PreprintPDF Available

The Australian ready-to-drink beverages tax missed its target age group

Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

During 2008 and 2009, the Australian Government increased the tax on ready-to-drink alcoholic beverages by 70% to discourage drinking among adolescents. To evaluate the tax, we use the difference-in-difference and comparative interrupted time series estimators, where age is used to define the control and treatment groups. This methodology is applied to the Household Income and Labour Dynamics in Australia survey. We show that the tax did not affect the alcohol consumption of those aged under 25 (the tax targeted age group) but substantially reduced drinking among those aged 25–69, reducing their average daily consumption of standard drinks by 8.9% from 2010 to 2018. The age group under 25 did not respond to the tax likely because of product substitution. Alcohol price policy may need to acknowledge complex substitute/complement relationships between beverages and consider a floor price on alcohol or a uniform volumetric tax per standard drink.
Content may be subject to copyright.
The Australian ready-to-drink beverages tax
missed its target age group
This document is a post-print; please cite the published version:
Alexeev, S., Weatherburn, D. “The Australian ready-to-drink beverages tax
missed its target age group. International Journal of Drug Policy 95 (2021):
Sergey Alexeev
Donald Weatherburn
National Drug and Alcohol Research Centre,
University of New South Wales
22-32 King Street Randwick, NSW 2052, Australia
September, 2021
Background: During 2008 and 2009, the Australian Government increased
the tax on ready-to-drink alcoholic beverages by 70% to discourage drink-
ing among adolescents.
Methods: To evaluate the tax, we use the difference-in-difference and com-
parative interrupted time series estimators, where age is used to define the
control and treatment groups. This methodology is applied to the House-
hold Income and Labour Dynamics in Australia survey.
Results: We show that the tax did not affect the alcohol consumption of
those aged under 25 (the tax targeted age group) but substantially reduced
drinking among those aged 25–69, reducing their average daily consump-
tion of standard drinks by 8.9% from 2010 to 2018.
Conclusion: The age group under 25 did not respond to the tax likely be-
cause of product substitution. Alcohol price policy may need to acknowl-
edge complex substitute/complement relationships between beverages and
consider a floor price on alcohol or a uniform volumetric tax per standard
Keywords: Australia; alcopops; ready-to-drink beverages; drinking.
In the early 2000s, there were growing concerns in Australia that ready-to-
drink (RTD) beverages were encouraging a culture of binge drinking by the
young people (e.g., Australian Drug Foundation 2004; Jones and Barrie 2011).
On 28 March 2008, former Prime Minister Kevin Rudd announced a ‘National
Binge Drinking Strategy’ to address what he described as a binge-drinking
epidemic among young Australians (Prime Minister of Australia 2008). A key
part of the strategy involved a 70% increase in the tax on RTDs. This raised
the price of RTDs from around $39.00 per litre to around $66.00 per litre of
alcohol and brought the excise on RTDs into line with the excise on straight
spirits (Gale et al. 2015). The alcohol industry initially responded by develop-
ing and selling RTDs based on beer and wine instead of spirits (Carragher and
Chalmers 2011). In late 2009 the Rudd Government responded by closing this
loophole so that beer-based and wine-based products that mimic spirit-based
RTDs were taxed equivalently.
Several studies have since been conducted to assess the impact of the
RTD tax on sales and alcohol consumption (Chikritzhs et al. 2009; Doran
and Digiusto 2011; Gilmore et al. 2020; Hall and Chikritzhs 2011; Mojica-
Perez, Callinan, and Livingston 2020). They leave little doubt that the tax
reduced the consumption of alcohol1(although for an exception, see: Kisely et
al. (2011)), however, most also conclude that the tax was effective in reducing
RTD consumption among the young people. Most reach this conclusion on the
basis that alcohol consumption (or its proxy measures) among the young de-
clined after the tax was introduced (Gale et al. 2015; Lensvelt et al. 2016). The
discovery that alcohol consumption among the young fell after the RTD tax
increase, however, is not strong evidence that the tax increase reduced alco-
hol consumption. Several western countries have reported declines in alcohol
consumption, including the United Kingdom (Health and Social Care Informa-
tion Centre 2018), the United States (Johnston et al. 2019), Sweden (Norstr¨
and Svensson 2014), New Zealand (Jackson et al. 2017) and Canada (Elgar,
Phillips, and Hammond 2011).
The study reported here has two major advantages over earlier research
in this area. Firstly we use individual-level rather than aggregate data. Our
measures of alcohol consumption and control variables are drawn from the
Household Income and Labour Dynamics in Australia (HILDA) survey, a na-
tionally representative panel study conducted annually since 2001. Secondly,
we employ more robust methods than past research to examine the effect of the
RTD tax on drinking by young people. We employ the difference-in-difference
(DD) and comparative interrupted time series (CITS) estimators where we use
the age group 70+ as the control group. National Drug Strategy Household
Survey (NDSHS) consistently shows that this age group does not drink RTDs
(Australian Institute of Health and Welfare 2008,2011; Srivastava and Zhao
2010), and the RTD producers tailor their marketing efforts to make RTDs par-
ticularly appealing to a younger consumer (Copeland et al. 2007; Gates et al.
1Figure S1 in Supplementary Material section Bplots the Australia national annual drink-
ing trends for 2002-2018 and shows a visible reduction in drinking starting 2010.
2006; Mosher and Johnsson 2005).2
The HILDA survey data used in the study consists of 16 waves of the survey
from 2002 until 2018 that are pooled together. Approximately 23,000 individ-
uals are sampled in each wave. Wave 2001 has slightly different coding for
drinking and is, therefore, excluded. The HILDA dataset consistently reports
drinking intensity and drinking frequency for all years. We recoded these two
variables into one variable of average daily consumption of standard drinks
(about 10 grams of alcohol). Abstinence is coded as zero.
The HILDA dataset also includes measures of smoking status, age, house-
hold post-government income, and state of residence. In order to look at the
income of household members on a comparable basis, we follow the OECD rec-
ommendations (OECD 2020) and equivalize the income measure by dividing
it by the square root of the household size.3The drinking is heavily skewed to
the right (heavy drinking is relatively rare). In addition, most of the values for
drinking are less than 1. We apply the inverse hyperbolic sine function (IHSF)
to drinking and income to stabilize the skewed data for linear data modelling
and prevent the loss of sample size. Except for minimal values, the IHSF is
identical to a standard log transformation, but it is defined for values close
to zero and performs better than the known alternative transformation tech-
niques (Burbidge, Magee, and Robb 1988; MacKinnon and Magee 1990).4
Difference-in-difference estimator
Our ability to estimate the impact on drinking using the DD estimator hinges
on the comparison group following the common trend over time. If this is the
case, we can use post-intervention outcomes to infer what would happen to
the treatment group without tax.5The conventional discrete DD specification
implemented using ordinary least square (OLS) regression provides no sense
of the dynamics of the tax increase: how quickly consumption reduced and
whether it accelerates, stabilizes, or mean reverts. To explore these dynamics,
we augment DD with leads and lags. Specifically, we add indicator variables for
each year of the survey, keeping 2009, the last pre-treatment year, as the omit-
ted category.6Because the dependent variable is effectively log-transformed
2Supplementary Material section Cuses NDSHS to demonstrate directly that the group
70+ does not drink RTDs.
3Table S1 in Supplementary Material section Breports mean and quantile values of vari-
ables used in the analysis.
4The IHSF is defined as sinh1(𝑥) := ln(𝑥+1 + 𝑥2). Its derivative is 1
/1+𝑥2which if 𝑥is
not too small approximates 1
/𝑥, the derivative of ln(𝑥).
5Figure S1 in Supplementary Material section Bplots the Australia national annual drink-
ing trends for 2002-2018 overall and by different age groups to confirm that common trend
assumption is satisfied.
6Not omitting one of the years leads to a dummy variable trap. The trap can be avoided by
dropping a constant from the model, but it complicates the interpretation of the coefficient.
and 2009 is the omitted category, the estimated coefficients can be interpreted
as the average annual per cent change in drinking relative to 2009.
Interrupted time series estimator
The DD estimator ignores the possibility that the tax may have different effects
on different age groups. The age group of 24 and younger is of particular
interest as the intention of the tax was to reduce their alcohol consumption.
That is why we also estimate the CITS model. This model can be seen as
a generalization of DD framework (Hallberg et al. 2018): a generalization that
accommodates multiple treatment groups with different pre-treatment trends.
More specifically, we follow Markowitz (2018) and Wong, Cook, and Steiner
(2015) and modify CITS to accommodate two treatment groups. This allows
us to use the respondents with age 70+ as the controls, while at the same time
having two treatment groups (the age groups 15–24 and 25–69) with different
pre-treatment trends. While the CITS specification relaxes the assumption of
common trends, it imposes the assumption of trends linearity. This regression
specification allows the tax effect to be reflected in a level shift and a shift in
the trend.7
Control variables
We add four controls to our DD and CITS models. They are individual smoking,
household income, interaction with the state of New South Wales (NSW) and
linear cohort effects. Explanations for each are provided below.
We control for smoking because, starting in 2013, the Australian Govern-
ment started to increase the tax on tobacco products gradually (Scollo and
Bayly 2020). Since alcohol and tobacco could be complementary goods (Decker
and Schwartz 2000; Tauchmann et al. 2013), the estimates for the year 2013
onward might be picking up some of the tobacco tax effects. The smoking
dummies partial some of this effect out. The variance of the control variables
is partialled out with year-specific parameters, allowing for a change in the
variables’ composition from year to year. The year dependence of the smok-
ing dummies accounts for the fact that some people who started or stopped
smoking during the study period because of the anti-smoking campaign also
changed their drinking behaviour.
We control for income as the Global Financial Crisis (GFC) has the potential
to influence our estimates. Theoretically, the control group 70+ nets out the
GFC’s influence, as this factor is common to all age groups. However, the older
control group might have been more resistant to the GFC shock due to larger
savings. An ideal control variable to equalize the treatment and control vari-
able would be the amount of assets. The HILDA dataset reports respondents’
total assets only in selected years, which does not meet our data requirements.
Household post-government income includes all recorded sources of family in-
come and available for all years.
Choosing to drop the pre-treatment year from an array of indicators (i.e., assigning zeros for
this year) is a standard practice in a dynamic DD setting.
7Supplementary Material section Eprovides formal exposition of econometric models.
Our models also account for various legislative reforms on drinking initiated
in NSW in 2008–2014 (Men´
endez, Tusell, and Weatherburn 2015; Roth and
Angus 2014). To account for the contemporaneous but spatially separated
anti-drinking reforms in NSW, we add a dummy for the residency in NSW to
our treatment variable and report treatment effects when the NSW dummy is
switched off. In addition, we also interact the year varying control variables
with this NSW dummy. This also allows for a change in composition in control
variables between NSW and other states.
A weak spot of the estimation design based on age-specific drinking prefer-
ences is the interaction between the age-based structure of our treatment and
control groups and the passage of time. For example, the birth cohort aged
62–70 belongs to the treatment group in 2010 but, due to the passage of time,
belongs to the control group in 2018. This would not be a problem if all birth
cohorts followed similar lifetime drinking patterns. However, a sizable body of
research shows that younger birth cohorts tend to drink less throughout their
lifetime (Callinan et al. 2020; Kraus et al. 2015; Livingston et al. 2016; Looze
et al. 2015; Meng et al. 2014). This migration of younger birth cohorts between
treatment and control groups that systematically drink less on average may be
interpreted as a violation of the stable unit treatment assumption. To control
for the observation that younger birth cohorts drink less, we add linear cohort
For all models, we use standard errors clustered at the household level and
a robust covariance matrix. We choose this level of clustering in response to
a growing concern regarding the correct level of standard error clustering in
policy evaluation (Abadie et al. 2017; Cameron and Miller 2015; Conley and
Taber 2011; MacKinnon 2019). We believe our clustering level is optimal be-
cause it follows the HILDA sampling methodology, which is centred around
households (Watson and Wooden 2020; Wooden, Freidin, and Watson 2002).
Our clustering choice accounts for the error term’s autocorrelation over dif-
ferent years for the same household, as well as for the similarities in drinking
within households between spouses9and across generations.10
Table 1reports the estimates of the overall effect of the RTD tax from 6 speci-
fications. Columns (1) reports estimates when no additional modifications are
added. The subsequent columns show results for each modification. Finally,
column (6) includes all modifications.
The coefficients on the tax leads (2002–2008) are zero, showing no evidence
of a violation of the common trend assumption (the key identifying assumption
for DD framework). To confirm the impression that the assumption is satisfied,
we also report the 𝑝-values of the null hypothesis that coefficients for 2002–
8Supplementary Material section Dshows that each cohort from 1970 drinks less than
the one before, and the change is linear.
9See, for example, Agrawal et al. (2006) and Howe et al. (2019) for the evidence of assortative
mating on drinking.
10See Alexeev (2021) and Windle and Windle (2012), among many others, for the evidence
of intergenerational transmission of drinking.
Table 1: The DD estimates of the effect of the RTD tax on overall drinking,
HILDA 2002-2018.
Dependent variable (DV):
IHSF of average daily consumption of standard drinks (10g of alcohol)
(1) (2) (3) (4) (5) (6)
2002 ×Treated -0.00432 -0.00335 -0.000739 -0.00721 0.00912 0.0238
(0.0212) (0.0214) (0.0214) (0.0213) (0.0220) (0.0240)
2003 ×Treated -0.00172 -0.00299 0.00434 -0.00450 0.00650 0.0142
(0.0210) (0.0211) (0.0211) (0.0210) (0.0219) (0.0237)
2004 ×Treated -0.00223 0.000629 0.00336 -0.00439 0.0115 0.0236
(0.0213) (0.0215) (0.0215) (0.0214) (0.0223) (0.0240)
2005 ×Treated -0.00408 -0.00692 -0.00154 -0.00530 0.00823 0.0187
(0.0212) (0.0214) (0.0216) (0.0213) (0.0222) (0.0242)
2006 ×Treated -0.00539 -0.00600 -0.00198 -0.00597 0.0127 0.0207
(0.0211) (0.0211) (0.0213) (0.0211) (0.0219) (0.0236)
2007 ×Treated 0.00750 0.00705 0.0129 0.00727 0.0218 0.0290
(0.0213) (0.0215) (0.0215) (0.0214) (0.0222) (0.0240)
2008 ×Treated 0.00491 0.00462 0.0102 0.00504 0.0124 0.0209
(0.0214) (0.0215) (0.0214) (0.0214) (0.0222) (0.0238)
2009 ×Treated (omitted)
2010 ×Treated -0.0128 -0.0159 -0.00844 -0.0124 -0.00368 -0.00776
(0.0211) (0.0213) (0.0212) (0.0212) (0.0219) (0.0234)
2011 ×Treated -0.0386* -0.0394* -0.0376 -0.0382 -0.0257 -0.0195
(0.0195) (0.0196) (0.0196) (0.0195) (0.0202) (0.0214)
2012 ×Treated -0.0560** -0.0563** -0.0564** -0.0552** -0.0392 -0.0351
(0.0195) (0.0196) (0.0196) (0.0195) (0.0202) (0.0214)
2013 ×Treated -0.0880*** -0.0885*** -0.0849*** -0.0868*** -0.0740*** -0.0661**
(0.0194) (0.0195) (0.0197) (0.0195) (0.0201) (0.0217)
2014 ×Treated -0.0867*** -0.0860*** -0.0813*** -0.0851*** -0.0738*** -0.0704***
(0.0193) (0.0194) (0.0194) (0.0193) (0.0200) (0.0211)
2015 ×Treated -0.101*** -0.0998*** -0.0987*** -0.0991*** -0.0894*** -0.0833***
(0.0193) (0.0194) (0.0194) (0.0194) (0.0200) (0.0212)
2016 ×Treated -0.0993*** -0.101*** -0.0955*** -0.0967*** -0.0914*** -0.0898***
(0.0192) (0.0193) (0.0193) (0.0192) (0.0198) (0.0209)
2017 ×Treated -0.104*** -0.102*** -0.101*** -0.101*** -0.0906*** -0.0860***
(0.0192) (0.0193) (0.0193) (0.0192) (0.0198) (0.0209)
2018 ×Treated -0.103*** -0.103*** -0.0989*** -0.100*** -0.0907*** -0.0889***
(0.0192) (0.0193) (0.0193) (0.0192) (0.0198) (0.0210)
𝐻0: 2002 2008 = 0 0.999 0.998 0.996 0.997 0.993 0.958
Individual smoking No Yes No No No Yes
Household income No No Yes No No Yes
Interaction with NSW No No No Yes No Yes
Linear cohort effects No No No No Yes Yes
Observations 226,232 223,979 223,979 226,232 223,979 223,979
Clusters 25,136 25,090 25,136 25,136 25,136 25,090
Adjusted 𝑅20.004 0.009 0.011 0.004 0.010 0.011
Mean of DV 0.574 0.573 0.574 0.574 0.574 0.573
Min of DV 0 0 0 0 0 0
Max of DV 3.260 3.260 3.260 3.260 3.260 3.260
*𝑝 < 0.05, ** 𝑝 < 0.01, *** 𝑝 < 0.001.
2008 are jointly statistically insignificant. We cannot reject the null for all
The tax effect accumulates over time. In 2010 – the first year after the
introduction of the tax – the effects are 1.28% from the most parsimonious
model in column (1) and 0.7% from the model with the most features in column
(6). In the last year, these effects are, respectively, 10.3% and 8.98%.
Table 2reports the estimates of the RTD tax effect on the age group 15–24
and 24–69 with another 6 specifications. This table clarifies that the effects are
Table 2: The CITS estimates of the differential effects of the RTD tax on drink-
ing, HILDA 2002-2018.
Dependent variable (DV):
IHSF of average daily consumption of standard drinks (10g of alcohol)
(1) (2) (3) (4) (5) (6)
Post ×Treated1524 -0.0276 -0.0265 -0.0347 -0.0296 -0.00263 -0.00116
(0.0177) (0.0178) (0.0179) (0.0178) (0.0184) (0.0191)
Post ×Treated2569 -0.0266 -0.0270 -0.0281 -0.0269 -0.0244 -0.0167
(0.0158) (0.0159) (0.0159) (0.0158) (0.0162) (0.0171)
Trend ×Post ×Treated1524 0.000756 0.00273 -0.00223 -0.000765 -0.00256 -0.00588
(0.00317) (0.00318) (0.00350) (0.00349) (0.00333) (0.00376)
Trend ×Post ×Treated2569 -0.0112*** -0.0106*** -0.0110*** -0.0114*** -0.0102** -0.0110**
(0.00308) (0.00310) (0.00313) (0.00310) (0.00318) (0.00337)
Individual smoking No Yes No No No Yes
Household income No No Yes No No Yes
Interaction with NSW No No No Yes No Yes
Linear cohort effects No No No No Yes Yes
Observations 226,230 223,977 226,230 226,230 226,230 223,977
Clusters 25,136 25,090 25,136 25,136 25,136 25,090
Adjusted 𝑅20.009 0.011 0.013 0.009 0.009 0.014
Mean of DV 0.574 0.573 0.574 0.574 0.574 0.573
Min of DV 0 0 0 0 0 0
Max of DV 3.260 3.260 3.260 3.260 3.260 3.260
*𝑝 < 0.05, ** 𝑝 < 0.01, *** 𝑝 < 0.001.
concentrated at the age group 25–69, not at the age group 15–24. Similarly
to Table 1, column (1) reports the most optimistic, whereas column (6) the
most conservative results. The first two rows report the level shifts in drinking
immediately after the tax increase separately by the age group of interest. The
last two rows report the change in trends in the post-treatment period by age
The level shifts reported by CITS models are not statistically significant:
echoing the results of the corresponding DD models. The most conservative
model in column (6) essentially reports a noisy zero for the age group 15–24.
Although not statistically significant, the estimated level shift for the group
25–69 is slightly higher than the corresponding estimate for 2010 reported
by the DD models. This is possibly because the CITS models fit linear trends.
Figure S3 from Supplementary Material section Bshows that the effects follow
the L-shape – stronger at the beginning and weaker at the end – the linear trend
fitted into this shape rotates slightly counterclockwise, which may artificially
increase the shift in the level parameter.
The tax’s accumulated effect is calculated after multiplying the last two
coefficients (annual linear trends) by 9. Column (6) implies that the effect
of the tax from 2010 to 2018 is 9.9%. This is not statistically different from
8.89% reported in column (6) in Table 1. The results are consistent across the
DD and CITS models.
Our aim in this study was to improve on previous efforts to examine the ef-
fectiveness of the RTD tax in curbing drinking by adolescents. The evidence
just presented indicates that the tax did not ultimately change the amount of
alcohol consumed by them. Instead, it reduced consumption by middle-aged
Australians. It appears that the pre-existing downward trend in adolescent
drinking might have been mistaken for a reduction in drinking caused by the
The age group under 25 did not respond to the tax likely because the de-
mand for alcohol for this age group exhibits high cross-price elasticity. Young
people have no problem switching to a different product if their preferred drink
increases in price. This is in line with Chikritzhs et al. (2009) and Mojica-Perez,
Callinan, and Livingston (2020), who show that, although the implementation
of the RTD tax in 2008 sharply reduced RTD consumption, this impact was
offset by increases in the consumption of spirits and other beverages. It is also
in line with arguments that some industry representatives used to contest the
tax – namely, that consumers would substitute RTDs with cheaper forms of
alcohol or drugs (Carragher and Chalmers 2011).
It is interesting to note in this connection that the 29% WET rebate of 2004
(Barton, Morgan, Pinto, et al. 2014) seems to noticeably increase drinking of
the age group under 25 (see Figure S3 in Supplementary Material section B).
The rebate was effectively a subsidy, propping up the production of cheap
wines (Daube and Stafford 2016). This further shows that the drinking of the
young Australian is sensitive to the presence of the cheapest drinking options.
One obvious implication of these results is that alcohol price policy should
avoid focusing on particular types of beverages because of the complex sub-
stitute and complement relationships between beverages. A better approach
would be to introduce a floor price on alcohol per standard drink or replace
the current taxation system on alcohol with a uniform volumetric tax per stan-
dard drink, applied across all alcohol products, as pioneered by the Northern
Territory (Taylor et al. 2021). These conclusions are not new and have been
advanced by many other earlier observers (e.g., Jiang and Livingston 2015;
Jiang et al. 2016; Sharma, Etil´
e, and Sinha 2016; Sharma, Vandenberg, and
Hollingsworth 2014; Srivastava et al. 2015).
Considering how decisive the Australian Government has been in its efforts
to protect young Australians with the RTD tax, it is to be hoped that the current
findings strengthen the case for reform of policy in relation to alcohol taxation.
Although the results are clear, like all research, the current study is not with-
out its limitations. The key limitation is that it relies on self-reporting. The
available evidence suggests that self-reports of alcohol consumption by cur-
rent drinkers are reliable (Gmel and Rehm 2004; Johnson and Mott 2001;
Lintonen, Ahlstr¨
om, and Metso 2004). However, we have no way of knowing
whether respondents in the HILDA survey give accurate accounts of their al-
cohol consumption.
The RTD tax was also aiming to reduce excessive drinking by younger peo-
ple, whereas we focus only on the mean values. The mean has to respond to
changes in excessive drinking as well, but if the social desirability or recall
biases truncate only excessive drinking, the mean may underestimate the ef-
fect of RTD tax. For example, Yang, Zhao, and Srivastava (2016) show that
binge drinking is often under-reported. In this sense, hospital admission or
other administrative data could be particularly useful to understand the ef-
fect of RTD on excessive adolescent drinkings. We encourage researchers to
replicate our methodology on administrative data to verify our results.
It is also possible that neither mean nor excessive drinking was affected by
the RTD, but mere switching to different types of beverages made adolescent
drinkers behave differently. For example, the work by Srivastava and Zhao
(2010) shows that RTDs (not wines) are most likely to be linked to risky be-
haviour such as property damage, stealing, and verbal and physical abuse un-
der alcohol influence. This might reconcile why some other metrics apart from
alcohol consumption (e.g., crime) may show a reduction for the young people
following the RTD tax (Weatherburn and Rahman 2021). The same amount
of alcohol consumed with a different type of alcoholic drink might reduce the
riskiness of the adolescent behaviour.
Supplementary material
A Past research
There is abundant evidence that the demand for alcohol is negatively related to
its price (Chaloupka, Grossman, and Saffer 2002; Fogarty 2010; Gallet 2007;
Wagenaar, Salois, and Komro 2009), so we would expect an increase in the
price of RTDs to reduce RTD consumption and most of the available evidence
suggests that it did. Chikritzhs et al. (2009) reported that 91 million fewer RTD
drinks were sold in the three months after the April 2008 tax increase than in
the same period in the previous year. Hall and Chikritzhs (2011), using data
from the Australian Bureau of Statistics, reported a two percent reduction in
per capita consumption of alcohol in the two years following the introduction
of the tax increase (after four years of rising per capita alcohol consumption).
Similarly, Doran and Digiusto (2011) reported a fall in RTD sales in the two
years following the increase in the RTD tax after four years of increasing RTD
sales before the tax change.
These simple ‘before and after’ comparisons involved no formal statistical
analysis of the trend in RTD consumption but subsequent studies employ-
ing interrupted time series methods have found the tax was followed by a
significant reduction in the emergency department (ED) admissions in Perth
(Lensvelt et al. 2016) and Sydney (Gale et al. 2015). Most recently, using data
from the NDSHS, Mojica-Perez, Callinan, and Livingston (2020) reported sta-
tistically significant reductions in self-reported alcohol consumption following
the introduction of the tax. The only negative result to date is that reported
by Kisely et al. (2011), who found no significant change in the proportion of
all ED admissions that were alcohol-related in two hospitals in Queensland
following the increase in the RTD tax.
The effect of the RTD tax increase on drinking by young (under 25-year-
old) people is not as clear. Four studies to date have examined the effect of
the tax on alcohol consumption by young people in Australia. The first of
these, Gale et al. (2015), involved an interrupted time series analysis of ED
admissions by age and gender across 39 hospitals in NSW. The study included
monthly retail liquor turnover to control for the effect of GFC in 2008. ‘Older
age groups’ (not otherwise described) were identified as a control, but their
alcohol consumption was not included as an explicit control in the analysis.
Testing for level and slope changes revealed no evidence of a change in the
level of ED admissions but a significant acceleration of the downward trend in
alcohol-related ED admissions for both males and females in the age groups
15–17, 18–24, 25–29 and females aged 50–64. No change in ED admissions
was observed amongst males aged 65 and over.
Lensvelt et al. (2016) employed Poisson regression to test for a significant
change in alcohol-related ED admissions to hospitals in Victoria and Western
Australia before and after the increase in the RTD tax. Separate tests were
conducted for ED admissions involving 12–15 and 15–19-year-old males in
both States. Immediately following the RTD tax increase, they found a sig-
nificant reduction in ED admissions in Western Australia for 12–15-year-olds
and a long-delayed reduction among 15–19-year-olds. No effects were found
in Victoria.
The third study to examine the effect of the RTD tax hike on young people
and the only one to date that sought to directly measure alcohol consump-
tion (instead of relying on ED admissions as a proxy for such consumption)
is Mojica-Perez, Callinan, and Livingston (2020). Using data from five waves
of NDSHS, they obtained alcohol consumption measures on the day preced-
ing the survey (ranging from 0 to 30 standard drinks). The drinks consumed
were grouped into four main categories (beer, wine, spirits, and pre-mixed
drinks). Consumption of pre-mixed drinks among those who reported having
consumed alcohol the previous day was found to be 59 percent lower in 2016
than it had been in 2007. However, significant reductions in the consumption
of all drink types were observed in the surveys conducted between 2004 to
2016 (2004, 2007, 2010, 2013, and 2016).
Gilmore et al. (2020) is the most recent study to have examined the effects
of the RTD tax increase on alcohol consumption by young people. The authors
conducted an interrupted time series analysis of Australian trends in chlamy-
dia (a sexually transmitted infection for which alcohol consumption is a known
risk factor) before and after the RTD tax increase. They found no significant
effect on their primary outcome measure (chlamydia notifications) but did find
a significant decrease in positive tests for chlamydia. They acknowledge the
fall might have been due to a decline in testing, but even if it were not and
even if we accept the assumption that trends in chlamydia are a good proxy
for trends in alcohol consumption by the young, the study sheds little light on
which specific age groups were most affected by the RTD tax increase.
B Descriptive statistics and trends in drinking
Table S1 shows the descriptive statistics of the variable used in the analysis. In
the table, the variables are pooled across all years. Note the 90th percentile of
the age variable. The control group takes up approximately 10% of the sample.
Drinking by years (overall and by age groups) are reported below.
Table S1: Descriptive statistics, HILDA 2002-2018.
Variable N Mean S.D. Min 0.25 Med 0.75 0.9 Max
Drinking 226,232 0.88 1.50 0.00 0.04 0.29 1.20 2.74 13.00
Age 344,513 36 23 0 17 35 54 69 102
Non-smoker 344,513 0.53 0.50 0 0 1 1 1 1
Income/1000 344,513 48.17 35.15 -2,000.00 27.26 41.15 60.53 929.38 989.67
NSW residency 344,513 0.30 0.46 0 0 0 1 1 1
We demonstrate the annual trends in drinking by various age groups using
the created drinking measures that are used in the econometric models (IHSF
of average daily consumption of standard drinks). Figure S1 shows overall
drinking trends. A reduction in overall alcohol consumption following the tax
is evident.
Figure S1: Australia national annual drinking trends, HILDA 2002-2018.
The figure visualizes the mean of the daily con-
sumption of standard drinks calculated using
all respondents in the survey. Vertical red line
indicates the introduction of the tax.
Figure S2 shows drinking for the age groups 70+ and 15–69. Prior to the
tax increase on RTDs, alcohol consumption by those aged under 70 closely
follows the trend for those aged 70+. The tax initiates a transition into a new
steady state. The transition roughly takes place during the period 2010–2014,
and the new equilibrium was achieved in 2015.
Figure S2: Trends in drinking: two age groups, HILDA 2002-2018.
Annual mean drinking by age groups. Vertical
red line indicates the introduction of the tax.
Figure S3 further breaks the age group 15–69 into the age groups 15–24
and 25–69. Drinking by those aged under 25 generally follows a downward
trend. The tax does not appear to have any influence on their drinking. There
is an increase in drinking in 2004, which is unrelated to the RTD tax and
likely related to the 29% wine equalization tax (WET) rebate of 2004 (Barton,
Morgan, Pinto, et al. 2014). The rebate was effectively a subsidy, propping up
the production of low-value wines (Daube and Stafford 2016). The effect of
the tax on the age groups 25–69 is easily distinguishable. The time trend for
this age group follows a clear mean-reverting process. The trend goes up right
before the tax increase and then goes down.
Figure S3: Trends in drinking: three age groups, HILDA 2002-2018.
Annual mean drinking by age groups. Vertical
red line indicates the introduction of the tax.
C Establishing the control group
Table S2: Age-specific drinking habits, NDSHS 2007.
Dependent variable (DV):
This is my usual type of alcohol drink (=1)
RTD Spirit Beer
(1) (2) (3) (4) (5) (6)
Treated1524 0.431*** 0.472*** 0.316***
(0.00677) (0.00889) (0.00818)
Treated2569 0.144*** 0.314*** 0.251***
(0.00269) (0.00354) (0.00325)
Treated1569 0.184*** 0.335*** 0.260***
(0.00258) (0.00331) (0.00302)
Control70+ 0.00924 0.00924 0.190*** 0.190*** 0.103*** 0.103***
(0.00624) (0.00645) (0.00820) (0.00825) (0.00754) (0.00754)
Observations 22,665 22,665 22,665 22,665 22,665 22,665
Adjusted 𝑅20.234 0.182 0.331 0.323 0.252 0.251
Mean of DV 0.159 0.159 0.315 0.315 0.238 0.238
Min of DV 0 0 0 0 0 0
Max of DV 1 1 1 1 1 1
*𝑝 < 0.05, ** 𝑝 < 0.01, *** 𝑝 < 0.001.
We use age-specific habits to delineate control and treatment groups. The
control groups should not consume RTDs. To verify the validity of our choice,
we appeal to NDSHS wave 2007 (the wave closest to the year of the tax im-
position). The advantage of NDSHS is that respondents are asked about their
usual type of alcohol drink, but the survey is conducted only every three years.
We need to verify that in the group 70+, nearly nobody states that their usual
drink is an RTD. Table S2 demonstrates the fraction of users by age groups
who affirm that their typical alcohol drink is an RTD. To recover the appro-
priate standard errors, we report those fractions by regressing the indicator
function on an array of dummies and suppressing the constant. For exam-
ple, regression results reported in column (1) show that 43.1% of respondents
aged 15–24 replied yes when asked if their preferred drink is RTD, a compa-
rable fraction of positive answers from the age group 25–69 is 14.4%. At the
same time, only 0.9% of the age group 70+ agreed that their preferred drink
is RTD, but this value proves to be statistically insignificant when the sam-
pling error is taken into account. Column (2) reports the estimates when the
treatment group is defined as respondents of age 15–69. As a confirmation
that this age-specific nature of drinking only typical for RTD, we estimated
similar equations for other types of drinks. Columns (3) to (6) show similar
results but for spirits and beer. For these types of drinks, 70+ is not an appro-
priate control group as this group also consume these types of drinks. This
confirms the uniqueness of RTD. RTDs target younger generations. Therefore
older generations could be used as a control group.
D Differences in drinking by birth cohorts
Figure S4: Trends in drinking: age and birth cohorts groups, HILDA 2002-
Annual mean drinking by age groups (red bars)
and by age groups and birth cohorts (con-
nected circles). The legend shows the birth co-
The red bars in Figure S4 show the average drinking by age in the dataset
used in the study. It increases sharply at the age 17–20, peaks at the late for-
ties, and slowly goes down. Figure S4 further shows average drinking by age
for different birth cohorts. Previous authors show that younger birth cohorts
tend to drink less throughout their lifetime (e.g., Livingston et al. 2016). Fig-
ure S4 confirms that it is the case in our data as well. It shows, for example,
that the age group 17–20 born in 1981–1990 drinks about 0.6standard drinks
(the black dot at the top of the red bar), whereas the group born in 1991–2000
only about 0.4(the blue dot at the bottom of the red bar).
Figure S5: Lifetime drinking habits by the year of birth, HILDA 2002-2018.
The plot shows the coefficients of the birth cohort fixed effects after
controlling for age fixed effects. The outcome variable is the IHSF of
daily standard drinks. Standard errors are robust and clustered at the
household level.
That is why in our model, we add cohort effect. We now show that parametriz-
ing the effect to be linear is sufficient. The differences in drinking by birth
cohorts follow a predictable pattern that can be easily seen in our data by
regressing our drinking measure on birth cohort fixed effects and age fixed
effects. The age fixed effects control for changes in drinking over the lifetime,
while the birth cohort fixed effects show drinking changes from one cohort to
another on average. This exercise is non-parametric and shows the data as it
is. Figure S5 visualizes the coefficients on the birth cohort fixed effects. Up
until the 1969 generation, the drinking is stable. Generations born in 1970
and after start drinking less. The reduction is linear.11 Since the change is
linear, we believe our model resolves the problem by including liner cohort
E Econometric models
Our DD specification:
𝑌𝑖𝑎𝑐𝑡𝑠 =
𝛽𝑡𝑠{𝑌 𝑒𝑎𝑟𝑡×𝑇 𝑟𝑒𝑎𝑡𝑒𝑑𝑎×𝑁𝑆𝑊𝑠}
𝑡𝑠{𝑌 𝑒𝑎𝑟𝑡×𝑋
𝑖𝑎𝑐𝑡𝑠 ×𝑁𝑆𝑊𝑠}
𝜆𝑡𝑊𝑡+𝛿𝑁 𝑆𝑊𝑠
+𝜓𝑉𝑐+𝜀𝑖𝑎𝑐𝑡𝑠 .
The variable 𝑌𝑖𝑎𝑐𝑡𝑠 is the drinking of person 𝑖in the age group 𝑎, belonging
to birth cohort 𝑐, in year 𝑡and state 𝑠. The variable 𝑇 𝑟 𝑒𝑎𝑡𝑒𝑑𝑎is an indicator
set equal to one for the age group 15–69. The variable 𝑌 𝑒𝑎𝑟𝑡is an indicator for
year 𝑡. The model allows for lags of post-treatment effects and leads of antic-
ipatory effects. The anticipatory effects are expected to be jointly statistically
insignificant if the common trend assumption is satisfied. The post-treatment
effects show the tax effect’s evolution before a new steady state of drinking is
reached. The coefficients 𝛾𝑎and 𝜆𝑡stand for age group and time fixed effect, re-
spectively. The array 𝑋includes individual-level characteristics. To control
for the observation that younger birth cohorts drink less, we add the variable
𝑉𝑐, which assigns 1to birth cohort 2003,2to birth cohort 2004, and so on. The
final term, 𝜀𝑖𝑎𝑐𝑡 , is an econometric error. The variable 𝑁 𝑆𝑊𝑠is a dummy for
NSW. Because the dependent variable is effectively log-transformed and the
year 2009 is excluded, the coefficients 𝛽𝑡can be interpreted as the average
annual per cent change in drinking relative to 2009.
Our CITS specification:
11The birth cohort of 1980 is somewhat unusual. It drinks more than all birth cohorts from
1971 to 1979, but it is likely a sampling error.
𝑌𝑖𝑎𝑐𝑡𝑠 =𝛼1𝑡+𝛼2𝑃 𝑜𝑠𝑡𝑡+𝛼3{𝑡×𝑃 𝑜𝑠𝑡𝑡}
𝛼4,𝑔{𝑡×𝑇 𝑟𝑒𝑎𝑡𝑒𝑑𝑔
𝛽1,𝑔𝑠{𝑃 𝑜𝑠𝑡𝑡×𝑇 𝑟𝑒𝑎𝑡𝑒𝑑𝑔
𝛽2,𝑔𝑠{𝑡×𝑃 𝑜𝑠𝑡𝑡×𝑇 𝑟𝑒𝑎𝑡𝑒𝑑𝑔
𝑡𝑠{𝑌 𝑒𝑎𝑟𝑡×𝑋
𝑖𝑎𝑐𝑡𝑠 ×𝑁𝑆𝑊𝑠}
𝜏×𝑇 𝑟𝑒𝑎𝑡𝑒𝑑1524
𝑎}+𝛿𝑁 𝑆𝑊𝑠
+𝜓𝑉𝑐+𝜀𝑖𝑎𝑐𝑡𝑠 .
The variable 𝑡is the time trend (e.g., 1 for 2002, 2 for 2003) and 𝑡=𝑡10.12
The control variables are the same as in Equation (1) and are also (and for the
same reason) interacted with an array of year dummies. The variable 𝑇 𝑟𝑒𝑎𝑡𝑒𝑑𝑔
is now trichotomous with 𝑔∈ {15 24,25 69,70+}. The omitted category is
still the age groups 70+.
The variable 𝑊
𝜏is an array of dummies for 2002, 2003, and 2004. This ar-
ray accommodates the irregularity that we see for the 15–25 age group drink-
ing before 2005. Effectively these dummies control for the WET rebate of 2004,
which increased drinking of the age group under 25 (note an increase in drink-
ing in 2005 for age groups 15–24 in Figure S3). Otherwise, the model assumes
that the trends are linear. This regression specification allows the tax effect to
be reflected in a level shift (𝛽1) and a shift in the trend (𝛽2).
Our baseline DD and CITS models do not have the variable 𝑁𝑆𝑊𝑠(and the
subscript 𝑠in the outcome variable and error term) as well as the parameters
𝑡𝑠 and 𝜓. Adding individual smoking and household income means introduc-
ing the parameters 𝜑
𝑡𝑠. Adding interaction with the state of NSW means adding
the variable 𝑁 𝑆𝑊𝑠(and introducing additional variation into the outcome and
the error term, as indicated by the subscript 𝑠). Adding linear cohort effects
means introducing the parameter 𝜓(and the subscript 𝑐).
Abadie, A., S. Athey, G. W. Imbens, and J. Wooldridge (Nov. 2017). When Should
You Adjust Standard Errors for Clustering? Working Paper 24003. National
Bureau of Economic Research. url:http: / /www . / papers/ w24003
(page 5).
12See Huitema and Mckean (2000) and Linden (2015), among many others, for a gentle
discussion of CITS models.
Agrawal, A., A. C. Heath, J. D. Grant, M. L. Pergadia, D. J. Statham, K. K.
Bucholz, N. G. Martin, and P. A. F. Madden (July 2006). “Assortative Mating
for Cigarette Smoking and for Alcohol Consumption in Female Australian
Twins and their Spouses”. In: Behavior Genetics 36.4, pp. 553–566. url: (page 5).
Alexeev, S. (2021). “The female vulnerability to parental drinking: evidence
from randomly assigned interviewers”. In: ResearchGate working paper.url:
to _ parental _ drinking _ evidence _ from _ randomly _ assigned _ interviewers
(page 5).
Australian Drug Foundation (2004). “Submission to the ‘Review of alcoholic
beverages that may target young people’ by the NSW Department of Gaming
and Racing”. In: Drug Issues: our views 204 (page 2).
Australian Institute of Health and Welfare (2008). 2007 National Drug Strategy
Household Survey: Detailed Findings. Drug statistics series. Australian In-
stitute of Health and Welfare. url:
OH-fAQAACAAJ (page 2).
(2011). 2010 National Drug Strategy Household Survey: Detailed Findings.
Drug statistics series. Australian Institute of Health and Welfare. url:https:
// (page 2).
Barton, G., A. Morgan, D. Pinto, et al. (2014). “The WET: is it a good drop?” In:
Tax Specialist 18.2, p. 54. url:;
dn=775868106360394;res=IELBUS (pages 8,12).
Burbidge, J. B., L. Magee, and A. L. Robb (1988). “Alternative Transformations
to Handle Extreme Values of the Dependent Variable”. In: Journal of the
American Statistical Association 83.401, pp. 123–127. url:http : / / www . (page 3).
Callinan, S., A. Pennay, M. Livingston, and E. Kuntsche (2020). “Patterns of
alcohol consumption in 16 cohorts of Australian young adults aged 15–24
between 2001 and 2016”. In: Addiction 115.8, pp. 1452–1458. url:https:
// (page 5).
Cameron, A. C. and D. L. Miller (2015). “A practitioner’s guide to cluster-robust
inference”. In: Journal of human resources 50.2, pp. 317–372. url:http://
(page 5).
Carragher, N. and J. Chalmers (2011). “What are the options? Pricing and tax-
ation policy reforms to redress excessive alcohol consumption and related
harms in Australia”. In: NSW Bureau of Crime Statistics and Research.url:
https://ndarc. /
What%20are%20the%20options_BOCSAR%20report.pdf (pages 2,8).
Chaloupka, F. J., M. Grossman, and H. Saffer (2002). “The effects of price on
alcohol consumption and alcohol-related problems.” In: Alcohol Research &
Health 26.1, pp. 22–34. url:
PMC6683806/ (page 9).
Chikritzhs, T. N., P. M. Dietze, S. J. Allsop, M. M. Daube, W. D. Hall, and K.
Kypri (2009). “The “alcopops” tax: heading in the right direction”. In: Medical
Journal of Australia 190.6, p. 294. url: system /
files/issues/190_06_160309/chi11362_fm.pdf (pages 2,8,9).
Conley, T. G. and C. R. Taber (2011). “Inference with “Difference in Differences”
with a Small Number of Policy Changes”. In: The Review of Economics and
Statistics 93.1, pp. 113–125. url:https://doi .org/10.1162/REST_a_00049
(page 5).
Copeland, J., R. J. Stevenson, P. Gates, and P. Dillon (2007). “Young Aus-
tralians and alcohol: the acceptability of ready-to-drink (RTD) alcoholic bev-
erages among 12–30-year-olds”. In: Addiction 102.11, pp. 1740–1746. url:
https :/ /onlinelibrary .wiley .com / doi/ abs/ 10. 1111/ j. 1360- 0443 .2007 .
01970.x (page 2).
Daube, M. and J. Stafford (2016). “Alcohol and tax – time for real reform”. In:
Medical Journal of Australia 204.6, pp. 218–219. url:
au/system/files/issues/204_06/10.5694mja16.00022.pdf (pages 8,12).
Decker, S. L. and A. E. Schwartz (Feb. 2000). Cigarettes and Alcohol: Substi-
tutes or Complements? Working Paper 7535. National Bureau of Economic
Research. url: (page 4).
Doran, C. M. and E. Digiusto (2011). “Using taxes to curb drinking: A report
card on the Australian government’s alcopops tax”. In: Drug and Alcohol
Review 30.6, pp. 677–680. url:
10.1111/j.1465-3362.2011.00309.x (pages 2,9).
Elgar, F. J., N. Phillips, and N. Hammond (2011). “Trends in Alcohol and Drug
Use among Canadian Adolescents, 1990–2006”. In: The Canadian Journal
of Psychiatry 56.4. PMID: 21507281, pp. 243–247. url:
10.1177/070674371105600408 (page 2).
Fogarty, J. (2010). “The demand for beer, wine and spirits: a survey of the
literature”. In: Journal of Economic Surveys 24.3, pp. 428–478. url:https:
/ /onlinelibrary . wiley. com /doi / abs/ 10 .1111 /j . 1467- 6419 . 2009. 00591 .x
(page 9).
Gale, M. et al. (May 2015). “Alcopops, taxation and harm: a segmented time
series analysis of emergency department presentations”. In: BMC Public
Health 15.1, p. 468. url:https : // doi. org/ 10 .1186 /s12889 - 015- 1769 - 3
(pages 2,9).
Gallet, C. A. (2007). “The demand for alcohol: a meta-analysis of elasticities”.
In: Australian Journal of Agricultural and Resource Economics 51.2, pp. 121–
135. url:https : / / onlinelibrary . wiley . com / doi / abs / 10 . 1111 / j . 1467 -
8489.2007.00365.x (page 9).
Gates, P., J. Copeland, R. J. Stevenson, and P. Dillon (Dec. 2006). “The influ-
ence of product packaging on young people’s palatability rating for RTDs
and other alcoholic beverages”. In: Alcohol and Alcoholism 42.2, pp. 138–
142. url: (page 2).
Gilmore, W., T. Chikritzhs, H. McManus, J. Kaldor, and R. Guy (2020). “The
Association between the Australian Alcopops Tax and National Chlamydia
Rates among Young People – an Interrupted Time Series Analysis”. In: Inter-
national journal of environmental research and public health 17.4, p. 1343.
url: (pages 2,10).
Gmel, G. and J. Rehm (2004). “Measuring Alcohol Consumption”. In: Contem-
porary Drug Problems 31.3, pp. 467–540. url: / 10 .1177/
009145090403100304 (page 8).
Hall, W. and T. Chikritzhs (2011). “The Australian alcopops tax revisited”. In:
The Lancet 377.9772, pp. 1136–1137. url:
science/article/pii/S0140673610614201 (pages 2,9).
Hallberg, K., R. Williams, A. Swanlund, and J. Eno (2018). “Short Comparative
Interrupted Time Series Using Aggregate School-Level Data in Education
Research”. In: Educational Researcher 47.5, pp. 295–306. url:https : / / (page 4).
Health and Social Care Information Centre (2018). Statistics on Alcohol, Eng-
land, 2018. United Kingdom National Health Service (NHS Digital). url:
statistics-on-alcohol/2018/part-5 (page 2).
Howe, L. J., D. J. Lawson, N. M. Davies, B. St. Pourcain, S. J. Lewis, G. Davey
Smith, and G. Hemani (Nov. 2019). “Genetic evidence for assortative mating
on alcohol consumption in the UK Biobank”. In: Nature Communications
10.1, p. 5039. url: (page 5).
Huitema, B. E. and J. W. Mckean (2000). “Design Specification Issues in Time-
Series Intervention Models”. In: Educational and Psychological Measure-
ment 60.1, pp. 38–58. url:https :/ /doi . org/ 10 . 1177 /00131640021970358
(page 16).
Jackson, N., S. Denny, J. Sheridan, T. Fleming, T. Clark, R. Peiris-John, and
S. Ameratunga (2017). “Uneven reductions in high school students’ alcohol
use from 2007 to 2012 by age, sex, and socioeconomic strata”. In: Substance
Abuse 38.1. PMID: 27782782, pp. 69–76. url:
08897077.2016.1252827 (page 2).
Jiang, H. and M. Livingston (June 2015). “The Dynamic Effects of Changes
in Prices and Affordability on Alcohol Consumption: An Impulse Response
Analysis”. In: Alcohol and Alcoholism 50.6, pp. 631–638. url:https://doi.
org/10.1093/alcalc/agv064 (page 8).
Jiang, H., M. Livingston, R. Room, and S. Callinan (2016). “Price elasticity
of on- and off-premises demand for alcoholic drinks: A Tobit analysis”.
In: Drug and Alcohol Dependence 163, pp. 222–228. url:http : / / www . (page 8).
Johnson, T. P. and J. A. Mott (2001). “The reliability of self-reported age of
onset of tobacco, alcohol and illicit drug use”. In: Addiction 96.8, pp. 1187–
1198. url:https : / / onlinelibrary . wiley .com /doi /abs /10 .1046 / j . 1360 -
0443.2001.968118711.x (page 8).
Johnston, L. D., R. A. Miech, P. M. O’Malley, J. G. Bachman, J. E. Schulenberg,
and M. E. Patrick (2019). “Monitoring the Future National Survey Results
on Drug Use, 1975-2018: Overview, Key Findings on Adolescent Drug Use.”
In: Institute for Social Research.url:
ED594190.pdf (page 2).
Jones, S. and L. Barrie (2011). “RTDs in Australia: Expensive designer drinks
or cheap rocket fuel?” In: Drug and Alcohol Review 30.1, pp. 4–11. url:
https :/ /onlinelibrary .wiley .com / doi/ abs/ 10. 1111/ j. 1465- 3362 .2010 .
00181.x (page 2).
Kisely, S. R., J. Pais, A. White, J. Connor, L.-H. Quek, J. L. Crilly, and D.
Lawrence (2011). “Effect of the increase in “alcopops” tax on alcohol-related
harms in young people: a controlled interrupted time series”. In: Medical
Journal of Australia 195.11-12, pp. 690–693. url:https://onlinelibrary. (pages 2,9).
Kraus, L., M. E. Tingh¨
og, A. Lindell, A. Pabst, D. Piontek, and R. Room (Mar.
2015). “Age, Period and Cohort Effects on Time Trends in Alcohol Con-
sumption in the Swedish Adult Population 1979–2011”. In: Alcohol and Al-
coholism 50.3, pp. 319–327. url:
(page 5).
Lensvelt, E., W. Liang, W. Gilmore, E. Gordon, M. Hobday, and T. Chikritzhs
(2016). “Effect of the Australian “Alcopops Tax” on Alcohol-Related Emer-
gency Department Presentations for Injury in Two States”. In: Journal of
Studies on Alcohol and Drugs 77.5. PMID: 27588531, pp. 730–739. url: (pages 2,9,10).
Linden, A. (2015). “Conducting Interrupted Time-series Analysis for Single-
and Multiple-group Comparisons”. In: The Stata Journal 15.2, pp. 480–
500. url: (page 16).
Lintonen, T., S. Ahlstr¨
om, and L. Metso (July 2004). “The reliability of self-
reported drinking in adolescence”. In: Alcohol and Alcoholism 39.4, pp. 362–
368. url: (page 8).
Livingston, M., J. Raninen, T. Slade, W. Swift, B. Lloyd, and P. Dietze (2016).
“Understanding trends in Australian alcohol consumption—an age–period–cohort
model”. In: Addiction 111.9, pp. 1590–1598. url:https:/ /onlinelibrary. (pages 5,13).
Looze, M. d. et al. (Mar. 2015). “Decreases in adolescent weekly alcohol use
in Europe and North America: evidence from 28 countries from 2002 to
2010”. In: European Journal of Public Health 25, pp. 69–72. url:https :
// (page 5).
MacKinnon, J. G. (2019). “How cluster-robust inference is changing applied
econometrics”. In: Canadian Journal of Economics 52.3, pp. 851–881. url: (page 5).
MacKinnon, J. G. and L. Magee (1990). “Transforming the Dependent Variable
in Regression Models”. In: International Economic Review 31.2, pp. 315–
339. url: (page 3).
Markowitz, A. J. (2018). “Changes in School Engagement as a Function of
No Child Left Behind: A Comparative Interrupted Time Series Analysis”.
In: American Educational Research Journal 55.4, pp. 721–760. url:https:
// (page 4).
endez, P., F. Tusell, and D. Weatherburn (2015). “The effects of liquor li-
censing restriction on alcohol-related violence in NSW, 2008–13”. In: Addic-
tion 110.10, pp. 1574–1582. url:https : / / onlinelibrary . wiley . com / doi /
abs/10.1111/add.12951 (page 5).
Meng, Y., J. Holmes, D. Hill-McManus, A. Brennan, and P. S. Meier (2014).
“Trend analysis and modelling of gender-specific age, period and birth co-
hort effects on alcohol abstention and consumption level for drinkers in
Great Britain using the General Lifestyle Survey 1984–2009”. In: Addiction
109.2, pp. 206–215. url:https ://onlinelibrary. wiley . com/doi/ abs / 10.
1111/add.12330 (page 5).
Mojica-Perez, Y., S. Callinan, and M. Livingston (2020). “Examining beverage-
specific trends in youth drinking in Australia before and after the imple-
mentation of the alcopops tax”. In: Drug and Alcohol Review 39.3, pp. 246–
254. url:https :/ / onlinelibrary.wiley . com/ doi /abs / 10. 1111 /dar. 13038
(pages 2,810).
Mosher, J. F. and D. Johnsson (Sept. 2005). “Flavored Alcoholic Beverages: An
International Marketing Campaign that Targets Youth”. In: Journal of Public
Health Policy 26.3, pp. 326–342. url:
jphp.3200037 (page 3).
om, T. and J. Svensson (2014). “The declining trend in Swedish youth
drinking: collectivity or polarization?” In: Addiction 109.9, pp. 1437–1446.
url: (page 2).
OECD (2020). OECD Income Distribution Database. Compare your income –
methodology and conceptual issues. OECD Income Distribution Database.
url: methodology.
pdf (page 3).
Prime Minister of Australia (2008). National Binge Drinking Strategy. Media Re-
lease. Parliament of Australia. url:
search/display /display.w3p;query=Id %3A%22media%2Fpressrel%2F3FVP6 % 22
(page 2).
Roth, L. and C. Angus (2014). Liquor licensing restrictions to address alcohol-
related violence in NSW: 2008 to 2014. NSW Parliamentary Research Ser-
vice Sydney. url:https : / / www . parliament . nsw . gov . au / researchpapers /
Documents / liquor - licensing - restrictions - to - address - alcohol / Liquor %
20in%20NSW%202008%20to%202014.pdf (page 5).
Scollo, M. and M. Bayly (2020). 13.2 Tobacco taxes in Australia. Tobacco in
Australia: facts and issues. Melbourne: Cancer Council Victoria. url:http:
/ / www . tobaccoinaustralia . org . au / chapter - 13 - taxation / 13 - 2 - tobacco -
taxes-in-australia (page 4).
Sharma, A., F. Etil´
e, and K. Sinha (2016). “The Effect of Introducing a Mini-
mum Price on the Distribution of Alcohol Purchase: A Counterfactual Anal-
ysis”. In: Health Economics 25.9, pp. 1182–1200. url:https://onlinelibrary. (page 8).
Sharma, A., B. Vandenberg, and B. Hollingsworth (Jan. 2014). “Minimum Pric-
ing of Alcohol versus Volumetric Taxation: Which Policy Will Reduce Heavy
Consumption without Adversely Affecting Light and Moderate Consumers?”
In: PLOS ONE 9.1, pp. 1–13. url:
0080936 (page 8).
Srivastava, P., K. R. McLaren, M. Wohlgenant, and X. Zhao (2015). “Disag-
gregated econometric estimation of consumer demand response by alco-
holic beverage types”. In: Australian Journal of Agricultural and Resource
Economics 59.3, pp. 412–432. url:
abs/10.1111/1467-8489.12095 (page 8).
Srivastava, P. and X. Zhao (2010). “What Do the Bingers Drink? Micro-Unit
Evidence on Negative Externalities and Drinker Characteristics of Alcohol
Consumption by Beverage Types*”. In: Economic Papers: A journal of applied
economics and policy 29.2, pp. 229–250. url:https://onlinelibrary.wiley.
com/doi/abs/10.1111/j.1759-3441.2010.00066.x (pages 2,9).
Tauchmann, H., S. Lenz, T. Requate, and C. M. Schmidt (Aug. 2013). “Tobacco
and alcohol: complements or substitutes?” In: Empirical Economics 45.1,
pp. 539–566. url: (page 4).
Taylor, N., P. Miller, K. Coomber, M. Livingston, D. Scott, P. Buykx, and T.
Chikritzhs (2021). “The impact of a minimum unit price on wholesale al-
cohol supply trends in the Northern Territory, Australia”. In: Australian
and New Zealand Journal of Public Health 45.1, pp. 26–33. url:https :
// (page 8).
Wagenaar, A. C., M. J. Salois, and K. A. Komro (2009). “Effects of beverage alco-
hol price and tax levels on drinking: a meta-analysis of 1003 estimates from
112 studies”. In: Addiction 104.2, pp. 179–190. url:https://onlinelibrary. (page 9).
Watson, N. and M. Wooden (2020). “The Household, Income and Labour Dy-
namics in Australia (HILDA) Survey”. In: Jahrb ¨
ucher f¨
ur National¨
und Statistik.url:
of-print/article-10.1515- jbnst-2020- 0029/article-10.1515-jbnst-2020-
0029.xml (page 5).
Weatherburn, D. and S. Rahman (2021). The Vanishing Criminal: Causes of De-
cline in Australia’s Crime Rate. Melbourne University Publishing. url:https:
// (page 9).
Windle, M. and R. C. Windle (2012). “Intergenerational Relations for Drinking
Motives: Invariant for Same- and Opposite-Sex Parent-Child Dyads?” In:
Journal of Studies on Alcohol and Drugs 73.1. PMID: 22152663, pp. 63–70.
url: (page 5).
Wong, M., T. D. Cook, and P. M. Steiner (2015). “Adding Design Elements
to Improve Time Series Designs: No Child Left Behind as an Example of
Causal Pattern-Matching”. In: Journal of Research on Educational Effective-
ness 8.2, pp. 245–279. url:
(page 4).
Wooden, M., S. Freidin, and N. Watson (2002). “The Household, Income and
Labour Dynamics in Australia (HILDA)Survey: Wave 1”. In: Australian Eco-
nomic Review 35.3, pp. 339–348. url:https: / /
doi/abs/10.1111/1467-8462.00252 (page 5).
Yang, O., X. Zhao, and P. Srivastava (2016). “Binge Drinking and Antisocial and
Unlawful Behaviours in Australia”. In: Economic Record 92.297, pp. 222–
240. url:https :// doi /abs/10.1111/1475- 4932.
12243 (page 8).
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Objective: The Northern Territory (NT) Government introduced a minimum unit price (MUP) of $1.30 per standard drink (10g pure alcohol) explicitly aimed at reducing the consumption of cheap wine products from October 2018. We aimed to assess the impact of the NT MUP on estimates of beverage-specific population-adjusted alcohol consumption using wholesale alcohol supply data. Methods: Interrupted time series analyses were conducted to examine MUP effects on trends in estimated per capita alcohol consumption (PCAC) for cask wine, total wine and total alcohol, across the NT and in the Darwin/Palmerston region. Results: Significant step decreases were found for cask wine and total wine PCAC in Darwin/Palmerston and across the Northern Territory. PCAC of cask wine decreased by 50.6% in the NT, and by 48.8% in Darwin/Palmerston compared to the prior year. PCAC for other beverages (e.g. beer) were largely unaffected by MUP. Overall, PCAC across the Territory declined, but not in Darwin/Palmerston. Conclusion: With minimal implementation costs, the Northern Territory Government's MUP policy successfully targeted and reduced cask wine and total wine consumption. Cask wine, in particular, almost halved in Darwin/Palmerston where the impact of the MUP was able to be determined and considering other interventions. Implications for public health: Implementation of a minimum unit price for retail alcohol sales is a cost-effective way to reduce the consumption of high alcohol content and high-risk products, such as cheap cask wine.
Full-text available
A national tax increase, which became known as the “alcopops tax”, was introduced in Australia on the 27th April 2008 on ready-to-drink alcoholic beverages, which are consumed predominantly by young people. The affordability of alcohol has been identified as the strongest environmental driver of alcohol consumption, and alcohol consumption is a well-known risk factor in the spread of sexually transmitted infections via its association with sexual risk-taking. We conducted a study to investigate whether there was any association between the introduction of the tax and changes in national chlamydia rates: (i) notification rates (diagnoses per 100,000 population; primary outcome and standard approach in alcohol taxation studies), and (ii) test positivity rates (diagnoses per 100 tests; secondary outcome) among 15–24 and 25–34-year-olds, using interrupted time series analysis. Gender- and age-specific chlamydia trends among those 35 and older were applied as internal control series and gender- and age-specific consumer price index-adjusted per capita income trends were controlled for as independent variables. We hypothesised that the expected negative association between the tax and chlamydia notification rates might be masked due to increasing chlamydia test counts over the observation period (2000 to 2016). We hypothesised that the association between the tax and chlamydia test positivity rates would occur as an immediate level decrease, as a result of a decrease in alcohol consumption, which, in turn, would lead to a decrease in risky sexual behaviour and, hence, chlamydia transmission. None of the gender and age-specific population-based rates indicated a significant immediate or lagged association with the tax. However, we found an immediate decrease in test positivity rates for 25–34-year-old males (27% reduction—equivalent to 11,891 cases prevented post-tax) that remained detectable up to a lag of six months and a decrease at a lag of six months for 15–24-year-old males (31% reduction—equivalent to 16,615 cases prevented) following the tax. For no other gender or age combination did the change in test positivity rates reach significance. This study adds to the evidence base supporting the use of alcohol taxation to reduce health-related harms experienced by young people and offers a novel method for calculating sexually transmitted infection rates for policy evaluation.
Full-text available
Alcohol use is correlated within spouse-pairs, but it is difficult to disentangle effects of alcohol consumption on mate-selection from social factors or the shared spousal environment. We hypothesised that genetic variants related to alcohol consumption may, via their effect on alcohol behaviour, influence mate selection. Here, we find strong evidence that an individual’s self-reported alcohol consumption and their genotype at rs1229984, a missense variant in ADH1B, are associated with their partner’s self-reported alcohol use. Applying Mendelian randomization, we estimate that a unit increase in an individual’s weekly alcohol consumption increases partner’s alcohol consumption by 0.26 units (95% C.I. 0.15, 0.38; P = 8.20 × 10⁻⁶). Furthermore, we find evidence of spousal genotypic concordance for rs1229984, suggesting that spousal concordance for alcohol consumption existed prior to cohabitation. Although the SNP is strongly associated with ancestry, our results suggest some concordance independent of population stratification. Our findings suggest that alcohol behaviour directly influences mate selection.
Full-text available
In many fields of economics, and also in other disciplines, it is hard to justify the assumption that the random error terms in regression models are uncorrelated. It seems more plausible to assume that they are correlated within clusters, such as geographical areas or time periods, but uncorrelated across clusters. It has therefore become very popular to use “clustered” standard errors, which are robust against arbitrary patterns of within‐cluster variation and covariation. Conventional methods for inference using clustered standard errors work very well when the model is correct and the data satisfy certain conditions, but they can produce very misleading results in other cases. This paper discusses some of the issues that users of these methods need to be aware of. Résumé Comment l'inférence robuste à l'échantillonnage est en passe de modifier l'économétrie appliquée. Dans de nombreuses branches de l'économie, mais également dans d'autres disciplines, il apparaît difficile de justifier l'hypothèse selon laquelle les termes d'erreurs aléatoires des modèles de régression sont indépendants. Il semble plus vraisemblable de considérer qu'ils sont corrélés à l'intérieur des échantillonnages, notamment dans le cadre d'aires géographiques ou de périodes de temps, mais qu'ils sont indépendants en dehors. Il est donc devenu très courant d'utiliser les écarts‐types d'échantillonnage, lesquels s'avérant robustes contre les modèles arbitraires de variations et de covariations à l'intérieur de ces mêmes échantillons. Les méthodes habituelles utilisant les écarts‐types d'échantillonnage pour évaluer l'inférence fonctionnent parfaitement lorsque le modèle est cohérent et que les données satisfont à certaines conditions. Néanmoins, dans d'autres circonstances, ces dernières peuvent aboutir à des résultats trompeurs. Cet article étudie certains des écueils que peuvent rencontrer les utilisateurs de cette méthode, et contre lesquels ils doivent se prémunir.
Introduction and aims: Alcohol consumption among young Australians has declined markedly since the early 2000s. As yet, there has been no data on how this decline has been spread across different beverages and instead high-level survey data with significant potential for recall and other bias has been used. Trends in beverage choice among young people following an increase in the 'alcopops' tax have also not received much attention. Design and methods: Data on 'yesterday' drinking occasions were obtained from five waves (2004, 2007, 2010, 2013, 2016) of the National Drug Strategy Household Survey. A total of 23 536 respondents aged 14-29 years were included in this study. Descriptive and regression analyses were conducted to explore trends in alcohol consumption and changes in beverage preferences. Results: Youth drinking declined by 45% across the study period, with declines of 66% in premix, 48% in spirits, 46% in beer and 33% in wine. Consumption of premix was significantly lower in 2013 and 2016 compared to 2007 amongst the overall sample, males, females, respondents aged 14-21 and 22-29 years, light and heavy drinkers. Significant reductions were also observed in the consumption of premix immediately following the tax (2010) for the younger age group, males and light drinkers. Discussion and conclusions: Youth consumption of alcohol has declined during the study period with significant variation across beverage types. We found some evidence of a separate impact for the alcopops tax, although for some groups, declines in premix consumption occurred well after the implementation of the tax.
Background and aims: Repeated cross-sectional surveys have identified substantial declines in adolescent drinking in Australia and some other countries in recent years. There is debate about whether these declines will be maintained as the cohort ages. This study modelled alcohol consumption over time to check for cohort effects reflecting a decrease in youth consumption and then used this model to predict how decreases in youth drinking will be sustained through to adulthood. Design: Longitudinal study using data from the Household Income and Labour Dynamics in Australia (HILDA) survey from 2001 to 2016. Piecewise latent growth models were estimated to assess consumption trajectories for each birth cohort from age 15 to 18 and 18 to 24. Setting: Australia PARTICIPANTS: This study focused on 5320 (51.9% female) respondents aged between 15 and 22 in Wave 1 (2001) to those aged between 17 and 24 in Wave 16 (2016). Measurement: Annual volume of alcohol consumption was calculated as the product of the quantity per occasion and the frequency of drinking expanded to represent drinking occasions per year. Findings: The model with best fit suggested consumption increased rapidly (b=0.67, SE = 0.05, p<.001) until the legal drinking age of 18 and then plateaued (b=-0.03, SE = 0.02, p=.088). More recent cohorts start with significantly lower levels of consumption (b=-0.15, SE = 0.01; p<.001) but increase at a faster rate (b=0.02, SE = 0.003, p<.001) between 15 and 18; however, not enough to catch up to earlier cohorts. Conclusion: Recent decreases in adolescent drinking in Australia may, at least in part, be attributed to lower consumption in recent cohorts of younger drinkers. Results indicate that this group may continue to drink less than previous cohorts as they age into their twenties.
Short comparative interrupted times series (CITS) designs are increasingly being used in education research to assess the effectiveness of school-level interventions. These designs can be implemented relatively inexpensively, often drawing on publicly available data on aggregate school performance. However, the validity of this approach hinges on a variety of assumptions and design decisions that are not clearly outlined in the literature. This article aims to serve as a practice guide for applied researchers when deciding how and whether to use this approach. We begin by providing an overview of the assumptions needed to estimate causal effects using school-level data, common threats to validity faced in practice and what effects can and cannot be estimated using school-level data. We then examine two analytic decisions researchers face in practice when implementing the design: correctly modeling the pretreatment functional form, which is modeling the preintervention trend, and selecting comparison cases. We then illustrate the use of this design in practice drawing on data from the implementation of the school improvement grant (SIG) program in Ohio. We conclude with advice for applied researchers implementing this design.
Around 20% of all empirical papers published by the American Economic Review between 2010 and 2012 estimate treatment effects using linear regressions with time and group fixed effects. In a model where the effect of the treatment is constant across groups and over time, such regressions identify the treatment effect of interest under the standard "common trends" assumption. But these regressions have not been analyzed yet allowing for treatment effect heterogeneity. We show that under two alternative sets of assumptions, such regressions identify weighted sums of average treatment effects in each group and period, where some weights may be negative. The weights can be estimated, and can help researchers assess whether their results are robust to heterogeneous treatment effects across groups and periods. When many weights are negative, their estimates may not even have the same sign as the true average treatment effect if treatment effects are heterogenous. We also propose another estimator of the treatment effect that does not rely on any homogeneity assumption. Finally, we estimate the weights in two applications and find that in both cases, around half of the average treatment effects receive a negative weight.