Phone surveillance, from scratch: Novel sample design features of the nationally representative Myanmar Household Welfare Survey (MHWS)

The first round of the Myanmar Household Welfare Survey (MHWS)–a nationwide phone panel consisting of 12,100 households–was implemented between December 2021 and February 2022. The objective of the survey was to collect data on a wide range of household and individual welfare indicators–including wealth, livelihoods, unemployment, food insecurity, diet quality, health shocks, and coping strategies–in a country exceptionally hard hit by conflict, severe economic collapse, and several damaging waves of COVID-19. The respondents interviewed in the MHWS were purposely selected from a large phone database aimed at being representative at the region/state level and urban/rural level in Myanmar.

In this paper, we discuss two key steps taken to ensure that the MHWS is nationally and subnationally representative at the state/region and urban/rural level. First, we used a quota-based sampling strategy by setting survey quotas for respondents’ geography, education, farming status, gender, and rural/urban residence. This sampling strategy is used to address the well-known drawbacks of phone survey samples (e.g., the over-sampling of more educated respondents) and the survey’s particular interest in over-sampling farm households and equally sampling men and women. Second, we constructed household, population, and individual level weighting factors to further ensure that the survey generates nationally and subnationally representative statistics.

To assess the effectiveness of these two strategies on achieving representativeness and consistency with previous surveys, we compare results from the MHWS to earlier nationally representative datasets, focusing on sample sizes of interviewed households for each state/region, and on education levels, farm/non-farm occupation, urban/rural residence, as well as respondents’ housing characteristics, which are unlikely to change substantially over short periods of time. We show that the phone-based MHWS has broader geographical coverage than previous national surveys, reaching 310 of Myanmar’s 330 townships. Moreover, our sampling approach was generally effective in reducing the education bias of phone surveys, except for a handful of states/regions. The MHWS is also unique in providing equal representation of male and female respondents. Additionally, the MHWS sampling and weighting strategies produce statistics on key indicators that closely mirror results from the two most recent national surveys in Myanmar. Overall, the results suggest that these strategies are successful in generating a subnationally representative phone survey that collected data on a rich array of household welfare indicators in exceptionally difficult political and economic circumstances.