Date Published: January 31, 2017
Publisher: Public Library of Science
Author(s): Sadie Boniface, Shaun Scholes, Nicola Shelton, Jennie Connor, C. Mary Schooling.
Previous studies have shown heavier drinkers are less likely to respond to surveys and require extended efforts to recruit. This study applies the continuum of resistance model to explore how survey estimates of alcohol consumption may be affected by non-response bias in three consecutive years of a general population survey in England.
Using the Health Survey for England (HSE) survey years 2011–13, number of contact attempts (1–6 and 7+) were explored by socio-demographic and drinking characteristics. The odds of drinking more than various thresholds were modelled using logistic regression. Assuming that non-participants were similar to those who were difficult to contact (the continuum of resistance model), the effect of non-response on measures of drinking was investigated.
In the fully-adjusted regression model, women who required 7+ calls were significantly more likely to drink more than the UK Government’s recommended daily limit (OR 1.19, 95% CI 1.06–1.33, P = 0.003) and to engage in heavy episodic drinking (OR 1.23, 95% CI 1.07–1.42, P = 0.004), however this was not significant in men in the fully-adjusted model. When the continuum of resistance model was applied, there was an increase in average weekly alcohol consumption of 1.8 units among men (a 12.6% relative increase), and an increase of 1.5 units among women (a 20.5% relative increase). There was also an increase in the prevalence of heavy episodic drinking of 2.5% among men (an 12.0% relative increase) and of 2.0% among women (a 15.8% relative increase), although other measures of drinking were less affected.
Overall alcohol consumption and the prevalence of heavy episodic drinking were higher among HSE participants who required more extended efforts to contact. The continuum of resistance model suggests non-response bias does affect survey estimates of alcohol consumption.
Non-response bias—where study participants’ answers to questions differ systematically from the potential answers of people who did not agree to take part in a study—has long been recognised as a limitation of surveys that measure alcohol consumption . Heavy drinkers are thought to be more difficult to contact and less likely to respond to surveys . One model that explores how non-response bias may influence survey estimates of alcohol consumption is the ‘continuum of resistance’ model. This assumes that non-participants (non-responders) are similar to each other, and that if they took part in the survey, they would be most similar to the participants who were the most difficult to recruit . The justification is that participants who required many attempts to contact would have been non-participants if the data collection had been stopped earlier, therefore non-participants are more similar to these participants than those who are interviewed after fewer attempts . Better understanding of non-response bias may allow for non-response weights in surveys to be improved. It has also been suggested that it may be preferential to use the continuum of resistance model which takes into account the amount of time and effort required to elicit a response (after Lin and Schaeffer ) to adjust for non-response than population weights based on demographics and geographical data , which do not necessarily eliminate the bias associated with non-response , and may even compound it.
Key demographic characteristics are shown by number of contact attempts in Table 2. There was significant variation in the proportion of participants who required 7+ contact attempts to elicit a response by age, economic activity, income, deprivation and region.
Using the assumptions of the continuum of resistance model to further adjust for potential non-response bias in survey estimates of alcohol consumption, there was a 12.6% relative increase in average weekly alcohol consumption (absolute increase of 1.8 units a week) in men and a 20.5% relative increase in women (absolute increase of 1.5 units a week). There was also a 12.0% relative increase in the prevalence of heavy episodic alcohol consumption (absolute increase of 2.5%) in men and a 15.8% relative increase among women (absolute increase of 2.0%). Other measures of drinking were less affected. This study provides evidence that non-response bias differentially affects survey estimates of alcohol consumption in a nationally-representative sample in England.