Attrition

Definition

The ‘wearing away’ or progressive loss of data in research.

Distinctive Features

Attrition occurs when cases are lost from a sample over time or over a series of sequential processes. One form of sample attrition occurs in longitudinal research when the subjects studied drop out of the research for a variety of reasons, which can include: unwillingness of subjects to continue to participate in research, difficulties in tracing original respondents for follow-up (for example, because of change of address) and nonavailability for other reasons (for example, death, serious illness). A survey of major longitudinal studies in the United States found that the average attrition rate was 17 per cent (Capaldi and Patterson, 1987, cited in Sapsford and Jupp, 1996). Therefore, quite a lot of cases may be lost.

A second sense of attrition relates to the loss of data from secondary sources in the process of collection. A commonly used example is the case of crime statistics, which are the end product of processes of decisions by victims to report crimes to the police, police decisions about recording and decisions by various agencies in the criminal justice process which may or may not result in prosecution, conviction and the imposition of a particular sentence. Using the British Crime Survey (BCS) (a random household survey of criminal victimization) as a base line, it is estimated that only 45 per cent of the crimes recorded by BCS are reported to the police, only 24 per cent are recorded as crimes, 2 per cent result in a conviction and 0.3 per cent result in a prison sentence (Home Office, 1999). Thus there is a considerable loss of data between the base source and subsequent measures.

Evaluation

The problem of attrition is not merely the problem of a reduction in sample size but, more importantly, it raises the possibility of bias. Those who ‘drop out’ may have particular characteristics relevant to the research aims (for examples, see Sapsford and Jupp, 1996: 9). The same point applies to the crime statistics example: certain crimes may be less likely to be reported, recorded or result in a conviction. For example, Kelly (2000), reviewing studies of attrition in rape cases, found that fewer than 1 per cent of these cases [Page 13]result in conviction. Researchers thus need to be aware of the problem of attrition and the threats to representativeness that can result.