Women’s Empowerment, Agricultural Extension, and Digitalization: Disentangling Information and Role Model Effects in Rural Uganda

In many developing countries, agricultural extension services are generally biased towards men, with information targeted mainly to male members of a farming household and in formats that are rarely tailored to female members. Nevertheless, female farmers may also benefit from such services as this may affect their ability to make informed decisions, resulting in increased farm productivity, household income, and welfare. We conduct a gendered field experiment among maize-farming households in eastern Uganda to test whether video-enabled extension messaging affects outcomes related to maize cultivation. In this experiment, men, women, and couples are shown randomly assigned videos about improved maize management practices in which male, female, or both male and female actors are featured. We first vary exposure to the videos by gender to test the effects of changes in intra-household information asymmetries, investigating whether involving women as recipients of information increases their ability to participate in household decision-making, and thus their involvement in household production choices. We then vary exposure to the gender of the actors in the videos to test for role-model effects, exploring whether involving women as information messengers challenges the idea that decision-making is a predominantly male domain, in turn affecting women’s outcomes. Results show that targeting women with information increases their knowledge about improved maize management practices, their role in agricultural decision-making, the adoption of recommended practices and inputs, production-related outcomes, and the quantity of maize women sell to the market. Results for the role-model effects are mixed, and are evident more in joint household outcomes than individual women’s outcomes. Overall, our findings suggest that in the context of our study, extension efforts aimed at directly addressing intra-household information asymmetries may be a first-best means of empowering women in agriculture. Other, more subtle means that seek to influence perceptions and norms about gendered roles in the household may not generate expected effects or work via expected impact pathways, though they remain worth further exploration.


INTERNATIONAL FOOD POLICY RESEARCH INSTITUTE
The International Food Policy Research Institute (IFPRI), established in 1975, provides researchbased 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.

Introduction
There are many well-established benefits to empowering women farmers in developing countries. A more prominent role for women in the farm household has been shown to result in more efficient allocations of scarce resources within the household, more equitable distributions of the returns to investments in household production, and general improvements in welfare and poverty (Polato e Fava & Arends-Kuenning, 2013: in Fiala & He, 2017McCarthy & Kilic, 2017;Croppenstedt et al., 2013;De Brauw et al., 2014). Involving women in the choice of crops may also lead to more nutritious dietary outcomes at the household level (Quisumbing & Maluccio, 2003;Duflo & Udry, 2004). More generally and from a human rights perspective, there is intrinsic value in empowering women (Kabeer, 1999;Meinzen-Dick et al., 2019).
Empowerment is seen as a process where those (women) who were previously denied the ability to make and act upon important choices in their lives gain such ability (Kabeer, 1999). This ability to make choices is based on three main components: access to resources including material assets, informational resources, and social capital; agency, including a person's ability to act on their decisions; and achievement, or social and economic outcomes such as income and well-being. Women's empowerment can imply that women have greater ability to make and act upon choices both individually and jointly within the household (Johnson et al., 2016(Johnson et al., , 2017. But empowering women in agriculture and achieving these welfare-improving outcomes is often constrained by multiple, related factors. These include the fact that women often have less access to information than men on a range of productivity-enhancing technologies and practices, which necessarily limits their participation in household decision-making on agricultural production (Fisher & Carr, 2015;Bryan et al., 2016;Lefore et al., 2019). The causal factors behind intra-household information asymmetries have been studied in considerable depth. For example, prior studies demonstrate how men rarely discuss agronomic practices or management choices with their wives or female partners (Magnan et al., 2015;Fisher & Carr, 2015). Other studies demonstrate how agricultural extension services rarely recognize women in their role as agricultural producers (Haug, 1999;Doss & Morris, 2000;Doss, 2001). Even though women often provide substantial amounts of labor, men are assumed to be responsible for farming and hence take most of the important decisions (Meinzen-Dick et al., 2011). As a result, agricultural extension-including rural advisory services and agricultural information campaigns-tend to target only men within the households, failing to recognize gendered power dynamics governing intra-household information exchanges, confirming a worldview in which women are largely absent from agricultural decision-making, and hindering empowerment of women in this domain (Due et al., 1997;Doss, 2001;Quisumbing & Pandolfelli, 2010;Netsayi et al., 2017).
Due partly to this neglect of women in agriculture and agricultural extension, there is now a growing literature that aims to demonstrate the importance of empowering women in agriculture and reaching women through extension services, and provide greater nuance on the topic (Quisumbing & Pandolfelli, 2010;Quisumbing et al., 2014;Doss et al., 2018). Recent studies of gender productivity gaps have pointed out that the use of modern productivity-enhancing inputs on female-managed plots is significantly lower as compared to male-managed plots (Ali et al., 2016). Considerable attention has been given to women's access to productivity-enhancing technologies and practices in sub-Saharan Africa, where women play a significant role in the management and production of crops (Ndiritu et al., 2014), but are generally not effectively reached by extension services (Fletschner & Mesbah, 2011;Ndiritu et al., 2014;Magnan et al., 2015;Kondylis et al., 2016;Lambrecht et al., 2016).
In this paper, we examine the way in which extension services aided by information and communications technologies (ICTs) affect women's access to informational resources, their agency, and their achievements in farming, all of which directly relate to women's empowerment in agriculture. Specifically, we conduct a field experiment among maize-farming households in eastern Uganda to test whether ICT-enabled extension messaging affects outcomes related to maize management and production. The intervention consists of short videos that provide viewers with information and recommendations on appropriate input use, management practices, and investment strategies that can be followed to obtain higher maize yields. Three versions of the video were created-one featuring a male actor, another featuring a female actor, and a third featuring both actors together-with no difference in the information contained in the video. Varying the exposure to the actors of different gender in this way allows us to test a first hypothesis related to the importance of role models to empower women. Additionally, men, women, and couples were randomly assigned to view the video, regardless of the version. This allows us to test a second hypothesis related to changes in intra-household information asymmetries.

Prior literature
This section offers a non-exhaustive review of the wide and rich literature on gender and agriculture. Specific emphasis is placed on the gendered dimensions of role models and intra-household asymmetries of information, drawing on examples that highlight their influence on technology adoption and extension service provision. We then tie these ideas to the hypotheses tested with our experimental approach.

Intra-household information asymmetries
Much of the literature on gender, agriculture, and technology adoption approaches intra-household information asymmetries as an example of inefficient intra-household resource allocation, where the key resource in question is information. Intra-household information asymmetries both affect and are affected by women's status, role, and welfare relative to other, typically male, household members (Fletschner & Mesbah, 2011;Doss et al., 2018). In general, information asymmetries are the result of differential costs on the acquisition and use of information that results in a lack of awareness about available resources or how to allocate them more efficiently, which, in turn, may result in forgone income-enhancing or assetaccumulation opportunities or acceptance of unnecessarily high risks by one of the spouses (Fletschner & Mesbah, 2011).
There is ample evidence that women face information deficiencies and asymmetries relative to men in a range of circumstances, and thus may hold weaker bargaining positions or be more vulnerable within their households or communities (Fletschner & Mesbah, 2011). In agriculture, women farmers have been shown to be more deprived of information regarding good agronomic practices (Doss & Morris, 2000;Doss, 2001;Lambrecht et al., 2016). In certain situations, this may be directly linked to unequal access to agricultural extension and advisory services between men and women farmers (Ndiritu et al., 2014;Lambrecht et al., 2016). This unequal access may contribute to lower adoption rates of improved agricultural practices and technologies among women, although women are found in many contexts to be willing to adopt such practices (Bryan et al., 2016). For example, studies show that with equal access to extension services, land, and labor, men and women farmers in male-headed households in Ghana would be as likely to adopt modern agronomic practices (Doss & Morris, 2000), while women in female-headed households would be as likely to adopt innovations in banana cultivation as men in male-headed households if they were as knowledgeable about the technology (Kabunga et al., 2012).
From this, we recognize that women's access to information-specifically, to information provided by agricultural extension services-is subject to both extra-household and intra-household constraints. The extra-household constraints can be infrastructural and logistic, for instance, women not being targeted for extension services, women lacking the money to travel to extension training locations, or women lacking the time to attend because of household management, domestic and reproductive responsibilities (Wodon & Blackden, 2006;Fletschner & Mesbah, 2011). Further, extra-household constraints may come in terms of information content or delivery: it may not be adapted to women's interests or needs, or may not recognize women's role as agricultural producers, and therefore may not appeal to women. Human capital constraints may also play a role given women's generally lower levels of education in many rural contexts in developing countries. Norms limiting women's mobility and women's interaction with men may impose additional constraints (Fletschner & Mesbah, 2011). These extra-household constraints to women's access to information mean that, in many situations, women may rely more on informal networks for gathering information, but where these networks are gender-specific and gender-segregated, they too may face similar problems associated with asymmetric information (Magnan et al., 2015).
Where men do not face those constraints, a (married) woman's interaction with her husband may often be her (intra-household) source of information. This assumption underpins most extension strategies that target the male household head (Fletschner & Mesbah, 2011). Yet the assumption requires that the preferences of male and female co-heads within a household align; that household resources, including information, are shared; and that households cooperate to reach Pareto-optimal outcomes. This unitary theory of the household, attributed to Gary Becker (1993), has been challenged in theoretical work (Lundberg & Pollak, 1994;Pollak, 1994;Alderman et al., 1995) and rejected in empirical work (Udry, 1996;Duflo & Udry, 2004;Ashraf, 2009;Iversen et al., 2011).
Non-cooperative models of the household posit that where preferences of the male and female co-heads conflict, an informed male co-head may choose not to share information and sustain a monopoly over information in order to allocate household resources towards his individual preferences (Fletschner & Mesbah, 2011). Information flows from male to female co-head only where the female co-head possesses some significant degree of bargaining power, or is an important intermediary in accessing resources (e.g., credit or transfer payments specifically designated for women). And even if preferences align and resources are pooled in the household, information flows from male to female co-head are never guaranteed: if each specializes within the household, information sharing may be deemed unnecessary or too costly.
Several studies explore non-cooperative models and intra-household information asymmetries in the context of developing-country agriculture and the provision of agricultural extension services. Kondylis et al. (2016) start from the observation that information about sustainable land management practices in Mozambique does not reach women farmers as effectively as men farmers, and is not perfectly shared between men and women farmers within the households. They find a positive effect on awareness and adoption among women farmers when they introduce additional female extension agents who reach out to women farmers. The positive effects, however, emerge over a longer-time period such that women were rarely among the early adopters. Pan et al. (2018) explore similar issues in Uganda with women model farmers who facilitate training and access to hybrid maize seed, particularly for fellow women smallholder 6 farmers. While Pan et al. (2018) only look at household-level outcomes, they do find significant positive effects of the intervention on the adoption of low-cost recommended agronomic practices and inputs, and on household food security.
Similarly assuming information does not freely flow within households, Lambrecht et al. (2016) investigate whether extension services are more effective if the information is provided to both the male and female cohead, the male co-head, or the female co-head in the household. Focusing on integrated soil fertility management practices in eastern DR Congo, they find that joint participation of male and female co-heads increases adoption most. They also find that participation of female co-heads affects adoption of laborintensive practices more than capital-intensive inputs, even on jointly-managed plots. Fisher and Carr (2015) explore explanations for the relatively low adoption rates for drought-tolerant maize varieties among women farmers (in male-headed households) in eastern Uganda. They do not find support for the idea that differences in technology preferences are at the basis, but that differences in adoption are related to awareness, and particularly whether the woman farmer received information about the varieties, whether credit was made available, and whether the plot was under legal or customary tenure.
Taken together, findings from these studies illustrate that information does not necessarily flow without friction between co-heads of a household, particularly from male co-head to female co-head. Furthermore, it shows that women may not have the bargaining power or access to intermediary resources necessary to convince their male co-heads to invest in certain types of technologies and practices, particularly those that are capital-intensive.

Role models and aspirations
There is also a growing literature that explores the socio-psychological dimensions of gender inequality. For instance, an emerging literature investigates the importance of role models in challenging gender stereotypes and empowering women in domains where they are active but lack voice and agency. Role models are defined as individuals who inspire people to make similar choices, or adopt a similar set of values, and to achieve comparable results (Madhavan & Crowell, 2014;Porter & Serra, 2019).
Role models are seen as important in stimulating aspirations and the development of an internal locus of control. They can update belief in one's own ability (self-efficacy) or beliefs about the returns to investments, especially for disadvantaged social groups that have few examples of success (Beaman et al., 2012;Riley, 2017). Updated beliefs in self-efficacy and returns to investments can, in turn, raise aspirations and increase people's ambitions, which create the motivation to work hard and attain the success projected by the role model (Riley, 2017). Inspiring films about successful farmers' life choices promoted welfareimproving aspirations among Ethiopian farmers (Bernard et al., 2015). Women chief village councilors in rural India raised parents' and girls' aspirations with regard to education and adult life opportunities Electronic copy available at: https://ssrn.com/abstract=3505653 Electronic copy available at: https://ssrn.com/abstract=3505653 (Beaman et al., 2012). In Nicaragua, proximity of women promoters of a conditional cash transfer program made women more optimistic about the future, happier in life, and less fatalistic (Macours & Vakis, 2014).
Female students in the U.S. were more likely to opt for economics after women alumni pointed out the importance of studying economics for success in their careers (Porter & Serra, 2019).
Evidence shows that role models not only increase aspirations, but also lead to changes in the choices they make. In Ethiopia, while the inspiring films did not specifically mention what choices made the role models successful, viewers were more likely to save and take out a loan, and to save or borrow higher amounts. The number of children enrolled in school and school-related expenditures went up as well (Bernard et al., 2015). Porter and Serra (2019) observed an increase in female student enrollment in an intermediate economics class after women alumni's testimonies. The proximity of women promoters in Nicaragua positively affected investments in education and nutrition (Macours & Vakis, 2014). Raghunathan et al. (2018) add nuance to the relation between role models and altered choices, in the context of rural India, showing that the extent to which greater aspirations are translated into altered choices depends on the type of aspirations-income, assets, or status-and the group that people compare themselves with.
The power of role models can also be linked to the fact that people's tendency to conform (i.e., not wanting to deviate from the group) can be challenged by an influential person whose behavior is non-conformist (Thaler & Sunstein, 2008). Role models thus have been found important in challenging role incongruity, which can be defined as prejudiced views and cognitive biases about the capabilities of specific social groups in specific social roles that arise from a combination of perceptions about the characteristics of members of that social group and perceptions about the capabilities and characteristics that specific social roles require (Eagly & Karau, 2002). Role models can affect women empowerment directly: as women start questioning cultural norms and gender stereotypes, their locus of control may shift, resulting in increased entry into traditionally male-dominated domains. Indirectly, role models can challenge beliefs and stereotypes about lesser abilities of another group held by the group whose abilities are not underestimated. For example, Beaman et al. (2009) show that the appointment of women leaders to Indian village councils improved men's perceptions of women's leadership abilities. Evans (2017) explains how, in Zambia, gender-status inequalities were reduced mainly because the example of economically successful women weakened gender beliefs.
Other mechanisms, including peer effects (which are linked to recognition and conformity) and gender homophily may also be at play in the case of role models of the same sex. Gender homophily is defined as the preference for interaction with individuals of the same sex, and is linked to having more trust in individuals of the same social group (Laniado et al., 2016). Such mechanisms imply that the information contained in a message brought by role models of the same sex is better understood and more trusted, thereby contributing to changes in an individual's choices and improving chances of success.

8
The extent to which role models actually help people become more successful is less evidenced in the literature to date. Riley (2017) find a direct positive effect of a movie projecting a poor Ugandan girl championing in chess and at school on the educational achievements of lower and higher secondary school students. Beaman et al. (2012) show that women leaders in Indian village councils not only raised aspirations, but also boosted girls' educational attainment and reduced girls' time spent on domestic chores. At the same time, there is inconclusive evidence on whether role-model effects accrue more to lowperforming people and people with modest aspirations, on the one hand, or high-performing people and people with high aspiration, on the other hand. Possibly, the perceived divergence between one's own performance and that of the role model partly explains the differential effects found in different studies (Raghunathan et al., 2018). More in general the effectiveness of role models depends on the social distance one feels towards the role model (proximity); the divergence of one's own achievements from those of the group or person that serve as the role model; and the extent to which that role model's (group) situation is similar and recognizable (Raghunathan et al., 2018). Even where aspirations are improved, other constraints may prevail and prevent the translation of higher aspirations into achievement gains (Beaman et al., 2012).

Testable hypotheses
Based on the literature, we can formulate several hypotheses about (a) the effect of involving women in receiving information, and (b) the effect of involving women in conveying information on various measures of women's empowerment. We consider women's empowerment along the lines of Johnson et al. (2016Johnson et al. ( , 2017, in which empowerment encompasses both an increase in women's individual and joint decisionmaking within the household with respect to the set of key livelihood-related variables. We consider empowerment as a positive change both for a woman and for the female and male co-heads jointly in (a) their knowledge about recommended agronomic practices and inputs, (b) their participation in intrahousehold decision-making on farm production, (c) the choices they make about the adoption of recommended practices and the use of inputs, (d) the output and yield from plots under either women's or joint management, and (e) their participation in selling farm output. A reduction in male domination over (a), (b), (c) and (e) is considered empowering as well. In the context of our study, these variables are specific to the types of smallholders defined by our population of interest (and described in detail below), but are generalizable to other smallholder agricultural contexts.
From the literature on intra-household information asymmetries, we can hypothesize that involving women in receiving information may strengthen women's knowledge about recommended practices and inputs, and possibly make common knowledge about other agronomic practices more salient. Here, the effects on decision-making and outcomes depend on several additional channels. The first is the effect of involving women in receiving information that would otherwise be monopolized by the male co-head if only he received that information. If the female co-head receives the information alone and does not monopolize that information or if both the female and male co-head of a household receive the information as a couple, we would expect an increase in both women's individual knowledge and the couple's overall (joint) knowledge where intra-household communication, consultation, and learning occur, which can lead to an increase in women's individual and joint decision-making and outcomes. If the female co-head receives the information alone, she might also monopolize that information, in which case changes in women's individual knowledge, decision-making and outcomes may be expected. The second channel is through a change in a woman's bargaining position that results from possession of an informational asset, which may prevent a male co-head from advancing his preferences at the expense of hers. These channels may be constrained by a woman's limited access to the markets-both material and financial-and if the woman is a late adopter, which may limit the realization of benefits. But notwithstanding these constraints, we expect that involving women in receiving information will positively influence women's participation, both alone or jointly with their spouses, in knowledge and in decision-making on crop production, the adoption and use of recommended crop management practices and inputs, and the production and the sale of farm output to the market.
Based on the literature on role-model effects, we also hypothesize that women's empowerment is positively affected by directly involving women in conveying information. In such instances, involving women in conveying information may affect recipients of the information via role-model effects that work via three inter-related channels: learning externalities through peer effects, gender homophily effects, and challenges to role incongruity. We hypothesize that peer effects occur where information recipients can relate to the information messenger as a peer-as an individual with a similar identity or experience, or as an individual possessing credible knowledge and information. Gender homophily effects occur where these peer effects are specifically based on a match in the sex of the messenger and recipient. Relatedly, we hypothesize that role incongruity effects occur where women update their beliefs about roles and responsibilities in the household as a result of receiving information from a messenger that is somehow inconsistent with their beliefs. Taken together or separately, we expect such effects to increase women's aspirations about their households and livelihoods, thereby causing them to make more ambitious choices. In our context, this is represented as involving a female actor in the video as a messenger of information, who may also function as a role model. Additionally, we might expect the involvement of women in conveying information to cause men to revise their beliefs about women's agency and capacity to make sound investment decisions on the farm. Possibly, this may also inspire men to share or leave more space for decision-making by their wife and to relax any inhibition to their wife's access to complementary inputs, their decision-making on the adoption of recommended practices, or the improvement of crop production. In sum, we can expect positive changes in women's participation in knowledge and decision-making about farm production, the adoption of recommended practices and inputs, the production and sale of farm output, both alone or jointly with their spouses, as a result of role-model effects following from involving women in conveying information through learning externalities, gender homophily, or role incongruity challenges.

Study context
To explore these effects, we conducted a field experiment in 2017 among smallholder maize-farming households in eastern Uganda. Participants in the field experiment were drawn from monogamous maizecultivating households residing in five districts where maize is particularly important, both as a staple and as a marketable crop (see Van Campenhout et al. (2018) for additional details). The experiment was conducted during the second maize-growing season which runs from approximately August to January of the following year, and is characterized by a shorter period of rainfall than the first maize-growing season. During this second season, the complete cycle from planting to harvest requires three to three and a half months, and farmers tend to cultivate early-maturing but lower yielding maize varieties. Fields are prepared in August, planted in September, and harvested from December onward.
Maize yields in the study area, and in Uganda more generally, are well below their potential. Research station trials in Uganda have demonstrated that yields range between 730 kg per acre and 1,820 kg per acre (Fermont & Benson, 2011). Yet, on-farm estimates are generally lower. A recent study of on-farm yield reports figures between 270 kg per acre and 995 kg per acre (Gourlay et al, 2019). There is also evidence that maize plots under female management are less productive than maize plots under male management. Sell et al. (2018) calculated that maize productivity on male-managed plots is produced at 57 percent of mean technical efficiency, while production on joint-and female-managed plots is 56 and 43 percent, respectively. They report that agrochemicals are used on only 10 percent of all female-managed maize plots compared to 28 percent of male-managed maize plots. Peterman et al. (2011), Larson et al. (2015, and Ali et al. (2016) observe similar differences in productivity between male-and female-managed plots in Uganda. In particular, Ali et al. (2016) link the productivity gap between male-and female-managed plots in Uganda (controlled for plot size but without distinguishing between type of crops) to an unequal distribution between men and women of responsibilities and resources, including modern inputs such as improved varieties, fertilizer, and agrochemicals.
Decision-making power over agricultural production is related to the actual or perceived ownership of the land. In Uganda, ownership of land for married women is constrained by patrilineal land inheritance customs and by customs and traditional rules prescribing that, through marriage, women gain access to land owned by their husband. Fisher and Carr (2015) show that in eastern Uganda, married women farmers are the primary owners of only 2.7 percent of all maize plots in the study's sample, and are the main decision-maker for the choice of maize variety on only 15.4 percent of the plots. They did not find evidence that land quality (in terms of soil fertility and slope) or type of tenure differ between female-and male-managed maize plots.
Decision-making power is also related to customary gender roles and responsibilities which, in turn, affect intra-household labor allocations. Based on data from the 2011/12 Uganda Living Standards Measurement Study (LSMS), Fisher and Carr (2015) found that, in total, more household labor was used on male-managed maize plots than on female-managed maize plots in male-headed households. On those female-managed plots, more female and child labor (and less male labor) was allocated to maize cultivation. In Uganda, generally, men tend to concentrate on marketable crops that generate cash income, while women tend to focus on other food crops for household consumption.

The experiment: A video extension information intervention in a factorial design
In this study, we test hypotheses about the (relative) effectiveness of involving women in conveying and receiving information on a range of outcomes related to knowledge accumulation, intra-household decisionmaking, adoption of maize cultivation practices, maize outputs and yields, and maize marketing, both for women individually and jointly with their male co-heads of households. To do so, we developed a 3x3 factorial design, in which one factor corresponds to the gender of the person (or persons) who receives the information and the other factor corresponds to the gender of the person (or persons) who delivers the information. 2 Each factor contained three levels: man alone, woman alone, or man and woman together (as a couple). The design is represented in Table 1 below, with sample sizes shown in each of the nine treatment combinations to indicate the number of households randomly allocated to one of these nine treatments. 3 2 In the context of this study, we refer to these factors in terms of "gender" and not "sex" because the implicit differences in the person(s) receiving or delivering the information are social and cultural in nature, and not simply biological.
3 Power calculations were based on a set of comparisons using different outcomes to power the complete 3x3 factorial design. We used simulation techniques that allowed us to sample from actual data on outcome variables (maize yields obtained from Uganda National Household Survey of 2005/06) instead of from a theoretical distribution with an assumed mean and standard deviation. Apart from the sample size in the control group, sample size in other treatment arms are the result of binding constraints for minimal sample size needed to test differences in two other factors of the design. Detailed information on the power calculations can be found in the pre-analysis in Van Campenhout et al.  To operationalize this design, we developed a series of videos that was shown to participating farmers according to this design. Corresponding to the recipient factor, the video was shown to one of the three recipients: the male co-head of the household; the female co-head; or the male and female co-heads together as a couple. To operationalize the messenger factor, we produced three versions of essentially the same video, with the only difference being the actor(s) featured in the video. In a first version of the video, a male actor-farmer is featured in the video. The second version of the video features a female actor-farmer.
In a third version of the video, both the male and female actor-farmers are featured. The videos can be found here.
During implementation of the experiment, one of these three versions of the video was screened to the participating individual(s) according to the treatment arm that the household was randomly assigned to. The videos were shown on 10-inch Android tablet computers by trained field enumerators during a private meeting with the participant (or participants if the recipient was the couple). Information on the sampled households and their treatment assignment was pre-loaded onto the tablet computers, such that the correct video was automatically queued and enumerators were only able to screen the appropriate version of the video. The video was shown twice to our study participants, once before the maize planting time ( The factorial design laid out in Table 1 allows us to test the different hypotheses described earlier by comparing outcomes across different treatment groups, which is visualized in Table 2. To examine the impact of providing women with direct access to extension information on different measures of women's empowerment-our intra-household information asymmetry hypothesis-we compare outcomes of the 2,192 households where the female co-head was involved in watching the video (either alone or as part of a couple and regardless of which version of the video that was shown) to outcomes of the 1,139 households where only the male co-head watched the video (regardless of which version of the video was shown) (  Next, to examine the impact of providing women and men with information provided by women as messengers-our hypothesis on role-model effects through peer, gender homophily and role incongruity challenges effects-we compare outcomes of the 2,219 households where a female actor featured in the video (either alone or as part of a couple and regardless of whom the video was shown to within the household) to outcomes of the 1,112 households where only a male actor featured in the video (regardless of whom the video was shown to within the household) ( Table 2 Model 2). To specifically assess (a) the impact of women receiving information conveyed by women-a specific test of our hypothesis of peer and gender homophily effects and challenges to role incongruity among women-we compare outcomes of the 754 households where the video in which a female actor was featured (either alone or as part of a couple) was shown to the female co-head within the household to outcomes of the 385 households where the video in which a male actor featured was shown to the female co-head (Table 2 Model 2A). We additionally zoom in on (b) effects following from men updating role incongruent beliefs when a woman is providing information and portrayed as a successful farmer by comparing outcomes of the 754 households where the male co-head was shown the video in which a female actor was featured (either alone or as part of a couple) to the 385 households where the male co-head was shown the video featuring a male actor (Table 2 Model 2B). 4

Data
Data were collected from households in five districts in eastern Uganda: Bugiri, Mayuge, Iganga, Namayingo, and Namutumba. From among these districts, we first removed town councils and two sub-counties that consisted of islands in Lake Victoria. We then used a two-stage cluster sampling approach to obtain a representative sample of this population. Specifically, we first selected parishes randomly and in proportion to the number of villages within each parish. In the selected parishes, all villages were included in the study.
Within each village, we then listed all households, from which we randomly selected households to be included in the study. Because we assume that outcomes within villages will be correlated (e.g., due to local weather conditions, or development programs that are implemented in certain areas), we used the village as a blocking factor such that in each village, we ensured that all possible treatment combinations related to the two factors were administered. Because the experimental design described above lays out nine different treatment combinations, a maximum of nine different households were selected in each village. Table 3 summarizes pre-treatment characteristics of the households that were enrolled in the study. The first column provides averages for the entire sample, with corresponding standard deviations in parentheses below. The 5 percent trimmed log maize yield (defined as kilograms produced per acre) was 5.3 on average.
The average household head is about 40 years old, and in only 37 percent of households did the head finish primary school. Households are large, consisting of more than 7.6 members. Only 11 percent of sampled households reported having access to extension in the previous year. Few farmers use inorganic fertilizer at baseline; the use of improved seed is higher. About 78 percent of households report to be owning a mobile phone.
The table also serves as a balance test for the different comparisons we will make. We observe baseline imbalances in observable characteristics that are statistically significant, and we further test for joint significance of the differences and find that certain groups are significantly different in the characteristics we measured at baseline. We will control for baseline characteristics that show up as significant in the balance tables by estimating analysis of covariance (ANCOVA) models when treatment and control outcomes are compared throughout the paper.

Indicator definition
Since this study aims to estimate effects on women's empowerment, we consider a woman as 'empowered' not only if she achieves things on her own, but also if she does so in partnership with the man co-head in the household. Thus, we examine effects on both women's outcomes and outcomes that are shared (joint) between the female and male co-heads of household. Those outcomes are cast as increases in (a) the female co-head's knowledge (individually and/or jointly with her male co-head) about the information promoted in the video; (b) the extent to which maize production decisions in the household are taken by the female cohead alone and/or jointly with her male co-head; (c) the adoption of the recommended practices and inputs decided upon by the female co-head individually and/or jointly with her male co-head; (d) output and yield on maize plots that were female-managed and/or jointly managed; and (e) maize sales decided upon by the female co-head individually or jointly with her male co-head. Additionally, we will examine effects on men's outcomes assuming that a reduction of men's individual decision-making and adoption can be empowering for women as well.
Knowledge about four different practices recommended in the video is measured by the extent to which respondents answered correctly to multiple choice questions about the practices. 5 A woman's (man's) knowledge score is based on responses from the female (male) co-head; the joint knowledge score is based on responses from both the female and male co-heads, where it was considered a correct joint answer if both of them got the answer correct, otherwise, not. We combine the outcomes of the four knowledge questions into a Women's Knowledge Index, a Joint Knowledge Index and a Men's Knowledge Index, which are constructed as the weighted mean of the individual standardized outcomes, using as weights the inverse of the co-variance matrix of the transformed outcomes (Anderson 2008). This, in addition to adjustments to the significance levels to account for Family Wise Error Rates (FWER), guards against over-rejection of the null hypothesis due to multiple inference. We used re-randomization to construct the joint null distribution for the family of outcomes we are testing. From this family-wise sharp null, we obtained the corresponding FWER-consistent significance thresholds by determining which cutoffs yield 10 percent, five percent and one percent significant hypothesis tests across all tests and simulations.
Agricultural decision-making is based on women's answers about whether she made particular decisions related to household maize production individually or jointly with her spouse, or whether her husband made the decisions by himself. We look at the following decisions: whether to plant maize on a particular plot; the timing of maize planting (the number of days after the start of the rain) on a particular plot; the spacing of maize plants and the number of seeds to plant per hill; how to control striga on a particular plot; and the timing of the first weeding on a particular plot. Per household, we consider the proportion of maize plots within the household on which the female co-head reported to have made the decision alone (jointly, or her husband alone). Following a similar procedure as discussed above, we constructed a Women's Decisionmaking Index, based on the proportion of the household's maize plots for which the female co-head made the decisions described above alone, along with a corresponding Joint Decision-making Index and Men's Decision-making Index. We also calculated FWER-consistent significance thresholds for the family of decision-making outcomes in the same manner as discussed above.
Next, we consider actual adoption of the practices recommended in the video upon which, according to the female co-head respondent, she individually made decisions or jointly with her male co-head, or her husband did. We specifically consider the proportion of the household's maize plots for which the female cohead decided about a particular practice alone and adopted the recommended practice by herself (respectively the proportion of the household's maize plots for which the female and male co-head decided and adopted a particular practice jointly, and for which the male co-head decided and adopted the practice alone). 6 We measure adoption of the following practices: planting within one day after the start of the rain, using the correct spacing and number of seeds per hill, removing striga before it flowers, and doing the first weeding in the third week after planting.
We measure use of inputs such as DAP (Diammonium phosphate), urea, organic fertilizer, maize hybrids, and open pollinated varieties (OPVs). We specifically consider the proportion of the household's maize plots for which a particular input was used and this was decided individually by the female co-head (jointly/individually by the male co-head) (based on the female co-head's answers). 7 As before, we used the same method to construct a Women's Adoption Index, a Joint Adoption Index, a Men's Adoption index, a 6 It is important to note that our intervention may change both the likelihood that a spouse makes a decision as well as the likelihood that the decision maker then adopts what was recommended in the video. In other words, we only observe the reduced form outcome of our intervention: P(Adoption by woman * Decision by woman) with P(Adoption by woman * Decision by woman) = P(Adoption by woman | Decision by woman) * P(Decision by woman). P(Adoption by woman * Decision by woman) is what estimated in Table 6 and P(Decision by woman) is estimated in Table 5.
For reference purposes, women's likelihood of adoption on condition of making the decision by herself = P(Adoption by woman | Decision by woman) = P(Adoption by woman * Decision by woman) / P(Decision by woman).
Women's Input Use Index, a Joint Input Use Index, and a Men's Input Use Index and defined FWERconsistent significance thresholds.
Next, we measured outcomes related to production, area of production, and productivity on maize plots under female and joint management. Female (jointly) managed plots are defined as plots on which, according to female co-head respondents, female co-heads alone (jointly) took at least three out of five decisions about planting maize on the plot, the timing of planting, spacing and seed rate, striga control, and the timing of the first weeding. We use the total amount of maize produced on female-managed maize plots within the household as our measure of production. The area of production is the total area (in acres) of female-managed maize plots in the household. Yield (in kg per acre) is the total amount of maize produced on female-managed plots divided by the total area of the female-managed maize plots in the household. A secondary, more subjective indicator for yield takes the value of one if the female co-head respondent believes the yield on at least one of her maize plots was greater than in a normal year. To capture the change in women's control over total household production, we measure both the share of household maize production that was produced on female-managed maize plots and the share of household acreage under maize cultivation that is on female-managed maize plots. We also defined indicators for jointly managed plots, and calculated FWER-consistent significance thresholds.
Finally, we looked at maize sales as an outcome. A first indicator takes the value of one if any amount of the maize produced in the household was reported by the female co-head respondent to have been sold independently by the female co-head (respectively, jointly and independently by the male co-head). 8 A second indicator is the amount of maize (in bags of about 100 kg) that the woman respondent reported to have sold independently (jointly/independently by the male co-head). 9

Knowledge of recommended practices
We begin our discussion of results by initially examining women's and joint (couple's) knowledge outcomes, focusing on the knowledge index described earlier as well as the individual recommended practices that respondents answered questions about.
First, we examine whether involving women as recipients of information leads to an increase in their knowledge. We do this by comparing: (a) groups in which female co-heads received the information, either 8 Selling independently by the female (male) co-head means all decisions regarding quantity, price and to whom the maize was sold were taken by the female (male) co-head according to the female co-head respondent. Jointly means all those decisions were taken together by the female and male co-head together. 9 Note that for the first indicator of the likelihood of selling independently we assumed women did not sell any maize independently if they did not know the amount sold independently. For the second indicator (amount of maize sold independently), we only considered positive amounts of maize sold independently. alone or with their male spouse, against (b) groups in which only the male co-head received the information (Table 2, Model 1). Results (Table 4, Model 1, Panel a) show a positive and statistically significant treatment effect on the women's knowledge index of 0.11. We further observe positive and significant treatment effects of involving women as recipients of information on women's knowledge about adequate spacing and the number of maize seeds per hill of 7 percentage points (pp). This translates into a 52 percent increase over the mean of the control group. We observe a similar 5 pp increase in women's knowledge about combining practices and starting small. These results remain consistent when the treatment group is restricted to: (a) the group in which the female co-head received the information alone (  Second, we examine joint knowledge outcomes, and present results in terms of a comparison between the same sets of groups (Table 2 Model 1). Results (Table 4, Model 1, Panel b) show no significant effects of the treatment on joint knowledge outcomes. There is an indication that new information (about spacing) received by female co-heads alone is somewhat less likely to be passed on to their spouses than when male co-heads received the information, which is evident from the negative treatment effect of -3 pp on joint knowledge (significant at 11 percent) (Table 4, Model 1A, Panel b). Men's knowledge index, and particularly his knowledge of new information (about spacing and starting small with a combination of practices), is negatively affected by involving women in receiving information (Appendix Table 4B, Model 1).
Third, we examine the knowledge outcomes that result from variation in the provider of information. We examine results for role-model effects by examining women's knowledge outcomes in terms of a comparison between: (a) groups in which a female actor in the video provides information (whether alone or with a male co-actor) against (b) groups in which only a male actor provides information (

Decision-making about maize production
Next, we explore the effect of the treatments on decision-making in the household. We do this by first examining whether involving women as information recipients (alone or as part of a couple) results in their participation in decisions about maize production on an average plot managed by the household. We do this by again making the comparison as described in the previous section and illustrated in Table 2, Model 1.
Results (Table 5, (Table 5, Model 1A, Panel a). No such increase is found when the treatment group is restricted to the group in which female co-heads received the information together with their male spouse (Table 2,   Model 1B; Appendix Table 5A, Model 1B, Panel a). This implies that the gains in women's individual decisionmaking are largely driven by providing only the female co-heads with information. We find no change in joint decision-making outcomes as a result of involving women as information recipients (alone or as part of a couple) based on a comparison between the same groups ( Table 2, Model 1;   Table 5, Model 1, Panel b). But, when the treatment group is restricted to the group in which female coheads received the information together with their male spouse (as in Table 2, Model 1B), there is a significant and positive effect on joint decision-making about the novel way of spacing and number of seeds (Appendix Table 5A, Model 1B, Panel b). We do find significant negative effects of involving women in receiving information on men's individual decision-making. The men's decision-making index decreases by 0.18 and men are less likely to make each of the decisions alone (Appendix Table 5B, Model 1). These negative effects on men's individual decision-making do not only follow from giving the female co-head the information alone (in which case the female co-head can monopolize the information) ( Appendix Table 5B, Model 1A), but also from giving the female and male co-head the information together ( Next, we explore decision-making outcomes that result from variations in the information provider by making the same comparison as described in the previous section and illustrated in Table 2, Model 2. Results (Table 5, Model 2, Panel a) show no evidence of a generalized role-model effect on women's decisionmaking. However, when we restrict this comparison to groups in which a female actor in the video provides information (whether alone or with a male co-actor) to female co-heads against groups in which a male actor provides information to female co-heads (Table 2, Model 2A), we do observe positive and statistically significant effects associated with role-model effects (Appendix Table 5A, Model 2A, Panel a). Specifically, we observe an increase of 0.16 in the women's decision-making index and increases in women's decisionmaking about the timing for planting maize, about spacing, about striga control measures, and the timing of weeding that amount to 7, 9 and 7 pp respectively, which translates into increases of 26, 34 and 39 percent, respectively, over the mean of the control group. Challenging role incongruity among male co-heads by showing them the video that involves a female actor (whether alone or with a male co-actor) instead of a male actor (as in Table 2, Model 2B) does not make a difference for women's decision-making (Appendix Table 5B, Model 2A, Panel a).
While joint decision-making does not change as a result of variations in the information provider (Table 5, Model 2, Panel b), men's decision-making does. Men's unilateral decision-making about whether to plant maize on the plot is reduced by 4 pp (an 11 percent reduction over the mean of the control) when a female actor is involved in providing the information (alone or together with a male actor) as compared to only a male actor, regardless of whom the video was shown to (Appendix Table 5B, Model 2). The fact that this reduction in men's decision-making about whether to plant maize is mirrored when male co-heads viewed the video that involved a female actor instead of the video with only a male actor (Table 2, Model 2B) demonstrates that role incongruity challenges may be at work among male co-heads (Appendix Table 5B, Model 2B).

Adoption of recommended practices
We apply the same approach to our presentation of adoption outcomes, tested by making the comparison illustrated in Table 2, Model 1. Results (Table 6, Model 1, Panel a) shows a positive and statistically significant treatment effect of involving women in receiving information, denoted by an increase of 0.12 in the women's adoption index. The likelihood of adoption of each of the practices individually decided upon by the female co-head increases between 1 and 5 pp. Given that adoption individually decided upon by the female co-head is generally low in the control group, these changes translate into sizable percent increases in women's decided-upon adoption. Effects remain generally consistent when the treatment group is restricted to the group in which the female co-head received the information alone (as in Table 2, Model 1A), in which case the women's adoption index increases by 0.15 (Table 6, Model 1A, Panel a). When the treatment group is restricted to the group in which the female co-head receives the information together with her male spouse (as in Table 2, Model 1B), the women's adoption index increases by 0.09 and only the adoption of striga control measures decided upon by the female co-head is positively affected (Appendix Table 6A, Model 1B, Panel a).
Second, we examine joint adoption outcomes, and present results in terms of a comparison between the same sets of groups as in Table 2 Model 1. Involving women in receiving information has a positive significant effect of 2 pp on the adoption of the novel practice of spacing and number of seeds per hill jointly decided upon by the female and male co-head (Table 6, Model 1, Panel b). Joint adoption outcomes follow from providing the information to the female and male co-head together in contrast to only providing the male co-head with information, as illustrated in Table 2, Model 1B. Results (Appendix Table 6A, Model 1B, Panel b) show an increase in the joint adoption index of 0.08, a 2 pp increase in the jointly decided-upon adoption of proper spacing, and a 5 pp increase in the jointly decided-upon adoption of striga control measures.
Next, we examine the effect on men's adoption outcomes by the making the same comparisons between groups. Involving women in receiving information, alone or as part of the couple, as compared to giving the information only to the male co-head (as in Table 2, Model 1), reduces men's individually decided-upon adoption of striga control measures by 3 pp (Appendix Table 6B, Model 1). Providing only the female cohead with information as compared to only the male co-head (as in Table 2, Model 1A) reduces men's individually decided-upon adoption of striga control measures by 4 pp and timely weeding practices by 6 pp, which come down to about a 30 percent reduction (Appendix Table 6B, Model 1A).
We proceed by assessing the effects of variation in the information provider in the same manner as before (as in Table 2, Model 2). Results (Table 6, Model 2, Panel a) do not indicate any difference in women's adoption outcomes as a result of involving women in conveying information. But when we make the comparison between the groups in which the female co-heads were shown the video with a female actor (alone or with a male co-actor) instead of the video with only a male actor (as in Table 2 Involving women in conveying information does not make any difference in joint adoption outcomes (Table   6, Model 2, Panel b). There are indications, however, that men's adoption outcomes are negatively affected by showing male co-heads the video with a female actor (whether alone or with a male co-actor) rather than with only a male actor (as illustrated in Table 2, Model 2B). Results (Appendix Table 6B, Model 2B) show a decrease of 3 pp (a 38 percent decrease relative to the mean of the control) in men's individually decidedupon adoption of the appropriate timing for planting maize.  Electronic copy available at: https://ssrn.com/abstract=3505653 Electronic copy available at: https://ssrn.com/abstract=3505653

The use of fertilizer and improved seeds
In this section, we explore the impact of the different treatments on the use of specific types of fertilizer (DAP, urea, and organic fertilizer) and improved seeds (hybrids or OPVs) by women who decided individually upon using the specified inputs, by female and male co-heads who jointly decided upon using those inputs, and by men who individually decided upon using those inputs.
First, we examine the effect of involving women in receiving information, alone or as part of the couple.
Results (Table 7, Model 1, Panel a) show a positive and significant increase of women's input use index by 0.08. The effect of providing only female co-heads with information as opposed to only male co-heads (as in Table 2, Model 1A) is an increase of the women's input use index by 0.13 and is based on an increase of the use of each of the listed inputs by women who decided individually upon their use (Table 7, Model 1A, Panel a). In contrast, providing the female co-heads together with their male co-heads with information as opposed to only the male co-heads (as in Table 2, Model 1B) has no effect on women's input use index (Appendix Table 7A, Model 1B, Panel a).
Next, we examine the effects of involving women in receiving information, alone or as part of the couple, on joint input-use outcomes. Results (Table 7, Model 1, Panel b) show a positive and significant effect on the joint input use index, which is increased by 0.06. Providing only female co-heads with information as opposed to only male co-heads (Table 2 Model 1A) increases the joint input use index by 0.04 (Table 7, Model 1A, Panel b). Providing female co-heads and their male co-heads with information as opposed to only the male co-heads (as in Table 2, Model 1B) increases the joint input use index by 0.09 (Appendix Table 7A, The effect of involving women in receiving information, alone or as part of the couple, on input use decided upon by men is mixed: it is negative for the use of the fertilizer DAP and positive for organic fertilizer ( Table   2, Model 1; Appendix Table 7B, Model 1). The negative effect on men's decided-upon use of DAP emerges when the female co-heads receive the information alone (Table 2, Model 1A; Appendix Table 7B, Model 1A), which may be linked to men lacking information about the use of DAP.
We continue by examining role-model effects with the comparison illustrated in Table 2   Involving women in conveying the information has no significant effects on joint input use (Table 7, Model 2, Panel b). Yet there are indications that jointly decided-upon use of urea increased by 2 pp as a result of female co-heads having seen the video with a female actor (alone or with a male co-actor) instead of the video with only a male actor (as in Table 2, Model 2A) (Appendix Table 7A, Model 2A, Panel b). The use of DAP individually decided upon by men, however, seems negatively affected by that treatment (Appendix ,   Table 7B, Model 2A). There are indications that the use of both urea and hybrid seed individually decided upon by men are positively affected as a result of showing male co-heads the video with a female actor (alone or with a male co-actor) instead of the video with only a male actor (Appendix Table 7B, Model 2B).

Production and productivity
Next, we assess effects on production and productivity of female-managed maize plots as a result of involving women in receiving information versus only targeting the man (as illustrated in Table 2, Model 1).
The results (Table 8, Model 1, Panel a) show a significant and positive increase of the total area of femalemanaged maize plots in the household by 0.07 acres, an increase of the total amount of household maize produced on female-managed plots by 35.8 kg, and an increase of the yield on female-managed maize plots by 50.4 kg/acre; and a 2 pp higher likelihood of a better-than-normal yield. Both the share of household maize production and maize area under female management increased by 6 pp. The positive effects on production and productivity outcomes on female-managed plots that result from providing only the female co-head with information instead of the male co-head (as in Table 2, Model 1A) are even larger in magnitude (Table 8, Model 1A, Panel a). In contrast, involving the female co-head as part of a couple receiving the information instead of only the male co-head (as in Table 2, Model 1B) shows no effect on production and productivity outcomes on female-managed plots (Appendix Table 8A, Model 1B, Panel a).
Involving women in receiving information, however, does not have any effect on production and productivity outcomes on jointly managed plots (Table 8 ,Model 1, Panel b). Even when the female and male co-heads received the information together (as in Table 2, Model 1B), there is no impact on production and productivity outcomes on jointly managed plots (Appendix , Table 8A, Model 1B, Panel b). Electronic copy available at: https://ssrn.com/abstract=3505653 Electronic copy available at: https://ssrn.com/abstract=3505653 Next, we examine the effect of involving women in conveying the information rather than only a man (as in Table 2, Model 2) on production and productivity outcomes of female-managed plots. Results (Table 8, Model 2, Panel a) point to a significant and negative effect on the yield on female-managed plots. Suggestive indications of the negative effects on yield on female-managed plots emerge both when female co-heads and when male co-heads have seen the video featuring a female actor (alone or with a male co-actor) instead of the video featuring only a male actor (respectively Model 2A and Model 2B in Table 2). In the latter case, there are also indications of a negative effect on production on female-managed plots (Appendix Table 8A There is no evidence that production and productivity outcomes of jointly managed plots changed as a result of involving women in conveying the information (alone or with a male co-actor) rather than only a man (as in  Table 2), the likelihood that the yield on jointly managed plots is better than normal is (indicatively) increased by about 4 pp. This translates into about a 40 percent increase over the mean of the control (Appendix Table 8A

Sales
Maize produced in the household independently sold by the female co-head is not affected by involving women in receiving information, alone or as part of the couple (Table 2 Model 1; Table 9, Model 1, Panel a).
Yet providing the information to only the female co-head as opposed to providing it to only the male cohead (as in Table 2, Model 1A) increases the likelihood that she independently sells maize by 4 pp (Table 9, Model 1A, Panel a). As few women in the control group actually sell maize independently, this translates into a substantial percentage increase of 46 percent over the mean of the control. There are indications of a positive effect on the amount of maize sold independently by women as well. The selling of maize by the female and male co-head jointly does not change by involving women in receiving information, alone or as part of the couple (Table 2 Model 1; Table 9, Model 1, Panel b). Neither does the selling of maize individually by the male co-head (Appendix Table 9B, Model 1).
Next, we look into the effect of involving women in conveying information (by making the comparison visualized in Table 2, Model 2). Results (Table 9, Model 2, Panel a) show a significant and negative effect on the amount of maize that women independently sell. There are indications of negative effects on the likelihood and the amount sold independently by women when male co-heads view the video featuring a female actor (alone or with a male co-actor) delivering the information instead of only a male actor (as in  Involving women in conveying information has a positive effect of 4 pp (though just under the FWERconsistent significance threshold) on the likelihood that maize produced in the household is jointly sold. This translates into a 12 percent increase over the mean of the control (Table 9,  The amount of maize independently sold by men, however, is significantly and negatively affected by involving women in conveying information (Appendix Table 9B, Model 2). This seems particularly the case when female co-heads alone viewed the video in which women are involved in conveying information (

Discussion
The results described above add nuance to prior studies that examine the role of gender in agricultural extension and technology adoption. To summarize these nuances, we provide an overview of the effects of different treatments on outcomes for women and for women jointly with their spouses in Table 10. As the withdrawal of men from individual decision-making may be empowering for women as well, we also included treatment effects on men's outcomes where relevant.

Women's access to information and extension services
Involving women in receiving extension information instead of giving it to only the male co-head has positive effects on women's individual knowledge. As gains are made by giving women access to information directly, we can infer that male co-heads do not necessarily pass the information to their female co-heads. Women's knowledge gains do not differ when a woman saw the extension video alone or when she saw it together with her male co-head, hence the presence of the male spouse in the exposure process apparently does not influence her knowledge.
Joint knowledge, however, does not benefit from providing women with direct access to information, not even when the two co-heads in the household receive it. Thus, we offer no evidence of a change in intrahousehold information flows between spouses as a result of the treatment. The fact that men's knowledge, particularly knowledge about new information, is negatively affected by giving the information to the female co-head alone, corroborates the finding that female co-heads do not necessarily share information with their spouse either. Somewhat surprisingly, men's knowledge also seemingly decreases when he receives the information together with his female co-head versus when he receives it alone. Model 1 -Involving the female co-head in receiving information (alone or as a couple) versus only the male co-head receiving the information Electronic copy available at: https://ssrn.com/abstract=3505653 Electronic copy available at: https://ssrn.com/abstract=3505653 Women gain individual agricultural decision-making when they are involved in receiving extension information, particularly when they receive it alone. Joint decision-making, however, remains unchanged.
These results suggest that women's agency in the domain of agriculture gains footing vis-à-vis men's by the reduction of information asymmetries between the two co-heads in the household. This may follow directly from her access to the required information for making a decision, or, indirectly, from a gain in bargaining power based on her-albeit unique-access to information. Conversely, men's individual decision-making is reduced as a result of involving women in receiving information, no matter if this happens by reaching the female co-head alone or together with the male co-head. If this reduction only occurred in the former case, men's reduced decision-making could have been related to a lack of information. The fact that the reduction occurs in both cases suggest that men's individual decision-making does not only decrease by not having direct access to information but also because of women's increased access to information.
Beyond women's decision-making, the adoption of recommended practices and use of inputs individually decided upon by women is also positively affected by providing women with information. Hence, women seem to use their increased agency as a result of reduced intra-household information asymmetries to become more involved in farm-related activities. Additionally, the adoption of recommended practices and use of inputs jointly decided upon by the female and male co-head increases as a result of involving women in receiving information. This implies that informed women can also participate more in joint action. The positive effect on jointly decided adoption of recommended practices and inputs when the female and male co-heads receive the information together is consistent with Lambrecht et al. (2016) who found a positive impact of joint participation in an extension program on fertilizer adoption on jointly (and men-) managed plots. In parallel with decision-making, the adoption of recommended practices and use of inputs individually decided upon by men is negatively affected by providing women with information.
As a result of providing only women with information, the total area of maize plots in the household managed by women increases and maize production on female-managed plots more than doubles. While increased women's decision-making about and adoption of each of the five recommended practices does not mean these all occurred on maize plots under female management (as the term is defined here), there is, however, an obvious positive correlation. We can thus infer that the increased decision-making about and adoption of recommended practices and inputs by women as a result of solely involving women as information recipients paid off in higher production, increased area, and higher productivity on femalemanaged maize plots. In contrast, providing the information to the female and male co-head together did not make a change for area, production and productivity of female-managed maize plots. This probably relates to the fact that providing the information to the female and male co-head together does not have the same strong positive effects on women's individual agency as providing the information to only the woman. Her lack of a unique possession of an informational asset seems to be in the way of capitalizing on her gain of information. There is no evidence of an increase of maize production, productivity, or area under cultivation under joint management as result of involving women in receiving information either, not even when the female and male co-heads were reached together. While we can also assume a positive correlation between joint decision-making and adoption and the likelihood that it concerns jointly managed plots (as defined here), we cannot infer that increased joint decision-making and adoption as a result of the female and male co-head receiving the information together paid off in higher production and productivity on jointly managed maize plots.
Women are also more likely to sell (more) maize independently if they alone received the extension information instead of the male co-head. At the same time, sales of maize individually by men is less likely, which indicates that women's increased agency as a result of unique access to information reduces men's dominance over maize sales. Giving the information to the female and male co-head together does not affect women's (nor men's) individual sales of maize. Joint sales of maize remain unaffected by involving women receiving information, even as part of a couple.
The fact that achievements in terms of production and sales by women individually when the information is given to the couple are absent whereas women can individually achieve gains in production and sales if they receive the information alone suggests that a woman's monopoly over information is essential for her individual achievements. Without that monopoly she seemingly cannot attain individual achievements. The fact that giving the information to the couple instead of only the male co-head increased women's agency in terms of joint decision-making and adoption but not joint achievement implies that, despite women's greater (joint) agency, these joint achievement are not different from what men individually achieve. This may be explained by the fact that the male co-head is involved in receiving the information in both cases, but it implies that women's greater involvement in joint decision-making and action did not lead to greater joint achievements.
A lingering question is the extent to which the increased efforts by women that result from exclusively giving them extension information, in terms of applying the recommended practices and inputs and larger areas of maize cultivation under their management, augment their work burden (see 5.3., 5.4., and 5.5). Indeed, we observe that the time women spent on preparing fields and weeding went up by 1.4 and 3.3 persondays/maize season, respectively, as a result of providing only the female co-head, instead of only the male co-head, with information. This increase in women's labor relates to the fact that women mainly apply laborbased intensification practices, which are likely more accessible to women than capital intensive practices such as the use of improved seeds or fertilizer. This suggests the need for further research on the labor and drudgery implications of these outcomes and the need to reflect upon making less labor-intensive intensification practices accessible to women.

Role-model effects
We expected that including women in the extension information videos portrayed as farmers, whom viewers can relate to and who become successful by applying the practices shown in the video, would have had rolemodel effects and improve women's outcomes. The evidence, however, is mixed. There is no impact on women's knowledge of agronomic practices, nor on women's individual decision-making and adoption of recommended practices, and indications of a negative effect on the yield on female-managed plots and sales of maize by women. Maybe not coincidentally, one of the limited positive role-model effects is on the use of organic fertilizer decided upon by women, which is a productivity-enhancing input that may be directly available to a woman and which she does not need to bargain for.
There are no effects of involving women in conveying information on joint outcomes, except for an indication that jointly selling maize has become more likely. The lack of effects on joint decision-making and adoption means that couples nor men were inspired to accept a greater participation of women in those decisions. There are, however, negative effects on men's individual decision-making, as well as on the amount of maize independently sold by men, which may make way for more involvement of women in decision-making and action.
Zooming in to role-model effects among female co-heads who viewed the extension information videos with women involved as messengers alone paints a slightly different picture. If women were the sole recipients of information provided by women, women's individual decision-making increased as well as women's individually decided-upon adoption of the appropriate timing for planting maize and use of organic fertilizer -again practices that are available to a woman at no or limited cost. Yet, the indication of a negative effect on yield on female-managed plots emerges in this case as well. There are some indications of positive effects on joint outcomes, such as use of fertilizer, improved yields on jointly managed plots and a higher likelihood of jointly selling maize. The amount of maize sold independently by men seems to decrease as a result. The effects on joint and men's outcomes of including women in the extension information videos shown to female co-heads as the sole recipients of the information seems to suggest that women who are inspired by the female role models not only gained individual agency but managed to make some changes at the couple level.
The effects of involving women in conveying information in the extension videos shown to male co-heads alone, which most likely mainly work through challenging role incongruity, seem to have reduced men's individual decision-making, particularly about whether to plant maize on a plot, men's individually decidedupon adoption, and improve yield of jointly managed plots. These results suggest that role-model effects and challenging role incongruity among male co-heads may have encouraged them to dominate agricultural decision-making less.
In the literature, the effects of role models on success were mixed (Beaman et al., 2012;Riley, 2017;Raghunathan et al., 2018). In this study they seem negative for women's achievements about maize productivity and sales. We believe there are three possible explanations for the decline in yield on femalemanaged plots as a result of role-model effects. First, the (indications of) increases in women's decided-upon use of organic fertilizer and timely planting of maize may have been insufficient to generate positive changes in production and productivity outcomes on woman management plots -and, as the sole productivityenhancing measures, may even have been counterproductive. Second, involving women in conveying information had negative effects on men individually deciding upon the planting of maize on the average household plot. Potentially, with men withdrawing from individually deciding to grow maize, women gained the opportunity to decide upon growing and focusing their efforts on other crops, which could explain a decline in women's maize production and sales. The effects of role models on success rather emerge in joint outcomes, in terms of (indications of) improved yields on jointly managed plots and a greater likelihood of joint maize sales, and particularly when the female co-head was the only one receiving the information by women role models. Third, the literature is inconclusive about whether to expect strongest role-model effects among highly performing or highly aspiring people or among lowly performing or aspiring people.
Involving women as role models in the extension videos may have had heterogeneous effects among women farmers with high and low aspirations, or among high-and low-performing women maize farmers. If such heterogeneous treatment effects exist and go in opposite directions, they could be a reason for the absence of significant effects and for seemingly overall negative effects.

Policy recommendations and conclusions
This paper examines the extent to which the reduction of women's extension information deficiency empowers women in agriculture and the extent to which information provided by women farmers as role models contributes to women's empowerment in agriculture. To examine the effect of those two factors, a field experiment with a factorial design was run in eastern Uganda with monogamous smallholder maize farmers. The intervention consisted of a short video shown twice in which farmer-actors explain and demonstrate various strategies and practices to intensify the production of maize. Three versions were randomized across farming households: one version portrays a man farmer, another a woman farmer, the third a couple formed by man and woman farmers. Three constellations of recipients of information in the household were randomized as well: the male co-head alone, the female co-head alone, the male and female co-heads together. As such, we can assess (a) the impact of involving women as extension information recipients, alone or as part of a couple, versus a man receiving information; and (b) the impact of involving women in conveying extension information, alone or as part of a couple, whereby these women can function as role models, versus a man giving information. We look at the impact on women's knowledge about agronomic practices and joint knowledge of the male and female co-heads, women's and joint decision-making about maize production, women's and jointly decided-upon adoption of practices and use of inputs, production and productivity on women-managed maize plots and jointly managed maize plots in the households, and sales of maize independently by women and jointly by the male and female co-heads as indicators of women's empowerment in agriculture. Additionally, we examine men's outcomes in terms of knowledge, decision-making, adoption and input use, and maize sales where a reduction could imply the potential for greater involvement of women.
The findings of this study are largely consistent with other studies on gender and agriculture (Doss & Morris, 2000;Kabunga et al., 2012;Fisher & Carr, 2015;Kondylis et al., 2016;Lambrecht et al., 2016). Yet, they give us pause for thought when we turn our attention to the role of ICTs in extension service provision. It is wellestablished that agricultural extension and rural advisory services tend to be male biased in both delivery and content, with predominantly male extension agents targeting male members of households with information that is often relevant to those crops, plots, or resources perceived to be under male control.
And while there may be tacit-or even explicit-recognition that women play a significant role in agricultural production, a key assumption within extension practice is that men will share information with women within their household. In reality, men may have incentives to keep (part of this) information private. The results of this study clearly show that significant advances in women's role in agriculture are made by giving women direct access to extension information. Furthermore, while the evidence is somewhat mixed, the results of this study suggest that women as role models, through peer effects, gender homophily effects, and/or challenges to role incongruity, influence women and men in different ways. Women are inspired to aspire when they receive information delivered by women but cannot turn that into individual success, in terms of yield and maize sales. The success emerges in joint outcomes. Men's individual decision-making and agency, however, seems to reduce when men receive information delivered by women, which may create space for women's participation in decision-making and action.
Three key policy recommendations follow from the results of this study. First, if the aim is to empower women by increasing their own agricultural knowledge, their independent decision-making, adoption of intensive production practices and, ultimately, to boost yields on plots that they autonomously manage, then there is considerable value in providing the necessary information directly to women. Second, if the aim is to empower women in collaboration with their male co-heads in the household, then providing information to female and male co-heads together can contribute to greater involvement of women in joint decision-making and joint action, but may not translate into better agricultural outcomes on jointly managed plots or increased joint sales. Besides, providing extension information to female and male co-heads together may not necessarily benefit women's independent agricultural outcomes. Third, including women as role models may prove particularly useful to reduce men's dominance in the domain of agriculture, which can create opportunities for greater involvement of women. Including women as role models in information campaigns targeted at women can stimulate women's individual decision-making and action and can engender improved joint production and sales. Care should be taken, however, to avoid negative impact on women's individual production and sales. Facilitating women's access to complementary inputs, including capital, may be needed.
Further research could explore the existence of heterogeneous treatment effects. Both video-based extension treatments may have had heterogeneous effects among women farmers with high and low aspirations, or among high-and low-performing women maize farmers. Heterogeneity that is linked to women's relative bargaining power in their household, or to women's (direct) access to complementary inputs needed for adoption, such as off-farm income or own capital, could exist as well. Finally, examining the effects of women's direct access to extension information and role models complemented by improved access to complementary inputs on women's empowerment in agriculture could answer questions about the role of such constraints to women's achievements.     Electronic copy available at: https://ssrn.com/abstract=3505653 Electronic copy available at: https://ssrn.com/abstract=3505653 Appendix