All children’s answers were coded into the SPSS programme for statistical analysis and the missing data were ignored. In the part relating to children’s knowledge about medicines, content analysis had been carried out manually to ensure that all the responses were related to their knowledge and the frequency distribution was used to reflect children’s general knowledge about medicines. For belief items, each item had different choices about the efficacy of medicines; the correct answer was scored 1 and an incorrect answer was scored 0. This scoring method about the efficacy of medicines had been adapted from the previous study relating to children’s perception about the benefit of medicines (Almarsdottir et al., 1996). The normality of the data was tested by Kolmogorov–Smirnov test; the total score of beliefs was skewed (p < .05). Therefore, non- parametric tests were used to find the differences between the groups. Mann–Whitney test was used to find the differences between the gender, age groups and parents’ job, while Kruskal–Wallis test was used to find the differences between the race, education level of the parents and the SES of the family. The value of p < .05 was considered significant with 95% confidence level.