Milking Gender Data for Improved Service Provision in the Kenyan Dairy Sector

Research Associate Sam Strimling argues that one must understand the specific needs of women to inform effective service delivery and presents an upcoming effort to gather these insights prior to launching a PAD advisory service aimed at Kenyan dairy farmers this summer. This post is the first of a two-part series, in which the subsequent installment will provide an analysis of the results for the survey described in this post. This is also the third in a series of blog posts on gender and digital development to mark International Women’s Day and Women’s History Month.  

PAD empowers smallholder farmers across the developing world with actionable information to improve their wellbeing. However, there is evidence that digital advisory services like ours are underused by women. This summer we are expanding our agricultural advisory service in Kenya to the dairy sector, an arena in which women play a central role. As we prepare to launch, we are looking beyond one-size-fits-all solutions to more effectively address the unique needs, preferences, responsibilities, and constraints of female farmers.

This post is the first of two about a survey we are conducting in April 2021 to better understand gender roles and power dynamics in Kenyan dairy production and to better target our new service to the underserved needs of women in this sector. Here we describe the motivation and design of this survey. In a follow-up post, we will share insights from the results.

Invisible Women Farmers

In her spectacular book Invisible Women, Caroline Criado Perez details case after case in which service providers failed to deliver because they neglected to collect data about female users’  specific needs, preferences, and priorities.

The book cites the example of an organization seeking to provide “improved” seeds for staple crops: Male farmers cared more about increasing outputs. Women farmers, on the other hand, were more concerned with their input of time and labor in the field and in cooking the staple for household consumption. Since, adopting this new technology would require those doing the planting —  women — to increase their time and effort, the seeds were not adopted.

Women often have less agency in agricultural and household financial decision-making. This is illustrated by Criado Perez who cites two failed attempts to introduce clean cookstoves capable of significantly reducing women’s exposure to toxic smoke. A model promoted in India required regular maintenance, which was considered men’s work. Since the new stoves improved women’s health, but didn’t directly affect men, stoves often fell into disrepair for want of maintenance. In Bangladesh, the fact that the clean cookstoves did not directly improve men’s outcomes, meant that they were not purchased at all, as men controlled household finances.

Understanding Context and Culture

Dairy is a sector of agricultural production in which tradition and gendered labor run deep. As we prepare to roll out our new advisory service, we commenced by focussing on a series of questions we need to answer to understand labor and decision-making dynamics in the sector, and within smallholder households more generally. These include:

  • Who performs which tasks in the household, and within dairy farming specifically?
  • Who holds the authority to make decisions?
  • How systematically do users think about farming and what level of knowledge do they have about dairy farming? Does this differ by gender?
  • Who controls household finances and for what types of purchase, and which accounts do they use?
  • To what degree do women and men communicate about decisions and trust each other to make decisions in the interests of the household, even if one is the final arbiter of a particular decision?

The phone survey we designed to collect this information is due to be deployed in the field in early April, and will be asked to pairs of spouses. The goal is to obtain information to enable us to better target our services to both genders. By asking both men and women within the same household the same set of questions, we will be able to learn about typical responses for each gender, and we will also be able to note whether there are some questions for which we receive vastly different responses from a husband and wife within the same household.

Our survey will consist of six sections:

  1. Household Tasks and Time Use: This section will collect information about the gender(s) and age(s) of those performing various household tasks, including crop farming, livestock grazing, and water fetching. This section also attempts to account for household members’ time spent on other tasks, such as jobs outside the home and schoolwork. Information arising from this section will allow us to measure, for each time-use category, whether an activity is primarily performed by men, women, girls, or boys, or a combination thereof.
  2. Dairy Tasks: This section is similar to the Household Tasks and Time Use section, but focuses specifically on dairy farming, since the research team is most interested in using the results of this gender survey for informing the design and distribution of dairy advisory messages. This section will allow us to measure whether the following dairy tasks are primarily performed by men, women, girls, or boys: cleaning milking equipment, observing animals, giving cows water, feeding cows/calves, milking, milk transportation/sale, cleaning the housing unit (if zero-grazing or semi zero-grazing), sourcing/threshing fodder, spraying/dipping (for tick prevention), deworming, seeking veterinary treatment / artificial insemination (AI) services, identification (ear tagging), debudding/dehorning, hoof trimming, and grooming.
  3. Scenario-Based Questions: These are questions for which answers seek to understand respondents’ critical thinking abilities. An example of such a question would be: “Mrs Choge realized that to boost milk production beyond 7 litres a day, she needed to give her cow sufficient concentrates. For every 1.5 litres of milk above 7 litres, she gave 1 kg concentrate. How much concentrate should she give a cow producing 10 litres of milk per day?” We will disaggregate the results by gender to determine whether, on average, men or women think more systematically about dairy farming. This type of question tests the respondent’s causal reasoning, logical deduction, and systematic planning.
  4. Knowledge-Based Questions: This section is similar to the scenario-based section, but instead tests respondents on their technical dairy knowledge. As in that section, these questions have correct answers, and we will disaggregate the results by gender to determine whether, on average, men or women have more technical knowledge about dairy farming.
  5. Financial Decisions: This section will allow us to draw conclusions about who within the household holds financial power, with regards to purchases of productive assets, personal assets, and consumption goods. We will measure this empirically by collecting information, disaggregated by gender, on transaction amount as well as account type (digital/cash, joint/solo).
  6. Public Goods Game: In this section, respondents will be asked whether they want to augment the amount earned for participating in the survey by allocating a portion to a common account with their partner. A random amount will also be added to the survey by the research team, such that it is not possible for the respondent to lose money by completing this exercise, only to increase their earnings by a variable amount depending on the contribution made by both members of the spousal pair.1 This section allows us to collect data on trust within each spousal pair, which we assume to be correlated with the total amount contributed into the common account.

We plan to publish our findings to inform our own platform development and to make them available to other service providers and policy makers concerned with smart service design. In doing so, we plan to add to the existing body of research on gender and international development — a more complete review of which was published by my colleague Theresa Solenski in an earlier blog post.

Surveying the Evidence

Specifically, our research will add to the empirical evidence base on asymmetric information within the household. One study in Ghana found that on average, both men and women commonly underestimate or overestimate their spouse’s expenditures by more than 75%. Concerningly, this research found implications for profitability: the more imperfect the husband’s information about his wife’s expenditure, the lower the wife’s output/profit, and vice versa (Chen & Collins, 2014). 

Another study from Uttar Pradesh provided an example in which women were less likely to adopt productivity-enhancing technology (in this case, laser land-leveling because they were less aware of its impact (Magnan et al. 2015)). Similarly, a study from Uganda in which agricultural videos were shown either jointly to a spousal pair, or separately to one member, found that information was not shared within the household when just one member was engaged (Lecoutere et al. 2019). These examples are particularly relevant to our work at PAD, as they highlight a way in which our service has the potential to fail: if our messages primarily reach men but women are charged with implementing the recommendations we provide (or vice versa), the suggested practices may not be adopted because we failed to get the information into the hands of women who can make it actionable.

Our survey will contribute to the literature on within-household asymmetric information in two ways. First, the sections on household tasks / time-use, dairy tasks, and finances will — in addition to providing information about gender roles — enable us to compare directly the wife’s knowledge of household affairs to that of her husband (and vice versa). Second, the sections quizzing respondents on dairy knowledge and testing of systematic thinking through scenario-based questions will measure within-household information asymmetries with respect to dairy farming.

Imperfect communication and limited trust also have implications for message adoption: Qualitative evidence from a focus group discussion with 113 male farmers and 141 female farmers in Malawi suggests that men often do not trust agricultural information received from their wives, and female farmers may be hesitant to suggest new agricultural practices for fear of retribution (Ragasa et. al., 2019). Another group of researchers partnered with Kakira Sugar Ltd., a large sugar company in Uganda that sources sugarcane through contracts with smallholder farmers, the vast majority of whom are men. This study implemented an intervention in which men were encouraged to transfer uncontracted blocks of sugarcane to their wives. Since both spouses needed to agree to the proposed intervention, whether the men were willing to do so provided in and of itself a measurement of trust. The researchers found a man was less willing to transfer the contract — i.e. was less trusting of his wife’s capabilities in this respect — if the household had fewer assets and/or expenditures, if his wife was not already involved in sugarcane farming, or if he had previously expressed a preference for managing the household’s finances himself (Ambler et al., 2021).

In our survey, we plan to collect novel data on trust within the household using the public goods game. Ambler et. al. applied a similar approach in their baseline survey, and collected time use data. They found that cooperation in the public goods game significantly predicted a husband’s willingness to transfer the contract to his wife, but time-use data collected did not have the same predictive power.

Finally, by collecting information on respondent demographics and knowledge alongside information about financial decision-making and intra-household communication and trust, our survey will add to existing literature about factors that influence bargaining power within the household. For example, a descriptive analysis of a Tanzania spousal dataset shows that a woman’s age, health, labor hours, and education were positively correlated with her authority to make decisions in the household; however, the number of children she had and her husband’s community standing were negatively correlated with this authority (Anderson et. al., 2017). In our survey, we plan to collect information on each respondent’s age, education level, and number of children, which we can use to disaggregate responses about financial decision-making and the amount contributed to the collective account in the public goods game. This will allow us to investigate whether factors similar to those observed in Tanzania hold weight in the Kenyan context.

Next Steps

We plan to begin the survey in early April and will interview approximately 600 dairy farmers via phone, over two rounds.2 We will first survey 300 respondents from dairy cooperatives, during which we will ask about contact information for their spouses. We will then complete a second round with the spouses of the first round participants.

The sample for the gender survey was constructed to select a representative sample for the two cooperatives interested in partnering with us for this survey, Sirikwa Dairies and General Limited and Katheri Dairy Farmers Cooperative Society. We chose these two cooperatives because they represent the two primary dairy producing regions in Kenya, and because they were able and willing to provide demographic information for their membership sufficient for us to stratify the sample by gender, total production, and total deductions on member accounts for expenses such as agrovet purchases, artificial insemination, veterinary services, and cash advances. This information will enable us to research intra-household gender dynamics for farmers of different levels of farming sophistication.

We expect this research will yield important insights, and we plan to publish the results in a subsequent blog post. The most important outcome, however, will be to improve our content and message delivery for the dairy farmers we serve.


1. This approach has the added benefit of preventing the respondent from obtaining exact information about his/her partner’s contribution. We recognize that particularly female respondents may fear retribution if her partner were to discover her contribution and deem it insufficient. Our contribution is intended to prevent this outcome.

2.  This number was arrived at using Cochrane’s (1977) formula with a margin of error 7.5% and confidence interval of 95% (both commonly accepted levels) to come to the following sample size of households for each dairy: 148 in Sirikwa and 145 from Katheri. Multiplying both numbers by two, we arrive at a minimum sample size of 586, which we round up to 600 for simplicity..