In general, in statistics, one wishes to find out about the population, but cannot sample all individuals from that population so a sample of individuals from that population is used with the aim of generalising the findings from that sample to the overall population. This can only be done well if the sampling and research process is free from bias. “Biases can be introduced at each stage of the research process. Systematic errors are potentially serious and can lead to bias and invalid conclusions. An example of a systematic error is where a nurse always measures blood pressure lower than colleagues. Random errors give rise to reduced precision but not in general to validity. Random errors can occur through data collection through questionnaire or equipment faults, observer error, responder mistakes, during data processing through coding, copying, data entry, programming and calculating errors. There are three main types of bias: selection bias; confounding bias and information bias. Selection bias occurs when the selected subjects (for the sample) differ in some systematic way from those not selected. This could be through high survey non-response, loss to follow-up of inappropriate choice of sampling frame (a list from which your sample is drawn) or sample. It can also occur through inappropriate choice of comparison group. Confounding bias occurs when researchers have failed to take into account an unknown or unrecorded factor that is associated with both the two factors being examined in the research (see Confounding). Age, gender, deprivation and socio-economic status are some common confounders. Information bias occurs due to systematically incorrect measurements or responses, or misclassifications of disease or exposure status which can result from questionnaire faults (culturally inappropriate questions, ambiguous wording, too many questions, etc.), observer errors (misunderstanding of procedures, misinterpretation, interviewer bias, etc.), responder errors (misunderstanding, faulty recall, wanting to give the ‘right’ answer, embarrassment, suspicion, etc.) and instrument errors (faulty calibration, incorrect dilution, inaccurate diagnostic tests, etc.)”. BR Kirkwood. Essentials of Medical Statistics. Blackwell Science, 1988.
In terms of selection bias or participation bias, in general, men are less likely to participate in a survey compared to women, although as people are more likely to participate in survey where they are interested in the topic, so the gender bias is not always the case. Generally, younger people are also less likely to participate in a survey compared to people in their ‘middle’ years. The response rate among older people can be variable as some retired people may have more time to participate in a survey whereas others have health issues that may make them less likely to participate, for instance, they may be more likely to have sight or hearing problems. People from minority ethnic backgrounds are generally less likely to participate in surveys as are people living in more deprived areas. There may also be further barriers to participation such as those relating to literacy, English (speaking, writing or reading English for people whose first language is not English), and technological issues such as lack of access to the internet or only having a mobile phone not a landline (some national surveys use randomisation of landline numbers to select individuals).
Also see: Causality, Confounding, Confidence Intervals, Effect Modification, Interaction, Small Numbers and Variation.