Correspondence to: Robert Scragg, Epidemiology & Biostatistics, School of Population Health, University of Auckland, Private Bag, Auckland, New Zealand. Fax: 09 3737 503; e-mail: firstname.lastname@example.org
Aims: To assess whether low attachment to parents is a consistent risk factor for adolescent smoking or is modified by ethnicity and parental smoking.
Methods: A national, cross-sectional multi-ethnic 2005 survey of 28,395 Year 10 students aged 14 and 15 years (18,934 Europeans, 4,769 Maori, 2,795 Asians, 1,897 Pacific Islanders). Students answered an anonymous questionnaire including measures of attachment to parents from the Inventory of Parent and Peer Attachment.
Results: The prevalence of adolescent smoking > monthly, going from the lowest to the highest quartile of parent attachment score, was 28%, 16%, 12% and 10%. The relative risk of smoking ≥ monthly for students in parent attachment quartile 1, compared to quartile 4, was increased in all ethnic groups, being 4.37 (95% confidence interval 3.00, 6.38) in Asian, 3.12 (2.77, 3.51) in European, 1.97 (1.45, 2.67) in Pacific Island, and 1.36 (1.23, 1.51) in Maori students, adjusting for sex and school socioeconomic decile. The ethnic variation in relative risk was explained mostly by a variable smoking prevalence in the reference category (quartile 4) which ranged from 3% in Asian students to 29% in Maori. A similar pattern was seen when students were categorised by parental smoking, with students in the lowest parent attachment quartile having a significantly (p<0.05) raised relative risk of smoking ≥ monthly compared to the highest quartile, regardless of whether their parents smoked or not.
Conclusions & Implications: Low parent attachment score is associated with an increased risk of adolescent smoking regardless of ethnicity and parental smoking.
In recent years awareness has increased about the potentially large influence that parents may have in determining whether or not their children smoke tobacco. A large body of evidence shows that adolescents with parents who smoke tobacco also have an increased risk of smoking.1 Personal example is the most obvious mechanism, but other mechanisms also appear to operate. Compared with non-smoking parents, parents who smoke cigarettes are more likely to allow smoking inside their home and provide their children with more pocket money.2 Both of these are risk factors for adolescent smoking.2–4 Adolescent smokers with smoking parents are more likely to get their cigarettes from a family member than those without a parent who smokes.2 While the effect of parental smoking on adolescent smoking is only moderate, in New Zealand a high proportion of adolescents are exposed to parental smoking so that, collectively, these parental factors explain about 60% of adolescents who are daily smokers, similar to the proportion explained by peer smoking.2
Other parental factors, not directly related to parental smoking, may also increase the risk of adolescent smoking. There is extensive international literature on how the quality of parent-child relationships influences the risk of adolescent smoking, underpinned by various models of adolescent learning and behaviour including social learning theory which proposes that behavior is learned observationally through modeling,5 problem behaviour theory which adds that behaviour is the result of person-environment interaction particularly to normative proscriptions and prescriptions6 and social attachment theory which proposes that attachment with parents in early life influences behaviour in adolescence and adulthood.7 Parent connectedness, attachment and support are associated with a decreased risk of adolescent smoking.8–13 Conflict between parents and their child increases the risk of adolescent smoking,14,15 whereas increased monitoring of children and authoritative parenting are associated with a decreased risk of adolescent smoking.14,16–18
Despite this large body of evidence, it is still unclear whether or not the effect of parental factors on adolescent smoking varies between sub-groups. With regard to ethnicity, for example, a national US multi-ethnic survey reported that positive parenting was associated with reduced smoking only in African-American adolescents.19 A Californian study found that parental communication was associated with decreased smoking prevalence among Latino youth, but not other ethnicities, whereas parental monitoring was associated with decreased smoking prevalence among all ethnic groups, except Asian youth.20
Parental smoking may also modify the effect of parent-child relationships on youth smoking. In a US study of 428 adolescents, low parent-child connectedness was associated with increased risk of smoking among youth with non-smoking parents, but not among those with parents who smoked.21 A Dutch study of about 800 adolescents found that the association between adolescent smoking and anti-smoking socialisation by parents (such as, setting rules about not smoking and warning children about the negative consequences of smoking) was not modified by parental smoking.22 However, a major limitation of these two studies is the relatively small size of their comparison sub-groups (less than 200) which reduced the power to detect interactions.
There do not appear to be any previous studies on parent attachment and adolescent smoking from Australia and New Zealand. We report results from a large New Zealand national survey, which collected information on parent attachment and adolescent smoking from a sample of over 30,000 Year 10 students. The large sample size, several times larger than previous studies, provided a unique opportunity to determine whether the effect of perceived attachment to parents was modified by other factors, such as parental smoking and ethnicity. Further, the sample had a 10-fold variation in smoking prevalence between its main ethnic subgroups — European, Polynesian and Asian — so the results found may apply to other communities across a wide variation in adolescent smoking prevalence. Understanding the contribution to differences in smoking prevalence by ethnicity in New Zealand is important, as smoking accounts for up to 10% of the 10-year gap in life expectancy between Maori and non-Maori.23
Participants were recruited for a national survey of Year 10 students in 2005, using similar methods to those used in previous surveys.2 A letter to the principals of all New Zealand schools with Year 10 students (n = 482) invited participation, and 278 schools agreed to participate (school response rate = 58%). The Ministry of Education classification of schools by socioeconomic decile (from one (low) to 10 (high) was used to code socioeconomic status (SES).24 Consent for the survey was obtained from school principals. The Ministry of Health Auckland Ethics Committee gave permission to survey.
Students answered an anonymous two page questionnaire on demographic characteristics (age, sex and self-assigned ethnicity) and their own smoking behaviour in answer to the question: “How often do you now smoke?” (response categories: daily, weekly, monthly, less often, previously and never). In addition, students were asked whether any ‘significant others’ smoked, including their mother, father, or best friend (response categories: Yes or No to each). Students also recorded the amount of pocket money they received in a usual week. Students who smoked were asked about the following possible sources for their cigarettes: either purchased personally, or obtained from a family member, or friend or someone else.
Students also answered a shortened list of questions from the Inventory of Parent and Peer Attachment (IPPA) about perceived attachment to their parents.25,26 A total of 12 parental questions (out of the original 28) were asked, using the four questions from each of the dimensions — communication, trust and alienation — previously shown to have highest item:total score correlation.26 Responses were processed using the method described by Nada Raja and colleagues.26 In summary, answers for each question were scored using a 4-point scale: never = 1, sometimes = 2, often = 3 and always = 4. A total score of parent attachment was calculated by adding the scores for each of the dimensions of trust, communication and alienation (the latter being reverse coded).
A total of 34,038 questionnaires were returned by schools from 44,428 on school rolls (77% student response). Initial analyses were restricted to 32,566 students who were 14 and 15 years old, with known sex, ethnicity, student smoking and school SES decile after excluding the following students: age 13 years (n=184), 16 years (476), other ages (159) or unknown age (158); unknown sex (87); unknown ethnicity (152); unknown student smoking status (219); and unknown school SES decile (195). Following Ministry of Health guidelines and New Zealand Census procedures, students could choose more than one ethnic group, so a priority system was used to classify any student choosing Maori as such, then any Pacific student as such, followed by any Asian student as such. For the remainder, ‘Other’ students (n=547, mainly from the Middle East and Africa) have been combined with 21,013 European students.
All statistical analyses were made using SAS callable SUDAAN (Release 9.0.0, 2004) which corrects standard errors for any design effect from clustering of students by school. The CROSSTAB procedure was used to compare percentages between sub-groups and to calculate adjusted Mantel-Haenszel relative risks and the MULTILOG procedure to calculate adjusted odds ratios. The REGRESS procedure was used to estimate adjusted means of the continuous outcome variable (parent attachment), with independent variables such as sex, age, ethnicity and smoking status entered as categorical and the Wald F-test was used to assess dose-response. The population attributable risk was calculated by estimating the attributable proportion for the exposed cases within each exposure category using standard methods.27
Final analyses were further restricted to the 28,395 students who answered all 12 questions in the parental attachment scale so that scores for this variable were not influenced by unanswered questions. For these students, the total score for parental attachment was normally distributed, ranging from 14 to 45. Compared to students with incomplete answers (n = 4,171), students with complete answers were more likely to be female (52% v. 49%; p=0.0034), of European and Asian ethnicity (67% and 10% v. 63% and 8%, respectively; p<0.0001), attend high SES decile 9 and 10 schools (28% v. 23%; p=0.0002) and be non-smokers (50% v. 45%; p<0.0001).
The sample analysed were aged 14 years (n=14,237) and 15 years (n=14,158); 13,542 were male and 14,853 were female. Their ethnic composition was: European 18,934, Maori 4,769, Asian 2,795 and Pacific Islander 1,897.
The percentages of students who answered ‘often’ or ‘always’ to the individual parent attachment questions are shown in Table 1. Percentages differed between male and female students for most questions, with no clear pattern to the direction of the difference between sexes.
Table 1. Percentage of students (completing all questions) who answered “often” or “always”.
(a) from χ2 test comparing male and female students
My parents respect my feelings
When I'm angry about something my parents try to be understanding
I wish I had different parents
My parents accept me as I am
I tell my parents about my problems and troubles
My parents help me understand myself better
If my parents know something is bothering me, they ask me about it
My parents have their own problems, so I don't bother them with mine
I don't get much attention at home
I get easily upset at home
Talking over my problems with my parents makes me feel ashamed & foolish
I feel angry with my parents
Table 2 shows mean parent attachment score by smoking status of students, parents and best friend, adjusted for demographic variables. Mean parent attachment score was highest in never smokers, significantly lower in students who previously smoked or smoked less than monthly, and lower again for students who smoked monthly, weekly or daily, with the latter three groups having scores that were nearly three units (8-9%) lower than never smokers. Students who had a best friend who smoked had a significantly lower parent attachment score than students whose best friend did not smoke, after adjusting for demographic variables.
Table 2. Mean (se) and mean difference in parent attachment score for categories of tobacco smoking by adolescents, parents and best friend, adjusted for sex, age, ethnicity and school SES decile.
Mean difference (se)
p-value for Wald F
p-value for Wald F
Best friend smokes
p-value for Wald F
Parental smoking was also associated with small differences in parent attachment score, which was lowest in students with both parents smoking, intermediate in students with either their mother or father smoking and highest in students with neither parent smoking (Table 2). Of interest, the mean score for students exposed to maternal only smoking was similar to that of paternal only smoking (mean scores 31.7 and 31.8 respectively).
Mean parent attachment score showed an inverted U-shape association with the amount of pocket money students received each week. The mean score was lowest in students who received no pocket money (mean 31.7), significantly (p<0.01) higher in students receiving $1-5 (32.4), $6-10 (32.4), $11-20 (32.2), $21-30 (32.1), $31-40 (32.4) and $41-50 (32.2), before reducing slightly in students receiving >$50 (31.9; p=0.34 compared with students receiving no pocket money).
Among students who smoked, those who purchased their cigarettes themselves had a significantly lower parent attachment score than those who did not (mean (se) = 30.1 (0.1) v. 30.4 (0.1); p=0.0086), but the mean score was not related to getting cigarettes from a family member, or from a friend or someone else (data not shown).
The percentage of students smoking, by quartile of parent attachment score, is shown in Table 3. Based on the mean parental adjustment scores for the student smoking categories in Table 2, students smoking daily, weekly or monthly were combined because of their similar mean scores, as were students smoking less often and those who had smoked previously (not in the last month). Students in the lowest parent attachment quartile had the highest prevalence of smoking monthly or more often (28%), more than twice that of students in the highest quartile (10%); while the pattern for the prevalence of never smoking was reversed (34% v. 62%). The variation in the prevalences across parent attachment quartiles was greater for regular smoking (monthly or more often) than for infrequent smoking (less than monthly), indicating that a low parent attachment score was more strongly associated with regular student smoking. However, the significantly increased relative risks of smoking less than monthly for quartiles one to three, compared with quartile four (parent attachment score >35), indicate that low parent attachment scores are also associated with infrequent student smoking patterns typical of experimentation (Figure 1). Logistic regression analyses showed that the addition of pocket money and best friend smoking to a model with socio-demographic variables did not change greatly the odds ratios of student smoking associated with parent attachment (data not shown).
Table 3. Percent (se) of student smoking, by quartile of parent attachment score.
Parent attachment score (quartile)
Smoke < monthly
Smoke > monthly
Row percent (se)
≥ 28 (1)
≥ 36 (4)
Table 4 shows the percentage of students who smoked monthly or more often, by quartile of parent attachment and ethnicity. Within each ethnic group, the smoking prevalence was lowest for students in the highest parent attachment quartile four and progressively increased in the lower quartiles. The largest difference in smoking prevalence occurred between quartiles one (score ≤ 28) and quartile two (score 29-32), being higher in quartile one by 10% for Asian students and 12% for all other ethnicities. The relative risk of smoking was significantly (p<0.05) higher in quartile one compared to quartile four in each ethnic group (Table 4), indicating that a low parent attachment score is associated with increased smoking prevalence regardless of ethnicity. The ethnic variation in relative risk is explained by a similar excess smoking prevalence between quartiles one and four (ranging from 12% in Maori up to 17% in European students), relative to a more variable prevalence in the reference category (quartile 4) which ranged from 3% in Asian students up to 29% in Maori.
Table 4. Prevalence and relative risk of student smoking monthly or more often associated with quartile of parent attachment score, by ethnicity.
Parent attachment score (quartile)
(a) Adjusted for sex and school SES decile
Prevalence: % (se)
≥ 28 (1)
≥ 36 (4)
Relative Risk (95% CI)a
≥ 28 (1)
≥ 36 (4)
Table 5 shows the percentage of students who smoked monthly or more often, by quartile of parent attachment and number of parents who smoked. Students in the highest parent attachment quartile without smoking parents had the lowest smoking prevalence (5%). Within each parental smoking category, student smoking prevalence was lowest in parent attachment quartile four (score ≥ 36) and highest in quartile one (score ≤ 28). The excess prevalence of student smoking between the highest parent attachment quartile (score ≥ 36) to the lowest quartile (score ≤ 28) was similar for the four parental smoking categories (ranging from 13% to 17%). Comparing parental smoking groups within each parent attachment quartile, student smoking prevalence was highest in those with both parents smoking, followed by students with their mother only smoking, then students with their father smoking only and lowest in students with neither parent smoking.
Table 5. Prevalence and relative risk of student smoking monthly or more often associated with quartile of parent attachment score, by parental smoking.
Parent attachment score (quartile)
Number of Smoking Parents
(a) Adjusted for sex, ethnicity and school SES decile
Prevalence: % (se)
≥ 28 (1)
≥ 36 (4)
Relative Risk (95% CI)a
≥ 28 (1)
≥ 36 (4)
Parental smoking and parent attachment had an additive effect on the excess risk in student smoking. When compared with students in the highest parent attachment quartile with neither parent smoking, the absolute increase in student smoking prevalence among students on the lowest parent attachment quartile who had both parents smoking (48% - 5%= 43%) was similar to the sum of the absolute increase between parent attachment quartiles one and four among those with neither parent smoking (20% - 5%= 15%) plus the absolute increase between students with both and neither parent smoking among students in the highest parent attachment quartile (31% - 5%= 26%).
Within each parental smoking category, the relative risk was still significant (p<0.05) when comparing the lowest and highest parent attachment quartiles (Table 5), indicating that low parent attachment score is associated with increased student smoking prevalence, regardless of parental smoking. However, for students with one or both parents smoking, the increased student smoking prevalence was only significant (p<0.05) among students in the lowest parent attachment quartile, when compared with quartile four. Because the absolute excess smoking prevalence between parent attachment quartiles one and four did not vary greatly between parental smoking categories (Table 5), the change in relative risk across parent attachment quartiles was greatest among students with neither parent smoking because of their lower smoking prevalence and smallest among students with both parents smoking because they had the highest smoking prevalence.
Thirty-two per cent of all students smoking monthly or more often could be attributed to parent attachment scores in the lowest three quartiles, based on the adjusted relative risks of smoking in Table 3 for parent attachment quartiles 1-3, with most of this proportion (22% of all student smokers) being for students in the lowest quartile.
We have shown in a large national survey that low parent attachment scores were associated with an increased risk of adolescent smoking. This association was stronger for regular smoking (monthly or more often) than for less frequent smoking patterns typical of experimentation (Table 3, Figure 1). The stronger association for regular smokers, who are further along the smoking trajectory than experimenters, suggests that the adverse effects from low parent attachment on smoking initiation occurred earlier in the lives of participating students, who were 14 and 15 years at interview. Although there was evidence of a dose-response in the effect of parent attachment on the relative risk of smoking monthly or more often (Table 3), the greatest increase in smoking prevalence occurred in the lowest parent attachment quartile (score ≤ 28), particularly in Maori and Pacific Island students, indicating there maybe a threshold-effect for these students (Table 4).
Low parent attachments were associated with an increased risk of adolescent smoking in all ethnic groups and was unaffected by parental smoking (Tables 4 & 5). This finding is in agreement with a previous study which found that the association between anti-smoking socialisation and adolescent smoking was not moderated by parental smoking,22 although it contrasts with three other studies that found associations in some ethnic groups but not others19,20 and only in students with non-smoking parents.21 The relatively small sample size in those studies, with limited statistical power to detect associations, is a likely explanation and not the differences in ethnicity since they had similar ethnic groups to our study, with the inclusion of whites and Asians. The much larger sample size of the current study allowed detection of associations between the parent attachment score and adolescent smoking in all ethnic subgroups. The variable relative risks associated with low parent attachment are due to variations in smoking prevalence between subgroups. Those with a high smoking prevalence in their reference category (e.g. Maori students) had much lower relative risks because the excess absolute smoking prevalence for the lowest parent attachment quartile (score ≤ 28) was reasonably constant, ranging from 12-17% in Table 4, much less than the range of 3-29% in the reference group (score ≥ 36). The same pattern occurred with parental smoking in Table 5, with relative risks being lowest for students with both parents being smokers.
Mean parent attachment score had an inverted U-shape association with the weekly amount of pocket money students received, being lowest at the extremes of the pocket money distribution. This indicates that the relationship between students and their parents is not related linearly to the amount of pocket money. However, students who bought their own cigarettes had a lower mean parent attachment score than other smokers who did not. Lower attachment between the former students and their parents may be one factor, among many, driving their purchasing of cigarettes.
Similarly, the lower mean attachment score for students with a best friend who smoked (Table 2) may reflect a greater attachment with their peers (at the expense of parent attachment), since high connectedness to peers is associated with increased risk of smoking.12 However, peer attachment was not measured for logistical reasons and is one of the limitations of this study. Other limitations include that it is cross-sectional and lacks temporal separation between factors, so cannot distinguish whether or not low parent attachment is either a cause or a consequence of tobacco smoking; that the instrument used to measure parent attachment has not been validated for non-European New Zealand children and that other dimensions of the parent-child relationship, which may also be risk factors for adolescent smoking, such as types of parenting,14,16–18 were not measured. In this regard, a recent report has found that positive family relationships are associated with reduced risk of smoking by Asian adolescents in New Zealand.28
Despite these possible limitations, our findings potentially have important implications for public health strategies to prevent adolescent smoking. The proportion of students smoking monthly or more often that is attributable to low parent attachment score — 32% for parent attachment quartiles 1-3, 22% for quartile one by itself — indicates that parent attachment potentially explains a substantial proportion of the variation in student smoking. The consistent inverse association between parent attachment score and risk of adolescent smoking in all ethnic groups and in all parental smoking categories strengthens the conclusion that low parent attachment is a likely risk factor for adolescent smoking. Overall, these findings suggest that prevention strategies, targeted at improving attachment between parents and their children, may help reduce the uptake of smoking by adolescents; along with further efforts to control parental smoking through cessation programs given our results showing that attachment to parents and smoking by parents are independently associated with adolescent smoking.
The survey was carried out by Action on Smoking and Health (ASH). The New Zealand Ministry of Health provided funds. Dr Reeder receives support from the Cancer Society of New Zealand Inc. and the University of Otago.