Definition of bias

Bias can be defined as a systematic error in an epidemiological study. Please see chapter 2 on "Biases in Epidemiological Studies" for more detailed information.

Information bias

The first main bias results from the type of information obtained (information bias). A recall bias is present when cases are more likely to remember their exposure than controls. In addition, the interviewer or observer bias can occur when different interviewers have different interpretations of the similar questions or their responses.

A recall bias can be reduced by increasing the timeliness of the survey, i.e. keeping the interval between the event of interest and the survey as short as possible. The interviewer bias can be reduced by developing a questionnaire with clear instructions. Also, interviewers should be trained thoroughly and perform test interviews prior to the start of the survey.

Selection bias

Selection bias results from a systematic error in the selection of the study population. For example,  if those who respond to the questionnaire differ from those who do not respond, a nonresponder bias is present (1). For example people who only own mobile phones will be excluded if the telephone interviews are restricted to landlines. Since mobile phone exclusive users tend to be younger and wealthier than the general propulation, a bias is introduced (2). 

It is therefore important to ensure an overall high response rate, for example by offering incentives to participate or send reminders to non-responders (3). Interview partners should be chosen randomly if telephone or household interviews are performed. For example, this can be done by asking to interview the person in a household whose birthday was last. If possible, information on demographic characteristics of the non-responders should be obtained. This could be achieved by a simple non-responder survey or by collecting demographic characteristics in the population registration office. However, obtaining information on on-responder is time-consuming and not always successful. A non-responder bias can also be corrected during the analysis , by standardising the results by age, sex etc..

References

1. Hennekens CH and Buring JE. Epidemiology in Medicine. Philadelphia 1987. PP 171-2

2. Kempf AM and Remington PL. New Challenges for Telephone Survey Research in the Twenty-First Century. Annu. Rev. Public Health 2007. 28:113-26

3. Silman AJ and Macfarlane GJ. Epidemiological studies. A practical guide. Cambridge 2002. PP 132-7