The Differential Impact of the 2000 Canadian Graphic Warning Label Policy on Smoking Prevalence by Sex and Education: A Difference-in-Difference-in-Difference Model

Abstract

Introduction

Using a quasiexperimental design, we compared the impact of the 2000 Canadian introduction of graphic warning labels (GWLs) on differences in smoking prevalence by sex and education, to the United States, where no GWLs were introduced.

Methods

We pooled 1999–2004 data from the Canadian Tobacco Use Monitoring Survey and the U.S. Behavioral Risk Factor Surveillance System. We used a difference-in-difference (DD) model to assess the impact of Canadian policy introduction on smoking prevalence, and a difference-in-difference-in-difference (DDD) model to examine differences in the policy impact by sex and education, comparing Canada (the treatment group) with the United States (the control group).

Results

From 1999 to 2004, smoking prevalence decreased from 23.7% to 18.6% in Canada, and from 21.7% to 20.0% in the United States. Results from the DD regression models showed that Canadian respondents reported lower odds of being a current smoker compared to the U.S. respondents following the 2000 introduction of GWLs (OR = 0.84, 95% CI = 0.74–0.94). The DDD model showed that the impact of the Canadian GWLs versus the United States did not differ by sex or education.

Conclusions

The 2000 Canadian GWL policy reduced smoking prevalence overall, with similar reductions for males and females and across education levels. The impact of the Canadian GWLs in reducing smoking prevalence did not reduce differences by sex or education. Although beneficial for all smokers, GWLs may not serve to decrease existing disparities, especially those by socioeconomic status.

Implications

Existing evidence shows that GWL implementation is associated with reductions in smoking prevalence. But there is limited evidence from past evaluation studies on whether the impact of GWLs on smoking prevalence differs by sociodemographic subgroup. Our findings confirm existing studies that the 2000 implementation of GWLs in Canada was significantly associated with an overall reduction in smoking prevalence in Canada compared to the United States. However, our study improves existing evidence by showing that the impact of the Canadian GWLs on smoking prevalence did not differ by sex or education, and thus did not reduce existing smoking disparities by educational levels.

Introduction

Cigarette smoking is a leading preventable cause of death in the world, accounting for over 8 million deaths annually.¹ In 2019, 1.14 billion people smoked cigarettes worldwide.² As restrictions on tobacco advertising and marketing have been implemented by many countries, as called for by the World Health Organization Framework Convention on Tobacco Control (FCTC), the cigarette pack has become an even more important vehicle for tobacco companies to advertise and promote their product.³ To counteract the increased use of the pack as a marketing vehicle, policies mandating prominent warning labels have become increasingly important and widespread, also consistent with the FCTC. In 2000, Canada became the first country to include colored graphic warning labels (GWLs) on cigarette packages, and as of October 2021, 134 countries, covering 70% of the world’s population, now require GWLs.^3,4 GWLs have a wide reach and potential for frequent exposure, as a person who smokes a pack-a-day could be exposed to the warnings more than 7000 times per year, each exposure being a potential reminder to smokers of the harms of cigarettes.⁵ GWLs have been found to lead to several positive outcomes, including improved knowledge of tobacco-related risks, increased intentions to quit, and smoking cessation.^6–8

Implementation of GWLs is associated with reduced smoking prevalence.^9–13 Comparative studies have shown a reduction in smoking prevalence in countries with GWL implementation relative to neighboring countries with none.^9,12 A trend analysis of Uruguay and Argentina showed that in Uruguay, the 2005 implementation of GWLs was associated with a decrease in smoking prevalence from 34.5% in 2001 to 28.7% in 2009, while Argentina, which did not implement GWLs, experienced a significantly lower decrease from 33.3% to 31.8% during the same period.¹² However, because Uruguay initiated a series of other comprehensive anti-smoking measures during the study period, the decrease in smoking prevalence may not be attributed solely to the introduction of GWLs.¹²

Other observational studies without comparison groups have also found a decrease in smoking prevalence following GWL implementation,^14,15 while other studies found marginally statistically significant¹⁰ and nonstatistically significant^11,13 associations between GWL policy implementation and reduction in smoking prevalence.^10,11 One of the studies that did not find a significant relationship between GWLs and smoking prevalence measured smoking prevalence as the number of daily smokers in the population, which did not capture the wider population of established smokers including nondaily smokers.¹¹ Another study used a short-time lagged post-policy period of 6 months following implementation, which could not account for those who may have quit smoking after 6 months.¹³

There is limited information on GWL policy effects by sociodemographic characteristics such as sex and socioeconomic status (SES).^14,16–19 A cross-sectional study of the Canadian GWLs using two waves of survey data, pre-policy (July–December 2000) and post-policy data (February–June 2001), found that higher SES individuals were less likely to smoke post-policy than lower SES individuals, but there was no difference in smoking prevalence by sex following the introduction of GWLs.¹³ However, this study did not test for interaction effects with the sociodemographic variables.¹³ Multicountry observational studies have shown that GWLs are more strongly associated with forgoing a cigarette and decreased smoking prevalence among less-educated individuals compared to higher-educated ones.^10,14

Experimental studies provide further evidence on the associations between smoking-related outcomes and sociodemographic characteristics.^16–18,20 For example, women perceive GWLs to be more effective at motivating cessation and preventing smoking initiation, compared to men.¹⁷ GWLs evoke more negative emotions, such as fear or disgust, among women on seeing GWLs on cigarette packs, and women rate their responses to GWLs more strongly than men do.^16–18 GWLs are also effective across diverse geographic and cultural contexts.²⁰ They effectively communicate health risks to low SES smokers¹⁴ and are viewed as more effective than text-only warnings by low SES smokers.²¹ Despite the existing evidence on the associations between sociodemographic characteristics and GWLs through both experimental and observational studies, more systematic research using pre-/post-policy designs with control groups is needed to evaluate whether GWL effects on smoking prevalence vary by sex and SES.

In June 2000, Canada passed regulations requiring 1 of 16 colored GWLs, in both English and French, to cover at least 50% of the main display area on both sides of the cigarette package.²² Each package also included messages inside of packs about cessation benefits and tips.²² All cigarette packages sold at retail stores were required to include these GWLs by December 2000.²² The 2000 GWL policy was effective in communicating health risks to smokers, encouraging smoking cessation, and reducing smoking prevalence.^5,9,15 A quasiexperimental study on the impact of the GWLs in Canada versus the United States, from 1991 to 2009, showed that the 2000 GWL policy relatively reduced smoking prevalence in Canada by 12%–20% compared to the United States, between 2001 and 2009.⁹ However, the differential effectiveness of the GWL policy across sociodemographic groups was not assessed. The goal of our study is to fill this important research gap. Using a quasiexperimental design, this study aimed to assess the relative impact of the 2000 Canadian GWL policy on smoking prevalence in Canada, compared to the United States, where there was no change in warnings during the study period, and to examine differences in the impact of GWLs by sex and education.

Methods

Study Population

We analyzed data from the Canadian Tobacco Use Monitoring Survey (CTUMS) and the U.S. Behavioral Risk Factor Surveillance System (BRFSS) from 1999 to 2004. We selected available data for 2 years pre-implementation of the GWLs in Canada and 4 years post-implementation to allow for a long enough period to evaluate time trends and to control for potential confounders such as cigarette prices. We restricted the sample to adults aged 25 years and older to allow for completion of education for most respondents.

CTUMS is a nationally representative cross-sectional survey, conducted annually from 1999 to 2013. The CTUMS was designed to produce regular estimates of tobacco use and monitor changes over time. The participants included noninstitutionalized Canadian residents 15 years and older, excluding residents in the Canadian territories. CTUMS employed a two-phase stratified random sampling design of households, selected from random digit dialing (RDD) methods, from which individuals were selected based on household composition. Data were collected using computer-assisted telephone interviews conducted monthly from February to December each survey year. A full description of the CTUMS methodology is available elsewhere.²³

BFRSS is an ongoing nationally representative cross-sectional study of U.S. adults, 18 years and older, conducted from 1984. The BFRSS is administered annually by the state health departments in all 50 states. It was designed to monitor health-related risk factors, including tobacco use, over time. Similar to CTUMS, BRFSS also employs a two-stage stratified random sampling of individuals selected from households, which were selected from RDD sampling. Data were collected monthly for each survey year. Additional details on the BFRSS methodology are available elsewhere.²⁴

Measures

Outcome Variable

Current smoking status was measured as a binary variable derived from measures of current (every day or some days) and established smoking (smoking 100 and more cigarettes in their lifetime). People who met the 100-lifetime cigarette threshold and reported currently smoking every day or some days were coded as 1, and people who did not report 100-lifetime cigarettes or who were not currently smoking every day or some days were coded as 0.

Policy Exposure Variable

Policy exposure was measured as the interaction between the country and a pre–post-implementation variable. The country variable was given a value of 1 for individuals living in the treatment country, Canada (1), and 0 for those living in the control country, United States (0). We created a pre–post-implementation variable to indicate the period before (before June 2001 = 0) and after (June 2001 and after = 1) the full GWL policy intervention. We utilized June 2001 as the cutoff period as all tobacco packages were expected to have health warnings by then.²⁵ Based on these definitions, the policy exposure variable was a value of 0 for all individuals living in both countries, Canada (1 * 0) and United States (0 * 0) before the GWL policy implementation, and a value of 1 for individuals living in Canada (1 * 1) after the policy implementation, while the United States remained 0 (0 * 1).

Third-Difference Variables

Sex (male/female) and education (high school degree or less, some college degree [tech/trade school/community college, or some university], and 4-year college degree or more [completed university or postgraduate]).

Covariates

In addition to sex and education, we included age group (25–39 years, 40–54 years, and 55+ years old) as a sociodemographic variable. We included a categorical Year variable representing all the sampled survey years to estimate prevalence, and a Time trend variable starting with 1 from January 1999 and increased by 1 each month throughout the study period, to estimate the pre- and post-prevalence differences when fitting the model. We included an annual cigarette price index variable for both countries: the national-level price index for Canada and the state-level price index for the United States.^26,27 Canadian dollars were converted to U.S. dollars in each year, converted to price per cigarette, and adjusted for inflation (indexed to 1 in November 2002).²⁸ Cigarette prices are a key tobacco control policy that influences smoking rates, hence the need to control for them.²⁹ We used the natural log transformation for the time trend and cigarette price index variables.

Statistical Sample and Analysis

We pooled data from the CTUMS and BFRSS. To describe the sample characteristics, we calculated the unweighted distribution by age group, sex, education, and smoking prevalence for each country and survey year. Overall weighted estimates of smoking prevalence for each country were then calculated, as well as by sex and education. Prevalence estimates were adjusted for sex, education, age group, and cigarette prices, computed for every sampled year, as well as pre- and post-June 2001 (GWL policy).

We utilized a quasiexperimental method to assess the causal impact of the Canadian policy intervention on smoking prevalence, comparing Canada (the treatment group) with the United States (the control group). Specifically, we used difference-in-difference (DD) regression models to estimate the pre- and post-policy differences in smoking prevalence in Canada relative to the United States controlling for potential confounders. A major benefit of comparing the United States and Canada is that both countries are similar geographically and culturally; hence, many uncontrolled factors tend to be similar. The DD model was specified as:

Model 1^a

$${\displaystyle \begin{array}{l}{\displaystyle \begin{array}{l}Logit(Smokingstatus=1)={\beta }_0+{\beta }_{1}Country+{\beta }_{2}PostPolicy\\ +{\beta }_{3}Country\ast PostPolicy+\\ {\beta }_{4}Agegroups+{\beta }_{5}Sex+{\beta }_{6}Education\\ +{\beta }_{7}ln(PriceIndex)+{\beta }_{8}ln(PriceIndex)\ast Country+\\ {\beta }_{9}ln(Timetrend)+{\beta }_{10}ln(Timetrend)\ast Country\end{array}}\end{array}}$$

^aThis DD model assumed that the time trend, the impact of cigarette price, and smoking prevalence are linear but different between countries.

Furthermore, we analyzed the differences in smoking prevalence stratified by sex and education using the difference-in-difference-in-difference (DDD) model. The DDD models are used to estimate the potential difference in the impact of policy intervention by sociodemographic subgroups.⁹ The DDD model used in this study was specified separately for sex and education as:

Model 2^b

$${\displaystyle \begin{array}{l}{\displaystyle \begin{array}{l}Logit(Smokingstatus=1)={\beta }_0+{\beta }_{1}Country\\ +{\beta }_{2}PostPolicy+{\beta }_{3}Country\ast PostPolicy+\\ {\beta }_{4}Agegroups+{\beta }_{5}Sex\\ +{\beta }_{6}Country\ast Sex+{\beta }_{7}PostPolicy\ast Sex\\ +{\beta }_{8}Country\ast PostPolicy\ast Sex+\\ {\beta }_{9}Education+{\beta }_{10}ln(PriceIndex)\\ +{\beta }_{11}ln(PriceIndex)\ast Country\\ +{\beta }_{12}ln(Timetrend)+\\ {\beta }_{13}ln(Timetrend)\ast Country\end{array}}\end{array}}$$

^bIn Model 2, all the variables are similarly defined as Model 1, with the addition of interaction terms for sex and country, sex and time trend, and country and time trend.

Model 3^c

$${\displaystyle \begin{array}{l}{\displaystyle \begin{array}{l}Logit(Smokingstatus=1)={\beta }_0+{\beta }_{1}Country+{\beta }_{2}PostPolicy\\ +{\beta }_{3}Country\ast PostPolicy+\\ {\beta }_{4}Agegroups+{\beta }_{5}Education+{\beta }_{6}Country\ast Education\\ +{\beta }_{7}PostPolicy\ast Education+\\ {\beta }_{8}Country\ast PostPolicy\ast Education+{\beta }_{9}Sex+{\beta }_{10}ln(PriceIndex)\\ +{\beta }_{11}ln(PriceIndex)\ast Country+\\ {\beta }_{12}ln(Timetrend)+{\beta }_{13}ln(Timetrend)\ast Country\end{array}}\end{array}}$$

^cIn Model 3, all the variables are similarly defined as Model 1, with the addition of interaction terms for education and country, education and time trend, and country and time trend.

All models and estimates were produced using logistic regression models. Each cross-sectional weight was rescaled to sample sizes for each country and for each year, to ensure estimates were comparable between waves and countries. Logistic regression results are presented as adjusted odds ratios (aORs), with all confidence intervals (CIs) and statistical significance tested at the 95% confidence level. All the analyses were conducted using SAS-Callable SUDAAN (v. 11).³⁰

Sensitivity Analyses

As described above, in the analysis for estimating smoking prevalence, we coded Year as a categorical variable in the models. In sensitivity analyses for the DD and DDD analyses, we removed the categorical Year variable and included the PostPolicy and Timetrend variables and their interaction terms for a pre- and post-label policy comparison. These two specifications allow comparisons of sensitivity on smoking prevalence estimates over time.

Results

Descriptive Statistics

Table 1 presents the unweighted descriptive characteristics of the sample population. BFRSS sample sizes were more than 10 times that of CTUMS. The final annual sample size ranged from n = 10 354 to 12 098 for Canadian participants and from n = 141 735 to 276 166 for U.S. participants from 1999 to 2004. Respondents aged 55 years and older represented at least a third of the sample population in all survey years in Canada (32.7%–38.1%) and the United States (34.5%–42.3%). There were more female respondents in all the survey years in Canada (55.8%–56.6%) and the United States (59.4%–61.4%). Between 40% and 50% of the population in Canada (39.4%–48.3%) and more than half in the United States (55.0%–59.2%) had more than a high school education.

Table 1.

Sociodemographic Characteristics of Sample Participants; Behavioral Risk Factor Surveillance System—United States and Canadian Tobacco Use Monitoring Survey—Canada (1999–2004)

		1999		2000		2001		2002		2003		2004
		n	%	n	%	n	%	n	%	n	%	n	%
Canada
Total		10 881		10 354		11 193		12 098		11 111		10 719
Sex	Male	4813	44.2	4574	44.2	4944	44.2	5249	43.4	4889	44.0	4717	44.0
	Female	6068	55.8	5780	55.8	6249	55.8	6849	56.6	6222	56.0	6002	56.0
Age group	25–39	3742	34.4	3439	33.2	3676	32.8	3646	30.1	3292	29.6	3082	28.8
	40–54	3582	32.9	3396	32.8	3651	32.6	3937	32.5	3590	32.3	3738	34.9
	55+	3557	32.7	3519	34.0	3866	34.5	4515	37.3	4229	38.1	3899	36.4
Education	High school or less	5663	52.0	5527	53.4	5849	52.3	6216	51.4	5479	49.3	6180	57.7
	Some college degree	2766	25.4	2537	24.5	2880	25.7	3093	25.6	3042	27.4	1784	16.6
	4-year college degree or more	2075	19.1	1940	18.7	2100	18.8	2391	19.8	2205	19.8	2440	22.8
Tobacco use	Smoking	10 847	22.4	10 318	21.7	11 162	19.2	12 061	18.9	11 101	18.8	10 716	17.3
United States
Total		141 735	100	162 239	100	184 909	100	219 762	100	237 360	100	276 166	100
Sex	Male	57 588	40.6	65 451	40.3	74 984	40.6	87 709	39.9	93 462	39.4	106 719	38.6
	Female	84 147	59.4	96 788	59.7	109 925	59.4	132 053	60.1	143 898	60.6	169 447	61.4
Age group	25–39	46 355	32.7	52 206	32.2	57 222	30.9	61 641	28.0	62 446	26.3	69 335	25.1
	40–54	46 038	32.5	54 041	33.3	62 492	33.8	73 686	33.5	79 565	33.5	90 144	32.6
	55+	49 342	34.8	55 992	34.5	65 195	35.3	84 435	38.4	95 349	40.2	116 687	42.3
Education	High school or less	63 458	44.8	68 982	42.5	77 965	42.2	92 184	41.9	96 294	40.6	112 238	40.6
	Some college degree	37 081	26.2	43 489	26.8	48 865	26.4	57 259	26.1	63 021	26.6	72 030	26.1
	4-year college degree or more	40 848	28.8	49 394	30.4	57 611	31.2	69 914	31.8	77 568	32.7	91 305	33.1
Tobacco use	Smoking	29 018	20.5	33 487	20.7	40 168	21.8	48 979	22.4	52 032	22.0	56 570	20.6

Smoking Prevalence

In Canada, smoking prevalence decreased from 23.7% in 1999 to 18.6% in 2004, while the United States experienced a smaller decrease from 21.7% to 20.0% during the same period (Table 2; Figure 1). In Canada, smoking prevalence decreased from 25.7% in 1999 to 21.2% in 2004 for males and from 21.8% to 16.0% for females in the same period (Table 2). In the United States, smoking prevalence decreased to a lesser extent from 1999 to 2004 among males (23.9%–22.2%) and females (19.8%–18.0%) (Table 2). Smoking prevalence also decreased across all educational levels in Canada and United States (Table 2). In Canada, smoking prevalence decreased from 1999 to 2004: among participants with a high school degree or less, from 27.3% to 21.1%; among those with some college degree, from 23.6% to 20.2%; and among those with a 4-year college degree or more from 16.6% to 13.1% (Table 2). In the United States, smoking prevalence decreased during the same period: among participants with a high school degree or less (27.1%–26.6%), those with some college degree (23.1%–22.0%), and those with a 4-year college degree or more (12.2%–10.7%) (Table 2).

Table 2.

Changes in Smoking Prevalence Overall, by Sex and Education, in Canada and the United States, 1999–2004

		1999		2000		2001		2002		2003		2004
		n	%	n	%	n	%	n	20.2	n	%	n	%
Canada
Overall		10 847	23.7	10 318	23.1	11 162	20.4	12 061	19.9	11 101	19.9	10 716	18.6
Sex	Male	4795	25.7	4555	24.9	4928	22.7	5227	22.1	4881	22.7	4715	21.2
	Female	6052	21.8	5763	21.4	6234	18.2	6834	18.4	6220	17.1	6001	16.0
Education	High school or less	5643	27.3	5508	27.1	5836	23.9	6200	23.0	5473	24.0	6179	21.1
	Some college degree	2758	23.6	2527	24.8	2871	21.2	3082	21.2	3040	21.0	1784	20.2
	4-year college degree or more	2069	16.6	1934	13.3	2094	12.3	2384	12.7	2204	11.9	2440	13.1
United States
Overall		141 344	21.7	161 774	21.3	184 343	21.9	219 147	21.9	236 723	21.3	275 278	20.0
Sex	Male	57 417	23.9	65 256	23.2	74 751	24.0	87 460	24.3	93 198	23.9	106 402	22.2
	Female	83 927	19.8	96 518	19.7	109 592	20.1	131 687	19.7	143 525	18.9	168 876	18.0
Education	High school or less	63 266	27.1	68 771	27.1	77 694	28.0	91 904	28.4	96 005	27.9	111 841	26.6
	Some college degree	37 002	23.1	43 373	22.8	48 734	23.5	57 129	23.6	62 865	23.0	71 780	22.0
	4-year college degree or more	40 744	12.2	49 259	12.0	57 454	12.2	69 723	12.0	77 386	11.8	91 077	10.7

Figure 1.

Changes in overall smoking prevalence in Canada and the United States, 1999–2004*. *Gray bar signifies policy transition period from the policy introduction to full implementation.

The overall adjusted prevalence estimates, as well as by sex and education, are reported by year (Supplementary Appendix A–C) and pre- and post-June 2001 (GWL policy) (Supplementary Appendix D). The estimates that were adjusted for sociodemographic variables and cigarette prices are more comparable between the two countries as they better reflect the impact of the new warning labels, and are consistent with the DD and DDD models.

DD and DDD Regression Models

Results from the DD regression models showed that Canadian smoking prevalence decreased following the 2000 policy intervention (Table 3; Model 1 [OR = 0.84, 95% CI = 0.74–0.94] for interaction between country and the policy timing). The DDD model showed that there were no statistically significant differences in the impact of the GWL policy on current smoking by sex or education, comparing both countries, after the 2000 policy intervention, as evidenced by the triple interaction with country, policy timing, and sex or education (Table 3, Models 2 and 3, respectively). In other words, the significantly greater decrease in smoking prevalence in Canada from 1999 to 2004 after the implementation of the GWLs versus the United States did not differ by sex or by education. Results from the sensitivity analyses were similar to the main models in magnitude and significance.

Table 3.

Difference-in-Difference Models Examining the Impact of the Canadian 2001 GWL Policy on Smoking Prevalence by Sex and Education, 1999–2004

Variable	Odds ratio (95% CI)
	Main model (Model 1)	Interactive models
		Sex interaction (Model 2)	Education interaction (Model 3)
GWL policy * country
Canada after June 2001	0.83 (0.73–0.93)	0.81 (0.71–0.94)	0.78 (0.68–0.90)
Canada (vs. United States)	2.30 (1.37–3.88)	2.30 (1.37–3.88)	2.20 (1.30–3.71)
GWL policy—after June 2001	1.06 (1.03–1.09)	1.04 (1.00–1.07)	1.07 (1.04–1.11)
Sex (ref: female)	1.30 (1.28–1.32)	1.26 (1.23–1.30)	1.30 (1.28–1.32)
Age group (vs. 55+)
25–39	2.36 (2.32–2.41)	2.36 (2.32–2.41)	2.36 (2.31–2.41)
40–54	2.33 (2.28–2.37)	2.33 (2.28–2.37)	2.33 (2.28–2.37)
Education (vs. High school or less)
Some college/trade/technical school	0.73 (0.71–0.74)	0.73 (0.71–0.74)	0.73 (0.70–0.75)
Completed university or postgraduate	0.31 (0.30–0.32)	0.31 (0.30–0.32)	0.32 (0.31–0.33)
Not stated	0.62 (0.53–0.72)	0.62 (0.53–0.72)
ln(trend) continuous	0.99 (0.98–1.01)	0.99 (0.98–1.01)	0.99 (0.98–1.01)
ln(cig price) continuous	0.67 (0.63–0.71)	0.67 (0.63–0.71)	0.67 (0.64–0.71)
ln(trend) continuous * country	0.96 (0.89–1.04)	0.96 (0.89–1.04)	0.97 (0.90–1.05)
ln(cig price) continuous * country	1.49 (1.22–1.83)	1.49 (1.22–1.83)	1.48 (1.21–1.82)
Sex interaction
GWL policy * sex
Female, after June 2001 policy		1.05 (1.01–1.08)
Country * sex
Canada, female		1.00 (0.90–1.12)
GWL policy * country * female
Canada, female, after June 2001 policy		1.03 (0.90–1.18)
Education interaction
GWL policy * education
Some college/trade/technical school, after June 2001			1.00 (0.96–1.04)
Completed university or postgraduate, after June 2001			0.94 (0.90–0.98)
Country * education
Canada, some college/trade/technical school			0.95 (0.84–1.08)
Canada, completed university or postgraduate			1.17 (0.99–1.37)
GWL policy * country * education
Canada, some college/trade/technical school, after June 2001			1.07 (0.91–1.26)
Canada, completed university or postgraduate after June 2001			1.15 (0.94–1.41)

GWL = graphic warning label.

Discussion

The findings of our quasiexperimental study are consistent with prior research showing that the Canadian GWL policy decreased the overall smoking prevalence in Canada compared to the United States.⁹ It also extends current research⁹ to find no differential policy impact by sex or educational attainment. Our study differed from Huang et al.’s quasiexperimental study⁹ in three ways. First, we conducted individual-level analyses on nationally representative data sets, thereby being able to estimate policy effects on smoking prevalence at the national level, whereas Huang et al. analyzed aggregate-level data. Second, we analyzed the data across a relatively shorter time lag pre-policy of 2 years, while Huang et al. used a 9-year pre-policy period. Third, our adult sample was restricted to adults 25 years and older to allow for education completion, while the 2014 study sample included individuals aged 15 years and older. The convergence of our individual-level findings with Huang et al.’s aggregate-level findings lends stronger support to the notion that GWLs lead to decreases in smoking prevalence.

The unique aspect of the present study was our examination of the impact of the Canadian GWLs on prevalence by key sociodemographic variables. We observed no differences in the impact of the Canadian GWL policy on smoking prevalence by sex. Although there have been studies on the impact of GWLs on key proximal indicators of warning effectiveness (eg, salience, thinking about health harms, thinking about quitting), there have been no such studies that have examined the impact of GWLs on smoking prevalence by sex.³¹ Regardless, our findings do not align with other observational and experimental GWL studies that have examined the effect of GWLs by sex.³¹ Prior evidence shows that women react more strongly to graphic portrayed warnings on tobacco packages than men^16–18,32 and women are more likely to experience negative emotional responses, such as fear and disgust, after viewing GWLs.^16,32 Negative emotional responses may contribute to message acceptance and are associated with thinking about the health risks of smoking and cessation-related behaviors.^6,33 Women are also more likely to develop negative perceptions about smoking³² and to consider quitting^17,32 after being exposed to GWLs. However, our results support a 2018 International Tobacco Control Report which found no sex differences in the impact of health warnings on proximal measures such as label salience and harm perceptions.³⁴ Our prevalence findings from 1999 to 2004 suggest Canadian female smoking prevalence may be declining faster than male smoking prevalence, which is consistent with other studies.³⁵ Although it is unclear if the change in trends is caused by the combination of tobacco control efforts in Canada, our results show there was no differential decline in smoking prevalence by sex following the introduction of the GWL policy. However, future research should constantly monitor smoking prevalence among women as they remain targets of the tobacco industry.³⁶

Our study also found no differences in the impact of the GWL policy on smoking prevalence by education. We did not find any quasiexperimental studies that have tested for the differential impacts of GWL policies on smoking prevalence by education. However, a cross-sectional study in Canada observed a decrease in smoking prevalence among all educational groups in Canada from 1999 to 2006 using the CTUMS data.³⁷ Our findings showing no differential change in smoking prevalence by education suggest that the GWL policy was equally effective across all educational groups, which supports the aforementioned study.³⁷ Tangential evidence from experimental and other observational study designs suggests that individuals with less than a high school education consider GWLs on cigarette packages more effective than individuals with more education.^21,38 A cross-sectional study on the predictors of GWL avoidance among a predominantly low SES sample observed that avoidance did not detract from the benefits of GWLs, and the use of GWLs was an effective communication technique among low SES groups.¹⁹ Furthermore, less-educated individuals are more likely to notice health warnings on cigarette packages and recall the content compared to their more-educated counterparts.^14,19 Our study did not observe differences in the effectiveness of the GWL policy by education, suggesting that although GWLs are effective for individuals from low SES backgrounds, the effectiveness is not greater than that for communities from higher SES backgrounds. Individuals with low SES backgrounds have lower cessation rates than individuals from higher SES backgrounds,³⁹ and likely require more targeted policy and intervention approaches to decrease the long-standing disparities in smoking across SES groups. While our results showed that the 2000 GWL policy was equally effective across all educational levels, it also highlights the burden of tobacco use among the less-educated population. Like women, less-educated individuals are targets of tobacco industry marketing,⁴⁰ and additional policies beyond GWL interventions, such as those to promote access and use of cessation services,⁴¹ may be necessary to reduce the persistent disparities in tobacco use across educational levels. In 2012, Canada added a quitline number to the GWLs on the cigarette packages, while also including additional information about cessation resources on pack inserts.²² Post-implementation research among Canadians who smoked found faster wearout of attention and cognitive responses to pack labeling among lower SES groups.⁴² However, pre–post-implementation studies are needed to better evaluate any differential effects of information about quitlines among individuals from lower socioeconomic groups. Indeed, Canada is the only country whose GWL policy includes complementary insert messages, and both experimental and observational studies suggest that insert messages can promote self-efficacy and sustained cessation attempts,^43,44 yet the health equity impact of this messaging approach has not been assessed.

Our study included a number of strengths. Because we used individual-level analysis within the quasiexperimental design framework, we were able to control for age, sex, education, and cigarette prices, as well as time trends, which are important potential confounders.²⁹ Our use of DD models allowed us to make stronger inferences about the causal impact of GWLs.⁴⁵ Among the many advantages of the DD model is the validity of the estimates as it controls for unmeasured potential confounders that may be related to the outcome across both countries.⁴⁵ As the U.S. FDA and other countries consider implementing a GWL policy to curb tobacco use, our results, as well as others,^6,9,15,37 provide evidence to support their decision.

Our study also has some limitations. First, our analyses excluded young adults, 18–24 years old, so that the resulting sample would allow for a more appropriate analysis of education levels, but this strategy also led to a limitation in that we were not able to assess the overall impact of GWLs on young adults. Future research should examine the sociodemographic impact of this GWL policy on the important subpopulation of young adults. GWLs discourage smoking initiation among young adults, so there may be a greater impact among this population.⁶ Second, we did not consider smoking frequency (ie, every day or some days) or smoking intensity (ie, the number of cigarettes smoked per day), as neither measurement was available in the BFRSS survey. Detailed examination of the impact of GWLs on smoking frequency or intensity would have provided more insight into interpreting our conclusions. Third, we were unable to test the effect of the GWL policy on race/ethnicity, as the variable was not available in the CTUMS survey. Future research should examine the impact of GWL policy on race/ethnicity as well as its intersection with SES.⁴⁶ Finally, parallel to introducing the 2000 GWL policy, Canada and all provinces increased cigarette tax prices.⁴⁷ While we controlled for the national-level tax increases, we did not account for provincial-level tax increases. However, our results are comparable with other studies that also account for national-level tax increases.⁹

Conclusions

The 2000 Canadian GWL policy was an effective tobacco control measure that reduced overall smoking prevalence in Canada relative to the United States. The impact of the Canadian GWLs in reducing smoking prevalence did not reduce disparities by sex and education. Although beneficial for all smokers, GWLs may not serve to decrease existing smoking disparities. Policy makers should consider policies that will reduce the burden of tobacco use among population subgroups that are most affected by smoking, including the less-educated groups.

Supplementary Material

A Contributorship Form detailing each author’s specific involvement with this content, as well as any supplementary data, are available online at https://academic.oup.com/ntr.

Funding

Research reported in this publication was supported by the National Cancer Institute of the National Institutes of Health (R37CA214787), which provided additional support to JFT (R01CA215466). Additional support for GTF was provided by a Senior Investigator Grant from the Ontario Institute for Cancer Research (IA-004) and the Canadian Cancer Society O. Harold Warwick Prize. Additional support to GM, MET, and GTF is provided by the Canadian Institutes of Health Research Foundation Grant (FDN-148477). The content is solely the responsibility of the authors and does not necessarily represent the official views of NIH.

Declarations of Interests

GTF and JFT have served as expert witnesses or consultants for governments defending their country’s policies or regulations in litigation. GTF served as a paid expert consultant to the Ministry of Health of Singapore in reviewing the evidence on plain/standardized packaging. All other authors have no conflicts of interest to declare.

Data Availability

In each country participating in the International Tobacco Control Policy Evaluation (ITC) Project, the data are jointly owned by the lead researcher(s) in that country and the ITC Project at the University of Waterloo. Data from the ITC Project are available to approved researchers 2 years after the date of issuance of cleaned data sets by the ITC Data Management Centre. Researchers interested in using ITC data are required to apply for approval by submitting an International Tobacco Control Data Repository (ITCDR) request application and subsequently to sign an ITCDR Data Usage Agreement. The criteria for data usage approval and the contents of the Data Usage Agreement are described online (http://www.itcproject.org). The authors of this paper obtained the data following this application process. They did not have any special access privileges. Others would be able to access these data in the same manner as the authors.