The Scottish Government

« Previous | Contents | Next »

Listen

SCOTTISH CRIME SURVEY 2003

APPENDIX B: METHODOLOGY

Full details of the methodology used in this survey can be found in the Technical Report produced by the survey company to accompany the dataset. An electronic version of this report can be obtained from the Scottish Executive on request.

Sample Design

The 2003 SCS, like previous sweeps of the survey, is a probability sample selected according to a multi-stage stratified design.

Selection of Sample Points

As in previous surveys, the sample was clustered in order to make it more economic than it might otherwise have been. This had the adverse effects of increasing the error associated with survey estimates, but was counteracted by using a large number of sampling points with a few interviews in each. Compared with 2000, the number of sampling points was increased from 334 to 455 and the number of interviews sought in each point was reduced from 15 to 11 in order to improve the accuracy of the survey estimates by lowering the design effects. Five thousand interviews were targeted.

Sampling points were stratified and spread across the whole of Scotland including the Highlands and the larger Islands. The sampling points were based on Census enumeration districts.

The primary stratifier was based on former local government regions. These regions no longer have any administrative function following local government reorganisation in 1996 but they are used for reporting crime statistics and they serve the primary purpose of ensuring a proportionate distribution of the survey sample. In previous surveys Scottish *ACORN had been used as the secondary stratifier within regions but in 2003, the Mosaic classification was used. However, apart from this minor change, the stratification was the same as that used in the 1996 and 2000 SCS.

Selection of primary sampling units and addresses

Since 1993 the Postcode Address File (PAF) has been used as the sampling frame for the SCS. This provides details of all postal delivery addresses in Scotland.

There are some potential disadvantages to PAF. First, it can contain a small number of properties which are not private residences. To overcome this, these were simply screened out during the survey. Second, and potentially more serious, is that the number of residences at each property is not always correctly identified by the number of delivery points listed in the PAF. For example, a single address may consist of a tenement block containing 8 separate flats. In most cases the existence of these additional addresses was indicated in the PAF, and the address was weighted to ensure each household has an equal chance of inclusion in the sample. Where the additional households are not identified in the PAF, the interviewer had to undertake a procedure to select one dwelling at random for interview. These cases should, in principle, be given an additional weight to balance the fact that these properties had a reduced likelihood of being sampled. In practice, because there are so few properties affected, and because the weights themselves would have no beneficial effect on the survey estimates, these weights were not applied.

Respondent selection

Once a household had been selected and they had agreed to take part in the survey, a respondent from that household had to be selected at random from all the adults resident at that address. To achieve this, the interviewer listed all adults (aged 16 and over) resident in the dwelling at the time of the survey. One of them was selected at random using a 'Kish grid'. No substitution of respondents was permitted.

The SCS Questionnaire

The questionnaire was divided into sections. In addition, two versions of the main questionnaire (Questionnaires A and B) were administered, each to half the sample. To summarise, the questionnaire sections and contents were:

Main Questionnaire A and B:

• introductory questions on fear of crime and general social issues
• screening questions on victimisation experiences (designed to identify any crimes that had affected the respondent since 1 January 2002, although very minor offences such as theft of a milk bottle from a doorstep were screened out at this stage)
• a check for eligibility for Victim Forms
• a follow up section, version A or B as appropriate
• socio-demographic questions.

Questionnaire A:

• Contact with and attitudes towards the police
• Contact with other parts of the criminal justice system
• Attitudes towards sentencing and the prison system.

Questionnaire B

• Attitudes towards personal safety in own neighbourhood
• Security features of home and car
• Experience of housebreaking, violent crime, nuisance telephone calls and anti-social behaviour.

Victim forms

Incidents identified in the screening section of the main questionnaire were followed up in Victim Forms, which collected more detailed information about the incident (or series of related incidents). The Victim Form also explored other factors that relate to the crime such as whether they think there was any racial motivation; the impact of the offence; whether the police were involved and what help they received.

The processes related to victim forms have changed somewhat over different sweeps of the survey, but in the 2003 SCS the 1996 practice of completing up to 5 Victim Forms in full was followed.

Self-completion module

A self completion module was asked with two halves: one on exposure to, and use of, illegal drugs and the other on experience of domestic violence. In 2003 it was decided to administer the self-completion questionnaire to all adult respondents rather than restrict it to adults under the age of 60 years as had been the case in previous years.

Fieldwork

Fieldwork for the 2003 SCS ran from March to July 2003. No interviews took place between 1 April and 1 May inclusive, owing to the 'purdah' period prior to the Scottish Parliamentary election on 1 May. This fieldwork period was approximately three months later than previous SCS fieldwork periods. The survey reference period - the period for which incidents and victimisation are recorded - remained the previous calendar year, January 2002 to December 2002.

Response rates

A total of 8,190 addresses were issued to interviewers from which 5,041 interviews were achieved. This represents an unadjusted response rate of 61.6 per cent. After taking account of ineligible addresses, the response rate was 68.1 per cent. Table B.1 shows the distribution of out-of-scope addresses and in-scope responses.

Table B.1: Summary of outcomes at issued addresses

4,665 adults completed the self-completion module representing 93 per cent of the sample who participated in the main survey and 63 per cent of the eligible sample.

Coding and Data Preparation

Offence classification

The purpose of the offence classification is to identify a single offence code for each Victim Form from which victimisation rates can be calculated and used for comparison with other statistics (e.g. police statistics and past crime surveys).

The framework for offence classification had been developed over the series of crime surveys. Prior to the 1993 SCS, offence coding instructions, consistent with previous British Crime Surveys were drawn up. These instructions were used for the 1996 and 2000 SCSs without significant amendments. However, some important changes to the coding of crimes have been made to the 2003 SCS. The definition of housebreaking now differs from both from the British Crime Survey's definition of burglary and previous SCS reports. This coding change was implemented to more accurately mirror Scottish police recorded crime definition of domestic housebreaking, by including domestic housebreakings to non-dwellings. This does not change the overall incidence of acquisitive crime, but merely changes the proportions of both housebreaking and 'other household theft'. The differences between the two definitions can be seen in Table B.2.

Table B.2: The difference between the old and new definition of housebreaking in estimates of the extent of victimisation in Scotland

Data Weighting

Weighting was conducted in four main stages:

• household weighting (main sample only)
• individual weighting
• weighting for grossing, victimisation and prevalence
• non-response weighting for self-completion questionnaires.

Each stage of the weighting looked at two different elements:

1. weighting to compensate for design elements of the survey, and specifically, elements that altered the probability of sampling units being from a simple random sample.
2. weighting to compensate for non-response bias where this was observable by comparing the survey data with population estimates.

Each of the stages of weighting will now be taken in turn.

Household weighting

This weights the data to adjust for factors relating to household selection. The sample for the survey was distributed nationally in proportion to the population so no design weights are necessary at this stage. However, weighting was necessary to compensate for non-response bias in terms of area bias where more or less than the target number of interviews (11) were achieved within a cluster.

Individual weighting

Within the main sample, all household data (e.g. tenure or property/household related offences) had to be weighted by the household weighting factor prior to analysis. Data that applies to individuals (e.g. attitudinal data or offences against the person) had to be weighted by a combination of the household weight and an individual weighting factor.

There are two reasons for weighting individuals. First, it is standard practice for surveys of adults to interview only one adult per household. The data therefore need to be weighted to account for the fact that each individual's chance of selection for interview varies depending on the number of adults in the household. Secondly, there can be systematic patterns of non-response to the survey and errors in respondent selection, both which can lead to bias.

Before bias could be assessed, the weighting for the number of adults had to be carried out in order to remove the natural bias in favour of single adult households, which tend to be biased towards older females. This weighted profile was then be compared with the profile of all adults in the household, which indicates selection bias. After accounting for selection bias, any residual bias observable by comparing the weighted profile of the sample with the profile of the adult population from the 2001 Census was dealt with.

Victim Form weighting

A victim form can represent either one incident or a series of similar and related incidents. Forms that represent a single incident got a weight of 1. Forms for a series of incidents were weighted by the number of incidents they represent. As in previous sweeps of the SCS the weight for the series victim forms was capped at 5 - partly to limit the impact on victimisation rates of outliers and partly to limit the impact of the weights on the survey standard errors.

Because the victim forms refer to either a household (in the case of property crimes) or an individual (in the case of personal crimes) the Victim Form weight was multiplied by the household weight (HH_WT) or the individual weight (IND_WT) for the record to which the form related.

Grossing weights

In addition to the basic design and corrective weights, the SCS requires weights that:

• express the data from the survey in terms of the number of households and adults across Scotland experiencing different types of incidents
• express incidents as a rate per 10,000 units (households, adults, households with access to a vehicle, households with access to a bicycle)
• express victimisation as a rate per 10,000 units (households, adults, households with access to a vehicle, households with access to a bicycle).

This was accomplished by multiplying the household and individual weights by additional weighting factors. These included population grossing weights to multiply up the total achieved sample to the population (aged 16 plus) total and the total number of households in Scotland.

Four weights were calculated for:

• household incidents, based on a household population of 2,192,346
• personal incidents, based on an adult population of 4,007,466
• motor vehicle incidents, based on a survey estimate of 1,468,799 households with use of a motor vehicle
• bicycle incidents, based on a survey estimate of 783,486 households with access to a bike.

Adult self-completion

The adult and young person forms were subject to the same selection probabilities and biases as the main survey, with additional scope for bias arising from non-response to that aspect of the survey.

The main weight for the adult self-completion form is the individual weight. However, additional bias in the response to the self-completion section was assessed by comparing the weighted profile of adults who completed a questionnaire with that of the adult population and small additional adjustments were made to the weights on the same basis as those used for adjusting the individual weight in the main data.

« Previous | Contents | Next »