This study utilized a cross-sectional design. To ensure better representativeness of the Chinese population, a large-scale survey was launched in the rural areas of 7 provinces located in the north (2 provinces), east (2 provinces), central (1 province), southwest (1 province) and south area (1 province) of China in 2018. A total of 6203 pairs of fourth to sixth grades children and their caregivers from 13 elementary schools were recruited as study participants. After excluding 592 refusals or invalid responses, 5611 valid questionnaires were obtained with an effective response rate of 90.5%. The socio-demographic characteristics of the surveyed children and their caregivers are presented in Table 1. In total, 2872 boys and 1739 girls with an average age of 11.4 years (SD = 1.02 years) participated in the survey. Furthermore, 56.1% of the caregivers were female and 89.2% were aged less than 60 years. Most caregivers were the children’s parents (80.7%), 17.8% were their grandparents and 1.5% were children’s relatives or others, indicating the existence of left behind children. In addition, 56.3% of caregivers were farmers, 82.2% did not receive a high school education, and 83.7% of their families had a monthly income of less than 6,000 RMB. These demographic characteristics are representative of current conditions in the rural areas of China.
Since the participants in the study are both children and their legal guardians, addressing ethical considerations, at the beginning of the questionnaire a statement was clearly explained the purpose of the study and informed that all participants (including caregivers and their children) completing the questionnaire was optional and if caregivers completed and returned the questionnaires, we would assume that they approved us to use the information both they and their children provided. Therefore informed consent to participate in the study has been obtained from all participants, including legal guardians of minors.
Trained researchers explained to the children of fourth to sixth grades how to complete the questionnaires. Children completed their part independently at school and took the rest of the questionnaire home. The major caregivers were asked to complete their sections and then children returned them to teachers within three days. The researchers collected all the returned questionnaires.
Rural caregivers’ parenting attitudes related to children’s eating habit and their segmentation
The rural caregivers’ parenting attitudes related to children’s eating habit was first determined. We gave a full consideration of the general requirements of the parental attitudes in the child feeding questionnaire in the literature  and the actual conditions of rural China and designed nine statements. Specifically, the descriptive statements with random number explored rural caregivers’ attitudes toward developing eating habit for nutritionally-balanced diet (Nos 1, 2, 3, 7 and 8) and food safety assurance (Nos 4), knowledge level toward obesity (Nos 5) and food safety (Nos 6), and attitudes about the necessity to guide children to properly purchase foods (Nos 9). The caregiver participants were asked to score their answers on a five-point Likert scale (1=strongly disagree, 2=fairly disagree, 3=agree, 4=fairly agree, 5=strongly agree) to indicate their agreement with each statement. An exploratory factor analysis was conducted to summarize rural caregivers’ parenting attitudes along major dimensions. Thereafter, K-means cluster methods were used to segment the participants into distinctive, homogeneous clusters based on their agreements with the dimensions.
Rural children and their caregivers’ unhealthy food choices under designed eight scenarios with safety risks.
Eight scenarios, all closely related to the real food consumption environment in the rural areas of China, were designed. Each of the eight scenarios involved one type of unhealthy food with safety risk, including foods that were moldy, dirty, unpackaged, foul smelling, imitated, out-of-date, sold by unsanitary street vendors, and had caused problems pertaining to foodborne diseases before. Both children and caregivers’ food choices under each scenario were examined and compared.
Effect of rural caregivers’ types, parenting attitudes, unhealthy food choice and other socio-demographic characteristics on their children’s unhealthy food choice
Caregiver factors influencing children’s choices on the unhealthy foods in the eight scenarios were examined. Since the sum of each child’s positive responses toward the eight scenarios takes values of (0-8), which are not interval in nature, we expressed the children’s choice for unhealthy foods as an ordinal ranking with three categories of arbitrary width: positive choices in zero to one scenario, positive choices in two to four scenarios, and positive choices in five to eight scenarios. Thus an ordered probit estimation technique was applied to assess the impacts of caregivers factors on the children’s choice. All of the analytic plan, including the construction of variable scales, was pre-specified. We chose the caregivers’ types, parenting attitudes, unhealthy food choice and other socio-demographic characteristics (sex, age, education level, occupation, and income) as independent variables. Children’s age and sex were set as control variables. The marginal eﬀects of the independent variables and control variables on children’s intention were also analyzed. When more unsafe foods were selected, a change in the sign of the marginal eﬀect from negative to positive reﬂects an increase in marginal utility, while a positive to negative change indicates a decrease in marginal utility. We also applied the data in the ordered logit model and OLS model and the results showed that the regression coefficients and significant regression results of each model were consistent with those of the ordered probit model and the robustness test was passed.
Factor-cluster analyses and children’s difference comparison were conducted using SPSS v21 (SPSS, Inc., Chicago). The ordered probit model analysis and marginal effects were evaluated in STATA 15.0.