Participants and procedures
The study sample includes 301 child-mother dyads recruited through Head Start programs in South Central Michigan. Head Start is a federally funded preschool program for low-income, high-risk families in the United States (US). Participants were drawn from a longitudinal cohort initiated in 2009 to investigate associations between stress and eating among low-income children. Children described in this study were between the ages of 4 and 8 years at the time of data collection. Inclusion criteria were: caregiver is fluent in English and does not have a college degree; and child is not in foster care, has no serious medical problems or history of food allergies and was born at ≥ 35 weeks gestation without significant perinatal or neonatal complications. For this analysis we only included children who were living with their biological mothers (as this represents the majority of this sample), who were living with only one sibling, and who had complete data on all variables (n = 102). Of those 102 children, we only included index children whose siblings were at least a year old (n = 86) on the premise that the processes via which infants may influence eating behavior of siblings could be fundamentally different. Mothers provided written informed consent for themselves and for their children. The University of Michigan Institutional Review Board approved this study.
During two study visits, mothers completed questionnaires, and trained staff members obtained child anthropometry. Three videotaped home mealtime observations were completed for each family. Each mother was asked to record three routine evening meals within a single week. Research assistants called each mother after the meal to obtain information regarding individuals present. These family mealtime observations (FMOs) followed standard procedures that have been previously described [25].
For the present study, inclusion criteria for the FMO videotape included that the index child (IC) was eating with his/her sibling, and that the IC was not eating with other children in addition to the sibling. We systematically selected one of the three FMO videos for each IC. We started video selection with the second FMO video on the premise that we would expect families to be more acclimated to the camera by the second home observation. If the second FMO video did not meet the inclusion criteria, we then assessed the third FMO video; if the third FMO video did not meet inclusion criteria, we assessed the first FMO video. After assessment of the FMO videos for each IC, a final sample of 75 index children was identified (8 from the first FMO, 55 from the second FMO, and 12 from the third FMO). The sample included in this analysis (n = 75) did not differ from the sample not included (n = 226) with regard to child sex, child race/ethnicity, birth weight z-score, and maternal age.
Measures
Demographic characteristics
Mothers reported information regarding IC’s birthdate, sex, and race/ethnicity (dichotomized for this report as non-Hispanic white vs. not) and mother’s birthdate and years of education (dichotomized as more than or equal to a high school education vs. not). Birthdates and dates of visits were used to calculate child and maternal age.
Sibling characteristics and birth order
For each individual living in the household, as well as for each individual on the FMO videotapes, mothers reported age, sex, and relationship to the IC. This information was used to determine the IC’s birth order (i.e., younger sibling vs. older sibling) and characteristics of the siblings.
Coding of interactions between index child and sibling
To evaluate mealtime sibling behaviors that may be most relevant to child obesity risk, we developed a coding scheme based on Bob and Tom’s Method of Assessing Nutrition (BATMAN) [10]. The BATMAN is an observational assessment used to evaluate parental behavior around food [10]. Although restrictive feeding behaviors are part of the BATMAN, we did not code these behaviors as they were not observed to occur between siblings with meaningful frequency. Although the BATMAN distinguishes between physical and verbal encouragements to eat, we did not observe frequent physical encouragements to eat between siblings and therefore focused our coding scheme on verbal encouragements to eat. The BATMAN defines verbal encouragements to eat as suggesting, demanding, directing, and making positive statements about food. We adapted some of the operational definitions to be consistent with theoretically important features of sibling interactions (i.e., parent-like interactions or “complementarity” and peer-like interactions or “reciprocity”) [12]. For example, food offers (representing complementarity) and statements about eating/finishing the food (representing reciprocity) were counted as verbal encouragements to eat.
Encouragements to eat delivered by the sibling and directed to the IC were coded in 5-min intervals from the videos. Ten percent of the videos were double coded and inter-rater reliability by intraclass correlation coefficient exceeded 0.80. Number of encouragements was summed across intervals to create the variable “total encouragements delivered to IC by sibling”.
Mealtime maternal presence
Siblings interact differently when their mother is present [26, 27]. In order to adjust for maternal presence, we coded whether the mother was sitting with the siblings during the meal in each 5-min interval (yes vs. no for each interval). Inter-rater reliability computed as Cohen’s kappa was 1.00. We created the variable “proportion of intervals in which mother is present” by dividing the total number of intervals in which the mother was sitting with the siblings by the total number of intervals.
Anthropometry
Staff members measured index children’s weight and height during study visits using standardized procedures. BMI was calculated and age and sex specific BMI z-score (BMIz) for the IC was calculated based on the US Centers for Disease Control and Prevention reference growth curves [28]. Mothers reported the IC’s birth weight, which was converted to z-scores based on National Centers for Health Statistics Natality Datasets [29]. Birth weight z-scores were missing and were imputed for 26 subjects using multiple imputation.
Statistical analysis
We conducted statistical analysis using Stata version 13 (StataCorp. 2013. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP). First, we calculated descriptive statistics for sample characteristics. Then, to test our hypothesis that encouragements to eat from the sibling is a mediating variable in the association of IC’s birth order and the sibling’s sex with IC’s BMIz, we conducted path analysis, which is an extension of the regression model comprised only of directly observed variables [30]. We ran our path model testing associations between IC’s birth order, the sibling’s sex, encouragements to eat directed to the IC from the sibling, and IC’s BMIz. We included the binary variables IC’s birth order (with “older sibling” as the reference category) and sibling’s sex (with “male” as the reference category) as predictors in the model. A Poisson distribution was used to model the mediating count variable “total encouragements delivered to IC from sibling”, and “number of meal intervals” was set as the offset variable to account for variations in length of the meal. The model was adjusted for maternal presence (i.e., proportion of intervals in which mother is present), sibling’s age, and the IC’s birth weight z-score. For all statistical analyses, significance level was set at 0.05.