.png){kind=icon}
Especially in studies on humans there is a complex terminology you might encounter on different types of bias that might creep into your sampling. We introduce these to you here so you are familiar with the terms, and also to help you to think broadly about all the ways bias might creep in.
Firstly, selection bias is a general term to describe any situation where the sample is systematically different from the population it is meant to represent in ways that might influence your study.
Sometimes in human studies you advertise the study and ask for people to volunteer to be part of your sample. Volunteer bias is a systematic difference between those people who volunteer and those eligible who do not volunteer in ways that might influence your study.
Non-response bias is similar but subtly different. Here you approach people and ask them to take part in your study; some will agree and some will refuse to take part. Systematic difference between those who are approached and agree and those who are approached and decline in ways that might influence the study is called non-response bias.
Ascertainment bias (sometimes called detection bias or information bias) occurs if there is systematic difference between the true value of traits of individuals in your sample and the recorded values in ways that could influence the study. For example, if you ask people how much they weigh then you might get systematically low values because people (consciously or unconsciously) give you an answer that is influenced by how much they would like to weigh. In this case the bias is driven by the participants, and this form of ascertainment bias is called response bias. Alternatively, you could imagine a study where the researcher photographs people at the beginning and end of the study and is required to record whether these photos suggest that the person has lost weight over the course of the study. If the researcher knows that these people have been on a calorie-restricted diet during the study, then they might (consciously or unconsciously) be more likely to record observed weight loss where none really exists. This type of ascertainment bias that is driven by the investigator rather than the participants is called assessment bias (or observer bias) and is discussed more fully in Section 11.6.3 of the book.
Imagine that we have a study where we ask people who have had a recent mild stroke about their lifestyle in the few years leading up to the event, and compare these answers to those of a control group of people who have not had a stroke. Imagine we ask participants if they have experienced times of particularly acute stress in the last three years. If people in the two groups differ systematically in their accuracy of recall in this regard in ways that might affect the study then this is called recall bias. You could imagine such bias occurring in the trial described because those who have had the stroke might be more motivated to search their memory for possible trigger events than the control individuals.
Imagine a study where a doctor has to allocate patients to either the conventional treatment or a new experimental alternative. If the doctor has more faith in the conventional treatment, then they might (consciously or unconsciously) have a tendency to allocate patients with more serious conditions to the conventional treatment. This would be an example of allocation bias; where there is systematic difference in a way that might affect the study in the subjects allocated to different treatment groups.
Imagine a study that aims to follow people who have completed a treatment program for alcohol misuse for three years afterwards to explore how effective the treatment has been in changing drinking behaviours. Participants are to be interviewed every six months for two years. Some people will inevitably drop out of the study during that time and so not be monitored for the whole three years. One reason for such drop-outs could be that people change address without informing the study, and so cannot be traced to give any further interviews. If those that drop out of the study are systematically different from those that remain in ways that might affect the study then this is attrition bias. In our example attrition bias might be of concern because you could imagine that those who relapse into problem drinking are less likely to be organized enough to inform the study about a change of address, and may also be more likely to change address (say if their problem drinking led to eviction).