Objectives. We investigated the relationship among father’s occupational group, daily smoking, and smoking determinants in a cohort of New Zealand adolescents.
Methods. The longitudinal Multidisciplinary Health and Development Study provided information on adolescents’ self-reported smoking behavior and potential predictors of smoking, such as social and material factors, personality characteristics, educational achievement, and individual attitudes and beliefs regarding smoking. Longitudinal logistic generalized estimating equation analyses were used.
Results. Adolescents whose fathers were classified in the lowest-status occupational group were twice as likely as those whose fathers occupied the highest-status occupational group to be daily smokers. This high risk of daily smoking among the adolescents from the lowest occupational group was largely predicted by their lower intelligence scores and by the higher prevalence of smoking among fathers and friends .
Conclusions. To prevent socioeconomic differences in smoking, school-based interventions should seek to prevent smoking uptake among adolescents, particularly those of lower socioeconomic status. Programs need to provide positive, nonsmoking role models consonant with the culture and norms of lower-socioeconomic-status groups. Adolescents need to acquire resistance skills and protective behaviors against social pressure and influences.
Attempts to describe and explain socioeconomic differences in unhealthy behavior have mainly focused on adults. But lifestyle patterns observed among adults are largely developed during adolescence and are perpetuated into adulthood. Little is known about the emergence of socioeconomic differences in unhealthy lifestyles during adolescence and even less about the determinants of this process. Such information would greatly facilitate the design of effective interventions to prevent the development of socioeconomic differences in behavior at an early stage.
The objective of our study was to examine patterns and predictors of socioeconomic differences in adolescent smoking behavior. The longitudinal Dunedin Multidisciplinary Health and Development Study followed a birth cohort of approximately 1000 individuals during their entire adolescence and thus provides a unique opportunity to describe and explain the relationship between father’s occupational group and adolescents’ daily smoking. This article is among the first to report on the contribution by a variety of determinants of adolescent smoking to the association between father’s occupational group and daily smoking among adolescents.
A review of the literature shows that adolescents of lower socioeconomic status (SES) smoke more often than do their peers of higher SES,1–13 though some studies fail to find such a relationship.14–16 This hypothetical relationship between parental SES and adolescent smoking might originate from a higher prevalence of determinants of adolescent smoking among lower-SES groups. To date, not many predictors of adolescent smoking behavior have been investigated regarding their association with SES, and only 2 other studies have analyzed the contribution of such predictors to socioeconomic differences in smoking behavior.4,12
The adolescent smoking literature emphasizes the effect of modeling behavior of parents and peers. Children who have smoking parents (or who live with people who smoke)4,5,8,10,12,14,15,17–27 or who have friends who smoke1,2,6,8,12,18–24,28 are more inclined to start smoking during adolescence, though some studies have not found such a relationship.6,7,28,29 Other social factors that are reported to predict smoking during adolescence are perceived approval or pressure to smoke,2,7,18,22,27,30–32 poor family support or control,6,12,14,22,24,26,33 poor social bonding,2 and high involvement in social activities.30
Seltzer and Oechsli9 addressed the predictive potential of personality traits with regard to adolescent smoking and reported that children with type-A personality traits, extraversion, and psychoticism are more likely to begin smoking during adolescence than are children without these characteristics. (A person that scores high on psychoticism will exhibit some qualities commonly found among psychotic persons. Examples of psychotic tendencies include recklessness, disregard for common sense, and inappropriate emotional expression.) External locus of control,11,18,34 low self-esteem,1,2,11,22,23,26,35 and deviant or risky behavior2,18,22,23,25,26,35 are also related to adolescent smoking. Individual positive attitudes and beliefs related to smoking2,6,17,18,27,28,30,36 predict adolescent smoking, though McNeill et al7 found no such a relationship.
Educational achievement also plays a role in adolescent smoking. Poorer school achievement,4,19,23 negative attitudes toward or poor adjustment in school,17,22 low academic expectations,2,17,18,24,26 and average or below-average school performance1,17,20,22,25 all predict smoking during adolescence. (Murray and colleagues,30 however, found no relationship between attitude toward school or truancy and smoking during adolescence.)
In general, material factors are considered important determinants of socioeconomic differences in health or health-related behavior.37–39 Smoking behavior during adolescence is predicted by the availability of money.2,7
Although the relationship between predictors of adolescents’ smoking behavior and SES has not been the subject of many studies, some of the above-mentioned predictors are reported to be more prevalent in lower SES groups. Adolescents of lower SES are more likely than those of higher SES to have smoking parents, friends, peers, and siblings10,12,23; they also experience more social pressure to smoke and positive norms involving smoking,10 and they more often report an external locus of control,11–12 lower self-esteem,12 and poorer academic achievement.12 We hypothesized that these determinants contribute to socioeconomic differences in adolescent smoking.
Data were obtained from the Dunedin Multidisciplinary Health and Development Study, which is a longitudinal investigation of the health, development, and behavior of a cohort of children from birth until adulthood.40 In brief, the sample consists of a cohort born in Dunedin’s only obstetric hospital between April 1, 1972, and March 31, 1973. The perinatal histories were documented soon after birth, but study participants were first enrolled in the longitudinal study at 3 years of age. Ninety-one percent of eligible births (i.e., of people still resident in the province of Otago) participated in the first assessment, providing a baseline sample of 1037 participants. Study participants were assessed every 2 years up to age 15 and then at ages 18, 21, and 26. Most participants were assessed within 2 months of their birthdays. Transportation to the research unit was provided for participants living in New Zealand (but outside Dunedin). During the assessment that was carried out when the study members were 21 years old, an interviewer traveled to those living overseas (almost all in Australia). This procedure resulted in very high follow-up rates: between 90% and 97% of the study participants included in the baseline sample, with a onetime low follow-up rate of 82% at age 13 years.40 Before the interviews, informed consent was obtained from a parent (for interviews of participants younger than 18 years) and from the participant (for interviews of participants aged 15 and older).
The sample was representative of the population of New Zealand’s South Island in terms of parent’s SES, children’s educational achievement, and ethnicity (primarily European).40
The questions on smoking behavior were first included in the study when study participants reached 9 years of age. When participants were 9, 11, and 13 years old, the interviews regarding smoking were conducted by the same trained interviewer. The interviews were conducted in private at the research unit as part of the series of assessments of health, development, and behavior during 1 day. A small portion of the sample was unable to attend the research unit for assessment at ages 9, 11, and 13 years; these participants were assessed at home or at school and were not asked about smoking. At ages 15, 18, and 21 years, smoking questions were included in the home, school, or workplace interviews. Most of these interviews were conducted by the same interviewer.
Regular daily smoking is often used as an indicator of the development of habitual smoking. In this sample, a comparison of self-reported smoking status with saliva cotinine concentrations showed high sensitivity (96%) and specificity (82%) of self-reported data.41
Father’s occupational group was assessed at the beginning of adolescence (age 9 years) and was categorized according to the Elley-Irving classification.42 This classification was designed for use in New Zealand but is internationally comparable to other occupational classifications, because it is based on the International Standard Classification of Occupations. Average income and education levels (based on the 1981 New Zealand census for males) were used to rate fathers’ occupations.43 We selected father’s occupation to indicate adolescent’s SES, with the selection criteria set out by Liberatos and colleagues, to maximize comparability of our study with other studies.44 When studying children or adolescents, social class is typically measured according to parental occupation, usually the father or the head of the household.4,9,10,16,45
Adolescents’ health is equally related to occupation-based SES measures and to nonoccupational SES measures, e.g., father’s education, household income, neighborhood deprivation, housing tenure, and car availability. However, on balance, the occupational measures seem to be the better discriminators.45 When information on father’s occupation was missing at a particular age, information collected at later measurements was used. Because of low numbers among the 2 lowest occupational categories, the semi-skilled and unskilled groups, we combined these groups.
The Dunedin Multidisciplinary Health and Development Study measured information on several potential predictors of smoking behavior among adolescents, including social and material factors, personality characteristics, personal attitudes and beliefs regarding smoking factors, and educational achievement.
Social factors included self-reported smoking behavior of mothers and fathers as well as parents’ reports regarding family relationships.46,47 The study participants reported on smoking of members of their household, smoking of close friends, their attachment to family and friends,48 their relationship with their parents, and their participation in organized groups, clubs, or other activities. Material factors were reported by the parents and covered receipt of pocket money by children, the number of children in the family, and occurrence of father’s unemployment in the 2 years preceding the interview.
Personality characteristics were measured with Rosenberg’s questions regarding self-esteem,49 Rutter’s neuroticism questionnaire,50 the Multidimensional Health Locus of Control scales,51,52 and the Quay and Peterson behavioral problems checklist.53
Individual attitudes and beliefs included attitudes toward smoking friends or adults, as well as toward smoking in general, beliefs that smoking is as bad for one’s health as people say and that smoking will affect health when one is older, and the number of reasons to smoke that could be enumerated by study participants.
Educational achievement was indicated by performance at school and the Wechsler intelligence score.54
To identify risk groups, we divided all continuous-scale variables into tertiles (3 equally sized groups) or into 2 groups, comparing the top or bottom quartile with the rest of the study population.
Analyses were conducted in 4 stages. At the first stage, we examined the relationship between father’s occupational group and daily smoking. We calculated the prevalence of daily smoking by SES for each of the assessments when participants were 9, 11, 13, 15, 18, and 21 years old. Because none of the participants smoked daily at ages 9 or 11, we omitted these age groups from further longitudinal analyses. We then fitted logistic regression models, adjusted for gender, for each measurement wave separately, with the highest-status occupational group as a reference category. Next, we fitted a longitudinal logistic generalized estimating equation (GEE) model accounting for the dependence between repeated measurements within the same individual, with the GENMOD procedure of SAS 8.0.55 We calculated occupational differences in daily smoking for the period between 13 years and 21 years by fitting a GEE model including gender, time, and father’s occupational group. This longitudinal GEE analysis used information on 947 study participants, of whom 50.8% were boys and 49.2% were girls. The magnitude of socioeconomic differences in daily smoking did not differ between boys and girls; that is, there was no significant interaction between father’s occupational group and the child’s gender.
At the second stage, we examined which variables longitudinally predicted daily smoking for the period between 13 years and 21 years by fitting GEE models including gender, time, and 1 potential determinant successively. Variables were considered to be predictors of daily smoking when the GEE analyses showed a significant likelihood ratio χ2 test (P < .05) and significantly increased odds ratios (P < .05).
At the third stage, for those predictors that showed significantly increased odds of daily smoking, we examined the distribution of the predictor among occupational groups.
Finally, at stage 4, we added significant predictors of daily smoking occurring more often among adolescents of lower SES to the first GEE model (including gender, time, and occupation) in an attempt to explain the relationship between father’s occupational group and adolescents’ daily smoking. The contribution of this predictor to the explanation of differences in smoking by SES was expressed by the percentage reduction in odds ratios of the different occupational groups owing to the inclusion of the predictor (all significantly increased odds ratios of occupation should decrease in value).
In this New Zealand cohort of adolescents born in 1972, daily smoking was observed to begin at 13 years (Table 1). From that time, the smoking prevalence increased drastically with each measurement wave, reaching adult levels of smoking prevalence at 18 years, when about one third of the respondents reported being daily smokers. Differences in daily smoking by father’s occupational group emerged at 15 years, at which age smoking prevalence clearly decreased with an increase in father’s occupational group (Table 1). Relative differences in daily smoking were statistically significant from age 15 onward, although only for the 2 lowest-status occupational groups (Table 1).
Longitudinal GEE analyses, which take into account all repeated measurements from the entire adolescent period beginning with 13 years, showed that adolescents in the 2 lowest-status occupational groups were twice as likely as those in the highest-status occupational group to be daily smokers, whereas adolescents in the second-lowest-status occupational group had odds of daily smoking more than 1½ times higher (Table 1). Longitudinal GEE analyses further revealed that occupational differences in adolescent daily smoking were stable during the entire period studied (P value = .4352 for occupation × time interaction).
Table 2 shows the potential predictors measured at baseline that predicted daily smoking during adolescence (13 years through 21 years). We found that several social factors—having a smoking (or formerly smoking) father or smoking friends, living with smokers, not belonging to any group or organization, having poor family relationships, and having low attachment to parents—significantly predicted daily smoking during adolescence (Table 2). The only material factor that significantly increased the odds of being a daily smoker was receiving pocket money (Table 2). Adolescents who reported behavioral problems showed significantly increased odds of daily smoking (Table 2). Furthermore, adolescents who had positive attitudes toward smoking friends or adults, who did not believe in the detrimental effects of smoking, who reported a higher number of reasons to smoke were significantly more likely to smoke daily (Table 2). Compared with adolescents who strongly agreed with the statements about the harmfulness of smoking, those who just agreed had higher odds of smoking daily (Table 2). Low and medium intelligence scores and average or below-average school performance significantly predicted daily smoking (Table 2).
Table 3 shows the relationship between father’s occupational group and statistically significant risk categories of predictors of daily smoking. A significant inverse relationship with father’s occupational group was observed for having a smoking father or friend, poor family relationships, behavioral problems, average or below-average school performance, or low intelligence scores. Also, number of reasons to smoke was significantly related to father’s occupational group; unexpectedly, adolescents of higher SES reported higher numbers of reasons to smoke (Table 3).
We also observed that some risk factors were clearly more prevalent in the lowest occupational group: living with smokers, not belonging to an organization, having positive attitudes toward smoking friends, and not believing in the adverse health effects of smoking.
We tested the explanatory potential of all predictors of daily smoking occurring significantly more often among groups of adolescents whose fathers reported lower-status occupational group (we excluded attachment to parents, receipt of pocket money, and reasons to smoke from stage 4 of the analyses). We found that in this New Zealand population, intelligence scores and the smoking behavior of father and friends explained the relationship between father’s occupational group and daily smoking during adolescence: together these predictors reduced the significantly increased odds ratios of skilled, semiskilled, and unskilled occupational groups to nonsignificant levels. The unequal distribution of intelligence scores across occupational groups contributed most to explaining the association between father’s occupational group and adolescent daily smoking (Table 4).
We observed a clear relationship between father’s occupational group and daily smoking during adolescence among this New Zealand cohort. Longitudinal analyses revealed that occupational differences in adolescent smoking were present and stable from the onset of daily smoking. The lower IQ scores of adolescents whose fathers reported a lower-status occupational group contributed most to the relationship between father’s occupational group and daily smoking. Also, the smoking behaviors of father and friends explained part of the differences in daily smoking among adolescents according to father’s occupational group. None of the material factors, personality characteristics, or individual beliefs and attitudes contributed to the relationship between father’s occupational group and adolescent smoking.
Some limitations of the data must be acknowledged. First, father’s occupational group might vary over time during the long study period. However, the significantly high correlation among the 4 measurements of father’s occupational group in this study—between 9 years and 15 years (r > 0.70, P = .0)—indicates that occupational group was largely stable. Second, to exclude all possible concerns about causality between predictors and smoking behavior, we chose to include factors measured at the baseline of our longitudinal analyses, that is, before or at age 13. Because we studied a relatively long time frame, it is possible that the effect of factors that affect smoking in the short term or that are likely to change is underestimated. For example, behavioral attitudes or material barriers are very likely to affect current behavior, but this effect might dissipate over time, resulting in weaker associations using longer time frames. Other factors, such as intelligence, are more likely to influence behavior throughout the adolescent period.
The most important predictor of occupational differences in daily smoking during adolescence was lower intelligence scores among children of lower SES. The rare studies that have investigated explanations for socioeconomic differences in adolescent smoking and substance use also have identified academic competence or achievement as one of the explanatory pathways.4,12 Less-intelligent adolescents might use smoking as a substitute for the satisfaction of academic success. They might also be less receptive to messages about the negative health consequences of antismoking programs.
Intelligence likely results not only from heredity but also from environmental influences.38,56–59 Reviews of long-term effects of early childhood education and day care found persistent positive effects of these interventions on academic achievement60,61 and future SES,60 as well as on IQ.58–60 Greater access to such interventions for lower-SES groups might help prevent, or at least delay, adolescent smoking.
The role of intelligence may reflect the mechanism by which socioeconomic differences in smoking among adults are established. Adolescents with lower intelligence levels are less likely to pursue higher education. In our New Zealand cohort, lower intelligence scores at age 13 were related to having a lower-status occupation at 21 years (P = .0162). Furthermore, adolescent smoking is known to lead to poor educational achievement and therefore low SES.62,63 To break the persistent cycle of socioeconomic differences in smoking, intervention programs should focus on preventing smoking uptake among adolescents, particularly those of lower SES. This prevention might be achieved by developing school-based interventions in lower-SES neighborhoods.
The contribution of the smoking behaviors of fathers and friends to the differences in adolescent smoking highlights the importance of modeling behavior during adolescence.64 The few studies that have attempted to explain socioeconomic differences in smoking among adolescents indicate an effect of parents’ and friends’ smoking behavior.4,12 Although adolescents of lower SES had more exposure to smoking role models, we found that they were not more vulnerable to the influence of these role models (nonsignificant interaction of occupation× smoking father/friends). Interventions to prevent smoking among adolescents should provide positive, nonsmoking role models who fit the culture and norms of lower-status occupational groups.31 Furthermore, adolescents can resist social pressure and influences in favor of smoking if they are taught resistance skills or protective behaviors.2 Relevant programs need to consider and effectively involve adolescents’ social environment, that is, their parents and social communities.1 This involvement accords with US guidelines for school health programs to prevent tobacco use and addiction.65 Such interventions should be targeted toward adolescents belonging to lower-SES groups, because they are disproportionately exposed to potent predictors of smoking and are thus at greater risk of becoming daily smokers.
Note. GEE = generalized estimating equation; OR = odds ratio. aOdds ratio of logistic regression analysis adjusted for gender for each measurement wave separately. bOdds ratio of longitudinal logistic GEE analysis, which included youths aged 13–21 years, adjusted for gender. cP value of likelihood ratio χ2 test of father’s occupational group. Note. OR = odds ratio. aOdds ratio of generalized estimating equation analyses, which included youths aged 13–21 years, adjusted for gender. bP value of likelihood ratio χ2 test of generalized estimating equation analysis. Values rounded to 4 decimal places. aHigher professional, administrative. bLower professional, technical. cClerical, highly skilled. dSkilled. eSemiskilled, unskilled. f P value of χ2 test. Values rounded to 3 decimal places. Note. OR = odds ratio. aOdds ratio of daily smoking during adolescence obtained by longitudinal logistic GEE analysis. bPercentage reduction in odds ratio of daily smoking owing to inclusion of predictor (OR basic model − OR [basic + predictor])/(OR basic model − 1). cBasic model for longitudinal logistic generalized estimating equation (GEE) analyses included youths aged 13–21 years, adjusted for gender. dP value of likelihood ratio χ2 test of father’s occupational group in GEE analysis.
Percentage (OR)a 9 11 13 15 18 21 ORb Father’s occupational group (distribution in population) Higher professional, administrative (14%) . . . . . . 1.1 (1.00) 8.1 (1.00) 22.3 (1.00) 29.6 (1.00) 1.00 Lower professional, technical (14%) . . . . . . 1.1 (0.98) 8.7 (1.06) 25.4 (1.17) 27.6 (0.90) 0.97 Clerical, highly skilled (26%) . . . . . . 1.5 (1.40) 14.1 (1.86) 28.9 (1.41) 31.5 (1.10) 1.24 Skilled (29%) . . . . . . 0.5 (0.42) 14.5 (1.94) 33.5 (1.75) 37.5 (1.43) 1.57 Semiskilled, unskilled (17%) . . . . . . 1.8 (1.65) 19.5 (2.86) 38.4 (2.21) 45.4 (1.99) 2.12 Overall prevalence 0 0 1.1 13.5 30.3 34.7 . . . Pc . . . . . . .8804 .0208 .0217 .0087 .0029 No. assessed 955 925 850 976 993 992 . . . No. interviewed on smoking 779 794 734 964 937 903 . . . ORa Pb Social factors Smoking behavior of mother .1332 Nonsmoker 1.00 Ex-smoker 1.15 Smoker 1.35 Smoking behavior of father .0091 Nonsmoker 1.00 Ex-smoker 1.54 Smoker 1.55 Smoking behavior in the house .0023 No 1.00 Yes 6.05 Smoking behavior of close friends ≤.0001 No one smokes 1.00 1 or more smoke(s) 3.26 Participates in organized groups, clubs, or activities .0019 Yes 1.00 No 1.74 Family relationships ≤.0001 Better relationships 1.00 Quartile with poorest relationship 1.93 Attachment to parents .0107 High 1.00 Medium 1.35 Low 1.74 Relationship with parents .0773 Okay 1.00 Not (always) okay 2.50 Attachment to friends .2274 Low 1.00 Medium 0.91 High 0.72 Individual attitudes and beliefs Attitude toward smoking friends .0070 Neutral 1.00 Prefer nonsmokers 0.65 Prefer smokers 6.54 Attitude toward smoking adults .0354 Not okay to smoke in moderation 1.00 Okay to smoke in moderation 1.43 General attitude toward smoking .4526 Lower scores 1.00 Most positive quartile 1.13 Believes smoking is as bad for your health as people say .0 Strongly agree 1.00 Agree 1.97 Disagree or strongly disagree 5.03 Believes smoking will affect health when you are older .0000 Strongly agree 1.00 Agree 1.79 Disagree or strongly disagree 12.53 Number of reasons to smoke .0002 Low 1.00 Medium 1.24 High 12.34 Educational achievement factors Performance at school .0000 Above average 1.00 Average 2.29 Below average 4.19 Intelligence ≤.0001 High 1.00 Medium 1.40 Low 3.21 Personality characteristics Self-esteem .4883 High 1.00 Medium 0.93 Low 1.25 Health locus of control .0687 Internal 1.00 Neutral 1.17 External 1.49 Neuroticism .3040 Lower scores 1.00 Highest quartile (most neurotic) 0.85 Behavioral problems ≤.0001 Low 1.00 Medium 1.54 High 2.46 Material factors Receives pocket money .0419 No 1.00 Yes 1.31 No. of children in family .5804 1 or 2 1.00 3 0.94 4 or more 1.04 Father registered as unemployed .0936 No 1.00 Yes 1.88 Father’s Occupational Group, % Lowesta Lower b Intermediatec Higher d Highest e Pf Social factors Smoking behavior of father .000 Ex-smoker 14.5 20.5 20.2 28.3 17.2 Smoker 56.5 45.4 41.2 33.6 25.9 Smoking in the house .129 Yes 11.9 6.0 6.1 4.3 7.7 Smoking behavior of close friends .002 1 or more smoke 49.5 36.0 35.1 33.0 22.2 Participates in groups, clubs, or activities .315 No 27.9 18.7 24.2 18.9 21.1 Family relationships .001 Quintile with poorest relationships 35.0 23.3 15.7 22.9 16.7 Attachment to parents .077 Low 35.5 39.7 32.3 34.4 23.6 Material factors Receives pocket money .913 Yes 55.6 56.5 59.6 59.0 59.7 Personality characteristics Behavioral problems .000 Medium 37.3 28.8 38.7 37.6 26.4 High 44.1 38.1 25.3 31.2 23.6 Individual attitudes and beliefs Attitude toward smoking friends .548 Prefer smokers 1.8 1.4 0.0 0.0 0.0 Attitude toward smoking adults .110 Okay to smoke in moderation 81.6 80.9 75.8 79.6 68.8 Believes smoking is bad for health .448 Agree 36.0 37.9 33.0 38.5 33.3 Disagree or strongly disagree 9.9 4.2 4.6 3.3 6.7 Believes smoking affects health when older .090 Agree 33.3 38.3 28.9 42.9 26.7 Disagree or strongly disagree 4.5 1.4 1.5 1.1 2.2 Number of reasons to smoke .001 High 29.7 28.5 35.6 38.3 54.4 Educational achievement factors School performance .000 Average 78.2 77.9 63.5 62.3 51.1 Below average 9.1 6.1 4.7 3.3 2.3 Intelligence .000 Low 50.8 42.1 29.8 22.9 10.9 ORa (% Reduction in OR From Basic Model)b Basic Modela,c Basic + Intelligence Basic + Smoking Father Basic + Smoking Friends Basic + All 3 Predictors Father’s occupational group (reference group: higher professional, administrative) Lower professional, technical 0.97 0.95 0.86 1.01 0.88 Clerical, highly skilled 1.24 1.05 1.08 1.00 0.71 Skilled 1.57 1.28 (51) 1.49 (14) 1.38 (33) 0.93 (100) Semiskilled, unskilled 2.12 1.88 (21) 1.93 (17) 2.07 (4) 1.28 (75) P value for occupationd .0029 .0692 .0090 .0281 .2752
Note. GEE = generalized estimating equation; OR = odds ratio.
aOdds ratio of logistic regression analysis adjusted for gender for each measurement wave separately.
bOdds ratio of longitudinal logistic GEE analysis, which included youths aged 13–21 years, adjusted for gender.
cP value of likelihood ratio χ2 test of father’s occupational group.
Note. OR = odds ratio.
aOdds ratio of generalized estimating equation analyses, which included youths aged 13–21 years, adjusted for gender.
bP value of likelihood ratio χ2 test of generalized estimating equation analysis. Values rounded to 4 decimal places.
aHigher professional, administrative.
bLower professional, technical.
cClerical, highly skilled.
f P value of χ2 test. Values rounded to 3 decimal places.
Note. OR = odds ratio.
aOdds ratio of daily smoking during adolescence obtained by longitudinal logistic GEE analysis.
bPercentage reduction in odds ratio of daily smoking owing to inclusion of predictor (OR basic model − OR [basic + predictor])/(OR basic model − 1).
cBasic model for longitudinal logistic generalized estimating equation (GEE) analyses included youths aged 13–21 years, adjusted for gender.
dP value of likelihood ratio χ2 test of father’s occupational group in GEE analysis.
The Van Walree Foundation and the Netherlands Organization for Scientific Research financially supported Mariël Droomers at the Alcohol and Public Health Research Unit, University of Auckland, New Zealand.
Droomers wishes to thank Karen Witten and Philippa Howden-Chapman for providing her the opportunity to work in New Zealand and the staff of the Alcohol and Public Health Research Unit for their moral support. Furthermore, the authors would like to thank Richie Poulton and Barry Milne for their indispensable help in accessing the Dunedin data, and Megan Pledger, Elisabeth Robinson, and Gerard Borsboom for their advice in statistical matters.
Human Participant Protection As part of the invitation to attend the research unit, full written explanations of all the procedures were provided, and the written consent of the parents and study participants was obtained.