Background Various sampling techniques have been implemented to assess child restraint use practices. However, little evidence exists about the impact of selection bias. This study explored differences in demographic characteristics and restraint practices between a random and self-selected sample of child restraint users.

Method Data gathered from two Australian observation studies were analysed. Both studies collected demographic data and errors in restraint use. Caregiver risk perception to misuse was assessed via survey. A multistage stratified cluster sampling plan was used to randomly select vehicles from long daycare centres, preschools and primary schools. The self-selected sample included persons who attended council-run restraint fitting days. Chi-square tests determined differences in demographic data by sampling method and logistic regression analyses examined the relationship between sampling method and restraint use practices, controlling for restraint type.

Results Total sample size was 277 (random n= 135, self-selected n =142). Child age (χ²(8) = 54.7, p<0.001), household income (χ²(2) = 8.7, p=0.013), Socio-Economic Indexes For Areas quartile (χ²(4) = 29.9, p<0.001) and risk (χ²(2) = 12.5, p =0.002) were significantly associated with sampling method. Errors of any kind were almost three times more likely in the self-selected sample than the random sample (OR=2.86, 95%CI 1.23–6.65), however, there were no differences in odds of serious errors between samples (p=0.91).

Conclusion Demographic and error profiles were influenced by sampling method, with any error more common in the self-selected sample, but no differences in serious errors. Awareness of systematic sampling biases is necessary when interpreting studies reporting child restraint practices.