Papua New Guinea Household Survey on Food Systems, 2018

Before designing the survey instrument and specifying the sample design, IFPRI undertook a scoping study of the districts to be included in the household survey and found little variance within communities. Given the low variance in key variables within communities, but greater variance across communities, the survey sample was designed to be as spatially extensive as possible. In order to provide baseline information to World Vision (WV) programs, the survey sample was split into WV communities and non-WV communities. We randomly selected WV and non-WV communities for survey implementation, however, a travel time restriction was imposed on the pool of non-WV communities from which sample communities were selected considering that mobility in PNG is logistically difficult, potentially unsafe, and time-consuming. In doing so, non-WV communities (as to survey clusters) were randomly selected based on a 1 to 4-hour travel time from the WV program communities using GIS analysis. The sample is comprised of one to three districts per survey area depending on the area. The Autonomous Region of Bougainville (ARoB) site comprises the Buin and Siwai areas of South Bougainville district; East Sepik comprises Maprik, Yangoro-Saussia and Wosera-Gawi districts; Madang comprises Middle Ramu and Usino-Bundi districts; and West Sepik comprises Nuku district. Within each selected WV or non-WV community, 15 households were randomly selected for enumeration. In total, 1,026 households were interviewed across the four areas.

To be clear, these data are not nationally representative, nor should they be considered representative at the provincial level. In short, the sample design aimed to collect data from a spatially expansive set of communities to obtain the largest variance possible across survey clusters to better represent the range of rural livelihoods in these areas.

Given that community in PNG does not have a household listing, and it was not financially feasible to collect a household listing, a ‘random walk’ protocol was used to select interviewed households. In doing so, community boundaries were drawn using satellite imagery, a start-point for the random walk was randomly selected by choosing a latitude and longitude coordinate within the boundary as a start-point for the random walk. The nearest household to the randomly selected coordinate was selected as the first household for interview. The following households were selected by identifying every third household along a navigable footpath or motorized track (depending on the community). Studies evaluating this ‘random walk’ method for household sampling have found that, while such a household selection method can provide accurate results, the objectivity, and independence of the household selection process can be easily compromised.