Inaccuracies in survey reporting of alcohol consumption

Here, we investigate whether the inclusion of the standard drink term and pictorial guide is associated with an adjustment in self-reported alcohol consumption among two cohorts of university/college students in New South Wales, Australia, and the US Pacific Northwest. The aims of the study were to: 1) establish the extent to which providing a definition of a standard drink led to within-subject differences in reporting of alcohol consumption; and 2) explore factors associated with differences in reporting of alcohol consumption. Both aims are considered from a cross-national perspective comparing samples from Australia and the USA in a convenience approach based on the researchers’ institutional affiliations and access to student populations. It is reasonable to expect differences in the cultural norms between the US and Australia given different tertiary education structures, as well as drinking, secondary supply and alcohol purchase laws.

Participants were recruited from one university in New South Wales (NSW), Australia, one in Washington, USA, and a college in Oregon, USA (Pacific Northwest sample). Recruitment materials invited students aged 18 years and above who drink alcohol to complete an anonymous, online survey.

In NSW, students were recruited using convenience, non-probability sampling through posts on the course management sites of courses across the university as well as through student networks such as residents groups. No compensation was offered for completion. Given the potential that content associated with alcohol consumption would be covered in some programs, participants were recruited from a range of faculties and programs. In Australia there is no legal drinking age per se, but the age for the legal purchase of alcohol is 18 years.

In Oregon, students at a small, liberal arts college were recruited via posters placed on campus and online posts. In Washington, students were recruited from two campuses of a large, public university. Students were recruited from the undergraduate psychology research pool including any student enrolled in a psychology course (not necessarily psychology majors). Students could select this study from a list and were compensated for completion via course credit. The legal drinking age in the USA is 21 years of age; however, illicit drinking is common among university students under the legal age of consumption [24].

Procedures and measures were identical for both cohorts apart from the use of country-specific standard drinks images. The survey contained three blocks of questions, with slightly different versions used in Australia and the US Pacific Northwest to account for different standard drinks definitions and drinking cultures. In Australia, the standard drink size is that which contains 10 g of pure alcohol, while the US equivalent is 14 g of pure alcohol. The first block contained three questions from the AUDIT-C (see https://​www.​integration.​samhsa.​gov/​images/​res/​tool_​auditc.​pdf for a reference guide) accompanied by an Australian Standard Drinks Guide or equivalent US image (see Fig. 1 a) and b)). The three questions were (1) “How often do you have a drink containing alcohol,” (2) “How many standard drinks do you have on a typical day when drinking,” and (3) “How often do you have six or more standard drinks on one occasion.” A second block acted as a diffuser, and the final block utilised a modified version of the AUDIT-C (mAUDIT-C), which asked the AUDIT-C questions, but replaced the term ‘standard drink’ with either ‘drinks’ or ‘serves’ and removed the pictorial guide. Blocks one and three were randomised for each participant. 50.5% of respondents saw the term ‘serves’ and 49.5% saw the term drinks. Participants were not allowed to return to previous question blocks.

Alcohol consumption volume arising from the questions in this block was reported as an ordinal variable, with respondents reporting the number of [standard] drinks consumed on a typical day as 1 or 2, 3 or 4, 5 or 6, 7 to 9, or 10 or more. The mid-point of these responses was taken as the number of drinks reported (1.5, 3.5, 5.5, 8, 12). We created a dichotomous variable to measure adjustment of alcohol consumption volume based on whether the number of ‘standard’ drinks reported was significantly different from the number reported in response to questions about drinks or serves only. This measure was used as the dependent variable in logistic regression analyses.

The second block of questions included items about self-reported drinking status (non-, occasional-, light-, party/social-, heavy-, or ex-drinker), the type of alcohol usually consumed (with ten options as well as an ‘other’ option provided along with locally relevant examples, e.g. ordinary [full strength] beer, low alcohol beer, wine, wine coolers, champagne or sparkling wine, alcoholic cider, alcoholic sodas, premixed spirits, spirits, liqueurs including premixed liqueurs), the places where the respondent usually drinks (In my own/spouse’s/partner’s home, at a friend’s house, at a party at someone’s house, at raves/dance parties, at restaurants/cafés, at licensed premises (e.g. pubs, clubs), at University, at my workplace, in public places (e.g. parks, beaches), in a car or other vehicle, or other), and a series of items relating to drinking intentions (thinking of all the times that you drink alcohol, how often would you drink for the following reasons …) and alcohol-related harms (e.g., being sick after drinking, violence, unwanted sexual activity). These items were selected based on the likelihood they would be associated with reported volumes, and participants were able to select multiple options.

One hundred twenty-two participants completed the survey in NSW and 285 in the Pacific Northwest USA. The participant demographics and differences between cohorts are summarised in Table 1. A majority of participants in both cohorts identified as female, the average age was 21 (20.98; SD = 3.47; Australian sample slightly older; p < 0.001), and most were in the first three to four years of an undergraduate program. The Australian sample were predominantly in the first two years of a program, while US students were more commonly in their third or fourth year (p < 0.001). Australian students were enrolled in 23 different programs, with Electrical Engineering, Oral Health, Pharmacy, and alternate entry programs most common. US students came from 28 different programs with Civil Engineering, Computer Systems Engineering, Professional Engineering, Energy Studies, or Linguistics most common.

In both cohorts, participants most commonly referred to themselves as party/social drinkers (51.5 and 68.8% of Australian and US cohorts, respectively), or occasional drinkers (33.6 and 23.5% of Australian and US cohorts, respectively). Age of initiation to drinking ranged from 9 to 21 years (mean 16.67; SD 1.82) and was not significantly different between cohorts. The types of drinks ‘usually’ consumed were spirits (56.0%), full strength beer (35.9%), wine or champagne (50.1%), pre-mixed drinks/alcoholic sodas (34.15%), and cider (25.1%) with differences between the cohorts in beverage selection (see Table 1). The most commonly identified locations of drinking were home/friend’s home/house party followed by licensed premises, rave parties, university, and cafes.

The average AUDIT-C score (based on the validated instrument using the standard drink term and image) was 4.53 and did not differ significantly between the groups. Forty-three percent of Australian and 44% of US respondents scored above 4 on the AUDIT-C, indicating possible hazardous drinking. In both samples, males were more likely to score above 4 than females (Australia male:female = 54.05%: 51.81%; US male:female = 68.18%: 60.27%). For a small number of women (n = 6 Australian and n = 4 US), all four points were from the first question which is suggestive of lower risk drinking.

Continuous variables were reported as mean (SD), and categorical variables as counts (%). Continuous variables were compared by means of the two-sample Student’s t-test for independent samples with equal and unequal variances (as appropriate), and ordinal variables were compared using the Pearson Chi-squared test and the Fisher’s Exact test (as appropriate). Paired data were compared using the McNemar-Bowker test. Due to the cross-over design, the effects of carryover, period and sequence were assessed and found to be non-significant. The analysis of response adjustments was modelled using Firth logistic regression, without the presence of these effects. Separate models for Australian and USA students were built using a two-stage process. First, univariate analyses were performed to determine the strength of association between independent variables and adjustment of drinking volume. Independent variables included gender, age, number of standard drinks consumed on a typical drinking day, frequency of consumption, frequency of consuming more than six standard drinks, type of alcohol consumed, places where alcohol is typically consumed, alcohol-related harms and alcohol expectancies. Second, variables with a p-value < 0.10 in the univariate analyses were included in the multivariable model. Firth’s logistic regression analysis was used to overcome the computational limitations and convergence issues caused by the sparseness (separation of the data [25]). Model validity was evaluated by performing the Hosmer-Lemeshow goodness-of-fit test and the Pregibon link test [26] on each model. A p-value< 0.05 by the two-tailed test was considered to be statistically significant. All analyses were performed using Stata Version 15 (StataCorp LP, College Station, TX). Post-hoc analysis was performed with the US sample age-stratified to assess any potential differences between under-age respondents and those legally able to purchase and consume alcohol.

Differences in reporting of alcohol consumption associated with the presentation of the term and guide for standard drinks were assessed. In the Australian sample, participants reported consuming an average of 5.41 (SD = 3.49) ‘standard’ drinks on a typical drinking day, and reports for drinks and serves were 5.16 (SD = 3.46) and 5.76 (SD = 3.84) respectively. In the US sample, the reported number of standard drinks was 6.58 (SD = 1.95), drinks was 4.75 (SD = 2.13), and serves was 4.65 (SD = 2.23). There were no statistically significant differences between reports of drinks or serves in either sample, so these were grouped in analysis. When reported standard drinks were compared with reported drinks/serves combined, there was almost no difference in the Australian sample (0.01, 95% CI: − 0.30 to 0.31, p = 0.977) while the values were significantly different in the US sample (1.88, 95% CI: 1.58 to 2.17, p < 0.001; Table 2). Differences were maintained in the US sample stratified for age (< 21 years versus 21 and over).

Frequency of consumption differed for reports of standard drinks, and drinks/serves across the entire sample (Table 3) No significant differences were found between reports of drinks and serves, so these were grouped. For standard drinks and drinks/serves, frequency also differed between the cohorts. Australian students reported more frequent consumption of standard drinks, with fewer reporting drinking monthly or less (25.0% vs 53.0%), and more drinking 2–4 times per month (40.8% vs 31.2%), 2 or 3 times per week (25.8% vs 12.3%), and 4 or more times week (8.3% vs 3.5%) compared to US students. Frequency of drinking six or more standard drinks, drinks, or serves did not differ significantly between cohorts or terms (not shown). Differences in frequency reports were statistically significant when comparing standard drinks and drinks/serves for US students (p < 0.001) but not for Australian students (p = 0.083). In the US sample, this difference was larger among students who were under 21 years of age, compared to ‘of age’ drinkers (p < 0.0001; not shown).

Ninety-one respondents (80.5% of valid responses) in the Australian sample and 63 respondents (22.1%) in the US sample made no adjustment to their responses of typical number of drinks consumed. The difference in adjustment between age groups in the stratified US sample was not significant, though there was a trend towards smaller adjustment among those aged less than 21 (Table 2). In the Australian sample, multivariable logistic regression of factors potentially associated with a lack of adjustment found that there was no association between age, gender, number of standard drinks/drinks/serves on a typical day, frequency of drinking, or frequency of consuming six or more drinks and the probability of adjustment (Table 4). Further, none of the secondary predictors (types of drinks consumed, reasons for drinking, harms associated with drinking and the locations of drinking) were significantly associated with adjustment at the univariate level and were therefore excluded from the regression model. The model with primary predictors only was found to have good fit (p = 0.675) and be appropriately specified (p = 0.846).

In the US model (Table 4), significant overall effects were found for the number of drinks/serves/standard drinks on a typical day (p = 0.003) and the frequency of drinking in past 3 months (p = 0.003) in predicting adjustment of survey response. Those who reported drinking more in terms of both frequency and volume were less likely to adjust their consumption. Having planned to get drunk was found to have a significant positive association with the probability of adjustment (p = 0.02). The model had good fit (p = 0.092) and was appropriately specified (p = 0.365.)