The Shengjin Lake National Nature Reserve (SJLNNR), which is located in Chizhou City in Anhui Province in China, facing Anqing City across the Yangtze River, was chosen as the study site (Figure 1). It is located at the junction of Dongzhi County and Guichi District, covering a total area of 33,340 ha. Known as “China Crane Lake,” SJLNNR is a representative wetland protected area that was established with the aim of protecting the freshwater wetland ecosystem and the rare and endangered waterfowl. Before the establishment of the SJLNNR, local residents closely relied on natural resources and supported themselves through agricultural cultivation, fishing and hunting. To achieve the protection goal of the SJLNNR, local residents are prohibited from catching fish in Shengjin Lake and hunting rare birds. In 2017, a total of 1,331 fishers stopped fishing, resettled, and went ashore to support themselves through other means. Meanwhile, the hunting of rare birds disappeared.

For better conservation, SJLNNR implements zoning control. It has been divided into core, buffer, and experiment zones, with areas of 10,150 ha, 10,300 ha, and 12,890 ha, respectively, accounting for 30.44%, 30.90%, and 38.66% of the total area of SJLNNR. The core zone is covered almost entirely by the lake and is under strict protection. In principle, all human activity is prohibited in the core zone, and no one is allowed to enter. The living areas of local residents in SJLNNR are mainly the buffer and experiment zones, involving 49 administrative villages in six towns (Shenli, Dongliu, Zhangxi, Dadukou, Niutoushan, and Tangtian), with a total population of approximately 64,200. Only scientific research containing no destructive influence is allowed in the buffer zone (Liu et al., 2022). In the experiment zone, multiple activities related to nature protection and sustainable development are permitted, such as scientific experiments, teaching practices, and ecotourism. In addition, crop farming and livestock breeding can only be conducted in the experiment zone. Currently, the agricultural population is approximately 22,000, whereas the rest mainly work away from their hometowns, especially young and middle-aged adults.

The data used in this study were collected from survey questionnaires administered to community residents in SJLNNR. The field survey was conducted in January 2022 by three field workers, and a stratified random sampling method was used to select the respondents. First, two administrative villages were randomly selected from each town in SJLNNR (Table 1). Community residents were randomly selected, with the family considered as the unit from each administrative village. Upon arrival in each administrative village, field workers first visited village cadres who were chosen and organized to serve community residents. Accompanied by village cadres, field workers gained more trust from residents and successfully completed the questionnaire survey. They read the questions and then participants gave them answer verbally. All respondents voluntarily participated in this survey, and most were the main decision-making members of their families.

In this study, community participation was defined as the willingness and behavior of local community residents to improve the management and construction of SJLNNR. Actual participation referred to community residents taking action in SJLNNR management, and participation behaviors mainly included decision-making, benefit-sharing, and daily supervision. At the beginning of this survey, the meaning of community participation and specific participation actions were fully explained to the respondents by every fieldworker. The respondents were then asked two questions: 1) Are you willing to participate in the management and construction of SJLNNR? 2) Have you ever actually participated in? If they chose “yes,” field workers would continue to ask about their specific willingness to participate or actual actions. If the answer was “no,” field workers would also ask respondents about their understanding of participation in SJLNNR. Thus, whether these respondents understood the meaning of participation willingness and actual behavior could be determined according to their responses. If not, field workers would correct the variation in how respondents interpreted the meaning of “participation.” During the survey, respondents could ask fieldworkers about technical terms in the questionnaire or other relevant questions at any time. Each questionnaire took 40 min to 1 h 200 questionnaires were distributed, and 197 were recovered, including 135 from Dongzhi County and 62 from Guichi District, with an effective rate of 98.50%. According to the total population in SJLNNR, the minimum sample size should be 100, at a confidence level of 95% and a precision level of 10% (Sarmah and Hazarika, 2012). A total of 197 valid questionnaires were collected, which was much greater than the minimum sample size.

The TPB was proposed by Ajzen et al. (1991), and was widely used to predict the willingness and behavior of an individual. According to the TPB, three main factors can directly influence behavioral intent: attitude, subjective norms, and perceived behavioral control. Behavioral intent, in turn, can directly determine the actual behavior of an individual (Ajzen, 1985) when the actual control conditions (e.g., individual capabilities, opportunities, resources) are sufficient. A deviation of behavior from willingness in individuals takes place owing to low facilitating conditions and intervening events (Gonçalves et al., 2021).Thus, the TPB connects individual willingness with behavior and provides theoretical support for the possibility of deviation between willingness and actual behavior of community residents in SJLNNR management.

Based on the TPB, attitude, subjective norms, and perceived behavioral control are the three dimensions that influence deviation in this study. In detail, attitude refers to the positive or negative evaluation of participation by community residents (Zhao et al., 2022); subjective norms refer to the external impetus and pressure caused by others when community residents make decisions on whether to participate (Jia et al., 2022); and perceived behavioral control refers to the perceived ease or difficulty based on personal control over resources, opportunities desires, and motives (Sultan et al., 2020) when community residents participate in the management of SJLNNR. In addition, relevant studies have also considered individual and family characteristics as potential influencing factors (Smith, 2012; Huber and Arnberger, 2021; Qiu et al., 2022). Therefore, this study analyzed the factors influencing deviation using four dimensions: attitude, subjective norms, perceived behavioral control, and individual and family characteristics. The deviation between participation willingness and the actual behavior of community residents in SJLNNR was taken as the explained variable, and its influencing factors were considered as explanatory variables.

From the descriptive statistics shown in Table 2, we can see that the average age of these community residents was approximately 58 years. Men and women accounted for approximately 74% and 26%, respectively, representing a large gap. This may be because the residents surveyed were mainly heads of households or family decision-makers, which were typically men in rural families in China. The education level of the samples generally reached that of junior high school. The average number of members living at home for six or more month during the year per family was four, and the average annual family income was approximately 65,700¥. Moreover, the proportion of village cadres was only approximately 7%.

As the explained variable (deviation between willingness and actual behavior) is dichotomous, a binary logistic regression model was used for the analysis. The logistic regression model was set as follows: p i = F y i = β 0 + ∑ j = 1 n β j X i j = e β 0 + ∑ j = 1 n β j X i j 1 + e β 0 + ∑ j = 1 n β j X i j (1) where y i is the deviation between the willingness and actual behavior of community residents in SJLNNR, defined as either 0 or 1, and X i j denotes the value of the j t h variable of the i t h local resident. Furthermore, p i is the probability of deviation, F y i is the probability distribution function, n is the number of explanatory variables, β j is the regression coefficient of the j t h explanatory variable, and β 0 is the intercept. Next, we calculate logarithms on both sides of Eq. 1 to transform the non-linear relationship between the explained and explanatory variables into a linear relationship. The simplified form is as follows: y i = ln p i 1 − p i = β 0 + ∑ j = 1 n β j X i j (2)

Based on Eq. 2, the probability of deviation between willingness and actual behavior to participate can be estimated, and the influencing factors can be further determined.

This study analyzes the hierarchical relationship among the influencing factors and discovers the direct, intermediate, and deep-rooted factors. The ISM was used to analyze the correlation between influencing factors. First, we determined the adjacency matrix R among the factors. It is assumed that the number of significant influencing factors is k, where S is used to denote the deviation between willingness and actual behavior and S i S j i , j = 1,2 , ⋯ , k is used to denote the factors affecting this deviation. The constituent elements of junction matrix R are defined as follows: r i j = 1 , s i   i s   r e l a t e d   t o   s j 0 , s i   i s n ′ t   r e l a t e d   t o   s j i = 0,1 , ⋯ , k ; j = 1,2 , ⋯ , k . (3)

Eq. 3 is used to determine the accessibility matrix M among the factors. The specific calculation formula for M is: M = R + I λ + 1 = R + I λ ≠ R + I λ − 1 ≠ ⋯ ≠ R + I 2 ≠ R + I (4) where I is the identity matrix, 2 ≤ λ ≤ k , and Boolean calculation is used for the power operation of the matrix. Next, we determine the level of influence of each factor. The basic calculation formula is as follows: P S i = S i | m i j = 1 , Q S i = S i | m j i = 1 (5) L 1 = S i | P S i ∩ Q S i = P S i , i = 1,2 , ⋯ , k . (6)

According to Eq. 5, the accessibility matrix is divided into an accessibility set P S i and an antecedent set Q S i , comprising the set of all factors that can be obtained from factor S i in the accessibility matrix. Further, m i j and m j i both denote the factors in the accessibility matrix. The highest level L 1 and the influencing factors that L 1 contains are determined by Eq. 6; then, the other level factors are determined. The specific operation is that rows and columns of the most influencing factors are removed from the accessibility matrix M to form the accessibility matrix M 1 , and the steps of Eqs 5, 6 are repeated to obtain the influencing level of the next factor, for all factors. Finally, a directed arrow is used to link the factors with the same level of influence and between adjacent levels.

As shown in Table 3, 83.76% of the respondents were willing to participate in SJLNNR management. In sharp contrast to the high willingness, only 16.24% of the respondents actually took actions. The willingness of community residents to participate did not fully translate into actual participation behavior. According to Rausch and Kopplin (2021), the deviation between willingness and actual behavior is generally divided into two different categories: 1) individuals that have willingness but no behavior and 2) individuals that have no willingness but behavior. In this study, 133 respondents expressed their willingness to participate but failed to perform actual behavior, accounting for approximately 67.52%. And these 133 respondents were considered for the following analysis. None of the respondents had no willingness but took action. Thus, a deviation exists between the willingness and actual behavior of local participation in SJLNNR, and there may be some factors that prevent willingness from completely converting into actual participation.

To ensure the reliability and validity of the empirical results, a sample data test must be conducted before using a binary logistic regression model. The test results are listed in Table 4, showing that the Cronbach’s α is 0.815 and the KMO coefficient is 0.840 (which is close to 1), with p < 0.05. This indicates that the questionnaire has good reliability and validity, and it is suitable to continue with the factor analysis. The variance inflation factors (VIF) of all explanatory variables were less than five, indicating that there was no multicollinearity problem among the explanatory variables. The p-value of the Hosmer-Lemeshow test is 0.156, which suggests that the null hypothesis of “the observation data and regression model fit well” is accepted and the results given by the binary logistic regression model to be analyzed next could reflect the true relationship between the original variables.

Binary logistic regression analysis was carried out on the sample data using the SPSS 26.0 software. As shown in Table 5, eight influencing factors were identified from the four dimensions. The detailed results are as follows:

(1) Attitude: The awareness by community residents of the necessity of establishing SJLNNR has a significant positive impact on deviation at the 1% level. This means that the more necessary community residents think the establishment of SJLNNR is, the more likelihood there is of a deviation between their willingness to participate and actual behavior. Their understanding of environmental protection laws and regulations and their satisfaction with SJLNNR both negatively affected the deviation at the 10% level. Specifically, community residents who are more familiar with laws and regulations on environmental protection and are satisfied with the governance of the SJLNNR authority are more likely to convert their willingness into actual participation. Additionally, awareness of the importance of self-participation had no significant impact on deviation.

From the results of the logistic regression model, eight factors that significantly influence the deviation between the willingness and actual behaviors of community residents to participate in the management of SJLNNR can be extracted. S 0 denotes the deviation, and S 1 , S 2 , ⋯ , S 8 denote the awareness of community residents of the necessity to establish SJLNNR, understanding of environmental protection laws and regulations, participation of important or recognizable people, future earnings expectations, satisfaction with SJLNNR authority, realization of environmental improvement effect, number of members living at home for 6 months or more during the year per family, and average annual family income. The logistic relationships of the eight factors were determined based on theoretical support and expert opinions. As shown in Figure 2, “V” indicates that row factors directly or indirectly affect column factors, “A” indicates that column factors directly or indirectly affect row factors, and “0” indicates that there is no direct or indirect relationship between row and column factors.

According to Figure 2 and Eq. 3, adjacency matrix R among the nine factors can be obtained as follows: R = S 0 S 1 S 2 S 3 S 4 S 5 S 6 S 7 S 8 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 1 0 0 0 1 1 0 0 1 1 1 0 1 1 1 0 0 1 1 0 0 0 0 0 0 0 1 1 0 0 1 0 0 0 0 1 1 0 0 1 1 0 0 0 1 1 0 0 0 1 0 0 1 1 1 1 0 1 1 1 0 0

Then, accessibility matrix M is calculated from adjacency matrix R, according to Eq. 4, using MATLAB software: M = S 0 S 1 S 2 S 3 S 4 S 5 S 6 S 7 S 8 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 1 1 1 0 1 1 1 0 0 1 1 1 1 1 1 1 0 0 1 1 0 0 1 0 0 0 0 1 1 0 0 1 1 0 0 0 1 1 0 0 1 1 1 0 0 1 1 1 0 1 1 1 1 1 1 1 1 0 1 1 1 0 1

Next, according to accessibility matrix M, Eqs 5, 6, this study obtains the highest-level element set, which is L 1 = S 0 . Furthermore, the 2nd-8th level element set can be obtained in turn, which are L 2 = S 1 , L 3 = S 4 , L 4 = S 5 , L 5 = S 6 , L 6 = S 2 , L 7 = S 3 , S 8 , and L 8 = S 7 . Reordering the rows and columns of accessibility matrix M according to L 1 , L 2 , …, L 8 , backbone matrix N can be obtained as follows: N = S 0 S 1 S 4 S 5 S 6 S 2 S 3 S 8 S 7 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0 0 1 1 1 1 0 0 0 0 0 1 1 1 1 1 0 0 0 0 1 1 1 1 1 1 0 0 0 1 1 1 1 1 1 1 0 0 1 1 1 1 1 1 0 1 0 1 1 1 1 1 1 1 0 1

Finally, from backbone matrix N, it can be concluded that these nine factors were divided into eight layers. According to established logical relationships, the factors at the same level and between adjacent levels are connected by directed edges, and the hierarchical structure and correlation among the eight factors can be determined.

As shown in Figure 3, the eight influencing factors are at different levels. Most importantly, the number of members living at home 6 months or more during the year per family is a deep-rooted influencing factor. Future earnings expectations from SJLNNR, satisfaction with SJLNNR authority, realization of environmental improvement effect of SJLNNR, understanding of environmental protection laws and regulations, participation of important or recognizable people, and average annual income per family are all intermediate indirect influencing factors. Awareness of the necessity of establishing SJLNNR by community residents is a direct influencing factor at the surface level. Furthermore, the deep-rooted factor indirectly influences the deviation by affecting income and other six internal psychological factors.

Based on the above analysis, policy suggestions from three dimensions are proposed to decrease the deviation and promote the conversion of willingness to actual participation by community residents.

In this study, using the question “Have you ever participated in the management of SJLNNR?” to measure whether community residents have actual participation behaviors is relatively insufficient. To avoid the misinterpretation of the word “participation,” we explained its specific meaning and provided examples to increase understanding during the survey. Future research could overcome this impediment by adding multiple-choice questions to the survey questionnaire, such as “What specific participation behavior have you ever had?” and list several types of participation behaviors for respondents to choose. Another limitation is that external contextual factors were not fully considered in this study, such as the types of jobs provided by SJLNNR, individual health status (Feng et al., 2022), and distance to Shenjing Lake (Qiu et al., 2022). In addition, future research can be extended to specific types of protected areas, such as wetland protected areas.