1. Harmonised outputs: where appropriate, tables including marital status, living arrangements, ethnic groups, tenure, economic activity, accommodation type, length of residence and general health have adopted the harmonised output categories described in the publication ‘Harmonised Concepts and Questions for Government Social Surveys’ London: ONS (1996). However, where data from the 2000 survey has been combined with data from earlier years to providesufficient sample size for analysis and where long established time series are shown, harmonised outputs may not have been used.

2. Classification variables: variables such as age and income, are not presented in a standard form throughout the report partly because the groupings of interest depend on the subject matter of the chapter, and partly because many of the trend series were started when the results used in the report had to be extracted from tabulations prepared to meet different departmental requirements.

3. Nonresponse and missing information: the information from a household which co-operates in the survey may be incomplete, either because of a partial refusal (eg to income), or because information was collected by proxy and certain questions omitted because considered inappropriate for proxy interviews (eg marriage and income data), or because a particular item was missed because of lack of understanding or an error. Household and individuals who did not co-operate at all are omitted from all the analyses; those who omitted whole sections (eg marriages) because they were partial refusals or interviewed by proxy are omitted from the analyses of that section. The ‘no answers’ arising from omission of particular items have been excluded from the base numbers shown in the tables and from the bases used in percentaging. The number of ‘no answers’ is generally less than 0.5% of the total and at the level of precision used on GHS the percentages for valid answers are not materially affected by the treatment of ‘no answers’. Socio-economic group and income variables are the most common variables which have too many missing answers to ignore.

4. Base numbers: Very small bases have been avoided wherever possible because of the relatively high sampling errors that attach to small numbers. In general, percentage distributions are shown if the base is 50 or more. Where the base is 20-49, the percentages are shown in square brackets. For some analysis several years data have been combined to increase the sample size to enable appropriate analysis.

5. Percentages: A percentage may be quoted in the text for a single category that is identifiable in the tables only by summing two or more component percentages. In order to avoid rounding errors, the percentage has been recalculated for the single category and therefore may differ by one percentage point from the sum of the percentages derived from the tables. The row or column percentages may add to 99% or 101% because of rounding.

6. Conventions: The following conventions have been used within tables:
.. data not available
- category not applicable
0 less than 0.5% or no observations
[] the numbers in square brackets are percentages on a base of 20-49. See note 4.

7. Statistical significance: Unless otherwise stated, changes and differences mentioned in the text have been found to be statistically significant at the 95% confidence level.

8. Mean: Throughout the report the arithmetic term ‘mean’ is used rather than ‘average’. The mean is a measure of the central tendency for continuous variables, calculated as the sum of all scores in a distribution, divided by the total number of scores.

9. Weighting: All percentages and means presented in the tables in the substantive chapters are based on data weighted to compensate for differential nonresponse. Both the unweighted and weighted bases are given. The unweighted base represents the actual number of people/households in the specified group. The weighted base gives a grossed up population estimate in thousands. Trend tables show unweighted and weighted figures for 1998 to allow direct comparison between 1998 and 2000 and to give an indication of the effect of the weighting. For the weighted data (1998 and 2000) the weighted base (000’s) is the base for percentages. Unweighted data (up to 1998) are based on the unweighted sample.

.