How should we measure household wealth in our surveys?

Those of us designing a household or an individual survey in low- and middle-income countries (LMICs) often grapple with household wealth measurement. In my work, I typically use the Demographic Health Survey (DHS) Asset module as a starting point and then refine the questions depending on how I want to use this data (e.g., do I want to measure absolute or relative wealth; whether some of the assets included in the long list are universally or rarely owned by my sample households; do I want to compare my survey data with the DHS data, etc.) and how long I want the survey to be.

I have always found this approach unsatisfactory. For starters, the DHS asset index (meant to be a proxy for wealth) focuses solely on asset ownership and does not capture their market value. If your study area doesn’t have much price variation across households, the lack of price data may not matter much, especially if all you care about is relative wealth to estimate heterogeneous treatment effects by, say, asset index terciles. But the valuation becomes important once you are making temporal comparisons, cross-country comparisons, or even within country comparisons with significant price variation across states. Secondly, it overlooks formal and informal financial arrangements by not collecting data on financial assets, debt, and other liabilities, which are crucial components of the net household wealth calculation.

With this in mind, I was happy to come across a recent publication from the World Bank’s Development Data Group introducing new guidelines on comprehensive measurement of wealth in LMICs. Here are a few takeaways that might be of interest to the readers of this blog:

• If possible, include questions that capture all items included in the definition of wealth (real assets, financial assets, debt and other liabilities) in your survey, collecting both their ownership and market value. Include relevant wealth items keeping in mind the socio-economic context of the country. So maybe don’t ask about car ownership if you are in an extremely poor setting, and perhaps there is no point in asking very urban households whether they own oxen.
• Helpfully, the guidelines offer a basic module and an extended module that researchers can utilize to collect household wealth data in LMICs. The basic module includes the essential set of questions needed to estimate net wealth, including (a) the ownership and value of real and financial assets, (b) associated debts, and (c) the business use of assets. The extended module offers richer detail by additionally including questions about documented ownership, tenure security, informal ownership rights, shared vs individual ownership, and how the wealth item was acquired. These modules will soon be released in the public domain.
• Consider including some questions on individual level ownership for at least the most important assets in your survey, such as land and dwelling. Ideally, one should ask the respondent about what they own, rather than a proxy-respondent approach (e.g., asking the husband about his wife’s asset ownership; also see this paper). The more commonly used measures such as per capita household consumption expenditure and per capita household wealth significantly underestimate individual-level wealth inequality. Using data from Cambodia, Ethiopia, Malawi, and Tanzania, the LSMS+ team show that intra-household wealth inequality explains about 12–30 percent of overall wealth inequality, depending on the country context (more details here). If you are interested in measuring intra-household distribution of asset ownership, check out the guidelines from the UN’s EDGE project and the World Bank’s LSMS team.
• Lastly, did you know that the oldest survey on household wealth in LMICs is the All-India Debt and Investment Survey (AIDIS), first conducted by the Reserve Bank of India (RBI) in 1951–1952? I didn’t.