Study Population and Data Collection
Data from the China Health and Nutrition Survey (CHNS,1991-2015), which is a large-scale, longitudinal household-based survey initiated in 1989 that consists of representative participants of varying economic status, health indicators, and geographic areas throughout China[24]. Further information on survey procedures and the sampling scheme was reported in detail elsewhere [25]. We excluded participants who were pregnant, lactating, missing key variables from the analysis. Hence, our final sample consisted of 75,348 person-year observations aged 18-65 years with complete data on demographic, smoking status, obesity indicators.
2.1 Obesity indicators
Well-trained health workers measured the height, weight, and wasit circumstance (WC) following standardized procedures. Dividing weight (in kg) by the square of height (in m2) we got BMI and grouped it into thin (<18.5 kg/m2), normal (18.5-23.9 kg/m2), overweight (24-27.9 kg/m2)and obesity (>28 kg/m2) [26]. We defined participants as having abdominal obesity if the WC ≥85 cm in females and ≥90 cm in males In accordance with the national health and family commission for Chinese adults (2013)[27].
2.2 Smoking indicators
In the CHNS questionnaire, two variables were constructed to measure smoking behavior. The first was smoking status, which was classified into “current smoker” or “nonsmoker”. The second was smoking intensity, based on the question about “number of cigarettes consumed daily”, current smokers were further divided into heavy (>25cigarette/d), moderate (15-24cigarette/d) and light smokers (1-14cigarette/d)[18].
2.3 Other variables
We grouped participants into three education level (primary/illiterate, middle, and high school/above), two marital statuses (single / married), two geographical regions (rural / urban), three income levels (low / medium / high) and two drinking status (yes/no). The per-capital annual income in each survey was inflated to values in 2015. Participants reported all physical activities (PA) in average hours per week, and we converted the time spent in each activity into a metabolic equivalent of task (MET) hours per week based on the Compendium of Physical Activities [28].Dietary intake at the individual level was assessed by using three consecutive 24-h dietary recalls in each wave of the CHNS [24]. Based on the Chinese Food Composition Table, we measured nutrition intake data by (1) the respondent’s daily calorie intake per day during a 3-day measurement period and (2) the dietary structure as indicated by the percentage of fat in the respondent’s daily calorie intake [26].
3.Statistical Analysis
Statistical analyses were performed using the STATA 13.0 (STATA, Stata Corp). We first employed a pooled ordinary least square (OLS) model covering from1991 to 2015. Outcome variables including dietary energy intake, the percentage of dietary intake from fat and physical activity.
The equation we used for this model is
K abc = α0 + α1SMOKEabc + α2Xabc + δc + αb + εabc (1)
K abc denote the lifestyle behavior of individual a in province b and year c.
SMOKE abc indicating whether the individual is current smoker or nonsmoker in year c;
X abc contains control variables;
δc and αb indicate time and province fixed-effects, respectively;
ε abc is the random error that varies with individual, province and year.
α 1 indicates the impact of smoking on the lifestyle behavior.
Then we employed Probit model for binary dependent variables. Outcome variables including drinking status, general obesity and abdominal obesity. The equation we used for this model:
K’ abc =α0 + α1SMOKEabc + α2Xabc + δc + αb + εabc (2)
P = (K abc = |SMOKE, X, δ, α) = P (K’abc = |SMOKE, X, δ, α)
=G (α 1 SMOKE abc + α2Xabc + δc + αb + εabc ) (3)
K’ abc is the latent variable specifying the tendency of drinking, being general obesity and abdominal obesity. If K’abc>0,then Kabc =1;Otherwise Kabc =0.
G (.) is the cumulative distribution function of random error εabc.
When we investigate the effects of smoking intensity on lifestyle behaviors. We change the SMOKEabc in model 1–3 to SMOKEI abc. SMOKEI abc is dummy variable indicating whether the current smoker is heavy smoker (1 = yes, 0 = no), moderate smoker (1 = yes, 0 = no) or light smoker (1 = yes, 0 = no).