Using supermarket loyalty card data to measure the differential impact of the UK soft drink sugar tax on buyer behaviour

Abstract

This paper explores the impact of the soft drinks sugar tax introduced in the UK in 2018 on the purchasing behaviours of different geo‐demographic consumer segments. We analyse data for a composite good comprising the most popular sugar‐sweetened drinks (SSDs) using loyalty card data from one of the UK’s largest supermarkets. We use pre‐levy data to predict the effect of the tax and corroborate our predictions by analysing actual consumption of the composite good in the first 5 months post‐levy. The results show that the impact of the sugar tax is likely to have the desired effect of reducing the purchase of SSDs. Moreover, though the impact of the tax is likely to vary across different geo‐demographic segments, the evidence suggests that its impact is likely to be greatest on the most vulnerable market segments – families on low incomes – who are among the highest consumers of SSDs in the UK.

Full text

J Agric Ec on. 2 02 1;0 0 :1– 17.
|
1
wileyonlinel ibrary.com/journal/jage
Received: 23 Novemb er 2020
|
Revis ed: 2 Sep tembe r 2021
|
Acce pted: 10 Se ptemb er 2021
DOI: 10.1111/1477-9552.12462
ORIGINAL ARTICLE
Using supermarket loyalty card data to measure the
differential impact of the UK soft drink sugar tax on
buyer behaviour
AndrewFearne1 | NataliaBorzino2 | BeatrixDe La Iglesia3 |
PeterMoffatt4 | MargaretRobbins3
© 2021 The Agricu ltura l Econo mics S ociet y.
1Norwi ch Busi ness S chool, Universit y of
East Anglia , Norwich, UK
2Institute for Env ironm ental D ecisions, ET H
Zürich, Si ngapore, Singapore
3School of Computi ng Sciences, Univer sity
of East A nglia, Norw ich, UK
4School of Econom ics, Univers ity of Eas t
Angl ia, Nor wich, U K
Correspondence
Andr ew Fearn e, Norw ich Busines s School,
University of Ea st Angl ia, Norwich , UK.
Email: a.fearne@uea.ac.uk
Abstract
This paper explores the impact of the soft drinks sugar tax
introduced in the UK in 2018 on the purchasing behav-
iours of different geo- demographic consumer segments.
We analyse data for a composite good comprising the most
popular sugar- sweetened drinks (SSDs) using loyalty card
data from one of the UK’s largest supermarkets. We use
pre- levy data to predict the effect of the tax and corrobo-
rate our predictions by analysing actual consumption of
the composite good in the first 5 months post- levy. The
results show that the impact of the sugar tax is likely to
have the desired effect of reducing the purchase of SSDs.
Moreover, though the impact of the tax is likely to vary
across different geo- demographic segments, the evidence
suggests that its impact is likely to be greatest on the most
vulnerable market segments – families on low incomes –
who are among the highest consumers of SSDs in the UK.
KEYWOR DS
geo- demographic seg mentat ion, sugar tax, super market loyalty card
data, U K
JEL CL ASSI FICATION
D12; D22; I18; L66; M38; Q18; Q28

2
|
FEARN E Et Al.
1 | INTRODUCTION
Loyalty schemes have become established weapons in the armoury of retailers seeking to gain
competitive advantage through more effective product ranging, in- store merchandising and
promotional activity tailored to the (heterogeneous) behaviour of their customers. However,
there is little evidence to date of this rich source of behavioural data being used by policy-
makers to inform or evaluate the development of policies, legislation or (public) interven-
tions designed to foster more sustainable purchasing decisions and consumption behaviour.
One area in which there is a compelling need for behavioural change is diet and health. This
paper illustrates the potential value of behavioural insights derived from supermarket loyalty
card data for policy- makers and other stakeholders (food retailers, food manufacturers and
NGOs) who wish to foster healthier choices in the mainstream food purchasing environment
of supermarkets.
Previous studies (Felgate et al., 2012; Yamoah et al., 2014) have demonstrated an important
advantage that loyalty card data has over scanner data, which is the ability to analyse pur-
chasing behaviour by distinct consumer segments and the differential impacts of interventions
designed to change their behaviour, including changes in the retail price. Such analysis is not
possible with store- level scanner data, as no association is made between the sale of an item
and the person who purchased it. Loyalty card information is routinely and systematically
used by retailers and food manufacturers to inform decisions about marketing (pricing, dis-
tribution, ranging, and merchandising) all of which have an impact on purchasing behaviour.
Moreover, the information associated with loyalty cards enables inferences to be drawn re-
garding consumption, as loyalty cards are typically associated with single households with
known (geo- demographic) characteristics.
We have a particular interest in consumer behaviour in relation to prices because we want
to understand the differential impact of the sugar tax, introduced in April 2018 in the UK. The
tax was announced by the UK government in 2016 and is being applied to sugar- sweetened
drinks (SSDs). Excessive consumption of SSDs presents a real problem for public health as
they provide little nutritional benefit while contributing to weight gain and probably to the risk
of diabetes, cardiovascular heart disease and dental caries (Malik et al., 2006, 2010; Ng et al.,
2012; Vartanian et al., 2007).
Through the lens of normative economics, there are two key arguments in favour of the
tax. First, given that some people may disregard the effects of over- consuming SSDs on their
current and future health, being misinformed or prone to cognitive biases (e.g., time inconsis-
tency), the tax is likely to guide such individuals towards more rational and healthy behaviour.
Such effects are termed ‘internalities’ (Griffith et al., 2018; Gruber, 2002). Second, the tax is
likely to result in wider societal benefits through the reduction of the burden of diet- related
diseases on the National Health Service (NHS).
The UK sugar tax consists of a tiered levy with two bands, one of 18 pence per litre for
soft drinks with more than 5g of sugar per 100ml (low tier) and a higher one of 24 pence per
litre for drinks with more than 8g per 100ml (high tier). Other drinks with lower than 5g of
sugar are not taxed. It is the first time that such a tiered industry levy has been used, as other
countries have opted for a sales tax instead. It is hoped that the levy will help to tackle the na-
tion's obesity problem by reducing consumption of sugar in SSDs, particularly among younger
adults. However, there are concerns that it could become a regressive tax if it disproportionally
affects poorer members of society.
We analyse consumer behaviour pre- tax as well as their behavioural response as a result of
the introduction of the tax. To do so, we use two panel databases containing SSD purchase
data from over 2million loyalty card holders across the UK, divided into 10geo- demographic
segments, derived from a major UK supermarket chain.

|
3
IMPACT OF THE UK SOFT DRINK SUGAR TAX
The objectives of our study are twofold. First, we analyse consumer behaviour using
100weeks of pre- tax data and predict the effect of the tax on purchasing patterns of SSDs for
the different consumer segments. This allows us to establish a methodology for predicting the
effect of similar policies in the future. Second, we compare our predictions with the purchases
observed for the first few months (23weeks) after the implementation of the tax (April 2018),
as this provides corroboration of our predictive approach using measurements on the initial
effect of the policy.
Our approach is based on a composite SSD product using the prices of the principal prod-
ucts relevant to the tax to obtain price elasticities per consumer segment. Then we use these to
predict the effect of a price change equivalent to the imposition of the tax on each consumer
segment. This is compared with the actual purchasing changes before and immediately after
the imposition of the tax.
Our results are of interest to public health experts, to the UK and other governments con-
sidering similar policies, and to the public. They provide clear and important evidence about
the UK sugar tax on the purchase (and hence supply) of SSDs, particularly amongst the most
vulnerable consumer segments. Furthermore, they provide an exemplar for using supermarket
loyalty card data for other policy interventions designed to influence food and drink purchas-
in g b ehaviour.
The paper is organised as follows. Section 2 presents a review of relevant literature. Section 3
describes the materials and methods. Section 4 reports the results and, finally, Section 5 presents
our conclusions and recommendations for further research.
2 | LITERATURE REVIEW
The UK was the first country to introduce a tiered tax on SSDs (Briggs et al., 2017). Producers
and/or retailers are free to pass on the full tax to consumers (i.e., price pass- through) or to
absorb it, fully or in part. Worldwide, there is evidence of price pass- through from SSD taxes
ranging from 40% to more than 100% (Aguilar et al., 2019; Alsukait et al., 2020; Arteaga et al.,
2017; Berardi et al., 2016; Bollinger & Sexton, 2018; Capacci et al., 2019; Cawley & Frisvold,
2017; Cawley et al., 2018; Etilé et al., 2018; Falbe et al., 2015; Grogger, 2017; Rojas & Wang, 2017;
Seiler et al., 2021; Zhong et al., 2018).
There are concerns that a tax such as this could become regressive if it is passed through to
consumers and disproportionally affect the less aff luent social groups (Dubois et al., 2017). In
this context, a recent study proposed how to calculate the optimal soda tax range(s) (Allcott
et al., 2019) to strike the right balance between corrective and redistributive motives to avoid
the possible regressive nature of such taxes.
Another feature of the UK sugar tax design is to incentivise producers to reformulate
products— reducing the sugar content to avoid the tax. However, this represents a risk for the
manufacturer given that the new recipes may not be well received by consumers (Geyskens
et al., 2018; Gonçalves & Pereira dos Santos, 2020). Other countries (e.g., France from 2013
and Portugal from 2017) have also structured their SSD taxes to encourage reformulation
(Goiana- da Silva et al., 2018).
More than 40 countries have implemented SSD taxes and in some the implementation has
been delegated to regional authorities, such as Catalonia in Spain and California, Berkeley,
Boulder, Colorado, Philadelphia and Pennsylvania in the USA (Global Food Research
Program, 2020). Several studies have analysed the effect of the taxes on SSD consumption. The
evidence suggests that taxes decrease the SSD consumption from 6% in Mexico to more than
20% in Berkeley and Philadelphia and 33% in Saudi Arabia (Aguilar et al., 2019; Alsukait et al.,
2020; Arteaga et al., 2017; Castello & Lopez- Casasnovas, 2018; Cawley et al., 2019; Colchero
et al., 2021, 2017; Eleonora Fichera et al., 2019; Falbe et al., 2016; Nakamura et al., 2018;

4
|
FEARN E Et Al.
Seiler et al., 2021; Silver et al., 2017; Taylor et al., 2019; Zhong et al., 2018;). The differences in
the reported impacts could be due to a number of factors, including the quality of the data, the
methodologies employed but also the design and scale of the taxes used.
With respect to the type of data used to analyse the effect of the levy, the literature so far has
relied primarily on surveys, scanner data (e.g., Kantar World Panel) or hand- collected data
on a small number of products or in a small number of stores (Berardi et al., 2016; Castelló &
López- Casasnovas, 2018; Seiler et al., 2021), with a small number of studies having access to
retail data (Gonçalves & Pereira dos Santos, 2020).
In terms of methodologies applied, studies predicting and simulating the impact of the levy
ex ante use theoretical models (e.g., Briggs et al., 2013, 2017). Other studies analyse the impact of
the tax ex post using counterfactuals (e.g., Colchero et al., 2016); difference- in- difference anal-
ysis (e.g., Gonçalves & Pereira dos Santos, 2020) or synthetic control methods (e.g., Grogger,
2017 ).
Looking at the effect of the UK tax, Briggs et al. (2013) focused on predicting changes in the
number and percentage of overweight and obese adults post- tax. They used survey data from
various sources, which is problematic as applying results from one dataset to another requires the
assumption that the samples are drawn from the same population, which is not always the case.
The study predicted that the tax would reduce obesity by 1.3% with the greatest effect occurring
in young people. No significant differences were predicted between different income groups.
In another study, Briggs et al. (2017) simulated how the different industry responses to the
tax could impact on health. They included product reformulation to reduce sugar content and
avoid the tax; an increase in price when the levy is passed on to consumers; and a change in
market share as consumers switch to the lower sugar alternatives. Their findings suggested
that the greatest health benefit would come with the reformulation of the products. As a result,
individuals aged younger than 18years and those aged older than 65years would benefit the
most from the predicted reduction in obesity, diabetes and dental decay.
A recent study has assessed the effect of the UK tax on British households one year after the
implementation of the policy (Pell et al., 2021). To our knowledge, this is the f irst paper that esti-
mates the effect of the intervention post- tax. The authors first used pre- tax data to analyse con-
sumer behaviour and predict the effect of the tax using a counterfactual. Then, they compared
the predictions with the observed changes in volume and sugar intake one year post- tax. They
analyse a composite product to represent the high/low/no levy drinks. Their findings suggest that
in the consumption of high tax tier (the one we study) SSDs decreased by 44.3% and the associ-
ated sugar consumption decreased by 45.9%. They also identified decreases in the consumption
of low tier drinks but no changes in volumes consumed for drinks that did not attract the tax.
Overall, considering all soft drinks, they found that the total volume consumed did not change
but associated sugar consumption fell by 9.8%. The authors conclude that reduction in sugar was
most likely the result of reformulation. The study used household scanner data from a panel of
households reporting their purchasing on a weekly basis (Kantar Worldpanel). However, their
data do not allow for analysis of differential effects among different consumer segments.
Our study is the first to analyse the differential impact of the levy across a diverse range of
(geo- demographic) consumer segments, using supermarket loyalty card data.
3 | MATERIALS AND METHODS
3.1 | Data
The data used for this study are obtained from one of the UK’s largest supermarkets. Their
loyalty scheme generates a panel dataset of over two million households, which is a 10% sample
of the population of loyalty card holders. A considerable advantage of loyalty card data over

|
5
IMPACT OF THE UK SOFT DRINK SUGAR TAX
TABLE 1 Su mmary of Cameo seg mentation
Cameo group Description
% of UK
households
% of customers
(Period −1)
% of SSD customers
(Period −1)
% of customers
(Pe riod−2)
YAS Young and affluent singles 3.5 2.2 2.3 3
WRN Wealthy retired neighbourhoods 3.6 3.3 2.7 3.8
AHO Aff luent homeowners 11.4 1 2.7 10.8 12.5
SPFH Sma ller private family homes 13.7 15. 8 14.0 14.3
CMN Comfortable m ixed neighbou rhoods 9.5 9.4 8.9 10.5
LAF Less aff luent famil ies 13.9 15.8 15.6 13.4
LASS Less a ff luent si ngles and students 6.1 5.8 6.4 9. 3
PWBCW Poorer white - and blue- collar workers 15.7 14.6 15.7 13.3
PFSPH Poorer families and single parent households 10.9 11. 2 13.0 11 .1
PCT Poorer council tenants 11.9 9. 2 10.6 8 .7

6
|
FEARN E Et Al.
data from a typical household survey is the level of product disaggregation: it provides access
to information on both prices and purchase quantities for thousands of individual products by
specific consumers (loyalty card holders) on a weekly basis. Hence, for each geo- demographic
consumer segment we have weekly sales by volume, weekly prices and weekly number of cus-
tomers for the entire soft drinks category, comprising hundreds of products and dozens of
brands. Since for SSDs we have unit size in millilitres, we can calculate the total volume of
sales in litres, litres per customer and price per litre for each product.
The sales data is segmented by 10 different geo- demographic segments, using the classifica-
tion provided by Cameo, which groups households in neighbourhoods that share certain char-
acteristics, such as life stage, lifestyle, affluence, ethnicity and employment. (Cameo, 2020). The
resulting segments are presented in Table 1, in order of affluence, and comprise:aff luentsingles
and couples in exclusive urban neighbourhoods (YAS);wealthy neighbourhoods nearing and en-
joying retirement(WRN); affluent home owning couples and families in large houses (AHO);
suburban homeowners in smaller private family homes (SPFH); comfortable mixed tenure neigh-
bours (CMN); less aff luent family neighbourhoods (LAF); less aff luent singles and students in
urban areas (LASS);poorerwhite- and blue- collarworkers (PWBCW); poorer family and single
parent households (PFSPH); and poorer council tenants including many single parents (PCT).
Geo- demographic segmentation dates back to the mid- 1970s and is defined by Birkin and
Clarke (1998) as ‘the study of the population types and their dynamics as they vary by geographical
area’. What distinguishes geo- demographic segmentation from other segmentation approaches is
that the unit of analysis is the neighbourhood rather than the individual. The fundamental ratio-
nale is that ‘the social context in which people live [has] a significant effect on their consumption
patterns as well as their attitudes, values’ (Webber, 2004). Cameo is one of a number of commer-
cial market research agencies that use a variety of data sources, including Census data, Household
Council Tax Band and Property Valuation Data, Consumer Credit data, and residency data from
the Electoral Roll, to classify every UK household into distinct marketing types.
Table 1shows the population stratification in percentages for the UK per Cameo segment
as well as the comparative size of the representative samples in our two SSD datasets, Period- 1
and Period- 2 as described below. We observe that our segmentation data is a good representa-
tion of the UK population. Groups such as AHO, SPFH and PFSPH (representative of larger
household sizes, e.g., families) are slightly over- represented in our sample data compared with
the UK population. There is also some small variation in the proportion belonging to some
groups (e.g., SPFH or LASS) from Period- 1 to Period- 2.
We used two separate but related datasets; they both include the weekly sales (by volume),
weekly number of customers and weekly prices for the top selling products in the soft drinks
category segmented by Cameo. The first dataset (Period- 1) contains data over a period of
100weeks, from June 2014 to May 2016 (i.e., period before the announcement of the tax). The
second dataset (Period- 2) contains data over a period of 104weeks from September 2016 to
September 2018 (i.e., after announcement and 5months post implementation of the tax). The
data from Period- 1 was used to estimate the model parameters and predict sales if the tax had
been imposed over that period. We then used the data from Period- 2 to measure the actual
impact of the tax. For the latter, we compared sales during the 23 weeks from 3 April to 4
September 2017 (before the tax) with sales for the corresponding period 12months later, from
2 April to 3 September 2018 (immediately after the tax).
3.1.1 | Period- 1 dataset
Our Period- 1 dataset contained a sample with over 700 SSDs. Many of the products in this
sample would be non- taxable given their sugar content in the initial period (i.e., <5mg/L), or
at the time of implementation, as many companies reduced the sugar content to avoid the tax.

|
7
IMPACT OF THE UK SOFT DRINK SUGAR TAX
However, the two top selling brands, Coca- Cola and Pepsi, did not change their formulations,
so provide us with a suitable sub- group for our analysis (high levy SSDs). The Coke products
contained sugar at a level of 10.6g/100ml and the Pepsi products 11.0g/ml. Of the total (64)
Coca- Cola and Pepsi products listed, the top seven, all of which contained sugar levels that
would attract the higher tax rate, accounted for 74% of the total volume of Coke and Pepsi
sold in the period, 55% of total customers for all SSDs and 63% of the total volume of SSDs
sold over the period. The balance of customers/volume included SSDs which would not have
attracted the tax (lower tier) represented a very small market share. We therefore used the
concept of a composite product for our analysis, focusing on those top seven high tax SSDs
which make the modelling manageable. Table 2shows summary statistics for the products
that make the composite in time Period- 1. Table 2 indicates that the products showing high
variation in price generally showed high variability in weekly volume and customers.
3.1.2 | Period- 2 dataset
The original dataset for the second period contained slightly over 800 products, of which 47
were Coca- Cola and 10 were Pepsi products, all of which were taxable at the high rate. The
data was for the 104weeks from 12 September 2016 to 3 September 2018. Prices of individual
products over this period were far more stable than in Period- 1, but many products were re-
moved and new products introduced, generally in smaller pack sizes. For example, 10 Coca-
Cola products that were selling in 2017 were unavailable during the 2018 post- levy period and
17 new products were introduced after the introduction of the levy. Of the 10highest- volume
products in the post- tax period, only one (Pepsi 2L) had significant sales prior to the imple-
mentation of the tax and eight were not available at all. For this reason, the composite product
in Period- 2 cannot be the same as in Period- 1. We use Period- 2 to analyse the actual changes
pre- and immediately post- tax, hence we can consider all 57 Coke and Pepsi products to be our
Period- 2 composite dataset. The summary statistics for the composite product in Period- 2 are
presented in Table 3 at the aggregate level as there are too many products changing over time
to present them individually. We present pre- and post- levy figures separately for comparison.
Table 3shows that from the pre- tax period (April– September 2017,) to the post- tax period
in 2018 there was a reduction in volume of the composite Coke+Pepsi product purchased of
33.6%, a reduction of 20% in customers purchasing and a reduction of 16.6% in litres per cus-
tomer purchased. This translates to a reduction of 16.8% in sugar purchased/week as the bal-
ance of Coke and Pepsi changed slightly between the two periods. The large reduction in total
volume may be associated with a reduction in package volume seen as a result of the tax. The
mean price of the composite product set increased by 39% after the introduction of the tax.
The expected increase was between 11% and 39% per product, depending on the original price,
with the most expensive products expected to see the smallest percentage increases. Hence, we
observed a high price pass- through for most products.
3.2 | Modelling SSDs as a composite good
The key feature of our modelling strategy is that the dependent variable in the empirical analysis
is not demand for individual SSDs, but rather the total demand for the composite good represent-
ing all SSDs. There are a number of justifications for this choice of approach. First, the focus
of the study is the impact of the tax on total demand for SSDs, not the demand for individual
SSDs. Second, the form of the econometric model developed here makes it very easy to predict
the effect of a proportionate change in the price of every individual good, which is exactly what
is required when considering the effect of the sugar tax. Third, using the composite good is a way

8
|
FEARN E Et Al.
TABLE 3 Su mmary statistics for the composite product set in Period- 2
Time period
Mean weekly sales
volume (L) Mean price (£/L)
Mean weekly
customers
Mean weekly purchase per
custo mer (L)
Mean weekly sugar
purchase per customer (g)
April– Sept. 2017
(p re - lev y)
1,658,102 £0.93 545,708 3.038 323
April– Sept. 2018
(p os t- l ev y)
1,101,629 £1.29 434,835 2.533 269
%Diff −3 3.6% +39% −2 0. 3% −16 .6 % −16.8 %
TABLE 2 Su mmary statistics for the composite product set in Period- 1
Var i ab l e Vari abl e Mean (CV) price (£/ L)
Mean (CV) weekly sale s
volume (L)
Mean (CV) weekly
customers
Mean (CV) weekly purchase
per customer (L)
p1Coca- Cola 1.75 L £0.68
(36%)
597,53 0
(40 %)
147,323
(37%)
4.1
(11%)
p2Coca- Cola 500ml £2.03
(12%)
56,075
(14%)
81,8 05
(13%)
0.7
(1%)
p3Pepsi 2 L £0.62
(39%)
216, 393
(62%)
57,821
(63%
3.7
(17%)
p4Coca- Cola 8 × 330m l £1.03
(23%)
161,5 64
(37%)
42,892
(28%)
3.8
(9%)
p5Coca- Cola 24 × 330ml £0.84
(29%)
351,109
(120 %)
36,782
(114%)
9.5
(7%)
p6Coca- Cola 330ml £1.97
(12%)
15,945
(12%)
34,127
(12%)
0.5
(1%)
p7Coca- Cola 1.25 L £0.89
(27%)
47,15 6
(52%)
28,130
(49%)
1.7
(8%)
PAll Coke +Pepsi £0.76 1,4 45,772 428,880 3.4

|
9
IMPACT OF THE UK SOFT DRINK SUGAR TAX
of side- stepping the problem, raised in Section 3.1.2, of individual products changing over time.
Theoretical underpinnings for the composite product approach have been provided by Lewbel
(1996) and the approach has been used recently in a context very similar to ours by Pell et al. (2021).
The principal explanatory variable is the weighted average of the prices of the individual
products, in the form of an index. Total demand for SSDs is measured using two variables: the
total number of customers purchasing one or more SSD; and the total volume (in litres) of
SSDs purchased. When we model the former, we are focusing on the ‘extensive margin’, that is,
the impact of price changes and taxes on the size of the market; when we model the latter, the
focus is on the ‘intensive margin’, that is, the impact on the behaviour of households who are
consumers both before and after the change.1
The model was estimated for each of the geo- demographic (Cameo) segments separately.
However, to minimise notational complexity, we do not include Cameo subscripts in the fol-
lowing specification of the model.
Let Nt be the total number of households who purchase SSDs in week t. Nt is one of our
chosen measures of demand for the composite good. The other measure is the total weekly
volume of sugary drinks purchased over all the loyalty card holders. To obtain this, we simply
sum the weekly volumes (in litres) of individual SSDs. Hence, if there are J sugary drinks, the
total volume in period t is:
where vjt is volume purchased of drink j in period t.
To obtain a price index for the composite good, we specify a weighted geometric mean of
the J individual product prices. That is, the price of the composite good in period t is assumed
to be:
where pjt is price per litre of drink j in period t, and αj, j = 1,···,J are parameters. Note that the αj
parameters capture the importance of each SSD in the budget, and also the responsiveness of
consumers to changes in the prices of each SSD.2 Thereason forthe minus sign applied to each of
theαj parametersis explained below.
Let Qt be the measure of demand under consideration; this will be one of the two measures
Nt and Vt, defined above. We assume that the demand function for the composite good has the
following reciprocal for m:
1A problem w ith the ‘number- of- cu stomer s’ vari ableis th at, although we k now the numb er- of- cus tomers purchasing e ach
individua lproduct, we do not know the numb er- of- cust omers purcha sing th ecompo siteproduct, wh ichis the focus of our
analy sis. Ou r chose nmeasure of numb er- of- customers i s the su m ofnumbe r- of- cu stome rs over products, but we acknowledge that
this r epresents anupper b oundfor the a ctual numbe r- of- custome rs(as a conseque nce ofdoub le- counti ngof house holds). In
contrast, for tot al volume of the composite g ood, we have an acc urate m easu re. We could ob tain volume- pe r- c ustomer by dividing
total volume by our estimate of the number- of- cu stomer s. However, th is would i ntroduce mea surem ent error, and thi s why we use
total volume in t he anal ysis of th e inten sive margin.
(1)
V
t=
J
∑
j
=
1
vjt t=1, …,T
(2)
P
t=
J
∏
j
=
1
p−𝛼j
jt t=1, …,
T
2The as sumpt ion that t he αj para meter s are f ixed over time g uara ntees the exogeneity of the pric e index defi ned in E quatio n (2), in
a model of c omposite consumpti on.
(3)
Q
t=
exp (𝛼0)
Pt
=exp (𝛼0)
J
∏
j
=
1
p𝛼j
jt t=1, …,T

10
|
FEARN E Et Al.
where the second equality is obtained using Equation (2). Note that theαjparameters now
have positive signs. Taking logs of both sides of Equation (3), we obtain the log- linear
equation:
Equation (4) makes it clear that the αj parameters represent the importance of each individ-
ual price in determining demand for the composite good.
Another important measure is the sum of all the αj parameters:
The quantity η defined in Equation (5) has the interpretation of a price elasticity: if the
prices of all J of the component goods rise by 1%, the quantity demanded of the composite
good will change by approximately η%. We expect η to be negative.
We also include a time trend variable to allow for changes in tastes over the sample
period. There is an upward spike in demand during the Christmas periods for most goods
including all carbonated drinks, so to allow for abnormal purchasing behaviour a set of
three ‘Christmas dummies’ (C1– C3) are also included, representing the first, second and
third week of the Christmas period. Finally, we add an error term. The resulting linear
regression equation is:
Under the tax, companies pay 18 p per litre if the product contains more than 5g of sugar
per 100ml, and 24 p per litre if it contains 8g of sugar per 100ml. Assuming that the tax is
passed fully to consumers, the post- tax prices (per litre) can be easily computed as:
where τj is either 0, 0.18 or 0.24, depending on the rate at which drink j is being taxed. For
all our seven products in Period- 1 (used to build the model) the amount of sugar is higher
than 8g per 100ml, hence the higher levy applies (i.e., £0.24/L). Assuming that all of the tax
is passed to consumers, the new prices are presented in Table 4 together with the percentage
increase over the price per litre they represent. Note that given the nature of the tax as a
fixed amount per litre, percentage increases are much more noticeable for the products with
lower prices per litre.
Finally, we predict the impact of the tax by combining the estimates from Equation (6) with
the assumed after- tax prices from Equation (7) to obtain the predicted consumption under a
scenario in which the tax is applied over the sample period. This prediction is obtained using:
(4)
ln
Qt=𝛼0+
J
∑
j
=
1
𝛼jlnpjt t=1, …,T
(5)
𝜂
=
J
∑
j
=
1
𝛼
j
(6)
ln
Qt=𝛼0+
J
∑
j
=
1
𝛼jlnpjt +𝛽t+
3
∑
k
=
1
𝛾kCkt +𝜀tt=1, …,T
(7)
p
TAX
jt
=pjt +𝜏jj=1, …,Jt=1, …,T
(8)
ln

Qt=𝛼 0+
J
∑
j
=
1
𝛼 jln pTAX
jt +
𝛽t+
3
∑
k
=
1
𝛾 kCkt t=1, …,T

|
11
IMPACT OF THE UK SOFT DRINK SUGAR TAX
where hats indicate estimates of the parameters in Equation (6), and p
TAX
jt
j=1,...,J are defined in
Equation (7). Equation (8) gives an unbiased prediction of ln Qt. To convert this to an unbiased
prediction of Qt, we apply Duan's smearing method (Duan, 1983). This is a non- parametr ic method
that provides consistent predictions whatever the distribution of the error term in Equation (6).
We compare this prediction of the tax effect from Period- 1 data to the actual consumerre-
sponse observedafter the tax using Period- 2 dataset. These comparisons are made in both
absolute and relative terms, separately, for each Cameo segment.
4 | RESULTS
In accordance with Equations (5) and (6), the full results of the model, showing the coeffi-
cients/elasticities for each term, are presented in the Appendix S1. Inorder totest the accuracy
of the model we also partitioned the Period- 1 dataset into two equal halves. The f irst 50 data
points we used to generate another model, which we tested against the second 50 data points
to obtain the in- sample and out- of- sample R2respectively. The out- of- sample R2 values are
significant in each case. These values are shown in the Appendix S1.
A summary of results with theoverallprice elasticities for each of the Cameo segments are
presentedinTable 5. Results include elasticities obtained from Equation (5); volume and cus-
tomers pre- tax (Pre- tax P- 1); predictions obtained using Equation (8) after applying Duan's
smearing (Post- tax P- 1), assuming the full pass- through of the tax; % change predicted (%
change P- 1); and actual percentage change observed in Period 2 (% change P- 2). Those are
presented for volume (intensive margin) and number of customers (extensive margin).
Looking at the predicted percentage change (P- 1) and actual percentage change (P- 2), we
note that the predictions are good for most groups but over- predict for YAS and WRNand to
a lesser extent for LASS. Recall from Table 1 that those Cameo groups contain the smallest
proportions of households and are under- represented in our dataset, which may make them
harder to model. Focusing on other groups, we see that price elasticities in Table 5 tend to rise
over the Cameo groups; that is, less affluent groups tend to be more sensitive to price changes.
The least aff luent group (PCT) has the largest elasticities: – 1.9 at the intensive margin; and – 1.4
at the extensive margin.
In Period- 1, we observe that the tax is predicted to decrease both volume and number of
customers for all Cameos, with the rate of decrease generally higher for less aff luent groups.
The predicted percentage reduction in volume is higher than the predicted reduction in number
of customers. This may be due to the additional reduction in volume that was caused by man-
ufacturers reducing the volume sold as a result of the tax (e.g., 1.75L bottle becomes 1.50L).
Table 5shows that Cameos YAS, WRN, LASS, and PCT are predicted to be highly affected by
TABLE 4 Prices in Per iod- 1 before and aft er the tax, assumi ng the ta x is ful ly passed on
Product
Average price before tax
(£/ L)
Average price after tax
(£/ L)
%
increase
Coca- Cola 1.75 L 0.68 0.92 35.5
Coca- Cola 500ml 2.03 2.27 11.8
Pepsi 2 L 0.62 0.86 38.6
Coca- Cola 8 × 330m l 1.03 1.27 23.4
Coca- Cola 24 × 330ml 0.84 1.08 28.5
Coca- Cola 330ml 1.97 2.21 12.2
Coca- Cola 1.25 L 0.89 1.13 26.9

12
|
FEARN E Et Al.
TABLE 5 Predicted i mpact i n Period- 1 (P- 1) of the sugar tax and a ctua l impact in Per iod- 2 (P- 2) on mean weekly total volu me and mean weekly number of customers by
geo- demographic (Cameo) segment
Cameo Segment
Mean weekly volume (l) Mean weekly customers
Elasticity
Pre- tax
(P−1) Post- tax (P−1)
% change
(P−1)
% change
(P−2) Elasticity
Pre- tax
(P−1) Post- tax (P−1)
% change
(P−1)
%
change
(P−2)
YAS −1.6 32 ,9 41 20,195 −38 .7 −24.9 −0.9 9,678 7,2 31 −25 .3 −12. 2
WRN −1.4 39, 817 24,551 −3 8.3 −31.2 −0.6 11, 576 8,719 −24 .7 −17.8
AHO −0.2 14 7,166 113,213 −23 .1 −31. 3 −0.02 4 6,14 5 37,755 −18. 2 −19. 2
SPFH −0.9 189,615 128,641 −32 .2 −32. 2 −0.3 60,078 46 ,733 −2 2.2 −2 0. 2
CMN −1.1 126 ,803 86,451 −31.8 −32 .1 −0.6 38,318 28,668 −25 .2 −18.9
LAF −0 .7 221,983 156,619 −29. 4 −35. 3 −0.3 66,724 51,837 −2 2.3 −23.4
LASS −1.4 9 5,763 59,019 −3 8. 4 −33.2 −0.8 27,465 20,618 −24.9 −18.7
PWBCW −1. 2 229,521 151,181 −34 .1 −35. 0 −0.6 67, 4 45 51,4 97 −23.6 −21.0
PFSPH −1. 2 196,077 129,748 −33.8 −35 .0 −0.6 55,839 42,356 −24.1 −2 0.9
PCT −1.9 16 6, 08 7 100,127 −3 9.7 −35.7 −1. 4 45, 613 31,515 −30.9 −21.9
Tot a l 1,4 45,772 969,744 −32 .9 −33. 6 428,880 326,930 −2 3. 8 −20 .3

|
13
IMPACT OF THE UK SOFT DRINK SUGAR TAX
the tax, as they show greater reductions both in volume and number of customers. The compos-
ite product had a mean sugar content of 10.66g/100ml (weighted by the balance between Coke
and Pepsi in Period- 1), and we therefore expect the changes in sugar purchased to be the same as
the predicted volume change, that is, 32.9%. For the actual changes (P- 2) portrayed in Table 5,
we see an inverse relationship between the affluence of the Cameo segments and percentage
change in volume purchased. This means that less aff luent groups show a stronger response.
A further test of this trend is obtained by performing a weighted linear regression with %
change in volume purchased as the dependent variable and Cameo group as the explanatory
variable, and with initial mean of total volume or number of customers as weights.3 The results
of these regressions are presented in Table 6.
The large negative intercept in these regressions indicates that even the most affluent groups
show a large negative response to the tax. The (significantly) negative slope confirms that the
response to the tax is even higher for less affluent groups.
5 | DISCUSSION AND CONCLUSIONS
In this paper we use a rich data source from a major UK retailer to estimate a demand equa-
tion over the 100weeks before May 2016 for the most popular SSDs treated as a composite
good. We have utilised these estimates to predict the behavioural response after the tax, not
only at aggregate level, but also across 10geo- demographic (Cameo) consumer segments. We
compare our predicted results with actual consumer response during the 5months after the
implementation of the tax and calculated both the actual effect of the tax at aggregate level and
by geo- demographic group. This allowed us not only to assess the effectiveness of the policy,
but also to corroborate our predicted results.
We show that, both in our predictions and the actual post- tax analysis: (1) mean weekly vol-
ume of SDDs purchased (i.e., intensive margin) reduced by 32.9% and 33.6%, respectively; and
3The reg ressions who se results ar e shown in Table 6 i mplicitly assume the ran king of Cameo g roups th at is routi nely as sume d by
others working w ith Cameo dat a (see,e.g.,Revor edo- Giha et al., 2009), and al so that t here is a n equal d ista nce be tween ranks. The
latte r assu mption c ould be avoided if we u sed a non- parame tric m easu re of associat ion, such as Spea rma n correlation. However,
the gr eat adva ntage of u sing a regres sion in this ca se is th at weare abl e toincorporate the we ights of e ach Cameo group
(usingweighted regre ssion) and thiswe ighti ngis imp orta nt. We therefore prefer t he regression approachfor the cu rrent p urpose,
even thou gh weac cept t hat the assumptions underlying this regr essio n may not be fu lly met.
TABLE 6 Results of weighted line ar regression— dependent variable: % i mpact of tax; independent var iable:
cameo group identifier (1– 10)a
Vol u me Customers
Predicted
Constant −26.43*** (3.73) −18.56*** (2.29)
Cameo −1.02* (0. 55 ) −0.83** (0.34)
N10 10
R20.30 0.43
Actual
Constant −28.98*** (1.11) −17.37*** (1.62)
Cameo −0.72 ** (0.16) −0.47* (0.24)
N10 10
R20.71 0.32
aStand ard errors in br ackets. *p < 0.1; **p < 0.05; ***p < 0.01.

14
|
FEARN E Et Al.
(2) mean weekly number of consumers that purchased SSDs (i.e., extensive margin) reduced by
23.8.% and 20.3%, respectively. A differential analysis, enabled by the loyalty card data, shows
that the tax had a significant impact on all consumer segments. However, the greatest impact
was felt by the less aff luent consumer segments.
We find, first, that the pattern of purchasing of soft drinks varies significantly between
consumer groups, with less affluent groups tending to be more likely to purchase SSDs. We
then analyse the demand for the composite good, made up of the seven most popular taxable
SSDs in Period- 1 and f ind that it appeared to be more sensitive to some prices than to others,
according to the estimates obtained.
Based on demand in Period- 1, we generate predictions of total mean weekly volume pur-
chased and total number of customers purchasing after the introduction of the tax. Our predic-
tions show a clear reduction in purchasing post- tax, both in terms of total volume and number
of customers. We also see that this negative impact of the tax is not homogeneous across
Cameo segments, being more marked for less affluent segments. Of all the Cameo groups, it
was the least aff luent (Poorer Council Tenants) that showed the highest percentage decrease in
purchasing (at both intensive and extensive margins).
Focusing on the percentage change in purchases, we predicted an overall decrease of 32.9%
in volume (intensive margin) and a decrease of 23.8% in the number of customers (extensive
margin), of the composite good across all Cameo groups. Overall, these results suggest that
the policy intervention would have a significant impact on SSD purchasing as expected by the
policy- makers.
This predicted overall decrease is consistent with the observed change in purchasing be-
haviour found in the 5months post- levy, which was a 33.6% reduction in volume and a 20.3%
reduction in customers. Our predictions for the individual Cameo groups showed a very good
match to actual reductions for SPFH, CMN, PWBCW, PFSPH, and to an extent for PCT. We
over- predicted for YAS and WRN, which are small groups and for LASS. We under- predicted
for AHO and LAF. The two groups for which the prediction error is highest appear to be the
first two (YAS and WRN); as previously remarked, this may be associated partly with the
higher sampling error arising from these being the smallest groups but may also be a result of
unobservable changes in the composition and/or the behaviour of these groups between the
period of estimation and the time at which the tax was introduced. Overall, we can conclude
that the model appears to be a good tool to predict buyer behaviour after the introduction of
the levy.
We have made an important distinction between the extensive and intensive margins. The
impact of the levy at the extensive margin is measured by the decrease in the number of cus-
tomers resulting from the introduction of the levy, and as previously noted this is 20.3%. The
impact on total volume is clearly higher than this, at 33.6%. The fall in volume per customer
which in real terms was about 16.6%, is what is meant by the intensive margin.
Our results show the policy intervention had more impact on the less affluent (higher rank)
Cameos compared with the more affluent ones (lower rank). However, because the group sizes
are quite different, we used a weighted regression to test this result. The results confirmed the
apparent pattern. When observing the effect actually seen in Period- 2, which included post- tax
data this differential impact was indeed significant.
It is worth noting that the levy being passed in full was expected to have an effect of an in-
crease in the average price per litre of between 11% and 39% depending on the original price,
with the more expensive products expected to see the smallest percentage increases. In fact,
in the observed period after the implementation, the average price of the composite good in-
creased by about 39%. This is consistent with the tax being passed in full and some over-
shifting of the tax, resulting in prices per litre higher than expected after the introduction of
the tax.

|
15
IMPACT OF THE UK SOFT DRINK SUGAR TAX
Our results are broadly consistent with the results reported by Pell et al. (2021) who also
analysed the effect of thetax in the UK one year afterthe implementation of the levy. They
found the high- tier SSDs (the comparable composite) purchased volumes decreased by 44.3%
and sugar purchased decreased by 45.9%. This is in line with our total observed 33.5% de-
crease in volume. The differences might be due tothe diversetime period considered in the
analysis(5months vs. 1year post- tax) and their population being biased towards lower income
groups for which we observed the largest reductions.
Finally, we conclude that loyalty card data affords us the possibility of a unique insight into
consumer purchasing behaviour and may enable us to study the impact of other policies (in-
volving taxation or other measures) aimed at influencing consumption behaviour on the food
and drinks market. This is one of the first evaluations of the impact of the sugar tax policy on
the UK and provides positive lessons for other ‘sin taxes’.
5.1 | Policy implications
Overall, from our predictions and from the observed period post- tax implementation, we con-
clude that the soft drinks tax is likely to have an overall sizeable negative impact on SSD
purchases (initially −32.9% in volume and −23.8% in customer numbers). The most important
aspect of our results is that the impact of the policy intervention is likely to be greatest on those
consumer segments with the highest propensity to purchase SSDs, who are also segments char-
acterised as less affluent— including single person households and less affluent families. This
indicates that the policy is not regressive, as the least aff luent consumers would not pay a
higher penalty by not changing their behaviour; instead, they appear to be more price sensi-
tive and are reducing their purchases of SSDs to a greater extent than more aff luent shopper
segments. This is an important point for governments and for society in general as further ‘sin
taxes’ are considered.
Our predicted negative impact on SSD purchasing behaviour is clearly a desirable re-
sult from the point of view of policy- makers, who introduced the soft drinks levy with
the intention of reducing the consumption of SSD and thereby the sugar intake of the UK
population.
5.2 | Limitations
An important limitation of our data is that we are able to observe Cameo segment purchasing
behaviour but not the behaviour of individual customers, nor their consumption behaviour.
We assume they consume what they purchase but this could be over varying periods of time.
Higher purchasing may be accounted for by a large family structure resulting in lower indi-
vidual consumption. Indeed, as households may vary in size and composition and we do not
have precise details for those, it would be difficult to translate our analysis to an individual
prediction per consumer or to see how it affects population across specific demographic char-
acteristics (e.g., old vs. young people).
ACKNOWLEDGEMENTS
The authors would like to acknowledge the constructive feedback received from the editor and
the anonymous reviewers, prompting changes which significantly improved the paper.
ORCI D
Andrew Fearne https://orcid.org/0000-0003-4910-046X

16
|
FEARN E Et Al.
REFERENCES
Agui lar, A., Gutierrez , E. & Seira, E. (2019) The effectiveness of sin food taxe s: evidence from Mexico. Instituto
Tecnológic o Autónomo de Méx ico (ITAM) Working Paper
Allcott, H., Lockwood, B. & Taubinsky, D. (2019) Regres sive Sin Taxes, with an Application to the Optimal Soda
Tax . The Qu arterly Journal of Economics, 134(3), 1557– 162 6.
Alsukait, R., Wilde, P., Bleich, S.N., Singh, G. & Folta, S.C. (2020) Evaluating Saudi Arabia’s 50% carbonated dri nk
excise tax: Changes in pric es and volu me sales. Econ Hum Biol, 38(100868), 10– 1016.
Arteaga, J.C., Flore s, D. & Luna, E. (2017) The effect of a soft- dr ink tax in Mex ico: a time ser ies approach.
Berardi, N., Sevestre, P., T'epaut, M. & Vigneron, A. (2016) The impact of a “soda tax” on prices: evidence from
french micro d ata. Applied Economics, 48(41), 3976– 3994.
Birki n, M. & Clarke, G. (1998) GIS, geodemographics, and spatial modeling in the U.K. financial ser vic e industry.
Journal of Housing Research, 9(1), 87 – 111.
Bollinger, B. & Sexton, S. (2018) Local excise taxe s, sticky prices, and spil lovers: ev idenc e from Berkeley's soda tax.
Sticky Pric es, and Spillovers: Evidence from Berkeley's Soda Tax (January 12, 2018).
Briggs, A., Mytton, O., Kehlbacher, A., Tiffin, R., Rayner, A. & Scarborough, P. (2013) 2013 Overall and income
specific effect on prevalence of over- weight and ob esity of 20% sugar sweetened drink tax in u k: econometric
and comparative risk as sessment modelli ng study. BMJ, 347, f6189.
Briggs, A., Mytton, O., Kehlbacher, A., Tiffin, R., Elhussei n, A., Rayner, M. et al (2017) Health impact asse ssment
of the UK soft drinks industry levy: a c omparative risk asse ssment modelling study. The Lancet Public Health,
2(1) , 15 – 22 .
Cameo. (2020). https://www.trans union.co.uk/produ ct/cameo
Capac ci, S., Al lais, O., Bonnet, C. & Ma zzocchi, M. (2019) The impact of the French soda tax on price s and pur-
chases. An Ex Post Evaluation. PLoS One, 14(10), e0223196.
Castelló, J.V. & López- Casasnovas, G. (2018) Impact of SBB taxes on consumption. CRES- UPF Working Pap er,
201804 , 110.
Cawley, J. & Frisvold, D.E. (2017) The pas s- through of t axes on sugar- sweetened beverages to retai l pric es: the c ase
of Berkeley, Califor nia. Journal of Policy Analysis and Ma nagement, 36(2), 303– 326.
Cawley, J., Frisvold, D., Hill, A. & Jones, D. (2019) The impa ct of the Ph iladelphia beverage tax on purchases and
consu mption by adults and children. Journal of Health Economics, 67, 102225.
Cawley, J., Willage, B. & Frisvold, D. (2018) Pass- through of a ta x on sugar- sweetened beverages at the Philadelphia
International Airport. JAMA, 319(3), 305 – 306.
Colchero, M., Popkin, B., Rivera, J. & Ng, S. (2016) Beverage purchases from stores in mexico under the excise tax on
sugar sweetened beverages: observational study. BMJ, 352, h6704. http://dx.doi.org/10.1136/bmj.h6704
Duan, N. (1983) Smeari ng esti mate: A nonparametric retransformation method. Journal of the American Statistical
Association, 78(383), 605– 610.
Dubois, P., Grif fith, R. & O’Connell, M. (2017) How well target ed are soda taxes? Toulouse School of Economics
(TSE) Working Pap er, December 2017, 17- 868.
Eleonora Fichera, E., Mora, T., Lopez- Valcarc el, B., & Roc, D. (2021) How do consumers respond to ‘sin taxes’? New
evidence from a tax on sugary drinks. Social Science & Medicine, 274, 113799.
Etilé, F., Leco cq, S. & Boizot- Szantai, C. (2018). The Inc idence of Soft- Drink Taxes on Consu mer Prices and Welfare:
Evidence from the French" Soda Tax".
Falbe, J., Rojas, N., Grummon, A.H. & Madsen, K.A. (2015) Higher retai l pric es of sugar- sweetened beverages
3 months after implement ation of an excise t ax in Berkeley. California. American Journal of Public Health,
105(11), 2194 – 2 201.
Falbe, J., Thompson, H.R., Becker, C.M., Rojas, N., McCulloch, C.E. & Madsen, K.A. (2016) Impact of the
Berkeley excise tax on sugar- sweetened beverage consumption. American Journal of Public Health, 106(10),
1865 – 18 71.
Felgate, M., Garcia, M. & Fearne, A. (2012) Using supermarket loyalty card dat a to explore the moderating
impact of shopper segments on promotional response. International Journal of Market Research, 54(2),
221– 240.
Geyskens, K., Grigoriev, A., Holtrop, N. & Nedelko, A. (2018) Optimal p olicy design for the sugar tax. arX iv pre-
pri nt arX iv:1810.07243.
Global Food Research Program. (2020) Sugary drink taxes arou nd the world.
Goiana- da- Silva, F., Nunes, A.M., Miraldo, M., Bento, A., Breda, J. & Araújo, F.F. (2018) Using pricing policies to
promote public health: the s ugar sweete ned beverag es taxation ex perienc e in Portugal. Acta Medica Portuguesa,
31(4), 191– 195.
Gonçalves, J. & Pereira dos Sa ntos, J. (2020) Brown sugar, how come you taste so good? The impact of a soda ta x on
pric es and consumption. Social Science & Medicine, 2 64,113332.
Griffith, R., O’Connell, M. & Smith, K. (2018) Corrective taxation and internalities from food consumption. Cesifo
Economic Studies, 6 4(1), 1 – 14 .

|
17
IMPACT OF THE UK SOFT DRINK SUGAR TAX
Grogger, J. (2017) Soda taxe s and the prices of sodas and other drinks: Evidence from mexico. American Journal of
Agricultural Economics, 99(2), 481– 498.
Gruber, J. (2002) Smoking’s internalities. Regulation, 25, 52.
Lewbel, A. (1996) Aggregation without separability: a general ize d composite com modit y theorem. The A merican
Economic Review, 86(3), 524– 543.
Malik, V.S., Popkin, B.M., Bray, G.A., Després, J.P., Willett, W.C. & Hu, F.B. (2010) Sugar- sweetened beverages and
risk of metabolic syndrome and t ype 2 d iabetes: a meta- analysis. Diabetes Care, 33(11), 2477– 2483.
Malik, V., Schul ze, M. & Hu, F. (2006) Intake of sugar- sweetened beverages and weight gain: a systematic review.
The American Journal of Clinical Nutrition, 84(2), 274– 288.
Nakamura, R., Mirelman, A.J., Cuadrado, C., Silva- Illane s, N., Dunstan, J. & Suhrcke, M. (2018) Evaluating
the 2014 sugar- sweetened beverage t ax in Chile: an obser vational study in urban areas. PL oS Me d, 15 (7),
e1002596.
Ng, S., Mhurchu, C., Jebb, S. & Popki n, B. (2012) Patterns and trends of beverage consumption among ch ildren and
adults in great britain, 1986– 2009. The British Journal of Nutrition, 108(3), 536– 551.
Pell, D., Mytton, O., Penney, T.L., Briggs, A., Cummins, S., Penn- Jones, C. et al (2021) Changes i n soft dr inks pur-
chased by British households as sociated with the UK soft dri nks industr y levy: controlled interrupted ti me
series analysis. BMJ, 372. http s://doi.org/10.1136/bmj.n254
Revoredo- G iha, C., Wrieden, W., Kupiec- Teahan, B., Leat, P., Fearne, A. & Cacciolatti, L. (2009) Analysis of red and
processed meat pu rchases in Scotland using representative supermarket panel data. Aberdeen: Food Standards
Agency i n Scotland.
Rojas, C. & Wang, E.Y. (2017) Do taxe s for soda and sugar y dri nks work? Scanner data evidence from Berkeley and
Washington. Scanner Data Evidence from B erkeley and Washing ton (September 23, 2017)
Seiler, S., Tuchman, A. & Yao, S. (2021) The impact of soda taxes: Pass- through, tax avoida nce, and nutritional
effects. Journal of Ma rketing Research, 58(1) 22– 49. https://doi.org/10.2139/ssrn.3302335
Silver, L., Ng, S., Ryan- Ibarra, S., Taillie, L., Indun i, M., Miles, D. et al (2017) Changes in pric es, sales, consumer
spending, and beverage consumption one year after a ta x on sugar- sweetened beverages in b erkeley, california,
US: A before- and- aft er study. PLoS Med, 14(4), 1– 19.
Taylor, R.L., Kaplan, S., Villas- Boas, S.B. & Jung, K. (2019) Soda war s: The effect of a soda tax elect ion on university
beverage sale s. Economic Inquiry, 57(3), 14 80– 14 96 .
Vartanian, L. R., Schwartz, M. B. & Brownell, K. D.(2007) Effects of soft dri nk con sumption on nutrition and
health: A systematic review and meta-analysis. American Journal of Pu blic Health, 97(4), 667– 675. htt ps://doi.
org/10.2105/AJPH.2 005.083782
Webber, R. (2004) Papers: Designi ng geodemographic classifications to meet contemporary business needs.
Interactive Marketing, 5(3), 219– 237.
Yamoah, F., Fearne, A. & Duf fy, R. (2014) Explori ng supermarket loyalty card analysis to identify who buys fair-
trade. The International Review of Retail, Distribution and Con sumer Research, 24(3), 328– 346.
Zhong, Y., Auchincloss, A., Lee, B. & Kanter, G. (2018) The short- ter m impacts of the philadelphia beverage tax on
bever age consu mption. Am erican Jo urnal of P reventive Med icine, 55 . ht tp s://d oi. org /10.1016 /j. ame pr e.2 018.0 2.017.
SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the
publisher’s website.
How to cite this article: Fearne, A., Borzino, N., De La Iglesia, B., Moffatt, P. &
Robbins, M. (2021) Using supermarket loyalty card data to measure the differential
impact of the UK soft drink sugar tax on buyer behaviour. Journal of Agricultural
Economics, 00, 1– 17. ht t p s ://doi. org/10.1111/ 1477- 9 552.12462