more emotion dysregulation in toddlers 1 year later (NICHD Early Child Care Research Network, 2004). Given that both emotional over- and undereating can be seen as responses to distress, it is reasonable to assume that children of parents who are less sen- sitive will cope less effectively with stress and thus display more of both eating behaviors. In addition to sensitivity, parental structuring may affect the development of emotional over- and undereating. Parental structuring may provide boundaries and rules about when and what to eat for their children. Thus, children of parents who structure adequately may simply have fewer opportunities for engaging in emotional overeating. Although scarce, the existing empirical evidence on parenting and emotional over- and undereating provides some support for these claims. Parental support can be considered a variation of parental responsiveness and sensitivity (Morris et al., 2007), and cross-sectional research shows that parental warmth and support are associated with less emo- tional overeating in school-aged children (Topham et al., 2011), whereas adolescents who lack parental support are inclined to engage in emotional overeat- ing (Snoek et al., 2007). It is also notable that women suffering from anorexia nervosa (retrospectively) report both a lack of parental support and emotional undereating in childhood (Kim et al., 2010). High levels of both parental sensitivity and structuring are crucial to the authoritative parenting style (Rhee, 2008), which is significantly related to less emotional overeating in children (Topham et al., 2011). Finally, one study showed that parental structure was corre- lated with less emotional overeating in school-aged children (Philips, Sioen, Michels, Sleddens, & De Henauw, 2014). To our knowledge, no prior study has investigated whether parental sensitivity and structuring specifically predict changes in emotional over- and undereating over time an assumption tested in the present inquiry. Interaction Effects Between Temperament and Contextual Factors? Throughout the preceding discussion, the impli- cation has been that children in general are affected by the conditions hypothesized to be associated with emotional over- and undereating. Yet, there is evidence that children vary in their susceptibility to a diverse array of environmental influences and that it may be children with more, rather than less, negative affectivity who are especially susceptible to environmental influences (Belsky & Pluess, 2009). Child factors that pose a risk (or benefit) may vary M in their impact depending on the quality of family functioning or parenting. Thus, high family func- tioning and sensitive or structuring parenting may buffer against the impact of other risk factors (Mas- ten & Shaffer, 2006). This line of reasoning suggests that low soothability in children may not necessar- ily pose a risk in instances of positive contextual factors (e.g., adequate family functioning). In other instances, such contextual factors may exacerbate the impact of alleged risk factors; highly emotion- ally reactive children who also live in a negative family environment may, therefore, be at the great- est risk for developing emotional dysregulation (Morris et al., 2007). There is also evidence that vul- nerabilities in emotion regulation and depression in 9- to 10-year-olds were indeed moderated by char- acteristics of the family environment (Feng et al., 2009). Following the reasoning outlined earlier, serious life events, poor family functioning, and less-sensitive and less-structured parenting might particularly promote emotional over- and undereat- ing in children who are difficult to soothe or have symptoms of depression-a hypothesis tested in the present inquiry. The Current Study In seeking to provide a better understanding of the etiology of emotional over- and undereating in childhood, we test an ecolgical model of child and contextual influences, drawing on a large, represen- tative sample of Norwegian children followed bien- nially from 6 to 10 years of age. We evaluate whether child temperament, symptoms of depres- sion, serious life events, family functioning, parental sensitivity, and structuring predict changes in chil- dren's emotional over- and undereating. Given evi- dence that emotional over- and undereating have some common etiology (Herle, Fildes, Steinsbekk, et al., 2017), but that neither theory nor research identifies specific factors related to one but not the other form of emotional eating, we do not make specific predictions with regard to distinctive deter- minants of over- and undereating. We hypothesize that both eating behaviors will increase over time when children score low on soothability, have more rather than fewer symptoms of depression, have experienced serious life events, are growing up in less well-functioning families, and have parents who are less sensitive and less structuring. We fur- ther hypothesize that the latter four contextual con- ditions will prove more influential among children who are difficult to soothe and children with symp- toms of depression. Temperament was measured using the short form of the Children's Behavior Questionnaire (CBQ-SF; Putnam & Rothbart, 2006) when the chil- dren were 6 years old. The falling reactivity/ soothability subscale (i.e., how quickly the child recovers from emotional arousal, including both distress and excitement) includes items such as "When angry about something, s/he tends to stay upset for 10 min or longer" and "Is easy to soothe when s/he is upset," and responses are rated from 1 ("extremely untrue of your child") to 7 ("ex- tremely true of your child"). To improve the inter- nal consistency of the original subscale (six items), one item was deleted ("Changes from being upset to feeling much better within a few minutes), resulting in a = .68, which is comparable to the internal reliability originally reported by the devel- opers of the short form of the CBQ (Putnam,& Rothbart, 2006). Because we expected soothability specifically, to predict emotional over- and undereating, a more detailed description of the four other subscales of the negative affectivity dimension (i.e., anger/frustration, discomfort, fear, and sad- ness) can be found in the note to Table 4. Anger/ frustration, discomfort, and fear had acceptable reli- abilities (x = .82; = .74; = .72, respectively), whereas the sadness subscale had an of 49, and it was not possible to improve this scale by remov- ing any items. It was nevertheless included because it is a component of negative affectivity. Depression Symptoms and Serious Life Events Depression symptoms and serious life events were recorded by means of the Preschool Age Psy- chiatric Assessment (PAPA; Egger et al., 2006) when the children were 6 years old. PAPA, which is a semistructured psychiatric interview of parents, was administrated by trained personnel using a structured protocol including both required and optional follow-up questions. The PAPA assesses depression according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (American Psychiatric Association, 1994), and a variable for depression was created using symptom counts (i.e., a continuous variable). Blinded raters recoded a random sample of 9% of the interview audio recordings, and the resulting interrater reliability (ICC: Intraclass Correlation Coefficient) was .90. To assess serious life events, parents were asked if their child had ever experienced any of 25 stressors that could potentially cause posttraumatic stress or death, or enduring physical or sexual abuse). The number of events experienced was summed to create a serious life events score. Family Functioning Family functioning was reported by parents using the general family functioning scale of the McMaster family assessment device (Epstein et al., 1983) at 6 years of age. This scale covers different aspects of the family (e.g., problem solving, com- munication, involvement, affective responsiveness) and comprises 12 items (e.g., "In times of crisis we can turn to each other for support"; "We can express feelings to each other"; = .90) rated on a 4-point scale from strongly agree to strongly disagree. Thus, a higher score indicates lower family function- ing. The general family functioning scale has previ- ously been shown to have high validity and test- retest reliability (Epstein et al., 1983). Parental Sensitivity and Structuring Parental sensitivity and structuring were mea- sured with the Emotional Availability Scale (EAS; Biringen, 2008). Sensitivity concerns a parent's abil- ity to develop a healthy emotional connection with her or his child, whereas structuring refers to ade- quately guiding the child's activities and setting appropriate limits for the child. When the children were 6 years old, parent-child interactions were videotaped during four successive 5-min sequences (free play, child-led play, parent-led play, and a clean-up task), and parental sensitivity and struc- turing were rated based on these observations. All raters were trained and certified as reliable by Biringen, one of the developers of the EAS. The interrater reliabilities (ICCs) among blinded raters of a random sample of 10% of the videotapes were .71 for sensitivity and .70 for structuring. Results All analyses were performed in Mplus version 7.4 (Muthn & Muthn, 1998-2015). Due to the screen- stratified sample in this study, all analyses were con- ducted with probability weights to produce accurate population estimates. The probability weights corre- sponded to the number of children in the population in a specific stratum divided by the number of par- ticipating children in that stratum. To provide robust standard errors, we used a robust maximum OMBO SAMSUNG Method Participants and Procedure Children born in 2003 and 2004 and their parents living in Trondheim, Norway, were invited to par- ticipate in the Trondheim Early Secure Study (TESS) by a letter sent to their homes (N = 3,456). The letter included the Strengths and Difficulties Questionnaire (SDQ) version 4-16 (Goodman, Ford, Simmons, Gat- ward, & Meltzer, 2000). When attending the ordi- nary health checkup for 4-year-olds, the parents brought the completed SDQ form. Almost all chil- dren in the two birth cohorts appeared at the health checkup (97.2%, 11 = 3,358); therefore, the sample is in effect a community sample. A health nurse informed about the study and asked the parents to participate (n = 3,016), and 2,475 gave written informed consent. The health nurse missed asking a total of 166 parents, who were thus excluded. Fur- ther, 176 were not eligible because they were not proficient in Norwegian. In order to increase sample variability and thus statistical power, children with higher scores on the SDQ were oversampled. This oversampling was accomplished by dividing SDQ total difficulty scores into four strata (cutoffs: 04, 5 8, 911, and 1240), and the probability for selection increased with higher SDQ scores (.37, .48, .70, and .89 in the four strata, respectively). In all, 1,250 con- senting families were drawn to participate. The Regional Committee for Medical and Health Research Ethics, Mid-Norway, approved the study. The child and one of the parents visited the uni- versity clinic for testing and observation about 2 weeks after the health checkup, and follow-up was on a biennial basis. Children's eating behavior was assessed from age 6 onwards. Therefore, the present study is based on data collected at age 6 (n = 797, Mage = 6.72 years, SD = .17), age 8 (n = 699, Mage = 8.80 years, SD = .24), and age 10 (n = 702, Mage = 10.51 years, SD = .17). The data collection took place from August 2009 to June 2011 (age 6), from August 2011 to June 2013 (age 8), and from August 2013 to June 2015 (age 10). Sample characteristics are presented in Table 1. Attrition was selective according to predictors but not according to emotional over- and undereat- ing at ages 6 and 8. Attrition at age 8 was higher among those with less-sensitive parents (OR = 0.67 [95% CI = 0.46, 0.96], p = .031) and those who had experienced serious life events at age 6 (OR = 1.27 [95% CI = 1.09, 1.48], p = .002). However, the com- bined effect of these variables was small (Nagelk- erke proxy R R = .040, Cox & Snell .023). Attrition at age 10 was higher among children who had OMBO Table 1 Sample Characteristics at Age 6 Gender of child Female Male Gender of parent informant Female Male % 50.2 49.8 81.1 18.9 Ethnic origin of biological mother Norwegian 93.0 Western countries 6.8 0.3 Other countries Ethnic origin of biological father Norwegian Western countries Other countries 93.0 6.5 0.5 experienced serious life events (OR = 1.38 [95% CI = 1.17, 1.62], p < .001) and children with higher levels of depression symptoms at age 6 (OR = 1.29 [95% CI = 1.06, 1.58], p = .013). Again, the com- bined effect was small (Nagelkerke proxy R = .045, .025). Cox & Snell Measures Emotional Over- and Undereating The emotional overeating scale (e.g., "My child eats more when anxious") and the emotional undereating scale (e.g., "My child eats less when upset") of the parent-reported Children's Eating Behaviour Questionnaire (CEBQ; Wardle et al., 2001) were used to assess children's emotional over- and undereating when they were 6, 8 and 10 years old. The responses were rated on a 5-point scale from never to always. The internal consistency of the emotional overeating scale (four items originally) was improved by removing one item ("My child eats more when s/he has nothing to do"), resulting in & = .81, .80, and .79 at age 6, 8, and 10, respec- tively, in the present sample. This specific item was also originally shown to load poorly on the emo- tional overeating scale, but it was provisionally retained by the developers to provide better compa- rability with other scales (Wardle et al., 2001). The emotional undereating scale (four items) had acceptable reliability (age 6: x = 75; age 8: = .76; age 10: x=78). The CEBQ has shown good test- retest reliability (Wardle et al., 2001) and has been validated against eating behavior observed in labo- ratory contexts (Carnell & Wardle, 2007). SAMSONS