Longitudinal and self‐attributed change in alcohol use among young adults during the COVID‐19 pandemic in Australia

Abstract

Introduction

The COVID‐19 pandemic introduced a unique concern regarding the potential for pandemic‐related increases in alcohol use. However, most studies which have measured pandemic‐related changes to date utilise self‐attribution measures of changes in alcohol use using cross‐sectional designs, which rely on accurate self‐attributions for validity. There has been minimal investigation of correspondence of self‐attributed and longitudinally measured changes in alcohol use during the pandemic. The current study seeks to examine this correspondence.

Methods

A total of 856 participants originally recruited from Australian secondary schools completed follow‐up surveys of an ongoing study at two timepoints (2018–2019, mean age 18.6 and 2020–2021, mean age 19.9; 65.3% female). Alcohol use was measured as any drinking (1+ drinks) and binge drinking (5+ drinks) frequency in the past 6 months. The correspondence and relationship between ‘longitudinal change’ measured from the first to the second timepoint and ‘self‐attributed change’ measured at the second timepoint were examined.

Results

For both any drinking and binge drinking frequency, moderate correspondence was observed between self‐attributed and longitudinal change in drinking (37.1% and 39.3%). Most participants with longitudinal increases in any drinking or binge drinking frequency failed to correctly self‐attribute this increase.

Discussion and Conclusions

The findings suggest that self‐attributed increases do not correspond well with longitudinally measured increases in pandemic‐related drinking and may underestimate increases measured longitudinally. Method of measurement needs to be taken into account if data are to be used to identify sub‐groups at risk of alcohol use increases and facilitate appropriate direction of public health efforts.

Key Points.

• Self‐attribution measures of pandemic‐related increases in alcohol use were found to largely underestimate increases in risky drinking behaviours relative to increases measured longitudinally via self‐report, in a sample of young Australian adults.

• Method of measurement needs to be taken into account if data are to be used to identify sub‐groups at risk of alcohol use increases and facilitate appropriate direction of public health efforts.

• Further study of the accuracy of self‐attribution measures of change in alcohol use is indicated.

1. INTRODUCTION

The COVID‐19 pandemic emerged as a global phenomenon in March 2020, accompanied by government measures aimed at reducing the spread of the virus [1]. Such measures include physical distancing, restrictions on social gatherings and orders to stay or work at home [1]. Restrictions were first implemented in Australia in March 2020 after the first national COVID‐19 case was confirmed in January 2020 [2, 3].

The pandemic intensified certain risk factors for alcohol misuse which are always of concern even outside this context. Poor mental health and stress are known risk factors for the onset of alcohol misuse as well as its maintenance [4]. An acute decline in community mental health under pandemic restrictions has been observed both in Australia and globally [5]. Increased alcohol use generally may contribute to diverse public‐health consequences, such as injuries, relationship damage, medical problems and increased risk of suicide [2, 6, 7]. Given the potential of social conditions such as isolation and financial insecurity to influence alcohol use and thus alcohol‐related harms, obtaining empirical data on alcohol use to understand patterns and harms is imperative for public health [8, 9].

Regarding the COVID‐19 pandemic, a systematic review of studies published during or prior to March 2021 found that in many countries under restrictions, most individuals decreased their alcohol consumption or overall population alcohol consumption decreased [1]. More recent systematic reviews of studies published [10, 11] have found a similar pattern of individual and overall decreases. However, all systematic analyses uncovered a general trend that those who drank heavily or in risky ways prior to the pandemic, or those suffering from mental health problems were more likely to increase their alcohol consumption during pandemic [1, 10, 11]. Young adults are a sub‐group who may be at particularly increased risk of pandemic‐related changes. In 10‐ to 24‐year‐olds, alcohol use is the leading contributor to disability‐adjusted life‐years [12] and in Australia people aged 18–24 reported the highest rates of alcohol‐related harms, binge drinking and drinking compared to age other groups in the population prior to the COVID‐19 pandemic [13] Several studies have reported the largest pandemic‐related drinking increases were observed in young adults, with others finding decreases in this group [1].

To assess changes in alcohol use in a sample of individuals over time, epidemiology may make use of cross‐sectional or longitudinal self‐report surveys [14]. Cross‐sectional measurement of change uses retrospective measures which ask participants to recall consumption over distant time points [14]. The accuracy of self‐report is likely to be reduced the more distant a time period that the participant has to recall [14]. Longitudinal surveys employ repeat assessments of alcohol use which may be spaced between distant points of time, reducing the compromise of accuracy by recall and aiding appraisal of trends in individuals [14]. Longitudinal surveys have their own limitations such as attrition which may influence the representativeness of a sample [14]. However, the advantages of longitudinal measurement, including relying less on recall, and using valid measures of alcohol use with established baselines, tends to recommend this method to accurately assess changes in alcohol use and other outcomes in individuals over time [2, 14, 15].

Therefore, we see this method predominate in practice generally in epidemiological studies assessing change in alcohol use. However, most studies assessing change in alcohol consumption in the context of the pandemic have relied on cross‐sectional assessment with retrospective self‐report measures [1, 2]. These studies asked participants to recall alcohol use prior to the pandemic and during, or in other words to give a self‐attribution of their change in consumption [1, 2]. As mentioned, scarce prospective or longitudinal data increases the potential for mis‐estimation and recall bias in self‐reported measures of change in alcohol consumption [16]. Others have already noted the need for longitudinal data which accurately assesses alcohol use prior to the pandemic [2, 12]. However, in the absence of such data, understanding the degree of accuracy of self‐attributed change in drinking relative to longitudinally measured change is a key methodological question. This question underpins the validity of current measurement of pandemic‐related changes in alcohol use.

Minhas et al. [16] investigated this correspondence. Their study found that self‐attributed change in drinking across the pandemic did largely correspond to change as measured by longitudinal data on the same participants [16]. However, their sample was relatively small, and somewhat limits generalisability to the wider population as the eligibility criteria required a history of heavy drinking [16] The current study seeks to add to the limited empirical evidence of correspondence between self‐attributed and longitudinal measures of alcohol consumption during the pandemic. The analysis has a specific focus on frequency of any drinking and binge drinking occasions, given their prevalence and impact among young people [16]. It examines this question in a large sample of young adults in Australia both prior to and during the COVID‐19 pandemic and associated restrictions. There was no eligibility criterion of alcohol use problems in the sample, unlike in Minhas et al. [16]. It is expected, in line with Minhas et al. [16], that there will be a general correspondence between self‐attribution and longitudinal measures.

2. METHODS

2.1. Ethics approval and consent

The data and sample used derived from the CLIMATE schools combined (CSC) long‐term follow‐up study, which was approved by the Human Research Ethics Committees of the University of Sydney (2018/906), the University of Queensland (2018002638/HREC/2018/906), Curtin University (HR92/2013) and Deakin University (2018‐103) [17]. The CSC study is registered with the Australian and New Zealand Clinical Trials registry, ACTRN12613000723785 and informed consent was obtained from participants for the original CSC study conducted in 2014 and for follow‐up surveys [17, 18].

2.2. Participants and procedure

The participants derived from an original cohort of 6386 year eight students originating from 71 schools in Australia (New South Wales, Queensland, Western Australia), in the 2014 CSC study [17]. Mean age at baseline was 13.5 (SD = 0.6), 54.8% were female and 81.2% were born in Australia [17]. The CSC study participants were followed up in 2018–2019 (T8, mean age 18.6) and again in 2020–2021 (T9, mean age 19.9). Participants were invited to complete a comprehensive follow‐up survey by email, Facebook, phone or school alumni mail‐outs [17].

The survey included items on alcohol use, alcohol‐related harms and mental health [18]. The wave 9 follow‐up sample consisted of 1877 participants [19]. For final analysis in the current study, we excluded participants without answers to all relevant survey items leaving a final sample of 856 participants. The current study and analysis were conducted at the Matilda Centre in Sydney, Australia.

2.3. Measures

Using a six‐point scale, T8 and T9 participants were asked about the frequency with which they consumed at least one standard alcoholic drink and, separately, five or more standard alcoholic drinks on one occasion in the past 6 months (never, less than monthly, once a month, 2–3 times a month, weekly, daily/almost daily). To assess participants' self‐attributed change in alcohol use from T8 to T9, two self‐attribution questions were added to the CSC survey at T9. Using a five‐point scale, participants were asked how frequently they are drinking now compared to before the COVID‐19 pandemic (a lot less frequently, a bit less frequently, at about the same frequency, a bit more frequently, a lot more frequently). One question referred to consuming at least one drink and the other referred to consuming five or more drinks on one occasion. These variables were used to reflect self‐attributed change in frequency of any drinking and binge drinking, respectively.

2.4. Statistical analysis

We reported descriptive statistics on gender, age and all drinking variables. We constructed two variables representing longitudinal change in frequency of drinking (any and binge frequency) by subtracting T8 frequency from T9 frequency for each participant. Any magnitude of change from T8 to T9 in terms of response options in the six‐point items was collapsed into ‘increase’ if the longitudinal change variable was positive, ‘decrease’ if it was negative and ‘no change’ if it was equal to zero. This resulted in a three‐category longitudinal change variable (‘no change’, ‘decrease’, ‘increase’). Similarly, the self‐attributed change items were collapsed into three‐category change variables, with any magnitude of self‐attributed increase or decrease in frequency from T8 to T9 coded as ‘increase’ or ‘decrease’, and self‐attribution of no change coded as ‘no change’. Longitudinal change was cross‐tabulated against self‐attributed change. In order to assess the correspondence between longitudinal and self‐attributed change three separate positive predictive values (PPV) with 95% confidence intervals (CI) were calculated. The first indexed the degree to which any longitudinal change in drinking frequency (either increase or decrease) was correctly self‐attributed as a change. The second indexed the degree to which a longitudinal increase in drinking frequency was correctly self‐attributed as an increase versus no change. The third indexed the degree to which a longitudinal decrease was correctly self‐attributed as a decrease versus no change.

The relationship between magnitude of longitudinal increase and the accuracy of self‐attribution of change was further explored. The hypothesis was that those who correctly self‐attributed change in drinking frequency would have larger longitudinal change (i.e., they would have changed by a greater number of response options on the frequency scale) compared to those who incorrectly self‐attributed change. To enable this, binary variables of correct or incorrect self‐attribution of increase and decrease were constructed for frequency of any drinking and binge drinking. In addition, corresponding binary variables representing the magnitude of longitudinal increase and decrease were constructed, with a one‐level change in the frequency scale defined as a ‘small’ magnitude change and a greater than one‐level change defined as a ‘large’ magnitude change. Chi‐squared tests of association were used to test this relationship.

IBM SPSS V28 was used [20]. For all statistical comparisons, p < 0.05 was regarded as significant. Analyses were not pre‐registered and should be considered exploratory.

3. RESULTS

3.1. Participant demographic and drinking characteristics

Tables 1 and 2 show the degree and patterns of change between T8 and T9 in any drinking and binge drinking frequency. Of the 856 participants 65.3% were female, 79% had consumed any alcohol at least once a month in the 6 months prior to the T9 survey, with the most common frequency of drinking being weekly drinking (31.4%; see Table 1). Binge drinking at T9 was common in the sample, with 43.7% consuming at five or more drinks on a single occasion at least once a month (see Table 2).

TABLE 1.

Any drinking frequency at T8 (past 6 months) against any drinking frequency at T9 (past 6 months)

		Any drinking frequency at T9
		Never	Less than monthly	Once a month	2–3 times a month	Weekly	Daily or almost daily	Total
Any drinking frequency at T8	Never	3	27	10	4	5	0	49
	Less than monthly	11	69	34	28	9	1	152
	Once a month	0	35	29	38	18	3	123
	2–3 times a month	1	28	36	130	81	3	279
	Weekly	1	3	14	52	149	16	235
	Daily or almost daily	0	0	1	2	7	8	18
	Total	16	162	124	254	269	31	856

Note: T8, CLIMATE schools participants combined followed up in 2018–2019; T9, CLIMATE schools participants combined followed up in 2020–2021.

TABLE 2.

Binge drinking frequency at T8 (past 6 months) against binge drinking frequency at T9 (past 6 months)

		Binge drinking frequency at T9 (N)
		Never	Less than monthly	Once a month	2–3 times a month	Weekly	Daily or almost daily	Total
Binge drinking frequency at T8 (N)	Never	121	71	10	9	2	0	213
	Less than monthly	36	156	56	37	11	0	296
	Once a month	6	54	38	31	10	0	139
	2–3 times a month	7	29	26	57	29	1	148
	Weekly	1	1	9	14	33	2	60
	Daily or almost daily	0	0	0	0	0	0	0
	Total	171	311	138	148	85	3	856

Note: T8, CLIMATE schools participants combined followed up in 2018–2019; T9, CLIMATE schools participants combined followed up in 2020–2021.

3.2. Longitudinal change versus self‐attributed change

Table 3 shows the correspondence between categories of longitudinal and self‐attributed change in any drinking frequency. Overall, moderate agreement was observed, with 318 of 856 total participants (37.1%) demonstrating correspondence between longitudinal and self‐attributed change in any drinking frequency. Self‐attributions of increase in those with a longitudinal increase in any drinking frequency were predominantly inaccurate, with only 58 of 277 participants (20.9%) correctly self‐attributing an increase (PPV 39.7%, 95% CI 34.5–44.9%). Self‐attributions of no change in those with no longitudinal change were relatively more accurate, with 135 of 388 participants (34.8%) correctly self‐attributing no change (PPV = 34.8%, 95% CI 31.6–38.0%). The category of longitudinal change which demonstrated the most accurate self‐attributions was longitudinal decrease, in which 125 of 191 participants (65.9%) correctly self‐attributed a decrease in any drinking frequency (PPV 72.7%, 95% CI 68.8–76.6%).

TABLE 3.

Self‐attributed change in any drinking frequency by categories of longitudinal change in any drinking frequency

		Longitudinal change in any drinking frequency across T8 and T9
		Decrease, N (%)	No change, N (%)	Increase, N (%)	Total, N (%)
Self‐attributed change in any drinking frequency across T8 and T9	Decrease, N (%)	125 (65.5)	196 (50.5)	131 (47.3)	452 (52.8)
	No change, N (%)	47 (24.6)	135 (34.8)	88 (31.8)	270 (31.5)
	Increase, N (%)	19 (9.9)	57 (14.7)	58 (20.9)	134 (15.7)
	Total	191 (100)	388 (100)	277 (100)	856 (100)

Note: Percentages (%) in cells represent column percentages. Bold represent those with correct self‐attribution of change, that is, self‐attribution of change corresponded with longitudinal change. T8, CLIMATE schools participants combined followed up in 2018–2019; T9, CLIMATE schools participants combined followed up in 2020–2021.

The observed patterns for binge drinking frequency were largely consistent with those for any drinking frequency (Table 4). Overall, there was moderate agreement between longitudinal change and self‐attributed change in binge drinking frequency, with 336 of 856 total participants (39.3%) demonstrating correspondence. Only 39 of 269 participants (14.5%) correctly self‐attributed an increase in binge drinking frequency when a longitudinal increase occurred (PPV 26.2%, 95% CI 21.5–30.8%). Self‐attributions of no change in those with no longitudinal change were relatively more accurate, with 177 of 405 participants (43.7%) correctly self‐attributing no change (PPV 43.7%, 95% CI 40.4–47.0%). Greatest accuracy was seen in self‐attributions of decrease in binge drinking frequency, with 120 of 182 participants (65.9%) correctly self‐attributed a decrease in binge drinking when a longitudinal decrease occurred (PPV 70.6%, 95% CI 66.8–74.4%).

TABLE 4.

Self‐attributed change in binge drinking frequency by categories of longitudinal change in binge drinking frequency

		Longitudinal change in binge drinking frequency across T8 and T9
		Decrease, N (%)	No change, N (%)	Increase, N (%)	Total, N (%)
Self‐attributed change in binge drinking frequency across T8 and T9	Decrease, N (%)	120 (65.9)	210 (51.9)	120 (44.6)	450 (52.5)
	No change, N (%)	50 (27.5)	177 (43.7)	110 (40.9)	337 (39.4)
	Increase, N (%)	12 (6.6)	18 (4.4)	39 (14.5)	69 (8.6)
	Total	182 (100)	405 (100)	269 (100)	856 (100)

Note: Percentages (%) in cells represent column percentages. Bold represent those with correct self‐attribution of change, that is, self‐attribution of change corresponded with longitudinal change.

3.3. Relationship between magnitude of longitudinal increase and self‐attribution of increase

The following section of analysis only focuses on those with longitudinal change in drinking (i.e., those who did not change over time are excluded). Of the 58 participants that correctly self‐attributed an increase in any drinking frequency, 37.9% reported a longitudinal increase that was large in magnitude (i.e., a jump of two or more response options on the frequency scale). For the 219 participants that failed to correctly self‐attribute an increase in any drinking frequency (i.e., they said their drinking frequency stayed the same or decreased), 26.9% reported a longitudinal increase that was large in magnitude. The difference between these percentages was not statistically significant (X ²(1269) = 2.68, p = 0.10). There was, however, a relationship between correct self‐attributions of decrease in any drinking frequency and the magnitude of longitudinal decrease (X ²(1269) = 6.34, p = 0.01).

The same broad pattern was observed when considering correct self‐attributions of increases and decreases in binge drinking frequency. However, in neither case was there a statistically significant association between correct self‐attributions of change and the magnitude of change (increase: X ²(1269) = 1.66, p = 0.20; decrease: X ²(1269) = 1.11, p = 0.29). In summary, the proportions of large magnitude drinking increase and decrease were greater amongst those who correctly self‐attributed an increase and decrease, respectively, yet statistically significant differences were only observed with respect to a decrease in any drinking frequency.

4. DISCUSSION

The current study explored the correspondence between longitudinal change in drinking from before to the COVID‐19 pandemic and related restrictions of 2020, and self‐attributed change in a sample of young adult Australians. For both drinking frequency variables, a significant relationship was found between longitudinal change and self‐attributed change. However, overall, there was only moderate agreement between longitudinal change and self‐attributed change. Participants predominantly failed to correctly perceive their longitudinal change or lack thereof. We also explored the effect of the magnitude of longitudinal change on whether a change in either variable was correctly self‐attributed. Although large magnitude changes (i.e., changes of two or more response options on the frequency scale) made up a slightly greater proportion of correctly self‐attributed increases in either measure, no significant relationship between magnitude of longitudinal increase and correct self‐attribution of increase was found. However, compared to those who incorrectly self‐attributed a decrease in any drinking frequency, those who correctly self‐attributed a decrease had a greater magnitude of longitudinal decrease. In other words, people are more accurate at correctly identifying decreases in their drinking over time than they are at identifying increases in their drinking over time.

The finding of moderate rather than strong correspondence between self‐attributions of change and longitudinal change departs from the correspondence found by Minhas et al. [16]. In Minhas et al. [16], longitudinal increases, no change and decreases all corresponded well to self‐attributed increases, no change and decreases respectively. In the current study, relatively strong correspondence between self‐attributed and longitudinal change was found in those who had a decrease in drinking, while the weakest correspondence was found in those who had a longitudinal increase in drinking. A potential explanation for this is that those with higher rates of drinking frequency tend to provide underestimates of their drinking behaviour in retrospective measures, thus there may be a negative influence of risky drinking on accuracy of self‐attributions of increase [10, 14]. It is noted, however, that Minhas et al. [16] had an eligibility criterion of risky drinking for their sample, and nevertheless found large correspondence between self‐attributed and longitudinal change measures.

The current study is subject to limitations. Self‐report measurement is subject to limitations to accuracy and validity including recall bias and social desirability bias in responding which tend to result in underestimation of alcohol use [21]. Similarly, alcohol use among young adults fluctuates over sometimes short periods of time and thus, fluctuations in alcohol use may reflect changes that would have been observed under non‐pandemic conditions. Furthermore, the alcohol use measures utilised at the two time‐points to create a measure of longitudinal change still relied on some degree of recall of behaviour over 6 months. Such measurement may be less accurate than more prospective measurement via daily diaries or other measures involving daily recording of alcohol consumption. The sample was self‐selected and therefore can be considered a convenience sample. Finally, given the study was conducted across different states in Australia with varying timelines for COVID‐19 related restrictions, participants were asked to self‐attribute change in drinking ‘now compared to before the COVID‐19 pandemic’ rather than given specific reference dates. Noting these considerations, the strength of the current study is that both the longitudinal data and the retrospective measure of self‐attributed change are comparable to those used in longitudinal studies and cross‐sectional studies of COVID‐19 pandemic‐related alcohol change [1, 2]. That these measures are present concurrently and applied to one sample of individuals allows important comparison between longitudinal change and self‐attribution, only investigated in one other study to date [16].

Regarding the sample, there was an over‐representation of females in the current sample (65.3%) compared to the general population in this age group [13]. Also, the proportion of males dropped from 45% at baseline in 2014 to 35% to follow‐up surveys in 2018–2021, indicating a non‐response bias [22]. Also, this was a sample of convenience from the 2014 CSC study. 43.7% of the present sample engaged in monthly single occasion risky drinking compared to 41% of 18‐ to 24‐year‐old Australians [13]. The gender composition and the fact that the sample is of convenience and from three Australian states limit generalisability of empirical findings to the wider population.

5. CONCLUSIONS

The finding that longitudinal increases in binge drinking frequency did not correspond well to self‐attributed increases is important. This means that certain recent studies on pandemic‐related changes in alcohol use, which utilise retrospective self‐attribution measures, may be missing or underestimating increases in alcohol consumption relative to measurement via longitudinal cohort studies. The related possibility that at‐risk sub‐groups may go unidentified has implications for public health. Targeted intervention and prevention efforts, clinical service provision, and health messaging may be directed towards those most at risk of increased alcohol consumption [4, 8]. Thus, it is crucial that alcohol use measures identify such individuals, which may not be the case currently. Further study should seek to add to the empirical evidence on the validity of self‐attribution measures of alcohol consumption relative to longitudinal measures in the context of the pandemic or other social events which necessitate short‐term solutions to uncovering empirical evidence of alcohol use changes. Heterogeneity observed in this correspondence in comparison to Minhas et al.'s [16] previous findings ought to be investigated, as there may be sub‐groups that consistently contribute to concordance or discordance between measures.

AUTHOR CONTRIBUTIONS

Each author certifies that their contribution to this work meets the standards of the International Committee of Medical Journal Editors.

FUNDING INFORMATION

The CSC long‐term study was funded by a National Health and Medical Research Council project grant (APP1143555) and Australian Rotary Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

CONFLICT OF INTEREST

There is no conflict of interest connected to this paper.

ETHICS STATEMENT

The data and sample used derived from the CLIMATE schools combined long‐term follow‐up study, which was approved by the Human Research Ethics Committees of the University of Sydney (2018/906), the University of Queensland (2018002638/HREC/2018/906), Curtin University (HR92/2013) and Deakin University (2018‐103).

Dolli I, Slade T, Teesson M, Chapman C. Longitudinal and self‐attributed change in alcohol use among young adults during the COVID‐19 pandemic in Australia. Drug Alcohol Rev. 2023;42(3):625–632. 10.1111/dar.13602