The survey has been conducted using a CATI (Computer Assisted Telephone Interview) facility on 1215 children age 6-10 years, in January 2010 in Italy. Calls have been conducted by experienced CATI operator. The sample has been stratified according to the definition of snacker, thus collecting information on 608 "snackers" and 607 "no-snackers" children.

One-thousand two-hundred fifteen mothers, randomly selected through an automated system afferent to the Italian landline phone directory have been asked to answer directly on their child and on the group of the closest friends of him/her, defined within a limit of 10 children per referenced child. The random sample was constructed in a way to approximately provide an equal geographical distribution of the families in the North, in the Center and in the South of Italy.

The study is based on telephone interview on a randomly chosen set of mothers living in Italy and with a cable-phone connection. Informed consent was obtained from people interviewed before starting the set of questions.

First, mothers answering to the phone calls were asked to provide information useful for stratifying the child as "snacker" or "no-snacker". Thus, the questionnaire was divided into 4 parts. The first part investigated on basic sociodemographic characteristics, like education, composition of the family and work. In the second section mothers were asked to answer about family's physical characteristics, like weight and height. The third part assessed the behavioral aspects of their children, like weekly hours of physical activity and more details on the consumption patterns of snacks. The latter part regarded children's social network; mothers were asked to answer on their children's friends, up to a maximum of 10 children, on snacking habits, physical activities and perception of their nutritional status.

The "snacker" has been defined as consuming at least 3 or more of the following products per day: filled cakes/sweet snacks, plain cakes/sweet snacks, chocolate snacks, sodas (no sugar-free) in can or bottle. This adds up to an average of at least 360-400 calories, which counts up to 18-22% of the recommended daily caloric intake for a child (about 1700-1900 calories per day in the age class 6-10) and it is coherent with the definition of the "snacker" as provided in the literature 8 .

Body mass index (BMI) was calculated as weight (kg)/height² (m). According to the present guidelines, in adults, normal weight, overweight and obesity were defined as a BMI < 25.0 kg/m², 25.0- 29.9 kg/m², and ≥ 30.0 kg/m², respectively 14 . Data on weight and height of the child, used to compute the BMI were updated in 76% of the cases to the last three months and they, in the remaining 24%, never exceeded the year. Children were defined as obese according to the percentile to which they were belonging, according to their age, as indicated from the CDC (Centers for Disease Control and Prevention) pediatric growths charts 15 ; thus, they were defined as "at risk for overweight" if their weight was between the 85^th and the 95^th percentile and as "overweight" is greater than the 95^th percentile 16 . CDC pediatrics growth charts were chosen as recommended when assessing growth's parameters of children older than 2 years 17 18 .

Continuous variables are expressed as median and inter-quartile difference and categorical variables as percentages and absolute numbers. Differences between groups were compared using Wilcoxon and McNemar or Chi-Square tests 19 20 , as appropriate.

All variables investigated are listed in Tables 1 and 2. Association of variables with the status of snacker has been investigated using a multivariable logistic regression model. Association of variables with the BMI level of the children in the "snacker" group has been investigated using a multivariable linear regression model. The following model building strategy applies to both models. Variables related (with a maximum p-value of the association, as resulting in the univariable analysis, of 0.25), to the network around the snacker have been also inserted in the model and final estimates have been adjusted for inter-network correlation using a Huber-White estimating equation approach.

All variables considered were entered into the model "as is", i.e. without any transformation or cutting-off. If a significant non-linearity using a score test was found, in relating the covariate's effect with survival, the specific covariate's effect was modeled using a restricted cubic spline. Selection criteria was the AIC (Akaike Information Criterion) applied backward for selecting significant covariates. The final model for each of the three steps was selected if superior in terms of AIC at a significance level of 0.05 and p-values have been explicitly indicated if below the 0.25 threshold, otherwise the "NS" indication is used.

To account for possible overfitting in the regression models secondary to high ratio between covariates and events, cross-validation and bootstrap (200 runs) techniques were applied. For the logistic regression model, Somer's concordance Index Dxy (the closer to one in absolute value the better) were obtained and evaluated for this purpose.

Multivariate Odds Ratios have been presented along with their 95% confidence intervals for the logistic regression model, whereas for the linear regression model effects on BMI scale have been presented along with their 95% confidence levels.

The statistical significance was settled at a p-value <0.05. The R System (release 2.7.0) statistical package and the Harrell's Design and Hmisc libraries were used for analysis.

The snacking activity, according to the definition adopted in the study, was eating at least three snacks per day. The distribution of snacks were: Plain cakes 230 (39.2%), Cream filled cakes 336 (57.2%), Chocolate cakes 227 (38.7%), Sugar sodas 299 (50.9%) and other types 16 (2.7%).

The network around the reference child is described in Table 3. The social network identified according the strict rules adopted in the study was composed by on average 1.27 friends per child surveyed, slightly but significantly higher for "snackers" than for "no-snackers" (1.40 vs 1.14, p = 0.042).

To understand which individual, familiar or social network characteristics are associated with the status of "snacker", a multivariable model has been built, following the variable selection procedure as described in the statistical section among all variables listed in Table 1 and in Table 2. Results of the multivariable model are presented in Table 4.

To understand which individual, familiar or social network characteristics are associated with any given BMI of the "snacker", a multivariable model has been built, following the variable selection procedure as described in the statistical section among all variables listed in Table 1 and in Table 2 but limited to the group of the 608 "snackers". Results from this multivariable analysis are shown in Table 5.

Results of the limited research on behavioral mediators of familial patterns of overweight indicate that parents' own eating behaviors and parenting practices influence the development of children's eating behaviors, mediating familial patterns of overweight. Parents provide food environments for their children's early experiences with food and eating, and overweight parents seem to select environments that promote overweight among their children 21 .

In our study two parental characteristics have been evaluated in relation to child eating behavior: parental BMI and parental habit to practice physical activity.

No association between fathers' BMI and child snacking consumption has been found, while the association between mothers' BMI and child snacking consumption observed in the univariate analysis disappears in multivariable analysis.

On the other hand, in our study children dietary habits seems to be strictly related to mothers' attitude toward physical activity. Indeed, mothers of no-snackers seem to be more prone to healthy behaviors, practicing at least 2 hours of sport activities during the week. But this not necessarily translates in a healthier lifestyle for their children, since no-snackers children tend have a more sedentary lifestyle.

On the contrary, snacking children profile is controversial, since, in according to the literature, they spend more time than no-snackers watching TV, but also, they seem to be more active, since the great part of them practices at least 4 hours per week of physical activity. Differences in family composition could explain this fact: while no-snacking children usually may come from classical families (father, mother and children) and may have not brothers or sisters, snacking children frequently come from numerous, enlarged families. Classical families having one child, may pay more attention on their children lifestyle and on their dietary habits, whereas, in an enlarged familiar environment, parents may have to delegate other people or relatives, such as grandparents, to take care of their children, ending up in a less strict control of their children's dietary habits.

Differently from what described in previous studies on snacking activities 22 , in which snacking children are largely described as children usually spending time in solitary activities, in our analysis snacking children, having a more active lifestyle than no-snackers, are part of a more numerous social networks. Thus, instead of observing the well-studied vicious circle in which more isolated children stay at home, spending time in sedentary activity and eating snacks 22 , we observed that snacking children are practicing physical activity more than no-snackers and are part of a larger social network. Indeed, in the snackers' social network, most of peers are both practicing sports and eating snacks. The attitude to "regress to the mean" in term of behaviors among peers, with the tendency of mimicking others' behaviors 23 could also contribute to explain these findings.

On the contrary, as a direct consequence of a more sedentary lifestyle, no-snackers children have less social relationships and, similarly to that observed among snackers, their peers tend to share the same eating and leisure time habits.

The existence of an association between snack consumption and risk of being overweight or obese has been previously evaluated in numerous studies obtaining contrasting results: is was shown in some studies but not in others 9 .

In our study we found that the status of snacker seems not to be associated with overweight and/or obesity being the BMI equivalent among snacking and not snacking children. The key factor in explaining such result might be the role of physical activity of the child. In addition to the positive effect of physical activity, by itself, on obesity prevention, also spending leisure time in sportive activities implies being part of a social environment which is definitely a positive one from the point of view of obesity control 24 . Indeed, results do not show an increase of overweight/obesity in relation to snack consumption.

In the present study, determinants of BMI in snacking group have been explored; according to previous large-scale researches which found correlations between paternal or maternal anthropometric measures and children's adiposity 25 26 27 , also in our analysis an increment in mother BMI seems to contribute to an increment in children BMI. As previously mentioned, this intergenerational relationship reflects genetic, social, cultural, and environmental components since, when parents make food consumption choices (or provide the feasible food consumption set) for children for meals eaten at home, this common environment generates a correlation between generations when both eat unhealthy (or healthy) food together 28 .

Moreover, a positive relationship between maternal age and children BMI has also been found, suggesting that younger mothers tend to pay more attention to their children's dietary habits or to encourage them to practice sport.

As expected, energy balance-related behaviors, including sport activity, studying and watching TV are related to children BMI: an increment in the first one induces a BMI reduction, while, more time spent in sedentary activity determines a BMI increasing, as expected.

Finally, as observed elsewhere 29 , a positive relationship between low income (indirectly measured here by the municipality income) and body weight seems to exist: the association of child and household food insecurity with childhood overweight has been demonstrated in several studies 30 ; moreover, children in low-income families could experience adverse psychosocial conditions which may contribute to poor eating habits and lower physical activity levels 31 .

This study is a survey, without direct interaction between the investigator and the interviewed people and thus relying only on self-reported measurements for what concerns anthropometric characteristics, physical activity levels and dietary patterns. Although this approach has been widely used 32 , it still presents the general limitation of such approaches, like reporting bias, barrier effect leading to interview refusal, tendency of ameliorant responses in presence of sensitive questions 31 . The authors used interview and analysis techniques to reduce each of such potential biases, like the involvement of only highly experienced interviewers in interview conduction and multivariable modeling for confounding adjustment.

No evidence of an association between snacking consumption and overweight status has been shown by our study. This fact can be explained by the different attitude toward physical activity observed in snackers and no-snackers: in fact, differently from common findings in scientific literature, in our study the snacking child can be described as practicing sport more than no-snakers and, as a consequence, having more active peer-to-peer social relationships, mostly related with sport activities. Among snackers, factors traditionally related to body weight such as mother BMI, energy balance-related behaviors and municipality income have been confirmed as major determinants of children BMI.