Each survey involved interviewing a large stratified probability sample of the general population, covering people living in households in England. It doesn’t include those who live in institutional settings (such as prisons or hospitals) or in temporary housing (such as hostels or bed and breakfasts) or those who sleep rough. The full age range was covered, with the youngest participants aged 16 and the oldest over 100. There were 7,546 respondents to the survey.
The questions covered the following topics:
- alcohol consumption (relating to how often a person drinks, how much they drink, and how often they drink heavily)
- alcohol-related harm (including feelings of guilt or remorse after drinking, blackouts, alcohol-related injury, and other concerns about alcohol consumption)
- symptoms of alcohol dependence (including a person’s difficulty in controlling how often or how much they drink, an increase in the importance of drinking to them, and morning drinking)
Initial questions about alcohol consumption were asked face to face by an interviewer. All participants who drank alcohol, even if just occasionally, were then asked to complete the remaining alcohol use questions. These were administered using computer-assisted self-completion interview (CASI), consistent with the approach used on the 2000 and 2007 surveys.
The resulting statistics for problem drinking have been age-standardised. This is because the prevalence of problem drinking is related to age and the age profile (the number of people of different ages within an ethnic group) can differ considerably between ethnic groups. This adjustment allows comparisons to be made between ethnic groups as if they had the same age profile.
If a participant could not take part in a long interview due to a physical or mental health condition, the interviewer recorded information about this on the doorstep. This information may be biased because often, the interviewer spoke to another member of the household.
Socially undesirable or stigmatised feelings and behaviours may be underreported. This is a risk for any study based on self-report data, but this study goes some way to minimising this by collecting particularly sensitive information in a self-completion format.
Some people selected for the survey could not be contacted or refused to take part. The achieved response rate was 57%. Weighting helps take account of those who were selected for the survey but didn’t take part. Weighting is used to adjust the results of a survey to make them representative of the population and improve their accuracy. For example, a survey which contains 25% women and 75% men will not accurately reflect the views of the general population, which we know has an even 50/50 split. Statisticians rebalance or ‘weight’ the survey results to more accurately represent the general population. This helps to make them more reliable. In this specific survey, weighting also helps correct the bias produced by non-respondents. Statistically significant tests have been conducted, using a regression model. If a significant association is found with an independent variable then post hoc tests are used to identify significant differences between individual levels of that variable.
Confidence Intervals Confidence intervals for each ethnic group for the detailed risk levels with AUDIT scores of 8 or above,are available in the ‘download the data’ section.
Confidence intervals provide a measure of the uncertainty associated with survey data. For example, 2.2% of White British people surveyed were drinking at harmful or mild dependence levels (AUDIT score between 16 and 19) in England 2014. This is a reliable estimate of the percentage of people who were drinking at this level. However, because the APMS results are based on a random sample of people aged 16 or older, it is impossible to be 100% certain of the true percentage. It is 95% certain, however, that somewhere between 1.8% and 2.7% of all White British people were drinking at harmful or mild dependence levels. In statistical terms, this is a 95% confidence interval. This means that if 100 random samples were taken, then 95 times out of 100 the estimate would fall in this range (ie, between the upper and lower confidence interval). But 5 times out of 100 it would fall outside this range.
The smaller the survey sample, the more uncertain the estimate and the wider the confidence interval. For example, fewer people from the Black ethnic group were sampled for this survey than British White people, so we can be less certain about the estimate for this smaller group (1.6% of all Black people). This greater uncertainty is expressed by the wider confidence interval for the percentage of Black people drinking at harmful or mild dependence levels of 0.6% and 3.9%.
Suppression rules and disclosure control
Individuals’ information has been kept confidential by:
- analysing people’s drinking levels by ethnicity and gender with no further breakdowns
- using broad categories for ethnicity and making the groupings of drinking levels sufficiently large
- maintaining large underlying populations for analysis in the survey
No further suppression or other disclosure control has been applied.
Percentages have been rounded to one decimal point.Quality and methodology information