Article Text

Download PDFPDF

Achieving the tobacco endgame: evidence on the hardening hypothesis from repeated cross-sectional studies in New Zealand 2008–2014
  1. Richard Edwards1,
  2. Danny Tu2,
  3. Rhiannon Newcombe2,
  4. Kate Holland3,
  5. Darren Walton4
  1. 1Department of Public Health, University of Otago, Wellington, New Zealand
  2. 2Research and Evaluation Unit, Health Promotion Agency, Wellington, New Zealand
  3. 3Health Promotion Agency, Wellington, New Zealand
  4. 4Psychology Department, University of Canterbury, Christchurch, New Zealand
  1. Correspondence to Dr Richard Edwards, Department of Public Health, University of Otago, 23A Mein Street, Newtown, Wellington 6021, New Zealand; Richard.Edwards{at}


Introduction The hardening hypothesis proposes that as smoking prevalence declines the proportion of ‘hardcore’ or ‘hardened’ smokers will increase. The possible constructs of hardening include reduced motivation to quit, increased levels of addiction, increased levels of disadvantage and reduced quit rates among continuing smokers. Most previous studies have investigated only a single facet of the hypothesis. We used data from a national population monitor to test the hypothesis using measures across all four hardening constructs.

Methods We analysed data from a biennial population-based survey of New Zealand adults (aged 15 years+) from 2008 to 2014. Data were collected through face-to-face computer-assisted personal interviews.

Results During a period of reducing smoking prevalence, there were no statistically significant changes in indicators of hardening including the proportion of smokers who were unmotivated to quit, unable to quit despite repeated attempts or receiving state benefits or on a low income. Quit rates did not change significantly over the study period. For 2014 versus 2008, the OR for recent (within last 1–12 months) quitting was 1.14 (95% CI 0.53 to 2.46) and for sustained (within previous 13–24 months) quitting was 1.88 (95% CI 0.78 to 4.54).

Conclusions This study provides evidence that robust tobacco control strategies that result in substantial declines in smoking prevalence are not accompanied by the hypothesised increase in ‘hardcore’ or ‘hardened’ smokers who are more addicted and less motivated and able to quit. The findings suggest that there is no need for substantial change in approach to achieve New Zealand's Smokefree 2025 goal on the grounds that the smoker population is becoming increasingly hardened.

  • Surveillance and monitoring
  • Addiction
  • End game
View Full Text

Statistics from


The hardening hypothesis emerged in the 1980s from the notion of the ‘hardcore’ smoker who is more addicted and less able or less motivated to quit.1 The hypothesis proposes that as smoking prevalence declines the proportion of ‘hardcore’ or ‘hardened’ smokers will increase.2 The logic of the hardening hypothesis is that less addicted and more motivated smokers are more likely to quit, which would leave a greater proportion of hardened smokers as prevalence falls. If true, the hypothesis has great potential relevance for tobacco control and smoking cessation practice. For example, as prevalence declines, tobacco control policies, health promotion campaigns and smoking cessation support services would need to be adapted so that they effectively target and support the growing proportion of hardened smokers.

However, it is uncertain whether the hypothesis is true. Two key reasons for this are: (1) that the population of smokers is a dynamic cohort and its composition will be influenced not only by the characterstics of the smokers who quit or die, but also by the characteristics of new smokers who join the cohort3 and (2) that continuing smokers will be affected by social, environmental and policy contexts, and hence may change their behaviour to become more or less ‘hardened’. Hence the hypothesis needs to be investigated and tested.

Many definitions of hardening and hardcore smokers have emerged. Hughes distinguished between causes and outcomes of hardening.3 Broadly speaking the causes can be divided conceptually into the first three constructs described in table 1. First, hardening could occur if the population of smokers become less motivated to quit, fewer try to quit, or those who do try to quit do so less often. This could occur because smokers enjoy smoking more, are less aware of the adverse health effects of smoking and benefits of quitting, or have less regrets about smoking. A possible indirect measure of motivation to quit is the degree to which smokers support tobacco control measures and goals, with a declining level of support representing a form of ‘attitudinal hardening’. Second, there could be an increasing proportion of smokers who are heavily addicted and hence find it more difficult to quit. Third, hardening could occur if smokers increasingly come from disadvantaged groups or suffer high levels of psychological distress and other comorbid mental health problems. This could reduce their ability, and possibly motivation, to quit. Hardening could also occur due to a combination of these causes. The fourth construct is an outcome. The net result of hardening due to one or more of the three causes of hardening would be reduced successful quit rates over time among the population of smokers.

Table 1

Possible causes and outcome of hardening among the population of smokers

Numerous cross-sectional studies have investigated the prevalence of hardened or hardcore smokers in a population at a point in time. However, to rigorously test the hypothesis more analytical studies are required in which changes in indicators of hardening are examined over time, for example, by using repeat cross-sectional or cohort studies.

One approach is to study subgroups of smokers. For example, an analysis from New Zealand investigated changes in levels of addiction among Quitline callers.4 The disadvantage of this approach is that any observed changes in indicators of hardening may be affected by selection bias if, for example, the degree of hardening in the subgroup is unrepresentative of all smokers or if the degree of hardening in the subgroup changes over time in a way that is not reflected among the whole population of smokers.

A second and methodogically stronger approach is to examine changes in indicators of hardening in population-based cohorts or repeat cross-sectional studies which include a representative sample of smokers. Such studies have mostly have used measures within a single causal construct from table 2. For example, Lund et al5 assessed the change in the proportion of hardcore smokers in Norway defined by their intention not to quit smoking and absence of attempts to quit during the past 12 months. Several studies have measured changes in the level of addiction among US smokers over time as measured by the mean numbers of cigarettes smoked,6–9 mean serum cotinine,6 ,7 proportion of heavy smokers9 ,10 and time to first cigarette.8 Two Australian studies have investigated the association between measures of socioeconomic disadvantage and psychological stress and smoking over time.11 ,12 In all of these studies, there was little evidence to support the hardening hypothesis.

Table 2

Daily smoking prevalence, 2008–2014 (age adjusted)

Some repeated cross-sectional studies have assessed changes in the proportion of hardened smokers over time using a definition based on measures across more than one causal construct. For example, in an Australian study Clare et al13 defined hardcore smokers as those with no quit attempt in the past 12 months, no plan to quit and smoking more than 15 cigarettes per day. A few studies have assessed hardening using measures drawn from each of the first three constructs. For example, Celebucki and Brawarsky14 assessed changes over time in motivation to quit and quit attempts, several measures of addiction and multiple indicators of socioeconomic status and mental health among smokers in Massachusetts. In both these studies there was little evidence of hardening.

Studies investigating changes in population quit rates are less common. One study explored changes in quit rates among participants of pharmacotherapy smoking cessation trials, and found reducing post-intervention quit rates over time in the treatment and control groups.15 This is one of the few studies to find evidence consistent with hardening. However, because the participants are a selected subgroup of smokers, other explanations are possible for the decline in quit rates observed. Selection bias is possible if the smoking-related characteristics of participants invited and agreeing to take part in trials changed over time. Quit rates could also be affected by changes to adjunct interventions. Other studies have assessed cessation rates at the population level among US smokers and found either no change or an increase in smoker quit rates over time.8 ,16

Finally, ecological studies investigate evidence of hardening at the population level. A recent study assessed whether, as predicted by the hardening hypothesis, declining prevalence was associated with increasing the mean number of cigarettes smoked per day and declining proportion making quit attempts in the past year among smokers in US States (1992–2011) and 31 European countries (2006–2012).17 The authors found mostly the opposite, and argued this was consistent with ‘softening’ of the smoking population occurring as smoking prevalence decreases.

New Zealand is a particularly important context in which to evaluate the hardening hypothesis. It has a relatively low smoking prevalence and was the first country to adopt a Government-endorsed ‘endgame’ goal—Smokefree 2025. This is commonly interpreted as achieving a prevalence of <5% by the year 2025.18 If hardening is occurring in the New Zealand population of smokers, it has obvious adverse implications for the likelihood of achieving the Smokefree goal, and for the tobacco control strategies that may need to be implemented. The aim of this study was to use data from a national population monitor (2008–2014) to investigate the hardening hypothesis using measures from all four of the possible constructs of hardening.


Survey method and sample

We used data from a biennial population-based survey of New Zealand adults—The Health and Lifestyle Survey (HLS), first in field in 2008.19 This survey uses a multistage, stratified, probability proportional to size sampling design with the aim of achieving a representative sample of adults aged 15 years or over, with oversampling of Māori and Pacific people. The surveys were approved by the New Zealand Ethics Committee.

The primary sampling units were a random sample of 350 meshblocks (the smallest geographical area used by Statistics New Zealand for census data) with oversampling of meshblocks with high Pacific (>20%) ethnicity populations. Dwellings within selected meshblocks were enumerated and a systematic sample drawn of around 10 dwellings per meshblock using a probability proportional to the size sampling method, with additional dwellings selected to achieve a boosted Māori/Pacific sample. Within each selected household an adult member was selected at random and asked to take part in the survey. Up to six calls were made at each sampled dwelling, at different times of the day and on different days of the week, before accepting that dwelling as a non-contact. Replacement dwellings and eligible adults were not recruited in the event of failure to make contact or a refusal to participate.

Data collection

Data were collected on four occasions from 2008 to 2014 through face-to-face interviews in participants' homes. Survey responses were recorded in a computer-assisted personal interview (CAPI).


We used HLS survey data to measure a range of hardening constructs. Motivational hardening was assessed as the proportion of daily smokers who reported making no quit attempts of 24 hours or longer in the previous year. Attitudinal hardening was assessed by measuring the proportion of daily smokers who agreed with two proposed tobacco control interventions (banning smoking in all public places where children are likely to go and reducing the number of places allowed to sell tobacco to make it less available) and who supported an ‘endgame’ type goal—cigarettes and tobacco should not be sold in New Zealand in 10 years time.

Level of addiction was not assessed directly in the HLS. We used as proxy measures the proportion of all current smokers who were daily smokers, and the proportion of daily smokers who had made four or more quit attempts of more than 24 hours in the past year.

We measured the degree of disadvantage using the proportion of daily smokers who were beneficiaries (receiving state benefits) and the proportion who were living in a low-income household (household income <$NZ20 000/year).

Finally, quit outcomes were assessed using two measures: (1) the recent quit rate, defined as the proportion of recent quitters within the past 1–12 months among the population of smokers eligible to quit in past year (=number ex-smokers who quit 1–12 months ago/(number ex-smokers who quit 0–12 months ago+number daily smokers)); (2) the recent sustained quit rate defined as smokers who quit within the past 13–24 months as a proportion of the population eligible to quit in past 2 years (=number ex-smokers who quit 13–24 months ago/(number ex-smokers who quit 0–24 months ago+number daily smokers)). In the former measure quitters within the past month were excluded from the denominator due to the very short time since quitting (very high risk of relapse). The second measure was designed as a measure of sustained recent quitting with abstinence for at least a year.

Statistical analysis

All statistical analyses were undertaken using STATA V.12.0 (StataCorp LP, College station, Texas, USA). We calculated population prevalence estimates, ORs and 95% CIs.

Three adjustments were made to the data. First, each respondent in each of the samples was given a selection weight, which adjusts for the survey sampling design. Second, data were adjusted for differing response rates in the meshblocks surveyed. Third, to ensure that the results are representative of New Zealand's resident population aged 15 years and over living in permanent, private dwellings, the survey data were benchmarked to the proportions for respective groups in the census of population for gender, age and prioritised ethnicity. Population growth factors have been applied in the 2008, 2010 and 2012 HLS based on the 2006 Census population counts. The 2014 HLS was benchmarked to 2013 Census population counts. A full description of the HLS methodology, questionnaire and further HLS publications is available.18

All analyses except quit rates (see above) were carried out for all current smokers (smoke at least once a month) and for daily smokers. We present data for daily smokers, but results were similar for current smokers. Prevalence estimates and occurrence of outcomes were adjusted for age group (15–24 years, 25–44 years, 45–64 years and 65+ years) using regression methods. Subgroup analyses were performed by ethnicity (Māori and non-Māori). We calculated ORs for occurrence of outcomes for each survey year in relation to the baseline year of 2008 adjusted by age group, gender and ethnicity (Māori and non-Māori). Māori and non-Māori subgroup ORs were adjusted for age and gender. We used logistic regression analysis to test for trends across the four surveys adjusted for age, gender and ethnicity with survey number entered as a continuous variable. The ORs represent the estimate of linear change in odds of the outcome between consecutive surveys. Complete case analysis was used for all these analyses, as the missing data was very low (0.04–0.31%).


Survey numbers and response rates

The sample size was 1608 (422 current smokers) in 2008 and 1740 (451 current smokers) in 2010; response rates were 63.7% and 56.7%, respectively. Following a change in survey provider in 2012 the sample size increased to 2672 (581 current smokers) and the response rate to 83.1%. In 2014, there were 2594 (580 current smokers) participants, with an 80% response rate. Full details are provided in online supplementary table S1 in the online supplementary material.

Smoking prevalence

Age-adjusted daily smoking prevalence decreased from 16.9% in 2008 to 14.4% in 2014 (table 2). Among Māori the decrease was more substantial over this period from 35.0% to 28.3% and for non-Māori from 14.5% to 12.3%. CIs (95%) for the falls in prevalence between 2008 and 2014 overlapped the null.

Low motivation to quit

The proportion of daily smokers who made no attempts to quit lasting more than 24 hours in the past 12 months increased slightly between 2008 and 2014. The adjusted OR for 2014 versus 2008 was 1.19 and for linear trend it was 1.04 with the 95% CIs including the null (table 3). There were no significant changes in motivation to quit in subgroups stratified by ethnicity. Findings were similar for current smokers (data not shown).

Table 3

Low motivation to quit among daily smokers, 2008–2014

Attitudes to tobacco control

Table 4 shows the levels of support for tobacco control measures and an endgame goal between 2008 and 2014 among daily smokers. Support for bans on smoking in outdoor areas where children are likely to go increased steadily - adjusted odds ratio for 2014 vs 2008 was 2.45 (95%CI 1.53 to 3.93) and for linear trend was 1.16 (95%CI 1.08 to 1.25). There was little change in the other two measures and confidence intervals for adjusted odds ratios for 2014 vs 2008 and for linear trend included the null. Results were similar for Māori and non-Māori—data presented in online supplementary table S2.

Table 4

Support for three tobacco control measures among daily smokers, 2008–2014

Levels of addiction

The proportion of all current smokers who were daily smokers and proportion of daily smokers who had made four or more quit attempts lasting 24 hours or more in the previous year did not change greatly over the four surveys (table 5). There was an initial trend from 2008 to 2012 for the proportion of smokers making multiple unsuccessful quit attempts to increase. This was not maintained in 2014. The patterns for the latter measure were similar for current smokers (data not shown).

Table 5

Measures of levels of addiction, 2008–2014

Levels of disadvantage

There was no evidence for a sustained increase in the level of disadvantage among daily smokers (table 6) or current smokers (data not shown). The increase in proportion of smokers who were beneficiaries seen in 2010 and 2012 most likely reflected an increase in unemployment levels following the global financial crisis.

Table 6

Disadvantage among daily smokers, 2008–2014

Quit rates

Recent, and sustained recent, quit rates are shown in table 7. Recent quit rates (in the past 12 months) changed little over time, if anything, increasing among Māori (although CIs overlapped the null). Recent sustained quit rates (in the past 13–24 months) were steady for 2008–2012 but slightly higher in 2014. Patterns were similar for current smokers (data not shown).

Table 7

Recent quit rates, 2008–2014


We found no evidence to support the hardening hypothesis in a multifaceted analysis using data from biennial New Zealand surveys from 2008 to 2014. This was during a time period when prevalence of smoking decreased (data from the NZ census and NZ health survey found similar size decreases in smoking prevalence over approximately the same time periods20), and hence if the hypothesis is true, evidence of ‘hardening’ of the smoker population should have been observed. Changes in either direction between 2008 and 2014 were mostly modest and the CIs included the null, suggesting that chance was a possible explanation for any changes observed. The findings align with results from most similar analyses carried out in other jurisdictions.

Key strengths of the study include the multifaceted nature of the investigation. We were able to investigate change using measures from all the main proposed causes of hardening, and the main outcome of hardening—recent quit rates. Data were sourced from a series of high-quality population surveys carried out using consistent and comparable methods. Weaknesses include some deficiencies in the measures used. For example, the survey did not include any direct measure of intensity of smoking or level of addiction such as the mean number of cigarettes smoked per day or time to first cigarettes. The proportion of smokers who were daily smokers is only an imprecise measure of level of addiction at the population level, as it is not sensitive to changes in intensity of smoking within the daily smoker population. The other measure used—four or more unsuccessful quit attempts in the past year—could be seen as an indicator of addiction by measuring the proportion of smokers who find it difficult to quit. However, conversely it could also be seen as a measure of motivation to quit, with an increase interpreted as growing levels of motivation to quit among smokers. A potential weakness was the increase in the survey's response rate following the change in fieldwork provider in 2012. However, the same survey method and protocol were used for measuring hardening, and there was no evidence of a consistent step change in findings between 2008/2010 and 2012/2014, and hence the change in response rate did not seem to have caused substantial selection bias. Another weakness was the small numbers of smokers, particularly for subgroup analyses. This resulted in many of the effect size estimates being imprecise with wide CIs.

There was little evidence for hardening in New Zealand during a period when there was intense tobacco control activity and smoking prevalence was falling. Tobacco control activity included the adoption of a world-leading national smoke-free goal, and implementation of several tobacco control polices (eg, removal of point of sale retail tobacco displays, annual above inflation tobacco tax increases (2010–2014), an expansion of smoke-free outdoor policies in many localities and the introduction of smoke-free prison. This setting therefore provided a good test of the hardening hypothesis.

However, the generalisability of the findings to other jurisdictions and time periods is uncertain as populations of smokers, societal norms, tobacco control environments, smoking prevalence and responses to tobacco control efforts may differ. For example, hardening may only emerge at very low levels of smoking prevalence. Therefore, further research to monitor smoker populations over time and investigate the hardening hypothesis should be carried out.

In conclusion, this study provides further reassuring evidence that robust tobacco control strategies that result in substantial declines in smoking prevalence are not accompanied by an increase in ‘hardcore’ or ‘hardened’ smokers who are more addicted and less motivated and able to quit. The findings suggest that there is no need for a substantial change in approach to achieve the Smokefree 2025 goal on the grounds that the smoker population is becoming increasingly hardened. Rather, evidence from modelling studies that the current trajectories in smoking prevalence are unlikely to result in Smoke-free 2025 being achieved,21 ,22 particularly for Māori, suggest further intensification of tobacco control efforts is required.

What this paper adds

  • The hardening hypothesis proposes that as smoking prevalence declines the proportion of ‘hardcore’ or ‘hardened’ smokers will increase.

  • There has been limited investigation of the hardening hypothesis across all its possible facets.

  • This study from New Zealand—a country with relatively robust tobacco control strategies and declining smoking prevalence to very low levels—found no evidence of an increase in any of the proposed facets of hardening among continuing smokers.

  • The findings do not support the hardening hypothesis.


The authors thank The Health Promotion Agency for supplying the data, producing the survey instruments, managing the data collection process and setting up, and maintaining the data sets. They also thank all the survey participants.


View Abstract


  • Contributors RE conceived the study. The original idea and agreement on the measures of hardening was developed further with all the coauthors. DT carried out the analysis and prepared the tables. RE wrote the first and subsequent drafts. All authors provided comments on the drafts and revisions.

  • Funding This project was not supported by external funding. It was undertaken as research activity under university and government agency employment.

  • Competing interests Although we do not consider it a competing interest, for the sake of full transparency we note that one of the authors (RE) has previously undertaken work for health sector agencies working in tobacco control New Zealand Ethics Committee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.