Implications of Choice of Second Stage Selection Method on Sampling Error and Non-Sampling Error : Evidence from an IDP Camp in South Sudan

The most common sampling approach for cross-sectional household surveys in the developing world is a stratified two-stage design, where the first stage is usually a sample from a census-based area frame, and the second stage is a random sample of households from each of the areas selected in the first stage. To overcome the problem of outdated census frame information, it is common to conduct a household listing operation within these areas. However, these listing operations come with severe implications for survey costs, timeframe, as well as quality. To avoid such second-stage operations, some surveys choose alternate approaches for their second-stage operation. This paper compares five of these approaches, namely, satellite mapping, segmentation, grid square, the north method, and random walk, through simulations based on a census conducted in a refugee camp in South Sudan. The paper compares the simulated approach with the estimates derived from the actual experiment and finds that all the resulting estimates are biased. Nevertheless, in addition to their practical challenges, the satellite mapping, segmentation, and grid square approaches exhibit the smallest bias. Although random walk shows the worst performance in the simulations, it regains ground in its implementation, especially vis-à-vis the north method, where implementation adds most significantly to its bias. In conclusion, most probability-based methods perform better than non-probability methods like random walk and are therefore preferrable when no traditional household listing can take place. Although it is important to consider the theoretical properties of sampling approaches, implementation is at least as important. Training, implementation modalities, and monitoring of compliance are key factors in the overall performance.