Maintenance mechanisms of child restraint system use based on the health action process approach theory: Evidence from China

Abstract Objective Adherence to the use of child restraint system (CRS) can greatly reduce child casualties in road traffic accidents. This study aims to develop the behavior maintenance model of CRS use based on the health action process approach (HAPA) theory to explore the maintenance mechanisms, suggesting appropriate intervention strategies. Methods Adopting the multi-stage stratified cluster random sampling method, a cross-sectional survey for guardians recruited from kindergarten and primary school was conducted using the questionnaire developed based on the HAPA theory in Nanjing, China, and 1135 valid samples were finally obtained. Rigorous analysis procedures of the partial least squares structural equation modeling (PLS-SEM) method were applied by using SmartPLS3.3.2 software. Results The results confirmed that action self-efficacy (β = 0.479, P < 0.001) and negative outcome expectancy (β = −0.187, P < 0.001) were valid predictors of intention in the non-intentional stage, intention (β = 0.488, P < 0.001) and maintenance self-efficacy (β = 0.334, P < 0.001) contributed significantly to action planning in the intentional stage, while action planning (β = 0.470, P < 0.001) and recovery self-efficacy (β = 0.256, P < 0.001) jointly determined behavior maintenance in the action stage. It was also revealed that action planning fully mediated the relation between intention and behavior maintenance (β = 0.229, P < 0.001). Conclusion Based on the above mechanisms, interventions aimed at the stages of non-intention, intention, and action should be designed and implemented, with the participation of the government, automobile dealers, and communities to promote the maintenance of improved CRS use behavior.


Introduction
Road traffic injury is the primary cause of death globally among children aged 5-19 years, killing more than 500 children every day (Safe Kids Worldwide 2021). In China, from 2010 to 2016, road traffic injury was likewise the primary cause of death among children aged 5-19 years with a mortality rate of 8.30/100,000 in 2016 (Dong et al. 2020). Indeed, injuries and fatalities from road traffic injuries can be lessened. Previous studies have confirmed that child restraint systems (CRS) are effective in reducing injuries and deaths of child passengers, decreasing fatalities by at least 60% in a traffic collision (World Health Organization 2018). Therefore, CRS use is receiving attention from governments. As of 2018, there were 84 countries with national legislation, 33 of which met best practice standards (World Health Organization 2018). In China, local regulations have been implemented in some cities headed by Shanghai since March 1, 2014, requiring guardians to equip and use CRS correctly for their children under four years of age. On June 1, 2021, the provision that guardians shall equip CRS for minors was incorporated into the Law of the People's Republic of China on the Protection of Minors and entered into force. However, there is no data available on the impact generated by national legislation currently, and local regulations have shown no satisfactory results. A survey of 12,061 children aged 0-12 years between 2015 and 2019 showed that the CRS use rate was only 6.42% in Shanghai, although the legislative approach can increase parental safety awareness (Chen et al. 2021).
Given the low use rate, researchers have carried out extensive work to explore the factors affecting the parents' use of CRS for their children. From the children's perspective, the CRS use rate declines as age, weight, and height increase (Moradi et al. 2019).
Children's reluctance, low travel frequency, and short travel distances can also discourage parents from using CRS (Deng et al. 2016;Moradi et al. 2019); from the parental perspective, higher level of education, better financial income, stronger safety precautions, and more regulated driving behaviors lead to a higher CRS use rate (Deng et al. 2016). In addition, parents' attitudes, knowledge, and relevant legal understanding of CRS play an essential role (Liu et al. 2016); from the CRS perspective, high price, unreliable quality, installation hassles, and insufficient space inside cars are deterrents for parents to purchase and use CRS for their children (Deng et al. 2016). Although these studies demonstrated multiple aspects of the influences on CRS use, they have not identified the intrinsic relationships and intensity among the contributing factors. Building a model based on sound behavior theory can effectively address the above issues and explore the behavioral formation mechanisms of parental CRS use for children systematically and in depth (Prestwich et al. 2015). Another critical point is that children should ride CRS every time they travel by vehicle and all through the travel for maximum protective effects (Kakefuda et al. 2008). A study on adherence to exercise programs has shown that most individuals will reenter previous behavior routines in a short time, even if they have changed their behavior (Robison and Rogers 1994). Therefore, it is necessary to conduct a theory-based study on the maintenance of CRS use behavior.
The health action process approach (HAPA) theory was proposed by German scholar Schwarzer in 1992 based on Bandura's self-efficacy theory and previous studies, which can not only clarify the mechanisms for the role of motivational factors, intention formation, and plan-making in initiating a health behavior but also explain the maintenance and recovery (Schwarzer 1992). In practical applications, the HAPA theory is usually divided into non-intentional, intentional, and action stages to explain health behaviors clearly.
In the non-intentional stage, risk perception (referring to the perception of absolute or relative health risks), outcome expectancy (referring to the expectation of the outcome after acting), and action self-efficacy (referring to the confidence in the ability to perform a behavior) determine behavior intention (referring to personal behavioral goals). In the intentional stage, action planning (referring to plans for when and how to perform a behavior) and coping planning (referring to plans for dealing with obstacles that may arise in adopting and maintaining a behavior) play the mediating role in transforming behavior intention into actual behaviors. Maintenance self-efficacy (referring to the confidence in the ability to maintain a behavior) promotes individuals to move to the next stage by incentivizing them to make plans and attempt actions. In the action stage, the initiation, maintenance, and recovery of behavior are determined by the combination of action planning, coping planning, maintenance self-efficacy, and recovery self-efficacy (referring to the confidence in the ability to resume interrupted behavior). The hypothetical model in this study was proposed based on the HAPA theory and divided outcome expectancy into positive and negative directions ( Figure 1).
Schwarzer proposes that the HAPA theory can be widely applied to the field of health behaviors, especially suitable for studies in the action stage, which has been confirmed in numerous articles. The HAPA model can predict the behavior of seat belt use, support maintenance of physical activity for coronary heart disease patients and more recently, encourage facemask use in the early stage of the COVID-19 pandemic (Schwarzer et al. 2007;Aliabad et al. 2014;Lao et al. 2021). Thus, previous indirect evidence suggests that it is potentially feasible to apply the HAPA theory to the maintenance of CRS use behavior. The primary objective of the present study was to explore maintenance mechanisms of CRS use by constructing a behavior model based on the HAPA theory to provide suggestions for developing multistage and targeted intervention programs.

Sample
The conservative sample size N ¼ 1296 was calculated by G Ã Power3.1 considering analyses of two subgroups. In December 2021, this study conducted a cross-sectional survey by distributing an online questionnaire using the Questionnaire Star platform in Nanjing, Jiangsu Province. Through multistage stratified cluster random sampling, in the first stage, three urban areas of Gulou District, Xuanwu District, and Jianye District, and four suburb areas of Qixia District, Pukou District, Liuhe District, and Jiangning District were randomly selected from eleven administrative districts of Nanjing. In the second stage, two kindergartens and two primary schools were randomly selected in each administrative district. In the third stage, we randomly selected one class in each kindergarten, along with one class respectively in low-grade (grades one, two, and three) and high-grade (grades four, five, and six) of each primary school for an online survey completed by the children's guardians. Inclusion criteria: (i) they were guardians of children selected by the sampling method described above and agreed to participate in this study; (ii) they had ever used CRS. In the order mentioned above, 149,121,185,246,165,137, and 228 questionnaires (1231 in total) were collected from each district. After removing invalid questionnaires containing outliers, logical errors, regular answers, etc., 1135 valid questionnaires were obtained with an effective rate of 92.2 percent.

Ethics approval
All participants have provided appropriate informed consent. The present study was approved by the Ethics Review Committee of Nanjing Medical University ( (2021) 661).

Measures
Demographic characteristics: Guardian's categories, degree of education, average monthly household income, and the number of children; children's age, gender, grade, frequency of ride by private car, and average time per ride.
The behavioral scale of CRS use: In this study, we designed the subscales using the literature retrieval method based on the operational definitions for the dimensions of the HAPA theory. All items were scored by a 5-point Likert scale, from "1 ¼ strongly disagree" to "5 ¼ strongly agree." The pretest of a small population had been conducted, and based on the results, the total scale was modified slightly to the final version with 10 dimensions (44 items). Among the dimensions, action self-efficacy (4 items), positive outcome expectancy (3 items), negative outcome expectancy (3 items), risk perception (5 items), intention (5 items), maintenance self-efficacy (5 items), action planning (5 items), coping planning (4 items), and recovery self-efficacy (5 items) were independent variables, while behavior maintenance (5 items) was the major dependent variable. Cronbach's alphas for all dimensions ranged from 0.840 to 0.978, while the total Cronbach's alpha was 0.958. Kaiser-Meyer-Olkin (KMO) was 0.958 and Bartlett's test was significant (P < 0.001), demonstrating the applicability of factor analysis.

Statistical analyses
The procedures of the PLS Algorithm, Bootstrapping (5,000 subsamples), Blindfolding (omission distance was set at 7), and Permutation (5,000 times) were executed applying SmartPLS3.3.2. The path weighting scheme, maximum iterations of 300, stop criterion of 1 Â 10 À7 , and equal indicator weights for the initialization were set. All iterative estimates were convergent, and all analyzes were performed at the significant level of 0.05 with a two-tailed test, while the acceptable statistical power was 80%. Post hoc analysis of statistical power was conducted using G Ã Power3.1. All variance inflation factors (VIFs) obtained in the model (ranged from 1.00 to 3.22) were lower than 3.3, indicating the absence of common method bias (Kock 2015).
As the reflective measurement models were applied in this study, their item reliability, internal consistency reliability, and convergent validity were evaluated by standardized factor loadings, composite reliability (CR), and average variance extracted (AVE), respectively. The standardized factor loading and CR greater than 0.7 are considered acceptable, and the threshold for AVE should be 0.5 or higher (Hair et al. 2019). The Heterotrait-Monotrait (HTMT) criterion was used to test the discriminant validity, which needs all HTMT values below 0.85 to establish the discriminant validity between constructs (Henseler et al. 2015).
The structural model was evaluated based on the standardized path coefficients (b), coefficients of determination (R 2 ) adjusted, the predictive relevance (Q 2 ), and the effect size (f 2 ). Empirically, the R 2 values of 0.670, 0.333, and 0.190 can be considered substantial, moderate, and weak. The Q 2 values above zero indicate that the predictive accuracy of the path model is acceptable for endogenous constructs (Hair et al. 2019). In addition, f 2 values of 0.02, 0.15, and 0.35 represent small, medium, and large effects of exogenous constructs, respectively (Cohen 1988). The f 2 value of less than 0.02 indicates no practical meaning for that path, although it may have statistical significance in the case of a large sample size. This study used the root mean square residual covariance (RMS theta ) to assess global fitness, whose threshold value lies between 0.12 and 0.14 (Henseler et al. 2014).

Demographics
Among the 1135 participants, 73.0% (828) were mothers, and 27.0% (307) were fathers or others. 57.2% (649) of respondents were below undergraduate, 42.8% (486) were undergraduate and above. 48.5% (551) of the participants had an average monthly household income of 15,000 CNY (about 2300 $) or more, while 51.5% (584) had less than 15,000 CNY (about 2300 $). In terms of children, 62.4% (708) of families had only one child, 37.6% (427) had two or more children. The children's mean age was 6.86 ± 2.31 years. In addition, 52.0% (590) of the children were boys, and 48% (545) were girls. 40.8% (463) of the children were from kindergarten and 59.2% (672) were from primary school. Only 11.8% (134) of the children rode in private cars less than once a month, but 22.8% rode daily or almost daily. Furthermore, 63.7% (723) of the children spent less than 30 minutes per ride, and 36.3% (412) spent 30 minutes or more (Table A1). Table A2 provides the mean (standard deviation) of each item and the test results of measurement models. The standardized factor loadings of all items (ranged from 0.809 to 0.973) and the values of CR for all constructs (ranged from 0.906 to 0.983) exceeded the threshold of 0.70. The AVE of each construct (ranged from 0.763 to 0.932) was above the recommended value of 0.5. Therefore, our measurement models satisfied the criteria of item reliability, internal consistency reliability, and convergent validity. All the HTMT values (ranged from 0.121 to 0.802) were below the cutoff value of 0.85, indicating the establishment of discriminant validity (Table A3).

Structural model
The estimated results of the structural model are presented in Table A4. The statistical power calculated by the smallest f 2 value with practical meaning in the structural model was over 99.9%, obviously exceeding the minimum requirement of 80%. The R 2 values for intention (0.386), action planning (0.524), and coping planning (0.397) were higher than the moderate bound of 0.333, while the R 2 value for behavior maintenance (0.654) was close to the substantial bound of 0.670. All Q 2 values (ranged from 0.127 to 0.584) of endogenous constructs were higher than the acceptable value of 0. The RMS theta of the global model was 0.112 (< 0.12), indicating the satisfactory fitness of this model (Figure 2). Table 1 exhibits the analysis results of the mediating effects on behavior maintenance. The action planning was the full mediator between intention and behavior maintenance (b ¼ 0.229, P < 0.001), since the effect of intention on behavior maintenance (b ¼ 0.013, P ¼ 0.643) was not statistically significant when the direct path was added. Similarly, action planning (b ¼ 0.157, P < 0.001) and recovery self-efficacy (b ¼ 0.149, P < 0.001) were also full mediators for the effect of maintenance self-efficacy on behavior maintenance.

Discussion
The results of this study suggest that action self-efficacy, negative outcome expectancy, intention, maintenance selfefficacy, action planning, and recovery self-efficacy have practical meaning in the context of child restraint use. In the non-intentional stage, action self-efficacy has a positive impact on behavior intention, while negative outcome expectancy has the opposite. In the intentional stage, behavior intention and maintenance self-efficacy can positively predict action planning. In the action stage, action planning  and recovery self-efficacy have positive impacts on the behavior maintenance of CRS use. Action planning is the full mediator, building a bridge over the intention-behavior gap. Hence, the health behavior model based on the HAPA theory provides clues to make guardians insist on using CRS for their children, and we proposed corresponding interventions and strategies. The intention of guardians to use CRS can be improved by reducing their expectation of negative outcomes in the use of CRS. According to Deng et al. (2016), the main negative outcome expectancy factors impacting guardians' intention to use CRS include increasing economic burden, time-consuming installation, occupying interior space, and damaging automobile seats. Relevant government departments can regulate the market price of CRS to reduce the economic burden of children's guardians and carry out rating and labeling CRS according to their ease of use to facilitate guardians purchasing easy-to-install CRS. In addition, automobile dealers can prompt children's guardians to consider CRS use when buying vehicles and actively popularize the knowledge of CRS use to reduce the negative outcome expectancies related to automobile use.
Interventions to increase guardians' self-efficacy at the stages of non-intention, intention, and action are indispensable, and the critical role of action planning in maintaining CRS use behavior cannot be overstated. According to Bandura's self-efficacy theory, self-efficacy comes from performance accomplishments, vicarious experience, verbal persuasion, and emotional arousal (Bandura 1977). To perform the cognitive roles of action self-efficacy, maintenance selfefficacy, and recovery self-efficacy, multistage behavioral intervention projects with the theme of CRS use can be designed based on the above four sources. For guardians in the non-intentional stage, interventions such as raising awareness of others' successful experiences and benefits from insisting on CRS use are recommended to increase their action self-efficacy; for guardians in the intentional stage, interventions demonstrating the correct use of CRS and then encouraging them to get started can be conducted to increase their maintenance self-efficacy; for guardians in the action stage, their recovery self-efficacy can be improved by summarizing the experience of successes and failures in the maintenance of CRS use and enhancing the ability to cope with situations that tend to interrupt the use of CRS. As formulating action plans is the required step for guardians with the intention of CRS use to transition to behavior maintenance, guiding guardians to develop and manage personalized CRS use plans is a necessary intervention. The plan should include at least the frequency of CRS checks, access to the knowledge of safe rides for children, when to replace more appropriate CRS according to children's growth, how to handle possible failures of CRS, and how to cultivate children's habit of riding CRS.

Limitation
First, although our sample is representative of Nanjing, it is not rigorous enough to infer the behavioral characteristics of Chinese guardians' CRS use, so the promotion of the results is limited. Second, both the demographic questionnaire and the scale designed based on the HAPA theory were filled out by the participants themselves, and there may be reporting bias leading to information distortion. Third, this study did not explore potential moderator variables (e.g., vehicle characteristics, driving habits, and perceived social support) that could have an impact on the behavior maintenance model of guardians' CRS use.