2.2.2. Breastfeeding

Mothers completed a retrospective questionnaire that asked whether they breastfed (78.3%), used formula (5.6%) or both (16%). In this study, we define exclusive breastfeeding as exclusive for a period of time with a minimum of one month. Non-exclusive breastfeeding is defined to include formula only or mixed methods. Mothers were also asked to report breastfeeding duration in months (mean = 8.79, range 0–24). Duration was categorized into three levels: less or equal to 7 months (25%), between 7 and 10 months (51%) and greater or equal to 10 months (24%).

#### 2.2.3. Maternal Interaction

Mothers were asked a follow-up question for (breast) feeding: "Did you talk to the child while (breast) feeding in the first two years" These responses were: 1 = Never, 2 = Sometimes, 3 = Always. In our study, not all mothers were always talking to the child while breastfeeding. We combined the two measured into one and named it "Feeding types and bonding", reflecting the beneficial effects of both nutritional and active communication. In our sample, we identified four groups. 34.8% of the mothers who always use breast milk and always talk to the child while feeding were classified into Breastfed and Active bonding (group 1); 43.5% who never talk to the child even when they used breast milk were assigned into Breastfed and no Active Bonding (group 2); 9% of the mothers were in the group of Non-breastfed and Active Bonding (group 3), and the final 12.7% in the Non-breastfed and no Active Bonding (group 4).

#### 2.2.4. Social Adversity

Parents were asked to fill in a sociodemographic questionnaire at the same time they completed the CBCL when children were six years old. A number of researchers have demonstrated the utility of combining of several individual measurements of psychosocial risk factors in studying child behavior problems [20,21]. The adversity index was created along lines similar to those described by Rutter and Moffitt [20,21]. A total adversity score was derived based on 11 variables. This score was created by adding 1 point (for 9 of the 11 indicators) or 2 points (for 2 indicators) for 11 adversity indicators: mother's low education (below middle school, 8.8%), father's low education (below middle school, 5.4%), mother's low occupational status (3-point scale: 0 = professional or skilled work, 22.2%; 1 = un-skilled worker, 38.9%; and 2 = no occupation, 31.8%), father's low occupational status (3-point scale: 0 = professional or skilled work, 26.8%; 1 = unskilled worker, 60.3%; and 2 = no job, 3.8%), mother's poor health status (2.1%), father's poor health status (3.8%), obstetric complication (bleeding, hypertension, diabetes, Caesarian section, difficult birth, low birth weight, difficulty breathing, 35.6%), divorce (3.3%), absence of biological mother (4.2%), house size below 70 m<sup>2</sup> (8.4%), and poor neighborhood (overcrowded neighborhood, noise pollution, damp, 35.6%). Details on these indicators are given in Liu *et al*. [22]. The adversity score ranged from 0 to 13 (M = 3.51, SD = 2.04).

#### *2.3. Statistical Methods*

To test whether there is any social adversity or internalizing behavior score differences between exclusive and non-exclusive breastfeeding groups, independent *t* tests were employed. Differences in proportions for gender were tested using Ȥ 2 tests. The same tests were conducted again to test differences between active bonding and non-active bonding group. Cohen's *d* was calculated to show the size of the effect between two comparison groups.

To test for the combined effects of feeding and bonding types on internalizing behaviors, a series of analysis of variance (ANOVA) were performed. "Feeding and bonding types" was a new variable (four groups defined in the method section) created from the two independent binary variables. It was the grouping variable in ANOVA, and the dependent variables were emotionally reactive, Anxious/Depressed, Somatic Complaints, Withdrawn and total internalizing problems respectively. Omega squared *w* 2 was computed as an index of effect size with several groups.

To test for the effect of social adversity as a potential confound, general linear models were fitted with adversity entered as covariate. The effect moderator of sex was assessed by entering the measure as a factor alongside feeding and bonding type. Gender × breastfeeding and bonding types interaction term was included in the model as well. The interaction effect of breastfeeding types and bonding types was tested in the models besides including these two variables independently in the models. To test for a dose-response relationship between breastfeeding duration and internalizing behavior, the grouping variable of duration took on three levels ( PRQWKV –10 months and PRQWKV IRU univariate ANOVAs. All results were considered significant if *P* < 0.05 using a two-tailed test. Statistical analyses were conducted by using SPSS version 20.0 (IBM SPSS Statistics).
