Cluster randomisation and design effects in school surveys: an exploration of the Irish Health Behaviours in School-aged Children data.

Abstract

School surveys are typically conducted in classrooms where children are clustered and thus each participating student is not fully independent of others. Such unavoidable cluster randomisation increases the size of standard errors, resulting in widened confidence intervals, larger probability values and therefor reduced statistical power and effective sample size. The need to address these issues is becoming more evident in the literature, with increasing impetus to publish on the effects of such sampling techniques and to consider more fully the analysis implications for complex survey data. A first step is to assess the design effects (DEFF), which is the ratio of the number of participants requred to achieve precision using cluster randomisation to the number that would be required if all participants were independent.