Annex 1: Using the information in this report
HOW DATA IS DISPLAYED IN TABLES
Tables are generally presented in the format 'dependent variable by independent variable' where the independent variable is being used to examine or explain variation in the dependent variable. Thus, a table titled 'housing tenure by household type' shows how housing tenures vary among different household types. Tables generally take three forms within the report; column percentages (with dependent variable in the rows), row percentages (dependent variable in the columns) and cell percentages which may show agreement or selection of a statement with one or a number of statements.
All tables have a descriptive and numerical base showing the population or population sub-group examined in it. While all results have been calculated using weighted data, the bases shown provide the unweighted counts, which have been rounded to the nearest 10 to comply with statistical disclosure control principles and the Code of Practice for Official Statistics. It is therefore not possible to calculate how many respondents gave a certain answer based on the results and bases presented in the report.
In general, percentages in tables have been rounded to the nearest whole number. Zero values are shown as a dash (-), values greater than 0 per cent but less than 0.5 per cent are shown as 0 per cent and values of 0.5 per cent but less than 1 per cent are rounded up to 1 per cent. Columns or rows may not add to exactly 100 per cent because of rounding, where 'don't know/refused' answers are not shown or where multiple responses to a question are possible.
In some tables, percentages have been removed and replaced with '*'. This is where the base on which percentages would be calculated is less than 50 and this data is judged to be insufficiently robust for publication.
VARIATIONS IN BASE SIZE FOR TOTALS
As the questionnaire is administered using computer assisted personal interviewing (CAPI), item non-response is kept to a minimum. Bases do fluctuate slightly due to small amounts of missing information (where, for example, the age or sex of household members has been refused and where derived variables such as household type use this information).
Some questions are asked of a reduced sample and the bases are correspondingly lower. From January 2012, the redesigned survey asked questions typically of full or one-third sample allocation. This concept of streaming was first introduced to the SHS in 2007, when some questions were streamed or changed in the course of the year and again the base size is lower. Further changes to streaming have been made in subsequent years.
The sample base annex (Annex 3) gives details of frequencies and bases for the main dependent variables.
All proportions produced in a survey have a degree of error associated with them because they are generated from a sample survey of the population rather than a survey of the entire population (e.g. Census). Any proportion measured in the survey has an associated confidence interval (within which the 'true' proportion of the whole population is likely to lie), usually expressed as ±x per cent. As a general rule of thumb, the larger the sample size for a given question, the smaller the confidence interval around that result will be (thus making it easier to detect real change year-on-year and differences between sub-groups.
It is possible with any survey that the sample achieved produces estimates that are outside this range. If the survey were to be run multiple times on the same population in the same year (i.e. under repeated sampling), the number of times out of a 100 surveys that the result achieved would be expected to lie within the confidence interval is also quoted; conventionally the level set is 95 out of 100, or 95 per cent. Technically, all results should be quoted in this way however, it is less cumbersome to simply report the percentage as a single percentage, the convention adopted in this report.
Where sample sizes are small or comparisons are made between sub-groups of the sample, the sampling error needs to be taken into account. There are formulae to calculate whether differences are statistically significant (i.e. they are unlikely to have occurred by chance) and Annex 4 provides a simple way to calculate whether differences are significant. Annex 4 also provides further explanation on statistical significance and on how confidence intervals can be interpreted.
Email: Andrew Craik