Data source and sample
Survey data from FLASHE, a cross-sectional, Internet-based study conducted from April to October 2014 and sponsored by the National Cancer Institute (NCI), were used for these analyses [19]. An online consumer opinion panel was used to recruit eligible parent-child dyads and surveys were administered via the web. Eligibility criteria included: at least 18 years old; at least one child aged 12–17 years living at least 50% of the time in the household; agreed to be contacted for study participation. One eligible child was randomly selected from eligible households. The household sample was created using balanced sampling and is similar to the general US population for sex, income, age, household size, and region [20]. A total of 1945 dyads (parent-caregiver and child) were enrolled. Study participation involved completion of six web surveys, three by the parent and three by the child. FLASHE was approved by the United States (US) Government’s Office of Management and Budget, the NCI Special Studies Institutional Review Board, and Westat’s Institutional Review Board. Additional details on study methods have been published elsewhere [20].
Demographic and anthropometric characteristics
For analytic purposes, child age was categorized into two groups, representing early adolescence (12–14 years) and middle adolescence (15–17 years). For parents and children, race/ethnicity were ascertained with two questions that were combined to create four categories – Hispanic, non-Hispanic black or African American only, non-Hispanic white only, and non-Hispanic other. Responses for parents’ education level included less than a high school degree, a high school degree or General Educational Development (GED) certification, some college but not a college degree, and a 4-year college degree or higher. Responses for parents’ marital status included married, divorced, widowed, separated, never married, and member of an unmarried couple. Divorced, widowed, and separated were combined into a single category in the public use dataset. Responses for household income (parent survey) included nine options ranging from $0–$9999 to $200,000 or more that were dichotomized to $0–$99,999 and $100,000 or more in the public use dataset. Using parent and child self-reported values, BMI was calculated as weight (kg) divided by height (m2). For parents, body weight was classified as underweight (BMI < 18.5), healthy weight (18.5 ≤ BMI < 25), overweight (25 ≤ BMI < 30), and obesity (BMI ≥ 30). For children, body weight was classified based on Centers for Disease Control and Prevention’s sex-specific 2000 BMI-for-age growth charts as underweight (BMI < 5th percentile), healthy weight (5th percentile ≤BMI < 85th percentile), overweight (85th percentile ≤BMI < 95th percentile), and obesity (BMI ≥ 95th percentile). For analytic purposes, body weight categories were collapsed to underweight/healthy weight and overweight/obesity.
Dietary intake
The FLASHE dietary screener (27 items) was used to capture parent and child dietary intake frequency for foods and beverages that have remained of interest in dietary guidance in the US [21]. Complete screener wording can be found on the FLASHE website [19]. Response options, based on the past 7 days, included no consumption, 1–3 times/week, 4–6 times/week, 1 time/day, 2 times/day, and ≥ 3 times/day. The responses 1–3 times/week and 4–6 times/week were converted to daily intake frequencies by dividing the median by 7 resulting in values of 0.29 an 0.71 times/day, respectively. The response ≥3 times/day was coded as 3 times/day. Daily frequencies of each food and beverage group were calculated by summing individual item frequencies based on groupings used in previous studies [22]. These groupings included fruits, vegetables, fruits and vegetables combined (with and without fried potatoes), dairy, added sugars (total and sugary drinks only), and whole grains. Specific foods and beverages related to added sugars intake included sugary cereal, candy and chocolate, chips, cookies, cake, frozen desserts, sweetened fruit drinks, regular soda, sports drinks, and energy drinks. Using programs developed to compare responses from the National Health and Nutrition Examination Survey dietary screener (similar to the FLASHE dietary screener) with the What We Eat in America 24-h dietary recall data [21], daily teaspoons of added sugars were estimated from daily frequencies [23]. For the current study, dyads were included in the analyses if they had non-missing data on parent and child daily intakes (teaspoons) of added sugars.
Junk food and sugary drinks (JS) parenting practices and legitimacy of parental authority
Six JS parenting practices were measured with one item each and represented two types of coercive control practices – negative emotions (allow JS when had bad day) and restriction (parent decides JS amount); three types of structure practices – monitoring (do not eat too much JS), availability (do not buy JS), and modeling (avoid eating JS when child around); and one type of autonomy support practice – child involvement (decide together JS amount). Additionally, legitimacy of parental authority regarding JS (JS-LPA) was measured with one item (okay to make rules about JS). On the surveys, junk foods were defined as foods that are high in calories and usually have added sugars and fat and include candy, cookies, potato chips, French fries, etc. Sugary drinks were defined as regular soda, sports drinks, fruit drinks, sweetened teas, and other drinks with added sugars. The items were taken or modified from valid and reliable instruments using cognitive testing; source information and full survey wording can be found on the FLAHSE website [19] and in Supplementary Table 1, Additional file 1. Items were included on parent and child diet surveys and responses ranged from strongly disagree (1) to strongly agree (5). For analytic purposes, responses were dichotomized as strongly disagree to neither agree nor disagree (1–3) and agree to strongly agree (4–5).
Statistical analysis
Statistical analyses were performed using SAS® software, version 9.4 (SAS Institute Inc., Cary, NC). The statistical significance level was set at 0.05. Dietary survey weights for parent and child cohorts were used for computing descriptive statistics for demographic characteristics, BMI, added sugars intake, JS parenting practices, and JS-LPA. Sample weights were not used for correlation analysis or latent class analysis because no dyadic analysis weights were provided (population control totals cannot be easily defined) [24]. Each parent was linked to only one child and child was linked to only one parent. Parents and children were asked the same questions to facilitate dyadic analysis. Dyadic analysis was performed using the parent-child dyad identifier. Comparisons between the analytic and excluded dyads on demographic characteristics were performed using chi square tests of association. To confirm associations among JS parenting practices, Spearman rank correlation coefficients (rs) were computed because variables were measured on an ordinal scale and some distributions were skewed. Correlation coefficients’ strength was based upon Cohen’s recommendations (weak < 0.30, moderate =0.30–0.49, and strong ≥0.50) [25].
To identify subtypes of parent-child dyads that exhibited similar patterns of food parenting practices, PROC LCA [26] was used to conduct latent class analysis based on 12 indicators (six parent- and six child-reported JS parenting practices) and step recommendations by Bray et al. [27]. Model selection was conducted using one through six class solutions. Information criteria, entropy, and interpretability of each latent class solution were used to select the appropriate number of classes. Entropy refers to the certainty of model selection with values near one indicating high certainty. Interpretability is based on how clearly classes are distinguished from one another based on item-response probabilities. Item-response probabilities represent the probability of a reporting agreement with a specific parenting practice given membership in a specific latent class. Missing data on parenting practice indicators were accounted for using full-information maximum likelihood estimation. The selected latent class model was re-fit with child age group (12–14 and 15–17 years) and parent and child sex, BMI category, added sugars intake, and JS-LPA included as covariates to produce posterior probabilities. The addition of covariates resulted in a set of regression coefficients, representing the increase in odds of belonging to a class relative to a reference class, corresponding to each covariate attribute. Maximum-probability assignment was used to assign dyads to the class for which they had the highest posterior probability of membership which allowed for descriptive (not inferential) comparisons between classes.