OBJECTIVES: This study compares methods for detecting and correcting the bias associated with nonresponse to postal questionnaires. METHODS: Questionnaires were sent out in three sequential stages to parents of all 5-year-old children examined in a clinical survey. Each stage progressively targeted nonresponders. Data on dmft and area measures of socioeconomic status were available for all children. Estimates for whole population dmft were produced by different methodologies comparing the relationship between dmft and stage of response and three area measures of socioeconomic status. RESULTS: A total of 1,776 children were examined and 1,437 questionnaires were obtained, a response rate of 80.9 percent. The mean dmft of the total population (1.49) was 17.3 percent more than responders (1.27). The dmft of the nonresponders was 2.41, 89.7 percent more than responders. There were significant linear trends in dmft and socioeconomic status across the mailing stages. The methodology using mailing stage regressed against dmft produced the most accurate adjusted dmft value (1.42). The methods using area measures of socioeconomic status produced nearly identical adjusted dmft values ranging from 1.31 to 1.32. CONCLUSIONS: Even with an "acceptable" response rate, nonresponse bias can still be present. Researchers should report the outcomes of analyses to detect nonresponse bias when publishing questionnaire studies.
OBJECTIVES: This study compares methods for detecting and correcting the bias associated with nonresponse to postal questionnaires. METHODS: Questionnaires were sent out in three sequential stages to parents of all 5-year-old children examined in a clinical survey. Each stage progressively targeted nonresponders. Data on dmft and area measures of socioeconomic status were available for all children. Estimates for whole population dmft were produced by different methodologies comparing the relationship between dmft and stage of response and three area measures of socioeconomic status. RESULTS: A total of 1,776 children were examined and 1,437 questionnaires were obtained, a response rate of 80.9 percent. The mean dmft of the total population (1.49) was 17.3 percent more than responders (1.27). The dmft of the nonresponders was 2.41, 89.7 percent more than responders. There were significant linear trends in dmft and socioeconomic status across the mailing stages. The methodology using mailing stage regressed against dmft produced the most accurate adjusted dmft value (1.42). The methods using area measures of socioeconomic status produced nearly identical adjusted dmft values ranging from 1.31 to 1.32. CONCLUSIONS: Even with an "acceptable" response rate, nonresponse bias can still be present. Researchers should report the outcomes of analyses to detect nonresponse bias when publishing questionnaire studies.
Authors: Anne-Marie Glenny; Helen V Worthington; Keith M Milsom; Eric Rooney; Martin Tickle Journal: BMC Med Res Methodol Date: 2013-09-04 Impact factor: 4.615
Authors: Timothy J Beebe; Donna D McAlpine; Jeanette Y Ziegenfuss; Sarah Jenkins; Lindsey Haas; Michael E Davern Journal: Health Serv Res Date: 2012-01-17 Impact factor: 3.402