Increasing questionnaire response: evidence from a nested RCT within a longitudinal birth cohort study
Published version
Peer-reviewed
Repository URI
Repository DOI
Change log
Authors
Abstract
Abstract: Background: High response rates are essential when questionnaires are used within research, as representativeness can affect the validity of studies and the ability to generalise the findings to a wider population. The study aimed to measure the response rate to questionnaires from a large longitudinal epidemiological study and sought to determine if any changes made throughout data collection had a positive impact on the response to questionnaires and addressed any imbalance in response rates by participants’ levels of deprivation. Methods: Data were taken from a prospective, comparative study, designed to examine the effects of the reintroduction of water fluoridation on children’s oral health over a five-year period. Response rates were analysed for the first year of data collection. During this year changes were made to the questionnaire layout and cover letter to attempt to increase response rates. Additionally a nested randomised control trial compared the effect on response rates of three different reminders to complete questionnaires. Results: Data were available for 1824 individuals. Sending the complete questionnaire again to non-responders resulted in the highest level of response (25%). A telephone call to participants was the only method that appeared to address the imbalance in deprivation, with a mean difference in deprivation score of 2.65 (95% CI -15.50 to 10.20) between the responders and non-responders. Conclusions: Initially, low response rates were recorded within this large, longitudinal study giving rise to concerns about non-response bias. Resending the entire questionnaire again was the most effective way of reminding participants to complete the questionnaire. As this is a less labour intensive method than for example, calling participants, more time can then be spent targeting groups who are underrepresented. In order to address these biases, data can be weighted in order to draw conclusions about the population.