First, children's BMI scores were transformed into BMI z-scores using the program developed by the Centers for Disease Control and Prevention. The program specifies the measurement procedures, reference population, and age- and sex-specific cut-off points of an internationally accepted definition of child overweight and obesity. Second, descriptive statistics of the samples were calculated in terms of means and standard deviations or frequencies and percentages. Third, to assess the differences between child's BMI categories (normal weight vs. obese) concerning the measured variables, a number of independent sample t-test was performed. Child's BMI category (normal weight vs. obese) was inserted as an independent variable, whereas parents’ BMI, SES, EOQ scores and PRFQ scores were used as dependent variables. Effect size was measured using Cohen's d, in which benchmarks are small (d ≥ .2), medium (d ≥ .5), and large (d ≥ .8) (Cohen, 1988). Fourth, based on the aforementioned studies, as well as the significant differences shown by t-test(s), hierarchical multiple regression was performed to investigate the role of parents’ BMI, SES and PRF on child's BMI. Because t-test(s) showed no differences in parents’ EOQ scores between the two groups, parents’ EOQ scores were not inserted as a variable in the hierarchical multiple regression. Instead, even though mother's BMI was not significantly different between the two groups (p ≥ .05), it was inserted in regression model because previous studies indicated that it has a crucial role on child's weigth, even more important than father's BMI. Also, it was necessary to control its effect in the relation between child's BMI and the other variables (i.e., father's BMI, SES, PRFQ). Finally, although p-value related to mothers’ BMI was not significant, effect size was almost medium. Therefore, parents’ BMI and SES were inserted first, given that their association with child's BMI has been previously demonstrated. PRFQ variables were inserted last, because they were taken into account for the first time. Our interest was to determine whether these newly added variables showed a significant improvement in R2, namely in explaining variance of child's BMI. Five different models were run. Mother's BMI was the first variable entered (Model 1), followed by father's BMI (Model 2), then SES (Model 3), the significant subscales on mothers’ PRFQ assessments (Model 4) and the significant subscales on fathers’ PRFQ assessments (Model 5). With regard to child's BMI, in order to utilize a continuous dependent variable gender and age weighted, child's BMI z-score was used in regression analysis. It was inserted as the dependent variable. The Akaike information criterion (AIC) of the five models, as well as ANOVA among them, were calculated to find the best model (as indicated by significant variation in ΔR2 and the lowest AIC). The R package (R Development Core Team 2012) and PASW Statistics 18 (SPSS Inc., 2009) were used for the analyses.

As shown in Table 1, overall, significant differences with medium-large effect sizes were found between the two groups on parents’ BMI, familial SES and the CMS subscale of the PRFQ. No differences in EOQ scores or in the PM and IC subscales of the PRFQ were found.

More specifically, the BMI of fathers and mothers of children with obesity was higher than that of parents of normal weight children. Although p-value referred to mothers’ BMI was not significant, effect size was almost medium. Familial SES and the CMS subscale of the PRFQ among parents of children with obesity were higher than those of parents of normal-weight children.

Table 2 shows the results of the five hierarchical models. Hierarchical multiple regression suggested that Model 4 was the best model because it showed a lower AIC than did other models (namely, the best trade-off between the goodness of fit of the model and the simplicity of the model), as well as a significant variation in R2(ΔR2) from Model 3 (namely, a significant increase in the variance of children's BMI explained by Model 4). Model 4 was significant (F(4,55) = 6.86, p < .001) and explained over 34% of the variance (R2 = .345) in children's BMI z-score. Although mothers’ BMI was a significant predictor of children's weight in Model 1, fathers’ BMI was a stronger predictor in Model 2. The higher the parents’ BMI, the higher the children's BMI z-score. Model 3 confirmed the relationship between BMI and SES. The higher the SES is, the lower the child's BMI z-score. In Model 4, mothers’ PRF became a significant predictor of children's weight, above and beyond the role of parents’ BMI and SES. Children's weight increased approximately .3 z-score for each unit increase in mothers' CMS. SES predicted a decrease of -.05 z-score in children's weight. Model 5, which included fathers’ PRF, did not add a significant increase in explained variance in children's BMI z-scores.

The present research aimed to explore whether childhood obesity was influenced in part by low PRF. A comparison of mothers and fathers of children with and without obesity showed that parents of children with obesity had higher BMI, lower SES and lower PRF than those of normal-weight children. Moreover, among the different hierarchical models run, the model that included parents’ weight and SES and mothers’ PRF was the best, as it was significant and explained over 34% of the variation in children's weight.

With regard to parents’ weight, these data confirm those of previous studies showing that higher parents’ weight predicted higher children's weight (Ejtahed et al., 2018; Linabery et al., 2013). In contrast to earlier research, paternal weight was a stronger predictor of children's weight than was maternal weight. It must be taken into account that few studies consider the key role played by both parents, as attention has been primarily given to maternal factors, and fathers are unequivocally underrepresented in studies on parenting and childhood obesity (Davison et al., 2016). Therefore, future studies should further explore the role of both parents’ weight as risk factors in childhood obesity. However, in our research, the significant difference in parents’ weight could not be caused by their emotional eating, as no difference emerged between the parents of the two groups.

With regard to familial SES, our results confirmed those of previous studies showing that children from low socio-economic families in industrialized countries are at higher risk of being obese (Lissner et al., 2016). In general, while children from all SES groups are considered vulnerable to the development of obesity, children whose parents have low SES appear to have a higher than average risk.

Referring PRF, parents’ CMS subscale scores were higher in the children with obesity group than that of normal-weight group. No differences in the PM and IC subscales of the PRFQ were found. Furthermore, mothers’ CMS was significantly related to children's BMI, above and beyond parents’ BMI and SES.

The CMS subscale assesses parents’ inability to recognize that children's mental states are not transparent, and it has been shown in parental accounts of their children's behaviours and attitudes that go far beyond the data; these accounts are characterized by undue certainty about the mental states of the child and are often distorted, intrusive and sometimes paranoid (Luyten et al., 2017).

Therefore, a greater maternal inability to recognize the opaqueness of children's mental states predicted greater children's weight. It seemed that mothers’ inability to understand that they have a limited ability to truly know what is in their children's minds and to tolerate the uncertainty of not knowing why their children are behaving in a certain way, which involves mothers’ awareness that their children's mental state is separate from her (Ordway, Sadler, Dixon, & Slade, 2014), constitutes a risk factor in the onset of childhood obesity.

The absence of a relationship between the interest and curiosity, and non-mentalizing subscales is also of interest suggesting that these modes of parental mentalization may be differentially related to, or independent from, the risk of onset of obesity in childhood. It may be that these dimensions of mentalization are less relevant to the specific characteristics of parenting in childhood obesity than the dimension of certainty about mental states. Parents of children with and without obesity seem not to differ both in their capacity to keep their child in mind (Interest and curiosity), and in difficulties to identify and understand their children mental states (Pre-mentalizing modes). A previous study found out that pre-mentalizing was associated with low levels of maternal emotional availability, except to maternal intrusiveness, while high level of certainty seemed to be associated with intrusiveness, as the parent assumed to know everything about their child's mental states (Luyten et al., 2017); hence, in the present study the specific combination of PRFQ subscales seemed to suggest that mothers of children with obesity show not general difficulties in emotional availability, but greater intrusiveness and difficulties in understanding that they have a limited ability to truly know what there is in their child's mind. Interestingly, this finding seems to be also consistent with clinical observations of the attitudes of mothers of children with obesity to be overprotective and to superimpose their own needs on those of their child, leading to a mismatch in the interaction between the dyad and difficulties in children's ability to learn to distinguish hunger from other internal or affective states (Bruch, 1941).

As the CMS subscale measures the extent to which the parent is unable to recognize that children's feelings, thoughts, and intentions are not always readily apparent, high certainty could consequently affect mothers’ capacity to mirror children's mental states and, therefore, to regulate his/her affects (Luyten et al., 2017). Because parents impact children's ability to learn to self-regulate and develop strategies for managing increasing levels of arousal, such as negative affects and stress responses, and because PRF is considered to serve as a modulating function once the parent-child relationship has been dysregulated, acting as a buffer against breakdowns in affect regulation during times of stress (Grienenberger et al., 2005), our results seemed to substantiate the issue of mothers’ dysregulated emotional response in childhood obesity. Future studies should investigate the supposed mediation role played by mothers’ emotion regulation strategies in the relationship between high maternal certainty and high child weight.

To date, only one study has investigated the potential role played by maternal reflective function in childhood obesity. In contrast to the hypotheses based on the literature, Keitel-Korndörfer et al. (2016) found that the general reflective functioning of 60 mothers with and without obese children aged 18 to 55 months was not directly associated with the child's weight; rather, general reflective functioning indirectly affected the child's BMI percentile via the level of security in the mother-child relationship. The research differed from our exploratory study in several respects. First, the sample selection criteria were mothers’ weight and not children's weight. Second, the sample was composed of pre-schoolers, in contrast to our sample composed of children in early middle childhood, a developmental period in which regulation issues are salient in both emotional and eating domains (Harrist, Hubbs-Tait, Topham, Shriver, & Page, 2013). Finally, maternal general capacity, rather than her PRF, was assessed to reflect on mental states. More recently, studies have paid greater attention to reflective functioning as being relationship-specific, suggesting that the capacity to reflect on a specific relationship with a significant other could differ from more general reflective processes (Sharp & Fonagy, 2008).

Overall, our findings showed that the combination of high maternal CMS, high parental BMI and low familial SES represents a risk factor for childhood obesity. It seemed that to grow up in a socio-economically disadvantaged environment in an industrialized country, with parents overly focused on eating and with mothers being too certain of what was in their children's minds, constituted an environment where children have a higher risk of becoming obese. Future research on childhood obesity is needed to better comprehend these issues.

With regard to the dissimilarities between mothers and fathers that emerged in our study, children's weight was affected mainly by maternal PRF and paternal weight. Differences in PRF capabilities related to parental gender have already arisen in a previous study (Pajulo et al., 2018; Pazzagli et al., 2018) and may be the manifestation of a difference between mothers and fathers with respect to how they engage in reflective interactions with their children (Arnott & Meins, 2007). This finding is also consistent with that of a previous study on youngsters with obesity that showed more maternal than paternal influence on children's emotion regulation (Vandewalle, Moens, & Braet, 2014). It seemed that in this socio-economically disadvantaged environment, mothers’ lack of capacity to give meaning to children's behaviour and to regulate negative emotions and distress had a strong influence on children's weight. In contrast, fathers played a more crucial role than mothers in constructing their children's health behaviours, as their weight was a stronger predictor of children's weight.

The study is exploratory in nature, and the findings need to be replicated before firm conclusions can be drawn. Future research should confirm the observed effects of PRF on childhood obesity and the supposedly mediating role of maternal emotion regulation strategies on the relationship between PRF and children's weight with a larger sample. The current study had other limitations, including the fact that only self-report measures were used. The use of observational methods would be more desirable, and these approaches should be integrated into future studies. As the study is cross-sectional in nature, it does not allow inferring causal relationship. It would be fruitful for future research to utilize longitudinal data to more definitely predict associations between the selected variables. Furthermore, the study aimed to focus on the role of PRF beyond parents’ weight and familial SES, but many other factors of the family environment might influence children’ obesity. More clarification is needed regarding the risk factors associated with childhood obesity and the relationships among different predictors of children's weight. To our knowledge, this is the first study aiming to explore the contribution of PRF to the prediction of children's weight above and beyond the contribution of mothers’ and fathers’ weight and SES. The results suggested that low maternal PRF could be an important contributing factor to children's weight, thus highlighting the need to take into account parental responses to children's emotions in the assessment and treatment of childhood obesity. As a deeper understanding of the multi-factorial contributors to obesity is of utmost importance to aid in the development of effective interventions aimed at reducing the rates of global obesity, the current study provides preliminary support for the perspective that childhood obesity prevention and treatment programmes expand beyond a focus on children's behavioural regulation, and also encompass the role of PRF in childhood obesity. As noted by Harrist et al. (2013), interventions that target emotion regulation strategies could be a useful new approach and would have the potential to impact eating behaviours and prevent weight issues among children.