745Probability-proportional-to-size cluster sampling as an effective methodology for large-scale participant recruitment

Abstract

Background: We demonstrate the utility of Probability-Proportional-to-Size Cluster Sampling (PPS-CS) to select participants/sites for large scale surveys. Methods: Post ethical and administrative clearance, PPS-CS was carried-out to estimate the prevalence of ‘sublexical dyslexia’, a reading impairment in children from III-VII grades in Udupi district of India. Schools were regarded as clusters. By performing PPS-CS, calculated sample size of 1812 children was systematically recruited. Sampling frame: List of all 1256 schools in Udupi district was retrieved. Kannada medium schools were shortlisted, which yielded 128 schools later classified individually as urban or rural. Results: Strength of schools was retrieved and cumulative strength (cs) was derived. With an average school size of 105 for rural and 98 for urban cluster, 18 schools were required to meet the sample size. Owing to a ratio of 3.5:1 of rural-to-urban students, 14 rural and 4 urban schools were selected. Ratio of ‘cs’ to ‘number-of-schools-required’ gave a sampling interval (SI) of 758 for rural, and 662 for urban cluster. A random number (R) was selected between one and SI. First school picked was that with cs > =R. Second was that with cs > =SI+R. The third was with cs > =(SI+R at II school)+SI. Progressively, 18 schools were identified. Conclusions: With disproportionate sizes of clusters, PPS-CS ensured that selected participants reflect population estimates accordingly. Prevalence of sublexical dyslexia in Udupi district was therefore accurately estimated using PPS-CS. Key messages: Large-scale studies on healthcare or businesses may be effectively carried-out using PPS-CS!