A few weeks back I blogged about the importance of incorporating household and family dynamics in our research to gain a more comprehensive understanding of how policies influence individual behaviors and outcomes. Building on that discussion, today I want to share a few thoughts on how we can improve our data collection, intervention designs, and analytical approaches to do so.
First and foremost, and before launching your intervention, try to understand the socio-cultural context in which you will be conducting your study. While this step may seem self-evident, it's often overlooked. Before you jump in and finalize your research design, conduct focus group discussions and qualitative interviews with diverse stakeholders, collaborate with local researchers, and read academic literature from adjacent disciplines such as sociology, history, and anthropology on your topic of interest. Immersing yourself in the field, exploring non-academic literature, and even consuming popular media like movies and TV shows can provide valuable insights into relevant family structures and the underlying social norms shaping intrahousehold hierarchies. Given how time intensive this exercise can be, it is not surprising that researchers tend to specialize in certain countries or regions. Essentially, we should assume that we don’t know everything and, worse still, may have biased beliefs about unfamiliar contexts and populations.
Make the most of your baseline survey by validating your initial assumptions, and the insights gained from the preceding exercise, within your study sample. In randomized experiments, baseline surveys serve a dual purpose: they facilitate identification and estimation by allowing us to assess balance and measure baseline values of key outcome variables, which can later serve as controls. However, beyond their statistical utility, baseline surveys offer invaluable contextual understanding, especially in scenarios where other representative data sources are lacking or in highly diverse contexts. The household roster is a good place to start. By examining who lives with whom, you can find out the prevalence of extended households and patrilocal residence, for instance. By including appropriate questions, you can identify key decision-makers within the household and gain deeper insights into familial structures and dynamics.
Beware of the nuclear family bias when designing survey questions. Take, for example, the Demographic Health Surveys, which are one of the most important sources of data from a wide range of countries. These surveys gather data on women’s autonomy, decision-making power, and preferences solely in the context of their relationships with their husbands through questions like:
“Who decides how the money you earn will be used: mainly you, mainly your husband, or you and your husband jointly?”
“Would you say that using contraception is mainly your decision, mainly your husband's decision, or did you both decide together?”
This approach inadvertently sidelines other familial dynamics and fails to capture the complexities of decision-making processes within non-nuclear family arrangements. Even questions that have an option for choosing “someone else” are framed with respect to the husband, and do not ask the respondent to specify who this other person is. For example:
“Who usually makes decisions about making major household purchases: mainly you, mainly your husband, you and your husband jointly, or someone else?”
Instead, I prefer the approach followed by the Indonesia Family Life Survey (IFLS). The IFLS-5 asks: In your household, who makes decisions about [XYZ]?”. The respondent can select multiple household members in response. A similar approach is adopted by the India Human Development Survey (IHDS). In IHDS-2, respondents are first asked “Please tell me who in your family decides the following things?”, again allowing for multiple selections, followed by “Who has the most say in the decision?”. The IHDS offers a smaller choice of response options than the IFLS, but both are significant improvements over the DHS decision-making questions.
Does this make a difference to our understanding of how the household functions? Yes. For example, according to the IHDS-2 data, in 28% of the rural households, a senior male (different from the respondent’s husband) has a say in decision-making about whether to buy land or property, and in 18% of the households, a senior male has the most say on this decision within the household. Similarly, a senior female or a senior male (i.e., not the respondent or her husband) have a say in decision-making on how much money is spent on a social function such as marriage in 33% of the households. If we only asked these questions the way the DHS does, we would miss out on capturing the intrahousehold decision-making process accurately.
Incorporate the socio-cultural variation in your design. Depending on your research question and the underlying variation in family or social structures in your study context, you can stratify your randomization along a particular dimension, and later examine heterogeneity in treatment effects in that dimension. To illustrate, Ashraf et al (2020) find that the effect of school construction programs on female education differs by the prevalence of bride price in a community – the program had no effect on the education of girls from groups without bride price but it had large positive effects for girls from groups with bride price. Although this paper was not based on an RCT, if one were to implement a similar program through a randomized experiment, it would be valuable to stratify by the prevalence of bride price customs in an ethnic group in advance. One could similarly stratify by the prevalence of polygamy, patrilineality, different inheritance norms and property rights, and extended families, among other things, when appropriate.
Hopefully, these ideas will be helpful as you plan your next research projects. If you have any additional thoughts or suggestions, do share them via the comments section below!
Join the Conversation