Hardware: 1982-1999, VAX 6410 mainframe.
Operating System: VMS, Microsoft Windows NT subsequently.
Application Software: BLAISE developed by Statistics Netherlands, offering: computer-assisted personal interviewing (CAPI), Computer-assisted telephone interviewing (CATI), and Computer-assisted self interviewing; Interactive data editing; forms-based data entry; structured questionnaire design; complex routing and checking; top-down editing with interactive tables; survey management; data manipulation and tabulation; interactive coding; export to statistical and database packages. Until 1999, data sent to the Home Office in SPSS was exported into P-STAT, software used by the Home Office to hold and manipulate BCS data. From 2001, this was changed to Quancept CAPI, produced by SPSS. The related Quanquest module was used to author the questionnaire, which was designed to replicate the BLAISE surveys.
User Interface: The datasets were accessed via a Windows-type interface. Files were opened and manipulated using SPSS Data Editor, in which data was displayed in a tabular format. Dialogue boxes could be used for basic manipulation; separately created syntax files were used for more complex manipulations.
Logical structure and schema: Before the migration of the BCS datasets from P-STAT to SPSS (in 1999), datasets consisted of separate tables which corresponded, broadly, to the main sections of the survey questionnaire. Since 1999 each dataset has primarily consisted of two tables: one for data from the Victim Forms section (the 'vf' table), the other for data from all other sections of the questionnaire (the 'nonvf' table). From 1994 the datasets have field names which normally correspond to the fields used in the interviewing software.
However, the 1984-1992 field names have prefixes indicating the section of the survey questionnaire to which they relate; including: M (Main Questionnaire), V (Victim Forms), FA (Follow-up A), FB (Follow-up B), D (Demographics) and SC (Self-completion).
How data was originally captured and validated: BCS data is gathered from a 'core' sample of adults aged 16 and over. The first four surveys (1982-1992) achieved successful interviews with a core sample of between 10,000 and 11,000 respondents in England and Wales. In 1994 it was decided to increase the core sample to around 15,000 adults in England and Wales; the sample was further increased to 19,411 for the 2000 survey, and 37,170 in 2001/2002.
In the first three surveys electoral registers were used as a sampling frame. By 1988 it was acknowledged that the use of electoral registers was flawed, because they do not cover all households. The 1992 survey and all subsequent sweeps of the BCS have based their sampling frame on the Postcode Address File (PAF).
In 2000 primary stratification by inner city and non inner city was abandoned; instead, primary sampling was done by Police Force Area (PFA). In response to the higher incidence of crime in inner city areas, every BCS from 1982 until 1998 oversampled these areas by a factor of about two. The 2000 BCS abandoned the oversampling in favour of an oversampling of the smallest PFAs, in order to ensure a minimum number of achieved interviews per PFA.
The 1988-1996 and 2000-2001/2 sweeps employed 'booster' samples of ethnic minorities. These methods produced booster samples of 1,349 ethnic minority respondents in 1988, increasing to a peak of 3,874 in 2000. The 1992 BCS also employed a one-off booster sample of children aged 12-15, from households where an adult had already been interviewed.
In 2000 the Home Office announced that, with effect from 2001, the BCS would be held on an annual rather than a biennial basis and would aim for a larger sample of around 40,000 adults (including ethnic minority booster). In preparation for the enlarged survey the Home Office commissioned a review of BCS methodology; it recommended:
- That data should be gathered through continuous interviewing.
- That the recall period for questions relating to victimisation should be the 12 months prior to the interview.
- That in order to measure the effect of the changes, the 2001 BCS should employ a mixture of the old and new methodology.
Paper forms were used in the BCS until the 1994 sweep, when Computer Assisted Personal Interviewing (CAPI) using BLAISE was introduced. In CAPI, answers to survey questions are logged by the interviewer onto a laptop computer. CAPI was used for all sections of the BCS questionnaire except for the Self-completion sections, where Computer Assisted Self Interviewing (CASI) was used. Under CAPI/CASI, the entire interview consists of one electronic questionnaire.
Coding and classification: The input, coding and checking responses was done by coders employed by the contractor(s) carrying out each sweep of the survey. This included coding the occupation, social class, etc. of the respondent and the head of household; and coding the offences recorded on Victim Forms, according to a standard scheme for classifying offences. The BCS's instructions for classifying offences were developed by the Home Office in consultation with a number of police forces.
After the introduction of CAPI/CASI in 1994, post-interview coding involved the editing and checking of data input as part of the interview. This included checking 'open-ended' questions, where a response was recorded in the respondent's own words. In the 1992 survey the contractor developed a computer program which suggested the most likely classification for an offence, based on a range of variables recorded in the Victim Forms. Similar programs were used in subsequent sweeps. The classifications suggested by the computer were confirmed or modified by coders.
Weighting: In each sweep data has been weighted at the data processing stage, to correct for imbalances in the sampling and design of the interview. Every BCS sweep has used a series weight to account for the fact that series incidents are recorded on a single Victim Form, and an inner-city weight to correct for inner city oversampling.
Other weights include:
- To account for the mismatch between electors and persons at addresses.
- For household crimes, where the victim is the household rather than the respondent.
- A dwelling unit weight (applied to from 1992 onwards), to correct for cases where more than one household was covered by an entry on the Postcode Address File.
- An individual weight (applied to surveys after 1992), to account for the fact that individuals living in larger households tend to be under-represented.
- Ethnic minority weights (used in the 1992 and 1996 BCS), to down-weight this booster sample.
- The 2000 and 2001 surveys employed a police force area weight, to correct for disproportionate sampling.
Constraints on the reliability of the data: The BCS faces difficulties in accurately representing the entire population. The Postcode Address File was adopted to help overcome this problem. May also tend to over-represent older age groups. Because of the small number of incidents, estimates of sexual offences are not likely to be reliable. It also undercounts crimes where the victim and offender know each other.
Validation performed after transfer: Details of the content and transformation validation checks performed by NDAD are contained in the catalogues of individual datasets.