Use of Global Positioning Systems and satellite photographs in survey sampling: A novel method

Background: Statistical analysis of surveys requires knowing the probability of sampling any unit. In high income countries, there is typically sufficient information to allow selection involving multiple stages and/or cluster samples. Elsewhere, one may require an alternative method. Global Positioning Systems (GPS) were used in a survey of mortality in Iraq following the US invasion. Kolbe and Hutson adapted the method in a survey in Haiti. We extended this approach for a population survey in southern Lebanon. Methods: We had limited information on the populations of towns. Still, we sampled towns with probability proportional to estimated size. Within each town, we randomly sampled several locations to within 1 m, and from satellite photographs identified all buildings within a 20 m radius. One was randomly chosen. Interviewers went to the locations and confirmed the building was residential; if the building contained more than one household, one was randomly chosen and approached for interview. An adult was randomly chosen from the household as the primary respondent. The probabilities of sampling a household and individual were the products of the probabilities of sampling at each step. Results: The method allowed us to identify households and respondents with known probabilities of selection and we could also account for the clustering of households / individuals within towns. Discussion: Our experience highlights several issues for discussion, including the following: satellite photos that are out-of-date; counting buildings may not be simple; circles around selected locations may overlap; and difficulty maintaining the safety of interviewers.