For jobs or housing in the Metropolis: A Life Course Analysis of Small-town Migrants’ Settlement Intentions Through Machine Learning

For jobs or housing in the Metropolis: A Life Course Analysis of Small-town Migrants’ Settlement Intentions Through Machine Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article For jobs or housing in the Metropolis: A Life Course Analysis of Small-town Migrants’ Settlement Intentions Through Machine Learning Xinyi Shu, Hurex Paryzat, Kaili Zhang, Shijie Chen, Zhigang Li, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7845488/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 10 You are reading this latest preprint version Abstract Migration from small towns to metropolises is a global trend, but migrants often face a dilemma between employment opportunities and prohibitive living costs. In China, this is exacerbated by institutional barriers like the hukou system, trapping many in a state of temporary residence without full welfare benefits. Settlement intention is thus a complex nonlinear process, yet understanding of these dynamics remains limited. Employing the XGBoost-SHAP methodology on China Migrants Dynamic Survey data, this study analyzes the non-linear effects of employment and housing on the settlement intentions of small-town migrants. The findings reveal significant threshold effects, where factors of urban attraction transform into repulsion. Counterintuitively, highly educated migrants exhibit a lower tolerance for metropolitan pressures, as their versatile human capital allows them to find a better work-life balance in medium-sized cities. In contrast, less-educated migrants show higher tolerance, compelled by constrained employment alternatives to accept precarious incomplete settlement. This educational divide is sharpened by life course events: for the highly educated, having children decreases their tolerance for megacities, while a structural dilemma forces the less-educated to remain even after having children. Scientific community and society/Geography Social science/Geography Social science/Science technology and society Social science/Sociology Settlement intention Small-town migrants Life course Machine learning Education divide China Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1 Introduction The concentration of population from small towns to large cities is a global trend, where better employment opportunities and income are available (Liu & Wang, 2020 ). In China, this trend has been exacerbated by large-scale infrastructure development, particularly the high-speed rail network (Wang & Duan, 2018 ; Wang, Acheampong, & He, 2020 ). However, due to high housing prices in large cities, migrant populations face the dilemma of being unable to balance employment opportunities with living conditions. This issue is further complicated in China by institutional factors such as the hukou system and its associated public service provision (Wang & Shen, 2023 ; Liu et al., 2016 ). As a result, a large number of migrant populations remain in a long-term state of “temporary residence”—having the need to work and live in large cities while struggling to obtain full urban welfare benefits and stable housing, namely long-term settlement capacity. Meanwhile, as migrant populations are one of the important driving forces behind the rapid economic growth of large cities globally, particularly in China, the issue of migrants’ settlement intentions has attracted increasing research attention (Gan et al., 2016 ; Liu & Wang, 2020 ). The effects of employment and housing on settlement intentions vary significantly across different life stages and populations. Different life events, especially marriage and childbearing, fundamentally influence individuals’ spatial choices by reconfiguring family responsibility networks and resource allocation patterns (Elder, 1994 ; Elder et al., 2003 ; Wagner & Mulder, 2015 ). Due to limited educational resources for non-local hukou holders, the impact of marriage and childbearing on migrants’ settlement intentions is more complex in China (Wang & Shen, 2023 ; Gan et al., 2016 ). Furthermore, through employment and income differentiation, groups with different educational backgrounds exhibit disparities in long-term settlement capacity (Fan, 2011 ; Liu & Wang, 2020 ). This gap has been further exacerbated by the recent emergence of "talent wars" among major cities, which implement differentiated preferential policies using educational level as a screening criterion. Although an increasing number of studies recognize the importance of educational level and life course on settlement intentions (Wang & Shen, 2022 ; Liu & Wang, 2020 ), gaps remain in research methods and study populations. On one hand, since migrants’ settlement intentions in large cities represent a complex non-linear problem, it is necessary to adopt machine learning algorithms to precisely identify the non-linear effects and threshold effects of variables, rather than relying on traditional linear regression models used in previous studies (De Jong & Graefe, 2008 ; Wagner & Mulder, 2015 ). On the other hand, since the gradual relaxation of the hukou system since the millennium, rural and non-rural hukou migrant populations, as the main focus of past research, face increasingly diminished differences in their challenges (Wang & Shen, 2022 ). Migrant populations from resource-poor small cities, including farmers, should replace the singular farmer perspective. They migrate to large cities due to limited development opportunities in their hometowns, yet remain in a long-term state of temporary residence, not fully integrated into large cities (Wang, 2024 ; Gu, 2021 ). Therefore, this study will analyze how employment and housing affect the settlement intentions of small-town migrant populations with different educational backgrounds in large cities, based on a life course perspective and utilizing machine learning algorithms. The paper is structured as follows: Section 2 reviews the literature on population mobility, life course, and settlement intentions, followed by an analytical framework for understanding the settlement intentions of small-town migrants in Section 3 , incorporating multiple influencing factors, and introduces the variables and methodology used in the study. Section 4 presents variable importance analysis and partial dependence analysis based on SHAP interpretations. Finally, the paper discusses findings and concludes in Section 5 . 2 Literature review 2.1 Working and Housing as Drivers of Population Mobility Working and housing constitute the core factors driving urban migration and settlement, but their mechanisms differ across educational groups (Fan, 2011 ; Khoo et al., 2008 ; Wang & Shen, 2022 ; Liu et al., 2021 ). In China, the interplay of institutional factors such as household registration (hukou), housing prices, and public service allocation makes the impact of working and housing on settlement intention particularly complex (Wu, 2004 ; Chan, 2010 ; Wang et al., 2021 ). Working factors encompass three core dimensions: job opportunities, working stability, and income level. Among these, income level has received extensive attention due to its quantifiable characteristics (Wang & Shen, 2022 ; Fan, 2011 ; Khoo et al., 2008 ). Income directly determines migrants’ quality of life and capacity for settlement, making it a significant factor influencing settlement intention (Zhu, 2007 ; Huang et al., 2017 ). Migrants with different educational levels show substantial differences in income acquisition and income expectations. Highly educated migrants typically earn higher incomes and focus more on urban wage levels and development potential (Wang, 2024 ). In contrast, less educated migrants face greater restrictions in the labor market and are more likely to be excluded from quality employment opportunities, making them more concerned with personal income levels (Wang, 2024 ; Wei et al., 2023 ). The accessibility, comfort, and stability of housing largely determine migrants’ living standards, thereby influencing their settlement intention (Liu et al., 2016 ; Yang & Guo, 2018 ). In China, homeownership is tied to the household registration system, providing not only housing but also access to local public services and welfare. Despite relaxed hukou entry requirements and economic support since the 2014 New-Type Urbanization Plan, persistently high housing prices keep homeownership unaffordable for most migrants. (Liu et al., 2016 ; Zheng et al., 2021; Luan et al.,2024). Consequently, many migrants enter diversified private rental markets, where housing accessibility is relatively high, but stability and comfort are often poor (Du et al., 2019 ; Zheng et al., 2021; Wu et al., 2019 ). Different rental types exhibit significant variations in stability, pricing, and associated public services (Lin et al., 2020 ). Furthermore, differences in migrants’ educational backgrounds shape distinct housing choice patterns and settlement decisions across groups by influencing their income levels and access to policy support (Wang, 2024 ; Liu et al., 2021 ). However, migrants in major cities often struggle to balance work and residence (Wang & Shen, 2022 ; Liu et al., 2021 ). Many migrants adopt strategies that sacrifice residential quality to ensure working stability, leading them into long-term temporary residence status, while only a few migrants with dual resource advantages achieve permanent settlement; the remainder who prioritize residential security choose to return to their hometowns (Zhu & Chen, 2010 ; Wang & Shen, 2023 ; Yue et al., 2010 ). Therefore, promoting the transition from temporary residence to permanent settlement among migrants represents a pressing practical issue. However, settlement intention as a complex multi-dimensional decision-making process requires more comprehensive multidimensional analytical frameworks for in-depth interpretation 2.2 Life Course and Migration Settlement Intention Life course theory provides a framework for understanding migration dynamics through the interaction between individual life events and their historical spatiotemporal contexts. (Elder, 1994 ; Elder et al., 2003 ). Marriage and childbearing events, as structurally significant turning points in the life course, fundamentally influence individuals’ spatial choices and settlement intentions by reconfiguring family responsibility networks and resource allocation patterns (Kulu & Milewski, 2007 ; Mulder & Wagner, 2015 ; Courgeau, 1990 ). In Europe and North America, married groups with children demonstrate significantly lower migration frequency than single groups and are more inclined to settle in areas with better educational resources and working opportunities (De Jong & Graefe, 2008 ; Findlay et al., 2015 ; Geist & McManus, 2008 ). Under China’s unique household registration system, the impact of marriage and childbearing on settlement intention exhibits distinctive complexity (Fan, 2011 ; Zhu & Lin, 2014 ). The strong linkage between marriage and housing makes housing an important prerequisite for marriage establishment, a cultural phenomenon that further reinforces married populations’ demand for housing stability (Wu & Gaubatz, 2020 ; Yang & Guo, 2018 ). Meanwhile, the high coupling mechanism between the household registration system and educational resource allocation results in multiple institutional barriers for migrant children without local household registration when enrolling in urban public schools, including additional borrowing fees and restrictions on secondary and higher education entrance exam policies (Wang & Shen, 2023 ; Wu, 2004 ). These educational access thresholds force many migrant families to face difficult choices when their children enter compulsory education: either bear the high costs of private education or send their children back to their hometowns for education, thus creating a contradiction between "high settlement needs and high institutional barriers" (Huang et al., 2020 ; Li & Liu, 2019 ; Hu et al., 2011 ). Although existing research has revealed the complex associations between life course and settlement intention (Cao et al., 2015 ; Fan, 2011 ), current analytical methods mostly rely on traditional linear regression and other econometric models, making it difficult to precisely capture the nonlinear effects and threshold effects involved. Therefore, this study employs machine learning methods to better identify and model the nonlinear influence mechanisms of life course events on settlement intention. 3 Methodology 3.1 Analytical framework [Figure 1 about here] As illustrated in Fig. 1 , this study constructs a threshold effect analytical framework grounded in push-pull theory. The threshold effect describes a nonlinear process where response variables undergo significant changes once stimuli exceed critical values (Zou et al., 2023 ). Based on this theoretical foundation, the framework systematically analyzes the long-term settlement decision-making mechanisms of small-town youth with different educational backgrounds in large cities through two core dimensions: working and residence. Working and residence factors exhibit threshold effect characteristics in metropolitan environments. Working elements, including income levels and career development opportunities, along with residence elements represented by housing prices and living conditions, constitute pull factors attracting migrants (Wang & Shen, 2023 ). When urban living costs exceed an individual’s critical tolerance threshold, pull factors transform into push factors, prompting individuals to reconsider their residential choices. This transformation process essentially reflects migrants’ trade-offs between working and residence: to obtain superior working conditions, migrants typically must bear higher housing costs and residential pressures, seeking equilibrium between career development potential and residential economic burden (Liu & Xiao, 2022 ). Educational background differentiates threshold sensitivity among migrant populations. Highly educated groups possess stronger economic capacity and policy support, enabling them to tolerate higher urban costs (Wang, 2024 ). Less educated groups face working limitations and economic constraints, encountering earlier threshold critical points (Khoo et al., 2008 ). This difference stems from varying abilities to access urban resources rather than different preference patterns (Wang & Shen, 2022 ). Life-stage characteristics further moderate threshold effects and priority rankings. Single individuals prioritize career advancement over maintaining living stability, willing to compromise residential comfort for professional development opportunities (Coulter & Van Ham, 2013 ). Families with children face multidimensional resource allocation challenges: they must simultaneously balance personal career development, family economic responsibilities, and children’s educational investments. The superposition of multiple obligations significantly compresses their urban adaptation buffer space (Gong et al., 2011 ). This differentiation pattern reflects the dynamic balancing strategies of individuals at different life stages between pursuing development, maintaining stability, and cultivating belonging (Vidal & Lutz, 2018 ). Based on this conceptual framework, we propose the following hypotheses: Hypothesis 1 In large cities, working and residence factors exhibit threshold effects on migration intentions, where pull factors transform into push factors once critical boundaries are exceeded. Hypothesis 2 Highly educated and less educated individuals have different threshold boundaries, with highly educated groups tolerating higher urban costs due to higher incomes. Hypothesis 3 Within the same educational level, life-course stages moderate threshold effects, with unmarried individuals showing highest tolerance for urban challenges, followed by married individuals without children, and married individuals with children showing lowest tolerance. 3.2 Research area and data description [Figure 2 about here] This study uses data from the 2017 China Migrants Dynamic Survey (CMDS) by China’s National Health Commission. The survey employed stratified, multi-stage probability proportional to size (PPS) sampling across all 31 provinces, targeting floating migrants aged 15 + with non-local hukou who had resided locally for over one month, yielding 169,989 samples. To examine small-to-large city migration patterns, systematic screening was applied: current residence limited to top 10 GDP cities in 2017 (Beijing, Shanghai, Shenzhen, Guangzhou, Chongqing, Suzhou, Chengdu, Wuhan, Hangzhou, Tianjin) and provincial capitals; hukou registration restricted to county-level areas and below; “employer" samples excluded for homogeneity. This yielded 64,623 valid observations (Fig. 2 ). Settlement intention is the dependent variable, derived from expected local residence duration and recoded as binary: 6 + years coded as 1 (long-term intention), others as 0 (short-term intention). Among samples, 42.60% expressed long-term settlement intentions. Table 1 shows variable definitions and descriptive statistics. Control variables include individual characteristics (education, life course, gender, age), economic variables (income, industry, income gap), and housing economic variables (price gap, affordability). Housing type serves as the core explanatory variable with nine categories from workplace accommodation to self-purchased housing. Table 1 Definitions and descriptive statistics of variables for small-town migrants Variable Type Definition and Coding Numbers Percentage Population Classification Education level = 0 if high school or below 48123 74.47% = 1 if college degree or above 16500 25.53% Life course = 0 if unmarried = 1 if married without children = 2 if married with children Avg. 1.62 (std. 0.75) Dependent Variable Settlement intention = 0 if short-term stay (≤ 5years) 37,081 57.40% = 1 if long-term settlement (≥ 6 years & permanent settlement) 27,540 42.60% Independent Variable Housing Variables Housing price gap Destination city average housing price minus origin city average housing price Avg. 14,345.06 (std. 16,260.69) Housing affordability Destination city average housing price × 90㎡ ÷ (annual household income) Avg. 20.68 (std. 32.05) Housing affordability gap Housing affordability in destination city minus housing affordability in origin city Avg. 13.94 (std. 27.34) Housing type = 1 if workplace accommodation = 2 if employer/unit housing = 3 if borrowed housing = 4 if other informal housing = 5 if government public rental housing = 6 if private rental housing = 7 if self-purchased commercial housing = 8 if self-purchased affordable housing = 9 if self-purchased limited property rights housing Avg. 5.80 (std. 1.72) Working Variables Monthly personal income Monthly personal income (yuan) Avg. 4,336.82 (std. 3,157.52) Average wage level Average monthly personal income in destination city Avg. 4,336.82 (std. 814.47) Average wage gap Destination city average income minus national average income Avg. 0.12 (std. 734.93) Control Variables Gender = 0 if female 32,655 50.53% = 1 if male 31,968 49.47% Number of children Number of children in household Avg. 1.28 (std. 0.92) Family number Number of family members Avg. 3.10 (std. 1.19) Birth year Year of birth Avg. 1981.56 (std. 10.52) Industry classification = 1 if employed in knowledge-intensive services = 2 if employed in labor-intensive services = 3 if employed in manufacturing = 4 if employed in raw materials and resources Avg. 2.23 (std. 0.74) [Table 1 about here] 3.3 XGBoost-SHAP Modeling Method Given the non-linear relationships among factors in settlement intention decisions and data imbalance, this study employs XGBoost (eXtreme Gradient Boosting) to construct a settlement intention prediction model. XGBoost, as a gradient boosting decision tree ensemble algorithm, automatically identifies non-linear relationships and complex interactions among variables. The study uses Random Forest for feature selection and importance ranking, then applies XGBoost with fixed hyperparameters: 100 trees, maximum depth of 5, learning rate of 0.05, subsample ratio of 0.8, feature subsampling of 0.8, minimum child weight of 1, and binary:logistic objective function. Feature matrix X contains core explanatory variables, while target variable y represents binary long-term settlement intention. Data is split 8:2 for training and testing. For model interpretability, Shapley Additive exPlanations (SHAP) is employed. Based on cooperative game theory, SHAP provides fair feature contribution allocation through the Shapley value formula: $$\:\phi\:ⱼ\:=\:\sum\:[S\subseteq\:F\backslash\:\{j\left\}\right]\:\left|S\right|!(n-|S|-1)!/n!\:\times\:\:\left[f\right(S\cup\:\left\{j\right\})\:-\:f(S\left)\right]$$ where S represents feature subsets excluding feature j, F is the complete feature set, n is total features, and f(S) is model output using subset S. Feature importance is calculated via global mean absolute SHAP values: $$\:Iⱼ\:=\:(1/N)\:\sum\:[i=1\:to\:N]\:\left|\phi\:ⱼⁱ\right|$$ where N is total samples and φⱼⁱ is the SHAP value of feature j for sample i. The model achieves approximately 74% training accuracy and 71–73% testing accuracy, demonstrating stable performance. SHAP values are computed using TreeExplainer, extracting positive class contributions to analyze feature influence on long-term settlement intention. 4 Results 4.1 OLS Analysis and Variable SHAP Interpretation [Figure 3 about here] To ensure the robustness of the model estimation, we first diagnosed and addressed multicollinearity among all potential predictors (Fig. 3 ). An analysis of the Variance Inflation Factor (VIF) for the initial set of variables revealed that housing affordability gap (VIF = 49.2), housing price gap (VIF = 51.1), average wage level (VIF = 178.1), and birth year (VIF = 223.2) exhibited severe multicollinearity, with VIF values substantially exceeding the conventional threshold of 5.0. Through the stepwise removal of collinear variables while retaining the most theoretically salient predictors, all VIFs in the final optimized set were reduced to acceptable levels (e.g., housing type, VIF = 5.0; monthly personal income, VIF = 3.1; average wage gap, VIF = 1.3). This process effectively mitigates the potential for multicollinearity to bias the subsequent model results. Based on this optimized set of variables, we first employed a logistic regression model to conduct a preliminary examination of the factors influencing settlement intention (Fig. 4 , left panel). The results indicated that education level, life course, personal housing affordability, housing type, monthly personal income, average wage gap, gender, and children number were all statistically significant predictors of settlement intention (p < .05). Specifically, the adjusted odds ratio (OR) for education level was 2.451 (95% CI: 2.3479–2.5586), indicating that individuals with higher education had significantly greater odds of intending to settle. Similarly, life course (OR = 1.6521), housing type (OR = 1.2992), monthly personal income (OR = 1.0551), and average wage gap (OR = 1.1089) were all positively associated with settlement intention, whereas personal housing affordability exerted a negative influence (OR = 0.9957). However, the logistic regression model inherently assumes a linear relationship between the predictors and the log-odds of the outcome, making it ill-suited for capturing more complex non-linear patterns and threshold effects. [Figure 4 about here] To overcome this limitation, we further employed an XGBoost model and utilized SHAP (SHapley Additive exPlanations) values for a more in-depth interpretation (Fig. 4 , right panel). The SHAP summary plot provides a comprehensive visualization of the global importance of each feature, as well as the direction and magnitude of its impact on the model's predictions. Housing type exhibited a distinct bifurcation in its SHAP values: low feature values (representing non-ownership) corresponded almost exclusively to negative contributions, while high values (representing ownership) yielded significant positive contributions, highlighting the decisive role of obtaining property rights. Education level showed a similarly stratified effect, with high feature values clustering in the positive SHAP range and low values concentrating in the negative range. Particularly noteworthy is the complex, non-linear relationship demonstrated by housing affordability: while low to moderate levels of housing pressure (blue dots) exerted a negative influence, extremely high levels (red dots) paradoxically showed a positive rebound effect on settlement intention. Furthermore, the wide, bidirectional distribution of SHAP values for life course, monthly personal income, and average wage gap strongly indicates that their effects are non-monotonic and characterized by significant thresholds. In contrast, gender and children number had more concentrated SHAP value distributions, suggesting they possessed comparatively weaker predictive power overall. 4.2 PDP Analysis: Nonlinear Relationships and Threshold Effects Partial dependence plots (Figs. 5 and 6 ) were employed to identify nonlinear influences and threshold effects on settlement intentions across education levels and life course stages. 4.2.1 Working and Settlement Intention [Figure 5 about here] For the high-education group, the influence of average wage gap on settlement intention reveals significant threshold effects and life-course heterogeneity (Fig. 5 , Panels A1-A3). Compared to their peers with children, unmarried and married-childless individuals exhibit greater tolerance for cities where wages are below the national average. For the unmarried group, SHAP values remain positive even when the wage gap is below -¥991, suggesting that cities with significantly lower wages can still hold some attraction. Beyond this point, the effect fluctuates narrowly around zero, indicating saturation. The married-childless group shows a significant positive effect when the gap is below -¥1,321 or above ¥208, but a slight negative effect within the intermediate range of -¥1,321 to ¥208, implying a sensitivity to moderate wage levels. In stark contrast, the married-with-children group has the lowest tolerance for low-wage cities, showing a significant negative effect in the -¥2,996 to -¥1,686 range. Once the gap exceeds -¥1,686, the effect flattens, reflecting a diminished sensitivity to wage gaps after a basic income threshold is met. The threshold effects of average wage gap for the low-education group present a starkly different pattern (Fig. 5 , Panels B1-B3). Contrary to their high-education counterparts, a negative wage gap has a negative impact on unmarried low-education individuals. Once the gap exceeds ¥920, the positive effect grows continuously, indicating this group prioritizes the absolute advantage of actual wages. For the married-childless low-education group, the effect is positive when the wage gap is below ¥384 but turns negative beyond this threshold—a pattern nearly opposite to that of their high-education peers. While the married-with-children low-education group shares a similar pattern with its high-education counterpart (i.e., low tolerance for low-wage cities), it is more sensitive to the threshold: any wage gap below zero exerts a negative influence, and the effect only turns positive after exceeding ¥863. This may reflect more severe household economic pressures, leading to stricter income requirements. The impact of monthly personal income on the high-education group also varies significantly across life course stages (Fig. 5 , Panels a1-a3). The unmarried group has an "optimal income range" between ¥4,949 and ¥27,719, within which the effect is positive. Incomes below or, notably, above this range produce a significant negative effect, suggesting a clear "satisfaction" threshold rather than a "higher-is-better" logic. The pattern for the married-childless group is inverted, showing a significant positive effect only after income surpasses ¥20,468, indicating that higher family living costs elevate their income threshold. The married-with-children group exhibits an inverted U-shaped pattern: the effect is positive within the ¥4,280 to ¥20,066 range, but beyond this upper limit, further income increases lead to a sharp decline into negative territory. This likely reflects the trade-offs between high-paying jobs, their associated pressures, and multiple household expenditures. In comparison, the response of the low-education group to monthly personal income is more linear and monotonic (Fig. 5 , Panels b1-b3). Unlike the complex patterns of the high-education group, the settlement intention of unmarried and married-with-children low-education individuals is largely positively correlated with their income. Once their monthly income crosses the thresholds of ¥4,147 and ¥5,217, respectively, the SHAP values turn positive and continue to increase with rising income. Only the married-childless subgroup is an exception, showing a positive effect within a high-income bracket of ¥19,899 to ¥34,448. Overall, the low-education group responds more directly to income gains, with lower activation thresholds for positive effects and little evidence of a negative effect at very high income levels. Synthesizing the analysis of these two employment factors reveals that the high-education group tolerates low-wage cities before having children but avoids them afterward. In contrast, the low-education group generally pursues wages that are absolutely higher than the national average. Regarding personal income, very high earnings can act as a push factor for unmarried and parent high-education individuals, whereas for their low-education counterparts, it consistently serves as a powerful pull factor. This divergence profoundly reflects the heterogeneous strategies employed by groups with different educational capital when balancing employment opportunities against the cost of living. 4.2.2 Housing and Settlement Intention [Figure 6 about here] The effect of housing affordability on settlement intention is predominantly negative, functioning as an inhibitor. However, its mechanism exhibits complex non-linear characteristics across different groups, most notably a counter-intuitive positive rebound under extreme pressure (Fig. 6 , Panels A1-B3). Among the high-education group, the settlement intention of unmarried individuals turns consistently negative once the housing affordability ratio exceeds 13.49 years. A similar trend is observed for the married-childless group, but after the ratio surpasses 61.37 years, the SHAP values unexpectedly rebound, exerting a significant positive influence. This rebound pattern also appears for the married-with-children group, with the threshold occurring around 319 years. The low-education group's response likewise generally follows this "low-pressure-inhibition, high-pressure-rebound" pattern, albeit with vastly different thresholds. Unmarried individuals in this group paradoxically show positive settlement intention within an extremely high-pressure range of 247.53 to 301.39 years, while the married-childless group's intention turns positive after the ratio exceeds 96.72 years. The most complex pattern belongs to the married-with-children group, who exhibit a positive effect across a broad range below 2221.99 years, with the effect only turning negative beyond this critical point. In contrast, the influence of housing type reveals a clear "property rights divide," with a significant step-change in effect between different tenure statuses (Fig. 6 , Panels a1-b3). Regardless of education level or life course stage, housing categories 7–9 (which encompass various forms of homeownership such as self-purchased commercial housing, affordable housing, and limited-property-rights housing) provide a strong and significant positive impetus to settlement intention. All other non-ownership categories, conversely, act primarily as inhibitors to varying degrees. A noteworthy exception is that among high-education unmarried and married-childless individuals, there is a certain tolerance for housing types 3–5 (e.g., borrowed housing, other informal housing, and government public rentals). Their SHAP values are slightly above zero, suggesting these transitional housing options still hold a modest attractive power for them. This characteristic is also observed among low-education individuals who are married with children. Overall, we find that housing affordability generally functions as a "push factor" that inhibits settlement intention. However, low-education groups, particularly those married with children, display a paradoxical tolerance for extreme housing pressure. This profoundly reflects the trade-offs they are forced to make to access employment opportunities in large cities. Simultaneously, the effect of housing type is more absolute, demonstrating a clear bifurcation based on property rights. The transition from renting to owning represents a decisive "pull factor" for all groups, dramatically enhancing settlement intention. This sense of stability and belonging conferred by homeownership is especially pronounced among the more resource-constrained low-education group, underscoring the pivotal role of acquiring property rights in their settlement decisions. 5 Discussion & conclusion Existing literature has firmly established the important role of economic incentives, urban amenities, and policy factors in talent mobility (Wang & Shen, 2022 ; Liu & Wang, 2020 ; Fan, 2011 ), but the non-linear mechanisms of employment and housing factors have received insufficient attention. Traditional research often employs econometric methods like linear regression to analyze migrants' settlement intention, assuming a monotonic relationship between influencing factors and the outcome (De Jong & Graefe, 2008 ; Wagner & Mulder, 2015 ). However, this linear assumption may fail to capture complex threshold effects and the mechanisms of critical point transitions. Employing the XGBoost-SHAP machine learning method and data from the China Migrants Dynamic Survey, this study develops an analytical framework based on the life course perspective to systematically examine the non-linear effects of employment and housing on the settlement intention of migrant populations with different education levels in small cities. The study identifies the critical boundaries where push-pull factors transform, pinpoints differentiated threshold sensitivities across educational divides, and reveals the institutional moderating mechanisms of life course events, offering a new theoretical perspective for understanding the underlying logic of talent competition in large cities. The results confirm Hypothesis 1 : employment (working) and housing factors exhibit significant non-linear threshold effects on settlement intention. This effect is particularly pronounced in the dimension of housing tenure (housing type): regardless of a migrant's educational background or life course stage, acquiring property rights is consistently the most powerful driver of settlement intention. More interestingly, we find that the “lower bound” of these thresholds plays a more critical role in decision-making than the “upper bound.” For the high-education group, the response to average wage gap and monthly personal income presents a seemingly paradoxical pattern: unmarried individuals show a moderate preference for cities with above-average wages but a strong aversion to cities with extremely high personal incomes; meanwhile, individuals with children significantly avoid cities with low average wages and are also negatively disposed toward cities with high personal incomes. This finding aligns with theoretical expectations from urban economics concerning the diminishing marginal returns of agglomeration (Duranton & Puga, 2004 ; Ottaviano & Peri, 2006 ). That is, once a city reaches a certain level of development, the economic advantages of matching, sharing, and learning are offset by high congestion costs and living expenses. Migrants' decisions, therefore, are not about maximizing returns but about making trade-offs within an acceptable satisficing range. Contrary to the common expectation of Hypothesis 2 , the analysis further reveals that high-education migrants actually have a lower tolerance for the pressures of large cities than their low-education counterparts. This anomaly is consistently validated across multiple indicators. Regarding wage levels, high-education individuals (especially the unmarried and married-childless) show greater tolerance for cities with below-average wages, whereas low-education individuals gravitate decisively toward regions with above-average wages. In terms of housing affordability, when the price-to-income ratio reaches extremely high levels, low-education individuals with children even exhibit a strong positive settlement intention, whereas high-education individuals turn negative at a much lower pressure threshold. We posit that this difference is rooted in the varying ways their human capital advantages are manifested in spatial choices. With their versatile human capital, high-education migrants can secure decent employment and a reasonable price-to-income ratio even in small and medium-sized cities, thus achieving a work-life balance (Li et al., 2022 ; Kerr et al., 2017 ). In contrast, for low-education migrants, high-quality job opportunities in large cities are almost the sole pathway to higher income (Asher & Novosad, 2020 ). Consequently, when housing prices in large cities cross a critical threshold, a large number of low-education individuals are forced to make a compromise between employment and residence, accepting a long-term state of non-propertied residence. This mode of settlement, lacking equal access to local public services and social welfare, is essentially a form of “incomplete” or “floating” settlement—a state of precarious sojourning rather than permanent integration. The analysis further validates the expectation of Hypothesis 3 regarding differentiated life course impacts, but also reveals significant heterogeneity in this effect across educational groups. Among the high-education group, the progression through the life course aligns with theoretical expectations: individuals married with children, facing child-rearing issues tied to the hukou system, have the lowest tolerance for expensive and competitive large cities, yet they also cannot accept small cities with inadequate infrastructure and opportunities. The low-education group, however, exhibits a counter-intuitive pattern: even after having children, they maintain a strong preference for megacities. We infer that this phenomenon stems from the unequal distribution of institutional support across social strata. High-education talent can more easily obtain urban hukou and access to public education through talent acquisition policies, enabling them to make trade-offs between cities of different tiers and ultimately settle in a medium-sized city with a higher quality of life. Conversely, due to institutional constraints, low-education individuals can often only meet their children's educational needs by returning to their hometowns. However, returning home means facing scarce job opportunities and lower income. This structural dilemma compels them to remain in large cities that offer better economic returns even after having children—a trend that may be mechanistically linked to the widespread phenomenon of “left-behind children” in China. This study expands the analytical framework of talent mobility research along two dimensions. On the theoretical front, first, by identifying threshold effects for employment and housing factors, the study transcends the linear framework of traditional push-pull theory, demonstrating that key factors have distinct points of critical transition and providing a more refined analytical lens for understanding migration complexity. Second, the study reveals the negative correlation between education level and tolerance for urban pressures, and clarifies how the moderating effect of life course events is heterogeneous by educational background. This finding challenges the conventional wisdom that higher human capital equates to greater adaptability in large cities, proposing that the true advantage of human capital lies not merely in enhancing survival capacity in large cities, but in expanding an individual’s spatial choice architecture, thereby fundamentally altering their migration decision-making logic. On the methodological front, this research demonstrates the potential of machine learning methods (XGBoost-SHAP) in social science research. This approach enables the effective identification of complex non-linear relationships and interaction effects that are often missed by traditional linear regression models, providing a powerful tool for precisely measuring thresholds and critical points. This not only enhances the robustness of our findings but also offers a new methodological toolkit for future migration research. This study also holds significant policy implications. First, in the current landscape of fierce talent competition, the most competitive cities may not be the megacities with prohibitive living costs, nor small towns lacking opportunities, but rather the medium-sized cities that offer an optimal balance between job opportunities, public services, and cost of living. These cities possess a unique competitive advantage in attracting and retaining high-skilled talent by providing both robust infrastructure and quality employment for them and their families. Second, policymakers must pay close attention to the long-term well-being of the low-skilled migrant population in large cities. This group, while contributing immensely to urban development, is often excluded from property ownership and quality public services (especially education for their children), leaving them in a long-term floating state of economic inclusion but social exclusion. This not only undermines social equity but may also pose long-term social risks. Therefore, this situation requires coordinated and more inclusive urbanization strategies from both sending and receiving governments, such as reforming the hukou system, expanding the supply of affordable housing, and opening up school enrollment channels for migrant children to help them achieve genuine social integration. This study provides a crucial foundation for future research on population or talent mobility. Subsequent research should further explore the dynamic nature of these threshold effects and delve deeper into their underlying psychological and institutional mechanisms. One promising avenue is to use longitudinal data to examine how threshold parameters evolve across different stages of urban development and how the interplay between housing and employment factors jointly shapes settlement decision thresholds. Another valuable line of inquiry is to analyze the interaction between life course events and the institutional environment, particularly the long-term impact of policy changes on individual decision-making trajectories. Furthermore, given the significant differences in institutional structures and cultural contexts across countries, conducting cross-national comparative studies to test the universality and particularity of these threshold effects and educational divides would undoubtedly enrich our understanding of talent mobility as a global phenomenon. Declarations Funding This research is based upon work supported by the National Natural Science Foundation of China (NSFC) [Grant No. 42271219, 42171203]; and the Fundamental Research Funds for the Central Universities [Grant No. 413100119]. Ethical Approval Statement This study is a secondary analysis of the 2017 China Migrants Dynamic Survey (CMDS) dataset. The data were collected by the National Health Commission of the People's Republic of China. All data used in this study were fully anonymized before we accessed them. As this research involved no direct contact with human participants and utilized a publicly available, de-identified dataset, it did not require separate ethical approval from an institutional review board, in accordance with the institutional and national research ethics regulations for secondary data analysis. The ethical procedures for the original data collection were the responsibility of the institution that conducted the survey, the National Health Commission of China, which performed the research in accordance with relevant national guidelines and regulations. Informed Consent Statement This study is based on a secondary analysis of a publicly available dataset. Therefore, the authors of this study did not obtain informed consent directly from the participants. Informed consent was obtained from all individual participants included in the original survey by the data collectors from the National Health Commission of China at the time of data collection. The publicly available dataset used in our analysis has been fully de-identified, ensuring participant confidentiality. References Asher S, Novosad P (2020) Costs and benefits of rural-urban migration: Evidence from India. J Dev Econ 146:104481. https://doi.org/10.1016/j.jdeveco.2020.104481 Cao G, Li M, Ma Y, Tao R (2015) Self-employment and intention of permanent urban settlement: Evidence from a survey of migrants in China. J Dev Econ 114:24–42. https://doi.org/10.1016/j.jdeveco.2015.02.004 Chan KW (2010) Fundamentals of China’s urbanization and policy. China Rev 10(1):63–93 Coulter R, Van Ham M (2013) Following people through time: An analysis of individual residential mobility biographies. Hous Stud 28(7):1037–1055. https://doi.org/10.1080/02673037.2013.783903 Courgeau D (1990) Migration, family, and career: A life course approach. In: Baltes PB, Featherman DL, Lerner RM (eds) Life-span development and behaviour. Psychology De Jong GF, Graefe DR (2008) Family life course transitions and the economic consequences of internal migration. Popul Space Place 14:267–282. https://doi.org/10.1002/psp.481 Du H, Hui EC, Chen L (2019) Measuring the housing precarity of migrant private tenants and assessing its impact on their life satisfaction. Cities 90:1–12. https://doi.org/10.1016/j.cities.2019.01.031 Duranton G, Puga D (2004) Micro-foundations of urban agglomeration economies. In J. V. Henderson & J. F. Thisse (Eds.), Handbook of Regional and Urban Economics (Vol. 4, pp. 2063–2117). Elsevier. https://doi.org/10.1016/S1574-0080(04)80005-1 Elder GH Jr. (1994) Time, human agency, and social change: Perspectives on the life course. Social Psychol Q 57(1):4–15. https://doi.org/10.2307/2786971 Elder GH Jr., Johnson M, Crosnoe R (2003) The emergence and development of life course theory. In J. T. Mortimer & M. Shanahan (Eds.), Handbook of the life course (pp. 3–22). Kluwer. https://doi.org/10.1007/978-0-306-48247-2_1 Fan CC (2011) Settlement intention and split households: Findings from a survey of migrants in Beijing’s urban villages. China Inform 25(2):253–274. https://doi.org/10.1177/0920203X11407906 Findlay A, McCollum D, Coulter R, Gayle V (2015) New mobilities across the life course: A framework for analysing demographically linked drivers of migration. Popul Space Place 21(4):390–402. https://doi.org/10.1002/psp.1956 Gan X, Zuo J, Ye K, Li D, Chang R, Zillante G (2016) Are migrant workers satisfied with public rental housing? A study in Chongqing, China. Habitat Int 56:96–102. https://doi.org/10.1016/j.habitatint.2016.05.006 Geist C, McManus PA (2008) Geographical mobility over the life course: Motivations and implications. Popul Space Place 14(4):283–303. https://doi.org/10.1002/psp.508 Gong F, Xu J, Fujishiro K, Takeuchi DT (2011) A life course perspective on migration and mental health among Asian immigrants: The role of human agency. Soc Sci Med 73(11):1618–1626. https://doi.org/10.1016/j.socscimed.2011.09.014 Gu H (2021) Understanding the migration of highly and less-educated labourers in post-reform China. Appl Geogr 137:102605. https://doi.org/10.1016/j.apgeog.2021.102605 Hu F, Xu Z, Chen Y (2011) Circular migration, or permanent stay? Evidence from China’s rural–urban migration. China Econ Rev 22(1):64–74. https://doi.org/10.1016/j.chieco.2010.09.007 Huang S, Hou J, Sun L, Dou D, Liu X, Zhang H (2017) The effects of objective and subjective socioeconomic status on subjective well-being among rural-to-urban migrants in China: The moderating role of subjective social mobility. Front Psychol 8:819. https://doi.org/10.3389/fpsyg.2017.00819 Huang Y, Liu W, Zhuo Y, Sun X (2020) Social insurance participation and urban settlement intentions of China’s floating population. Popul Space Place 26(8):e2366. https://doi.org/10.1002/psp.2366 Kerr SP, Kerr WR, Özden Ç, Parsons C (2017) Global talent flows. J Economic Perspect 31(3):83–106. https://doi.org/10.1257/jep.31.3.83 Khoo SE, Hugo G, McDonald P (2008) Which skilled temporary migrants become permanent residents and why? Int Migrat Rev 42(1):193–226. https://doi.org/10.1111/j.1747-7379.2007.00118.x Kulu H, Milewski N (2007) Family change and migration in the life course: An introduction. Demographic Res 17:567–590. https://doi.org/10.4054/DemRes.2007.17.19 Li L, Liu Y (2019) Understanding the gap between de facto and de jure urbanization in China: A perspective from rural migrants’ settlement intention. Popul Res Policy Rev 39:311–338. https://doi.org/10.1007/s11113-019-09532-1 Li Y, Chen X, Wang Z (2022) Tolerance for housing unaffordability among highly skilled young migrants: Evidence from the Zhejiang Province of China. Int J Environ Res Public Health 20(1):616. https://doi.org/10.3390/ijerph20010616 Lin S, Wu F, Li Z (2020) Social integration of migrants across Chinese neighbourhoods. Geoforum 112:118–128. https://doi.org/10.1016/j.geoforum.2020.04.009 Liu T, Wang J (2020) Bringing city size in understanding the permanent settlement intention of rural–urban migrants in China. Popul Space Place 26(4):e2295. https://doi.org/10.1002/psp.2295 Liu T, Xiao W (2022) The ‘intention to stay’ of the floating migrant population: A spatiotemporal meta-analysis in Chinese cities. Popul Space Place 28(4):e2574. https://doi.org/10.1002/psp.2574 Liu Y, Liu Y, Lin Y (2021) Upward or downward comparison? Migrants’ socioeconomic status and subjective wellbeing in Chinese cities. Urban Stud 58(12):2490–2513. https://doi.org/10.1177/0042098020968965 Liu Z, Wang Y, Chen S (2016) Does formal housing encourage settlement intention of rural migrants in Chinese cities? A structural equation model analysis. Urban Stud 54(8):1834–1850. https://doi.org/10.1177/0042098016634979 Luan X, Paryzat H, Chu J, Liu W, Liu Z, Li C (2024) Different roads take me home: The nonlinear relationship between distance and flows during China’s Spring Festival. Humanit Social Sci Commun 11:1356. https://doi.org/10.1057/s41599-024-03779-8 Ottaviano GI, Peri G (2006) The economic value of cultural diversity: Evidence from US cities. J Econ Geogr 6(1):9–44. https://doi.org/10.1093/jeg/lbi002 Vidal S, Lutz K (2018) Internal migration over young adult life courses: Continuities and changes across cohorts in West Germany. Adv Life Course Res 36:45–56. https://doi.org/10.1016/j.alcr.2018.03.003 Wagner M, Mulder CH (2015) Spatial mobility, family dynamics, and housing transitions. Kölner Zeitschrift für Soziologie und Sozialpsychologie , 67 (Supplement 1), 111–135. https://doi.org/10.1007/s11577-015-0327-4 Wang C, Shen J (2022) Settlement intention of migrants in urban China: The effects of labor-market performance, employment status, and social integration. Appl Geogr 147:102773. https://doi.org/10.1016/j.apgeog.2022.102773 Wang C, Shen J (2023) How subjective economic status matters: The reference-group effect on migrants’ settlement intention in urban China. Asian Popul Stud 19(1):105–123. https://doi.org/10.1080/17441730.2021.2012025 Wang CL, Shen J, Liu Y (2021) Administrative level, city tier, and air quality: Contextual determinants of hukou conversion for migrants in China. Comput Urban Sci 1(1):1–16. https://doi.org/10.1007/s43762-021-00015-0 Wang L, Acheampong RA, He S (2020) High-speed rail network development effects on the growth and spatial dynamics of knowledge-intensive economy in major cities of China. Cities 105:102772. https://doi.org/10.1016/j.cities.2020.102772 Wang L, Duan X (2018) High-speed rail network development and winner and loser cities in megaregions: The case study of Yangtze River Delta, China. Cities 83:71–82. https://doi.org/10.1016/j.cities.2018.06.010 Wang Y (2024) Individual preferences: Jobs or amenities? Understanding the destination choices of highly and less educated migrants in China. Appl Geogr 162:103169. https://doi.org/10.1016/j.apgeog.2023.103169 Wei YD, Wu Y, Xiao W, Wu X (2023) Preferences and barriers of vulnerable urban populations in affordable housing in urban China. J Urban Plan Dev 149(4):05023030. https://doi.org/10.1061/JUPDDM.UPENG-4271 Wu F (2004) Residential relocation under market-oriented redevelopment: The process and outcomes in urban China. Geoforum 35(4):453–470. https://doi.org/10.1016/j.geoforum.2004.01.004 Wu F, Gaubatz P (2020) The Chinese City. Routledge. https://doi.org/10.4324/9780429052682 Wu W, Stephens M, Du M et al (2019) Homeownership, family composition and subjective wellbeing. Cities 84:46–55. https://doi.org/10.1016/j.cities.2018.07.008 Yang S, Guo F (2018) Breaking the barriers: How urban housing ownership has changed migrants’ settlement intentions in China. Urban Stud 55(16):3689–3707. https://doi.org/10.1177/0042098017741265 Yue Z, Li S, Feldman MW, Du H (2010) Floating choices: A generational perspective on intentions of rural–urban migrants in China. Environ Plann A 42(3):545–562. https://doi.org/10.1068/a4213 Zhu Y (2007) China’s floating population and their settlement intention in the cities: Beyond the Hukou reform. Habitat Int 31(1):65–76. https://doi.org/10.1016/j.habitatint.2006.04.002 Zhu Y, Chen W (2010) The settlement intention of China’s floating population in the cities: Recent changes and multifaceted individual-level determinants. Popul Space Place 16(4):253–267. https://doi.org/10.1002/psp.544 Zhu Y, Lin L (2014) Continuity and change in the transition from the first to the second generation of migrants in China: Insights from a survey in Fujian. Habitat Int 42:147–154. https://doi.org/10.1016/j.habitatint.2013.12.002 Zou D, Li Q, Zhou Y, Liang S, Zhou S (2023) Understanding the nonlinear relationship between the built environment and non-mandatory travel: A machine learning approach. Appl Geogr 161:103118. https://doi.org/10.1016/j.apgeog.2023.103118 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 11 Apr, 2026 Reviews received at journal 09 Apr, 2026 Reviews received at journal 14 Feb, 2026 Reviewers agreed at journal 16 Jan, 2026 Reviewers agreed at journal 15 Jan, 2026 Reviewers invited by journal 14 Jan, 2026 Editor assigned by journal 11 Nov, 2025 Editor invited by journal 11 Nov, 2025 Submission checks completed at journal 09 Nov, 2025 First submitted to journal 09 Nov, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7845488","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":575512986,"identity":"a1a24f64-bfb7-4673-9c5e-38015faafbda","order_by":0,"name":"Xinyi Shu","email":"","orcid":"","institution":"Wuhan University","correspondingAuthor":false,"prefix":"","firstName":"Xinyi","middleName":"","lastName":"Shu","suffix":""},{"id":575512989,"identity":"6a46dd92-9ed6-4831-98e8-ef391200eefb","order_by":1,"name":"Hurex Paryzat","email":"","orcid":"","institution":"Wuhan University","correspondingAuthor":false,"prefix":"","firstName":"Hurex","middleName":"","lastName":"Paryzat","suffix":""},{"id":575512991,"identity":"f9005e9e-930e-4099-b9e1-c0a6576bb871","order_by":2,"name":"Kaili Zhang","email":"","orcid":"","institution":"Wuhan University","correspondingAuthor":false,"prefix":"","firstName":"Kaili","middleName":"","lastName":"Zhang","suffix":""},{"id":575512993,"identity":"826e5ebb-3aec-4216-9b41-8a49930806df","order_by":3,"name":"Shijie Chen","email":"","orcid":"","institution":"Wuhan University","correspondingAuthor":false,"prefix":"","firstName":"Shijie","middleName":"","lastName":"Chen","suffix":""},{"id":575512996,"identity":"2350e033-272a-447d-97d4-d1206970bb1a","order_by":4,"name":"Zhigang Li","email":"","orcid":"","institution":"Wuhan University","correspondingAuthor":false,"prefix":"","firstName":"Zhigang","middleName":"","lastName":"Li","suffix":""},{"id":575512998,"identity":"e8f1e2bc-e75d-4f5e-b294-ab1158135622","order_by":5,"name":"Xiaofan Luan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA90lEQVRIiWNgGAWjYDACZiCuAGI+EOcDwwEQZUBYyxkgZgNixhlEaWFA0sLMQ4wWg+PMDx8cqLnDwCaR/OyxbdudxAb25m0SDEARXECymc3Y4MCxZ0AtaebGuW3PEht4jpVJMABFcAF+ZgYz6Q9sh4FaEsykc7cdTmyQyDGTYGw4jFMLGzP7N4kD/0Ba0r9JW4K0yL/Br4WfmcdM4mAbSEuOmTQj2BYe/Fokm3mKDQ72HeZh43lTJtn777BxG09asUXCMdxaDM4f3/jgwLfDcvzs6dskfpw5LNvPfnjjjQ81uLXAAA/CdyAigaCGUTAKRsEoGAX4AAD6SlD9smGGHgAAAABJRU5ErkJggg==","orcid":"","institution":"Wuhan University","correspondingAuthor":true,"prefix":"","firstName":"Xiaofan","middleName":"","lastName":"Luan","suffix":""}],"badges":[],"createdAt":"2025-10-13 07:08:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7845488/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7845488/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":100596030,"identity":"13250bb5-3294-4cbb-8cf8-c9fe1add3483","added_by":"auto","created_at":"2026-01-19 13:50:23","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":62730,"visible":true,"origin":"","legend":"","description":"","filename":"anonymizedmanuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/b895e9c395ca609419bf855b.docx"},{"id":100574380,"identity":"3fe438c3-e889-4d09-89a3-5037148e2688","added_by":"auto","created_at":"2026-01-19 10:10:06","extension":"png","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":80869,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.AnalyticalFramework.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/940116a47126e4b869e7ef12.png"},{"id":100595681,"identity":"3bc56d36-bc76-4fbb-a20c-11510ae0ce30","added_by":"auto","created_at":"2026-01-19 13:49:07","extension":"png","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6552991,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.MigrationFlowODDiagramfromSmallCitiestoLargeCities.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/94e808c5ab84e459e83d8c64.png"},{"id":100595688,"identity":"16a35fb1-6a85-4284-b6d9-9405fe2c6d1b","added_by":"auto","created_at":"2026-01-19 13:49:08","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":78837,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.Multicollinearityanalysisandvariableoptimization.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/c393f81f78c481ae3731d848.png"},{"id":100595538,"identity":"12e1c6ec-df12-48a6-b894-6795f623d88f","added_by":"auto","created_at":"2026-01-19 13:48:44","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":285409,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.OLSbasedVariableSignificanceAnalysisandXGBoostbasedSHAPInterpretation.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/2ac28fc92c420f97f8de59e9.png"},{"id":100595685,"identity":"e9a408a7-94c7-4eb3-9fe4-c613f86abd0f","added_by":"auto","created_at":"2026-01-19 13:49:08","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1978141,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.Dependenceplotsfortheimpactofworkingonmigrantsettlementintentions.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/00695f0151ae1661e47ba0d6.png"},{"id":100595603,"identity":"69d64fbf-0cf4-46aa-abf8-d56078c56da9","added_by":"auto","created_at":"2026-01-19 13:48:53","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1797777,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.Dependenceplotsfortheimpactofhousingonmigrantssettlementintentions.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/954186571284bbabaf662742.png"},{"id":100595693,"identity":"69396a0d-db1a-437b-b3aa-172d876c1129","added_by":"auto","created_at":"2026-01-19 13:49:10","extension":"json","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7588,"visible":true,"origin":"","legend":"","description":"","filename":"3dc2981524784191a1d3a93bedbc5754.json","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/958a5a11fd1c532135933e7e.json"},{"id":100595690,"identity":"f297fbce-e188-4b3f-ae5c-2f37410a69b0","added_by":"auto","created_at":"2026-01-19 13:49:09","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":133518,"visible":true,"origin":"","legend":"","description":"","filename":"3dc2981524784191a1d3a93bedbc57541enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/a7d086aaa3bfe2cc1f47f9b5.xml"},{"id":100574391,"identity":"651bf93b-819e-4ebe-8ae1-637838aea853","added_by":"auto","created_at":"2026-01-19 10:10:06","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":80869,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.AnalyticalFramework.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/f72beaabd46ab18a82c13a6d.png"},{"id":100595677,"identity":"e4a0a051-42e6-447d-a6d6-6d3ed6836134","added_by":"auto","created_at":"2026-01-19 13:49:06","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6552991,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.MigrationFlowODDiagramfromSmallCitiestoLargeCities.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/53cab3f4028efbfa4cca90ce.png"},{"id":100595563,"identity":"510c47b1-e6e8-4c22-a341-947a8e06bcf4","added_by":"auto","created_at":"2026-01-19 13:48:47","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":78837,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.Multicollinearityanalysisandvariableoptimization.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/d6ce6077d900e590b5ee38ba.png"},{"id":100596089,"identity":"9a9d492d-ab37-4a7d-859b-2dbb06e22669","added_by":"auto","created_at":"2026-01-19 13:51:10","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":285409,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.OLSbasedVariableSignificanceAnalysisandXGBoostbasedSHAPInterpretation.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/f5885ae693d67d235c2a5fdd.png"},{"id":100595555,"identity":"f284015b-61c5-4121-a7fe-8d11b5562ceb","added_by":"auto","created_at":"2026-01-19 13:48:46","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1978141,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.Dependenceplotsfortheimpactofworkingonmigrantsettlementintentions.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/f7d6fbe8f30f54bf27e10e00.png"},{"id":100595896,"identity":"b615107e-6519-4b08-bf79-c9124d400666","added_by":"auto","created_at":"2026-01-19 13:49:37","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1797777,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.Dependenceplotsfortheimpactofhousingonmigrantssettlementintentions.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/afd8850191c74f0a898c9a64.png"},{"id":100595655,"identity":"9f4ca918-48e8-4906-95bf-4fbe6cf6acca","added_by":"auto","created_at":"2026-01-19 13:49:02","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":59619,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure1.AnalyticalFramework.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/8cb9aa16e264083838d99b61.png"},{"id":100596034,"identity":"f5aa1d95-2be1-4d3f-9fa7-60663e91012b","added_by":"auto","created_at":"2026-01-19 13:50:25","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":971646,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure2.MigrationFlowODDiagramfromSmallCitiestoLargeCities.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/43b501ff976fc36e237a8057.png"},{"id":100595310,"identity":"e6085434-47a1-48e8-a81c-b470640d9514","added_by":"auto","created_at":"2026-01-19 13:48:11","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":71164,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure3.Multicollinearityanalysisandvariableoptimization.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/028006d866b5aff7d25d7919.png"},{"id":100595463,"identity":"b9e71fe1-1a32-47ee-8672-9540fd5a6d00","added_by":"auto","created_at":"2026-01-19 13:48:32","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":83336,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure4.OLSbasedVariableSignificanceAnalysisandXGBoostbasedSHAPInterpretation.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/6309e8ef1823048c835fa386.png"},{"id":100595135,"identity":"fbbfe333-22b3-4296-b27e-c6ad03cb6a64","added_by":"auto","created_at":"2026-01-19 13:47:35","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":309525,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure5.Dependenceplotsfortheimpactofworkingonmigrantsettlementintentions.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/3a258ec94dd9e88f4c27ccaa.png"},{"id":100595682,"identity":"51e5ce42-3b1f-483f-bece-b32813df80c9","added_by":"auto","created_at":"2026-01-19 13:49:07","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":276105,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure6.Dependenceplotsfortheimpactofhousingonmigrantssettlementintentions.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/b92e5e9cb2c5f66728ad3ded.png"},{"id":100595948,"identity":"610509be-26de-4654-8003-ad2b4737d4a3","added_by":"auto","created_at":"2026-01-19 13:49:49","extension":"xml","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":131215,"visible":true,"origin":"","legend":"","description":"","filename":"3dc2981524784191a1d3a93bedbc57541structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/0e4d85d7068879331995994a.xml"},{"id":100596051,"identity":"50c38f89-5943-4275-a107-ba84077c2a49","added_by":"auto","created_at":"2026-01-19 13:50:28","extension":"html","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":140559,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/d6186ec7d9246002b6efdc61.html"},{"id":100595623,"identity":"0087c738-63c8-48b8-b7bc-765793d10772","added_by":"auto","created_at":"2026-01-19 13:48:57","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":692796,"visible":true,"origin":"","legend":"\u003cp\u003eAnalytical Framework\u003c/p\u003e","description":"","filename":"Figure1.AnalyticalFramework.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/c4d29d2a9707eaf0cb4f89f0.png"},{"id":100595924,"identity":"afd4b129-302e-4e23-bdec-c0c2491961d8","added_by":"auto","created_at":"2026-01-19 13:49:42","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1625531,"visible":true,"origin":"","legend":"\u003cp\u003eMigration Flow OD Diagram from Small Cities to Large Cities\u003c/p\u003e","description":"","filename":"Figure2.MigrationFlowODDiagramfromSmallCitiestoLargeCities.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/13466bdf14f41a426fbb4290.png"},{"id":100595143,"identity":"8e30d01b-74f2-457a-ae24-250e131d352e","added_by":"auto","created_at":"2026-01-19 13:47:38","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":367444,"visible":true,"origin":"","legend":"\u003cp\u003eMulticollinearity analysis and variable optimization\u003c/p\u003e","description":"","filename":"Figure3.Multicollinearityanalysisandvariableoptimization.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/c9eb177126553616b071c3bc.png"},{"id":100574384,"identity":"573ed32c-d9f7-4f35-bcb3-6074b336eaf2","added_by":"auto","created_at":"2026-01-19 10:10:06","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":459736,"visible":true,"origin":"","legend":"\u003cp\u003eOLS-based Variable Significance Analysis and XGBoost-based SHAP Interpretation\u003c/p\u003e","description":"","filename":"Figure4.OLSbasedVariableSignificanceAnalysisandXGBoostbasedSHAPInterpretation.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/4ba251a5f466d75b5e72c93d.png"},{"id":100574388,"identity":"4ef5a1cf-e0a3-4a21-b3e7-537da7834386","added_by":"auto","created_at":"2026-01-19 10:10:06","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1159097,"visible":true,"origin":"","legend":"\u003cp\u003eDependence plots for the impact of working on migrant settlement intentions. \u003cem\u003ePanels A1-A3 and B1-B3 respectively illustrate the SHAP analysis for the effects of average wage gap on high- and low-education groups across three life course stages (unmarried, married-childless, married-with-children). Panels a1-a3 and b1-b3 correspondingly depict the mechanism for monthly personal income under the same groupings\u003c/em\u003e.\u003c/p\u003e","description":"","filename":"Figure5.Dependenceplotsfortheimpactofworkingonmigrantsettlementintentions.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/4b6592ac5b29c061318183b6.png"},{"id":100596048,"identity":"e341a6d0-bdf7-455a-944c-74263612a13a","added_by":"auto","created_at":"2026-01-19 13:50:27","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1082667,"visible":true,"origin":"","legend":"\u003cp\u003eDependence plots for the impact of housing on migrants' settlement intentions. \u003cem\u003ePanels A1-A3 and B1-B3 respectively illustrate the SHAP analysis for the effects of housing affordability on high- and low-education groups across three life course stages (unmarried, married-childless, married-with-children). Panels a1-a3 and b1-b3 correspondingly depict the mechanism for housing type under the same groupings\u003c/em\u003e.\u003c/p\u003e","description":"","filename":"Figure6.Dependenceplotsfortheimpactofhousingonmigrantssettlementintentions.png","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/39b56d3377b59c112e3c8585.png"},{"id":100597478,"identity":"ca7306cf-f50b-4f0d-9ce7-34f1c6a8d2f2","added_by":"auto","created_at":"2026-01-19 14:18:17","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6154078,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7845488/v1/0d815227-fab9-4c31-8e04-332a3db93813.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"For jobs or housing in the Metropolis: A Life Course Analysis of Small-town Migrants’ Settlement Intentions Through Machine Learning","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eThe concentration of population from small towns to large cities is a global trend, where better employment opportunities and income are available (Liu \u0026amp; Wang, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). In China, this trend has been exacerbated by large-scale infrastructure development, particularly the high-speed rail network (Wang \u0026amp; Duan, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Wang, Acheampong, \u0026amp; He, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). However, due to high housing prices in large cities, migrant populations face the dilemma of being unable to balance employment opportunities with living conditions. This issue is further complicated in China by institutional factors such as the hukou system and its associated public service provision (Wang \u0026amp; Shen, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). As a result, a large number of migrant populations remain in a long-term state of \u0026ldquo;temporary residence\u0026rdquo;\u0026mdash;having the need to work and live in large cities while struggling to obtain full urban welfare benefits and stable housing, namely long-term settlement capacity. Meanwhile, as migrant populations are one of the important driving forces behind the rapid economic growth of large cities globally, particularly in China, the issue of migrants\u0026rsquo; settlement intentions has attracted increasing research attention (Gan et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Liu \u0026amp; Wang, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe effects of employment and housing on settlement intentions vary significantly across different life stages and populations. Different life events, especially marriage and childbearing, fundamentally influence individuals\u0026rsquo; spatial choices by reconfiguring family responsibility networks and resource allocation patterns (Elder, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e1994\u003c/span\u003e; Elder et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Wagner \u0026amp; Mulder, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Due to limited educational resources for non-local hukou holders, the impact of marriage and childbearing on migrants\u0026rsquo; settlement intentions is more complex in China (Wang \u0026amp; Shen, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Gan et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Furthermore, through employment and income differentiation, groups with different educational backgrounds exhibit disparities in long-term settlement capacity (Fan, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Liu \u0026amp; Wang, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). This gap has been further exacerbated by the recent emergence of \"talent wars\" among major cities, which implement differentiated preferential policies using educational level as a screening criterion.\u003c/p\u003e \u003cp\u003eAlthough an increasing number of studies recognize the importance of educational level and life course on settlement intentions (Wang \u0026amp; Shen, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Liu \u0026amp; Wang, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), gaps remain in research methods and study populations. On one hand, since migrants\u0026rsquo; settlement intentions in large cities represent a complex non-linear problem, it is necessary to adopt machine learning algorithms to precisely identify the non-linear effects and threshold effects of variables, rather than relying on traditional linear regression models used in previous studies (De Jong \u0026amp; Graefe, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Wagner \u0026amp; Mulder, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). On the other hand, since the gradual relaxation of the hukou system since the millennium, rural and non-rural hukou migrant populations, as the main focus of past research, face increasingly diminished differences in their challenges (Wang \u0026amp; Shen, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Migrant populations from resource-poor small cities, including farmers, should replace the singular farmer perspective. They migrate to large cities due to limited development opportunities in their hometowns, yet remain in a long-term state of temporary residence, not fully integrated into large cities (Wang, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Gu, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Therefore, this study will analyze how employment and housing affect the settlement intentions of small-town migrant populations with different educational backgrounds in large cities, based on a life course perspective and utilizing machine learning algorithms.\u003c/p\u003e \u003cp\u003eThe paper is structured as follows: Section \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e reviews the literature on population mobility, life course, and settlement intentions, followed by an analytical framework for understanding the settlement intentions of small-town migrants in Section \u003cspan refid=\"Sec5\" class=\"InternalRef\"\u003e3\u003c/span\u003e, incorporating multiple influencing factors, and introduces the variables and methodology used in the study. Section \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents variable importance analysis and partial dependence analysis based on SHAP interpretations. Finally, the paper discusses findings and concludes in Section \u003cspan refid=\"Sec14\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e"},{"header":"2 Literature review","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Working and Housing as Drivers of Population Mobility\u003c/h2\u003e \u003cp\u003eWorking and housing constitute the core factors driving urban migration and settlement, but their mechanisms differ across educational groups (Fan, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Khoo et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Wang \u0026amp; Shen, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In China, the interplay of institutional factors such as household registration (hukou), housing prices, and public service allocation makes the impact of working and housing on settlement intention particularly complex (Wu, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Chan, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWorking factors encompass three core dimensions: job opportunities, working stability, and income level. Among these, income level has received extensive attention due to its quantifiable characteristics (Wang \u0026amp; Shen, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Fan, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Khoo et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). Income directly determines migrants\u0026rsquo; quality of life and capacity for settlement, making it a significant factor influencing settlement intention (Zhu, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Huang et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Migrants with different educational levels show substantial differences in income acquisition and income expectations. Highly educated migrants typically earn higher incomes and focus more on urban wage levels and development potential (Wang, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In contrast, less educated migrants face greater restrictions in the labor market and are more likely to be excluded from quality employment opportunities, making them more concerned with personal income levels (Wang, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Wei et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe accessibility, comfort, and stability of housing largely determine migrants\u0026rsquo; living standards, thereby influencing their settlement intention (Liu et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Yang \u0026amp; Guo, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). In China, homeownership is tied to the household registration system, providing not only housing but also access to local public services and welfare. Despite relaxed hukou entry requirements and economic support since the 2014 New-Type Urbanization Plan, persistently high housing prices keep homeownership unaffordable for most migrants. (Liu et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Zheng et al., 2021; Luan et al.,2024). Consequently, many migrants enter diversified private rental markets, where housing accessibility is relatively high, but stability and comfort are often poor (Du et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Zheng et al., 2021; Wu et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Different rental types exhibit significant variations in stability, pricing, and associated public services (Lin et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Furthermore, differences in migrants\u0026rsquo; educational backgrounds shape distinct housing choice patterns and settlement decisions across groups by influencing their income levels and access to policy support (Wang, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHowever, migrants in major cities often struggle to balance work and residence (Wang \u0026amp; Shen, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Many migrants adopt strategies that sacrifice residential quality to ensure working stability, leading them into long-term temporary residence status, while only a few migrants with dual resource advantages achieve permanent settlement; the remainder who prioritize residential security choose to return to their hometowns (Zhu \u0026amp; Chen, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Wang \u0026amp; Shen, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Yue et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). Therefore, promoting the transition from temporary residence to permanent settlement among migrants represents a pressing practical issue. However, settlement intention as a complex multi-dimensional decision-making process requires more comprehensive multidimensional analytical frameworks for in-depth interpretation\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Life Course and Migration Settlement Intention\u003c/h2\u003e \u003cp\u003eLife course theory provides a framework for understanding migration dynamics through the interaction between individual life events and their historical spatiotemporal contexts. (Elder, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e1994\u003c/span\u003e; Elder et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). Marriage and childbearing events, as structurally significant turning points in the life course, fundamentally influence individuals\u0026rsquo; spatial choices and settlement intentions by reconfiguring family responsibility networks and resource allocation patterns (Kulu \u0026amp; Milewski, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Mulder \u0026amp; Wagner, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Courgeau, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e1990\u003c/span\u003e). In Europe and North America, married groups with children demonstrate significantly lower migration frequency than single groups and are more inclined to settle in areas with better educational resources and working opportunities (De Jong \u0026amp; Graefe, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Findlay et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Geist \u0026amp; McManus, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2008\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eUnder China\u0026rsquo;s unique household registration system, the impact of marriage and childbearing on settlement intention exhibits distinctive complexity (Fan, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Zhu \u0026amp; Lin, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). The strong linkage between marriage and housing makes housing an important prerequisite for marriage establishment, a cultural phenomenon that further reinforces married populations\u0026rsquo; demand for housing stability (Wu \u0026amp; Gaubatz, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Yang \u0026amp; Guo, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Meanwhile, the high coupling mechanism between the household registration system and educational resource allocation results in multiple institutional barriers for migrant children without local household registration when enrolling in urban public schools, including additional borrowing fees and restrictions on secondary and higher education entrance exam policies (Wang \u0026amp; Shen, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wu, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). These educational access thresholds force many migrant families to face difficult choices when their children enter compulsory education: either bear the high costs of private education or send their children back to their hometowns for education, thus creating a contradiction between \"high settlement needs and high institutional barriers\" (Huang et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Li \u0026amp; Liu, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Hu et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2011\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAlthough existing research has revealed the complex associations between life course and settlement intention (Cao et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Fan, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), current analytical methods mostly rely on traditional linear regression and other econometric models, making it difficult to precisely capture the nonlinear effects and threshold effects involved. Therefore, this study employs machine learning methods to better identify and model the nonlinear influence mechanisms of life course events on settlement intention.\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Methodology","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Analytical framework\u003c/h2\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e about here]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, this study constructs a threshold effect analytical framework grounded in push-pull theory. The threshold effect describes a nonlinear process where response variables undergo significant changes once stimuli exceed critical values (Zou et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Based on this theoretical foundation, the framework systematically analyzes the long-term settlement decision-making mechanisms of small-town youth with different educational backgrounds in large cities through two core dimensions: working and residence.\u003c/p\u003e \u003cp\u003eWorking and residence factors exhibit threshold effect characteristics in metropolitan environments. Working elements, including income levels and career development opportunities, along with residence elements represented by housing prices and living conditions, constitute pull factors attracting migrants (Wang \u0026amp; Shen, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). When urban living costs exceed an individual\u0026rsquo;s critical tolerance threshold, pull factors transform into push factors, prompting individuals to reconsider their residential choices. This transformation process essentially reflects migrants\u0026rsquo; trade-offs between working and residence: to obtain superior working conditions, migrants typically must bear higher housing costs and residential pressures, seeking equilibrium between career development potential and residential economic burden (Liu \u0026amp; Xiao, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eEducational background differentiates threshold sensitivity among migrant populations. Highly educated groups possess stronger economic capacity and policy support, enabling them to tolerate higher urban costs (Wang, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Less educated groups face working limitations and economic constraints, encountering earlier threshold critical points (Khoo et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). This difference stems from varying abilities to access urban resources rather than different preference patterns (Wang \u0026amp; Shen, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eLife-stage characteristics further moderate threshold effects and priority rankings. Single individuals prioritize career advancement over maintaining living stability, willing to compromise residential comfort for professional development opportunities (Coulter \u0026amp; Van Ham, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). Families with children face multidimensional resource allocation challenges: they must simultaneously balance personal career development, family economic responsibilities, and children\u0026rsquo;s educational investments. The superposition of multiple obligations significantly compresses their urban adaptation buffer space (Gong et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). This differentiation pattern reflects the dynamic balancing strategies of individuals at different life stages between pursuing development, maintaining stability, and cultivating belonging (Vidal \u0026amp; Lutz, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBased on this conceptual framework, we propose the following hypotheses:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eHypothesis 1\u003c/strong\u003e \u003cp\u003eIn large cities, working and residence factors exhibit threshold effects on migration intentions, where pull factors transform into push factors once critical boundaries are exceeded.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eHypothesis 2\u003c/strong\u003e \u003cp\u003eHighly educated and less educated individuals have different threshold boundaries, with highly educated groups tolerating higher urban costs due to higher incomes.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eHypothesis 3\u003c/strong\u003e \u003cp\u003eWithin the same educational level, life-course stages moderate threshold effects, with unmarried individuals showing highest tolerance for urban challenges, followed by married individuals without children, and married individuals with children showing lowest tolerance.\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Research area and data description\u003c/h2\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e about here]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis study uses data from the 2017 China Migrants Dynamic Survey (CMDS) by China\u0026rsquo;s National Health Commission. The survey employed stratified, multi-stage probability proportional to size (PPS) sampling across all 31 provinces, targeting floating migrants aged 15\u0026thinsp;+\u0026thinsp;with non-local hukou who had resided locally for over one month, yielding 169,989 samples.\u003c/p\u003e \u003cp\u003eTo examine small-to-large city migration patterns, systematic screening was applied: current residence limited to top 10 GDP cities in 2017 (Beijing, Shanghai, Shenzhen, Guangzhou, Chongqing, Suzhou, Chengdu, Wuhan, Hangzhou, Tianjin) and provincial capitals; hukou registration restricted to county-level areas and below; \u0026ldquo;employer\" samples excluded for homogeneity. This yielded 64,623 valid observations (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSettlement intention is the dependent variable, derived from expected local residence duration and recoded as binary: 6\u0026thinsp;+\u0026thinsp;years coded as 1 (long-term intention), others as 0 (short-term intention). Among samples, 42.60% expressed long-term settlement intentions.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows variable definitions and descriptive statistics. Control variables include individual characteristics (education, life course, gender, age), economic variables (income, industry, income gap), and housing economic variables (price gap, affordability). Housing type serves as the core explanatory variable with nine categories from workplace accommodation to self-purchased housing.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDefinitions and descriptive statistics of variables for small-town migrants\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDefinition and Coding\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNumbers\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePercentage\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePopulation Classification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eEducation level\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;0 if high school or below\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e48123\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e74.47%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;1 if college degree or above\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e25.53%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLife course\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;0 if unmarried\u003c/p\u003e \u003cp\u003e=\u0026thinsp;1 if married without children\u003c/p\u003e \u003cp\u003e=\u0026thinsp;2 if married with children\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 1.62 (std. 0.75)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDependent Variable\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSettlement intention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;0 if short-term stay (\u0026le;\u0026thinsp;5years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37,081\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e57.40%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;1 if long-term settlement (\u0026ge;\u0026thinsp;6 years \u0026amp; permanent settlement)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e27,540\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e42.60%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIndependent Variable\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHousing Variables\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHousing price gap\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDestination city average housing price minus origin city average housing price\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 14,345.06 (std. 16,260.69)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHousing affordability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDestination city average housing price \u0026times; 90㎡ \u0026divide; (annual household income)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 20.68 (std. 32.05)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHousing affordability gap\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHousing affordability in destination city minus housing affordability in origin city\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 13.94 (std. 27.34)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHousing type\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;1 if workplace accommodation\u003c/p\u003e \u003cp\u003e=\u0026thinsp;2 if employer/unit housing\u003c/p\u003e \u003cp\u003e=\u0026thinsp;3 if borrowed housing\u003c/p\u003e \u003cp\u003e=\u0026thinsp;4 if other informal housing\u003c/p\u003e \u003cp\u003e=\u0026thinsp;5 if government public rental housing\u003c/p\u003e \u003cp\u003e=\u0026thinsp;6 if private rental housing\u003c/p\u003e \u003cp\u003e=\u0026thinsp;7 if self-purchased commercial housing\u003c/p\u003e \u003cp\u003e=\u0026thinsp;8 if self-purchased affordable housing\u003c/p\u003e \u003cp\u003e=\u0026thinsp;9 if self-purchased limited property rights housing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 5.80 (std. 1.72)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWorking Variables\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMonthly personal income\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMonthly personal income (yuan)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 4,336.82\u003c/p\u003e \u003cp\u003e(std. 3,157.52)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage wage level\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAverage monthly personal income in destination city\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 4,336.82\u003c/p\u003e \u003cp\u003e(std. 814.47)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage wage gap\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDestination city average income minus national average income\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 0.12 (std. 734.93)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eControl Variables\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;0 if female\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32,655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e50.53%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;1 if male\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e31,968\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e49.47%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of children\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNumber of children in household\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 1.28 (std. 0.92)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFamily number\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNumber of family members\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 3.10 (std. 1.19)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBirth year\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYear of birth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 1981.56 (std. 10.52)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIndustry classification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e=\u0026thinsp;1 if employed in knowledge-intensive services\u003c/p\u003e \u003cp\u003e=\u0026thinsp;2 if employed in labor-intensive services\u003c/p\u003e \u003cp\u003e=\u0026thinsp;3 if employed in manufacturing\u003c/p\u003e \u003cp\u003e=\u0026thinsp;4 if employed in raw materials and resources\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAvg. 2.23 (std. 0.74)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e[Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e about here]\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.3 XGBoost-SHAP Modeling Method\u003c/h2\u003e \u003cp\u003eGiven the non-linear relationships among factors in settlement intention decisions and data imbalance, this study employs XGBoost (eXtreme Gradient Boosting) to construct a settlement intention prediction model. XGBoost, as a gradient boosting decision tree ensemble algorithm, automatically identifies non-linear relationships and complex interactions among variables.\u003c/p\u003e \u003cp\u003eThe study uses Random Forest for feature selection and importance ranking, then applies XGBoost with fixed hyperparameters: 100 trees, maximum depth of 5, learning rate of 0.05, subsample ratio of 0.8, feature subsampling of 0.8, minimum child weight of 1, and binary:logistic objective function. Feature matrix X contains core explanatory variables, while target variable y represents binary long-term settlement intention. Data is split 8:2 for training and testing.\u003c/p\u003e \u003cp\u003eFor model interpretability, Shapley Additive exPlanations (SHAP) is employed. Based on cooperative game theory, SHAP provides fair feature contribution allocation through the Shapley value formula:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\phi\\:ⱼ\\:=\\:\\sum\\:[S\\subseteq\\:F\\backslash\\:\\{j\\left\\}\\right]\\:\\left|S\\right|!(n-|S|-1)!/n!\\:\\times\\:\\:\\left[f\\right(S\\cup\\:\\left\\{j\\right\\})\\:-\\:f(S\\left)\\right]$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere S represents feature subsets excluding feature j, F is the complete feature set, n is total features, and f(S) is model output using subset S.\u003c/p\u003e \u003cp\u003eFeature importance is calculated via global mean absolute SHAP values:\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:Iⱼ\\:=\\:(1/N)\\:\\sum\\:[i=1\\:to\\:N]\\:\\left|\\phi\\:ⱼⁱ\\right|$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere N is total samples and φⱼⁱ is the SHAP value of feature j for sample i.\u003c/p\u003e \u003cp\u003eThe model achieves approximately 74% training accuracy and 71\u0026ndash;73% testing accuracy, demonstrating stable performance. SHAP values are computed using TreeExplainer, extracting positive class contributions to analyze feature influence on long-term settlement intention.\u003c/p\u003e \u003c/div\u003e"},{"header":"4 Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e4.1 OLS Analysis and Variable SHAP Interpretation\u003c/h2\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e about here]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo ensure the robustness of the model estimation, we first diagnosed and addressed multicollinearity among all potential predictors (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). An analysis of the Variance Inflation Factor (VIF) for the initial set of variables revealed that housing affordability gap (VIF\u0026thinsp;=\u0026thinsp;49.2), housing price gap (VIF\u0026thinsp;=\u0026thinsp;51.1), average wage level (VIF\u0026thinsp;=\u0026thinsp;178.1), and birth year (VIF\u0026thinsp;=\u0026thinsp;223.2) exhibited severe multicollinearity, with VIF values substantially exceeding the conventional threshold of 5.0. Through the stepwise removal of collinear variables while retaining the most theoretically salient predictors, all VIFs in the final optimized set were reduced to acceptable levels (e.g., housing type, VIF\u0026thinsp;=\u0026thinsp;5.0; monthly personal income, VIF\u0026thinsp;=\u0026thinsp;3.1; average wage gap, VIF\u0026thinsp;=\u0026thinsp;1.3). This process effectively mitigates the potential for multicollinearity to bias the subsequent model results.\u003c/p\u003e \u003cp\u003eBased on this optimized set of variables, we first employed a logistic regression model to conduct a preliminary examination of the factors influencing settlement intention (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, left panel). The results indicated that education level, life course, personal housing affordability, housing type, monthly personal income, average wage gap, gender, and children number were all statistically significant predictors of settlement intention (p\u0026thinsp;\u0026lt;\u0026thinsp;.05). Specifically, the adjusted odds ratio (OR) for education level was 2.451 (95% CI: 2.3479\u0026ndash;2.5586), indicating that individuals with higher education had significantly greater odds of intending to settle. Similarly, life course (OR\u0026thinsp;=\u0026thinsp;1.6521), housing type (OR\u0026thinsp;=\u0026thinsp;1.2992), monthly personal income (OR\u0026thinsp;=\u0026thinsp;1.0551), and average wage gap (OR\u0026thinsp;=\u0026thinsp;1.1089) were all positively associated with settlement intention, whereas personal housing affordability exerted a negative influence (OR\u0026thinsp;=\u0026thinsp;0.9957). However, the logistic regression model inherently assumes a linear relationship between the predictors and the log-odds of the outcome, making it ill-suited for capturing more complex non-linear patterns and threshold effects.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e about here]\u003c/p\u003e \u003cp\u003eTo overcome this limitation, we further employed an XGBoost model and utilized SHAP (SHapley Additive exPlanations) values for a more in-depth interpretation (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, right panel). The SHAP summary plot provides a comprehensive visualization of the global importance of each feature, as well as the direction and magnitude of its impact on the model's predictions. Housing type exhibited a distinct bifurcation in its SHAP values: low feature values (representing non-ownership) corresponded almost exclusively to negative contributions, while high values (representing ownership) yielded significant positive contributions, highlighting the decisive role of obtaining property rights. Education level showed a similarly stratified effect, with high feature values clustering in the positive SHAP range and low values concentrating in the negative range. Particularly noteworthy is the complex, non-linear relationship demonstrated by housing affordability: while low to moderate levels of housing pressure (blue dots) exerted a negative influence, extremely high levels (red dots) paradoxically showed a positive rebound effect on settlement intention. Furthermore, the wide, bidirectional distribution of SHAP values for life course, monthly personal income, and average wage gap strongly indicates that their effects are non-monotonic and characterized by significant thresholds. In contrast, gender and children number had more concentrated SHAP value distributions, suggesting they possessed comparatively weaker predictive power overall.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e4.2 PDP Analysis: Nonlinear Relationships and Threshold Effects\u003c/h2\u003e \u003cp\u003ePartial dependence plots (Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e and \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e) were employed to identify nonlinear influences and threshold effects on settlement intentions across education levels and life course stages.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003e4.2.1 Working and Settlement Intention\u003c/h2\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e about here]\u003c/p\u003e \u003cp\u003eFor the high-education group, the influence of average wage gap on settlement intention reveals significant threshold effects and life-course heterogeneity (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, Panels A1-A3). Compared to their peers with children, unmarried and married-childless individuals exhibit greater tolerance for cities where wages are below the national average. For the unmarried group, SHAP values remain positive even when the wage gap is below -\u0026yen;991, suggesting that cities with significantly lower wages can still hold some attraction. Beyond this point, the effect fluctuates narrowly around zero, indicating saturation. The married-childless group shows a significant positive effect when the gap is below -\u0026yen;1,321 or above \u0026yen;208, but a slight negative effect within the intermediate range of -\u0026yen;1,321 to \u0026yen;208, implying a sensitivity to moderate wage levels. In stark contrast, the married-with-children group has the lowest tolerance for low-wage cities, showing a significant negative effect in the -\u0026yen;2,996 to -\u0026yen;1,686 range. Once the gap exceeds -\u0026yen;1,686, the effect flattens, reflecting a diminished sensitivity to wage gaps after a basic income threshold is met.\u003c/p\u003e \u003cp\u003eThe threshold effects of average wage gap for the low-education group present a starkly different pattern (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, Panels B1-B3). Contrary to their high-education counterparts, a negative wage gap has a negative impact on unmarried low-education individuals. Once the gap exceeds \u0026yen;920, the positive effect grows continuously, indicating this group prioritizes the absolute advantage of actual wages. For the married-childless low-education group, the effect is positive when the wage gap is below \u0026yen;384 but turns negative beyond this threshold\u0026mdash;a pattern nearly opposite to that of their high-education peers. While the married-with-children low-education group shares a similar pattern with its high-education counterpart (i.e., low tolerance for low-wage cities), it is more sensitive to the threshold: any wage gap below zero exerts a negative influence, and the effect only turns positive after exceeding \u0026yen;863. This may reflect more severe household economic pressures, leading to stricter income requirements.\u003c/p\u003e \u003cp\u003eThe impact of monthly personal income on the high-education group also varies significantly across life course stages (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, Panels a1-a3). The unmarried group has an \"optimal income range\" between \u0026yen;4,949 and \u0026yen;27,719, within which the effect is positive. Incomes below or, notably, above this range produce a significant negative effect, suggesting a clear \"satisfaction\" threshold rather than a \"higher-is-better\" logic. The pattern for the married-childless group is inverted, showing a significant positive effect only after income surpasses \u0026yen;20,468, indicating that higher family living costs elevate their income threshold. The married-with-children group exhibits an inverted U-shaped pattern: the effect is positive within the \u0026yen;4,280 to \u0026yen;20,066 range, but beyond this upper limit, further income increases lead to a sharp decline into negative territory. This likely reflects the trade-offs between high-paying jobs, their associated pressures, and multiple household expenditures.\u003c/p\u003e \u003cp\u003eIn comparison, the response of the low-education group to monthly personal income is more linear and monotonic (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, Panels b1-b3). Unlike the complex patterns of the high-education group, the settlement intention of unmarried and married-with-children low-education individuals is largely positively correlated with their income. Once their monthly income crosses the thresholds of \u0026yen;4,147 and \u0026yen;5,217, respectively, the SHAP values turn positive and continue to increase with rising income. Only the married-childless subgroup is an exception, showing a positive effect within a high-income bracket of \u0026yen;19,899 to \u0026yen;34,448. Overall, the low-education group responds more directly to income gains, with lower activation thresholds for positive effects and little evidence of a negative effect at very high income levels.\u003c/p\u003e \u003cp\u003eSynthesizing the analysis of these two employment factors reveals that the high-education group tolerates low-wage cities before having children but avoids them afterward. In contrast, the low-education group generally pursues wages that are absolutely higher than the national average. Regarding personal income, very high earnings can act as a push factor for unmarried and parent high-education individuals, whereas for their low-education counterparts, it consistently serves as a powerful pull factor. This divergence profoundly reflects the heterogeneous strategies employed by groups with different educational capital when balancing employment opportunities against the cost of living.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section3\"\u003e \u003ch2\u003e4.2.2 Housing and Settlement Intention\u003c/h2\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e about here]\u003c/p\u003e \u003cp\u003eThe effect of housing affordability on settlement intention is predominantly negative, functioning as an inhibitor. However, its mechanism exhibits complex non-linear characteristics across different groups, most notably a counter-intuitive positive rebound under extreme pressure (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, Panels A1-B3). Among the high-education group, the settlement intention of unmarried individuals turns consistently negative once the housing affordability ratio exceeds 13.49 years. A similar trend is observed for the married-childless group, but after the ratio surpasses 61.37 years, the SHAP values unexpectedly rebound, exerting a significant positive influence. This rebound pattern also appears for the married-with-children group, with the threshold occurring around 319 years. The low-education group's response likewise generally follows this \"low-pressure-inhibition, high-pressure-rebound\" pattern, albeit with vastly different thresholds. Unmarried individuals in this group paradoxically show positive settlement intention within an extremely high-pressure range of 247.53 to 301.39 years, while the married-childless group's intention turns positive after the ratio exceeds 96.72 years. The most complex pattern belongs to the married-with-children group, who exhibit a positive effect across a broad range below 2221.99 years, with the effect only turning negative beyond this critical point.\u003c/p\u003e \u003cp\u003eIn contrast, the influence of housing type reveals a clear \"property rights divide,\" with a significant step-change in effect between different tenure statuses (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, Panels a1-b3). Regardless of education level or life course stage, housing categories 7\u0026ndash;9 (which encompass various forms of homeownership such as self-purchased commercial housing, affordable housing, and limited-property-rights housing) provide a strong and significant positive impetus to settlement intention. All other non-ownership categories, conversely, act primarily as inhibitors to varying degrees. A noteworthy exception is that among high-education unmarried and married-childless individuals, there is a certain tolerance for housing types 3\u0026ndash;5 (e.g., borrowed housing, other informal housing, and government public rentals). Their SHAP values are slightly above zero, suggesting these transitional housing options still hold a modest attractive power for them. This characteristic is also observed among low-education individuals who are married with children.\u003c/p\u003e \u003cp\u003eOverall, we find that housing affordability generally functions as a \"push factor\" that inhibits settlement intention. However, low-education groups, particularly those married with children, display a paradoxical tolerance for extreme housing pressure. This profoundly reflects the trade-offs they are forced to make to access employment opportunities in large cities. Simultaneously, the effect of housing type is more absolute, demonstrating a clear bifurcation based on property rights. The transition from renting to owning represents a decisive \"pull factor\" for all groups, dramatically enhancing settlement intention. This sense of stability and belonging conferred by homeownership is especially pronounced among the more resource-constrained low-education group, underscoring the pivotal role of acquiring property rights in their settlement decisions.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"5 Discussion \u0026 conclusion","content":"\u003cp\u003eExisting literature has firmly established the important role of economic incentives, urban amenities, and policy factors in talent mobility (Wang \u0026amp; Shen, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Liu \u0026amp; Wang, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Fan, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), but the non-linear mechanisms of employment and housing factors have received insufficient attention. Traditional research often employs econometric methods like linear regression to analyze migrants' settlement intention, assuming a monotonic relationship between influencing factors and the outcome (De Jong \u0026amp; Graefe, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Wagner \u0026amp; Mulder, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). However, this linear assumption may fail to capture complex threshold effects and the mechanisms of critical point transitions. Employing the XGBoost-SHAP machine learning method and data from the China Migrants Dynamic Survey, this study develops an analytical framework based on the life course perspective to systematically examine the non-linear effects of employment and housing on the settlement intention of migrant populations with different education levels in small cities. The study identifies the critical boundaries where push-pull factors transform, pinpoints differentiated threshold sensitivities across educational divides, and reveals the institutional moderating mechanisms of life course events, offering a new theoretical perspective for understanding the underlying logic of talent competition in large cities.\u003c/p\u003e \u003cp\u003eThe results confirm Hypothesis \u003cspan refid=\"FPar1\" class=\"InternalRef\"\u003e1\u003c/span\u003e: employment (working) and housing factors exhibit significant non-linear threshold effects on settlement intention. This effect is particularly pronounced in the dimension of housing tenure (housing type): regardless of a migrant's educational background or life course stage, acquiring property rights is consistently the most powerful driver of settlement intention. More interestingly, we find that the \u0026ldquo;lower bound\u0026rdquo; of these thresholds plays a more critical role in decision-making than the \u0026ldquo;upper bound.\u0026rdquo; For the high-education group, the response to average wage gap and monthly personal income presents a seemingly paradoxical pattern: unmarried individuals show a moderate preference for cities with above-average wages but a strong aversion to cities with extremely high personal incomes; meanwhile, individuals with children significantly avoid cities with low average wages and are also negatively disposed toward cities with high personal incomes. This finding aligns with theoretical expectations from urban economics concerning the diminishing marginal returns of agglomeration (Duranton \u0026amp; Puga, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Ottaviano \u0026amp; Peri, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). That is, once a city reaches a certain level of development, the economic advantages of matching, sharing, and learning are offset by high congestion costs and living expenses. Migrants' decisions, therefore, are not about maximizing returns but about making trade-offs within an acceptable satisficing range.\u003c/p\u003e \u003cp\u003eContrary to the common expectation of Hypothesis \u003cspan refid=\"FPar2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the analysis further reveals that high-education migrants actually have a lower tolerance for the pressures of large cities than their low-education counterparts. This anomaly is consistently validated across multiple indicators. Regarding wage levels, high-education individuals (especially the unmarried and married-childless) show greater tolerance for cities with below-average wages, whereas low-education individuals gravitate decisively toward regions with above-average wages. In terms of housing affordability, when the price-to-income ratio reaches extremely high levels, low-education individuals with children even exhibit a strong positive settlement intention, whereas high-education individuals turn negative at a much lower pressure threshold. We posit that this difference is rooted in the varying ways their human capital advantages are manifested in spatial choices. With their versatile human capital, high-education migrants can secure decent employment and a reasonable price-to-income ratio even in small and medium-sized cities, thus achieving a work-life balance (Li et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Kerr et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). In contrast, for low-education migrants, high-quality job opportunities in large cities are almost the sole pathway to higher income (Asher \u0026amp; Novosad, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Consequently, when housing prices in large cities cross a critical threshold, a large number of low-education individuals are forced to make a compromise between employment and residence, accepting a long-term state of non-propertied residence. This mode of settlement, lacking equal access to local public services and social welfare, is essentially a form of \u0026ldquo;incomplete\u0026rdquo; or \u0026ldquo;floating\u0026rdquo; settlement\u0026mdash;a state of precarious sojourning rather than permanent integration.\u003c/p\u003e \u003cp\u003eThe analysis further validates the expectation of Hypothesis \u003cspan refid=\"FPar3\" class=\"InternalRef\"\u003e3\u003c/span\u003e regarding differentiated life course impacts, but also reveals significant heterogeneity in this effect across educational groups. Among the high-education group, the progression through the life course aligns with theoretical expectations: individuals married with children, facing child-rearing issues tied to the \u003cem\u003ehukou\u003c/em\u003e system, have the lowest tolerance for expensive and competitive large cities, yet they also cannot accept small cities with inadequate infrastructure and opportunities. The low-education group, however, exhibits a counter-intuitive pattern: even after having children, they maintain a strong preference for megacities. We infer that this phenomenon stems from the unequal distribution of institutional support across social strata. High-education talent can more easily obtain urban \u003cem\u003ehukou\u003c/em\u003e and access to public education through talent acquisition policies, enabling them to make trade-offs between cities of different tiers and ultimately settle in a medium-sized city with a higher quality of life. Conversely, due to institutional constraints, low-education individuals can often only meet their children's educational needs by returning to their hometowns. However, returning home means facing scarce job opportunities and lower income. This structural dilemma compels them to remain in large cities that offer better economic returns even after having children\u0026mdash;a trend that may be mechanistically linked to the widespread phenomenon of \u0026ldquo;left-behind children\u0026rdquo; in China.\u003c/p\u003e \u003cp\u003eThis study expands the analytical framework of talent mobility research along two dimensions. On the theoretical front, first, by identifying threshold effects for employment and housing factors, the study transcends the linear framework of traditional push-pull theory, demonstrating that key factors have distinct points of critical transition and providing a more refined analytical lens for understanding migration complexity. Second, the study reveals the negative correlation between education level and tolerance for urban pressures, and clarifies how the moderating effect of life course events is heterogeneous by educational background. This finding challenges the conventional wisdom that higher human capital equates to greater adaptability in large cities, proposing that the true advantage of human capital lies not merely in enhancing survival capacity in large cities, but in expanding an individual\u0026rsquo;s spatial choice architecture, thereby fundamentally altering their migration decision-making logic. On the methodological front, this research demonstrates the potential of machine learning methods (XGBoost-SHAP) in social science research. This approach enables the effective identification of complex non-linear relationships and interaction effects that are often missed by traditional linear regression models, providing a powerful tool for precisely measuring thresholds and critical points. This not only enhances the robustness of our findings but also offers a new methodological toolkit for future migration research.\u003c/p\u003e \u003cp\u003eThis study also holds significant policy implications. First, in the current landscape of fierce talent competition, the most competitive cities may not be the megacities with prohibitive living costs, nor small towns lacking opportunities, but rather the medium-sized cities that offer an optimal balance between job opportunities, public services, and cost of living. These cities possess a unique competitive advantage in attracting and retaining high-skilled talent by providing both robust infrastructure and quality employment for them and their families. Second, policymakers must pay close attention to the long-term well-being of the low-skilled migrant population in large cities. This group, while contributing immensely to urban development, is often excluded from property ownership and quality public services (especially education for their children), leaving them in a long-term floating state of economic inclusion but social exclusion. This not only undermines social equity but may also pose long-term social risks. Therefore, this situation requires coordinated and more inclusive urbanization strategies from both sending and receiving governments, such as reforming the \u003cem\u003ehukou\u003c/em\u003e system, expanding the supply of affordable housing, and opening up school enrollment channels for migrant children to help them achieve genuine social integration.\u003c/p\u003e \u003cp\u003eThis study provides a crucial foundation for future research on population or talent mobility. Subsequent research should further explore the dynamic nature of these threshold effects and delve deeper into their underlying psychological and institutional mechanisms. One promising avenue is to use longitudinal data to examine how threshold parameters evolve across different stages of urban development and how the interplay between housing and employment factors jointly shapes settlement decision thresholds. Another valuable line of inquiry is to analyze the interaction between life course events and the institutional environment, particularly the long-term impact of policy changes on individual decision-making trajectories. Furthermore, given the significant differences in institutional structures and cultural contexts across countries, conducting cross-national comparative studies to test the universality and particularity of these threshold effects and educational divides would undoubtedly enrich our understanding of talent mobility as a global phenomenon.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research is based upon work supported by the National Natural Science Foundation of China (NSFC) [Grant No. 42271219, 42171203]; and the Fundamental Research Funds for the Central Universities [Grant No. 413100119].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Approval Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study is a secondary analysis of the 2017 China Migrants Dynamic Survey (CMDS) dataset. The data were collected by the National Health Commission of the People's Republic of China. All data used in this study were fully anonymized before we accessed them. As this research involved no direct contact with human participants and utilized a publicly available, de-identified dataset, it did not require separate ethical approval from an institutional review board, in accordance with the institutional and national research ethics regulations for secondary data analysis. The ethical procedures for the original data collection were the responsibility of the institution that conducted the survey, the National Health Commission of China, which performed the research in accordance with relevant national guidelines and regulations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformed Consent Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study is based on a secondary analysis of a publicly available dataset. Therefore, the authors of this study did not obtain informed consent directly from the participants. Informed consent was obtained from all individual participants included in the original survey by the data collectors from the National Health Commission of China at the time of data collection. The publicly available dataset used in our analysis has been fully de-identified, ensuring participant confidentiality.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAsher S, Novosad P (2020) Costs and benefits of rural-urban migration: Evidence from India. J Dev Econ 146:104481. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jdeveco.2020.104481\u003c/span\u003e\u003cspan address=\"10.1016/j.jdeveco.2020.104481\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao G, Li M, Ma Y, Tao R (2015) Self-employment and intention of permanent urban settlement: Evidence from a survey of migrants in China. J Dev Econ 114:24\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jdeveco.2015.02.004\u003c/span\u003e\u003cspan address=\"10.1016/j.jdeveco.2015.02.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChan KW (2010) Fundamentals of China\u0026rsquo;s urbanization and policy. China Rev 10(1):63\u0026ndash;93\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoulter R, Van Ham M (2013) Following people through time: An analysis of individual residential mobility biographies. Hous Stud 28(7):1037\u0026ndash;1055. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02673037.2013.783903\u003c/span\u003e\u003cspan address=\"10.1080/02673037.2013.783903\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCourgeau D (1990) Migration, family, and career: A life course approach. In: Baltes PB, Featherman DL, Lerner RM (eds) Life-span development and behaviour. Psychology\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Jong GF, Graefe DR (2008) Family life course transitions and the economic consequences of internal migration. Popul Space Place 14:267\u0026ndash;282. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/psp.481\u003c/span\u003e\u003cspan address=\"10.1002/psp.481\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDu H, Hui EC, Chen L (2019) Measuring the housing precarity of migrant private tenants and assessing its impact on their life satisfaction. Cities 90:1\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2019.01.031\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2019.01.031\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuranton G, Puga D (2004) Micro-foundations of urban agglomeration economies. In J. V. Henderson \u0026amp; J. F. Thisse (Eds.), \u003cem\u003eHandbook of Regional and Urban Economics\u003c/em\u003e (Vol. 4, pp. 2063\u0026ndash;2117). Elsevier. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1574-0080(04)80005-1\u003c/span\u003e\u003cspan address=\"10.1016/S1574-0080(04)80005-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElder GH Jr. (1994) Time, human agency, and social change: Perspectives on the life course. Social Psychol Q 57(1):4\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2307/2786971\u003c/span\u003e\u003cspan address=\"10.2307/2786971\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElder GH Jr., Johnson M, Crosnoe R (2003) The emergence and development of life course theory. In J. T. Mortimer \u0026amp; M. Shanahan (Eds.), \u003cem\u003eHandbook of the life course\u003c/em\u003e (pp. 3\u0026ndash;22). Kluwer. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-0-306-48247-2_1\u003c/span\u003e\u003cspan address=\"10.1007/978-0-306-48247-2_1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFan CC (2011) Settlement intention and split households: Findings from a survey of migrants in Beijing\u0026rsquo;s urban villages. China Inform 25(2):253\u0026ndash;274. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0920203X11407906\u003c/span\u003e\u003cspan address=\"10.1177/0920203X11407906\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFindlay A, McCollum D, Coulter R, Gayle V (2015) New mobilities across the life course: A framework for analysing demographically linked drivers of migration. Popul Space Place 21(4):390\u0026ndash;402. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/psp.1956\u003c/span\u003e\u003cspan address=\"10.1002/psp.1956\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGan X, Zuo J, Ye K, Li D, Chang R, Zillante G (2016) Are migrant workers satisfied with public rental housing? A study in Chongqing, China. Habitat Int 56:96\u0026ndash;102. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.habitatint.2016.05.006\u003c/span\u003e\u003cspan address=\"10.1016/j.habitatint.2016.05.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGeist C, McManus PA (2008) Geographical mobility over the life course: Motivations and implications. Popul Space Place 14(4):283\u0026ndash;303. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/psp.508\u003c/span\u003e\u003cspan address=\"10.1002/psp.508\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGong F, Xu J, Fujishiro K, Takeuchi DT (2011) A life course perspective on migration and mental health among Asian immigrants: The role of human agency. Soc Sci Med 73(11):1618\u0026ndash;1626. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.socscimed.2011.09.014\u003c/span\u003e\u003cspan address=\"10.1016/j.socscimed.2011.09.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGu H (2021) Understanding the migration of highly and less-educated labourers in post-reform China. Appl Geogr 137:102605. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.apgeog.2021.102605\u003c/span\u003e\u003cspan address=\"10.1016/j.apgeog.2021.102605\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu F, Xu Z, Chen Y (2011) Circular migration, or permanent stay? Evidence from China\u0026rsquo;s rural\u0026ndash;urban migration. China Econ Rev 22(1):64\u0026ndash;74. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.chieco.2010.09.007\u003c/span\u003e\u003cspan address=\"10.1016/j.chieco.2010.09.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang S, Hou J, Sun L, Dou D, Liu X, Zhang H (2017) The effects of objective and subjective socioeconomic status on subjective well-being among rural-to-urban migrants in China: The moderating role of subjective social mobility. Front Psychol 8:819. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyg.2017.00819\u003c/span\u003e\u003cspan address=\"10.3389/fpsyg.2017.00819\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang Y, Liu W, Zhuo Y, Sun X (2020) Social insurance participation and urban settlement intentions of China\u0026rsquo;s floating population. Popul Space Place 26(8):e2366. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/psp.2366\u003c/span\u003e\u003cspan address=\"10.1002/psp.2366\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKerr SP, Kerr WR, \u0026Ouml;zden \u0026Ccedil;, Parsons C (2017) Global talent flows. J Economic Perspect 31(3):83\u0026ndash;106. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1257/jep.31.3.83\u003c/span\u003e\u003cspan address=\"10.1257/jep.31.3.83\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhoo SE, Hugo G, McDonald P (2008) Which skilled temporary migrants become permanent residents and why? Int Migrat Rev 42(1):193\u0026ndash;226. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1747-7379.2007.00118.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1747-7379.2007.00118.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKulu H, Milewski N (2007) Family change and migration in the life course: An introduction. Demographic Res 17:567\u0026ndash;590. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4054/DemRes.2007.17.19\u003c/span\u003e\u003cspan address=\"10.4054/DemRes.2007.17.19\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi L, Liu Y (2019) Understanding the gap between de facto and de jure urbanization in China: A perspective from rural migrants\u0026rsquo; settlement intention. Popul Res Policy Rev 39:311\u0026ndash;338. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11113-019-09532-1\u003c/span\u003e\u003cspan address=\"10.1007/s11113-019-09532-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Y, Chen X, Wang Z (2022) Tolerance for housing unaffordability among highly skilled young migrants: Evidence from the Zhejiang Province of China. Int J Environ Res Public Health 20(1):616. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijerph20010616\u003c/span\u003e\u003cspan address=\"10.3390/ijerph20010616\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin S, Wu F, Li Z (2020) Social integration of migrants across Chinese neighbourhoods. Geoforum 112:118\u0026ndash;128. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.geoforum.2020.04.009\u003c/span\u003e\u003cspan address=\"10.1016/j.geoforum.2020.04.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu T, Wang J (2020) Bringing city size in understanding the permanent settlement intention of rural\u0026ndash;urban migrants in China. Popul Space Place 26(4):e2295. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/psp.2295\u003c/span\u003e\u003cspan address=\"10.1002/psp.2295\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu T, Xiao W (2022) The \u0026lsquo;intention to stay\u0026rsquo; of the floating migrant population: A spatiotemporal meta-analysis in Chinese cities. Popul Space Place 28(4):e2574. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/psp.2574\u003c/span\u003e\u003cspan address=\"10.1002/psp.2574\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Y, Liu Y, Lin Y (2021) Upward or downward comparison? Migrants\u0026rsquo; socioeconomic status and subjective wellbeing in Chinese cities. Urban Stud 58(12):2490\u0026ndash;2513. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0042098020968965\u003c/span\u003e\u003cspan address=\"10.1177/0042098020968965\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Z, Wang Y, Chen S (2016) Does formal housing encourage settlement intention of rural migrants in Chinese cities? A structural equation model analysis. Urban Stud 54(8):1834\u0026ndash;1850. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0042098016634979\u003c/span\u003e\u003cspan address=\"10.1177/0042098016634979\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuan X, Paryzat H, Chu J, Liu W, Liu Z, Li C (2024) Different roads take me home: The nonlinear relationship between distance and flows during China\u0026rsquo;s Spring Festival. Humanit Social Sci Commun 11:1356. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1057/s41599-024-03779-8\u003c/span\u003e\u003cspan address=\"10.1057/s41599-024-03779-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOttaviano GI, Peri G (2006) The economic value of cultural diversity: Evidence from US cities. J Econ Geogr 6(1):9\u0026ndash;44. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/jeg/lbi002\u003c/span\u003e\u003cspan address=\"10.1093/jeg/lbi002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVidal S, Lutz K (2018) Internal migration over young adult life courses: Continuities and changes across cohorts in West Germany. Adv Life Course Res 36:45\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.alcr.2018.03.003\u003c/span\u003e\u003cspan address=\"10.1016/j.alcr.2018.03.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWagner M, Mulder CH (2015) Spatial mobility, family dynamics, and housing transitions. \u003cem\u003eK\u0026ouml;lner Zeitschrift f\u0026uuml;r Soziologie und Sozialpsychologie\u003c/em\u003e, \u003cem\u003e67\u003c/em\u003e(Supplement 1), 111\u0026ndash;135. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11577-015-0327-4\u003c/span\u003e\u003cspan address=\"10.1007/s11577-015-0327-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang C, Shen J (2022) Settlement intention of migrants in urban China: The effects of labor-market performance, employment status, and social integration. Appl Geogr 147:102773. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.apgeog.2022.102773\u003c/span\u003e\u003cspan address=\"10.1016/j.apgeog.2022.102773\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang C, Shen J (2023) How subjective economic status matters: The reference-group effect on migrants\u0026rsquo; settlement intention in urban China. Asian Popul Stud 19(1):105\u0026ndash;123. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/17441730.2021.2012025\u003c/span\u003e\u003cspan address=\"10.1080/17441730.2021.2012025\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang CL, Shen J, Liu Y (2021) Administrative level, city tier, and air quality: Contextual determinants of hukou conversion for migrants in China. Comput Urban Sci 1(1):1\u0026ndash;16. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s43762-021-00015-0\u003c/span\u003e\u003cspan address=\"10.1007/s43762-021-00015-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang L, Acheampong RA, He S (2020) High-speed rail network development effects on the growth and spatial dynamics of knowledge-intensive economy in major cities of China. Cities 105:102772. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2020.102772\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2020.102772\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang L, Duan X (2018) High-speed rail network development and winner and loser cities in megaregions: The case study of Yangtze River Delta, China. Cities 83:71\u0026ndash;82. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2018.06.010\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2018.06.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Y (2024) Individual preferences: Jobs or amenities? Understanding the destination choices of highly and less educated migrants in China. Appl Geogr 162:103169. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.apgeog.2023.103169\u003c/span\u003e\u003cspan address=\"10.1016/j.apgeog.2023.103169\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei YD, Wu Y, Xiao W, Wu X (2023) Preferences and barriers of vulnerable urban populations in affordable housing in urban China. J Urban Plan Dev 149(4):05023030. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1061/JUPDDM.UPENG-4271\u003c/span\u003e\u003cspan address=\"10.1061/JUPDDM.UPENG-4271\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu F (2004) Residential relocation under market-oriented redevelopment: The process and outcomes in urban China. Geoforum 35(4):453\u0026ndash;470. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.geoforum.2004.01.004\u003c/span\u003e\u003cspan address=\"10.1016/j.geoforum.2004.01.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu F, Gaubatz P (2020) The Chinese City. Routledge. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4324/9780429052682\u003c/span\u003e\u003cspan address=\"10.4324/9780429052682\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu W, Stephens M, Du M et al (2019) Homeownership, family composition and subjective wellbeing. Cities 84:46\u0026ndash;55. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cities.2018.07.008\u003c/span\u003e\u003cspan address=\"10.1016/j.cities.2018.07.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang S, Guo F (2018) Breaking the barriers: How urban housing ownership has changed migrants\u0026rsquo; settlement intentions in China. Urban Stud 55(16):3689\u0026ndash;3707. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0042098017741265\u003c/span\u003e\u003cspan address=\"10.1177/0042098017741265\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYue Z, Li S, Feldman MW, Du H (2010) Floating choices: A generational perspective on intentions of rural\u0026ndash;urban migrants in China. Environ Plann A 42(3):545\u0026ndash;562. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1068/a4213\u003c/span\u003e\u003cspan address=\"10.1068/a4213\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu Y (2007) China\u0026rsquo;s floating population and their settlement intention in the cities: Beyond the Hukou reform. Habitat Int 31(1):65\u0026ndash;76. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.habitatint.2006.04.002\u003c/span\u003e\u003cspan address=\"10.1016/j.habitatint.2006.04.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu Y, Chen W (2010) The settlement intention of China\u0026rsquo;s floating population in the cities: Recent changes and multifaceted individual-level determinants. Popul Space Place 16(4):253\u0026ndash;267. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/psp.544\u003c/span\u003e\u003cspan address=\"10.1002/psp.544\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu Y, Lin L (2014) Continuity and change in the transition from the first to the second generation of migrants in China: Insights from a survey in Fujian. Habitat Int 42:147\u0026ndash;154. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.habitatint.2013.12.002\u003c/span\u003e\u003cspan address=\"10.1016/j.habitatint.2013.12.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZou D, Li Q, Zhou Y, Liang S, Zhou S (2023) Understanding the nonlinear relationship between the built environment and non-mandatory travel: A machine learning approach. Appl Geogr 161:103118. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.apgeog.2023.103118\u003c/span\u003e\u003cspan address=\"10.1016/j.apgeog.2023.103118\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Settlement intention, Small-town migrants, Life course, Machine learning, Education divide, China","lastPublishedDoi":"10.21203/rs.3.rs-7845488/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7845488/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMigration from small towns to metropolises is a global trend, but migrants often face a dilemma between employment opportunities and prohibitive living costs. In China, this is exacerbated by institutional barriers like the hukou system, trapping many in a state of temporary residence without full welfare benefits. Settlement intention is thus a complex nonlinear process, yet understanding of these dynamics remains limited. Employing the XGBoost-SHAP methodology on China Migrants Dynamic Survey data, this study analyzes the non-linear effects of employment and housing on the settlement intentions of small-town migrants. The findings reveal significant threshold effects, where factors of urban attraction transform into repulsion. Counterintuitively, highly educated migrants exhibit a lower tolerance for metropolitan pressures, as their versatile human capital allows them to find a better work-life balance in medium-sized cities. In contrast, less-educated migrants show higher tolerance, compelled by constrained employment alternatives to accept precarious incomplete settlement. This educational divide is sharpened by life course events: for the highly educated, having children decreases their tolerance for megacities, while a structural dilemma forces the less-educated to remain even after having children.\u003c/p\u003e","manuscriptTitle":"For jobs or housing in the Metropolis: A Life Course Analysis of Small-town Migrants’ Settlement Intentions Through Machine Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-19 10:10:01","doi":"10.21203/rs.3.rs-7845488/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-11T10:09:24+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-09T09:55:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-14T13:55:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"141664457398973497578521030024785078407","date":"2026-01-16T09:40:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"144132857955307119213336382155384284905","date":"2026-01-15T13:59:11+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-01-14T14:23:55+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-11-11T09:06:49+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-11-11T09:01:59+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-11-09T08:39:25+00:00","index":"","fulltext":""},{"type":"submitted","content":"Humanities and Social Sciences Communications","date":"2025-11-09T08:36:16+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"8d2c2358-859a-4c1e-aeaf-29a5eb4b25c2","owner":[],"postedDate":"January 19th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[{"id":61233821,"name":"Scientific community and society/Geography"},{"id":61233822,"name":"Social science/Geography"},{"id":61233823,"name":"Social science/Science technology and society"},{"id":61233824,"name":"Social science/Sociology"}],"tags":[],"updatedAt":"2026-04-11T10:24:54+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-19 10:10:01","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7845488","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7845488","identity":"rs-7845488","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}