WORKING
P A P E R
A Discrete Time Hazards
Model of Smoking
Initiation Among West
Coast Youth from Age 5
to 23
MARIA ORLANDO, JOAN S. TUCKER,
AND PHYLLIS L. ELLICKSON
WR-395
June 2006
This product is part of the RAND
Health working paper series.
RAND working papers are intended
to share researchers’ latest findings
and to solicit informal peer review.
They have been approved for
circulation by RAND Health but
have not been formally edited
or peer reviewed. Unless otherwise
indicated, working papers can be
quoted and cited without permission
of the author, provided the source is
clearly referred to as a working paper.
RAND’s publications do not necessarily
reflect the opinions of its research
clients and sponsors.
is a registered trademark.
1
A Discrete Time Hazards Model of Smoking Initiation
Among West Coast Youth from Age 5 to 23
Maria Orlando, Joan S. Tucker, and Phyllis L. Ellickson
RAND Corporation
Santa Monica, CA
Acknowledgments: The research reported in this article was funded by Grants 10RT-0062 and
13RT-0019 from the Tobacco-Related Disease and Research Program (TRDRP) administered by
the University of California
2
Abstract
Purpose: Adolescents who initiate smoking at a younger age are at increased risk for tobacco
dependence and continued use, tobacco-related health problems, and psychosocial and
behavioral difficulties during adolescence and emerging adulthood. Hazards models can identify
factors associated with the age of smoking initiation and help distinguish vulnerable periods for
initiation according to those factors.
Methods: Discrete time hazards analysis was used to model smoking initiation as a function of
age (5-23), demographic and familial influence variables (measured at age 13), and their
interactions in a cohort of 6,255 youth who completed six assessments over a 10-year period
from age 13 to 23 years.
Results: Half of the sample had initiated smoking by age 12, and the hazard for initiation was
greatest between ages 13 and 14. In addition to differences associated with race/ethnicity,
protective factors for age of initiation included high parental education and having an intact
family of origin. Having an older sibling, and participant’s exposure to smoking by the adult
most important to them were both risk factors, with the latter having a stronger effect for females
and African Americans. Although having an intact nuclear family was protective, the existence
of a nuclear family was not able to counteract the risk associated with both having an older
sibling and being exposed to an important adult smoker, regardless of gender. For all effects, the
strength of the association differed across the age range.
Conclusions: The impact of demographic and family influence factors on smoking initiation
varies over time. However, children are at maximum risk for initiation in their early teen years,
and the range of considerable vulnerability is during middle school and high school.
[Key words: smoking initiation, demographics, adolescents and young adults, survival analysis]
3
A Discrete Time Hazards Model of Smoking Initiation
Among West Coast Youth from Age 5 to 23
Adolescents who initiate smoking at a younger age are at increased risk for tobacco
dependence and continued use 1-3, as well as tobacco-related health problems 4, 5Early onset
smokers also differ from other adolescents on a broad range of risk and protective factors that
may foreshadow future problems 6. Adolescents who are smoking by grade 7, for example, are
more likely than their peers to report current academic problems, alcohol and drug use, and
delinquent behavior; further, these early smokers show significantly greater rates of such risky
behavior five years later, even after adjusting for these initial problems 7. Given the heightened
risk associated with early onset, it is important to identify and understand factors associated with
the age of smoking initiation and discover potential pathways to delaying initiation.
Demographic characteristics and familial influences are among the most widely studied
predictors of adolescent smoking initiation. Much of the research on demographic characteristics
has focused on race/ethnicity and gender. The majority of studies on ethnicity and smoking
initiation report earlier initiation and higher rates throughout adolescence among White and
Hispanic/Latino youth relative to African and Asian Americans 3, 8-14. However, one study
reported higher initiation rates for Whites than Hispanics 15, and our own study of West Coast
youth found that rates of lifetime smoking at age 13 were slightly higher among African
Americans compared to Whites 16.
In terms of gender, some results indicate similar initiation rates for boys and girls 14, 15,
while others have reported slightly lower rates for girls 11, 13, or gender differences within
racial/ethnic subgroups 9. Several studies have found that boys are more likely to initiate early
4
(i.e., before age 13), but girls’ initiation increases rapidly in the early teen years resulting in
equal initiation rates by about age 15 3, 10, 17, 18. Other demographic risk factors for initiation that
have been identified across studies include parental divorce or separation 6, 15, 19 and lower
parental education 6, 12.
Familial influences most relevant to smoking initiation include smoking by parents 6, 15, 20,
21
and older siblings 6, 17. Both increase the risk of smoking initiation, with several studies
suggesting that parental smoking may be particularly influential among Whites and girls 10, 17, 22.
Family smoking appears to be more strongly related to initiation than to the transition to regular
smoking 23, perhaps because it allows for easy access to cigarettes and opportunities for
experimentation. At least one study has found that initiation often takes place in the presence of
an older sibling 24.
Despite the considerable empirical attention that has been devoted to adolescent smoking
initiation and its correlates, relatively few studies have used hazard or survival modeling to focus
specifically on issues related to the timing of initiation such as when initiation is most likely to
occur, what risk factors are associated with early initiation, and how the impact of these risk
factors may change during the course of adolescence. Studies of this type that do exist have
limitations including using retrospective cross-sectional data 8, 25, 26, estimating the conditional
hazard of initiation (i.e., at wave 2 for non-smokers at wave 1; 15), or using a restricted age range
(i.e., 9-14; 11). One study, based on a sample from the UK, used incidence data from ages 12-22,
but the primary focus was on regular smoking and the authors did not examine whether the
impact of predictors on initiation varied over time 17.
The present study used a discrete time hazard model approach 27 to examine smoking
initiation from childhood to young adulthood as a function of age, demographic and familial
5
influence variables, and their interactions, in a cohort of 6,255 youth who completed multiple
assessments over a 10-year period from age 13 to 23 years. Retrospective age at initiation was
used for initiation prior to the first wave of data collection and incidence data was used for
initiation during the study period, providing a broad age range (from 5-23) that largely covers the
timing of initiation. Six demographic and familial influence variables, assessed at age 13, were
selected for inclusion in this study based on prior research indicating their relevance to
adolescent smoking initiation, as well as their stability (or relative stability) over time: gender,
race/ethnicity, parental education, having an intact nuclear family, presence of an older sibling,
and having an important adult who smoked. We expected that each of these variables would
predict smoking initiation, but were particularly interested in examining whether and how the
impact of these variables on initiation varied as a function of time. Based on previous literature,
we were also interested in testing the interaction between gender and ethnicity, and the
interaction of both of these factors with the important adult smoking variable. A final interest
was to examine the combined effect of different family characteristics on age of initiation to
determine whether certain protective factors are strong enough to counteract the influence of risk
factors.
Methods
Participants
Subjects included in this study were participants in the RAND Adolescent/Young Adult
Panel Study, a multi-year panel study originally conducted to evaluate Project ALERT, a drug
prevention program for middle school children (see Ellickson & Bell, 1990). The study sample
included students originally drawn from 30 California and Oregon middle or junior high schools.
6
The schools were chosen to reflect a wide range of school and community environments, and
encompassed urban, suburban, and rural school districts. Nine schools had a minority population
of 50% or more, and 18 schools drew students from neighborhoods with household incomes
below their state median.
The baseline sample in 1985 consisted of 6,527 seventh-graders whose parents had given
implied consent for their participation and who, themselves, gave active consent. This group
represented nearly the entire cohort (85%) of the 30 original schools; analyses of the eligible
subjects not surveyed at baseline showed that the sample was unbiased with respect to age,
gender, ethnicity, and grades 28.
Participants with missing information about their age of smoking initiation (N=263) or
ethnic background (N=9) were excluded from this study. Approximately half (49%) of the
sample of 6,255 participants in this study were female, and the majority were Caucasian (68%),
with 10% African American, 8% Asian, 10% Hispanic, and 4% of other ethnicity. Less than half
(44%) of participants reported that one or both of their parents had attended some college, over
half of the participants came from intact nuclear families (59%), many participants had at least
one older sibling (71%), and approximately half the sample (49%) had an important adult
smoker in their lives. Over half (56%) of the sample had smoked at least once prior to the
baseline survey.
Missing Data
The percentage of missing values for the variables used in this study was low (range <1%
to 5%), but listwise deletion of cases with any missing values would have resulted in a loss of
19% of the sample. To avoid this loss, we imputed the missing values for each predictor variable
7
based on regression from all the other variables using an imputation procedure provided by Stata
software 29.
Measures
Age of smoking initiation. For the 56% of participants who had initiated before Grade 7
(baseline assessment), age at initiation was based on participants’ self-reported age of first trying
a cigarette. However, 263 of these respondents were excluded from analyses because their
reported initiation age was missing or was less than 5 years old. For the remaining 44% who had
not initiated prior to the baseline survey, age at initiation was assigned as their approximate age
at the midpoint of the interval between the first wave at which the respondents said they had ever
smoked (even just a puff) and the previous wave (in which they said they had never smoked).
Predictor variables. Because many of the 6,255 participants had already initiated at the
time of the baseline survey, predictors in the model were limited to those that could be assumed
relatively constant prior to the baseline survey. These predictors included indicators for female
gender, ethnicity (African American, Hispanic, Asian, Other), existence of an intact nuclear
family (i.e., living with both biological parents), one or both parents having some college
education, having an older sibling, and whether the adult who was most important to the
respondent smoked cigarettes.
Analytic Approach
The age of smoking initiation was estimated with a discrete-time hazards model 27, 30.
This approach accounts for the right-censored data using logistic regression with a person-period
dataset and was implemented with SAS PROC LOGISTIC 31.
As a first step in fitting the model, different representations of time were tested (e.g., age
in years modeled as a series of indicator variables, or modeled as a polynomial function). These
8
representations were evaluated based on Fnested model comparison tests. To get a
representation of the overall trend of initiation, a simple model was fitted that modeled initiation
solely as a function of age and age-squared. Next, a more complex model was fitted that
incorporated the predictor variables. For this model, first the main effects were added, then the
planned interactions among factors were tested (between gender and ethnicity, and between these
two factors and important adult smoker), as were interactions between the predictor variables and
the age function. These interaction terms were evaluated based on the significance of the beta
coefficients. Finally, plots of the hazard and survival functions were constructed based on the
final model to represent the effects of several predictor variables.
Results
Of the 6,255 respondents included in this study, 1,175 (19%) left the study while still at
risk for smoking initiation and were treated as right-censored observations in all analyses. A
quadratic polynomial function (age and age-squared) provided an accurate and parsimonious
representation of the passage of time. Results from the simple discrete-time hazards model that
included this polynomial function as a predictor of initiation showed that the probability of
initiation increased over time (age: OR=1.26, p<.01), but decelerated during the later years (agesquared: OR=0.96, p<.01). Table 1 displays the hazard and survival rates over time for this basic
model and gives a general picture of smoking initiation across the age span. Examination of the
hazard rates show that the risk for initiation was greatest between ages 10 and 17. The survival
rates indicate that nearly 25% of the sample had initiated by age 9, over 50% by age 12, and
more than 75% had initiated by age 16. Overall, an estimated 16% of the sample did not smoke a
cigarette at some point before or during the study period. Additionally, most initiation in this
sample occurred before age 20. The hazard rates for age 20 and above were very low, and
9
survival rates also indicate a leveling off after age 19.
Odds ratios and confidence intervals from the full model that included the demographic
and family influence variables, as well as interactions are shown in Table 2. All of the predictor
variables had a significant impact on initiation either as main effects or in interaction terms.
Initial testing of the gender by ethnicity interactions revealed a significant gender by Asian
effect. However, when the model had been fully developed (i.e., when significant interactions
among predictors and the age function were added) these effects became non-significant, so all
gender by ethnicity effects were excluded from the final model. The interaction between gender
and important adult smoker was significant. There was also a significant interaction between
African American ethnicity and important adult smoker. Finally, there were several significant
interactions between the predictor variables and the polynomial age variables (age and agesquared). As a general rule, only significant interactions were included in the final model.
However, there were some instances where a variable had a significant interaction with agesquared, but the interaction with age and/or the main effect were not significant. In these cases,
all terms were included in the final model.
The presence of multiple significant interaction terms makes interpretation of the final
model results difficult. To aid in interpretation, plots of the hazard and survival functions were
generated (Figures 1 – 5) to illustrate the effects of most interest. Figure 1 contains the hazard
and survival functions for Whites, Hispanics, Asian Americans, and those of “other” ethnicity
(African Americans are excluded from this figure because of their interaction with adult smoker).
For these four ethnic groups, the risk of initiation peaked at age 13-14. In addition, consistent
with other findings, Asian Americans were at a lower risk of initiation across the age span and
had a higher survival rate than those in the other three ethnic groups. In contrast, Hispanics and
10
those in the “other” ethnicity group tended to have higher hazard rates. The absence of
interactions with age for Asian Americans and the “other” ethnic group indicates that their
effects were stable over time. For Hispanics, the hazard of initiation was initially lower, but
increased more rapidly and decreased more sharply than that for Whites, so at its’ peak the
hazard function was higher than that for Whites, but the range of relatively high risk was slightly
narrower.
The interaction between African American ethnicity and important adult smoker is
displayed in Figure 2. On average, the survival rates for African Americans were slightly lower
relative to Whites in the early years, but higher by the end of the study period indicating that
African Americans were at higher risk early on relative to Whites, but their risk diminished over
time. For Whites, the presence of an important adult smoker was a risk factor across the age
span. Although its’ influence waned in later years, White youth with an important adult smoker
had higher hazard rates up to age 17, and lower survival rates across the age span relative to their
counterparts without an important adult smoker. In contrast, for African Americans the presence
of an important adult smoker was actually a protective factor in the pre-teen years; the hazard for
initiation among African American youth without an important adult smoker was higher up to
age 11 and peaked at age 12, whereas the hazard for those with an adult smoker was higher after
age 11 and peaked at age 13. By age 15, the survival rates of African Americans without an adult
smoker caught up and exceeded rates of those with an adult smoker indicating that the early
protective effect of having an adult smoker had eroded.
The effects of gender and its’ interaction with important adult smoker are displayed in
Figure 3. Consistent with other studies, results show that boys tended to have higher hazard rates
in the early years, but that girls were at higher risk during mid- to late-adolescence. Also as has
11
been reported previously, the presence of an important adult smoker had a much stronger impact
for girls than for boys. The hazard function for girls with an important adult smoker spiked
sharply after age 10 and continued to be higher than the hazard for boys with an important adult
smoker throughout the teen years. In contrast, for those without an adult smoker, although the
hazard for girls exceeded that of boys in the early teen years, this occurred later, lasted less time,
and was not as dramatic. From age 15 on, the survival rates for girls without an adult smoker
were highest, while girls with an adult smoker had the lowest survival rates, averaging out to a
nearly equal survival rate between young men and women by age 24.
The effect of nuclear family was very striking (see Figure 3), indicating a strong
protective factor across the age span until about age 19, beyond which the hazard for initiation
was low regardless of nuclear family. Those with intact nuclear families at age 13 had a higher
survival rate across the age span relative to those without an intact family. The impact of high
parental education was inconsistent, showing a small protective factor in the early years, but
actually reversing at later ages; those whose parents had high levels of education had higher
hazard rates between ages 17 and 24 than those whose parents had lower levels of education, and
the overall survival rate at age 24 was very similar for the two groups, indicating that parental
education may delay initiation, but does not have a lasting protective effect. Throughout
childhood and the teen years, the presence of an older sibling was a risk factor for initiation.
Although the impact of the older sibling waned in the later years, it had a lasting effect on
survival rates.
A final set of plots, generated separately for boys and girls, illustrates the combined effect
of the different family characteristics (Figure 5). In these plots, the overall protective effect of
having an intact nuclear family (dotted line vs. solid line) is combined with risk factors (adult
12
smoker and older sibling). For both genders, the nuclear family remains somewhat protective
when there is an older sibling, and for boys the nuclear family’s protective effect also remains to
some degree in the presence of an important adult smoker. However, for girls, the protective
effect of the nuclear family is counteracted by the presence of an important adult smoker.
Finally, for both genders, the nuclear family’s protective effect erodes in the presence of both an
older sibling and an adult smoker, and this erosion is much stronger for girls.
Discussion
This study used a discrete time hazards analysis to model smoking initiation among a
large cohort of West Coast youth. As in other studies of smoking initiation 8, 11, 26, the greatest
risk for initiation occurred during the pre-teen and teen years, peaking at about age 13-14 and
then declining. The effects of demographic and familial influence variables on smoking
initiation in this study were also largely consistent with previous research (e.g., 3, 10, 17, 18).
However, the rich longitudinal data set and inclusion of interactions between predictive factors
and age in the model produced findings that significantly extend previous research by elucidating
the distinct periods during early adolescence during which identifiable subgroups are most
vulnerable to initiation. Furthermore, the ability to examine survival rates as well as hazard
functions allowed us to determine which demographic and familial influence variables had a
lasting impact on initiation, and which did not.
Results of this study further delineate the age of risk for smoking initiation. It is striking
that nearly 25% of the sample had initiated before age 10. This result emphasizes the need for
early prevention efforts before children reach middle school. Additionally, it appears that
children are at greatest risk for initiation in their early teen years. Although initiation rates
continue to increase throughout the teen years, there is a substantial deceleration in initiation
13
rates that nearly levels off after age 19, implying that the range of greatest vulnerability is during
middle school and high school.
Studies generally find that Asians initiate smoking at lower rates than Whites 8, 26 and this
was also true in our study. What other studies have not found is that the hazard of initiation for
Hispanics relative to Whites appears to vary over time. Although Hispanics had a lower hazard
of initiation during childhood compared to Whites, their hazard increased more rapidly during
early adolescence and decreased more sharply during late adolescence relative to Whites
indicating a narrower time frame for high vulnerability. In general, African Americans in this
sample tended to be at greater risk relative to Whites in their pre-teen years, but at lower risk
thereafter. By age 24, African Americans tended to have lower initiation rates than all other
ethnic groups with the exception of Asian Americans. However, as discussed below, differences
between African American and other youth in the hazard of initiation depended on the presence
of an important adult smoker.
In terms of family influences, having an older sibling was a risk factor for initiation,
although its impact waned in later years. Because variable selection was limited to those
considered as relatively stable prior to the baseline survey, the presence of an older sibling
smoker was not used as a predictor in this model. However, based on reported findings in the
literature on this topic as well as supplementary analyses, it is likely that the significance of the
older sibling variable is a marker for the influence of smoking by an older sibling, rather than
indicative of risk due to the presence of an older sibling per se. A substantial percentage (37%)
of respondents with older siblings reported that those older siblings were smokers. In addition, as
a sensitivity analysis, we estimated a model that had both older sibling non-smoker and older
14
sibling smoker as predictors. In this model, only older sibling smoker had a significant effect on
initiation.
More complex is the impact of smoking by an important adult on initiation, which differs
as a function of the youth’s race/ethnicity and gender. Among non-African American youth, the
presence of an important adult smoker was a risk factor for initiation from childhood through
emerging adulthood. However, the presence of an important adult smoker was actually
protective for black youth during the pre-teen years. Although unexpected, previous studies
have found that African Americans are more deterred by parental disapproval of smoking 32, 33.
Perhaps those who disapprove most effectively are those who are smoking themselves.
Incidentally, a relatively large percentage of African American respondents in this sample
reported that the adult most important to them smoked (58% as compared to 48% for nonAfrican Americans).
In the case of gender, the overall trend for girls and boys in this study was consistent with
other findings 3, 10, 17, 18 and suggested that on average boys were at greater risk than girls of
initiating smoking during childhood (until about 10) but that girls’ risk increased to exceed that
of boys after age 11 or so. It has been documented that girls are more likely to initiate smoking
if they are concerned with dieting 22, 34. It is possible that girls’ increased hazard relative to boys
at age 11 is associated with their growing awareness of body image and weight. In addition, the
presence of an important adult smoker was a much stronger risk factor for girls than boys.
Although other studies have also reported this result 10, 17, 22, it is not clear why girls are more
strongly influenced. Future research is necessary to understand the mechanism behind this effect,
but in the meantime, the strong impact of the important adult smoker in this and other studies
15
suggests a need for prevention programs to explicitly counteract this influence, especially among
girls.
The importance of social influences is demonstrated in our findings regarding nuclear
family. Although having an intact nuclear family was protective until age 19, its protective
effect eroded in the presence of an important adult smoker among girls, and the presence of both
an important adult smoker and an older sibling among boys. Finally, we found that higher
parental education had a small protective effect on initiation until about age 15, but reversed
directions and was a risk factor for initiation between ages 17 and 24. The overall survival rate
at age 24 is very similar regardless of parental education indicating that this factor may delay
initiation, but does not have a lasting protective effect. Perhaps the later initiation among those
whose parents had high levels of education is associated with going away to college and
heightened experimentation with substance use that often occurs during this period 35.
Strengths of this study include use of a large sample and a rich longitudinal data set that
allowed for the modeling of smoking initiation across the entire age range for which this event is
likely to occur. The study also benefited from the modeling of interactions that highlighted the
ways in which influential factors may change over time and according to demographic
characteristics. Limitations of this study include the fact that our findings are based on a single
sample of West Coast youth, which might restrict the generalizability of our results. However,
the schools from which participants were recruited were chosen for their diversity and the sample
includes high school dropouts. Second, attrition occurred over the 10-year follow-up period,
necessitating the use of sample weights to account for most of the bias due to source. A third
limitation is the reliance on self-report measures. Reports of current smoking at grades 7 and 8
were found to be highly accurate when externally validated and subjected to internal consistency
16
checks 36 and there is no reason to suspect that these self-reports became less accurate over time.
Nonetheless, it was not feasible to obtain external validation of smoking reports at subsequent
waves of data collection.
The results from this study comprise a rich picture of smoking initiation, the factors that
influence it, and how the strength of those influences change as youth approach young
adulthood. One of the more practical implications of these findings can be gleaned from the
significance of the family variables, especially the impact of the important adult smoker.
Recently, researchers have advocated that primary care physicians identify the smoking status of
parents of their adolescent patients 37. Results from this study indicate that this knowledge may
be useful in the prevention or delay of smoking initiation, especially among adolescent girls.
17
References
1.
Breslau N, Kilbey MM, Andreski P. Vulnerability to psychopathology in nicotinedependent smokers: an epidemiologic study of young adults. Am J Psychiatry.
1993;150(6):941-946.
2.
Chen J, Millar WJ. Age of smoking initiation: implications for quitting. Health Rep.
Spring 1998;9(4):39-46(Eng); 39-48(Fre).
3.
Everett SA, Warren CW, Sharp D, Kann L, Husten CG, Crossett LS. Initiation of
cigarette smoking and subsequent smoking behavior among U.S. high school students.
Prev Med. Nov 1999;29(5):327-333.
4.
Lando HA, Thai DT, Murray DM, et al. Age of initiation, smoking patterns, and risk in a
population of working adults. Prev Med. Dec 1999;29(6 Pt 1):590-598.
5.
Wiencke JK, Thurston SW, Kelsey KT, et al. Early age at smoking initiation and tobacco
carcinogen DNA damage in the lung. J Natl Cancer Inst. Apr 7 1999;91(7):614-619.
6.
Wills TA, Resko JA, Ainette MG, Mendoza D. Smoking onset in adolescence: a personcentered analysis with time-varying predictors. Health Psychol. Mar 2004;23(2):158-167.
7.
Ellickson PL, Tucker JS, Klein DJ. High-risk behaviors associated with early smoking:
results from a 5-year follow-up. J Adolesc Health. Jun 2001;28(6):465-473.
8.
Chen X, Unger JB, Cruz TB, Johnson CA. Smoking patterns of Asian-American youth in
California and their relationship with acculturation. J Adolesc Health. May
1999;24(5):321-328.
9.
Epstein JA, Botvin GJ, Diaz T. Ethnic and gender differences in smoking prevalence
among a longitudinal sample of inner-city adolescents. J Adolesc Health. Sep
1998;23(3):160-166.
18
10.
Griesler PC, Kandel DB. Ethnic differences in correlates of adolescent cigarette smoking.
J Adolesc Health. Sep 1998;23(3):167-180.
11.
Harrell JS, Bangdiwala SI, Deng S, Webb JP, Bradley C. Smoking initiation in youth: the
roles of gender, race, socioeconomics, and developmental status. J Adolesc Health. Nov
1998;23(5):271-279.
12.
Headen SW, Bauman KE, Deane GD, Koch GG. Are the correlates of cigarette smoking
initiation different for black and white adolescents? Am J Public Health. Jul
1991;81(7):854-858.
13.
Unger JB, Chen X. The role of social networks and media receptivity in predicting age of
smoking initiation: a proportional hazards model of risk and protective factors. Addict
Behav. May-Jun 1999;24(3):371-381.
14.
Wallace JM, Jr., Bachman JG, O'Malley PM, Schulenberg JE, Cooper SM, Johnston LD.
Gender and ethnic differences in smoking, drinking and illicit drug use among American
8th, 10th and 12th grade students, 1976-2000. Addiction. Feb 2003;98(2):225-234.
15.
Kandel DB, Kiros GE, Schaffran C, Hu MC. Racial/ethnic differences in cigarette
smoking initiation and progression to daily smoking: a multilevel analysis. Am J Public
Health. Jan 2004;94(1):128-135.
16.
Ellickson PL, Orlando M, Tucker JS, Klein DJ. From adolescence to young adulthood:
racial/ethnic disparities in smoking. American Journal of Public Health. Feb
2004;94(2):293-299.
17.
Swan AV, Creeser R, Murray M. When and why children first start to smoke. Int J
Epidemiol. Jun 1990;19(2):323-330.
19
18.
Warren CW, Kann L, Leavy Small M, Santelli JS, Collins JL, Kolbe LJ. Age of initiating
selected health-risk behaviors among high school students in the United States. Journal
of Adolescent Health. 1997;21:225-231.
19.
Kirby JB. The influence of parental separation on smoking initiation in adolescents. J
Health Soc Behav. Mar 2002;43(1):56-71.
20.
Barman SK, Pulkkinen L, Kaprio J, Rose RJ. Inattentiveness, parental smoking and
adolescent smoking initiation. Addiction. Aug 2004;99(8):1049-1061.
21.
Fleming CB, Kim H, Harachi TW, Catalano RF. Family processes for children in early
elementary school as predictors of smoking initiation. J Adolesc Health. Mar
2002;30(3):184-189.
22.
Robinson LA, Klesges RC, Zbikowski SM, Glaser R. Predictors of risk for different
stages of adolescent smoking in a biracial sample. J Consult Clin Psychol. Aug
1997;65(4):653-662.
23.
Chassin L, Presson CC, Sherman SJ, Corty E, Olshavsky RW. Predicting the onset of
cigarette smoking in adolescents: A longitudinal study. Journal of Applied Social
Psychology. 1984;14(3):224-243.
24.
Presti DE, Ary DV, Lichtenstein E. The context of smoking initiation and maintenance:
findings from interviews with youths. J Subst Abuse. 1992;4(1):35-45.
25.
Chen K, Kandel DB. The natural history of drug use from adolescence to the mid-thirties
in a general population sample. Am J Public Health. Jan 1995;85(1):41-47.
26.
Chen X, Unger JB. Hazards of smoking initiation among Asian American and non-Asian
adolescents in California: a survival model analysis. Prev Med. Jun 1999;28(6):589-599.
20
27.
Singer JD, Willett JB. Applied Longitudinal Data Analysis. NY, NY: Oxford University
Press; 2003.
28.
Bell RM, Gareleck C, Ellickson PL. Baseline Nonresponse in Project ALERT: Does it
Matter? Santa Monica: The RAND Corporation; April 1990. N-2933-CHF.
29.
StataCorp. Stata Statistical Software: Release 6.0. College Station, TX: Stata
Corporation; 1999.
30.
Cox DR, Oakes D. Analysis of Survival Data. London: Chapman and Hall; 1984.
31.
SAS Institute. SAS/STAT user's guide. Vol 1. Cary, NC: SAS Institute, Inc.; 1990.
32.
Catalano RF, Morrison DM, Wells EA, Gillmore MR, Iritani B, Hawkins JD. Ethnic
differences in family factors related to early drug initiation. J Stud Alcohol. May
1992;53(3):208-217.
33.
Clark PI, Scarisbrick-Hauser A, Gautam SP, Wirk SJ. Anti-tobacco socialization in
homes of African-American and white parents, and smoking and nonsmoking parents. J
Adolesc Health. May 1999;24(5):329-339.
34.
Austin SB, Gortmaker SL. Dieting and smoking initiation in early adolescent girls and
boys: a prospective study. Am J Public Health. Mar 2001;91(3):446-450.
35.
Arnett JJ. Emerging adulthood. A theory of development from the late teens through the
twenties. Am Psychol. 2000;55(5):469-480.
36.
Reinisch E, Bell RM, Ellickson PL. How Accurate are Adolescent Reports of Drug Use?
The RAND Corporation; 1991. N-3189-CHF.
37.
Key JD, Marsh LD. Missed opportunities for prevention: failure to identify smoking in
the parents of adolescent patients. Subst Abus. Dec 2002;23(4):215-221.
21
Table 1. Fitted hazard and survival probabilities from a basic model predicting initiation as a
function of age and age-squared.
Age
Fitted Hazard
Probability
Fitted Survival
Probability
5
0.016
0.984
6
0.028
0.956
7
0.047
0.911
8
0.070
0.847
9
0.098
0.764
10
0.126
0.668
11
0.151
0.567
12
0.169
0.471
13
0.178
0.387
14
0.175
0.319
15
0.162
0.268
16
0.140
0.230
17
0.113
0.204
18
0.085
0.187
19
0.059
0.176
20
0.037
0.169
21
0.022
0.165
22
0.012
0.163
23
0.006
0.162
24
0.003
0.162
Note: Shaded hazard probabilities indicate age range of highest risk, bolded survival rates
indicate approximate quartiles of the survival distribution.
22
Table 2. Hazards ratios and 95% confidence intervals for discrete-time hazard model predicting
smoking initiation as a function of age, demographics, and family influence variables.
Predictor
Hazard
Ratio
95% Wald
Confidence Limits
Age
1.254
(1.212– 1.297)**
Age-squared
0.965
(0.959 - 0.970)**
Female
0.373
(0.285 – 0.488)**
Female*Age
1.089
(1.063 – 1.116)**
Female*Age-squared
0.990
(0.986 – 0.994)**
African American
2.503
(1.361 – 4.602)**
African American*Age
0.938
(0.886 – 0.992)*
African American*Age-squared
0.989
(0.981 – 0.998)*
Asian
0.714
(0.636 – 0.802)**
Hispanic
0.491
(0.316 – 0.762)**
Hispanic*Age
1.096
(1.051 – 1.144)**
Hispanic*Age-squared
0.988
(0.979 – 0.996)**
Other ethnicity
1.221
(1.049– 1.420)**
Nuclear family
0.684
(0.531 – 0.881)**
Nuclear family*Age
0.995
(0.972 – 1.020)
Nuclear family*Age-squared
1.005
(1.001 – 1.010)*
Parents education
0.918
(0.711 – 1.185)
Parents education*Age
0.989
(0.965 – 1.014)
Parents education*Age-squared
1.007
(1.002 – 1.011)**
Adult smokes
2.519
(1.973 – 3.215)**
Adult smokes*Age
0.945
(0.926 – 0.965)**
Has older sibling
1.078
(0.816 – 1.423)
Has older sibling*Age
1.020
(0.993 – 1.047)
Has older sibling*Age-squared
0.995
(0.990 – 0.999)*
Female *Adult smokes
1.232
(1.091 – 1.391)**
African American*Adult smokes
0.154
(0.070 – 0.338)**
African American*Adult smokes *Age
Note: * indicates p<.05; ** indicates p<.01
1.157
(1.078 – 1.241)**
23
Figure Captions
Figure 1. Hazard and Survival Functions for Four Ethnic Groups
Figure 2. Hazard and Survival Functions for White and Black Ethnicity by Adult Smoker
Figure 3. Hazard and Survival Functions for Gender by Adult Smoker
Figure 4. Hazard and Survival Functions for Nuclear Family, High Parental Education, and Older
Sibling. Note: Comparison represents respondents with non-nuclear families, parents of high
school education or less, and no older siblings.
Figure 5. Hazard and Survival Functions for Combinations of Family Factors by Gender. Note:
Comparison represents respondents with non-nuclear families, no older siblings, and no
important adult smoker.
24
0.25
1
0.2
0.8
Fitted Survival Probability
Fitted Hazard Probability
Hazard and Survival Functions for Four Ethnic Groups
0.15
0.1
0.05
0
0.6
0.4
0.2
0
5
7
9
11
13
15
17
19
21
23
5
7
9
Age First Cigarette
---¨
11
13
15
17
19
21
23
21
23
Age First Cigarette
Hispanic
-x-- Other
—— White
----- Asian
0.25
1
0.2
0.8
Fitted Survival Probability
Fitted Hazard Probability
Hazard and Survival Functions for White and Black Ethnicity by Adult Smoker
0.15
0.1
0.05
0
0.6
0.4
0.2
0
5
7
9
11
13
15
17
19
21
23
Age First Cigarette
— White, No Adult Smoker
--- Black, No Adult Smoker
5
7
9
11
13
15
17
19
Age First Cigarette
¨ White, Adult Smoker - -ż- Black, Adult Smoker
25
0.25
1
0.2
0.8
Fitted Survival Probability
Fitted Hazard Probability
Hazard and Survival Functions for Gender by Adult Smoker
0.15
0.1
0.6
0.4
0.2
0.05
0
0
5
7
9
11
13
15
17
19
21
23
5
7
9
13
15
17
19
21
23
Age First Cigarette
Age First Cigarette
— Boys, No Adult Smoker
11
--- Girls, No Adult Smoker
¨ Boys, Adult Smoker - -ż- Girls, Adult Smoker
0.25
1
0.2
0.8
Fitted Survival Probability
Fitted Hazard Probability
Hazard and Survival Functions for Nuclear Family, High Parental Education, and Older Sibling
0.15
0.1
0.6
0.4
0.2
0.05
0
0
5
7
9
11
13
15
17
19
21
23
5
7
9
13
15
17
Age First Cigarette
Age First Cigarette
— Comparison ---- Nuclear Family
11
ż- High Parental Ed.
¨ Older Sibling
19
21
23
26
Hazard and Survival Functions for Combinations of Family Factors by Gender
0.25
1
0.2
0.8
Fitted Survival Probability
Fitted Hazard Probability
Girls
0.15
0.1
0.6
0.4
0.2
0.05
0
0
5
7
9
11
13
15
17
19
21
23
5
7
9
11
13
15
17
19
21
23
19
21
23
Age First Cigarette
Age First Cigarette
0.25
1
0.2
0.8
Fitted Survival Probability
Fitted Hazard Probability
Boys
0.15
0.1
0.05
0.6
0.4
0.2
0
0
5
7
9
11
13
15
17
19
21
23
Age First Cigarette
— Comparison
7
9
11
13
15
17
Age First Cigarette
---- Nuclear Family
¨ Nuclear Family and Adult Smoker
5
ż- Nuclear Family and Older Sibling
x- Nuclear Family and Older Sibling and Adult Smoker