Changes in Smoking Behavior over Family Transitions: Evidence for Anticipation and Adaptation Effects

The study of changes in smoking behaviors over the life course is a promising line of research. This paper aims to analyze the temporal relation between family transitions (partnership formation, first childbirth, separation) and changes in smoking initiation and cessation. We propose a discrete-time logistic model to explore the timing of changes in terms of leads and lags effects up to three years around the event in order to measure both anticipation and adaptation mechanisms. Retrospective biographical data from the Santé et Itinéraires Professionnels (SIP) survey conducted in France in 2006 are used. Partnership formation was followed for both genders by a fall in smoking initiation and an immediate rise in smoking cessation. Childbirth was associated with increased smoking cessation immediately around childbirth, and additionally, females showed an anticipatory increase in smoking cessation up to two years before childbirth. Couple separation was accompanied by an anticipatory increase in smoking initiation for females up to two years prior to the separation, but this effect only occurred in males during separation. Our findings highlight opportunities for more targeted interventions over the life course to reduce smoking, and therefore have relevance for general practitioners and public policy elaboration.


Introduction
Over the past decades, the dissemination of information on the harmful effects of tobacco and the implementation of public policies have contributed to a gradual and continuous decrease in smoking prevalence. In fact, this decline in smoking was largely attributed to a decrease in prevalence in the most privileged groups and was more contrasted for women [1][2][3] . These evolutions have led to a sharp increase in social inequalities in smoking and a convergence of male and female prevalence.
At the present time, differences between men and women remain in the age pattern of smoking for the most recent cohorts: in the most educated group, women more frequently smoke regularly than men before the age of 25, and do so less frequently after this age. These differences can probably be explained by a greater cessation of women at the approach of the their first pregnancy, especially among the most educated, and could also reflect the specificities of smoking patterns by gender and social group [4][5][6] .
Exploring the dynamics of changes in smoking behaviors over the life course appears to be a promising line of research to understand the persistence as well as the evolution of smoking disparities over cohorts and over ages.
Smoking during childhood and adolescence has been studied in detail since this period of life is crucial to the understanding of the context of experimentation and transition to daily use of cigarettes. This literature has highlighted the responsiveness of smoking initiation to price and regulation factors as well as family and peer contexts explaining smoking initiation [7][8][9] . But the period of smoking initiation goes beyond adolescence and corresponds to the period of the transition into adulthood. This period of life is associated with the transitions to adult work and family roles and the related changes in health behaviors 10,11 , which could explain differences in smoking across social groups 12 . To date, the literature on the impact of family life transitions on smoking behaviors is very limited. The present study is therefore an attempt to fill this gap by focusing on the impact of family transitions on smoking behavior, especially in relation to couple formation, first childbirth and separation.
Empirical findings on smoking behaviors and family events are mostly based on contemporaneous correlations in cross-sectional data, with of a focus on the effect of family status rather than on family events. A protective effect of being in couple on smoking has been shown with a greater influence of living with a partner for men, whereas women are more affected by separation 13,14 which is consistent with the greater vulnerability of women to stressful events 15 . These effects can also be explained by an influence of partner on smoking, but for that it is necessary to distinguish the effects of matching or bargaining: matching between partners would come from a tendency of individuals to choose a partner sharing the same lifestyles or preferences whereas bargaining refers to a negotiation process between spouses or partners about smoking 16 . Some authors have shown that the correlation of smoking between partners would rather be due to the effects of matching than bargaining 17 but an influence of partner on smoking have also been demonstrated 18 . Concerning the birth of a child, the literature shows few changes in smoking of men in comparison with women. For women, there would be relatively little long-term effect because of relapse after pregnancy [19][20][21] . Studies of men's smoking behaviors at the time of the birth of their child are rare and show little changes in smoking behavior or only among the besteducated 22 , with the exception of changes in their habits such as not smoking in the home 23 .
One shortcoming of available studies is that they have considered family transitions as discrete events and only considered the subsequent changes. Viewing family transitions as discrete events per see is questionable, as those transitions are the outcome of a long process and their consequences develop over time. For instance, marital relations start to deteriorate long before the partners break up, and there is often a long time elapsing from the initiation of the project of having a child to the actual birth 14 . Family transitions may therefore lead to changes in behavior occurring before the happening of the event ("anticipation effects"). They may also lead to changes after the transition has taken place, and those may be temporary or permanent ("adaptation effects"). Another aspect which has hardly been addressed is that of the gendered dimension of the long-term processes. In other words, do transitions within the

Empirical strategy
The aim of this paper is to estimate the impact of family events on smoking trajectory, and particularly the changes in smoking behavior before (anticipation), during (contemporary) and after (adaptation) the transition to a different family situation.

A general approach
We favor a discrete time approach with panel econometric strategies to distinguish between individual variations and within individual variations in longitudinal data structure. This strategy is usual in multilevel analysis and estimates both within and between individual effects by including individual means of time-varying variables, called within-between estimator or Mundlak specification in the econometric literature 24,25 . The formulation allows to test within and between effects in the same model.
An alternative approach often used in linear model is to estimate within effects separately using fixed effects model. For the case of a binary outcome variable, such as smoking status, we favor the withinbetween strategy instead of the conditional logit fixed effects model that would have excluded individuals without longitudinal variation. This approach has the advantage to include ever-smokers and never-smokers in the analysis.
We consider a discrete-time logistic regression of smoking with the following formulation of the latent variable of smoking for the individual i at age t: * = .

A specific approach to measure leads and lags
The general approach presented before assumes a contemporary relationship between variables which could be very restrictive. Here, we propose a specific approach to measure leads (anticipation) and lags We propose to consider a three years window around life events, with dummies of leads effect (3 years before, 2 years before, on year before), the contemporary effect and dummies of lags effect (one year after, two years after, three of more years after) : * = .
We assume here a reference category for the interpretation of this dynamic, which is "being at least three years before the occurrence of the event" (including never experiencing it).
The effect of the different transitions are estimated jointly in the same model to adjust for successive events. We estimate a model for smoking initiation and another for cessation, separating men and women in each of them.

Data
This empirical work is based on the data derived from the national representative SIP ( (current smokers), we constructed a smoking binary indicator for every year between the age of 10 and the age at interview, whose value is one if a respondent smoked or zero otherwise.
We use information collected on a biographical grid to construct the family status variables. Partnership episodes are reported and allow to date the age at union and separation periods ("Avez-vous déjà vécu en couple ? Si oui, situez la ou les périodes"). The questionnaire defined a partnership episode as a cohabitation period (not necessarily a marriage) during at least one year. Shorter episodes are reported only in the case of the birth of a child. We also used information on the age at first birth ("Avez-vous eu ou bien adopté des enfants ? Si oui, notez les années de naissance").
Adjustment was made for activity status as a time-varying variable in 5 categories: long-term employment (five years or more), short-employment (less than five years, or inactivity and unemployment of less than one year), unemployment (more than one year), inactivity (more than one year) and schooling. The following variables were also adjusted for: social class from last or current (6 categories); number of major life events during childhood (handicap or serious illness, health problems or death of a relative, family conflict or separation, violence, financial difficulties, etc.); educational level; migration status; residence in a rural municipality at the time of the survey; birth cohort, and; log of cigarette prices of the period (see Table S2 in supplementary materials for descriptive statistics of the variables).
The distributions of age at smoking initiation and cessation and of family events are presented in supplementary materials Table S1 (see Figures S1 and S2 in supplementary materials for graphical representations). This retrospective information is used to construct a person-years database of smoking status, family status and other time-varying or fixed individual variables. Given the age of the first family events, we restricted the person-year database of analysis from 17 years old to 50 years old. This dynamic effect has to be interpreted in reference to the baseline level which represents at least three years before the event or never having experienced it.

Anticipation and adaptation in smoking behaviors with family transitions
The graphs in Figure 1 refer to couple formation. Living with a partner has a protective effect on smoking initiation: it delays and decreases smoking initiation after 3 years of cohabitation for men and at least after two years for women (top panel). This protective effect of living in couple occurs as a break in behavior as smoking initiation was significantly higher for women 2 years before this event and as similar but non-significant effects are observed for men.

Figure 1: Leads and lags in smoking -Living in couple (Marginal effects)
Living with a partner has also an effect on smoking cessation but this effect is mainly a contemporaneous effect which is significant for men and women at the time of the event (bottom panel). This protective effect is consistent with the hypothesis of bargaining effect between partners at the time of couple formation leading to a change in behavior or with the pleasure or ease of being in couple. There is also a long term effect of living with a partner as it increases slightly smoking cessation for men after at least 2 years from this transition (although no such effect is observed in women). Those results confirm the behavioural change following couple formation and the long-term effects of living with a partner, the effects being larger for men than for women.
The graphs in Figure 2

Figure 2 : Leads and lags in smoking -First childbirth (Marginal effects)
Further to that, there are adaptation effects related to the birth of a child. We find an increase in smoking initiation at least three years after first childbirth for women and to a lesser extent for men (top panel). This result can be interpreted as a response to the stress and constraints related to the joint exercise of the work and family roles 31 , that would limit the protective effect of childbirth in the long term, especially for women.
The dynamics of separation is also characterized by a contrast between anticipation and adaptation effects and a strong gender pattern (Figure 3). Anticipation effects precede separation for women with an increase in smoking initiation two years and one year before the separation, while the effect is only contemporaneous for men (top panel). Additionally, there is an adaptation effect, with an increase in smoking cessation after three years of separation for men (column 1, bottom panel) and a decrease in smoking initiation after three years of separation for women (column 2, top panel).

Discussion
We studied the associations between family transitions and tobacco smoking with a method that distinguishes anticipation, contemporaneous and adaptation effects. This takes into account the fact that

Summary of the findings
Men and women tend to behave in the same way with two exceptions: the first childbirth and the couple formation. The first childbirth accelerates cessation and delays initiation among women, while it only has a positive effect on cessation for men, which is mainly contemporaneous and fades quickly.
These results are in line with previous publications regarding anticipation of pregnancy among women and reduced effects in men 5,23 . But this beneficial effect of childbirth may also be temporary for some women who initiate smoking after the birth of their child.
Similarly, living with a partner deters initiation for both genders, while it has a long-term enhancing effect on cessation for men but not for women (for whom only a contemporaneous effect is found). This is in line with the literature that emphasises the fact that marriage is protective against smoking 14,32 and more generally favourable in terms of health behaviours 16 and coherent with findings showing that being married/cohabiting is important for men's cessation success but not for women's. No anticipation effect is clearly visible. Finally a separation is accompanied by an increase in initiation the years before and an increase in cessation some years after. This accords with previous studies concluding that marital disruption is connected to increased smoking risks 14,33 , but we further reveal that the adaptation effects lead in the long term to a return to more balanced behaviours.

Limitations
Our study is based on retrospective data which may lead to some biases and measurement errors. First, differential mortality among smokers could lead to an underestimation of the prevalence of smoking for people older than 70 as suggested by a study testing the validity of retrospective data on smoking 34 .
This bias may be limited because only a very small part of our sample is aged more than 70. The use of retrospective data could also entail recall bias and reporting errors. It was demonstrated that these biases were lower for heavy smokers 35 but we cannot address this problem because of lack of information on the number of cigarettes that former smokers were smoking. These biases may vary with level of education and lead to underestimating educational inequalities in smoking in the case of under-reporting by the less educated. The reconstruction of smoking itineraries is also limited by the recall of a single smoking period per individual, not allowing for temporary quitting, which is commoner among the less educated, whose attempts to quit are more numerous and less successful 36,37 . This bias could in fact minimize the short term effect of some family events or transitions as adult ex-smokers are more likely to start smoking again. Nevertheless, the use of retrospective data on smoking avoids attrition and mortality with age which can be encountered with longitudinal studies. Furthermore, the questionnaire was based on an age-event grid, a validated technique of high reliability that allows a homogenous and coherent recall of family and occupational life events 38 . In addition, the advantage of our data is that the unit of time is the year, which is shorter than what is used in the vast majority of longitudinal surveys in which subjects are surveyed every two or five years without any retrospective report of change in smoking behaviour. This allows a much more detailed analysis, especially the measurement of lags and leads effects.
It is worth noting that we focus on some key family events related to the life cycle, adjusting on the mean time passed in each configuration (couple, first childbirth and separation/single) during the whole life. Couple formation and separation may occur many times during life and it has been shown that multiple broken partnerships reduce the probability of smoking cessation in men 39 , while it is likely that the anticipation or adaptation effect would vary over age and repetition of the same events. And yet, the succession of the events and their timing are likely to moderate the effects on tobacco smoking: for example maternal age is strongly linked to an initiation after birth 40 . Our measure of the average effects of the family events is a strong argument for the robustness of our results.
Contrarily to most papers published on the subject, we did consider cohabitation and living with a partner rather than marital status: this definition is broader and less specific, but maybe more adapted to contemporary life. Nystedt (2006) found that the cohabitation was linked to an increase in smoking compared to marriage, for both genders, which is consistent with a long term protective effect of partnership.

Conclusions
Family life transitions influence health behaviors, leading to permanent or more temporary changes, and  Tables   Table S1: Distributions of age at smoking initiation and cessation and at family events   Table S2: Descriptive statistics Table S3: Discrete-time logistic regression of smoking initiation -Men and Women (Odds-ratio)

Role of Funding Source
The data collection was funded by the National Institute for Health Education and Promotion (INPES) which is a public health agency. The analysis was funded by the National Institute for Demographic Studies which is a public research centre.