2.1 Data and study design
To further access the health vulnerability of residents under climate change risk in western China, a disaster reduction education program targeted to residents and children was funded jointly by the Public Health Emergency Center of the Chinese Center for Disease Control and Prevention and the United Nations Children’s Fund. This program was conducted in three pilot areas; Shifang City, Yuexi County, and Lu County. These mountainous areas are exposed frequent floods, mudslides, and landslides caused by heavy rain storms and the region’s dense network of rivers.
The data collection was a part of the project on disaster risk reduction in western China, and the pre-designed questionnaire in this study was developed from the reports published by UNICEF (United Nations Children’s Fund 2015), with a special focus on children’s risk awareness and preparedness to specific climate change events. Only aggregated data was used and participants will remain anonymous. Sun Yat-Sen University has provided guidelines for this study procedures. Additionally, the datasets collected and analyzed during the current study are available from the corresponding author on request.
During the survey, the primary school teachers were trained by the disaster education program to monitor the students to fill the pre-designed questionnaire to assess their climate change risk perception. Due to children’s limited cognitive ability, pictures and short videos were used to simulate specific disaster scenes and guide children to complete the questionnaire. Besides, external validation including unified training for investigators and quality control after investigation were conducted to guarantee the quality of data collection. Specifically, after the questionnaire is filled out by the investigator, it shall be checked by the quality control personnel with missing items, wrong items and logic errors, and placed on file after signed by the investigators and inspectors together. Besides, the Propensity Score Matching Method was applied in the statistical analysis to further control the sample selection bias and avoid the self-reported problematic associations.
2.2 Sampling and questionnaire design
A multistage stratified random sampling method was adopted to recruit respondents. First, the project team randomly selected 45 primary schools from each township in these three counties, including Shifang City 16 schools, Lu County 20 schools and Yuexi County 9 schools. Then, for each school, a random class of grade 4 to grade 5 was selected and at least 50 students with age 8-12 were surveyed. Specifically, if the class size is more than 50 students, 50 students are randomly selected to participate in the survey. If the class size is less than 50 students, the whole class will participate in the survey, and another class will be selected and some students will be randomly selected to conduct the survey.
In October of 2016, 2250 questionnaires were distributed in each of the three counties (i.e., Shifang City 800, Lu County 1000 and Yuexi County 450), after excluding 58 questionnaires due to data-quality issues, we finally collected the individual data of a number of 2192 (97.42%) students (baseline group). In October of 2018, another 1800 students (i.e., Shifang City 640, Lu County 800 and Yuexi County 360) conducted the survey, and we finally collected 1710 (95%) samples in these primary schools in 2018 post-intervention. Additionally, we did not follow up the same students group because children’s perception increased significantly with their age growing (Christiansen et al. 2018; Mudavanhu et al. 2015).
The questionnaire had several parts, including individual and family information (e.g., age, grade, gender, ethic, disaster experience and household income), climate change risk perception (the knowledge of climate change and its secondary disasters, including flood evacuation, gastrointestinal disease, respiratory disease, safety telephone identification, earthquake evacuation and emergency preparedness), and parent-child interactions and parenting strategy (children’s left-behind status and child-rearing style).
2.4 Measurement
In line with Inventory of Parent and Peer Attachment (IPPA) (Armsden and Greenberg, 1987, Xie et al. 2019), the measurement of parent-child attachment in this study included four dichotomous questions about children’s attachment relationship with their parents: did they ask their parents for help in time; did they tell their parents about their crisis; did they attend disaster drills with their parents; and, did their parents pay attention to their demands. Responses were reconstructed as a continuous variable, ranging from 0 (almost no interaction) to 4 (very good interaction) in order to indicate the quality of the parent-child attachment. Parenting strategies included the child’s left-behind status (left-behind or not left-behind) and child-rearing style (raised by parents together or single-raised, and non-parental caregivers), and both variables were constructed as two dummy variables. We further classified them into nine dummy variables to identify the parenting strategies (i.e. completely left-behind, non-left-behind, father works outside, mother works outside, raised by parents together, raised by father alone, raised by mother alone, raised by grandparents, raised by other relatives). Several control variables such as children’s gender, grade, household income, and survey regions were included, and we also considered the confounding effects of disaster risk reduction programs, namely whether the child participated in any intervention programs at school (i.e. drinking water safety, food safety, personal hygiene, prevention of infectious diseases, emergency call for help, escape and rescue from climate extremes). The detailed descriptive statistics of all the study’s variables are presented in the Table 1.
Table 1. Descriptive statistics of variables in this study
Variables
|
Definition
|
Mean
|
Min
|
Max
|
Dependent variables
|
|
|
|
Climate change risk perception
|
Principal Component Analysis scores for 23 questions of individual cognition of knowledge of climate extremes and the secondary disasters 1.
|
|
|
|
-8.43E-10
|
-1.468
|
0.219
|
Explanatory variables
|
|
|
|
Parenting strategies
|
|
|
|
Children’s left-behind status
|
Partly left-behind =1 Completely left-behind =0
|
0.706
|
0
|
1
|
Father Work Outside
|
Yes =1 No =0
|
0.313
|
0
|
1
|
Mother Work Outside
|
Yes =1 No =0
|
0.065
|
0
|
1
|
Non-Left-behind
|
Yes =1 No =0
|
0.328
|
0
|
1
|
Child-rearing style
|
Raised by parents together =1
Single raised, and non-parental caregivers =0
|
0.534
|
0
|
1
|
Raised by Father alone
|
Yes =1 No =0
|
0.062
|
0
|
1
|
Raised by Mother alone
|
Yes =1 No =0
|
0.154
|
0
|
1
|
Raised by Grandparents
|
Yes =1 No =0
|
0.229
|
0
|
1
|
Raised by other Relatives
|
Yes =1 No =0
|
0.021
|
0
|
1
|
Parent-child attachment
|
Frequency of interaction between parents and children.
|
3.206
|
0
|
4
|
Control variables
|
|
|
|
|
Gender
|
Boy =1 Girl =0
|
0.470
|
0
|
1
|
Minority
|
Minority =1 Hanzu =0
|
0.736
|
0
|
1
|
Disaster experience
|
Yes =1 No =0
|
0.397
|
0
|
1
|
Less than 10,000
|
Yes =1 No =0
|
0.214
|
0
|
1
|
Between 10,000 and 30,000
|
Yes =1 No =0
|
0.295
|
0
|
1
|
Between 30,000 and 50,000
|
Yes =1 No =0
|
0.180
|
0
|
1
|
Between 50,000 and 70,000
|
Yes =1 No =0
|
0.104
|
0
|
1
|
More than 70,000
|
Yes =1 No =0
|
0.208
|
0
|
1
|
Grade3
|
Yes =1 No =0
|
0.012
|
0
|
1
|
Grade4
|
Yes =1 No =0
|
0.100
|
0
|
1
|
Grade5
|
Yes =1 No =0
|
0.677
|
0
|
1
|
Grade6
|
Yes =1 No =0
|
0.211
|
0
|
1
|
Shifang City
|
Yes =1 No =0
|
0.351
|
0
|
1
|
Yuexi County
|
Yes =1 No =0
|
0.387
|
0
|
1
|
Lu County
|
Yes =1 No =0
|
0.263
|
0
|
1
|
Received DRR education
|
Received =1 Not received =0
|
0.884
|
0
|
1
|
Note: Household income: More than 70,000=ref, Grade: Grade 6=ref, Survey regions: Lu County=ref.
|
2.4 Analytical Strategy
To identify the effect of parent-child attachment and parenting strategy on children’s climate change risk perception, the collected data was analyzed using Stata14.0 software. First, in line with previous research, Principal Component Analysis was used to calculate the children’s climate change risk perception scores in this study (Slovic 1987; Burns and Slovic 2012). The dependent variable was a continuous variable, so the Ordinary Least Square Regression was applied as a basic assumption in Model 1 to identify the effects of parent-child attachment and parenting strategy on children’s climate change risk perception.
Second, to capture the indirect effects of parent-child attachment and parenting strategy on children’s climate change risk perception, the baseline model was extended with two interaction items between parent-child attachment and parenting strategies in Model 2. We included two separate interaction terms between parent-child attachment and parenting strategies: parent-child attachment × children’s left-behind status and parent-child attachment × child-rearing style. Additionally, to identify how the effects of parent-child attachment on children’s climate change risk perception varies in different parenting strategies, the children’s left-behind status and child-rearing style were first divided into nine different caregiver subgroups and then regression analyses were conducted accordingly.
Third, to reveal the co-influencing mechanisms of parent-child attachment and parenting strategies on the children’s climate change risk perceptions, we illustrated how children left-behind status and child-rearing style modified the effect of parent-child interaction frequency on the children’s climate change risk perceptions, respectively.
Finally, based on the previous attachment literature, we established a conceptual framework of parent-child attachment patterns for children’s climate change risk perception based on a mathematical quadrant between frequent parent-child interactions and available attached figures. The parent-child attachment patterns were then classified into four types: securely attached, avoidant attached, ambivalent attachments, and disorganized attachments, and children’s climate change risk perception were further accessed by categorizing different patterns.