Factors Influencing Climate-Induced Evacuation in Coastal Cities: The Case of Shanghai

Abstract

Against the backdrop of global climate change, extreme weather events, such as heavy rainfall, typhoons, tsunamis, and rising sea levels, have become frequent, posing unprecedented challenges to human society. As an important strategy for coastal cities to respond to climate change, climate-induced evacuation is influenced by complex and diverse factors. This study delves into the driving mechanisms of population migration willingness, revealing the dynamic balance of push, pull, and resistance factors and their interaction with individual value orientations affecting migration decisions. By constructing a Logistic Regression Model, this research quantitatively analyzes the significant impacts of personal circumstances, family characteristics, living conditions, risk perception, compensation relocation, and supportive policies on climate-induced migration willingness, using Shanghai as a case study. The findings indicate that age, education level, household size, housing type, risk perception, and compensation policies are key factors. Building upon the multidimensional capital interaction mechanisms and dynamic threshold response patterns identified in the research, this study proposes a three-phase progressive policy framework: initially, establishing an integrated human–material–social capital framework to implement tiered relocation incentive programs, which address decision window constraints through cognitive empowerment and asset replacement strategies; subsequently, creating a dynamic compensation adjustment mechanism by developing policy toolkits aligned with inverted U-shaped utility curves while enhancing synergistic effects between cultural cognition transformation and vocational training; and ultimately, innovating an institutional–cultural co-governance paradigm that rebalances public service dependency and place attachment through spatial equity redistribution. Specific recommendations encompass designing modular risk education curricula, establishing social network transplantation mechanisms, piloting climate citizenship regimes, and constructing cross-border governance knowledge platforms. These multidimensional interventions encompassing capital restructuring, threshold responsiveness, and cultural adaptation offer valuable policy insights for resolving the “development resilience–migration inertia” paradox in coastal cities.

1. Introduction

In the context of global climate change, extreme weather events, such as heavy rainfall, typhoons, tsunamis, and rising sea levels, have become frequent, posing unprecedented challenges to human society [1]. Coastal cities, as important hubs of global economy and culture, are among the most sensitive and vulnerable areas affected by climate change due to their unique geographical locations and ecological environments [2]. Shanghai, as one of the most representative coastal cities in China, not only bears the responsibility for the country’s economic development but also faces multiple pressures from climate change [3]. Therefore, conducting an in-depth analysis of the factors influencing climate-induced migration in coastal cities, using Shanghai as a case study, holds significant theoretical value and practical significance.

The severe situation of climate change prompts us to focus on disaster prevention and reduction issues in coastal cities [4]. With the continuous increase in global greenhouse gas emissions, the trend of climate warming has become increasingly apparent, and the intensity and frequency of extreme weather events have significantly increased, creating unprecedented disaster prevention pressures for coastal cities [5]. The inundation of low-lying areas caused by rising sea levels, flooding disasters triggered by typhoons and heavy rainfall, and the impact of climate change on water resources and ecosystems all directly threaten the safety, economic development, and social stability of coastal city residents [6]. Therefore, studying disaster evacuation strategies in coastal cities under the backdrop of climate change, particularly the factors influencing climate-induced migration, is of great significance for enhancing urban disaster prevention and reduction capabilities and ensuring the safety of residents’ lives and property [7].

Climate-induced migration is an important strategy for coastal cities to respond to climate change, and its influencing factors are complex and diverse [8]. On one hand, changes in the natural environment serve as direct driving forces for climate-induced migration [9]. For instance, the deterioration of living conditions caused by rising sea levels and safety hazards resulting from frequent extreme weather events can compel residents to consider leaving their original residences. On the other hand, socio-economic factors also play a significant role in influencing climate-induced migration [10]. Differences in economic development levels, fluctuations in employment opportunities, the allocation of educational resources, and the degree of social security system improvement all profoundly impact residents’ migration decisions [11]. Additionally, factors such as the policy environment and cultural psychology should not be overlooked [12]. Therefore, a comprehensive analysis of the factors influencing climate-induced migration requires considering multiple dimensions, including the natural environment, socio-economic conditions, and policy environment, to form a systematic and holistic understanding [13].

Recent years have witnessed extensive and in-depth scholarly investigations into climate-induced evacuation in coastal cities across global academia. Research domains encompass climate change trajectories, urban vulnerability assessments, and residents’ evacuation willingness and behavioral patterns [14]. Particularly regarding the dynamic mechanisms underlying disaster-avoidance migration willingness, empirical analyses have elucidated the dynamic equilibrium of push–pull forces and resistance factors, alongside their interactions with individual value orientations in shaping migration decisions [15]. However, despite these advancements in conceptual frameworks and theoretical insights, critical limitations persist, particularly in quantitatively analyzing specific determinants, constructing predictive models, and formulating targeted policy recommendations. This study adopts Shanghai as a paradigmatic case to systematically examine multidimensional determinants influencing disaster-avoidance migration willingness. Through constructing a Logistic Regression Model, we quantitatively assess the significant impacts of personal circumstances, household characteristics, residential conditions, risk perceptions, compensation resettlement, and supportive policies on evacuation intentions. Our findings not only enrich existing scholarship but also provide robust theoretical foundations and empirical validation for developing scientifically informed climate evacuation policies. The academic significance lies in revealing the dynamic mechanisms and influencing factors of disaster-avoidance migration willingness, thereby offering novel conceptual frameworks and methodological tools for coastal cities confronting climate change. Furthermore, building upon the identified multidimensional capital interaction mechanisms and dynamic threshold response patterns, this study proposes a tripartite progressive policy framework. Specifically, it advocates three-dimensional interventions encompassing capital restructuring, threshold responsiveness, and cultural adaptation to resolve the “development resilience–migration inertia” paradox in coastal cities. These policy innovations furnish valuable references for sustainable urban development and resilience-building initiatives in climate-vulnerable regions.

2. Mechanisms of Population Migration Dynamics

In the context of climate change, the dynamics of population migration are complex and multifaceted, primarily influenced by the frequent occurrence of extreme weather events, increased difficulties in resource acquisition, and the deterioration of living environments [16]. These factors compel people to migrate in pursuit of more stable living conditions and safer environments. The mechanisms underlying this formation involve the interplay of various economic, social, and environmental factors [17]. According to push–pull theory, population migration is driven by various endogenous and exogenous forces [18]. The “endogenous forces” include push factors that compel individuals or groups to leave their original residences, while pull factors attract them to new locations. In contrast, “exogenous forces” refer to the obstacles encountered during the migration process. Changes in the natural, economic, and social conditions of the origin area can create varying degrees of pressure that act as push factors driving migration [19]. Conversely, the destination area typically offers more attractive natural, economic, and social conditions, drawing individuals or groups toward it. Resistance factors generally encompass distance, material conditions, and differences in customs. Based on individuals’ value judgments regarding these factors, they collectively form the resistance experienced during the migration process [20]. Thus, population migration represents a balanced state at the individual level generated by the combined effects of push, pull, and resistance factors.

The long-term, complexity, and multi-scale nature of climate change leads to delayed and uncertain impacts on human life and property safety [21]. Potential risk factors in the origin area, along with adverse socio-economic conditions, create push factors for voluntary migration. In contrast, the reliable natural environment and favorable socio-economic conditions of the destination area form pull factors that encourage people to migrate [22]. Coastal regions, characterized by relatively developed economies, well-established infrastructure, and sound social security systems, present unique resistance challenges for climate migrants compared to other populations. On one hand, the combined effects of rising sea levels and ground subsidence are often weakly observable, and the advanced risk forecasting systems in coastal areas contribute to a state of ambivalence or even indifference among people regarding “climate migration”. The extent to which migrants perceive the risks associated with rising sea levels and ground subsidence, along with variations in individual cultural and material factors, means that they remain in a state of limbo between different risk zones, without achieving genuine population migration [23]. On the other hand, the misalignment between migration demands and the protection of migrant rights is a key factor creating resistance to migration [24]. A limited capacity for absorbing population in the destination area, differences in living customs, and barriers posed by the household registration system are inevitable challenges for resettling migrants [25]. When the endogenous forces exceed the exogenous forces—meaning that the combined push and pull factors surpass the resistance—population migration can be effectively realized [26]. The mechanisms for forming migration willingness are illustrated in Figure 1.

3. Research Methods and Data Sources

3.1. Study Area

Shanghai, located in the eastern coastal region of China, lies on the southern edge of the Yangtze River Delta and is a typical coastal city. It faces the vast East China Sea to the east, is bordered by Hangzhou Bay to the south, adjoins Jiangsu Province to the west, and overlooks the mouth of the Yangtze River to the north. Its advantageous geographical position makes it an important hub connecting China’s inland areas with overseas regions. The latitude and longitude range of Shanghai is approximately 120°52′ to 122°12′ E and 30°40′ to 31°53′ N. This geographical location not only establishes it as a significant gateway to eastern China but also endows it with a unique natural environment and climate conditions. The location map of Shanghai is shown in Figure 2.

Shanghai’s terrain is primarily flat and is part of the alluvial plain of the Yangtze River Delta, characterized by a low average elevation. These topographical features make Shanghai more susceptible to direct impacts from rising sea levels and extreme weather events. In recent years, with the intensification of global warming, rising sea levels have become one of the significant challenges that Shanghai must confront. Climatically, Shanghai belongs to the subtropical monsoon climate zone, with distinct seasons and simultaneous rainfall and heat. Summers are hot and rainy, while winters are mild and relatively dry, creating favorable conditions for agricultural production and residents’ livelihoods. However, as climate change exacerbates, Shanghai also faces an increased risk of frequent extreme weather events, such as typhoons, heavy rainfall, and high-temperature droughts, which raise higher demands for the city’s disaster prevention and reduction efforts.

Additionally, Shanghai boasts abundant natural resources, particularly water and biological resources. The Huangpu River, as the largest river in Shanghai, not only provides an important water source for the city but also serves as a crucial carrier for economic and cultural development. Moreover, Shanghai has vast marine resources, with the development of marine fisheries and marine tourism injecting new vitality into the city’s economy. However, against the backdrop of climate change, finding ways to better address the disaster risks posed by climate change has become an important issue that Shanghai needs to confront.

3.2. Data Sources

Based on the calculated sea-level rise risk assessment values for various districts in Shanghai, we selected districts with higher risk assessment values for this survey: Chongming District, Pudong New Area, Xuhui District, and Baoshan District, as shown in Figure 3. From these districts, two communities were chosen for surveys and interviews based on characteristics such as their proximity to the sea, ground subsidence, and experience with marine disasters. The information of the selected communities is shown in

Table 1. Information on the selected communities in each district.

District	Community
Pudong New Area	Sidu Village, Huinan Town, and Zhizhu Village, Heqing Town
Xuhui District	Tianlin Community, Tianlin Street, and Tianping Community, Tianping Street
Baoshan District	Haijiang New Village Community, Wusong Street, and Linjiang Park Community, Youyi Road Street
Chongming District	Wujianing Community, Chengqiao Town, and Yingchen Community, Chenjia Town

.

Details are provided in Figure 4. Following the principle of random sampling, a total of 610 questionnaires on the migration willingness of community residents for disaster avoidance were distributed, with 603 valid responses collected.

3.3. Research Methods

The Logistic Regression Model is a widely used classification method in statistics, data analysis, and machine learning, particularly suited for handling binary classification problems. This model constructs a Logistic function (usually the Sigmoid function) to map the predictions of a linear regression model into the interval (0, 1), thereby providing a probability estimate for the occurrence of an event. This characteristic allows the Logistic Regression Model to directly output the predicted probabilities for classification tasks rather than merely classification labels.

In the Logistic Regression Model, the relationship between the independent variables (or features) and the dependent variable is represented through a linear combination, which is then transformed using the Sigmoid function to ensure that the output values fall within a reasonable probability range. The form of the Sigmoid function is 1/(1 + e^(−z)), where z is the result of the linear combination, specifically the weighted sum of the model parameters plus the intercept term. This transformation allows the model to capture complex nonlinear relationships between the independent and dependent variables, especially in binary classification scenarios. Parameter estimation in the Logistic Regression Model typically employs the method of maximum likelihood estimation, which seeks to solve for model parameters by maximizing the likelihood function of the observed data. This process is usually implemented using iterative optimization algorithms such as gradient ascent or more efficient algorithms like the Newton method.

In addition to its mathematical rigor, the Logistic Regression Model is favored for its ease of implementation, strong interpretability, and high computational efficiency. It is widely used in various fields such as medicine, marketing, and credit scoring to predict key business metrics like customer behavior, disease risk, and loan defaults, providing decision makers with robust data support.

In general linear regression analysis, the dependent variable y is a continuous numerical variable, and its linear relationship with the independent variable x is represented as follows:

$$\widehat{\mathrm{y}}{=\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{x}}_1+\cdots +{\mathsf{\beta }}_{\mathrm{k}}{\mathrm{x}}_{\mathrm{k}}$$

(1)

In practice, when predicting migration willingness, the dependent variable is a categorical variable. Using linear fitting methods in this case is clearly inappropriate. Therefore, the Logistic function is often employed for the following transformation:

$$\mathrm{p}\left(\mathrm{y}={\displaystyle \frac{1}{{\mathrm{x}}_1}},{\mathrm{x}}_2\dots {\mathrm{x}}_{\mathrm{k}}\right)={\displaystyle \frac{1}{1+{\mathrm{e}}^{-\left({\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{x}}_1+\cdots +{\mathsf{\beta }}_{\mathrm{k}}{\mathrm{x}}_{\mathrm{k}}\right)}}}$$

(2)

$$\begin{array}{c}\\ \mathrm{L}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}\left(\mathrm{p}\right)=\mathrm{l}\mathrm{n}{\displaystyle \frac{\mathrm{p}}{1-\mathrm{p}}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{x}}_1+{\mathsf{\beta }}_2{\mathrm{x}}_2+...+{\mathsf{\beta }}_{\mathrm{k}}{\mathrm{x}}_{\mathrm{k}}\end{array}$$

(3)

In Equation (2), P represents the probability of willingness to migrate, β is the regression coefficient, β_k denotes the regression coefficient for the k-th influencing factor, and x_k is the k-th independent variable, where k indicates the number of influencing factors. Among these, the following are true:

β₀ is the constant term, representing the influence of factors that are independent of the variable x_k.

If β_k = 0, it indicates that P is independent of the variable x_k, meaning that the willingness to migrate is not determined by the factor x_k.

If β_k > 0, it indicates that P is related to the variable x_k, suggesting that the variable x_k is a positive factor influencing the willingness to migrate.

If β_k < 0, it indicates that P is related to the variable x_k, suggesting that the variable x_k is a negative factor affecting the willingness to migrate.

The strength of the relationship between the factors and the willingness to migrate is measured using the odds ratio (OR), calculated as shown in Equation (4):

$$\mathrm{O}\mathrm{R}={\mathrm{e}}^{{\mathsf{\beta }}_{\mathrm{k}}}$$

(4)

The Logistic model, as an advanced modern analytical tool, demonstrates significant advantages in processing survey data regarding the willingness of disaster migrants to relocate. It skillfully transforms diverse individual information (such as age, asset status, etc.) into feature variables and then employs its unique Logistic function mechanism to deeply analyze the impact probabilities and directions (positive or negative) of these features on the binary outcome of migration willingness. Through multiple iterations of regression analysis, the model effectively eliminates irrelevant factors, providing a solid basis for formulating targeted strategies. The popularity of this model stems not only from its scientific rigor, accuracy, and objectivity but also from the intuitiveness of its probability expressions and computational efficiency, which make the analysis process both rapid and convenient. Furthermore, the Logistic model is not limited to predictions within the sample; its robust generalization capabilities allow for effective predictions on external data, equipped with mechanisms for result comparison and validation, ensuring the interpretability and reliability of the predictive results.

3.4. Independent Variable Selection and Definition

Based on a review of the relevant literature on the factors influencing disaster migration, expert consultations, and the actual conditions of the survey areas and adhering to the principles of objectivity, representativeness, and operability in selecting indicators, the following variables that impact disaster migration were determined as independent variables; these mainly include six parts: (1) variables related to individual basic information, including gender, age, household registration type, residency status, educational level, main source of income, nature of work, years of work experience, and attachment to hometown; (2) variables related to family basic information, including number of family members, annual family income, family asset status, number of school-age children, and marital status; (3) variables related to current living conditions, including housing type, living arrangement, and housing area; (4) variables related to risk perception, including familiarity with the situation of rising sea levels in Shanghai and familiarity with ground subsidence in Shanghai; (5) variables related to compensation and resettlement, including resettlement destination, housing compensation method, housing resettlement method, and employment resettlement method; and (6) variables regarding supportive policies, including policy guarantees and training needs. For details, see Figure 5.

Based on the characteristics of the survey sample and the research hypotheses, this study selected a total of 25 input variables across six dimensions to analyze the factors influencing disaster migration willingness. The willingness to migrate (Y) serves as the output variable. According to the design of the survey questionnaire, if respondents are willing to migrate, Y is assigned a value of 1; otherwise, Y is assigned a value of 0. Using survey data from 603 residents in eight communities across four districts of Shanghai, this study conducts an empirical analysis of the factors affecting disaster migration willingness in response to rising sea levels. Each option for the independent variables has been scientifically assigned values, as shown in

Table 2. Factors influencing the disaster migration willingness of Shanghai residents.

Impact Dimensions	Variable Name	Variable Assignment
Y: Disaster migration willingness (dependent variable)		1 = willing; 0 = unwilling
A: Individual basic information	X1, gender	[1,0] = male; [0,1] = female
	X2, age	1 = 0~16; 2 = 17~40; 3 = 41~60; 4 = 61~90
	X3, household registration type	[1,0] = urban household registration; 0,1] = rural household registration
	X4, household registration status	[1,0,0] = local household registration [0,1,0] = non-local household registration (no intention to transfer to Shanghai household registration) [0,0,1] = non-local household registration (intention to transfer to Shanghai household registration)
	X5, education level	1 = primary school and below; 2 = junior high school; 3 = senior high school; 4 = undergraduate (including associate degrees) and above
	X6, primary source of income	[1,0,0,0,0] = salary; [0,1,0,0,0] = agriculture and animal husbandry; [0,0,1,0,0] = asset income; [0,0,0,1,0] = business; [0,0,0,0,1] = parental support
	X7, nature of employment	[1,0,0,0,0] = public institution [0,1,0,0,0] = state-owned enterprise; [0,0,1,0,0] = private enterprise; [0,0,0,1,0] = individual business owner; [0,0,0,0,1] = freelancer
	X8, years of employment	1 = 0~5; 3 = 6~10; 4 = 11~60
	X9, local attachment	1 = shallow; 2 = average; 3 = deep
B: Family basic information	X10, household size	Continuous variable: 1~8
	X11, household annual income situation	1 = 0–200,000; 2 = 200,001–500,000; 3 = 500,001 and above
	X12, household asset situation	1 = 0~600,000; 2 = 600,001~6,000,000; 3 = 6,000,001 and above
	X13, number of school-age children	Continuous variable: 0~3
	X14, marital status	[1,0,0] = unmarried; [0,1,0] = married; [0,0,1] = divorced or widowed
C: Current living conditions	X15, housing type	[1,0] = rural housing; [0,1] = urban housing
	X16, living arrangement	[1,0] = family residence; [0,1] = non-family residence
	X17, housing area	1 = 0~50m2; 2 = 51~100m2; 3 = 101~160m2; 4 = 161~200 m2; 5 = 201m2 and above
D: Risk perception level	X18, unfamiliarity with sea-level rise in Shanghai	1 = familiar; 2 = average; 3 = unfamiliar
	X19, unfamiliarity with ground subsidence in Shanghai	1 = familiar; 2 = average; 3 = unfamiliar
E: Compensation and resettlement situation	X20, resettlement destination	[1,0,0,0,0] = within the county district; [0,1,0,0,0] = other counties in the same city; [0,0,1,0,0] = adjacent city; [0,0,0,1,0] = first-tier and provincial capital cities; [0,0,0,0,1] = other cities
	X21, housing compensation method	[1,0,0] = calculated by registered population; [0,1,0] = calculated by unified standards of housing area and structure; [0,0,1] = compliant with arrangements indifferently
	X22, housing relocation method	[1,0,0] = the government centrally builds affordable housing, which is purchased according to policy; [0,1,0] = the government directly provides individual housing subsidies, allowing for self-directed purchasing; [0,0,1] = designate residential areas for self-construction outside urban towns
	X23, employment placement method	[1,0,0,0] = agriculture; [0,1,0,0] = employment in industrial parks; [0,0,1,0] = government-arranged employment positions; [0,0,0,1] = self-employment
F: Supporting policies	X24, policy protection	[1,0,0,0,0] = social security continuity policy; [0,1,0,0,0] = employment and entrepreneurship support; [0,0,1,0,0] = child education support; [0,0,0,1,0] = provide relocation subsidy; [0,0,0,0,1] = housing relocation guarantee
	X25, training needs	[1,0,0,0] = employment training; [0,1,0,0] = entrepreneurship training; [0,0,1,0] = agricultural and livestock breeding technology training; [0,0,0,1] = social integration training

. These settings ensure a comprehensive analysis of migration willingness and lay the groundwork for subsequent regression analysis.

3.5. Research Hypothesis

Based on theories of population migration and behavioral choices and in conjunction with the actual characteristics of the study area, this research focuses on community residents in Shanghai. The factors influencing the willingness of disaster-relief migrants to relocate are categorized into six main categories: individual basic information, family basic information, current living conditions, risk perception level, compensation and resettlement situation, and supporting policies. These categorizations informed the formulation of the research hypotheses presented in

Table 3. Research hypotheses.

Hypothesis	Content
H1	The individual basic information of residents in coastal cities has an impact on their willingness to migrate for disaster relief.
H2	Different family circumstances of residents in coastal cities affect their willingness to migrate for disaster relief.
H3	Current living conditions influence the willingness to migrate for disaster relief.
H4	The level of risk perception impacts the willingness to migrate for disaster relief.
H5	Compensation and housing arrangements influence the willingness to migrate for disaster relief.
H6	Supporting policies affect the willingness to migrate for disaster relief.

.

4. Results

4.1. Descriptive Statistical Analysis

This study employed the Cochran formula for sample size calculation:

$$\mathrm{n}={\displaystyle \frac{{\mathrm{Z}}^2\cdot \mathrm{p}\left(1-\mathrm{p}\right)}{{\mathrm{e}}^2}}$$

(5)

where Z = 1.96 (corresponding to 95% confidence level), the anticipated proportion p = 50% (maximum variance conservative estimation), and the margin of error e = 5%. The calculated minimum sample size required was 385. Ultimately, 603 valid questionnaires were collected, achieving a 98.53% valid response rate, exceeding the baseline requirement to ensure subgroup analysis reliability.

To control for response bias, three strategic measures were implemented, as shown in

Table 4. Strategies to mitigate response bias in survey design.

Strategy	Content
Pretesting	Optimized question formulation through 30 pilot surveys to eliminate ambiguous items
Anonymization protocol	The exclusion of personally identifiable information to reduce social desirability bias
Stratified compensation	The implementation of face-to-face interviews for low-response demographics (≥61 years)

.

Among the 603 respondents who completed valid surveys, the gender distribution was 352 males and 251 females. The age structure was as follows: 0 respondents aged 16 and below, 362 respondents aged 17–40, 196 respondents aged 41–60, and 45 respondents aged 61 and above. Regarding marital status, 62 respondents were unmarried, 538 were married, and 3 were divorced or widowed. The household registration type distribution was 491 urban residents and 112 rural residents. The distribution of local vs. non-local household registration was 473 with local residency and 130 with non-local residency. In terms of educational background, 20 respondents had a primary school education or below, 61 had completed junior high school, 218 had completed high school (including vocational schools), and 304 had a bachelor’s degree (including associate degrees) or higher. The occupational structure was as follows: 29 respondents worked in public institutions, 85 were employed in state-owned enterprises, 297 worked in private enterprises, and 70 were self-employed. Among the respondents, 379 explicitly indicated a willingness to relocate, accounting for 62.9%, while 224 expressed an unwillingness to relocate, accounting for 37.1%. Most migrants indicated willingness to relocate in the future for disaster risk mitigation. For further details, see

Table 5. Basic population characteristics of the sample.

Statistical Characteristics	Sub-Indicators	Number of People	Proportion (%)
Gender	Male	352	58.37
	Female	251	41.63
Age	0~16	0	0
	17~40	362	60.03
	41~60	196	32.5
	61~90	45	7.46
Marital status	Unmarried	62	10.28
	Married	538	89.22
	Divorced or widowed	3	0.5
Household registration type	Urban household registration	491	81.43
	Rural household registration	112	18.57
Household registration status	Local household registration	473	78.44
	Non-local household registration (no intention to transfer to Shanghai household registration)	32	5.31
	Non-local household registration (intention to transfer to Shanghai household registration)	98	16.25
Education level	Primary school and below	20	3.32
	Junior high school	61	10.12
	Senior high school	218	36.15
	Undergraduate (including associate degrees) and above	304	50.41
Willingness to migrate out	Willing	379	62.85
	Unwilling	224	37.15

.

4.2. Results of Logistic Regression Analysis

This study employs binary Logistic Regression Modeling to quantitatively analyze both objective and subjective determinants of migration intention, aiming to elucidate the magnitude of influence exerted by various factors on relocation willingness. These analytical outcomes will provide valuable references for future disaster-avoidance migration decision making and resettlement planning. Using SPSS 27.0 software, Logistic Regression coefficient significance analysis was performed on 603 valid samples. The study implemented Tolerance tests through linear regression procedures in SPSS 27.0 to examine multicollinearity among independent variables, following these operational steps: initially designating gender as the dependent variable while configuring other variables as independent variables to establish a regression model, thereby calculating Tolerance values, and subsequently iteratively altering the dependent variable through successive regressions against other independent variables to derive respective Tolerance values. The analytical results indicate that the minimum Tolerance value of 0.352 across all independent variables exceeds the 0.2 threshold, demonstrating the absence of multicollinearity in the model, thus confirming the reliability of subsequent Logistic Regression analysis. Detailed regression outcomes are presented in

Table 6. Initial results of the Logistic Regression Model calculations.

	Variables in the Equation	B	S.E	Wald	f	Sig.	Exp(B)
Step 1	Gender	−0.076	0.094	0.654	1	0.419	0.927
	Age	−0.189	0.106	3.168	1	0.075	0.827
	Marital status	0.041	0.207	0.039	1	0.844	1.041
	Household registration type	−0.230	0.119	3.724	1	0.054	0.795
	Local attachment	−0.290	0.079	13.547	1	0.000	0.749
	Education level	0.275	0.075	13.630	1	0.000	1.317
	Primary source of income	−2.533	0.564	20.150	1	0.000	0.079
	Nature of employment	0.078	0.041	3.608	1	0.057	1.081
	Years of employment	−0.446	0.147	9.252	1	0.002	0.639
	Household size	−0.120	0.043	7.678	1	0.006	0.887
	Household annual income situation	−0.106	0.079	1.786	1	0.181	0.900
	Household asset situation	0.182	0.092	3.953	1	0.047	1.200
	Housing type	0.504	0.138	13.296	1	0.000	1.656
	Living arrangement	−0.086	0.236	0.132	1	0.716	0.918
	Housing area	−0.061	0.051	1.433	1	0.231	0.941
	Unfamiliarity with sea-level rise in Shanghai	−0.263	0.069	14.494	1	0.000	0.769
	Unfamiliarity with ground subsidence in Shanghai	−0.046	0.078	0.359	1	0.549	0.955
	Resettlement destination	0.190	0.094	4.052	1	0.044	1.209
	Housing compensation method	0.071	0.064	1.243	1	0.265	1.074
	Housing relocation method	−0.025	0.059	0.185	1	0.667	0.975
	Employment placement method	−0.054	0.045	1.462	1	0.227	0.948
	Policy protection—social security continuity policy	−0.086	0.095	0.828	1	0.363	0.917
	Policy protection—employment and entrepreneurship support	0.159	0.091	3.035	1	0.081	1.172
	Policy protection—child education support	0.040	0.093	0.183	1	0.669	1.041
	Policy protection—provide relocation subsidy	−0.123	0.094	1.729	1	0.189	0.884
	Policy protection—housing relocation guarantee	0.081	0.100	0.663	1	0.416	1.085
	Training needs—employment training	0.338	0.099	11.537	1	0.001	1.402
	Training needs—entrepreneurship training	−0.202	0.093	4.738	1	0.030	0.817
	Training needs—agricultural and livestock breeding technology training	0.144	0.100	2.084	1	0.149	1.155
	Constant	0.251	0.999	0.063	1	0.802	1.285

.

In conducting significance tests on the variables, factors such as household registration type, place attachment, education level, primary income source, occupation type, annual household income, asset status, housing type, unfamiliarity with sea-level rise, unfamiliarity with land subsidence, choice of migration destination, housing compensation standards, housing resettlement method, and employment resettlement method were introduced into the model. This study employs the Backward Stepwise (Likelihood Ratio) method for variable selection, a methodology consistent with theoretically driven modeling logic: Beginning with a full model encompassing all predetermined variables, this approach avoids the risk of omitting critical variables inherent in forward selection methods while controlling for Type I error inflation through stringent elimination criteria. Initially, all predetermined independent variables are incorporated into the model for significance testing. Subsequently, statistically insignificant variables are progressively eliminated until all remaining variables in the equation attain statistically significant levels. After 28 rounds of Logistic Regression analysis, the final data showed that the regression coefficients of all variables had p-values (Sig.) less than 10% in the significance tests. Detailed results are provided in

Table 7. Final results of the Logistic Regression Model.

	Variables in the Equation	B	S.E	Wald	f	Sig.	Exp(B)
Final	Age	−0.174 *	0.099	3.075	1	0.080	0.840
	Household registration type	−0.289 **	0.115	6.309	1	0.012	0.749
	Local attachment	−0.280 ***	0.076	13.606	1	0.000	0.756
	Education level	0.270 ***	0.071	14.495	1	0.000	1.310
	Primary source of income	−2.376 ***	0.544	19.084	1	0.000	0.093
	Nature of employment	0.078 **	0.040	3.864	1	0.049	1.081
	Years of employment	−0.449 ***	0.111	16.247	1	0.000	0.639
	Household size	−0.117 ***	0.039	9.044	1	0.003	0.890
	Household asset situation	0.210 **	0.088	5.645	1	0.018	1.234
	Housing type	0.491 ***	0.135	13.183	1	0.000	1.634
	Unfamiliarity with sea-level rise in Shanghai	−0.279 ***	0.067	17.547	1	0.000	0.756
	Resettlement destination	0.167 *	0.092	3.291	1	0.070	1.182
	Training needs—employment training	0.257 ***	0.096	7.215	1	0.007	1.293
	Training needs—entrepreneurship training	−0.202 **	0.089	5.106	1	0.024	0.817
	Constant	−0.304 *	0.793	0.147	1	0.702	0.738

Note: ***, **, and * indicate significance at the 1%, 5%, and 10% statistical levels, respectively.

.

The analysis of the results from the econometric model indicates that 13 variables—age, household registration type, place attachment, education level, primary income source, occupation type, years of work experience, household size, asset status, housing type, unfamiliarity with sea-level rise, resettlement destination, and training needs—passed the significance test at the 10% level. Among these, age, years of work experience, place attachment, household size, and unfamiliarity with sea-level rise were negatively correlated with the willingness to migrate for disaster relief. In contrast, education level and asset status were positively correlated with the willingness to migrate for disaster relief. For categorical variables, household registration type, primary income source, occupation type, housing type, resettlement destination, and training needs all showed some correlation with the willingness to migrate for disaster relief.

5. Discussion

5.1. Interactive Mechanisms of Individual and Household Characteristics

In disaster-induced migration decision-making processes, individual life courses and household resource profiles constitute an interactive relational framework [27]. Grounded in life-cycle theory, institutional constraints and place attachment jointly influence migration decision thresholds [28]. The consolidation effect of residential inertia among elderly populations significantly accelerates social capital depletion, serving as the primary driver of delayed decision making [29]. Age gradient disparities within coastal urban–rural dual structures reveal that spatial equity differentiation stemming from household registration systems and urban residents’ policy dependency form a compound moderating mechanism [30]. Empirical analysis demonstrates that household registration status exerts a coefficient of −0.289 (p < 0.01) on migration willingness, with explanatory power substantially exceeding traditional push–pull theory predictions [31]. Second, educational capital significantly reduces psychological migration costs through cognitive restructuring mechanisms. The positive association between educational attainment and migration willingness (β = 0.270, p < 0.001) corroborates knowledge acquisition’s catalytic effect on decision optimization, forming a dual reinforcement loop with household assets (β = 0.210, p < 0.01). Household structural differentiation (β = −0.117, p < 0.001) induces decision-efficacy stratification: simplified social networks in nuclear families diminish decision-making capacity, while multigenerational households reinforce place dependence through kinship ethics, creating distinctive adaptation dilemmas under escalating climate risks [32]. Third, material capital demonstrates threshold effects, with household assets surpassing critical values significantly weakening residential preference constraints [33]. However, occupational stability’s inhibitory effect (β = −0.449, p < 0.001) reveals that long-term career commitments compress decision-making windows, exhibiting temporal hysteresis fundamentally diverging from Todaro’s expected income theory [34]. Gradual environmental changes reconstruct risk perception pathways, while place identity rooted in cultural substrates forms psychological immunization mechanisms against policy interventions.

Migration decision making essentially constitutes a dynamic gaming system of multidimensional capital, where individual cognitive capacities and household structural features may either foster decision synergy or trigger resource competition [35]. Data analysis indicates migration willingness increases 2–3 times above baseline levels in populations with both high education and household assets, challenging linear assumptions in conventional decision models [36]. Decision hysteresis reflects the disjuncture between policy logic and local cognition in modernization processes: when institutional designs deviate from place-based experiential frameworks, economic compensation mechanisms fail to activate migration behaviors [37].

5.2. Theoretical Reconstruction of Risk Perception and Migration Decision Making

Under the interactive lens of cultural cognition and psychometric evaluation, the theoretical framework of risk perception necessitates transcending the explanatory boundaries of traditional rational choice models [38]. Empirical analysis reveals that residents’ cognitive gaps regarding sea-level rise (β = −0.279, p < 0.001) significantly delay decision-making processes, being rooted in informational asymmetries between local experiential systems and scientific early-warning mechanisms [39]. As disaster uncertainties intensify, differentiation in institutional trust exacerbates place-dependent psychology, leading to systematic deviations between risk cognition and migration behaviors. The phenomenon of decision hysteresis fundamentally represents a coupling of risk anxiety and affective decision-making patterns [40]. Field investigations demonstrate pronounced cultural–cognitive disparities in coastal residents’ risk perception, particularly manifesting as nonlinear attenuation trajectories in policy trust over time [41]. This intergenerational transmission effect exhibits heterogeneous characteristics across groups: collectivism-oriented groups are more susceptible to social norm constraints, generating collective delay effects in migration decisions [42].

Policy compensation efficacy demonstrates significant correlations with cultural capital levels. When economic incentives mismatch local cultural–cognitive frameworks, their impact pathways are readily neutralized by affective decision-making modalities [43]. Urban and rural populations exhibit marked divergences in decision thresholds, with urban residents’ institution-dependent thresholds substantially elevated compared to their rural counterparts [44]. Furthermore, these disparities display dynamic regulatory characteristics evolving temporally throughout incremental disaster progression [45].

Culturally sensitive information strategies reduce psychological resistance through cognitive restructuring, yet their effectiveness remains constrained by the inclusivity of local knowledge systems [46]. In contemporary disaster and emergency education practices, integrating traditional survival wisdom with modern risk mitigation knowledge can exponentially enhance cognitive upgrading efficiency, thereby necessitating revisions to the linear transmission mechanisms of protection motivation theory [47].

Migration decision-making reconstruction essentially constitutes an adaptive alignment process between institutional provision and individual cognition [48]. This study’s data analysis indicates that policy interventions penetrating cultural memory barriers effectively mitigate credibility decay, with effect intensity exhibiting complex correlations with resettlement zone economic development. These findings offer novel perspectives for resolving adaptation paradoxes: catalyzing risk perception paradigm shifts through cultural–cognitive transformation [49].

5.3. Synergistic Effects of Residential Environment and Resettlement Policy Interventions

The synergistic interplay between residential environments and compensation policies manifests multidimensional dynamic linkages, with mechanisms rooted in the diachronic interaction of spatial dependency and institutional trust [50]. Institutional disparities in urban–rural housing attributes engender significant gradient differentiation in spatial attachment levels, originating from the asymmetric distribution of public service networks: urban residents develop risk-buffering perceptions through accumulated institutional trust, while rural dwellings’ embedded localized cultural memories reinforce traditional path dependency in migration decisions. This spatial heterogeneity manifests as dual systems of expected benefit evaluation in migration economic models, where economic incentives encounter diminishing marginal utility when policy designs deviate from local cognitive frameworks [51].

Decision hysteresis arising from resettlement location selection exposes deficiencies in institutional articulation mechanisms, with migration distance extension significantly amplifying social network reconstruction costs [52]. Theoretical deconstruction reveals critical thresholds in push–pull dynamics generated by economic disparities between origin and resettlement areas: when regional development gaps surpass specific boundaries, migration cost absorption mechanisms trigger systemic obstructions [53]. Such nonlinear relationships generate unique tensions in coastal urbanization processes: affluent populations exhibit dampened migration responsiveness due to institutional inertia, while underdeveloped regions face exacerbated decision paralysis from social capital fragmentation risks.

Policy portfolio efficacy demonstrates deep coupling with human capital enhancement. The significant impact of vocational training on migration willingness (β = 0.257, p < 0.001) corroborates knowledge iteration mechanisms’ pivotal role. While this mechanism reduces psychological costs through risk perception remodeling, its effectiveness remains strictly constrained by cultural adaptability. Training programs dissociating traditional expertise from modern technologies risk reinforcing place attachment. Empirical comparisons reveal that skill development programs integrating local production logics significantly enhance decision efficiency, outperforming singular economic compensation measures [54].

Optimizing compensation policies requires transcending economic rationality assumptions to integrate temporal evolution patterns of socio-psychological factors [55]. The economic utility of migration compensation follows an inverted U-shaped temporal trajectory, peaking at 6–12 months before entering decay phases due to lagging social network reconstruction [56]. Institutional designs must establish anticipatory guidance frameworks to prolong policy efficacy through security reinforcement [57]. When compensation measures resonate with cultural governance systems, migration willingness amplification can reach multiples of baseline levels, demonstrating particular value in addressing incremental climate risks like sea-level rise [58].

5.4. Hypothesis Validation and Integrated Impact Mechanism Synthesis

Through iterative corroboration of theoretical construction and empirical verification, this study systematically validates six initial research hypotheses (H1-H6), elucidating the multidimensional dynamic mechanisms underlying disaster-avoidance migration decision making in coastal cities. The H1 individual characteristic influence mechanism is comprehensively confirmed, with age, household registration type, and educational capital constituting a decision-making triad whose interactions create generationally differentiated migration thresholds. The H2 household characteristic pathway exhibits nonlinear features, where antagonistic effects between household assets and population size validate the dialectical relationship between material capital accumulation and social network reconstruction. The H3 residential condition mechanism centers on housing tenure attributes, revealing institutional constraints surpassing physical spatial characteristics as core decision variables. H4’s theoretical reconstruction of risk cognition uncovers deep couplings between perceptual thresholds and cultural–cognitive frameworks, transcending explanatory limitations of conventional risk communication models. The H5 spatiotemporal effects of compensation–resettlement manifest policy resonance across relocation site selection and vocational training dimensions, with efficacy decay curves challenging static assumptions of economic compensation. H6’s institutional resilience of supportive policies emerges through differentiated impacts of social security and entrepreneurial support, confirming synergistic potential in policy portfolio design. The validation status of all hypotheses is detailed in

Table 8. Validation of research hypotheses.

Hypothesis	Content	Validation Status
H1	The individual basic information of residents in coastal cities has an impact on their willingness to migrate for disaster relief.	Supported
H2	Different family circumstances of residents in coastal cities affect their willingness to migrate for disaster relief.	Supported
H3	Current living conditions influence the willingness to migrate for disaster relief.	Supported
H4	The level of risk perception impacts the willingness to migrate for disaster relief.	Supported
H5	Compensation and housing arrangements influence the willingness to migrate for disaster relief.	Supported
H6	Supporting policies affect the willingness to migrate for disaster relief.	Supported

.

The study reveals that disaster-avoidance migration decision making systems exhibit a tripartite nested structure of “individual capital–household networks–institutional environments”: at the micro-level, educational capital transcends risk perception thresholds through cognitive upgrading; at the meso-level, household asset accumulation reconfigures the elastic boundaries of spatial attachment; and at the macro level, temporally designed compensation policies reshape the dynamic equilibrium of institutional trust [59]. This cross-scale interaction generates a unique “decision window period” phenomenon: when cumulative climate risks surpass individual psychological thresholds, household assets reach critical thresholds for migration cost absorption, and compensation policies operate at peak efficacy, migration willingness exhibits exponential growth characteristics. The research concurrently uncovers a paradox in coastal megacity migration governance: advanced public service systems simultaneously reduce survival risks while reinforcing place attachment through institutional dependency, creating a novel contradiction between “developmental resilience” and “migration inertia”.

The study’s limitations manifest in three aspects: First, cross-sectional data inadequately capture temporal evolutionary patterns of climate risk perception and migration decision making [60]. Second, the model’s incomplete incorporation of social capital indicators like community network structures may underestimate informal institutions’ influence [61]. Third, the Shanghai-centric case study necessitates cautious generalization to other coastal cities, requiring consideration of regional development gradients. Future research should deepen investigations in three directions: establishing multi-wave panel survey databases to decode dynamic evolutionary mechanisms of decision processes; developing hybrid models integrating social network analysis and spatial econometrics to unveil community-level contagion effects; and conducting cross-national comparative studies to identify universal governance principles across institutional–cultural contexts. These advancements will facilitate constructing more robust theoretical frameworks for climate migration decision making, providing scientific foundations for global coastal cities confronting sea-level rise.

6. Conclusions and Recommendations

This study delves into the complex structure of the driving mechanisms behind migration willingness. At the core of this mechanism is the dynamic balance between push, pull, and resistance forces and how these forces interact with individuals’ value orientations to influence their migration decisions [62]. To quantify the impact of these factors, we carefully constructed a Logistic Regression Model. This model was chosen for its ability to handle binary dependent variables (such as willingness to migrate), effectively revealing the strength and direction of the associations between independent variables (such as personal and family characteristics, living conditions, risk perception, and compensation and resettlement policies) and the dependent variable [63]. During the model construction process, we rigorously selected and defined variables to ensure that each one accurately reflected its potential impact on migration willingness. Additionally, we collected high-quality data through multiple channels and applied scientific preprocessing methods to ensure the representativeness and reliability of the analysis. Using Shanghai as a case study, we applied the Logistic Regression Model to conduct an empirical analysis of the factors influencing disaster-relief migration in the context of sea-level rise. The results show that various factors, including personal characteristics (e.g., age, household registration type, and education level), family characteristics (e.g., household size and family assets), current living conditions (e.g., housing type), risk perception (e.g., awareness of sea-level rise), compensation and resettlement situation, and supporting policies (e.g., resettlement destination and training needs), significantly influence the willingness of Shanghai residents to migrate for disaster relief. These findings not only enrich the theoretical research on migration but also provide valuable insights for policymakers. They reveal that in addressing global environmental challenges such as sea-level rise, a comprehensive consideration of individual, family, and societal factors is essential for formulating precise and holistic policies. Such measures are needed to facilitate the smooth migration and resettlement of affected residents, ensuring their livelihoods and well-being.

The study reveals that disaster-avoidance migration decision-making systems exhibit three regular patterns of multidimensional capital interaction, dynamic threshold response, and institutional-cultural synergy: First, the formation of migration willingness essentially constitutes a nonlinear superposition process involving human capital (educational attainment), material capital (household assets), and social capital (household registration systems and family structures), whose interactions generate decision window effects through mechanisms of cognitive restructuring, risk buffering, and network viscosity. Second, policy intervention efficacy is governed by dynamic threshold response patterns: the marginal utility of economic compensation follows inverted U-shaped decay across migration cycles, while synergistic effects of cultural–cognitive transformation and vocational training can transcend temporal limitations of conventional incentives. Third, the compatibility between institutional provision and local experiential frameworks determines migration governance resilience, where compounded effects of public service dependency and place attachment engender a “development paradox,” necessitating dynamic rebalancing of institutional trust through spatial equity redistribution and social network resilience building. Based on these findings, the following recommendations are proposed:

(1)

Construct a multidimensional capital-integrated policy framework:

Leveraging the interactive enhancement effects of human, material, and social capital, design tiered relocation incentive programs [64]. For populations with both high educational attainment and substantial household assets, implement a “cognitive empowerment + asset replacement” composite strategy, accelerating the activation of decision windows through modular risk education curricula and property conversion financial instruments [65]. For low-education/high-asset households, adopt “culturally embedded compensation” by transforming traditional residential memories into cultural landmark development in resettlement zones, thereby reducing the psychological costs of spatial attachment.

(2)

Develop dynamic threshold-responsive policy tools:

Establish a migration willingness monitoring and early-warning system to dynamically adjust compensation standards and resettlement plans based on risk accumulation rates and household asset growth curves. Implement “graduated housing subsidies” providing high rental coverage ratios for the first six months, followed by phased reductions aligned with social network reconstruction progress. Concurrently introduce “occupational transition buffer funds” to mitigate decision hysteresis caused by economic disparities in resettlement areas [66].

(3)

Implement an institution–culture co-governance model:

Reconstruct risk communication paradigms by translating scientific sea-level rise monitoring data into localized experiential narratives. Facilitate the integration of scientific cognition and place-based wisdom through cultural mediums such as community disaster memory museums and cross-generational risk workshops. Embed a “social network transplantation mechanism” in resettlement planning, utilizing digital twin technologies to simulate original community relational structures for optimized neighborhood matching [67].

(4)

Innovate spatial equity redistribution mechanisms:

Address migration barriers stemming from household registration systems by piloting a “climate citizenship” regime that incorporates disaster risk exposure into public service allocation weights, permitting high-risk zone residents to select cross-jurisdictional resettlement sites while retaining original household registration benefits [68]. Establish a “resilience credits system” converting participation in disaster preparedness training and community mutual aid into resettlement priorities and housing subsidy coefficients.

(5)

Strengthen decision window intervention efficacy:

For regional governments, establish climate migration response funds integrating land swaps, tax incentives, and entrepreneurial support packages [69]. Develop a migration decision support system to generate personalized relocation plans based on individual life-cycle stages and household structural characteristics, with blockchain technology ensuring transparent compensation process tracing [70].

(6)

Establish a dynamic migration adaptation assessment system:

Construct a tripartite evaluation model incorporating institutional resilience, cultural adaptability, and social network reconstruction indices for quarterly policy impact monitoring [71]. Institute migrant committees to participate in resettlement planning, employing deliberative democracy tools to resolve cultural conflicts. Align traditional lunar cycles with disaster seasons to design emergency drills and policy communication rhythms congruent with local temporal cognition.

(7)

Promote cross-border governance knowledge-sharing platforms:

Establish a climate migration information platform for coastal megacities, compiling transnational case libraries and policy experimentation databases. Develop a migration policy simulator embedded with culturally contextualized decision algorithms to provide cultural adaptability adjustments for institutional transplantation [72]. Create an international climate migration expert mobility mechanism to foster the iterative integration of local experiences and global governance frameworks.