process (see also Doepke and Tertilt, 2019). Let there be two spouses h and w and let ch and cw be their consumption bundles. According to their definition, spouse w has a preference for executive agency on her consumption if: uw(cw, D h w) < uw(cw, D w w) (2) where consumption cw is kept constant, Dh w means that spouse h decides the consumption bundle cw for w, and Dw w means w decides for herself. The above inequality implies that when w (or h) prefers to decide, she is willing to accept a lower level of material welfare in exchange for having executive agency on her consumption. 4This logic is behind a series of recent papers on Engel curves within households (e.g., Bargain and Donni, 2012; Bargain et al., 2014; Brown et al., 2021; Calvi, 2020; Dunbar et al., 2013; Lechene et al., 2022; Sokullu and Valente, 2022; Tommasi, 2019). 4 In this paper we do two things: 1) we experimentally assess the decision process by which couples allocate consumption between themselves and estimate spouses’ demand for control and agency within that process; and 2) we estimate the causal effect that a full-scale external intervention — an unconditional cash transfer (UCT) — has on the decision process and on the demand for agency of each spouse. We know, from an evaluation of the UCT, that this UCT increased women’s self-employment and income and that it raised traditional indicators of female empowerment focusing on material welfare and on decision–making in farm production (Papineni et al., 2024). We want to know whether it also affected agency and, if yes, along which dimensions. Our experiment starts by eliciting true allocative preferences in consumption, which we identify by letting each spouse independently decide how to allocate various consumption budgets y between cw and ch, as well as between individual and joint consumption. This is achieved by setting up stalls that carry gender-specific goods. One advantage of this approach is that it does not assume that each spouse is solely motivated by self-interest: it allows subjects to manifest (possibly different) altruistic preferences in household con- sumption.5 This differs significantly from the canonical model of intrahousehold allocation in which differences in material allocation between spouses are seen as a manifestation of different bargaining weights. When spouses are differentially altruistic towards each other, the canonical model cannot identify how power is distributed within the household. Fur- thermore, with altruistic preferences, it is possible for both spouses to divide a budget y in the same manner cw + ch = y while at the same time both wanting to make that decision themselves: uw(cw, ch, D h w) < uw(cw, ch, D w w) and uh(cw, ch, D w h ) < uh(cw, ch, D h h) (3) We experimentally measure this demand for executive agency by allowing subjects to defer to their spouse the division of a budget y between consumption bundles cw and ch. Deferral refers to a subject’s willingness to delegate consumption decisions. It indicates a reluctance to hold sole executive agency over household consumption decisions. In our model, this preference can be represented by reversing the sign of equation (2): uw(cw, ch, D h w) > uw(cw, ch, D w w) (4) Demand for executive agency can be measured experimentally: we let subjects independently and secretly decide how to divide a budget y into cw and ch on their own and then we ask 5Altruistic behavior can arise from a variety of sources, such as social norms, moral and philosophical principles, affection, or reaction to emotional stimulus. Our experiment is not designed to disentangle them. 5 them whether they would prefer to delegate that division to their spouse. We can also look at whether both husband and wife delegate the decision to their spouse, suggesting a joint preference towards a negotiated outcome. Delegating the final decision to someone else does not, however, imply an absence of demand for agency: the person may still wish to influence that decision, e.g., because of asymmetric information about preferences between spouse (e.g., Tagat et al., 2023) or be- cause agency is valuable in its own right (e.g., Afzal et al., 2022). Measuring demand for consultative agency is more complex because consultation involves at minimum three distinct steps: truthful communication of a preferred split (cw, ch) w by w to h; consultation by h of this communication made by w; and accommodation of that communication by h who revises his original division budget allocation (cw, ch) h at least partially in the direction of the split communicated by w. To capture these ideas, we invite subjects to communicate a preferred allocation to their spouse. Ideally, this allocation should correspond to their true preferred allocation, in case the spouse decides to implement it. It also may involve some strategic misrepresentation whereby the subject distorts their true preferences when communicating with their spouse, e.g., to avoid disapproval or retaliation. Misrepresentation may also indicate an unwillingness to reveal one’s true preferences due to self-image considerations (e.g., not wanting to appear too selfish). The absence of misrepresentation therefore signals a relatively serene approach to joint decision–making among spouses and is a pre-condition for consultative agency to achieve allocative efficiency in consumption. Communicating one’s preferences to a spouse may fall on a deaf ear, though: the spouse may refuse to hear them. Willingness to listen to someone else’s preferences is what we call consultation. Two-way communication between the spouses is thus a necessary condition for a subject to influence their spouse’s decision. But it is not sufficient: the communicated preferences must also be taken into account by the deciding spouse, i.e., they must be ac- commodated. Taken together, truthful communication, consultation, and accommodation therefore create the conditions for a negotiated allocation of the household consumption budget. Even in the presence of deferral, the combination of these three processes confers to the deferring spouse the right to be heard and influence decisions — a process we call consultative agency. It is measured as the combined effect of communication (including mis- representation) by the subject and of their spouse’s willingness to consult and accommodate. Formally, let cww be the consumption for self, chosen by subject w in isolation; let chw be the consumption for w chosen by subject h in isolation; let kh = 1 be h’s decision to consult w’s communicated preferences and 0 otherwise; and let ch accom w be the consumption for w chosen by h after being told w’s communicated preference. Consultative agency is then 6 defined as the extent to which chw moves towards cww as a result of the consultation process. It is given by the following formula: Ch w = 1− cww − (khc h accom w + (1− kh)c h w) cww − chw (5) where we have normalized by cww − chw, the difference between the true preferences of w and h. Note, first, that Ch w is undefined if cww = chw; if the husband and wife both prefer the same amount to for the wife, we can’t observe accommodation. Whenever the initial preferences are different, however, Ch w = 1 means that w receives her true preference while 0 means that she receives her husband’s preference. Any value between 0 and 1 indicates a compromise between the spouses. It is also possible for Ch w < 0, in case h revises his initial choice chw away from cww, for instance, to manifest disapproval, or for Ch w > 1 in case h overcompensates by moving beyond what w asked for. To facilitate the analysis, we constrain Ch w to only take values between 0 and 1. This expression can be decomposed into those parts which are due to misrepresentation, willingness to consult, and extent of accommodation: Ch w = 1− (cww − cw comm w cww − chw + (1− kh)(c w comm w − chw) cww − chw + kh(c w comm w − ch accom w ) cww − chw ) (6) where cw comm w is the choice communicated by w to her husband. The first fractional term is the loss of agency due to misrepresentation by w, the second fractional term is the loss of agency due to lack of consultation by h, and the last fractional term is the loss of agency due to the lack of accommodation. An equivalent formula can be derived for the consultative agency of the husband – should the wife be granted sole executive agency. We also borrow aspects of the design developed by Afzal et al. (2022) to study the decision to defer when it has no instrumental value. This allows us to parse out the role played by instrumental motives from preferences over process — a distinction that mirrors that between consequentialist/utilitarian and deontological ethics (e.g., Benabou et al., 2024). In this part of the experiment, w chooses one of four physical consumption bundles for herself to consume on the spot. She is then invited to pick a bundle for h without knowing which bundle h prefers and is finally asked whether she wants to defer the choice to h. Mirror decisions are independently made by h. Here, not deferring the choice to the spouse has no instrumental value — w cannot appropriate any of h’s bundle and vice versa. It is therefore always inefficient since choosing for the spouse has a strong chance of deviating from his or her preferred bundle. Hence choosing not to defer is a measure of demand for pure control 7 (e.g. Afzal et al., 2022; Ashraf, 2009; Jakiela and Ozier, 2016)6. Finally, we examine how spouses handle the trade-off between efficiency and self-interest in household production decisions, using a simplified version of a game developed by Fafchamps and Kebede (2022). The literature has shown that assigning (legal or customary) control over financial or productive assets within the couple can affect the allocation of consumption expenditures between spouses (e.g., Browning et al., 1994; Lundberg et al., 1997). In some cases, this can even lead to inefficient choices (e.g., Lundberg and Pollak, 1993; Udry, 1996). The purpose of this part of the experimental is to test whether subjects are capable of taking efficient production decisions even if it means reallocating productive endowments to their spouse. Deviation from the efficient choice again captures demand for pure control, in this case over the production process. A simple theory of emancipation Our experiment, by combining these different ele- ments, produces evidence on how executive agency is allocated between spouses and the extent to which spouses are granted consultative agency in the absence of external interven- tion. We then use the experiment to test how this equilibrium is affected by two separate interventions: secrecy in executive agency, which is granted to some subjects in the experi- ment; and a large UCT intervention targeted to the wife. The objective of these interventions is to disentangle three broad mechanisms of subjugation which we refer to as oppression, indoctrination, and alienation. Oppression refers to the fear of retaliation: a spouse may wish to exert control over the allocation of household consumption, but fears disapproval or reprisal if this action is observable by the spouse (e.g., Buchmann, 2022). Secrecy may also enable individuals to insulate their income from social pressures from the spouse (e.g., Zhang, 2023a,b) or from outside the household (e.g., Boltz et al., 2019). By providing the opportunity for a subject to exert executive agency in secret (e.g., Ashraf, 2009), we elicit what we call ‘pent-up demand’ for agency, that is, demand for executive agency that is repressed by the spouse and can only be expressed in secret (e.g., Ashraf et al., 2014). Reprisal may also take a financial form, e.g., by diverting household resources away from the offending spouse. In this case, providing more financial autonomy to an oppressed spouse should offer protection against such reprisal. Hence an intervention that helps that spouse gain financial independence — such as the UCT (see Papineni et al., 2024) — should reduce the fear of reprisal, thereby allowing the spouse to openly exert more agency, either executive or consultative, without the need for secrecy. 6For a deeper understanding of the difference between instrumental and intrinsic control, see the following papers on intimate partner violence (IPV) (e.g., Angelucci, 2008; Bobonis et al., 2013; Haushofer and Shapiro, 2016; Hidrobo et al., 2016; Perova and Vakis, 2013; Roy et al., 2018). 8 Indoctrination refers to the internalization of gender norms. Other researchers have sought to document the effect of a change in such norms — or in the perception of such norms — on behavior and agency (Bursztyn et al., 2020). In this paper, we do not attempt to change norms or perceptions directly. Nonetheless we are hopeful that gender norms may improve in the aftermath of a sustained intervention that, by providing regular monetary transfers to married women, helped them start a business and gain financial autonomy (Papineni et al., 2024). If this is the case, we expect the UCT intervention to modify the household budget allocation in favor of women, to induce husbands to grant more consultative agency to their wife — i.e., by consulting them and accommodating their stated preferences — and to induce wives to be less willing to defer consumption decisions to their husband and more willing to truthfully disclose their true preferences to their husband. The third main channel for subjugation is alienation, which often manifests itself through a lack of aspiration for agency: women do not imagine that they could make decisions on their own (e.g., Bernard et al., 2023; Orkin et al., 2023). In individualistic societies, executive agency over own consumption is taken for granted. But in collectively-minded societies (e.g., Enke, 2019; Roland, 2020), making important decisions without consulting with the group is often regarded as unwise or illegitimate. Consulting others and deferring to those in charge of the group is regarded as beneficial to the group and its members — and in patriarchal societies the application of this principle to women and other dependents means that they do not imagine being capable of making correct decisions by themselves. This makes it all the more important to capture consultative agency in contexts, such as our study, where collective decision–making is generally seen as the norm (Thomas et al., 2023). If women refrain from expressing a demand for agency because of collective thinking, the decisions they take should be the same whether they are taken in secret or not: there should be no pent-up demand for agency that is expressed in secret. An external intervention that specifically targets funds to one member of the household does, however, create a new reality in which the individual agency of married women is recognized by an external actor. This in turn may favor a rise in individualism that would express itself in more selfish consumption choices and a demand for executive agency. If such demand may trigger reprisal from the spouse or the social group, it may only be expressed in secret. 3 Experimental Design and Background We conducted a lab-in-the-field experiment in 38 sessions with married couples in 27 villages from two rural regions in Kebbi State, in northwest Nigeria. The experiment was embedded in a larger randomized controlled trial of an unconditional cash transfer (UCT) program 9 targeted at ultra-poor women in this region. The majority of households in this region live in extreme poverty. These households are predominantly of Hausa ethnicity and Islamic faith, known for their patriarchal social structures and women face restrictive social norms, low levels of education, and low labor force participation, resulting in limited agency (Braimah, 2014; Ogu et al., 2016). The UCT baseline survey confirms women’s limited agency, especially regarding labor outside the home which is decided on by husbands, and it documents the existence of social norms restricting women’s work and mobility, with 40% of men considering it inappropriate for women to accept paid jobs outside the home. Female labor force participation was low before the intervention, with most women engaged in household work or childcare. Only 36% of women reported any income-generating activities in the twelve months prior the baseline survey, mostly on household-operated farms (Papineni et al., 2024). The UCT, randomized at the household level within study villages, provided a total of 75,000 Nigerian Naira (roughly USD 693 PPP for 2015) over fifteen months to the primary female decision-maker of the household. This is a considerable amount of money for the participating households, amounting to approximately half of their annual consumption. Indeed 92% of sample households lived on less than USD 1.90 a day according to the baseline survey. The cash transfers did not come with any explicit conditions of how the money should be spent or shared. However, during a sensitization campaign, households were told by traditional community leaders that the money was for the female recipient. As noted by Papineni et al. (2024), this light intervention could have influenced the female recipient’s ability to keep more of the cash transfer (e.g., Benhassine et al., 2015). Bastian et al. (2017) report that 54% of the cash transfer (in the same program) was kept by female recipient and 26% was passed on to her husband; per capita consumption increased by 25%; and the value of household assets (mainly small animals) increased by 30%. Papineni et al. (2024), similarly, show an increase in a women’s empowerment index as measured by a modified versions of the Women’s Empowerment in Agriculture Index (WEAI). The experimental protocol is detailed in the rest of this section. Further details are available in Sections C of the Online Appendix. 3.1 Budget allocation decisions At the heart of our design is a set of budget allocation decisions that subjects are asked to make under several treatment conditions. In the first part of the experiment, each spouse independently and secretly chooses how to split a budget of 2500 Naira (around 20 USD) 10 between two categories of goods.7 This choice is never revealed to their spouse but it has a positive probability of being implemented, making revealed preferences incentive compatible. These choices can be seen as defining the social utility of each spouse if he/she were given full control over household consumption — i.e., they measure Wi(x) for a particular x vector and fixed budget. y = 2500. For three of the four domains (A, B and C) subjects are given a budget and asked how they wish to allocate it to pre-stipulated categories of goods they can only purchase from us at the end of the experiment, thereby reducing the chance of ex post transfers outside the experiment. Once that budget has been assigned to a particular category, it cannot be converted into cash or spent in another category. To facilitate comparison with other experiments in which spouses are given cash, we also include a separate allocation decision (Domain D) in which subjects divide a cash amount between themselves and their spouse. To implement this design, three market stalls are constructed in the lab, each of which contains only one category of items. These items were pre-tested to be recognizable and desirable, and they include some goods that are easy to buy locally and some goods that only available in town. All participants are shown pictures of the items available in each stall. One stall includes goods intended to appeal to women: jewelry, colorful fabric, and dresses. The second stall includes male-oriented goods such as hats, caps, belts, shoes, and fabric. The third stall offers common household items such as cleaning supplies, mats, cups, plates, and mosquito coils.8 The choice of these items was carefully designed such that male goods would not appeal to women and vice versa. Given how strongly gendered the local context is, this was not particularly difficult to do. At the end of the experiment, husband and wife are reunited and, if that decision was randomly selected for implementation, couples are given vouchers redeemable in specific stalls. Thus, when a subject allocates funds to a stall expected to appeal to their spouse, the subject may intend to purchase something for their spouse or to let their spouse choose – in the context of this experiment, both choices are observationally equivalent. We did not record the items that subjects purchased or who made the choice. By making two of the stalls gender-specific, the design limits the scope for ex post real- location of the objects between spouses. There still remains the possibility that subjects sell or give the objects to others after the experiment. To avoid this, we include a fifth domain in which subjects separately consume a food and drink of their choice directly in the lab, as in Afzal et al. (2022). In that case, reallocation across spouses is impossible. 7This amount is roughly equivalent to six days of household consumption or around two weeks of the UCT payments. It was chosen to be sufficiently large to incentivize participants to take the experimental tasks seriously. 8See Online Appendix photographs C3, C4, and C5, respectively. 11 Throughout the experiment, each spouse answers various allocation decisions regarding their preferred budgetary allocations over four possible choice pairs: A female vs male items purchased in the lab stalls B household items vs male items purchased in the lab stalls C household items vs female items purchased in the lab stalls D cash for wife vs cash for husband9 J two kinds of cookies and two flavors of juice for both self and spouse. Choice A measures the social utility Wi(xw, xh) of consumption bundles xw and xh for individuals i = {w, h}, conditional on the choice of goods available in our stalls. Choice D is over money and can thus potentially be undone outside the lab, but it offers the advantage that consumption is not restricted to the goods on sale in the lab. It measures the utility Wi(xw, xh) of consumption budgets xw and xh for individuals i = {w, h} that can be achieved from the choice of goods available outside the lab. Choices B and C are between private goods and household goods, thereby revealing how husbands and wives differ in the extent to which they care for household public goods, including child consumption (xc). They measure Wi(xw, xc) and Wi(xh, xc), respectively. In each case A-D, the subject is given a budget to divide — in multiples of 100 Naira — between each of the two options listed above. In the rest of the paper, we refer to each of these choice sets using the letters A to D above. The food-and-drink game is only played once with real stakes — i.e., actual food and drinks given to the subject at the end of the experiment. At the end of the experiment, subjects are also asked to choose between two input allocations that determine individual incomes — see below. 3.2 Decision blocks The experiment is divided into a series of decisions grouped into blocks that each subject makes silently and in isolation from their spouse — who is in another room. In terms of sequencing, Block 1 always comes first while Blocks 4, 5 and 6 always come last, in that order. The order of Blocks 2 and 3 is permuted at random across sessions. At the end of the experiment, one decision from one of Block 1 through 4 and 6 is selected at random for each couple to determine the cash and vouchers received by the couple. This 9The decision was presented as money for self or money for spouse but we flip the husbands decisions and always present it as money for wife versus money for husband. 12 means that each decision a subject makes is fully incentivized. 1 provides examples of the flow of decisions across blocks 1 to 4 and how final payoffs are determined. We delineate which choices were secret for everyone and which ones were secret only in the secret treatment, and, in the next sub-section, explain how choices were hidden. Block 1 Allocating a budget across different types of items or payments [split and resplit]: Each participant is sequentially asked to split a budget of 2500 Naira (about $23) in 100 Naira increments. This is done in each of the four domains: splitA-D. Each subject is then asked to split a budget of 2100 Naira10 in domains A and D: resplitA and resplitD. The order of both sets of choices is the same for all subjects in a session and randomized across sessions. These choices are never revealed to the spouse. The purpose of this block is to elicit each subject’s true preferences over all possible choice pairs. Block 2 Choosing whether to allow one’s spouse to make the decision instead of making one’s own [defer]: Subjects are reminded of their initial allocation across splitA-D and can choose to retain this allocation or to replace it with their spouse’s allocation, which they have not seen (deferA-D). The object is to elicit subjects’ willingness to defer budget allocation to their spouse. This decision is not revealed to the spouse in the secret treatment. Block 3 Choosing whether to allow one’s spouse to make the decision instead of making one’s own with additional costs [defercost, deferbenefit]: Having already made decisions in Block 1, subjects are first reminded of their initial allocation across splitA and splitD and then given a choice between their decision over a budget of 2500 or their spouse’s decision over a budget of 2100 (defercostA and defercostD). In addition, subjects are given a choice between their decision over a budget of 2100 Naira (resplitD) or their spouse’s decision over 2500 Naira. This decision applies only to domain D, splitting money between husband and wife (deferbenefitD). For defercost it is costly to defer the decision and for deferbenefit it is costly to retain the decision. The purpose of this block is the elicit subjects’ willingness to pay to defer — or not defer — budget allocation to their spouse. These choices are not shown to the spouse in the secret treatment. Block 4 Communicating preferences, consulting over preferences and revising decisions [communicateB-D, consultB-D, reviseB-D] Subjects are then asked to pick an allocation across domains B, C and D that will be communicated to their spouse communicateB- D). The purpose of this part of the experiment is to determine whether subjects choose to misrepresent their true choices from their spouse. These are the only decisions in the secret treatment that are always visible to the spouse without shrouding. 10The difference of 400 Naira was chosen so as to be just large enough to identify variation in deferral rates among women. In the last two sessions (10% of subjects), we increased the difference by reducing the amount to 1800 Naira to see whether we could induce deferral among men. 13 After having done this, subjects are asked, if they would like to see the communication of their spouse (consultB-D). The purpose of this question is to ascertain whether subjects are interested in learning about their spouse’s preferences — something they would want to do in case they are willing to accommodate these preferences, in part or in full. Finally, whether or not the subject chose to see their spouse’s communication, the subject is shown the communication11 of their spouse and asked if they would like to revise their split decisions made in Block 1. It is indeed possible that a subject does not wish to be informed of his or her spouse’s preference but, once informed, cannot resist the mental pressure of accommodating their wishes. This phenomenon, if observed, would be reminiscent of the findings of DellaVigna et al. (2012) whose subjects avoid being asked for something they have difficulty refusing when asked. Variable reviseB-D records the revised decision after receiving communication: if a subject did not want to change their split, revise is the same as the original split. The purpose of this part of the experiment is to determine the extent to which each subject accommodates the revealed preferences of their spouse. Block 5 Food and drink choices [deferF] In this block, each subject is asked to select one of two different cookies (Food) and one of two different drinks (Juice) for themselves and for their spouse. Having chosen for themselves and their spouse each is given the option to defer their decision over both food and juice to their spouse (deferF) without knowing what their spouse chose for them. The purpose of this question is to elicit the subject’s willingness to defer their private consumption of food and drink to their spouse even when that own consumption has no effect on the spouse’s own consumption and their consumption is not observed by the spouse, as in Afzal et al. (2022). Subjects who do not defer receive the food and drink of their choice; those who defer receive the food and drink chosen for them by their spouse. The food and drink are consumed privately by each subject in their gender- specific room so the spouse cannot determine what the subject selected for themselves and the consumption of the spouse does not affect own consumption. In about a third of the sessions (13 sessions with 166 subjects), there is a penalty for retaining the decision: the subject receives one cookie and a half glass of juice of their own selection if they do not defer; but they receive two of cookies and a full glass of the options that their spouse chose for them if they do defer. The purpose of this design is to elicit a bound on subjects’ willingness to pay to defer their own consumption choice to their spouse. Table 1 shows two examples of how decisions are made and paid out in Blocks 1 through 5 for domain B (male vs. household goods). Note that the first woman chooses to spend 1400 on male goods, communicates to her spouse that she wants to spend 1200, asks to see her spouse’s preferences, and then moves towards his communication. The second woman 11All subjects retained the right not to look at the information they were given. 14 chooses to spend 1300 on male goods, communicates her true preferences, does not want to see her husband’s preferences and, when shown them anyway, does not alter her decision. Note that, in both examples, by chance, the husband’s decision is implemented, but the first couple had the woman’s choice to defer selected, and the second couple had the man’s original decision selected. In Block 5, the first woman wanted round cookies and coke, guessed what her husband wanted and then chose to consume what he wanted for her (not shown). In the second example, she chose to consume what he chose for her. Block 6 Allocating inputs across two production functions [efficiency] In this final block, the focus of the choice is not consumption but production. Each subject is asked to pick one of four possible input allocations between themselves and their spouse. Each input allocation maps into a cash income for themselves and a cash income for the spouse.12 Subjects answer two versions of the game.13 Each game has an efficient choice that maximizes the total income of the household. In Game 1, the efficient choice implies less output is assigned to self and in Game 2, the efficient outcome assigns less to the spouse. 3.3 The secret treatment We randomized the sessions equally into a secret treatment in which most decisions are shrouded and a no-secret treatment in which most decisions are not shrouded. Other than Block 5 (cookie and juice), which is always implemented during the lab experiment, all other choices are only implemented if selected at the end. Some decisions are always shrouded from spouses, irrespective of the secret condition. In particular, all Block 1 decisions (initial allocations) are always hidden from the spouse. In contrast, the communication sent to the spouse in Block 4 and 5 is never hidden. All the other decisions, if selected at the end of the experiment, are either potentially revealed or shrouded depending on the secret condition. Shrouding of these decisions is achieved as follows. At the end of the experiment, when a decision is selected for implementation, if the decision is to be shrouded, the actual decision is first mixed with a random decision, one of the two is drawn at random and then implemented. This ensures credible deniability in the secret condition: the spouse never knows whether the implemented allocation was randomly selected or selected by the subject. The precise process is described in further detail in Section C.4 of the Online Appendix. Once this process is over, for domains A, B and C, the selected individual receives tokens to be spent in the stalls, and the couple is brought together to the lab stalls associated with that decision. For domain D and Block 6 (the production game), the money is split between 12The games mimic Udry (1996) in which men and women farm separate plots and fertilizer should be allocated across the plots to achieve the maximum level of output. 13For a small number of sessions, only one version was presented to subjects. 15 the husband and wife according to the choices (or shrouded choices) of the individual whose decision was selected to be implemented. Block 5 is always selected and is implemented (either the choice or the decision to defer) privately in the gender-separated room during the experiment. 3.4 Implementation The UCT randomized controlled trial (RCT) was designed by the Africa Gender Innovation Lab of the World Bank. The UCT intervention, supported by USAID and implemented by the Catholic Relief Services, took place between September 2015 and March 2017. The baseline survey for the RCT took place between April and June 2015. A midline survey was conducted approximately one month after the last UCT monthly payment, i.e., between April and June 2017. The endline survey was conducted one year after the end of treatment, in May to July 2018. The lab-in-the-field experimental sessions took place in 27 villages from the UCT impact evaluation sample in 38 distinct lab sessions between March and May 2018, one year after the UCT intervention ended and just before the endline survey began.14 The one-year gap between the RCT and our experiment is intended to diminish the pure income effects of cash receipts to identify longer-term effects on decision–making within the household. To ensure that participants could easily walk to a location that allowed sufficient privacy, we selected villages from the evaluation sample with a nearby school. This means that our sample is not necessarily representative of the full UCT study sample. Lab sessions were conducted either on off days or after regular school hours. Participants for the lab sessions were recruited among married women included in the UCT experiment either as control or as treated. Female participants were invited to partic- ipate to the lab experiment with their husband. Upon arrival, participants were told that they would receive a compensation of 500 Naira (around $5) as show-up fee. Additionally, they were informed that they could earn significantly more, either in cash or in-kind, based on their decisions in the experimental games, in the sense that one of all the decisions made by both husband and wife that day would be selected for implementation and additional pay-out. Husbands and wives were then sent to two separate gender-specific rooms where the bulk of the experiment took place. The female room only included female enumerators; the male room only had male enumerators. As is common in settings where illiteracy is widespread, all instructions were read to subjects by a room supervisor or an enumerator 14Nine of them took place in villages that received an intervention from the Feed the Future Nigeria Livelihoods Project (FNLP). This intervention is modeled on an ultra-poor graduation program focusing on focuses on education, coaching and savings groups – but without monetary transfers. 16 who sat with each subject and asked them to make decisions by pointing to pictures using experimental prompts. This setup allowed us to reach a wide range of subjects and avoided participants revealing their choices to others verbally. When allocating money to different decisions or budgets, participants used laminated photocopies of Nigerian Naira denomina- tions with which they were familiar. Participants spent approximately 2.5 to 3 hours in the lab session and earned up to an equivalent of 3000 Naira (around 25 USD) – 500 as show-up fee and up to 2500 in cash or in kind, depending on the random draw. Recruitment was stratified to ensure that half of the lab participants had received the UCT and the other half had not. Appendix Table A1 shows that the lab sample is balanced on observables across UCT treated and controls.15 It also indicates that, on average, women in the lab sample are about ten years younger than their husbands and are unlikely to be considered adequately empowered at baseline, according to various empowerment indicators, including the A-WEAI. In terms of power, we have 506 participants in 38 lab sessions and four equally-sized treatment cells resulting from the crossing of the UCT and the secret condition. Given that we cluster all standard errors at the level of the lab sessions, we are powered enough to identify large effects of combinations of the UCT intervention and the secret condition: the minimum effect size to reject the null with 80% probability is 14 percentage points in all the dichotomous choices – such as the decision to defer or consult. Given that the UCT was found to have effects exceeding 25 pp on many outcomes, including empowerment (Papineni et al., 2024), we can reasonably expect causal effects of a similar magnitude in our experiment. For continuous variables such as split choices, these decisions are typically unaffected by the secret condition. For these, the minimum effect size that we can reject with large probability is 85 Nairas – which is small (3.4%) relative to the choice budget of 2500 Nairas. 4 Preferences over outcomes and process in the UCT control group To properly appreciate the impact of the UCT intervention, we examine, first, the behavioral patterns of experimental participants in the control group, that is, those who did not receive the UCT treatment. 15The test of balance between secret and no secret treatments also shows no significant differences. A test across all four treatment cells (not reported) shows that men and women in the UCT+secret treatment cell are 4 and 2 years older than average, respectively. All other variables are balanced. 17 4.1 Budget allocation decisions We present, in Table 3, summary statistics of all the key behavioral variables, broken down by gender. A t-statistic for the test that the means (pairwise within couples) are equal is also provided, together with the associated p-value. The first panel of Table 3 shows the four main split decisions with the full budget of 2500 Naira. Most split decisions are significantly different by gender, but the average differences between husbands and wives are not large in magnitude. Both spouses tend to divide budgets more or less equally on average, and this is true for all four splitting decisions (1250 would be an equal split) — a finding reminiscent of equal sharing in dictator games. However, these averages hide a lot of variation across the sample with significant proportions of both men and women allocating much less or much more to themselves.16 Note that men tend to allocate a larger budget share to female goods sold in the lab (1298>1250), but less to money for their wives (1062<1250); money that could spent outside the lab. Note that if we convert the amount a man gives to his wife in goods or money into the amount he keeps for himself, women are more selfish in goods than men are (1410 > 1202, p-value<0.00) and women are less selfish in money than men are (1265 < 1438, p-value<0.00). This may reflect the fact that the items sold in the lab shop appeal more to women than men or that men had planned to buy female goods on their wife’s behalf both in and outside of the lab.17 Given that these splitting decisions were always shrouded, the findings violate the idea that, given the opportunity, spouses would like to appropriate a large share of the offered budget to themselves. This is true of women but also of men who, as we shall see shortly, wield most of the power in our sample population. From this evidence we conclude that spouses have social preferences, meaning that they incorporate the expenditures of each other in their own utility function. We also note that women do not, contrary to common perception, wish to spend much more than men on household goods. While it is true that husbands and wives have relatively similar divisions of expenditures on average, the same does not hold within individual households. In nearly 50% of the couples, the difference between the allocation chosen by the wife and that chosen by her husband is more than 400 Naira. Few couples (around 10%) have an identical allocation. This implies that, while spouses have social preferences, they need not agree on how to divide a budget between specific expenditure categories. Hence, they may wish to influence 16Thirteen and 20 percent of women allocate less than 1000 to themselves in goods and money respectively and 30 and 22 percent of women allocate less than 1000 to their husband in goods and money respectively. For men, 22 percent and 11 percent allocate less than 1000 to themselves in goods and money and 20 and 36 percent allocate less than 1000 to their wife in good and money respectively. 17Recall that, although a couple must spend tokens in the stall for which they are designated, we have no way of forcing couples to buy things chosen by either the husband or wife. 18 household expenditure decisions in the direction of their own social preferences. 4.2 Decision Process Four variables are used to characterize the decision process of the couple. Two of these (defer and consult) are measured directly; the other two (misrepresent and accommodate) are constructed from the sequence of decisions described above. The first of these is deferral (defer): whether they wish to use final allocation preferred by their spouse. Our findings, summarized in Panel II of Table 3, offer arguably the most striking contrast between spouses in our study: for all split choices, wives are much more likely to defer (67-68%) than husbands (20-26%). The difference is highly significant and there is no noticeable difference in women’s proclivity to defer depending on the type of choice. This is consistent with the existence of strong social norms that men’s control of household finances is expected to be acknowledged by their wife. Furthermore, women are not less likely to defer even when expenditures are fully rival — i.e., choices A and D — then when they are not – i.e., choices B and C. This suggests that deferral decisions by women are not affected by instrumental considerations, a point we revisit below. The second measure in our experimental decision tree is the decision to misrepresent one’s true preferences from the spouse. This is measured as the difference between the communicated split (communicateB-D) and the original split (splitA-D). Panel III of Table 3 presents the average splits that subjects choose to communicate to their spouse. Comparing the allocations reported to their husbands to those they chose when decisions were shrouded, wives allocate slightly more to their husbands in choices B and D and slightly less to themselves in choice C. The magnitude, however, of these changes is small, suggesting minimal misrepresentation. This being said, we also find that, in all three cases, choices communicated by wives shrink allocations towards the middle. A similar, but much more muted, process is observed among husbands. The third measure is the decision to consult (consultB-D): subjects are asked whether they wish to be told what allocation their spouse chose. The decision to consult is always secret, but individuals may be planning to accommodate and therefore believe consulting is necessary. Panel IV of Table 3 shows the subjects’ willingness to consult their spouse’s choice. We see that wives are much more likely to consult than husbands, with few differences across goods. The only surprise is that wives’ propensity to consult is less than their propensity to defer — perhaps because consultation is not required when the choice is deferred anyway. The fourth decision is whether to accommodate or dismiss the preferred split communi- cated by the spouse. Subjects may ‘stick to their guns’ and keep their original split choice 19 even if it diverges from their spouse’s; or they may opt to partially or fully accommodate their spouse’s wishes. We regard this measure as the closest to the concept of ‘procedural empow- erment’, meaning that a subject feels entitled to impose his or her consumption allocation preferences on their spouse. To capture this idea, we construct a categorical variable that compares the revised split (revisedB-D) to the original split (splitA-D). If the subjects ‘sticks to their guns’, then their revised split is the same as their original split; if they fully accommodate their spouse’s choice, their revised split is equal to the communicated split of their spouse. Partial accommodation is when the revised decision goes in the direction of the spouse’s choice, but not completely. Over-accommodation is when the subject’s revised allocation overshoots their spouse’s. It is also conceivable that subjects are contrarian in the sense that they revise their allocation away from their spouse’s preferred choice. Panel V of Table 3 shows the revised splitting decisions of husbands and wives after having been shown their spouse’s communication. We see husbands allocating significantly less than wives to female goods in choice C while in choice D wives allocate more money to themselves than their husbands do. For both women and men, however, these differences were already present in their original splitting decisions. In Panel VI of Table 3 we examine in more detail the extent to which husbands and wives accommodate the communicated split of their spouse when revising their original choice. Here, accommodation is measured compared to the communication received from the spouse: do subjects move toward their spouse’s preferences? We see that mean accommodation by wives is much higher than that of husbands, and the difference is highly significant. This is presented in more detail in Table A2. We see that full accommodation is the modal behavior for women: in 57, 60 and 38 percent of choices. In contrast, zero accommodation despite discordant choices is the overwhelming response of husbands (84, 84 and 91 percent of choices). What these results indicate is a strong procedural inequality between husbands and wives in the study area — but much less inequality in the allocation of consumption ex- penditures. This suggests that intrahousehold allocative fairness is achieved through social preferences—which are largely (albeit not fully) shared between husbands and wives— not through procedural equality. 4.3 The demand for agency The discussion so far has focused on decisions where the interests of the two spouses are potentially divergent. In this context, agency has instrumental value because it allows each spouse to allocate the household’s consumption budget in a way more in line with their pref- 20 erences. To investigate the possibility of non-instrumental demand for agency, we examine the choices of food and drink that they make in Block 5. The stated preferences indicate which of the four private consumption bundles ci, has the highest private utility Ui(ci) for individual i. Since consumption is non-rival — what the husband consumes does not affect the wife’s choice set, and vice versa — interfering with the consumption decision of a spouse has no instrumental value. Hence if i defers his/her consumption decision to spouse j, it can only satisfy j’s desire for control. This gives us a clean measure of deferral (deferF) for non-instrumental reasons, i.e., as a way for i to increase j’s non-material utility from the decision process itself. In the last line of Panel II of Table 3 we see that 68% of wives and 19% of husbands delegate the selection of their food and drink to their spouse. Do they believe their spouse knows what they want? Since there are only two options, congruent choices should occur with a 50% probability if couples do not know each other’s preferences. This is indeed what we find: the proportion of congruent choices is 54, 54, 52 and 49% for husbands and wives for cookies and husbands and wives for drink, respectively.18 Partners do a bad job of picking for their spouse. This suggests that, on average, deferral has a material utility cost for subjects: they are less likely to consume the items they prefer. In Table 4, we examine the rate at which subjects defer when they are faced with varying costs of deferral. Since the change in the size of the budget represents a cost of deferral, observed choices map out the demand for agency as a function of the cost. In domain A (women’s versus men’s goods) subjects made three sets of deferral choices: 1) when deferral used the spouse’s decision over 2500 compared to retention using their own decision over 2100; 2) when the budgets were both 2500 and 3) when deferral used the spouse’s decision over 2100 compared to retention using their own decision over 2100. In domain D (money for the woman versus money for the man) subjects made two sets of deferral choices: 1) when deferral used the spouse’s decision over 2500 compared to the retention using their own decision over 2100; and 2) when the budgets were both 2500. In domain J (non-rival choices over food and drink) subjects made one choice but with two randomly allocated deferral budgets: 1) deferral was for two cookies and a full glass of juice while retention was for one cookie and half a glass and 2) deferral and retention had the same budget. Note that, in domain J, no subject faced both choices: the differences are across, not within subjects. Table 4 shows a downward sloping demand for the choice to defer for women: the higher the price of deferring (smaller budgets for deferring), the less likely women are to defer their 18In a similar experiment, Afzal et al. (2022) show that subjects are no better informed about the prefer- ences of their spouse than a stranger. 21 choice. In contrast, there is no change for men and most men chose to retain their choices, even when there is a cost for doing so — a low and price-inelastic demand for deferral. It is possible that women (and men) defer because they believe their spouse will chose for them something that is more advantageous than what they would have chosen for themselves. We investigate this possibility here. We observe both what someone chooses for themselves and what their spouse chooses over the same choice set and we can compare the individual payoffs for both choices to see if deferring is, indeed, sometimes optimal. Since subjects were not asked what they expect their spouse to do in case of deferral, we do not know what each of them expects to gain or lose from deferral but if subjects form rational expectations on splits, on average their guesses should be correct. With these assumptions, we can compare subjects’ private material payoff without deferral to the private material payoff they would receive if they deferred. For Table 5 we calculate the proportion of cases in which it would be optimal for subject to defer. We see that, on average, women over-defer relative to what would be optimal, a difference that is always significant. In contrast, husbands massively under-defer: 20-21% when it would be optimal to defer in 60-64% of optimal cases, a difference that is always significant. Taken together, these results indicate that deferral is driven primarily by non-instrumental considerations. This is particularly clear for men, who seldom defer and, when they do, show no responsiveness to instrumental concerns. Women, in contrast, tend to over–defer. There is no evidence that deferral is a method for improving material outcomes, but, like a normal good, demand for deferral does respond to costs. This serves as further confirmation that the deferral decision is mostly driven by intrinsic motives, such as respect for the husband’s authority or adherence to an internalized social norm.19 4.4 Aggregating agency at the level of couples So far we have examined individual decisions related to agency. We now turn to the impact on the collective behavior of spouses that is implied by our experimental results. Executive agency We start by examining the combined deferral behavior of spouses across five experimental decisions: A, B, C, D, and J (see Panel II in Table 3). Note that although the proportion of female and male deferrals are quite similar across A, B, C, D, and J (see Panel II in Table 3), they are not perfectly correlated within subject; for A to D, 19We cannot entirely rule out the possibility that women’s adherence to local social norms is reinforced by a kind of Hawthorne effect: being observed encourages participants to be on their best behavior. We can, however, rule out experimenter effects: neither female nor male participants show any inclination to follow Western gender norms. 22 correlations coefficients vary between 0.51 and 0.60 for women and between 0.36 and 0.58 for men. Correlation coefficients between deferral in domains A through D and deferral in the food-and-juice experiment are even lower: between 0.22 and 0.31. In addition, deferral decisions are not synchronized across spouses: in only about half of the couples do we observe the wife deferring to her husband and the husband not deferring to his wife, and this is true for all five experimental decisions. In 26-28% of couples, neither the wife nor the husband defer; in 12-14% of them, both husband and wife defer; and in 8-11% of them, the husband defers and the wife does not. Thus, there is significant variation in deferral within individuals and little evidence of coordination in deferral behavior across spouses. Consultative agency Next we turn to our measure of consultative agency, Ch w, which measures how much of the gap (cww − chw) between the preferred allocations of the wife and her husband is bridged by the husband through the process of misrepresentation, consulting, and accommodation. To recall, Ch w = 1 means that the allocation implemented through consultation is that preferred by the wife, and Ch w = 0 means it is that of the husband. A number between 0 and 1 implies a compromise between the two. The husband’s consulta- tive agency Cw h is computed in a similar manner. By construction, Ch w and Cw h cannot be computed if cww = chw. This happens in around 10% of the cases – see column 1 of Table 6. In column 2 of Table 6, we report the average values of Ch w and Cw h for choices B, C, and D. We see that Ch w is very low – between 2.6% and 5.2%, indicating that the consultation process reproduced in the lab moves the allocation chosen by the husband only 2.6 to 5.2% in the direction of the wife’s preferred choice. In contrast, Cw h takes values between 31.7% and 40.2%, indicating that wives grant a lot more consultative agency to their husband when given executive agency by the experimenter. Next we decompose Ch w and Cw h in three separate components (see equation 6 which we reproduce below for convenience): misrepresentation (i.e., self-censoring); willingness to consult the spouse; and accommodation when consulting. Ch w = 1− (cww − cw comm w cww − chw + (1− kh)(c w comm w − chw) cww − chw + kh(c w comm w − ch accom w ) cww − chw ) The results are shown columns 3, 4 and 5 of Table 6. Column 3 reports first term cww−cw comm w cww−chw . This term can lie outside the cww to chw range. We see that, for choices B and C – which involve allocating a budget between a private good or a household good – there is little misrepresentation/self-censoring: values are close to 0. Things are different for choice D – which divides a monetary budget between the spouse: here wives reduce what they report 23 to their husband by 18.9% of the difference between their private preferred allocation cww and that of their husband. In contrast, all values are uniformly small for husbands, indicating no self-censoring on their part. The second term (1−kh)(c w comm w −chw) cww−chw captures the loss of consultative agency because the spouse refuses to be informed of the participant’s chosen allocation.20 It is shown in column 4. We see that this is where most of the loss of consultative agency occurs: for wives, this term accounts for between 67.6 and 83.8% of the difference between cww − chw; for husbands it is smaller but still represents between 45.6 and 46.9%. The third term kh(c w comm w −ch accom w ) cww−chw is shown in the fifth column and represents the loss of consultative agency due to lack of accommodation by a consulting spouse. This term is not negligible, accounting for between 10.9 and 18.2% of the gap for wives and between 6.2 and 28.4% for husbands. But it is smaller in magnitude than the second term largely because spouses seldom consult: 15 to 20% of husbands and 52 to 56% of wives consult, depending on the allocation decision. The experiment allows us to compare Ch w and Cw h to the counterfactual allocation Ĉh w = 1− cww−ch accom w cww−chw that would result if spouses were forced to hear the allocation that their spouse communicated to them. The results of this calculation are presented in column 6 of the Table. It shows a slight increase in the consultative agency of women, rising by a factor of 2.4 to 3.8 times relative to the values reported in column 2. The modesty of this improvement reflects the fact, already documented earlier in the paper, that husbands do not accommodate the communicated allocation of their wife. In contrast, forcing consultation would improve the consultative agency of husbands by 45 to 68% – assuming counterfactually that they would delegate executive agency to their wife. Table A3 in the online appendix examines whether material allocations of consumption budgets lean more towards women’s preferences when they have executive vs consultative agency – i.e., we compare cww to khc h accom w + (1− kh)c h w. We find that the material agency of women is slightly but significantly better when they have executive agency, but the differences are quite small on average. Additional analysis, not presented here, nonetheless shows that these small differences in the average allocation hide large within-couple differences: how husbands allocate the consumption budget is rarely what wives would have chosen. This means that, for women in our sample, lack of executive agency is consequential: consumption is not allocated the way they want. 20Note that, because we always show participants the communication of their spouse, this is not observed in the experiment, but can be inferred. 24 4.5 Allocative efficiency in production We now turn to Block 6, which allows allocative inefficiency in production, either due to a desire to increase one’s individual income, or driven by fairness considerations in input allocation. Most of the theoretical literature on intrahousehold allocation implicitly or ex- plicitly assumes efficiency in production decisions. Yet some evidence suggests that input endowment effects can impede production efficiency of households in Africa (see Udry, 1996). The two production games played in Block 6 examine whether subjects make decisions that are efficient for the couple. Recall that, in one game, maximizing household production requires giving more input to the spouse and in the other, maximizing household production requires giving more input to oneself. We see from Panel VII of Table 3 that, in three of the four cases, about half of the subjects choose the efficient outcome that maximizes their joint income. The one exception is that, unlike female subjects, male subjects are less likely to choose the efficient allocation when it yields less for themselves — a statistically significant gender difference. This is a priori surprising: since the subjects are paid in cash in front of each other for this block, they could easily compensate each other for making efficient choices after the experiment. The fact that most men do not opt for the efficient outcome when it benefits their wife suggests that they do not, in fact, expect to be able to costlessly appropriate the surplus ex post.21 Combined with our earlier results showing that subjects often defer to their spouse even when doing so reduces the household surplus, these findings indicate the existence of prefer- ences over the process by which a particular allocation is obtained: it is preferable to choose the “right” allocation rather than the “best” outcome and, ex post, reallocate. With this understanding in mind, we now examine whether offering an unconditional cash transfer to women modifies these preferences. 5 Effect of the UCT on intrahousehold agency Equipped with a better understanding of the decisions made in the lab by couples who did not receive the unconditional cash transfer (UCT) intervention, we now examine the effect of the UCT intervention on female empowerment using the full sample of treated and control participants. The impact evaluation study by Papineni et al. (2024) examines the effects of cash trans- fers on various economic and social outcomes, with a particular focus on women’s empow- erment. The findings reveal that, one year after the program ended, beneficiary women 21In Online Appendix B we examine the full set of choices and show that the best model of decision–making is a blend of preferences over both input and output allocations. 25 experienced a 20 percentage point increase in enterprise ownership, while non-beneficiary women saw a 13 percentage point increase. This surge in female-led entrepreneurship sig- nificantly enhanced household consumption and reduced food insecurity. However, while the cash transfers effectively mobilized previously underutilized female labor, they did not significantly alter societal attitudes toward women working outside the home. As part of the impact evaluation of the UCT intervention, RCT participants were asked a series of questions about decision–making in their households before and after receiving the program. These questions were used to create indices of empowerment based on the Women’s Empowerment in Agriculture Index (WEAI).22 The impact evaluation study by Papineni et al. (2024) reports significant positive impacts on this index for the full RCT sample. Specifically, the cash transfer treatment led to a 0.69 standard deviation increase in the pro-WEAI index for beneficiary households compared to the control group. In Table 7 we reproduce these findings for our much smaller sample of lab participants: while the A-WEAI index constructed from answers to the baseline survey is balanced across UCT treatment and control, we observe a significant effect on pro-WEAI, the index collected at endline – implying an improvement in female empowerment in our sample as well. Significant improvements are also reported by Papineni et al. (2024) for specific compo- nents of the pro-WEAI index, such as group membership, inputs into productive decisions, and ownership of assets and plots. Additionally, there was a 0.28 standard deviation increase in the decision–making index, which captures control over enterprise activities and income use. Overall, this shows that the cash transfer program significantly enhanced women’s eco- nomic empowerment by increasing their participation in decision–making and control over resources. In spite of these effects, the program did not significantly impact psychosocial components of the index, like self-efficacy and attitudes toward gender-based violence. In this section, we complement these findings by testing whether the UCT intervention affected the gender distribution of executive and consultative agency within couples, separately under the no- secret and secret conditions. 5.1 Testing strategy To investigate the effect of the UCT treatment and the secret condition in the lab, we estimate a model, in equation 7, that examines the interaction of the secret condition (S) and the UCT treatment (T) as four independent categories (with Ti = 0 and Si = 0 as the 22One, referred to as A-WEAI by Malapit et al. (2019), is based on 6 questions collected at baseline, and another, referred to as pro-WEAI, is based on 12 different questions asked at endline. 26 omitted category) and also, in equation 8, as the interaction of two treatments. xi = β0 + β1[T 1 i S 0 i ] + β2[T 0 i S 1 1 ] + β3[T 1 i S 1 i ] + ϵi (7) xi = β′ 0 + β′ 1Ti + β′ 2Si + β′ 3TiSi + ϵ′i (8) xi is any of the budget allocations or other decisions made by the subject, variables Ti and Si indicate the UCT treatment and secret condition and ϵi is an error term which we cluster at the level of the session to control for unobserved session effects that would lead to correlated effects. Note that β1 = β′ 1, β2 = β′ 2, ϵi = ϵ′i and β3 = β′ 1 + β′ 2 + β′ 3. Since the two regressions are almost exactly the same, we only report the coefficients for equation 7 and the p-value of the coefficient β′ 3. Similar regressions are estimated for the two subsequent allocation decisions subjects are asked to make. The three dichotomous measures of procedural empowerment — defer, consult, and accommodate — are regressed on treatment. If the UCT treatment empowers women in a procedural sense and makes them more openly assertive, we should observe that their decisions to defer, consult, and accommodate all fall with treatment. The need to misrepresent would also fall. Furthermore, if the treatment also reduces the power of husbands, or makes them take their wife’s wishes into consideration, we should observe increases in the rate in which men defer, consult, and accommodate. 5.2 Budget allocation decisions We begin with the initial split decisions of husbands and wives. As explained in Section 3, initial split decisions (Block 1) are always kept secret in the experiment. Thus, choices made by subjects can be interpreted as representing their true preferences and any treatment effect we find can be seen as a shift in preferences induced by the UCT.23 To investigate this, we show in Table 8 the estimates obtained by regressing split decisions on a UCT treatment dummy. Since choices are shrouded, we expect no effect of the no- secret treatment.24 We see that women who experienced the UCT increased the allocation to themselves in goods and money by 67 Naira in domains A and D, for a total change of 134 Naira. Only the total allocation is significantly different between those who received the UCT and those who did not (the p-values for domain A and D are 0.103 and 0.107 respectively). The total change in A and D for men, by comparison is 12 Naira, an amount 23A shift in preferences may arise because the utility function of the subject has changed or because of an income effect induced by the UCT. 24We confirm in Online Appendix Table A4 that the secret/no-secret treatment has no separate significant effect on split decisions. 27 both statistically and economically non-significant. There is also suggestive evidence that women allocated more to the household versus their husband (domain B). The increase in household allocation is 51 Naira (p-value = 0.116) The UCT treatment changed the amounts women are willing to allocate to themselves (and their household compared to their husband). The changes are marginally significant, but economically small. The total change in the amount women allocate to themselves after a eighteen-month cash transfer program is 134 Naira, 5% of the average allocation across domains A and D. Since men do not change their allocation, the treatment can be seen as widening (slightly) the divide between women’s and men’s preferred consumption bundles. This finding is consistent with a slight rise of individualism among women who received the UCT. 5.3 Decision Process In Table 9 we estimate regression model (7) for deferral decisions. Unlike the split deci- sion, defer decisions are only shrouded in the secret condition, something participants were informed of before making a decision. Consequently, we include both a UCT and a secret effect. We find that, in all four rival domains, wives defer much less often when they receive the UCT treatment under the secret condition. The difference is large in magnitude – be- tween 12 and 17 percentage points — and it is significant in all cases and in the aggregate (column A-D). No such changes are observed for husbands who, as we noted earlier, are much less likely to defer on average. We see the results of the same magnitude for the non- rival food and juice (J) but the coefficients are not significant. Combined with the findings from Table 8, these results suggest that the UCT treatment has made women want more consumption and agency. But it has not made them more vocal or openly assertive: they continue to defer to their husband if this decision is observed. This is consistent with deferral being a social norm internalized by participating women, since it is also present in the Secret treatment. The fact that this behavior changes with the UCT indicates that the treatment changed this internalized norm for some female recipients, but did not change the fear of possible retribution for breaking it. This is quite a remarkable outcome, and not one that would be observable outside this experiment. Table 10 presents a similar analysis for the decisions to misrepresent, consult, and ac- commodate. Note that the initial allocation is always secret, the communicated allocation is never secret and the decision to consult is always secret. The decision to accommodation could be revealed in the no-secret condition, since the spouse knows both their communica- tion and the final allocation. Despite the fact that the secret condition only directly affects 28 the accommodation decision, it might indirectly affect misrepresentation and consult, and therefore we examine all four treatment conditions. In the decision to misrepresent, presented in Panel A, the dependent variable is the difference between the communicated split and the initial (secret) split: a negative coefficient implies that the communicated split is smaller than the secret split. We find that women who received the UCT tend to misrepresent more in the secret condition but the effect is only significant at the 10% level in one of the three regressions. For husbands we observe more misrepresentation as a result of treatment, but only in choice D. The lack of systematic pattern across choices makes us suspect the result is not robust. There is no evidence for an effect of the UCT treatment alone. In the second panel of Table 10 the dependent variable equals 1 if the subject manifested a desire to see their spouse’s communicated split. In five of the six regressions, we observe a large fall in the likelihood of consultation as a result of treatment, with all but one significant effect being concentrated in the secret condition. The magnitude of these effects is large, especially in the secret condition: wives reduce the likelihood of consulting by a combined 13 to 27 percentage points, while husbands reduce it by 16 to 23 percentage points (starting from a much lower base.) The fact that these changes are significant primarily in the secret condition suggests the presence of hidden tension between treated spouses, tensions that they are trying not to learn about. The third panel of Table 10 focuses on accommodation. Here we find little evidence of treatment effects: except for one significant coefficient at the 10% level, there is no dominant pattern across choices. Even in secret, treated women are not less willing to accommodate their husbands’ communicated allocation; the pressure is too direct. These findings contribute to a coherent picture of the effect of treatment on procedural fairness in couples: treated women become secretly more demanding and less willing to defer and consult, as long as these decisions can be hidden; treated husbands also tend to shift their budget allocation towards a more selfish posture, although the effect is not statistically significant; and they consult less often, especially if this decision is less observable. 5.4 The Demand for Agency In the control group, we saw that there is a demand for agency: the willingness to defer is sensitive to the cost of deferral. In a parallel analysis, we show that the secret treatment shifts the demand curve, increasing the demand for agency at every price. While changes in consumption bundles or increased demand for agency could reflect an income effect of the UCT treatment on household assets and female labor force participation, the impact of the 29 secret treatment cannot be an income effect as the secret treatment was randomized across individual on the day of the experiment. We examine the effect of the UCT and no-secret treatments in Table 11 where we com- bined all the deferral decisions subjects make in domains A (female v. male goods) and D (cash for the wife v. the husband) and in the food and drink domain J . By combining decisions made in Blocks 2 and 3, we can estimate the response of the demand for agency (i.e., non-deferral) to its cost.25 The results confirm what we found earlier: that increasing the cost of agency reduces women’s willingness to exert agency. For men, the cost of agency has no significant effect on deferral, except for domain D (cash to the wife) where increasing the cost to deferral actually increases the propensity to defer.26 In the secret condition the deferral decision cannot be discovered by the spouse; otherwise, it can. We expect the secret condition to potentially have an effect on deferral: if a wife is enticed to demand more agency as a result of the UCT treatment but is afraid of retribution if it is revealed that she did not defer to her husband, she may refrain from deferring only in the secret condition. The impacts of the treatment match those shown above: women (but not men) increase their demand for agency if they received the UCT treatment and are in the secret condition. This result is significant both in total — i.e., UCT recipients in the secret condition are different from non-recipients in the no-secret condition — as well as on the margin — i.e., the combined impact of the UCT and secret treatments is different from the sum of the UCT treatment and secret condition. 5.5 The distribution of agency within the couple So far we have examined the impact of the UCT and secret treatments on individual behavior. We now turn to its impact on the collective agency of spouses. We start by creating four variables that average the proportion of paired spousal decisions across experimental choices A to J. The first variable represents the proportion of paired decisions in which the wife defers and the husband does not, and so on. Equal agency is achieved when both defer – and it can be seen as a premise for a negotiated outcome. In contrast, disharmony in the couple can be expected when neither defer. We already know that, on average, the combination of UCT and secrecy induces less deferral by wives without changing husbands’ behavior. But these averages could mask shifts in opposite directions across couples: if men who deferred to a non-deferring wife no longer do as a result of treatment (i.e., they oppose the treatment), 25Coefficient estimates for the cost variable in the J domain are across-subject comparisons since no subject made a decision over deferring a cookie and drink at two different costs. 26This can be read as a reaction to experimental demands for deferral, but should not be interpreted as a reaction to female agency; men are pushing back on the experimenter, not necessarily their wives. 30 we could observe both an increase in the proportion of equal agency couples (both defer) as well as in the proportion of disharmonious couples (neither defer). To investigate this issue, we regress each of these four proportions on the UCT and secret treatments .27 Results, shown in Table 12, show that the combination of the UCT and the secret treatment causes a transfer of around 8-9% of couples from the “Wife defers, husband does not” to “No one defers” – and another 4-5% transfer from “Both defer” to “Husband defers, wife does not” (significant only for the former). These findings indicate no correlation between deferral among husbands and the change in deferral by their wife – i.e., it is not the case that husbands react negatively to a reduction in deferral by their spouse. We nonetheless note a reduction in the proportion of couples who both defer from around 16% to 11% and an increase in the proportion of no-deferral couples from approximately 23% to 32%. In other words, in our sample, equal agency has gone down and the risk of conflict has increased – a finding that is largely due to the fact that husbands do not accommodate the increased demand (or rising aspirations) for women’s agency that the UCT induced. Given this lack of accommodation, it is no surprise that pent-up demand for executive agency remains hidden: it is only expressed in secret. May husband nonetheless have accommodated their wives’ increased demand for agency by consulting them more? To investigate this possibility, we regress Ch w and Cw h on the UCT and secret treatments, as done in Table 12 for deferral. We find no evidence on an effect of treatment on the consultative agency of either husbands or wives.28 The same holds for the aggregate consultative index when the spouse is forced to consult. We also find no evidence of any effect of the UCT or secret treatments on material allocations resulting from consultation among spouses.29 This indicates that the UCT intervention has had no effect on husbands’ willingness to consult their wife over the allocation of consumption in the household, suggesting that the intervention has not changed gender norms in favor of women. This evidence also contradicts the collective thinking hypothesis: if collective thinking was the reason for women not to want to exert executive agency over consumption, then we should observe husbands consulting them – which we do not. 5.6 Aggregate Welfare and Secrecy Here, we examine the effect of treatment on the expected payoff of experimental subjects. As explained in Section 2, one of a long list of possible choices made by subjects is drawn at 27Similar but slightly more significant results obtain using a fractional logit model. 28Of the 18 estimated coefficients (3 regressors x 3 decisions x 2 spouses), only one is significant at the 10% level – see Appendix Table A5. 29See Appendix Table A6. 31 random to determine final payoffs. We focus on the most important choices made by both spouses, namely splitA-D and deferA-D. Using simple rules described in the note to Table 13, we construct a variable that measures the payoffs πf and πm that a male and female subject can expect to receive based on their own decisions. These expected payoffs are what subjects can expect to receive purely for themselves at the end of the experiment based on their own split and defer decisions and the decisions of their spouse. Coefficient estimates are presented in Table 13. Although the point estimate for the UCT is positive, it is not significant. The material payoff of women only increased in the UCT/secret treatment. The increase in this category is about 5 percent of the total possible payouts. To the extent that the rule for selecting payoffs in the experiment mimics what happens at home, this suggests that the UCT treatment is most likely to raise the material welfare of women if it is accompanied by some form of secrecy. Without secrecy, women overwhelmingly delegate consumption decisions to their husband, whose social preferences are largely unaffected by treatment. 5.7 Allocative Efficiency in Production Finally, we examine the impact of the UCT treatment and other experimental conditions on the efficiency of couple’s production decisions. Based on the existing literature, we have no reason to believe that the UCT treatment should increase or decrease efficiency, since theory suggests efficiency is a dominant strategy in all households, irrespective of the decision– making process. Moreover, the game does not allow deferring production decisions to one’s spouse, so there is no way to express or avoid agency — and thus no possible effect of the UCT through demand for agency. Table 14 summarizes the impact of the UCT treatment and secret conditions on decision– making in the production games. We examine two measures of efficiency: subjects who played both versions of the game are strongly efficient if they chose the efficient outcome in both games; subjects who played only one version are weakly efficient if they chose the efficient outcome in that game. In the top Panel of Table 14, we see that, among control participants (no-UCT) in the secret condition, women chose the efficient option less frequently. This is true for both sets of subjects, i.e., those who played both versions of the game and those who only played one version. Based on model predictions, this finding indicates lower social preferences for women in the condition that mimics a non-cooperative state. There is no similar effect of the secret condition on men (see Panel II) but, on average, they choose the efficient option less frequently — suggesting that, unlike women, men act equally selfishly in both the secret 32 and no-secret conditions. Turning to the UCT, we see that it induces a large drop in the frequency of efficient choices by women in the no-secret condition. Since this l