Scottish Government

ANNEX B - TECHNICAL DETAILS

The Scottish Social Attitudes series

1. The Scottish Social Attitudes ( SSA) survey was launched by the Scottish Centre for Social Research ³¹ (part of the National Centre for Social Research) in 1999, following the advent of devolution. Based on annual rounds of interviews with 1,500-1,600 people drawn using random probability sampling (based on a stratified, clustered sample) ³², its aims are to facilitate the study of public opinion and inform the development of public policy in Scotland. In this it has similar objectives to the British Social Attitudes ( BSA) survey, which was launched by the National Centre in 1983. While BSA interviews people in Scotland, these are usually too few in any one year to permit separate analysis of public opinion in Scotland (see Park, et al, 2004 for more details of the BSA survey).

2. SSA is conducted annually and has a modular structure. In any one year it will typically contain four or five modules, each containing 40 questions. Funding for its first two years came from the Economic and Social Research Council, while from 2001 onwards different bodies have funded each year's individual modules. These bodies have included the Economic and Social Research Council, the Scottish Executive and various charitable and grant awarding bodies, such as the Nuffield Foundation and Leverhulme Trust.

The 2007 survey

3. The 2007 survey contained modules of questions on:

• attitudes to government and public services in post-devolution Scotland (funded by the Scottish Government Office of Chief Researcher from 2004-2007)
• attitudes to drinking alcohol and the role of alcohol in Scottish culture (funded by the Scottish Government)
• the 2007 Scottish Parliament and Local Government elections, in collaboration with John Curtice at the University of Strathclyde, David McCrone and Nicola McEwen at University of Edinburgh and Michael Marsh at Trinity College Dublin (funded by the Leverhulme Trust and the Economic and Social Research Council), and
• views on the funding, provision and delivery of public services in Scotland, funded as part of a suite of surveys (including modules on the 2007 British Social Attitudes survey, Northern Ireland Life and Times and a stand alone survey in Wales) under the ESRC's public services programme.

4. Findings from the 2007 Scottish Government modules will be available in reports published by the Government, while a programme of dissemination and publication is planned for each of the other modules. This technical annex is designed to accompany Scottish Government reports. It covers the methodological details of the 2007 survey as well as further discussion of the analysis techniques used in the reports.

Response rates

5. The Scottish Social Attitudes survey involves a face-to-face interview with respondents and a self-completion questionnaire, completed by around nine in ten of these people (87% in 2007). The numbers completing each stage in 2007 are shown in Table 1. See Bromley, Curtice and Given (2005) for technical details of the 1999-2004 surveys, Given and Ormston (2006) for details of the 2005 survey and Cleghorn, Ormston and Sharp (2007) for the 2006 survey.

Table 1: 2007 Scottish Social Attitudes survey response

Notes to table
The table shows a 'lower' and an 'upper' response rate. The former is calculated on the assumption that all addresses whose eligibility to participate was unknown were in fact eligible to take part. The latter is calculated on the assumption that they were all ineligible (because they were empty/derelict, non-residential, etc). The 'true' response is likely to lie somewhere between the two, since some addresses whose eligibility was unknown are likely to have been 'deadwood' while others may have been eligible. See Lynn et al (2001) ³³ for a discussion of treatment of unknown eligibility in calculating response rates.
1 This includes empty / derelict addresses, holiday homes, businesses and institutions.
2 'Unknown eligibility' includes cases where the address could not be located, where it could not be determined if an address was a residence and where it could not be determined if an address was occupied or not. For the lower response rate, this is shown as a % of 'in scope' addresses. For the upper response rate, it is shown as a % of issued addresses, since these addresses are excluded from 'in scope' for the purposes of calculating the upper response rate.
3 Refusals include refusals prior to selection of an individual, refusals to the office, refusal by the selected person, 'proxy' refusals made by someone on behalf of the respondent and broken appointments after which a respondent could not be re-contacted.
4 Non-contacts comprise households where no one was contacted after at least 4 calls and those where the selected person could not be contacted.
5 'Other non-response' includes people who were ill at home or in hospital during the survey period, people who were physically or mentally unable to participate and people who with insufficient English to participate.

Sample design

6. The survey is designed to yield a representative sample of adults aged 18 or over living in Scotland. The sample frame is the Postcode Address File ( PAF), a list of postal delivery points compiled by the Post Office. The detailed procedure for selecting the 2007 sample was as follows:

I. 134 postcode sectors were selected from a list of all postal sectors in Scotland, with probability proportional to the number of addresses in each sector for addresses in urban areas and a probability of twice the address count for sectors in rural areas (i.e. the last 3 categories in the Scottish Government's 6 fold urban-rural classification). Prior to selection the sectors were stratified by region, percentage of household heads recorded as being in non-manual occupations ( SEG 1-6 and 13, taken from the 2001 Census) and the Scottish Government classification of urban and rural areas ³⁴ (see below for a description of this).

II. Once 134 sectors had been selected, half this postcode sector was selected at random (to reduce travel time for interviewers working in very large postcode sector areas).

III. In 2007, the number of addresses selected from each sampled postcode sector varied based on known non-response, to try and ensure that the achieved sample matched the geographic spread of the population (after taking account of rural over-sampling) as closely as possible, and to try and ensure that interviewers working in different types of areas were able to achieve similar numbers of interviews from their sample batches. This approach has recently been used on the Scottish Household Survey.

IV. Interviewers called at each selected address and identified its eligibility for the survey. Where more than one dwelling unit was present at an address, all dwelling units were listed systematically and one was selected at random using a computer generated random selection table. In all eligible dwelling units with more than one adult aged 18 or over, interviewers also had to carry out a random selection of one adult using a similar procedure.

Weighting

7. The weights applied to the SSA 2007 data are intended to correct for three potential sources of bias in the sample:

I. Differential selection probabilities

II. Deliberate over-sampling of rural areas

III. Non-response.

8. Data were weighted to take account of the fact that not all households or individuals have the same probability of selection for the survey. For example, adults living in large households have a lower selection probability than adults who live alone. Weighting was also used to correct the over-sampling of rural addresses. Differences between responding and non-responding households were taken into account using information from the census about the area of the address as well as interviewer observations about participating and non-participating addresses. Finally, the weights were adjusted to ensure that the weighted data matched the age-sex profile of the Scottish population (based on 2006 mid-year estimates from GROS).

9. Prior to the 2005 dataset, SSA data was only weighted to take account of differential selection probabilities and over-sampling in rural areas. The decision to introduce non-response weighting and 'calibration' weighting to match the sex-age profile of the population was taken following experimentation with the 2004 British Social Attitudes ( BSA) dataset. Both BSA and SSA weights now incorporate these new elements, which are designed to reduce non-response bias.

10. When reporting time-series analysis, there is of course a small possibility that changes to the weighting scheme could disrupt the results and suggest changes that would not have been found had the old weighting scheme been used. The SSA 2007 dataset therefore included a variable based on the old weighting scheme, and for any time-series reporting the 2007 figures were rerun using the old weighting structure to ensure that this did not present a radically different picture. However, unless otherwise specified, in this report all percentages are weighted using the new weighting scheme described above. The unweighted sample sizes are shown in the tables.

Fieldwork

11. Fieldwork ran between late May and early November 2007 (with 84% completed by the end of August). An advance letter was sent to all addresses and was followed up by a personal visit from a Scottish Centre for Social Research interviewer. Interviewers were required to make a minimum of 4 calls at different times of the day (including at least one evening and one weekend call) in order to try and contact respondents, although in practice interviewers often made many more calls than this. All interviewers attended a one day briefing conference prior to starting work on the study.

12. Interviews were conducted using face-to-face computer-assisted interviewing (a process which involves the use of a laptop computer, with questions appearing on screen and interviewers directly entering respondents' answers into the computer). All respondents were asked to fill in a self-completion questionnaire which was either collected by the interviewer or returned by post. Table 1 summarises the response rate and the numbers completing the self-completion in 2007.

Analysis variables

13. A number of standard analyses have been used in the five reports. Most of the analysis variables are taken directly from the questionnaire and to that extent are self-explanatory. These include age, sex, household income, and highest educational qualification obtained. The main analysis groups requiring further definition are set out below.

The Scottish Government six-fold urban-rural classification (2005/6)

14. The six categories used in analysis are: 1) large urban, 2) other urban, 3) small accessible towns, 4) small remote towns, 5) accessible rural, 6) remote rural. For more details see Hope, S. et al (2000).

National Statistics Socio-Economic Classification ( NS-SEC)

15. The most commonly used classification of socio-economic status used on government surveys is the National Statistics Socio-Economic Classification ( NS-SEC). SSA respondents were classified according to their own occupation, rather than that of the 'head of household'. Each respondent was asked about their current or last job, so that all respondents, with the exception of those who had never worked, were classified. The seven NS-SEC categories are:

• Employers in large organisations, higher managerial and professional
• Lower professional and managerial; higher technical and supervisory
• Intermediate occupations
• Small employers and own account workers
• Lower supervisory and technical occupations
• Semi-routine occupations
• Routine occupations

The remaining respondents were grouped as 'never had a job' or 'not classifiable'.

Scottish Index of Multiple Deprivation ( SIMD)

16. The Scottish Index of Multiple Deprivation ( SIMD) ³⁵ 2006 measures the level of deprivation across Scotland - from the least deprived to the most deprived areas. It is based on 37 indicators in seven domains of: Current Income, Employment, Health, Education Skills and Training, Geographic Access to Services (including public transport travel times for the first time), Housing and, new for 2006, Crime. SIMD 2006 is presented at data zone level, enabling small pockets of deprivation to be identified. The data zones are ranked from most deprived (1) to least deprived (6,505) on the overall SIMD 2006 and on each of the individual domains. The result is a comprehensive picture of relative area deprivation across Scotland.

17. The analysis in this report used a variable created from SIMD data indicating the level of deprivation of the data zone in which the respondent lived in quintiles, with 1 being the most deprived and 5 being the least deprived. ³⁶

Analysis techniques

Regression

18. Logistic regression models are used to assess whether there is reliable evidence that particular variables are associated with each other.

19. Regression analysis aims to summarise the relationship between a 'dependent' variable and one or more 'independent' explanatory variables. It shows how well we can estimate a respondent's score on the dependent variable from knowledge of their scores on the independent variables. This technique takes into account relationships between the different independent variables (for example, between education and income, or social class and housing tenure). Regression is often undertaken to support a claim that the phenomena measured by the independent variables cause the phenomenon measured by the dependent variable. However, the causal ordering, if any, between the variables cannot be verified or falsified by the technique. Causality can only be inferred through special experimental designs or through assumptions made by the analyst. All regression analysis assumes that the relationship between the dependent and each of the independent variables takes a particular form.

20. The reports on the 2007 Scottish Social Attitudes Core Module use logistic regression - a method that summarises the relationship between a binary 'dependent' variable (one that takes the values '0' or '1') and one or more 'independent' explanatory variables. The tables in this report show how the odds ratios for each category in significant explanatory variables compares to the odds ratio for the reference category (always taken to be 1.00).

21. Taking Model 1 (below) as an example, the dependent variable is based on whether or not people had a below average score on the 'life satisfaction' question (one of the five subjective well-being measures included in the SSA 2007). If the respondent gave their satisfaction with their life as whole a score between 0 and 7 on the 11-point scale, the dependent variable takes a value of 1. If not, it takes a value of 0. An odds ratio of above 1 means that, compared with respondents in the reference category, respondents in that category have higher odds of having a below average score for life satisfaction. Conversely, an odds ratio of below 1 means they have lower odds of thinking this than respondents in the reference category. If we look at self-perceived financial hardship, we can see people who said they were 'coping', 'living comfortably' or 'very comfortably' on their household income all have odds ratios of less than 1, indicating that they have lower odds of having a below than average score on 'life satisfaction' than those in the reference category (i.e. people who say they are finding it 'difficult' or 'very difficult' to cope on household income).

22. The significance of differences between the reference category and other categories are indicated by 'P'. A p-value of 0.05 or less indicates that there is less than a 5% chance we would have found such a difference just by chance if in fact no such difference exists, while a p-value of 0.01 or less indicates that there is a less than 1% chance. P-values of 0.05 or less are generally considered to indicate that the difference is highly statistically significant, while a p-value of 0.06 to 0.10 may be considered marginally significant. As shorthand to aid interpretation, we have used symbols to summarise statistically significant differences:

• '+' denotes results that are significantly different from 0 at the 10% level (p = 0.06-0.10)
• '*' denotes results that are significant from 0 at the 5% level (p = 0.015 - 0.05) and
• '**' denotes results that are significantly different from 0 at the 1% level (p = 0.01 or below)
• ' NS' denotes results that are not significantly different from the reference category.

23. It should be noted that the final regression models reported below were produced following a process involving several stages of analysis:

1. First, forward stepwise regression analysis was conducted in SPSS 12.0. The variables entered into these initial models are noted below each final model, below.

2. Second, those variables found to be significantly associated with the dependent variable by these SPSS forward stepwise models were entered into final regression models run through STATA. Unlike SPSS 12.0, STATA can account for complex sample designs (in particular, the effects of clustering and associated weighting) when calculating odds ratios and determining significance. The models shown below include only those variables found to be significant after the regression models were run in STATA.

3. In some cases, two models were run for one dependent variable - for example, running a model including demographic factors only in the first instance, then running a second model including significant demographic factors from the first stage plus subjective factors such as self rated health and hardship. Running the analysis in these stages allowed for the exploration of how much each additional set of factors added to the ability to explain the dependent variable. Further, it revealed interesting demographic variations that might have been masked had self rated health and hardship been included in this analysis from the outset.

4. Where more than one model was created for one dependent variable, only the final model has been reported below. These include significant factors after all the various demographic and attitudinal variables listed have been taken into account. Copies of any additional models are available from ScotCen on request.

Regression models

Model 1 Satisfaction with life as a whole

Cases included in model = 1,472
This is the final STATA model, including all significant factors.
Independent variables included in initial forward stepwise models in SPSS:
Demographic factors: sex, age, education, socio-economic classification ( NS-SEC), economic activity, income, marital status, children in household, long term illness, area deprivation ( SIMD quintiles), Scottish Government urban rural classification, Strathclyde area or not.
Subjective factors: self rated health, self perceived hardship, social trust

Model 2 Happiness

Cases included in model = 1,479
This is the final STATA model, including all significant factors.
Independent variables included in initial forward stepwise models in SPSS:
Demographic factors: sex, age, education, socio-economic classification ( NS-SEC), economic activity, income, marital status, children in household, long term illness, area deprivation ( SIMD quintiles), Scottish Government urban rural classification, Strathclyde area or not.
Subjective factors: self rated health, self perceived hardship, social trust

Model 3 Satisfaction with job

Cases included in model = 819
This is the final STATA model, including all significant factors.
Independent variables included in initial forward stepwise models in SPSS:
Demographic factors: sex, age, education, socio-economic classification ( NS-SEC), economic activity, income, marital status, children in household, long term illness, area deprivation ( SIMD quintiles), Scottish Government urban rural classification, Strathclyde area or not.
Subjective factors: self rated health, self perceived hardship, social trust.

Model 4 Standard of living

Cases included in model = 1,441
This is the final STATA model, including all significant factors.
Independent variables included in initial forward stepwise models in SPSS:
Demographic factors: sex, age, education, socio-economic classification ( NS-SEC), economic activity, income, marital status, children in household, long term illness, area deprivation ( SIMD quintiles), Scottish Government urban rural classification, Strathclyde area or not.
Subjective factors: self rated health, self perceived hardship, social trust

Model 5 Satisfaction with 'family or personal life'

Cases included in model = 1,483
This is the final STATA model, including all significant factors.
Independent variables included in initial forward stepwise models in SPSS:
Demographic factors: sex, age, education, socio-economic classification ( NS-SEC), economic activity, income, marital status, children in household, long term illness, area deprivation ( SIMD quintiles), Scottish Government urban rural classification, Strathclyde area or not.
Subjective factors: self rated health, self perceived hardship, social trust