Skip to main content
PMC home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2016 Jul 11.
Published in final edited form as: Int J Epidemiol. 2008 Jan 6;37(3):615–624. doi: 10.1093/ije/dym250

Educational attainment and cigarette smoking: a causal association?†

Stephen E Gilman 1,2,*, Laurie T Martin 3, David B Abrams 4, Ichiro Kawachi 1, Laura Kubzansky 1, Eric B Loucks 5, Richard Rende 6, Rima Rudd 1, Stephen L Buka 7
PMCID: PMC4939617  NIHMSID: NIHMS779597  PMID: 18180240

Abstract

Background

Despite abundant evidence that lower education is associated with a higher risk of smoking, whether the association is causal has not been convincingly established.

Methods

We investigated the association between education and lifetime smoking patterns in a birth cohort established in 1959 and followed through adulthood (n = 1311). We controlled for a wide range of potential confounders that were measured prior to school entry, and also estimated sibling fixed effects models to control for unmeasured familial vulnerability to smoking.

Results

In the full sample of participants, regression analyses adjusting for multiple childhood factors (including socioeconomic status, IQ, behavioural problems, and medical conditions) indicated that the number of pack-years smoked was higher among individuals with less than high school education [rate ratio (RR) = 1.58, confidence interval (CI) = 1.31, 1.91]. However, in the sibling fixed effects analysis the RR was 1.23 (CI = 0.80, 1.93). Similarly, adjusted models estimated in the full sample showed that individuals with less than high school education had fewer short-term (RR = 0.40; CI = 0.23, 0.69) and long-term (RR = 0.59; CI = 0.42, 0.83) quit attempts, and were less likely to quit smoking (odds ratio = 0.34; CI = 0.19, 0.62). The effects of education on quitting smoking were attenuated in the sibling fixed effects models that controlled for familial vulnerability to smoking.

Conclusions

A substantial portion of the education differential in smoking that has been repeatedly observed is attributable to factors shared by siblings that contribute to shortened educational careers and to lifetime smoking trajectories. Reducing disparities in cigarette smoking, including educational disparities, may therefore require approaches that focus on factors early in life that influence smoking risk over the adult life span.

Keywords: Education, smoking, causality, disparities

Introduction

Educational differences in smoking were not observed prior to the discovery that smoking causes death. However, following the 1964 Surgeon General’s report1 and subsequent public health campaigns, researchers established a large educational difference between those who do and do not smoke. In the United States, for example, the annual decline in smoking prevalence between 1974 and 1985 in the National Health Interview Survey was approximately five times higher among the most educated than among the least educated.2 The gap in smoking rates between individuals with the highest and lowest levels of education is greater than at any time in the past.3–6 Educational differences in the risk for daily smoking initiation, development of nicotine dependence,7 and in smoking cessation2 contribute to this gap.

A causal effect of education on smoking could provide an additional input into policy decisions regarding tobacco control. However, the possibility that the association between education and smoking is not causal casts doubt on the public health relevance of studies showing a link between education and smoking. This possibility was raised as early as 1982 by Farrell and Fuchs,8 who concluded that because smoking patterns were established by age 17, they could not be influenced by years of schooling. They argued instead that ‘there are apparently one or more ‘third variables’ that affect both smoking and years of schooling’,8 (p. 228) and hypothesized that such ‘third variables’ include intelligence9 and time discount,10 e.g. the propensity to refrain from smoking in anticipation of future health benefits. Few of the numerous studies reporting on educational differences in smoking attempted to rule out these or other11–15 alternative explanations for the association between lower educational attainment and higher rates of smoking.

Establishing causality in the absence of experimental data is a virtually intractable problem; however, data arising from ‘natural experiments’ can be used to overcome some of the limitations of observational studies when it is not feasible to conduct a randomized trial. One such natural experiment occurs in the context of a family study, by comparing outcomes of ‘discordant siblings’.16 In this design, siblings with different levels of education (i.e. discordant for education) are compared with respect to their smoking outcomes; this design removes the confounding effects of factors that are shared by both siblings. This includes both environmental and genetic factors that siblings have in common, but not factors such as individual-specific experiences in the social environment or genetic susceptibility that is not shared with siblings. The objectives of the current study are to investigate the association between education and smoking using analyses adjusting for potential confounders measured prior to school entry, and using sibling fixed effects models that adjust for unmeasured familial vulnerability.

Methods

Participants were offspring of pregnant women enrolled in the National Collaborative Perinatal Project (NCPP) between 1959 and 1966.17 The original aims of the NCPP were to identify the developmental consequences of pregnancy and delivery complications. Detailed social and medical histories were obtained from mothers at the time of enrolment. Information on offspring birth outcomes and subsequent growth and development was obtained several times during the first year of life, and again at ages 4 and 7.

The New England Family Study (NEFS) was established to locate and interview the adult NCPP offspring at the Providence, Rhode Island and Boston, Massachusetts sites (n = 17921) between 2001–04. Participants in the current study were selected through a multi-stage sampling procedure as part of the Brown-Harvard Transdisciplinary Tobacco Use Research Center, which involved a core assessment interview and three component studies. Screening questionnaires were mailed to 4579 of the 15 721 Boston and Providence NCPP offspring who survived until age 7. Of the 3121 questionnaires returned (68.2%), 2271 were eligible for participation based on the combined inclusion criteria of the three component studies. In total, we enrolled 1674 NCPP offspring. Participants enrolled in the NEFS had a somewhat higher level of education (e.g. 64.1% with at least some college education) than participants who were eligible but not enrolled (e.g. 51.8% with at least some college education). Data from 49 individuals were excluded from the final sample because of participation in a pilot version of the survey (n = 4) or because of problems with the interview administration (n = 45). This yielded 1625 completed adult assessments. The analysis sample for the current study was restricted to participants who reported having smoked at least once in their lifetime and had complete data on all key study variables.

Measures

Educational attainment

Education was assessed during the NEFS follow-up interview and was classified according to five categories: (i) less than high school or GED; (ii) high school degree; (iii) high school degree plus additional technical training or certificate; (iv) some college and (v) college degree.

Smoking

Smoking histories were obtained by the Life Interview of Smoking Trajectories and Quitting Methods Questionnaire, developed by the Methods and Measurement core of the Brown-Harvard Transdisciplinary Tobacco Use Research Center. These instruments obtain detailed information on participants’ experiences with smoking beginning from experimentation, progression to regular smoking, levels of consumption, nicotine dependence and patterns of quit attempts. Regular smoking was defined as a positive response to the question ‘Did you ever become a weekly smoker (that is, smoke at least once per week for two months or longer)?’ We created a summary measure of cigarette consumption using data on participants’ smoking intensity and duration during their heaviest smoking phase; similar to measures of ‘pack-years’,18 this was calculated as the number of years of participants’ heaviest smoking phase × number of cigarettes per day/20. Nicotine dependence was defined according to Diagnostic and Statistical Manual, Fourth Edition, criteria (DSM-IV),19 and was assessed using the Composite International Diagnostic Interview (CIDI).20 Smoking desistance (among participants who became regular smokers) was defined by the number of 24 h quit attempts, number of 3 month abstinence periods and smoking cessation. Cessation was coded as positive if participants did not smoke during the year preceding the interview. Participants also provided information on their ages of onset of regular smoking and nicotine dependence symptoms, and age of smoking cessation.

Potential confounders

Information on early childhood factors was collected during the NCPP upon the mother’s enrollment and again during the 7-year follow-up interview. Measures of parental socioeconomic status from both assessments include occupation, educational attainment and an indicator of household income below the United States poverty threshold.21 Additional characteristics of early childhood environment between birth and age 7 were household crowding, family disruption by age 7, number of moves and father’s unemployment. Maternal smoking during pregnancy, defined as the maximum number of cigarettes smoked per day, was coded as none, less than one pack (1–19 cigarettes), and a pack or more (20+ cigarettes).

Measures of early childhood physical health included birthweight, number of medical conditions during the child’s first year of life and history of asthma by age 7. Psychological development included full-scale IQ at age 7,22 psychologist’s abnormal behaviour rating at age 7 and the presence of neurological soft signs at 7 years.23

Analysis procedures

Analyses were conducted among participants who had smoked at least once in their lifetime, and from whom complete data were available on the early childhood measures obtained during the NCPP. The study was limited to lifetime smokers because experimentation with cigarettes was pervasive in the sample (91.1%), and because our primary concern was with smoking intensity and frequency rather than lifetime abstinence.

Discrete-time survival analysis24 was used to investigate the development of nicotine dependence (among regular smokers) and smoking cessation (also among regular smokers). Educational attainment was coded as a time-varying covariate, with person-years prior to school completion coded as ‘in-school’; the coefficients for education therefore reflect the effects of participants’ final educational attainment on nicotine dependence and smoking cessation. Analyses of cigarette consumption (pack-years) and quit attempts were conducted in the sample of regular smokers using negative binomial regression.25 Regression coefficients from this model, when exponentiated, indicate the ratio of the number of quit attempts (or rate of consumption) associated with a unit change in each covariate.

For each outcome, we present the results of two sets of analyses: (i) analyses based on the full analysis sample and (ii) analyses of siblings discordant for education, i.e. individuals with a sibling in the NEFS follow-up study who had a different level of education. For the survival26 and negative binomial27 regressions in the full sample, a random intercept for each family was included to account for the interdependence of data from siblings using the GLIMMIX procedure in SAS.28 In the discordant sibling analyses, we used conditional logistic regression to estimate discrete-time survival models, and estimated fixed effects negative binomial models of count data.29 These approaches adjust the effect estimates for all between-family variation in the smoking outcomes.30

Results

Of the 1625 participants who were interviewed, 1445 (88.9%) provided data on all childhood covariates. A total of 92.1% of these participants (n = 1311) reported lifetime smoking, and therefore comprised the analysis sample for the current study. A comparison of demographic characteristics between the full interviewed sample of 1625 and the analysis sample of 1331 lifetime smokers is shown in Table 1. The samples are similar with respect age, sex, race/ethnicity and the number of siblings per family. The mean (SD) age of the analysis sample is 39.1 years (1.8); the sample is 59.5% females (n = 780), and 84.0% Whites (n = 1101). 10.8% of the sample has less than a high school education (n = 142), while one-third has a college degree (n = 365). The number of siblings in the full and analysis samples is also shown in Table 1. The analysis sample represents 1036 families; 793 participants did not have a sibling in the study, whereas the remaining 518 participants represent 243 families. The age range of siblings is an approximate indicator of the extent of shared environmental experiences in early childhood, as siblings closer in age are likely to have more shared experiences than siblings further apart. The age difference between siblings in the analysis sample was quite narrow [mean (SD): 2.2 (1.3) years].

Table 1.

Demographic characteristics of participants in the New England Family Study (NEFS) and participants in the analysis sample

All NEFS
participants
Percent (n)
(n = 1625)		 Analysis
sample
Percent (n)
(n = 1311)
Age, Mean (SD) 39.1 (1.9) 39.1 (1.8)
Female sex 59.2 (962) 59.5 (780)
Race/ethnicity
  White 83.5 (1354) 84.0 (1101)
  Black 9.6 (155) 9.1 (119)
  Other 7.0 (113) 6.9 (91)
Educational attainment
  Less than high
school or GED		 11.4 (185) 10.8 (142)
  High school degree 13.7 (222) 13.8 (181)
  High school plus
additional training		 16.1 (262) 16.7 (219)
  Some college 30.0 (488) 30.8 (404)
  College graduate 28.8 (468) 27.8 (365)
Number of siblings per family
  1 56.8 (923) 60.5 (793)
  2 34.5 (560) 32.8 (430)
  3 7.2 (117) 5.7 (75)
  4 1.2 (20) 0.6 (8)
  5 0.3 (5) 0.4 (5)
Open in a new tab

Patterns of smoking uptake and quitting are shown in Table 2. A total of 63.3% of the sample progressed to regular smoking (n = 826). Among regular smokers, 66.1% (n = 506) subsequently met diagnostic criteria for nicotine dependence and 43.1% (n = 355) quit smoking. The majority of participants who became regular smokers did so in the teen years [mean age (SD): 15.9 (4.1)], with the mean (SD) onset of nicotine dependence occurring 9.4 (7.0) years afterwards. The mean age (SD) of smoking cessation among regular smokers who had quit by the adult follow-up interview was 27.7 (6.7). On average, regular smokers reported smoking 8.4 (9.6) pack-years, made 9.6 (19.6) quit attempts that lasted at least 24 h, and achieved 1.6 (2.1) periods of abstinence that lasted at least 3 months. The remaining columns in Table 2 show the proportion of participants with missing data for each outcome, the number of participants included in each analysis, and the distribution of the smoking outcomes among discordant siblings. While the distributions of the smoking outcomes in the sample of discordant siblings are similar to those in the full analysis sample, the sample size is smaller.

Table 2.

Distribution of cigarette smoking outcomes and patterns of missing data in the full sample, and the sample of siblings discordant for educational attainment, New England Family Study

Full analysis sample
 Discordant siblings only
Mean (SD)
or per cent (n)		 Sample
size		 Per cent
missing (n)		 Mean (SD)
or per cent (n)		 Sample
size
Smoking uptake
Onset of regular smokinga 63.3 (826) 1305 0.5 (6) 62.8 (215) 348
Onset of nicotine dependenceb 66.1 (506) 765 7.4 (61) 66.2 (100) 151
Mean cigarette consumption (pack-years)b 8.4 (9.6) 787 4.7 (39) 7.7 (9.6) 148
Smoking desistance
Mean 24 h quit attemptsb 9.6 (19.6) 785 5.0 (41) 9.8 (20.4) 144
Mean 3 month abstinence periodsb 1.6 (2.1) 807 2.3 (19) 1.6 (2.1) 163
Smoking cessationb 43.1 (355) 824 0.2 (2) 44.6 (74) 166
Open in a new tab
a

The prevlance of regular smoking was estimated in the sample of lifetime smokers (n = 1311).

b

Estimated in the sample of regular smokers.

Smoking trajectories in relation to educational attainment

The majority of participants who became regular smokers (84.7%, n = 700) did so prior to completing school. Therefore, we do not present analyses of the relation between education and smoking initiation. Among participants who became regular smokers, there was no association between educational attainment and the development of nicotine dependence according to DSM-IV criteria. In contrast, there was a strong education gradient in lifetime smoking patterns (Table 3). Participants without high school degrees smoked ~50% more pack-years than participants with college degrees; rate ratios (RRs) in the full sample were 1.63 (1.38–1.92) in Model I (unadjusted), and 1.58 (1.31–1.91) in Model II (controlling for participant demographic factors and childhood covariates assessed during the NCPP). However, the RR for the effect of low education on pack-years was attenuated in the sibling fixed effects model [1.23, confidence interval (CI): 0.80–1.93].

Table 3.

Educational attainment and lifetime smoking patterns in the New England Family Study

Cigarette consumptiona
 Number of 24-hr quit attemptsb
Educational attainment Model Id
(Full sample)
RR (95% CI)		 Model IIe
(Full sample)
RR (95% CI)		 Model IIIf
(Discordant siblings)
RR (95% CI)		 Model Id
(Full sample)
RR (95% CI)		 Model IIe
(Full sample)
RR (95% CI)		 Model IIIf
(Discordant siblings)
RR (95% CI)
Less than high school or GED 1.63 (1.38, 1.92) 1.58 (1.31, 1.91) 1.23 (0.80, 1.93) 0.38 (0.23, 0.62) 0.40 (0.23, 0.69) 1.04 (0.39, 2.82)
High school degree 1.46 (1.24, 1.72) 1.41 (1.19, 1.67) 1.28 (0.91, 1.81) 0.70 (0.43, 1.14) 0.77 (0.47, 1.27) 1.97 (0.77, 5.06)
High school plus additional training 1.43 (1.23, 1.67) 1.34 (1.15, 1.57) 1.10 (0.78, 1.56) 0.69 (0.44, 1.09) 0.69 (0.44, 1.09) 1.33 (0.52, 3.40)
Some college 1.36 (1.19, 1.55) 1.33 (1.16, 1.53) 1.22 (0.93, 1.60) 0.58 (0.39, 0.87) 0.70 (0.47, 1.04) 1.30 (0.62, 2.73)
College graduate 1 1 1 1 1 1
Number of 3-month abstinence periodsb
 Smoking cessationc
Educational attainment Model Id
(Full sample)
RR (95% CI)		 Model IIe
(Full sample)
RR (95% CI)		 Model IIIf
(Discordant siblings)
RR (95% CI)		 Model Id
(Full sample)
OR (95% CI)		 Model IIe
(Full sample)
OR (95% CI)		 Model IIIf,g
(Discordant siblings)
OR (95% CI)
Less than high school or GED 0.50 (0.37, 0.66) 0.59 (0.42, 0.83) 0.61 (0.26, 1.46) 0.22 (0.13, 0.37) 0.34 (0.19, 0.62) 1.01 (0.28, 3.64)
High school degree 0.61 (0.47, 0.80) 0.63 (0.47, 0.85) 0.47 (0.27, 0.82) 0.50 (0.33, 0.77) 0.67 (0.41, 1.08)
High school plus additional training 0.78 (0.61, 1.00) 0.84 (0.64, 1.09) 0.98 (0.57, 1.71) 0.53 (0.36, 0.79) 0.64 (0.42, 1.00) 0.88 (0.20, 3.80)
Some college 0.84 (0.68, 1.03) 0.88 (0.69, 1.10) 0.70 (0.42, 1.14) 0.65 (0.46, 0.92) 0.78 (0.53, 1.15) 1.13 (0.31, 4.14)
College graduate 1 1 1 1 1 1
Open in a new tab
a

Cigarette consumption defined as the number of pack-years smoked during heaviest smoking phase. RRs obtained from negative binomial regression models, offset by the logarithm of the number of years’ duration of heaviest smoking phase.

b

Rate ratios obtained from negative binomial regression models.

c

Odds ratios obtained from discrete-time survival models including person-years at risk from age at regular smoking through age at cessation or censoring/age at interview.

d

Model I includes a random intercept for each sibling set.

e

Model II includes a random intercept for each sibling set, and adjusts for age, sex, race/ethnicity, age at regular smoking, parental occupation, parental education, household poverty, household crowding, parental separation, residential instability, paternal unemployment, maternal smoking during pregnancy, IQ, abnormal behaviour ratings, neurological soft signs, birthweight, medical conditions during first year of life and childhood asthma.

f

Model III is a fixed-effects/conditional analysis among siblings discordant for education, with adjustment for age, sex, age at regular smoking, maternal smoking during pregnancy, IQ, abnormal behaviour ratings, neurological soft signs, birthweight, medical conditions during first year of life and childhood asthma.

g

Less than high school or GED and high school degree categories combined because of sparse data.

Short-term (24 h) quit attempts and long-term (3 month) periods of abstinence were less frequent among participants with a high school degree or less, with approximately half as many quit attempts as those participants who graduated from college. Adjusted rate ratios (and 95% CI) for short- and long-term quit attempts were 0.40 (0.23–0.69) and 0.59 (0.42–0.83), respectively. The effects of less than high school educational attainment on short-term (RR: 1.04; CI: 0.39–2.82) and long-term (RR: 0.61; CI: 0.26–1.46) quit attempts were attenuated in the sibling models. We observed a similar pattern in the analyses of smoking cessation, wherein low education was associated with a lower odds of cessation in Models I (OR: 0.22; CI: 0.13–0.37) and II (OR: 0.34; CI: 0.19–0.62), but not among discordant siblings in Model III.

Discussion

The objective of the current study was to evaluate evidence for a potential causal effect of educational attainment on lifetime patterns of cigarette smoking. This association has emerged repeatedly from epidemiologic studies in the United States, 31–33 Europe,34,35 Australasia36 and several emerging market economies.37–39 The relevance of this evidence for public health and social policy will remain uncertain until researchers can establish whether educational attainment is itself a causal factor in the aetiology of cigarette smoking rather than alternative factors that confer vulnerability to attain less education and to smoke.

We evaluated whether there is evidence consistent with a causal association between education and smoking patterns by controlling for a wide range of potential confounders that were measured prior to school entry and by analysing differences in the smoking outcomes of siblings with different levels of education. After controlling for measured confounders, lower education was associated with more pack-years of smoking, fewer quit attempts and a lower likelihood of cessation. The effect of education was characterized by a higher risk of smoking frequency/intensity among participants without high school degrees compared to those who had graduated college. However, after controlling for between-family variation in smoking outcomes in the sibling fixed effects analyses, evidence for an effect of education on smoking was substantially weaker. CIs were also wider among discordant siblings due to the smaller number of participants in these analyses.

Among smokers, there was no association between education and nicotine dependence. The absence of an association between education and nicotine dependence among regular smokers in the NEFS sample is consistent with findings of Breslau et al.40 from the National Comorbidity Survey, but is different from the study of Hu et al.7 of a much younger sample of smokers from the National Longitudinal Survey of Adolescent Health. Inconsistencies may be due to age differences between the samples41 and to differences in the conceptualization of nicotine dependence.42

Results from this study underscore the importance of differentiating the processes underlying the progression to regular smoking43 and the pathways to desistance.44 This necessitates a developmentally sensitive analysis that takes into account the age distributions of each smoking transition. While aspects of the school environment may contribute to the development of regular smoking, it is unlikely that level of completed education was a major contributor to smoking initiation given the early onset of smoking. However, once regular smoking behaviours were established, our initial results indicated that educational attainment had a significant impact on adult smoking trajectories as indexed by pack-years, quit attempts and cessation. Adjusting for a wide range of factors measured prior to school entry (e.g. parental socioeconomic status, maternal smoking, IQ) had little impact on the effect estimates for educational attainment. Rather, the attenuation of the effects of education in the discordant sibling analyses suggests that other, unmeasured factors operating at the family level (i.e. shared by siblings), contribute to the relation between education and smoking behaviours.

Limitations

Our analyses rely on retrospective reports of lifetime smoking histories. Self-reported smoking status has a high level of agreement with serum cotinine-defined smoking status.45,46 The correspondence between self-reported smoking and serum cotinine concentrations has not been found to differ according to education or other indicators of socioeconomic status.45,47 With respect to cigarette consumption, studies comparing consumption levels reported retrospectively to those obtained prospectively show a moderate to high level of agreement.18,48,49 Fewer studies examined retrospectively reported ages of smoking initiation, although there is some evidence in support of 50,51 and against52 the reliability of such reports.

The NCPP cohort was not designed to be a representative sample of all births in Rhode Island and Massachusetts, and the adult offspring of NCPP participants who were included in the current study on the basis of several layers of inclusion criteria cannot be considered a representative sample of adults from this geographic area. The discordant sibling analyses are limited by small sample size, thereby reducing the precision of effect estimates. As these analyses rely solely on within-family variation, ignoring all variation in smoking outcomes between-families, the consequences of lower sample size on statistical power are potentially substantial.

While information was available on parental socioeconomic status and early childhood physical and cognitive development, information was not available on all potential confounders. To the extent that these potential confounders represent vulnerability to smoking that is shared by siblings, their effects were accounted for in the fixed effects analyses. The discordant sibling design does not control for exposures that are not shared by the siblings. A potentially relevant class of non-shared factors includes those that cause siblings to attain different levels of education. However, controlling for covariates such as IQ, medical conditions and parental employment in these analyses strengthens the discordant sib design, as these would be a likely source of sibling differences in education.

Implications of our findings for understanding the relation between education on smoking

Our analyses of education differentials in smoking in the full sample are consistent with prior studies showing strong protective effects of schooling on smoking behaviours; these effects remained even after adjusting for a wide range of potential confounding factors measured prior to school entry. The fixed effects models that we estimated among siblings in the NEFS suggest that some portion of the effect of education on smoking is due to unmeasured familial vulnerability. Several prior studies53–55 have also attempted to control for unobserved confounders. Using the method of instrumental variables, they reported statistically significant effects of education on smoking, although the magnitude of the effects was substantially smaller than in standard regression models. The validity of the instrumental variable approach rests on identifying appropriate instruments—in this case, factors that influence smoking exclusively through education. Sander,53 however, used family background characteristics as instruments, which we have shown to predict smoking independent of education.11 It is unclear whether using stronger instruments would lead to a further attenuation of the effect of education on smoking. In contrast, the fixed effects method used in our study reduces confounding due to shared sibling vulnerability, but suffers from a loss of efficiency because estimates are based solely on within-family variation in smoking behaviours. Taken as a whole, we are hesitant to conclude that there is no causal effect of education on smoking solely on the basis of the fixed effects analyses in our study, given the strong effects that emerged from the full sample analyses and the limitations of the sibling models that we describe earlier. It is therefore important to identify, and estimate the magnitude of effects of, the specific factors that represent a common source of vulnerability to attain lower levels of education and to persist in smoking through adulthood.

Explanations for education differentials in smoking behaviours include differential valuations of the health consequences of smoking,54,56 and differences in access to and effectiveness of cessation treatments.57 Lower education may also confer risk for persistent smoking due to lower occupational status and financial strains,36 and be a more frequent mechanism for coping with stress among individuals with lower levels of education.58

Evidence that aspects of the school59 and neighbourhood environments60 increase risk for smoking among adolescents suggests that school experiences are relevant in addition to the educational degree attained. Relevant aspects of the school environment that could contribute to the development of smoking include school smoking policies, affiliation with deviant peers, and economic status.61,62 Evidence that smokers have shortened educational careers raises the possibility of a reciprocal effect of educational achievements and smoking behaviours over time.63,64

Inequalities in smoking by educational attainment are a major contributor to educational inequalities in mortality.33,65,66 Therefore, tobacco control efforts could have a significant impact on reducing health disparities. The identification of lower education as a marker of ‘hard core’ smoking, both in community67 and clinical samples,57 suggests that multiple types of tobacco control efforts will be needed, including community-wide programs,68 workplace69 and school interventions.70

KEY MESSAGES.

  • There is abundant evidence that cigarette smoking is more prevalent among individuals with lower education.

  • It remains unresolved whether low education is a causal risk factor for smoking, or whether educational disparities in smoking are attributable to factors that confer risk for low education and for smoking.

  • We investigated the relation between education and smoking in a birth cohort followed through middle adulthood, controlling for a wide range of potential confounders measured prior to school entry. We also estimated fixed effects models to control for unmeasured confounders shared by siblings.

  • In fully adjusted models, we observed education differentials in cigarette consumption, frequency of quit attempts, and likelihood of quitting that were similar to prior reports. These differentials were reduced in sibling models controlling for unmeasured confounders.

  • A portion of the education differential in smoking is attributable to factors shared by siblings that contribute to shortened educational careers and to lifetime smoking trajectories.

Acknowledgments

The authors acknowledge collaborators on the Brown-Harvard TTURC for the development of the Life Interview of Smoking Trajectories, Quitting Methods Questionnaire and Socioeconomic Status Interview (David Abrams, Laurie Chassin, Melissa Clark, Suzanne Colby, Raymond Niaura, Edmond Shenassa and Saul Shiffman). We sincerely appreciate the efforts and contributions of Michelle Rogers and Kathleen McGaffigan for data management and programming. This research was supported in part by a Transdisciplinary Tobacco Use Research Center (TTURC) Award (P50 CA084719) and grant AG023397 (Buka) from NIH, and by the Robert Wood Johnson Foundation.

Footnotes

†

A previous version of this work was presented at the Society for Epidemiologic Research/North American Congress of Epidemiology, June 2006, Seattle, WA, USA.

Conflict of interest: None declared.

References

  • 1.United States. Smoking and Health; Report of the Advisory Committee to the Surgeon General of the Public Health Service. Washington: U.S. Department of Health, Education, and Welfare, Public Health Service; 1964. Surgeon General’s Advisory Committee on Smoking and Health. [Google Scholar]
  • 2.Pierce JP, Fiore MC, Novotny TE, Hatziandreu EJ, Davis RM. Trends in cigarette smoking in the United States. Educational differences are increasing. JAMA. 1989;261:56–60. [PubMed] [Google Scholar]
  • 3.Giovino GA, Henningfield JE, Tomar SL, Escobedo LG, Slade J. Epidemiology of tobacco use and dependence. Epidemiol Rev. 1995;17:48–65. doi: 10.1093/oxfordjournals.epirev.a036185. [DOI] [PubMed] [Google Scholar]
  • 4.Iribarren C, Luepker RV, McGovern PG, Arnett DK, Blackburn H. Twelve-year trends in cardiovascular disease risk factors in the Minnesota Heart Survey. Are socio-economic differences widening? Arch Intern Med. 1997;157:873–881. [PubMed] [Google Scholar]
  • 5.Escobedo LG, Peddicord JP. Smoking prevalence in US birth cohorts: the influence of gender and education. Am J Public Health. 1996;86:231–236. doi: 10.2105/ajph.86.2.231. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Centers for Disease Control and Prevention. Cigarette smoking among adults—United States, 2004. MMWR. 2005;54:1121–1124. [PubMed] [Google Scholar]
  • 7.Hu MC, Davies M, Kandel DB. Epidemiology and correlates of daily smoking and nicotine dependence among young adults in the United States. Am J Public Health. 2006;96:299–308. doi: 10.2105/AJPH.2004.057232. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Farrell P, Fuchs VR. Schooling and health: the cigarette connection. J Health Econ. 1982;1:217–230. doi: 10.1016/0167-6296(82)90001-7. [DOI] [PubMed] [Google Scholar]
  • 9.Kubicka L, Matejcek Z, Dytrych Z, Roth Z. IQ and personality traits assessed in childhood as predictors of drinking and smoking behaviour in middle-aged adults: a 24-year follow-up study. Addiction. 2001;96:1615–1628. doi: 10.1080/09652140120080741. [DOI] [PubMed] [Google Scholar]
  • 10.Khwaja A, Silverman D, Sloan F. Time preference, time discounting, and smoking decisions. J Health Econ. 2007;26:927–949. doi: 10.1016/j.jhealeco.2007.02.004. [DOI] [PubMed] [Google Scholar]
  • 11.Gilman SE, Abrams DB, Buka SL. Socioeconomic status over the life course and stages of cigarette use: initiation, regular use, and cessation. J Epidemiol Community Health. 2003;57:802–808. doi: 10.1136/jech.57.10.802. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Chassin L, Presson CC, Sherman SJ, Edwards DA. Parent educational attainment and adolescent cigarette smoking. J Subst Abuse. 1992;4:219–234. doi: 10.1016/0899-3289(92)90031-r. [DOI] [PubMed] [Google Scholar]
  • 13.Flay BR, Hu FB, Siddiqui O, et al. Differential influence of parental smoking and friends’ smoking on adolescent initiation and escalation of smoking. J Health Soc Behav. 1994;35:248–265. [PubMed] [Google Scholar]
  • 14.Peterson AV, Jr, Leroux BG, Bricker J, et al. Nine-year prediction of adolescent smoking by number of smoking parents. Addict Behav. 2006;31:788–801. doi: 10.1016/j.addbeh.2005.06.003. [DOI] [PubMed] [Google Scholar]
  • 15.Wolfinger NH. The effects of parental divorce on adult tobacco and alcohol consumption. J Health Soc Behav. 1998;39:254–269. [PubMed] [Google Scholar]
  • 16.Pomerleau CS, Pomerleau OF, Snedecor SM, Gaulrapp S, Kardia SL. Heterogeneity in phenotypes based on smoking status in the Great Lakes Smoker Sibling Registry. Addict Behav. 2004;29:1851–1855. doi: 10.1016/j.addbeh.2004.03.036. [DOI] [PubMed] [Google Scholar]
  • 17.Niswander KR, Gordon M. The Women and their Pregnancies: The Collaborative Perinatal Study of the National Institute of Neurological Diseases and Stroke. Washington: National Institute of Health; 1972. [Google Scholar]
  • 18.Bernaards CM, Twisk JW, Snel J, Van Mechelen W, Kemper HC. Is calculating pack-years retrospectively a valid method to estimate life-time tobacco smoking? A comparison between prospectively calculated pack-years and retrospectively calculated pack-years. Addiction. 2001;96:1653–1661. doi: 10.1046/j.1360-0443.2001.9611165311.x. [DOI] [PubMed] [Google Scholar]
  • 19.American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders: DSM-IV-TR. 4th. Washington, DC: American Psychiatric Association; 2000. [Google Scholar]
  • 20.Kessler RC, Wittchen H-U, Abelson JM, et al. Methodological studies of the Composite International Diagnostic Interview (CIDI) in the U.S. National Comorbidity Survey. Int J Methods Psychiat Res. 1998;7:33–55. [Google Scholar]
  • 21.U.S. Bureau of the Census. [Accessed 11 December, 2007];Weighted Average Poverty Thresholds for Families of Specified Size: 1959 to 2006. Available at: http://www.census.gov/hhes/www/poverty/histpov/hstpov1.html.
  • 22.Wechsler D. Wechsler Intelligence Scale for Children; Manual. New York: Psychological Corp.; 1949. [Google Scholar]
  • 23.Shaffer D, Schonfeld I, O’Connor PA, et al. Neurological soft signs. Their relationship to psychiatric disorder and intelligence in childhood and adolescence. Arch Gen Psychiatry. 1985;42:342–351. doi: 10.1001/archpsyc.1985.01790270028003. [DOI] [PubMed] [Google Scholar]
  • 24.Cox DR. Regression models and life-tables. J Roy Statl Soc B Met. 1972;34:187–220. [Google Scholar]
  • 25.McCullagh P, Nelder JA. Generalized Linear Models. 2nd. London, New York: Chapman and Hall; 1989. [Google Scholar]
  • 26.Barber JS, Murphy SA, Axinn WG, Maples J. Discrete-time multilevel hazard analysis. Sociol Methodol. 2000;30:201–235. [Google Scholar]
  • 27.Cameron AC, Trivedi PK. Regression Analysis of Count Data. Cambridge, UK, New York, NY, USA: Cambridge University Press; 1998. [Google Scholar]
  • 28.SAS Institute. Base SAS 9.1.3 Procedures Guide. Cary, NC: SAS Pub.; 2004. [Google Scholar]
  • 29.Allison PD, Waterman R. Fixed effects negative binomial regression models. In: Stolzenberg RM, editor. Sociological Methodology. Oxford: Basil Blackwell; 2002. pp. 247–265. [Google Scholar]
  • 30.Allison PD. Fixed Effects Regression Methods for Longitudinal Data using SAS. Cary, NC: SAS Institute; 2005. [Google Scholar]
  • 31.Wagenknecht LE, Perkins LL, Cutter GR, et al. Cigarette smoking behavior is strongly related to educational status: the CARDIA study. Prev Med. 1990;19:158–169. doi: 10.1016/0091-7435(90)90017-e. [DOI] [PubMed] [Google Scholar]
  • 32.Winkleby MA, Schooler C, Kraemer HC, Lin J, Fortmann SP. Hispanic versus white smoking patterns by sex and level of education. Am J Epidemiol. 1995;142:410–418. doi: 10.1093/oxfordjournals.aje.a117649. [DOI] [PubMed] [Google Scholar]
  • 33.Clay CM, Dyer AR, Liu K, et al. Education, smoking and non-cardiovascular mortality: findings in three Chicago epidemiological studies. Int J Epidemiol. 1988;17:341–347. doi: 10.1093/ije/17.2.341. [DOI] [PubMed] [Google Scholar]
  • 34.Cavelaars AE, Kunst AE, Geurts JJ, et al. Educational differences in smoking: international comparison. Br Med J. 2000;320:1102–1107. doi: 10.1136/bmj.320.7242.1102. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Giskes K, Kunst AE, Benach J, et al. Trends in smoking behaviour between 1985 and 2000 in nine European countries by education. J Epidemiol Community Health. 2005;59:395–401. doi: 10.1136/jech.2004.025684. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Siahpush M, Heller G, Singh G. Lower levels of occupation, income and education are strongly associated with a longer smoking duration: multivariate results from the 2001 Australian National Drug Strategy Survey. Public Health. 2005;119:1105–1110. doi: 10.1016/j.puhe.2005.03.004. [DOI] [PubMed] [Google Scholar]
  • 37.Sorensen G, Gupta PC, Pednekar MS. Social disparities in tobacco use in Mumbai, India: the roles of occupation, education, and gender. Am J Public Health. 2005;95:1003–1008. doi: 10.2105/AJPH.2004.045039. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Pampel FC. Patterns of tobacco use in the early epidemic stages: Malawi and Zambia, 2000–2002. Am J Public Health. 2005;95:1009–1015. doi: 10.2105/AJPH.2004.056895. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Chen X, Li X, Stanton B, et al. Cigarette smoking among rural-to-urban migrants in Beijing, China. Prev Med. 2004;39:666–673. doi: 10.1016/j.ypmed.2004.02.033. [DOI] [PubMed] [Google Scholar]
  • 40.Breslau N, Johnson EO, Hiripi E, Kessler R. Nicotine dependence in the United States: prevalence, trends, and smoking persistence. Arch Gen Psychiatry. 2001;58:810–816. doi: 10.1001/archpsyc.58.9.810. [DOI] [PubMed] [Google Scholar]
  • 41.Colby SM, Tiffany ST, Shiffman S, Niaura RS. Measuring nicotine dependence among youth: a review of available approaches and instruments. Drug Alcohol Depend. 2000;59:23–39. doi: 10.1016/s0376-8716(99)00163-5. [DOI] [PubMed] [Google Scholar]
  • 42.Moolchan ET, Radzius A, Epstein DH, et al. The Fagerstrom test for nicotine dependence and the diagnostic interview schedule: do they diagnose the same smokers? Addict Behav. 2002;27:101–113. doi: 10.1016/s0306-4603(00)00171-4. [DOI] [PubMed] [Google Scholar]
  • 43.Mayhew KP, Flay BR, Mott JA. Stages in the development of adolescent smoking. Drug Alcohol Depend. 2000;59(Suppl 1):S61–S81. doi: 10.1016/s0376-8716(99)00165-9. [DOI] [PubMed] [Google Scholar]
  • 44.Hennrikus DJ, Jeffery RW, Lando HA. The smoking cessation process: longitudinal observations in a working population. Prev Med. 1995;24:235–244. doi: 10.1006/pmed.1995.1039. [DOI] [PubMed] [Google Scholar]
  • 45.Vartiainen E, Seppala T, Lillsunde P, Puska P. Validation of self reported smoking by serum cotinine measurement in a community-based study. J Epidemiol Community Health. 2002;56:167–170. doi: 10.1136/jech.56.3.167. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Patrick DL, Cheadle A, Thompson DC, Diehr P, Koepsell T, Kinne S. The validity of self-reported smoking: a review and meta-analysis. Am J Public Health. 1994;84:1086–1093. doi: 10.2105/ajph.84.7.1086. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Graham H, Owen L. Are there socioeconomic differentials in under-reporting of smoking in pregnancy? Tob Control. 2003;12:434. doi: 10.1136/tc.12.4.434. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Persson PG, Norell SE. Retrospective versus original information on cigarette smoking. Implications for epidemiologic studies. Am J Epidemiol. 1989;130:705–712. doi: 10.1093/oxfordjournals.aje.a115392. [DOI] [PubMed] [Google Scholar]
  • 49.Krall EA, Valadian I, Dwyer JT, Gardner J. Accuracy of recalled smoking data. Am J Public Health. 1989;79:200–202. doi: 10.2105/ajph.79.2.200. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Huerta M, Chodick G, Balicer RD, Davidovitch N, Grotto I. Reliability of self-reported smoking history and age at initial tobacco use. Prev Med. 2005;41:646–650. doi: 10.1016/j.ypmed.2005.01.011. [DOI] [PubMed] [Google Scholar]
  • 51.Johnson TP, Mott JA. The reliability of self-reported age of onset of tobacco, alcohol and illicit drug use. Addiction. 2001;96:1187–1198. doi: 10.1046/j.1360-0443.2001.968118711.x. [DOI] [PubMed] [Google Scholar]
  • 52.Engels RC, Knibbe RA, Drop MJ. Inconsistencies in adolescents’ self-reports of initiation of alcohol and tobacco use. Addict Behav. 1997;22:613–623. doi: 10.1016/s0306-4603(96)00067-6. [DOI] [PubMed] [Google Scholar]
  • 53.Sander W. Schooling and smoking. Econ Educ Rev. 1995;14:23–33. [Google Scholar]
  • 54.Cowell AJ. The relationship between education and health behavior: some empirical evidence. Health Econ. 2006;15:125–146. doi: 10.1002/hec.1019. [DOI] [PubMed] [Google Scholar]
  • 55.de Walque D. Does education affect smoking behaviors? Evidence using the Vietnam draft as an instrument for college education. J Health Econ. 2007;26:877–895. doi: 10.1016/j.jhealeco.2006.12.005. [DOI] [PubMed] [Google Scholar]
  • 56.Power B, Neilson S, Perry IJ. Perception of the risks of smoking in the general population and among general practitioners in Ireland. Ir J Med Sci. 2004;173:141–214. doi: 10.1007/BF03167928. [DOI] [PubMed] [Google Scholar]
  • 57.Fernandez E, Schiaffino A, Borrell C, et al. Social class, education, and smoking cessation: long-term follow-up of patients treated at a smoking cessation unit. Nicotine Tob Res. 2006;8:29–36. doi: 10.1080/14622200500264432. [DOI] [PubMed] [Google Scholar]
  • 58.Wills TA, Sandy JM, Yaeger AM. Stress and smoking in adolescence: a test of directional hypotheses. Health Psychol. 2002;21:122–130. [PubMed] [Google Scholar]
  • 59.Novak SP, Clayton RR. The influence of school environment and self-regulation on transitions between stages of cigarette smoking: a multilevel analysis. Health Psychol. 2001;20:196–207. [PubMed] [Google Scholar]
  • 60.Diez Roux AV, Merkin SS, Hannan P, Jacobs DR, Kiefe CI. Area characteristics, individual-level socioeconomic indicators, and smoking in young adults: the coronary artery disease risk development in young adults study. Am J Epidemiol. 2003;157:315–326. doi: 10.1093/aje/kwf207. [DOI] [PubMed] [Google Scholar]
  • 61.de Vries H, Candel M, Engels R, Mercken L. Challenges to the peer influence paradigm: results for 12–13 year olds from six European countries from the European Smoking Prevention Framework Approach study. Tob Control. 2006;15:83–89. doi: 10.1136/tc.2003.007237. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62.Kumar R, O’Malley PM, Johnston LD. School tobacco control policies related to students’ smoking and attitudes toward smoking: national survey results, 1999–2000. Health Educ Behav. 2005;32:780–794. doi: 10.1177/1090198105277451. [DOI] [PubMed] [Google Scholar]
  • 63.Cook PJ, Hutchinson R. Smoke Signals: Adolescent Smoking and School Continuation (Working Paper No 12472) Cambridge, MA: National Bureau of Economic Research; 2006. [Google Scholar]
  • 64.Fersterer J, Winter-Ebmer R. Smoking, discount rates, and returns to education. Econ Educ Rev. 2003;22:561–566. [Google Scholar]
  • 65.van Oort FV, van Lenthe FJ, Mackenbach JP. Material, psychosocial, and behavioural factors in the explanation of educational inequalities in mortality in The Netherlands. J Epidemiol Community Health. 2005;59:214–220. doi: 10.1136/jech.2003.016493. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66.Blakely T, Wilson N. The contribution of smoking to inequalities in mortality by education varies over time and by sex: two national cohort studies, 1981–84 and 1996–99. Int J Epidemiol. 2005;34:1054–1062. doi: 10.1093/ije/dyi172. [DOI] [PubMed] [Google Scholar]
  • 67.Emery S, Gilpin EA, Ake C, Farkas AJ, Pierce JP. Characterizing and identifying “hard-core” smokers: implications for further reducing smoking prevalence. Am J Public Health. 2000;90:387–394. doi: 10.2105/ajph.90.3.387. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68.Chou CP, Montgomery S, Pentz MA, et al. Effects of a community-based prevention program on decreasing drug use in high-risk adolescents. Am J Public Health. 1998;88:944–948. doi: 10.2105/ajph.88.6.944. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 69.Sorensen G, Glasgow RE, Topor M, Corbett K. Worksite characteristics and changes in worksite tobacco-control initiatives. Results from the COMMIT study. J Occup Environ Med. 1997;39:520–526. doi: 10.1097/00043764-199706000-00006. [DOI] [PubMed] [Google Scholar]
  • 70.Tengs TO, Osgood ND, Chen LL. The cost-effectiveness of intensive national school-based anti-tobacco education: results from the tobacco policy model. Prev Med. 2001;33:558–570. doi: 10.1006/pmed.2001.0922. [DOI] [PubMed] [Google Scholar]

RESOURCES