The Scottish Government

« Previous | Contents | Next »

Listen

CHAPTER 2: METHODOLOGY

The survey of 17-39 year old drivers

The approach adopted for the survey had, as a key objective, to provide a nationally representative sample of 17-39 year old drivers to provide robust comparisons with the survey conducted in 2000. The broad approach was the same - the absence of a sampling frame of drivers meant screening households to identify potentially eligible respondents and, within households, recruiting them to participate in the survey.

The data for the survey, therefore, was intended to be representative of the general population of drivers aged 17-39 years. Although the estimates can be compared with data from other studies such as the Scottish Crime Survey, or studies of drug use among sub-populations such as clubbers, people arrested for driving offences or people killed or injured in road traffic incidents, comparisons need to be mindful of the different bases on which the data are collected.

The sample was selected using Census output areas ( OAs) as primary sampling units. These have two advantages. First, they are small areas, averaging 50 households in Scotland and, therefore, represent an administratively efficient unit for an interviewer to be expected to screen all addresses in the area. Second, they are linked to Census data giving information on car ownership and the structure of the population.

Although Census results are available for every output area, the Census does not allow us to identify 17-39 year old drivers directly. It separately identifies the proportion of households with access to one or more cars and the proportion of the household population aged 17-39 years. Either variable on its own is not an adequate basis for sampling since there is the possibility that an area with high car ownership would contain few 17-39 year olds or that an area with a high proportion of 17-39 year olds would have low car ownership.

The sampling approach was as follows.

1. For every OA a proxy indicator of eligibility was calculated. This was the product of the proportion of car owning households and the proportion of the household population aged 17-39 years. For example, an area with 70% car ownership and 25% of the population aged 17-39 years would have a value of 0.175. An area with the same proportion of 17-39 year olds but only 30% car ownership would have a value of 0.075. This indicator therefore gives higher probability of selection to areas with high car ownership and a high proportion of 17-39 year olds.

2. The resulting indicator was multiplied by the total number of households in the OA to give an indication of the minimum number of eligible households in the area. Sample selection would also therefore take account of the potential size the eligible population.

3. Standard sampling assumptions based on the outcomes from the 2000 survey were used to convert the number of eligible households into a minimum interview yield.

4. Areas with an expected yield of less than 6 were combined with a neighbouring OA to ensure that an interviewer was able to work in each sampled area for three days to screen addresses and make repeated calls at addresses.

5. The remaining areas were sampled with probability proportionate to the number of potentially eligible households and every address within sampled OAs was screened to identify and interview eligible households.

Fieldwork outcomes

The survey achieved its target of just over 1,000 interviews although screening proved more difficult than expected. Compared with the 2000 survey, which reported an individual response of 76%, the response rate for the 2005 survey was 74% when calculated on the same basis.

Table 1 - Fieldwork outcomes comparable with 2000 survey

This suggests that the 2005 survey was, in fieldwork terms, as successful as the 2000 survey, achieving a high proportion of successful interviews in households that were successfully screened. Table 2 shows a more conventional calculation of the response rate, which shows the success rates for the survey at each stage of the household contact and screening process.

Table 2 - Survey response rate with allowance for unsuccessful screening.

This gives a response rate of 77% of adults known to be eligible to participate but it highlights the fact that a number of households could not be screened and their eligibility is therefore not known. It is likely that some of these households would contain 17-39 year old drivers and that the true response rate is lower than 77%. If we assume that these households contained eligible respondents in the same proportion as was found in the other addresses that were screened ( i.e. that 67% would be screened out because they did not contain any 17-39 year olds or did not contain any drivers) the response rate would be 58%.

Survey interviewing

Although all the survey screening was undertaken by MORI interviewers, most of the data collection from respondents involved the person inputting their answers directly into handheld computers in a manner similar to a self-completion survey. The objective of this was to offer respondents as much anonymity as possible. Compared with normal face-to-face interviewing, the principal benefit of this type of self-completion interviewing for respondents is that questions do not need to be read out and answers given verbally. Respondents read the questions and answer them directly. Unlike paper questionnaires, routing is rigorously controlled and respondents' answers 'disappear' into the computer. However, respondents in this age group are unlikely to believe that their answers are truly hidden and so, in spite of reassurances, there remains the possibility that responses are affected by concerns about anonymity and confidentiality. This is discussed in subsequent chapters.

Weighting

There are two types of weighting normally used in surveys: weights to correct for aspects of the sample design and corrective weights to accommodate deficiencies in the achieved sample. In this survey the selection of one adult in households containing more than one eligible respondent needs to be accounted for by weighting. In terms of the sample design, this is the only weight required.

After this weight was applied the sample was analysed both in terms of its geographical distribution and the age and sex structure of the sample. Geographical factors were important since it seems reasonable to expect differences in the prevalence of drug use and driving between individual local authorities, between urban and rural areas and between areas of relative deprivation and affluence. The analysis of geographical variables indicated that the sample required some correction between local authorities and in terms of the urban/rural balance of interviews. In essence, response was better in the more rural areas. Corrective weighting based on 2001 Census data and SHS data was used to remove this imbalance.

In terms of the demographic profile of the sample, the principal concern was that there may be differences in the likelihood of young drivers to respond to the survey, mainly because they are less likely to be home when interviewers call. This type of non-response is potentially a significant source of bias. However, the age profile of 17-39 year old drivers in this survey is not substantially different from the profile of 17-39 drivers in the SHS and it was decided that no further weighting was required.

Table 3 - Weighted and unweighted age profile of 2005 sample compared with 2003/2004 Scottish Household Survey and 2000 Drugs and Driving survey

Follow-up interviews

Qualitative depth interviews were a key part of meeting the study objectives relating to obtaining a more complete understanding of the motivations and circumstances of drug driving.

Follow-up interviews were undertaken with four groups:

• survey respondents who said they had driven while impaired by drugs in the previous 12 months
• problem drug users who were involved with the Drug Outcome Research In Scotland ( DORIS) project operated by the Centre for Drugs Misuse Research.
• survey respondents who had ever driven while impaired by drugs but not in the previous 12 months
• survey respondents who had ever been a passenger of someone driving while impaired by drugs but who had never done so themselves. Some problem drugs users who had been passengers of someone impaired by drugs were also interviewed.

The survey respondents were all people who had agreed to take part in follow-up research as part of the questionnaire. Overall, 74% of survey respondents gave consent to be re-contacted and among the target sub-groups the corresponding rates were 78% for drivers who had ever driven while impaired by drugs, 81% for those who had driven while impaired in the previous year and 70% for passengers of impaired drivers.

The interviews were conducted with respondents according to a topic guide agreed with the project advisory group. Interviews were recorded and transcribed before analysis.

The number of potential respondents was limited by the number of each group identified in the survey. Also, respondents who had recently driven while impaired were prioritised over 'ever' drivers so most difficulty was experienced with this group. At the end of the qualitative fieldwork the target number of interviews had been achieved in all but one group.

DUI 12 months (year) - 15 interviews completed

DUI Ever - 10 interviews completed compared with a target of 15

Problem drug users - 30 interviews completed

Passengers - 20 interviews completed.

« Previous | Contents | Next »