Smoking prevalence following the announcement of tobacco tax increases in England between 2007 and 2019: an interrupted time–series analysis

Abstract Aims This study aimed to evaluate the impact of announcement of tax increases on factory‐made (FM) and roll‐your own (RYO) cigarettes in England. Design, Setting and Participants Autoregressive integrated moving average with exogeneous input (ARIMAX) time–series modelling in England, UK. Data were aggregated monthly on 274 890 participants between 2007 and 2019 taking part in the Smoking Toolkit Study (STS). Measurements The association of sustained step level changes for tax rises for FM cigarettes and temporary pulse effects for tax rises for RYO cigarettes with smoking, quit attempt and quit success prevalence as well as per‐capita self‐reported cigarette consumption and cost per cigarette was assessed. Findings A 10% rise in tax on RYO cigarettes was associated with a temporary 21.1% decline [95% confidence interval (CI) = –30.4 to −10.7] in smoking prevalence, and 20.7% decline (95% CI = –32.4 to −7.0) in per‐capita self‐reported cigarette consumption; while a 3% rise of tax on RYO cigarettes was associated with a temporary 20.7% decline (95% CI = –33.3 to −5.8) in the amount paid per RYO cigarette. For tax increases on FM cigarettes, a 5% above inflation tax rise was associated with a step‐level increase of 33.1% (95% CI = 18.4–49.5) in quit success rates. However, some of the findings were sensitive to model specification and temporally specific. Conclusion The announcements of tax increases for cigarettes in England between 2010 and 2019 were inconsistently associated with temporary reductions in smoking prevalence, per‐capita self‐reported cigarette consumption and improved quit success. Paradoxically, reductions in the cost for roll‐your‐own cigarettes were also found. The results were not robust in all sensitivity analyses.


INTRODUCTION
England has a strong tobacco control climate, as evidenced by the introduction of the first national tobacco control strategy 'Smoking Kills' in 1998. This strategy resulted in a range of policies aimed at increasing the retail price of cigarettes. These included a rise in taxation [1,2] and the Minimum Excise Tax (MET) which ensures a minimum price for tobacco and minimum pack size (i.e. of 20 cigarettes) and therefore high purchase price [3,4]. Studies throughout several countries have shown that tax increases are cost-effective interventions for reducing smoking prevalence by promoting quitting and reducing uptake, with those in lower socio-economic groups being most responsive [5,6].
Data are available from the Smoking Toolkit Study (STS), a population survey of adults aged 16+ in England, to evaluate the impact of tax increases in England [7]. For the first three years after the STS was established in November 2006 taxes rose in line with inflation, as measured by the Retail Prices Index (RPI). In 2009, tobacco duties were increased to ensure that the overall level of taxation remained broadly unchanged following a temporary reduction in VAT, while in 2010 taxation was 1% above inflation. A duty escalator followed this in 2011 at 2% above inflation, with an additional 10% increase on rollyour own (RYO) tobacco. Duty was increased by 5% above inflation in the 2012 budget, and in the 2013 budget the tax rise reverted to 2% above inflation. In 2016, tax on RYO tobacco increased by 5% above inflation and in 2018 by 3% above inflation (see Table 1) [8].
Recently, an empirical study assessed the immediate and longerterm effects of tax increases on the prevalence of smoking in Australia [9], finding that large increases (25%) were effective in reducing smoking prevalence. The authors used an interrupted time-series design to account for confounding by other tobacco control policies and secular trends. Similarly, time-series analyses of US data found that an increase in the price of cigarettes to more than $4 per 20-cigarette pack was associated with a significant decrease in smoking among younger people [10]. Decreases were also seen in asthma hospitalizations, acute myocardial infarction hospitalization and sudden cardiac death rates. In other studies, significant increases in tobacco taxes have been shown to encourage current tobacco users to stop using and reduce consumption among those who continue to use, with the greatest impact on the young and disadvantaged groups [6].  report that ≥ 50% of the cigarettes they smoked are roll-your-own cigarettes [13]. The prevalence of predominant FM cigarette smoking in each month was calculated as the proportion of participants who report that < 50% of the cigarettes they smoked were RYO cigarettes.
Per-capita self-reported cigarette consumption Smokers were asked on average how many cigarettes they smoked per day. The per-capita consumption per day was then calculated as the summation of cigarettes smoked per day (which was set to zero for all non-current smokers, i.e. non-smokers and ex-smokers) divided by the entire population. Consumption within the population rather than among smokers was used to reflect quitting, i.e. a reduction may reflect an increase in recent ex-smokers without a change in consumption by continuing smokers.

Data on explanatory variables
We modelled the date of the announcement of the tax increases rather than the implementation. Tobacco industry tactics are implemented following the announcement: increasing prices on top of tax increases, so that both the price and tax increase are passed on to consumers (known as overshifting), absorbing the tax increase so it is not passed on to consumers (undershifting) or passing the tax increase on to consumers in full (fully shifting) [15]. Thus, modelling the annoucmenet means that the impact of these immediate tactics on smoking behaviour would be detected. Moreover, the announcement of the tax rise is likely to be the moment where consumers become aware of the cost of cigarettes going up and is therefore a useful benchmark.
For the overall primary analysis, we modelled two step level changes (sustained effects) for the tax increases on FM cigarettes: (1) the shift to above inflation tax increases on FM cigarettes from

ANALYSIS Amendments
In the original analysis plan we had included ever smoking (as an indicator of uptake) as an outcome variable. However, effect size estimates from the models including ever smoking appeared implausibly large. These data are provided on the Open Science Framework, together with the results of the pre-planned analysis for ever smoking (https://osf.io/kz3bc/).
During the same internal review, it was also decided to add new outcome variables in the primary and sensitivity analyses: (i) average population cigarette consumption (primary analysis), (ii) average population FM and RYO cigarette consumption per day (sensitivity analysis) and (iii) average cost per FM and RYO cigarette (primary analysis). These variables may be more sensitive to FM/RYO-specific tax increases and have been used widely in previous literature [6,14,16,17].
Following concerns that the size of the significant pulse and step level effects may have been sensitive to randomness in the monthly data, it was also decided to run additional unplanned sensitivity analyses using smoothed data at the quarterly rather than monthly level.

Primary analysis
All data were analysed in R studio [18]. The data frame and analysis plan were pre-registered on the Open Science Framework (https:// osf.io/kz3bc/). Alpha was set to 0.05; 95% confidence intervals (CIs) are reported.
We used an interrupted time-series analysis to account for autocorrelation among monthly observations by fitting autoregressive integrated moving average with exogeneous input (ARIMAX) models [19,20]. For the primary analysis, we modelled step level changes as a result of the rise in tax on FM cigarettes and pulse effects as a result of the rise in tax on RYO. We pre-specified modelling the tax increases in this way as: (1) tax increases for RYO were infrequent and therefore we hypothesized temporary rather than longer-term changes, and (2) tax increases for FM were frequent and consistent over time and therefore were better modelled as step changes. In the analysis these models were also the most stable and provided an adequate fit.
Bayes factors were calculated for non-significant findings for the primary analysis in R using code described by Dienes [21,22]. This helps to determine if there is evidence for the null hypothesis of no difference or the data are insensitive to detect an effect.

Availability of data and materials
The data frame was pre-registered on the Open Science Framework (https://osf.io/kz3bc/).  Figure 1 shows the prevalence of smoking, quit attempts, the success of quit attempts, per-capita self-report cigarette consumption and the cost per FM/RYO cigarette during the course of the study.

Primary analysis
In adjusted models, there was a significant association between a temporary pulse in the rise in tax by 10% on RYO cigarettes in March 2011 and smoking prevalence (see Table 2) and per-capita selfreported cigarette consumption per day (see Table 3). It was estimated that the rise in tax was associated with a 21. A pulse in the rise in tax by 3% above inflation on RYO in November 2018 was also associated with the average cost per RYO cigarette (see Table 3). Paradoxically, the tax rise was associated with a 20.7% decline in the cost per RYO cigarette. Fitted values from the model indicated that this equated to a decrease from £0.15 in the month before to £0.12 in November 2018.
Supporting information,  Modelling significant associations using quarterly data When re-running the significant associations identified in the primary analysis using quarterly rather than monthly data, no statistically significant associations were found between the pulses of the rise in tax on RYO during the first quarter of 2011 and the cost per RYO cigarettes and per-capita self-reported cigarette consumption. The association between the pulse in the first quarter of 2011 for the tax rise on RYO and smoking prevalence was of a similar magnitude to the primary analysis using monthly data (−18.9 versus −21.1), although not significant. Bayes factors confirmed that the data were insensitive (see Supporting information, Table S12).

Effect of FM-targeted tax increases (step level effects)
Primary analysis The step level change reflecting the rise in tax above inflation for FM cigarettes from March 2012 was associated with a 33.1% increase in the quit success rate (see Table 2). Fitted values from the model suggested that the rise in tax above inflation increased the success rate of quit attempts from an average of 13.9% before March 2012 to 16.6% after March 2012.
Supporting information, Table S3 shows the results for the unadjusted models and Supporting information, Table S4 shows the results for the models adjusted for all other tax increases. Bayes factors indicated that the data were largely insensitive to detect changes for the non-significant findings across the outcome measures of interest (see Supporting information, Table S5).  Table S11).

Modelling significant associations using quarterly data
When re-running the significant associations identified in the primary analysis using quarterly rather than monthly data, a step-level change from the first quarter of 2012 to the last quarter of 2019, reflecting the rise in tax on FM cigarettes, was positively associated with the success of quit attempts (see Supporting information, Table S12).

Summary
The announcement of large tax increases on RYO cigarettes in England appear to have been associated with a temporary reduction in smoking prevalence, in mean cigarette consumption per day and, perhaps surprisingly, in the amount spent per RYO cigarette. These associations were only present among those of lower SES after strati-

fication. The announcement of tax increases on FM cigarettes in
England also appeared to be associated with a sustained increase in the prevalence of successful quit attempts. This association was only present among those of higher SES after stratification. However, some of the findings were sensitive to model specification and temporally specific. Bayes factors also indicated that our data were insensitive to detect other associations.

Limitations
This study had several limitations. First, it was unexpected to find that the largest increase in tax to 5% above inflation from March 2012 to February 2013 was associated with a step level increase in quit success rates. Although we adjusted for mass media spend and tobacco control policies, other population level factors may account for this association, including the rise in use of e-cigarettes in 2012 [23]. Secondly, the effect sizes for the pulse effects and confidence intervals are relatively large. The plausibility of these effects occurring during a 1-month period as a consequence of announcing the tax increases needs to be considered. However, the analysis allowed for pulse effects to decay gradually or to decay abruptly and quarterly data were used in a sensitivity analysis. Thirdly, the STS requires participants to recall quit attempts in the past 12 months which could have introduced bias, although we have no reason to believe that reporting would differ over time. The STS also requires participants to recall their cigarette consumption and amount spent on cigarettes. STS estimates of cigarettes per day has been shown to align with sales data [12]. Fourthly, the findings might not generalize to other countries. Finally, the impact of several population-level polices were only modelled as pulse effects. It is possible that findings might be different if a more comprehensive evaluation had been undertaken; for example, a consideration of step-level changes or permanent changes. This was not possible with the number of different factors included in the models in the current study, but warrants further investigation. This study could also not adjust for several important policies, including the introduction and rise in MET, which creates a minimum price and discourages the selling of cheaper FM products. This occurred during the same month as the tax increases being investigated (November 2017), and therefore it was possible to disentangle effects. The MET has been shown to be effective in reducing tobacco industry revenues [3].

Comparison with previous studies
In line with previous studies, the finding of a decrease in smoking prevalence following the accouncement of tax increases on RYO cigarettes suggests that this may be an effective intervention for reducing tobacco use [5,6,9,10]. The fact that this effect was only present among those of lower SES indicates that this may be one strategy to reduce widening of social inequalities in health. Previous studies also report that smokers in lower SES groups are the most responsive to tobacco price increases [5], with RYO cigarettes being more common among those from disadvantaged backgrounds [24].
The finding of an increase in quit success as a function of increasing tax on FM cigarettes is also in line with previous studies [25,26], as is the increased strength of this association among those of higher SES. Although Wilkinson et al. [9] found that those of lower SES had a larger immediate reduction in smoking prevalence in response to the 2010 tax than the higher SES group, it was not sustained. A possible explanation is a greater incidence of relapse among the lower SES group. Smokers in lower SES groups are generally more vulnerable to relapse due to their higher nicotine dependence and decreased quitting self-efficacy [27]. More disadvantaged smokers also have a lower level of FM use and are therefore less affected by FM tax increases [24].
This study also found a reduction in the amount paid per RYO as a function of tax increases on RYO cigarettes-a finding which was also concentrated among those of lower SES. used cost-minimizing strategy by smokers [28].
The increased tax on RYO cigarettes was also associated with a reduction in per-capita self-reported cigarette consumption, predominantly among those of lower SES, which may be indicative of increased quitting activity and/or reduced uptake as a response to anticipation of the increased price of cigarettes. The latter would be consistent with the finding of a reduced smoking prevalence but not increased quit success rate among those of lower SES in response to tax rises. A surprising finding in the sensitivity analysis was that the 5% rise in tax on RYO in March 2016 was associated with an increase in per-capita self-reported cigarette consumption of predominant RYO cigarettes. This might reflect smokers pre-empting the tax rise by reducing the amount of tobacco per cigarette, resulting in compensatory smoking. Although this is speculative, previous studies have reported a reduction in the weight of tobacco used in RYO cigarettes in response to price rises [15]. A decrease in the cost of RYO by the tobacco industry could also account for this finding. At the point when tobacco taxes are increased in March/April each year, the industry overshifts the tax increase on the more expensive brands, while absorbing the tax increase on lower price brands [29].
Bayes factors suggested that the failure to find an impact of smaller tax increases reflected data insensitivity, and the nonsignificant results do not necessarily indicate no associations between these increases and key smoking outcomes. Some of the findings could also not be easily be explained and were unexpected. This introduces uncertainty about the findings from the models. For example, the above inflation tax on FM cigarettes initiated in March 2010 was associated with a step level 12.7% decline in smoking prevalence of predominant RYO cigarettes.

Implications
This study found evidence for a positive impact of reasonably large increases in taxation, targeting FM and RYO cigarettes. Although we are cautious to infer cause and effect from this study, it lends support to the argument that the government should consider sizeable and sustained tax increases in future budgets, perhaps in line with the increases seen in other countries such as Australia (25%) [9]. We used the date of the announcement of the budget as the primary implementation date, and therefore any temporary pulse effects may reflect the announcement forewarning smokers about tax changes, while step level changes may reflect more of the impact of tax increases themselves. Anticipated future costs are one of the main determinants of current smoking [30]. Mass media campaigns are effective in reducing prevalence of smoking and thus it may be important to consider publicity around budget changes in the future [31].
This study also highlights the possible role that tobacco industry tactics may play in undermining tobacco increases (as evidenced by the associations with the cost of cigarettes) and strategies used by smokers to mitigate increased cost (as evidenced by possible switching between FM and RYO and from premium to non-premium brands, and the reduction in consumption). However, tax rises nonetheless still had an impact on actual behaviour in the current study, in the form of increasing the rate of successful attempts to quit smoking.

CONCLUSION
The announcement of the largest tax increases on cigarettes in England (5% on FM and 10% on RYO) appear to be associated with a temporary reduction in smoking prevalence, per-capita self-reported cigarette consumption but also with the cost paid per RYO cigarettes.
They are also associated with a sustained increase in the success rate of quit attempts.