Scotland's New Ethnicity Classification for Scottish Official Statistics and Recommended for Scotland's 2011 Census

4. Constraints for Classification Development

4.1 The development of the classification has, in part, been restricted by a number of constraints. These preclude some options outright and limited the viability of others. The constraints are outlined below.

Specific census constraints

4.2 Question layout. Completed census forms (from approximately two million households, including over five million people) must be scanned electronically before data can be analysed. The layout of the form and its appearance, are standardized across questions to maximise scanning efficiency. The appearance of the new classification adheres to the standard layout.

4.3 Form space. Space on the census form is limited, with strong competition for a variety of different questions. Scotland's 2011 Census will have 4 pages of individual questions per person (2 columns per page). GROS will give one full column to the ethnicity classification. This is more space than any other question, and more than in the 2001 Census. But it still limits the number of tick boxes which can be fitted into the classification. There is no space (in this or any other question) for notes to help the respondent understand the question, which must therefore be self-explanatory.

4.4 UK harmonisation. As detailed in paragraphs 2.9 to 2.13, UK census offices have agreed to adhere to the principle of UK harmonisation, where possible, so statistics can be produced for GB and the UK. However, a balance had to be struck between meeting the needs of UK harmonisation and also the specific circumstances of each UK country. For example, Scotland has a different ethnic profile and the findings of Scotland's evidence-base differed from the rest of the UK in some cases.

Survey and census constraints

4.5 Data outputs. The classification must meet the main information needs of data users and users must be able to apply it in practice. The ethnic groups outputted from the classification must be meaningful and manageable (in number). In part, this determined the number of tick boxes which were included (now 21, previously 14). It precludes an exhaustive list of ethnic groups or a single write-in box for respondents to self-identify in any way they wish.

4.6 Single or multiple responses. If only a single response to the question is allowed, information can be provided about 21 ethnic groups (plus write-in responses), which is a useable total and gives a fuller breakdown than the 14 groups provided for in the 2001 classification. If people were allowed to tick as many boxes as they wished, there would be hundreds, and possibly thousands, of different combinations which would be extremely difficult to output or use meaningfully. Multiple responses may cause people of the same ethnicity to identify using different categories. This makes it difficult to obtain a robust single count of each ethnic group (a key requirement of data users). This is considered in more detail in paragraphs 5.13 to 5.20.

4.7 Small and disclosive numbers. Official Statistics and the census must avoid very small numbers being published in a way which is disclosive ( i.e. allows individuals to be identified, breaking the confidentiality of their data). Small numbers can also prevent robust cross-tabulations of the ethnicity question with other questions on the census (or in surveys), for example, ethnicity analysed by educational attainment or occupation etc (a key requirement of data users). Providing an exhaustive list of ethnic groups or allowing multiple ethnic group combinations to be outputted (from multiple responses), increases the likelihood of very small numbers in some of the data.

4.8 Response errors. The classification must try to minimise inaccurate responses or errors to ensure high quality data. Three main inaccuracies or errors are possible: multiple responses, not using the best or most appropriate category and crossing out an initial response then giving another response lower down the question. In such a long question such as this one, with several sub-sections, these errors can be reduced by labelling each sub-section ( e.g. A 'White', B 'Asian' etc) and a prominent completion instruction to give a single response also helps (as detailed in section 5 on page 24 onwards). Previous testing by ONS also found that people with 'mixed or multiple ethnicities' are less likely to multi-tick or not notice their category if it is positioned near the top of the classification rather than at the bottom.

Specific methodological constraints

4.9 Length of the classification. To reflect the diversity of Scotland's ethnic profile, a lot of categories and tick boxes are needed. So the classification needs to be very long - much longer than most statistical classifications. Because of this, respondents need help (visual cues) to guide them through it. Without these cues, testing has shown that a significant proportion of respondents do not read the whole question and fail to find the best response category for them (often using a less good category). So a visual cue is necessary. Labelling categories 'A' to 'E' was shown to be an effective visual cue to perform this methodological function (this is explained in more detail in paragraphs 5.21 to 5.27)

4.10 Double banking. As space on the census form is limited, some questions save space by positioning tick boxes side by side rather than in a longer list (called 'double banking'). It is recognised methodologically, that when tick boxes are double banked, some respondents are less likely to see (and so miss) the tick boxes to the right hand side. This is less problematic for questions where respondents make concrete responses, e.g. a 'yes' or 'no' as they will search for the appropriate response. However, this is less likely for questions where there is choice or many options. This applies to the ethnic group question and so double banking has been avoided.

User and stakeholder constraints

4.11 Historical comparability. Consultation found that, in general, data users recognise the need for a modernised classification but require a degree of comparability with the previous classification. Users need to compare ethnicity data over time to monitor the effectiveness of their policies to promote equality and tackle discrimination. A balance had to be struck between this comparability requirement and ensuring that the classification reflects Scotland's current (and future) ethnic profile and that it is acceptable to people filling in the census (and other surveys or questionnaires).

4.12 Consultation coverage. The classification applies to all ethnic groups in Scotland, but equally it causes specific concerns for certain communities and stakeholders. A balance had to be struck between gathering a broad range of views about ethnicity (using large public consultations) and consulting specific organisations, users and stakeholders about their concerns (using seminars, correspondence, meetings etc). It was not possible for SG and GROS to consult every stakeholder in Scotland; however a very broad range of views were gathered and fed into the development of the classification.

4.13 Balancing viewpoints. Stakeholders and data providers expressed a wide variety of views about many aspects of ethnicity. Some of these views were polarised or opposing, for example the acceptability of colour terms (particularly 'Black') and classification of 'Sikh' and 'Jewish' as ethnic groups. Where views were opposing, it was sometimes necessary to adopt compromise positions as a way of moving forward. It was also necessary to consider all the points of view put forward (sometimes conflicting), whether or not these were expressed by many or few people, and to strike an appropriate balance.