Measuring mass displacement of urban renewal in Shenzhen,China: Using longitudinal mobile phone trajectory data

Abstract

A large-scale urban renewal programme in many Chinese cities has resulted in residential displacement, raising concerns about its negative consequences. However, quantitative evidence is scarce. Utilising mobile signalling data that records continuous individual movements, we devise a strategy for measuring mass displacement caused by urban renewals, where a large number of migrant tenants are forced to move at the same time. Focusing on multiple urban renewal projects in Shenzhen, a pioneer city in urban renewal practices in China, we estimate the effects of mass displacement on the living conditions of displaced residents using both a difference-in-differences approach and a machine learning approach. The results show that, compared with relocations unaffected by renewal, displaced residents relocated to areas with worse housing quality and poor access to urban amenities, and experienced longer commutes, the pattern of which is more severe for urban renewals in the central area of the city. The aggregate displacement indices derived from the support vector machine model indicate that 25% of the displaced experienced a worsening of living conditions following the relocation. Our findings suggest significant adverse consequences of mass displacement as a result of large-scale urban renewal.

Keywords

Urban redevelopment displacement urban village relocation residential mobility

Introduction

Urban renewal programmes have been implemented by many governments worldwide as means of addressing urban issues and promoting urban prosperity (Nachmany and Hananel, 2023). This includes HOPE VI programme in the USA, the Big Cities Policy in the Netherlands, the Housing Market Renewal Pathfinders and ‘New Urban Renewal’ in England and urban renewal in China and France (Goetz, 2002; Kleinhans and Kearns, 2013; Lees, 2014; Li et al., 2019). However, most urban renewals face tensions with respect to residents’ outcomes. Despite improving the quality of life for new residents in renewal sites, urban renewal may result in the displacement of incumbent low-income residents, exacerbating both economic and social injustice (Zuk et al., 2018). Scholarly interest in displacement dates back to the mid-20th century, when post-war slum clearance processes in the US and UK prompted the earliest criticism of the destructive effects of urban renewal (Kearns and Mason, 2013). Since then, residential displacement has become a dominant analytical lens for understanding the adverse effects of urban renewal (Doff and Kleinhans, 2011; Goetz, 2002; Kearns and Mason, 2013; Reades et al., 2023). Measuring residential displacement as an outcome of urban renewal is no easy, however, which may explain the mixed evidence in the literature (Li et al., 2019; Easton et al., 2020).

Primarily, the analysis of displacement is complicated by its multifaceted nature (Elliott-Cooper et al., 2020; Wang, 2020). As a result of urban renewal, many forms of displacement can occur, such as direct displacements due to physical changes and indirect displacements as a result of economic factors (Marcuse, 1985). Residents can experience very different outcomes as a result of different types of displacement, which are further complicated by a variety of market and policy contexts, as well as different renewal approaches and relocation practices (Kleinhans and Kearns, 2013; Li et al., 2019; Reades et al., 2023). In order to encompass the diversity of displacement, its conceptual meaning is further expanded, yet this causes debates regarding its negative narrative and warnings that the concept might become obscure and illusory. The complexity of displacement cannot be adequately grasped without appropriate data (Reades et al., 2023). While scholars use various qualitative and quantitative methods to address who and how many people are displaced and what are the consequences of displacement, most analyses either suffer from small sample sizes through survey or from weaknesses in spatial resolution and temporal resolution through census data (Easton et al., 2020).

In this article, we deviate from the line of displacement analysis that moves beyond Marcuse (1985) and focuses on direct, measurable displacement as a result of forced relocations due to urban renewal. Rather than dealing with the conceptual complexity, we address the practical limitations caused by the lack of longitudinal data in measuring displacement, which is necessary for assessing the changes of the living conditions of displaced residents over time. By utilising longitudinal mobile signalling data with high spatiotemporal resolution, we devise a strategy for tracking mass displacement caused by urban renewals, where a large number of residents are forced to move at the same time. A clear identification of such displacements is crucial to drafting policy that addresses the harmful effects of urban renewal. In the empirical analysis, we first apply a difference-in-differences (DID) approach to assess the living conditions of displaced residents, considering housing and neighbourhood characteristics as well as travel behaviours. Next, we use a support vector machine (SVM) model to analyse these factors collectively and construct a displacement index for the sampled urban renewal projects. In this regard, we can perform a more accurate and comprehensive assessment of the adverse effects of displacement.

Our study case is Shenzhen, a pioneer city in urban renewal practices in China. The large-scale urban renewal in Shenzhen shares similarities with the gentrification processes discussed in Western academia, such as the substantial transformation of neighbourhoods and the displacement of low-income residents (Dai et al., 2023; Liu et al., 2018a). However, it also diverges in key aspects due to differences in political, economic and social contexts. Specifically, Shenzhen’s urban renewal involves extensive demolition and redevelopment of urban villages, leading to a level of displacement that surpasses that seen in many other cities. Informal settlements in urban villages, originally rural areas now encircled by urban developments, accommodate a significant number of migrant workers who are excluded from the formal housing market (Liu et al., 2018a, 2018b; Wu, 2016). When large-scale renewal occurs in urban villages, migrant workers, typically tenants, will be evicted on a large scale within a short period of time without compensation (Huang et al., 2018). In spite of this, displaced tenants have remained largely invisible to researchers since it is difficult to find them (Liu et al., 2018a; Wang, 2020). With the help of longitudinal mobile signalling data, we can identify the mass displacement of tenants caused by multiple urban renewal projects in Shenzhen and provide estimates of the displacement effects.

Our findings suggest significant adverse effects of mass displacement due to large scale urban renewals. Compared to relocations unaffected by renewal, displaced residents moved to poorer locations and ended with longer commutes, the pattern of which is more severe for urban renewals in the central area of the city. We find consistent evidence using the aggregate displacement indices. It shows that, on average, 25% of those displaced experienced worse living conditions following their relocation. Given the challenges of tracking hard-to-reach migrants, this work represents a significant contribution to the quantitative analysis of renewal-induced displacement in the Chinese context. The paper is structured as follows: The next section reviews displacement in urban renewal and discusses displacement resulting from Shenzhen’s urban renewal. Section 3 details our methodology, including mass displacement identification, relocation variables and empirical models. Section 5 presents the results, and the final section concludes.

Displacement led by urban renewal

Literature review

Since the post-war slum clearance processes in the US and UK in the 1950s, displacement has been a central concept in evaluating the social effect of urban renewal. It refers to the process of residents being forced to move from their original residence (Marcuse, 1985). Typically, direct displacement occurs when incumbent occupants are evicted before demolition begins, and it is considered to be the most extreme type of displacement, sometimes involving violent tactics such as cutting off gas and electricity (Lees, 2014; Marcuse, 1985). Even if incumbent residents manage to stay put, rent increases and sociodemographic shifts may make neighbourhoods less affordable and liveable, leading to indirect displacement that occurs in less violent ways (Marcuse, 1985). Residents may be more adversely affected by direct displacement than by indirect displacement, a topic that has sparked considerable debate in recent years. Particularly, residents subject to direct displacement have to cope with time pressure to relocate and face near-term housing insecurity, causing anxiety and stressful experiences (Baeten et al., 2017; Li et al., 2019). After forced relocation, residents are likely to experience worsening living conditions and reduced accessibility to urban opportunities (Huang et al., 2018), disruption to their social networks (Yang et al., 2024), and the loss of place attachment and difficulty adapting to new lifestyles (Watt, 2022). When central cities continue to be gentrified through large-scale renewal projects, low-income residents may face fewer affordable housing options, forcing them to leave and exacerbating social segregation (Zuk et al., 2018).

Despite widespread concerns over the adverse effects of displacement, some scholars have challenged this predominately negative narrative (Kirk, 2024; Kleinhans and Kearns, 2013; Li et al., 2019). Many studies reporting negative outcomes of displacement are based on small samples, which may not capture the broad picture (Baeten et al., 2017; Liu et al., 2018a; Watt, 2022). To assess the impact of displacement with statistical representativeness, researchers often compare patterns of involuntary and voluntary relocations, but it is difficult to differentiate between the two. It is possible that residents living in redeveloped areas, sometimes the worst neighbourhoods in a city, already intend to move and see forced relocation as an opportunity to improve their living conditions (Li and Song, 2009; Bolt and van Kempen, 2010; Huang et al., 2018). As with voluntary movers, their location choice will be determined by their socio-economic status and the housing market conditions (Li et al., 2019). Empirical evidence from the Netherlands shows no significant difference between displaced residents and regular movers in terms of the types of destination neighbourhoods they relocate to (Bolt and van Kempen, 2010). In addition, compensation and assistance mechanisms may enable displaced residents to relocate to alternative or even better housing (Goetz, 2002; Kearns and Mason, 2013). Reades et al. (2023) examined the impacts of estate regeneration on London’s social housing residents after forced relocation. Their findings show that many residents made short-distance relocations within the same borough despite inadequate monetary compensation, yet ended up in more deprived neighbourhoods. As a result, the outcomes of displacement involve a complex interplay between institutional contexts, market forces and individual characteristics, and some residents are affected by displacement more severely than others (Elliott-Cooper et al., 2020).

Accurately identifying displacement is a prerequisite for analysing which types of urban renewal contexts are more likely to result in displacement. However, empirical studies of displacement have long been hindered by a lack of adequate data (Easton et al., 2020; Zuk et al., 2018). Tracing displacement requires longitudinal data of residential movements at a sufficiently high spatiotemporal resolution (Easton et al., 2020). Some state-led renewal projects conducted official surveys to document the relocation outcomes of affected residents, but these surveys were unable to provide adequate information relevant to researchers (Doff and Kleinhans, 2011). Further, official records frequently failed to trace those displaced residents living in the most precarious housing conditions (Baeten et al., 2017). A majority of previous studies relied on interviews and questionnaires, supplemented by censuses or other administrative records, to examine whether and how displacement occurs (Bolt and van Kempen, 2010; Goetz, 2002; Kearns and Mason, 2013). The last few years have seen efforts to ‘move beyond conventional census-based measures’ (Easton et al., 2020). New data sources are opening up opportunities to trace residential movements with greater accuracy. Several empirical studies conducted in the US and UK have utilised detailed consumer and administrative data to analyse the relationship between residential mobility, gentrification and urban redevelopment (Reades et al., 2023). In light of these studies, it is important to use continuous individual-level data to reveal changes in residents’ relocation patterns and to provide a more accurate understanding of the consequences of displacement.

Displacement resulting from Shenzhen’s urban renewal

Intertwined with gentrification, the study of displacement caused by urban renewal in the Chinese context is relatively recent (Huang et al., 2018; Li et al., 2019; Liu et al., 2018a; Niu et al., 2024; Wang, 2020). As Chinese cities rapidly urbanised, local governments frequently implemented a demolition-redevelopment strategy to restructure urban spaces and make way for new development (He and Wu, 2005). As China’s first Special Economic Zone (SEZ), Shenzhen has been a pioneer in urban renewal, undertaking demolition and redevelopment at an unprecedented scale. However, existing studies have not sufficiently examined displacement resulting from Shenzhen’s large-scale renewal, a pressing issue for two key reasons.

Firstly, a large number of urban renewal projects in Shenzhen involve urban villages that accommodate a substantial number of migrants and low-income workers. Urban villages have grown rapidly as a result of the city’s rapid urbanisation and dual ownership of rural and urban lands (Wu et al., 2013). Lax development controls in urban villages have led to extensive illegal construction, resulting in numerous unauthorised and noncompliant housing units owned by rural collective villagers (Wu et al., 2013). These housing units offer essential accommodations for migrants and low-income workers, with significantly lower rents (Wu, 2016). In the era of urban renewal, however, these urban villages have become major land resources to sustain urban development. The original villagers, who are few in number, receive economic compensation, but the large number of tenants cannot participate in the urban renewal decision-making process and they are not entitled to renewal compensation (Liu et al., 2018a). In addition, tenants in urban villages rarely sign formal contracts with their landlords, making them the most vulnerable group in the urban renewal process (Wu, 2016). As a result, mass displacement may have a more adverse impact on the relocation destinations of displaced tenants and the overall quality of their lives post-displacement.

Secondly, urban renewal practices in Shenzhen are implemented at an accelerated rate with the participation of market players. Urban renewal processes in Shenzhen exemplify what Wu (2018) described as a combination of ‘planning centrality’ and ‘market instruments’. The local government is strongly motivated to restructure its inner city, and the real estate industry is eager to reap the benefits of favourable policies (He and Wu, 2005). Shenzhen developed the country’s first legal framework for urban renewal in 2009, establishing marketised operations with government guidance as the guiding principle for the city’s urban renewal practice. The active involvement of market participants has contributed to the unprecedented scale of urban renewal in the city. By the end of 2022, Shenzhen had included 1010 projects in its urban renewal plan, including 996 demolition-redevelopment projects totalling approximately 8400 hectares, and over 10% of the projects involve urban villages. More than half of the projects have entered the implementation stage or have been completed, according to statistics compiled by the authors from multiple government websites.

Therefore, Shenzhen has been undergoing extensive urban renewal projects that involve demolition and redevelopment of urban villages, resulting in more direct displacement than in other cities. This displaced group, which is large in number and predominantly composed of tenants, is difficult to trace due to their high mobility. Liu and her co-authors investigated how migrants in Shenzhen experienced different types of displacement during the redevelopment of urban villages and the consequences of such displacement (Liu et al., 2018a, 2018b). According to their findings, displaced migrants faced increased living costs and lost job opportunities, and they struggled to remain in nearby areas to maintain their social and economic networks (Liu et al., 2018a). However, the generalisability of these results, which are based on a few case studies, requires further investigation. In a recent quantitative study, Dai et al. (2023) documented similar findings examining the demolition and redevelopment of Baishizhou Village in Shenzhen. Based on mobile signalling data that tracked the residential trajectory of over 50,000 tenants, they found that the majority of displaced tenants experienced a decreased quality of life following relocation. However, Dai et al. (2023) did not establish a causal link between urban renewal and displacement, which could lead to inaccurate assessments of the effects of urban renewal. Therefore, there is still very little known about the life of tenants who were displaced and began an uncertain journey following urban renewal displacement.

Using a post-displacement lens, we examine the mass displacement caused by urban renewal projects in Shenzhen in order to shed light on the negative consequences of such displacements. Unlike indirect displacement, where residents are gradually priced out of gentrifying neighbourhoods, mass displacement occurs when a large number of residents are forced to move at short notice. The sudden rise in demand for affordable housing makes it more difficult to find alternative housing in a short period of time. In this regard, we expect mass displacement to have a more detrimental effect on displaced residents, which can be evaluated from two aspects.

First, it is important to know where displaced residents end up moving after they are displaced. A number of studies have shown that displaced residents tend to move into nearby neighbourhoods, reflecting their attachment to their original communities (Bolt and van Kempen, 2010; Liu et al., 2018a; Reades et al., 2023). However, tenants living in urban villages tend to be more vulnerable and thus less likely to find new residences nearby. Under mass displacement, tenants displaced from urban villages may relocate to more disadvantaged neighbourhoods, resulting in worse living conditions and reduced access to facilities (Hypothesis 1). Second, we are interested in how relocating to a new location will affect the livelihood of displaced tenants. Due to the possible relocation of displaced residents to more disadvantageous areas, we anticipate an increase in travel costs for them, both for commuting and non-commuting purposes (Hypothesis 2). Those displaced tenants from urban villages in the central area of the city should experience more severe effects of the above (Liu et al., 2018a). In this study, we define the original Special Economic Zone of Shenzhen (SEZ) as the central area, which includes the four administrative districts of Nanshan, Futian, Luohu and Yantian.

Data and method

Identification of mass displacement

Utilising mobile signalling data that records continuous individual movements, we identified mass displacement due to Shenzhen’s urban renewal projects. There are four key milestones in Shenzhen’s urban renewal projects, namely the launch of the project, the announcement of the draft plan, the announcement of the final plan and the confirmation of the developer. It typically takes 5 years from the launch of a project to the confirmation of the developer, following which the demolition process will begin. During this period, however, the exact date of tenant clearance remains unclear. There are some tenants who were warned about the demolition in advance and began looking for new dwellings several months in advance, while there are others who only received notification from the landlord at the last minute (Liu et al., 2018a). We therefore rely on outmigration patterns in urban renewal areas to identify the timing of mass displacement. We determined residents’ movements using 6 months of cellular signalling data (November 2018, March 2019, November 2019, March 2020, November 2020 and March 2021) from one of the three mobile network operators in China. The cellular signalling data can capture the location and duration of cellphone users’ stay points, allowing us to identify the residence and the workplace of each user based on their longest time of stay during nighttime and daytime, respectively.

The mass displacement identification consists of several steps. First, we detected relocated residents who changed their residence in two consecutive months. Due to data protection rules, we divided the study area into 1000-metre square grids. In this regard, we obtained grid-grid relocation trajectory records, which contain information about the origin and destination of the relocation, as well as the distance travelled and the number of residents that relocated. The relocations within 1 km were removed in order to reduce the influence of the inaccuracy of cellular data location. Then, we matched these relocation records with a dataset of urban renewal projects in Shenzhen. We compiled this dataset from the websites of municipal and district governments, detailing the geographical location of each renewal project as well as its four milestones. Following that, we calculated the number of residents who migrated out of renewal areas during each observation period and assessed whether there was a surge in out-migration from renewal areas during all observation periods. When outmigration residents were considerably higher during a period than its prior and subsequent periods, as well as when compared to neighbouring grids during the same period, we considered them as mass displacements. Consequently, by analysing the temporal and spatial variations of resident movements, we determined the timing of mass displacements. We verified this identification by comparing the mass displacement period with the timeline reported in official announcements and news releases regarding each project.

In total, 22 urban renewal projects, 12 of which are located within the SEZ, have been detected as causing mass displacement. During the mass displacement period, a total of 147,085 residents moved out, compared to 67,591 residents during the prior observation period and 70,876 in the subsequent period. Figure 1(a) illustrates the spatial and temporal distributions of all sampled renewal projects. The identified mass displacement in most urban renewal projects occurs between the announcement of the draft plan and the confirmation of the project by the developer, which is to prepare for the new development. Clearly, the outmigration patterns of displaced residents in the SEZ area differ from those in the non-SEZ area, especially in terms of relocation distance and the number of affected residents. Figure 1b shows an example of mass displacement in Beitou Village, where we found a prominent surge of out-migration during November 2019 to March 2020.

Variables of relocation

We examined a number of variables related to neighbourhood and housing conditions as well as travel behaviours following the relocation of residents. Our calculation of relocation distance was based on the physical distance from the original homes of the residents to their new homes. Specifically, we assessed housing conditions using two variables, namely the ratio of relocated residents who moved into neighbourhoods dominated by informal housing and the unit rent at the destination neighbourhood (Figure 1b). As informal housing does not conform to formal legal and regulatory standards, and often lacks formal property rights, building permits and safety and sanitation regulations, its quality is inferior to that of formal housing (Wu, 2016). Therefore, moving to informal housing would indicate poorer living conditions. The amount of rent paid by a relocated resident can also be an indicator of the quality of the conditions in which the resident lives, since higher rents are often associated with better living conditions. We collected the data on urban village housing and rent level from the website of Shenzhen Housing and Construction Bureau. Moreover, we utilised point of interest (POI) data from Amap (https://www.amap.com) to measure neighbourhood attributes (Figure 1b). Building on previous studies (Yang et al., 2023; Zhang et al., 2022), we calculated the distance from the centre of a resident’s residence grid to the nearest school and hospital, as well as the number of food services, recreational facilities and bus stations within each grid, using these as proxies for accessibility to urban services. The number of firms within the grid serves as a proxy for access to job opportunities.

We also measured the relocation outcome using a set of travel variables derived from mobile signalling data. Short commutes are generally considered to be beneficial for residents, whereas long commutes may limit their leisure time spent on non-commuting trips (Cui et al., 2024). The start and end stay points and duration for each trip enable us to identify whether the trip is a commute or a non-commute. We additionally identified whether the trip is completed by subway by monitoring signals connected exclusively to that system’s towers. After relocation, we calculated changes in the number of trips for each type, as well as changes in travel time and distance. Rather than measuring actual proximity, we calculated changes in the ratio of subway use during each type of trip to determine whether a resident has greater or reduced access to the subway system.

Econometric models

The displacement effect can be more accurately estimated by comparing relocations resulting from urban renewal with those not affected by such process during displacement periods. To achieve this, we follow the literature to apply the DID approach to quantify the effects of mass displacement on relocation outcomes (Niu et al., 2024). We selected the treated (subject to mass displacement) and control groups (not subject to mass displacement) in a similar vein as in Niu et al. (2024). The treated group includes residents who relocated from renewal areas during the identified displacement period, referred to as displaced residents. The control group is composed of movers less likely to be affected by urban renewal; however, they share similar relocation origins with the treated group, namely residents who relocated out of the areas adjacent to urban renewal projects. We constructed the following model to determine whether there are differences between the two groups exclusive to the displacement period:

Y_{i t} = β_{0} + β_{1} M D_{i t} + β_{2} X_{i t} + β_{3} N_{i t} + δ_{i} + θ_{t} + ε_{i t}

(1)

where $Y_{i t}$ represents the outcome variables, including relocation distance, housing conditions, accessibility to facilities and jobs and changes in travel behaviour indicators. $M D_{i t}$ is a dummy variable that equals 1 if residents moved out of the treated grid $i$ during the displacement period $t$ , and 0 otherwise. $β_{1}$ is the key coefficient to be estimated, representing the average effect of mass displacement. The control variables $X_{i t}$ and $N_{i t}$ capture the attributes of relocated residents (including their number and travel behaviour before relocation measured by average travel distance and time) and their original neighbourhoods (including the number of incumbent residents and accessibility characteristics), respectively. $δ_{i}$ represents the time-invariant grid-level fixed effects. $θ_{t}$ is the estimator of time fixed effects to control for general movement trends. $ε_{i t}$ signifies the error term.

We also investigated how displacement outcomes vary between the SEZ and non-SEZ areas by introducing an interaction term as follows:

Y = γ_{0} + γ_{1} M D_{i t} \times S E Z_{i} + γ_{2} M D_{i t} + γ_{3} S E Z_{i} + γ_{4} X_{i t} + γ_{5} N_{i t} + δ_{i} + θ_{t} + ε_{i t}

(2)

where $S E Z_{i}$ equals 1 if the grid is located within the SEZ, and 0 otherwise. $γ_{1}$ reveals the heterogenous effects of mass displacement. Other variables are the same as in equation (1). Columns (1) and (2) in Table 1 present the summary statistics of the outcome variables.

Table 1.

Summary of variables.

Variable	(1)	(2)	(3)	(4)	(5)
Variable	Mean value of the full sample	Standard Deviation	Mean value of the treated group	Mean value of the control group	t sig. of differences between (4) and (3)
Characteristics of relocation
Relocation distance (metre)	5197	1478	4290	5358	3.176***
Ratio of moving into informal housing	0.58	0.19	0.60	0.58	−0.475
Unit rent at destination (CNY)	59.54	10.57	55.44	60.29	3.015***
Characteristics of the living neighbourhoods
Distance to nearest school at destination (metre)	636.54	70.81	667.86	630.98	−2.574**
Distance to nearest hospital at destination (metre)	1367.56	185.52	1447.51	1353.36	−3.211***
Number of nearby food services at destination	148.70	62.98	129.34	152.13	1.788*
Number of nearby sport and recreational facilities at destination	23.59	4.79	21.65	23.93	2.768***
Number of nearby firms at destination	185.50	38.46	173.03	187.72	1.921*
Number of nearby bus stations at destination	6.44	0.51	6.14	6.49	4.308***
Changes in travel behaviour
Changes in subway use ratio for commuting	0.05	0.08	0.01	0.06	2.400**
Changes in average commuting distance	148.10	1982.76	1378.70	−70.48	−4.464***
Changes in average commuting time	−0.30	196.36	93.33	−16.93	−2.812***
Changes in subway use ratio for non-commuting	0.02	0.05	−0.02	0.02	4.361***
Changes in average non-commuting distance	−283.33	2499.18	755.84	−467.92	−3.253***
Changes in average non-commuting time	−54.81	186.54	−26.69	−59.81	−1.072

, ** and *** represent significance at the 10%, 5% and 1% levels, respectively.

Machine learning-based relocation classification

On the basis of the DID analysis of displacement effects on discrete aspects of residents’ living conditions, we proceed to a more comprehensive assessment of displacement caused by renewal. Specifically, we used a SVM model to calculate an aggregate displacement indicator at the project level, which accounts for the trade-offs displaced residents made between neighbourhood conditions and access to employment in their relocation decisions. This displacement index at the project level can provide a better indication of the extent to which a particular project has a negative impact on displaced residents.

Outcomes of displacement were classified based on variables mentioned in the DID analysis, such as changes in location, neighbourhood resources and travel behaviours. The SVM applied under the proposed classification framework provides an effective way to classify activity patterns such as relocation (Allahviranloo and Recker, 2013; Dai et al., 2023). Building on the approach of Dai et al. (2023), we classified relocations into two types: downward relocations, where residents faced both significantly reduced accessibility and increased travel costs – representing the worst displacement outcomes – and all other relocations. There are 18,876 relocation flows in the full sample, including mass displacements uncovered in the first step and moves in the control area. In order to train and test, we selected 9000 relocation flows from the full sample. This data was then split into a training set (7200 flows) and a testing set (1800 flows). In the training phase, the SVM gradually searched for a classification hyperplane that maximises the margin between the two relocation types. We adopted a soft-margin model, which allows some data points to fall within the margin. The formulation of this classification problem is:

\begin{matrix} \underset{w, b, ξ}{m i n} (\frac{1}{2} {‖ w ‖}^{2} + C \sum_{i = 1}^{n} ξ_{i}) \\ s . t . : y_{i} (w^{T} x_{i} + b) \geq 1 - ξ_{i}, ξ_{i} \geq 0 \end{matrix}

(3)

where $x$ is the input vector with multiple features. w and b define the decision hyperplane: $w$ is the normal vector to the hyperplane and $b$ is the intercept. $ξ$ is a slack variable that represents margin violations, while $C$ is the penalty term that balances the margin size and misclassification. The decision function obtained by solving the above problem is:

f (x) = s i g n (w^{T} x + b)

(4)

The sign of the function output denotes the predicted class of relocation flows. Based on this classification, we calculated the percentage of downward relocations among all relocation flows induced by each renewal project, namely the displacement index.

Results

Regression results

We began by looking at the descriptive comparison of relocation characteristics between the treated and control groups (Columns 3–5 in Table 1). Compared to the control group, the treated group is featured with a slightly higher ratio (not statistically significant) of residents moving to informal housing and significantly lower housing rents. Residents in the treated group also had significantly weaker access to various facilities and job opportunities compared to the control group. The control group used subway more frequently regardless of the purpose of the trip, which was less apparent among the treated group. These descriptive statistics are consistent with Hypothesis 1, however the average relocation distance in the treated group is 1.0 kilometres shorter than in the control group. We also found some descriptive evidence for Hypothesis 2. The treated residents had to travel a greater distance after they relocated: their commuting distance and non-commuting distance increased by approximately 1.4 kilometres and 0.7 kilometres, respectively. The control group, on the other hand, experienced a reduction in travel distances following relocation.

The descriptive statistics provide some suggestive evidence; however, we rely on regression analyses to provide more rigorous support. In general, we found significantly different effects of mass displacements in the SEZ and non-SEZ areas. In panel A of Table 2, column (1) shows that mass displacement has a negligible effect on relocation distance on average, but in the case of renewal projects within SEZs, the interaction term in column (2) suggests an exclusive significant and positive impact. That is, residents displaced by urban renewal projects in the central area moved over longer distances, and this difference does not exist in the non-SEZ area. As shown in the next four columns, displaced residents in the non-SEZ area had a greater likelihood of moving into informal housing, but their rents were not different from those of movers unaffected by renewal. Displaced residents within the SEZ area, however, relocated to neighbourhoods with significantly lower rents than movers in the control group. Thus, displaced residents were left with neighbourhoods that had lower rents or were dominated by informal housing.

Table 2.

Regression results on characteristics of the new neighbourhoods and changes in residents’ travel behaviour.

Panel A	Relocated distance		Moving to informal housing		Housing rents at destination
Panel A	(1)	(2)	(3)	(4)	(5)	(6)
MD	0.064 (0.052)	−0.036 (0.065)	0.019 (0.024)	0.060* (0.031)	−0.018 (0.014)	0.020 (0.017)
MD × SEZ		0.184** (0.074)		−0.075** (0.036)		−0.071*** (0.020)
Observations	179	179	167	167	167	167
R²	0.775	0.787	0.152	0.186	0.488	0.543
Panel B	Distance to nearest school at destination		Distance to nearest hospital at destination		Number of nearby food services at destination		Number of sport and recreational facilities at destination		Number of firms at destination		Number of nearby bus stations at destination
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
MD	0.037*** (0.013)	0.041** (0.019)	0.031** (0.013)	−0.014 (0.019)	−0.063*** (0.018)	−0.011 (0.026)	−0.071*** (0.016)	−0.014 (0.023)	−0.057*** (0.020)	0.002 (0.028)	−0.039*** (0.009)	−0.019 (0.013)
MD × SEZ		−0.006 (0.023)		0.074*** (0.022)		−0.085*** (0.031)		−0.093*** (0.028)		−0.099*** (0.034)		−0.033** (0.016)
Observations	179	179	179	179	179	179	179	179	179	179	179	179
R²	0.311	0.312	0.360	0.418	0.969	0.971	0.689	0.717	0.745	0.763	0.422	0.445
Panel C	Change in subway use ratio for commuting		Change in subway use ratio for non-commuting		Change in average commuting distance		Change in average non-commuting distance		Change in average commuting time		Change in average non-commuting time
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
MD	0.014 (0.014)	0.053*** (0.020)	0.012 (0.007)	0.033*** (0.011)	688.238 (469.884)	335.052 (696.141)	22.041 (494.317)	1151.785 (719.312)	72.279* (43.254)	−14.044 (63.247)	56.086 (36.092)	62.733 (53.578)
MD × SEZ		−0.065*** (0.025)		−0.036*** (0.013)		584.908 (848.947)		−1870.958** (877.205)		142.959* (77.130)		−11.008 (65.339)
Observations	179	179	179	179	179	179	179	179	179	179	179	179
R²	0.660	0.680	0.806	0.819	0.335	0.338	0.532	0.550	0.123	0.149	0.616	0.616

All models include control variables, time fixed effects and grid fixed effects. Notably, we used the logarithmic form for non-negative outcome variables. The standard errors are presented in parentheses.

, ** and *** represent significance at the 10%, 5% and 1% levels, respectively.

We found consistent evidence based on accessibility variables in panel B of Table 2. All urban renewal projects led displaced residents during mass displacement to neighbourhoods that were farther from schools, according to the positive and significant coefficients of MD in columns (1) and (2). The remaining columns in panel B of Table 2 show significant coefficients of MD, primarily due to the interaction term. In the new neighbourhoods, residents displaced by urban renewal projects in the SEZ area experienced a reduction in the accessibility of hospitals, food services, recreational facilities, employment opportunities and bus stations. Collectively, mass displacement caused by urban renewal, especially in SEZs, forced residents into more disadvantaged neighbourhoods. These results are in line with Hypothesis 1.

As shown in panel C of Table 2, we found less significant changes in residents’ travel behaviour following displacement. For the average effects, only the coefficient for commuting time is significant at the 10% level (column 9), suggesting longer commutes after displacement. This provides suggestive evidence for Hypothesis 2. Consistent with previous findings, we found that the impacts of mass displacement varied between SEZ and non-SEZ areas. Based on columns (1) to (4) of panel C, displacement did not result in significant changes in the share of subway use for either purpose on average, however this is due to offsetting effects in SEZs and non-SEZs. As indicated by the positive and significant coefficients of MD in columns (2) and (4), residents who previously lived in the non-SEZ area used the subway more frequently after relocation. Thus, although these displaced residents moved to less accessible locations, this was partially mitigated by increased subway use. In contrast, a significant reduction in subway usage was observed for displaced residents from the SEZ area. They also exhibited shorter trip distances for non-work activities that are considered beneficial to their quality of life (Cui et al., 2024).

SVM-based displacement index

Utilising the SVM model, we calculated the displacement index for each sampled renewal project. The prediction accuracy of the model for test set is 83%. Based on the prediction result of the full sample, we then calculated the displacement index at the project level (Table 3). Overall, 25% of displaced residents experienced a decline in overall living conditions, compared to 9% in the control group. Consistent with the DID analysis, mass displacements in SEZ areas led to a higher rate of downward relocations than in non-SEZ areas, reinforcing Hypothesis 2. Moreover, project-level indices reveal significant variation in the share of downward relocations across different projects, even within the same district. Notably, some projects with extreme indices warrant further investigation.

Table 3.

Project-level displacement indices.

District	Project	Displacement period	Downward relocation
District	Project	Displacement period	Project level (Treat)	District level (Treat)	District level (Control)
SEZ
Nanshan	Hengtong	November 2019–March 2020	15.1%	22.0%	5.6%
	Guimiao	November 2018–March 2019	34.3%
	Baishizhou	March 2019–November 2019	27.8%
	Zhuguang	November 2020–March 2021	5.0%
	Beitou	November 2019–March 2020	14.7%
Futian	Shangsha	November 2018–March 2019	10.9%	14.1%	9.3%
	Nanhua	November 2018–March 2019	5.2%
	Huanggang	November 2019–March 2020	34.9%
	Shixia	November 2019–March 2020	9.2%
Luohu	Caiwuwei	March 2019–November 2019	42.0%	49.9%	18.9%
Luohu	Hubei	November 2019–March 2020	67.3%	49.9%	18.9%
Yantian	Shatoujiao	November 2019–March 2020	7.9%	7.9%	5.3%
Non-SEZ
Longhua	Xinwei	November 2019–March 2020	60.7%	41.5%	15.7%
Longhua	Gonghe	November 2019–March 2020	17.7%	41.5%	15.7%
Longgang	Tianan	November 2019–March 2020	28.2%	18.2%	4.5%
	Guanyun	November 2019–March 2020	3.3%
	Ailian	November 2019–March 2020	8.7%
Baoan	Anhua	November 2019–March 2020	15.3%	9.6%	3.2%
	Changying	November 2020–March 2021	1.2%
	Huaide	November 2019–March 2020	10.9%
Guangming	Dasan	November 2019–March 2020	11.4%	10.5%	3.4%
Guangming	Jiuwu	November 2019–March 2020	9.9%	10.5%	3.4%
Total				25.69%	9.38%

A majority of displaced residents living in Luohu District ended up in worse living conditions as a result of urban renewal projects, compared to 19% among control group movers. Luohu’s central location provides many urban services and job opportunities, which may explain why residents who moved out struggled to find comparable neighbourhoods. One of Luohu’s most notable renewal projects is Hubei Village, which gained prominence due to public debate. Scholars and citizens advocated for its preservation as part of the historical Shenzhen Market (O’Donnell, 2019). In response to public pressure, the developer designated a preservation area to prevent wholesale demolition. However, our findings suggest that despite these revisions, significant displacement still occurred. Another project with exceptionally high displacement index is Xinwei, where 60% of displaced residents experienced downward relocation. A possible explanation is the concentration of multiple renewal projects in the area. Although these projects are not implemented simultaneously, their clustering may reduce the local rental housing supply and limit access to facilities, exacerbating the negative effects of mass displacement. Some SEZ projects exhibited relatively low displacement indices. Notably, three in Futian District had downward relocation rates comparable to the control group. This may stem from distinct renewal approaches, as these projects are led by rural collectives or local governments rather than private developers, potentially influencing displacement outcomes (Wong et al., 2024). However, further research is needed to explain why some projects have unusually high or low downward relocation rates.

Discussion and conclusion

This study examined the social effects of urban renewal through the lens of displacement. We introduced the concept of mass displacement to highlight the phenomenon of large-scale, direct displacement caused by urban renewal. Choosing Shenzhen as the study case, where urban villages are a primary target of its renewal programme, we foregrounded the experiences of migrant tenants who were worst affected during this process. Utilising mobile signalling data over multiple years, we examined the spatial and temporal variation in residents’ relocation patterns to identify mass displacement, providing a clear distinction between forced relocations and regular movements. The effects of mass displacement on displaced residents were then estimated by using both a DID statistical approach and a machine learning approach. Using the former, we examined the causal relation between mass displacement and discrete aspects of residents’ living conditions, while using the latter, we assessed the extent to which displacement negatively impacts residents’ overall living conditions.

In general, the results demonstrate significant negative effects of mass displacement on displaced residents. Consistent with previous studies, we found strong evidence that displaced residents relocated to more disadvantaged neighbourhoods, featured by poor accessibility to public facilities, lower rent levels and a higher ratio of informal housing (Dai et al., 2023; Liu et al., 2018a; Niu et al., 2024). There was also suggestive evidence that residents experienced longer commuting distances as a result of displacement. There may be a trade-off between commuting costs and living conditions for displaced residents, which is also reported by Huang et al. (2018). While the proximity of new residences to workplaces was prioritised, for instance, displaced residents had little choice but to live in substandard housing with limited access to urban amenities. Overall, the aggregate displacement indices of the SVM model indicate that 25% of the displaced residents experienced worse living conditions after relocation.

The negative effects of mass displacement are not evenly distributed. We found that displaced residents in the central area had longer commutes, travelled shorter for non-work activities and ended up with worse living conditions. We nevertheless found that displaced residents in the suburbs were less adversely affected by displacement. This spatial heterogeneity of displacement effects may be explained by variations in housing affordability. The continuous demolition of urban villages over the past decade in Shenzhen has significantly reduced the supply of affordable housing in the central area. The Baishizhou Village, for example, provided substantial housing for displaced residents when a nearby village was demolished a decade ago (Liu et al., 2018b). Nevertheless, when Baishizhou’s residents were displaced by a new round of renewal, the decreased availability of low-rent housing nearby forced them to relocate to more peripheral areas. The issue was particularly acute because, during mass displacement, the large-scale outmigration at the same time created a surge in the demand for affordable housing in the surrounding areas (Liu et al., 2018b). In the suburbs, instead, the shortage of affordable housing was less severe, making it easier for displaced residents to find alternative housing. Our SVM-based displacement indices have revealed additional heterogeneity between projects, which requires further investigation.

This study makes two major contributions to the displacement literature. First, the findings of this study add to the ongoing debate regarding the impacts of displacement based on a context of large-scale urban renewal. Unlike previous studies that broadly interpret displacement (Easton et al., 2020; Elliott-Cooper et al., 2020), this research focuses specifically on forced relocations triggered by urban renewal. These relocations disproportionately impact migrant tenants, a group often overlooked in urban renewal planning but particularly vulnerable to the adverse effects of displacement. A number of studies have examined the displacement of homeowners in Chinese cities, but they have largely overlooked the experiences of displaced tenants (Liu et al., 2018a; Wang, 2020). This group is difficult to trace since they are highly mobile, which is a common problem when quantifying displacement caused by the demolition of informal settlements in the Global South (Liu et al., 2018a; Easton et al., 2020). Our detailed analysis of displacement demonstrates that displaced tenants experience substantial adverse impacts, contradicting recent studies that highlight a localised displacement effect (Bolt and van Kempen, 2010; Liu et al., 2018b; Reades et al., 2023) and challenging the growing narrative framing displacement as a relatively benign process (Li and Song, 2009; Huang et al., 2018; Kirk, 2024). Yet, we do not intend to label migrant tenants as definitive victims of urban renewal. The underlying context plays an important role, as many scholars have suggested (Kearns and Mason, 2013; Li et al., 2019; Wang, 2020). In Shenzhen, the extensive urban renewal of informal settlements, combined with limited legal protections for tenants and lack of compensation, has likely intensified the negative impacts of displacement.

Second, it demonstrates the potential of novel data sources to estimate the displacement effects. Trajectory and mobility data collected from mobile devices, with their high sampling rate, broad geographic coverage and precise time and spatial information, have been increasingly used in relocation studies (Cui et al., 2024) but are rarely applied to displacement research (Dai et al., 2023; Niu et al., 2024). By combining mobile data with appropriate quantitative analysis tools, it becomes possible to accurately identify displaced individuals and assess their living conditions before and after displacement, providing valuable insights into the effects of such displacements. However, mobile data is not as detailed sociologically as questionnaires and interviews, which can provide a more comprehensive understanding of the effects of displacement. Grid-based analyses using mobile data are also susceptible to possible inaccuracies and overestimation due to the use of grids rather than individual data because of privacy concerns.

Our findings may not be generalisable to other contexts, but there are policy implications that are worth considering. Contemporary urban renewals claim to improve social inclusion and ensure a better quality of life for all (Nachmany and Hananel, 2023). Yet, according to our findings, migrant tenants are disproportionately affected by urban renewal. The local government in Shenzhen should consider different approaches of urban renewal, avoiding mass displacement due to wholesale demolition. If a demolition-redevelopment approach has to be used, there should be a series of plans to reduce the negative effects of urban renewal. An advanced analysis of the scale of displacement resulting from urban renewal is essential to develop targeted mitigation measures for vulnerable groups, particularly migrant tenants. These measures could include increasing the supply of affordable housing and expanding compensation schemes to better support displaced residents. When forced relocations are inevitable, tenants should be kept well-informed about project progress and the clearance timeline, so they may have enough time to secure alternative accommodations, potentially avoiding large-scale outmigration within a short period.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by funds from the National Science Foundation of China (Grant No. 42401226); Humanities and Social Science Fund of Ministry of Education of China (Grant No. 24YJC840017); Guangdong Province Basic and Applied Basic Research Fund (Grant No.2023A1515110288).

ORCID iDs

Ling Li

Yiru Tan

Jianming Liang

References

Allahviranloo

Recker

(2013) Daily activity pattern recognition by using support vector machines with multiple classes. Transportation Research Part B Methodological 58: 16–43.

Baeten

Westin

Pull

, et al. (2017) Pressure and violence: Housing renovation and displacement in Sweden. Environment and Planning A 49(3): 631–651.

Bolt

van Kempen

(2010) Dispersal patterns of households who are forced to move: Desegregation by demolition: A case study of Dutch cities. Housing Studies 25(2): 159–180.

Cui

Zhao

, et al. (2024) A new model for residential location choice using residential trajectory data. Humanities and Social Sciences Communications 11(1): 1–13.

Dai

Tong

Chu

(2023) Involuntary resettlement outcomes following the regeneration of informal communities: The case of Baishizhou urbanizing village in Shenzhen, China. Population Space and Place 29(2): 1–14.

Doff

Kleinhans

(2011) Residential outcomes of forced relocation: Lifting a corner of the veil on neighbourhood selection. Urban Studies 48(4): 661–680.

Easton

Lees

Hubbard

, et al. (2020) Measuring and mapping displacement: The problem of quantification in the battle against gentrification. Urban Studies 57(2): 286–306.

Elliott-Cooper

Hubbard

Lees

(2020) Moving beyond Marcuse: Gentrification, displacement and the violence of un-homing. Progress in Human Geography 44(3): 492–509.

Goetz

(2002) Forced relocation vs. voluntary mobility: The effects of dispersal programmes on households. Housing Studies 17(1): 107–123.

10.

(2005) Property-led redevelopment in post-reform China: A case study of Xintiandi redevelopment project in Shanghai. Journal of Urban Affairs 27(1): 1–23.

11.

Huang

Dijst

van Weesep

(2018) Rural migrants’ residential mobility: Housing and locational outcomes of forced moves in China. Housing Theory and Society 35(1): 113–136.

12.

Kearns

Mason

(2013) Defining and measuring displacement: Is relocation from restructured neighbourhoods always unwelcome and disruptive? Housing Studies 28(2): 177–204.

13.

Kirk

(2024) Legitimising displacement: Academic discourse, territorial stigmatisation and gentrification. Urban Studies 61(13): 2492–2512.

14.

Kleinhans

Kearns

(2013) Neighbourhood Restructuring and residential relocation: Towards a balanced perspective on relocation processes and outcomes. Housing Studies 28(2): 163–176.

15.

Lees

(2014) The urban injustices of new labour’s ‘new urban renewal’: The case of the Aylesbury Estate in London. Antipode 46(4): 921–947.

16.

S-M

Song

Y-L

(2009) Redevelopment, displacement, housing conditions, and residential satisfaction: A study of Shanghai. Environment and Planning A 41(5): 1090–1108.

17.

Liu

Geertman

van Oort

, et al. (2018a) Making the ‘invisible’ visible: Redevelopment-induced displacement of migrants in Shenzhen, China. International Journal of Urban and Regional Research 42(3): 483–499.

18.

Liu

Lin

, et al. (2018b) Towards inclusive and sustainable transformation in Shenzhen: Urban redevelopment, displacement patterns of migrants and policy implications. Journal of Cleaner Production 173: 24–38.

19.

van Ham

Kleinhans

(2019) Understanding the experiences of relocatees during forced relocation in Chinese urban restructuring. Housing Theory and Society 36(3): 338–358.

20.

Marcuse

(1985) Gentrification, abandonment, and displacement: Connections, causes, and policy responses in New York City. Washington University Journal of Urban and Contemporary Law 28(1): 195–240.

21.

Nachmany

Hananel

(2023) The urban renewal matrix. Land Use Policy 131: 106744.

22.

Niu

, et al. (2024) After forced relocation: Assessing the impact of urban regeneration on residents’ living spaces in Wuhan, China. Housing Policy Debate 1–19.

23.

O’Donnell

(2019) Heart of Shenzhen: The movement to preserve ‘Ancient’ Hubei Village. In: Banerjee

Loukaitou-Sideris

(eds) The New Companion to Urban Design, 1st edn. London: Routledge, pp.480–493.

24.

Reades

Lees

Hubbard

, et al. (2023) Quantifying state-led gentrification in London: Using linked consumer and administrative records to trace displacement from council estates. Environment and Planning A: Economy and Space 55(4): 810–827.

25.

Wang

(2020) Beyond displacement – Exploring the variegated social impacts of urban redevelopment. Urban Geography 41(5): 703–712.

26.

Watt

(2022) Displacement and estate demolition: Multi-scalar place attachment among relocated social housing residents in London. Housing Studies 37(9): 1686–1710.

27.

Wong

Chen

Tang

, et al. (2024) Neoliberal state intervention and the power of community in urban regeneration: An empirical study of three village redevelopment projects in Guangzhou, China. Journal of Planning Education and Research 44(2): 619–631.

28.

(2016) Housing in Chinese urban villages: The dwellers, conditions and tenancy informality. Housing Studies 31(7): 852–870.

29.

(2018) Planning centrality, market instruments: Governing Chinese urban transformation under state entrepreneurialism. Urban Studies 55(7): 1383–1399.

30.

Zhang

Webster

(2013) Informality and the development and demolition of urban villages in the Chinese peri-urban area. Urban Studies 50(10): 1919–1934.

31.

Yang

Qian

(2023) Measuring the accessibility deprivation of concentrated resettlement communities in China: An integrated approach of space syntax and multi-criteria decision analysis. Applied Geography 150: 102839.

32.

Yang

Wang

Yang

(2024) Return or relocation? Evolving neighbourhood attachment of work-unit residents after gentrification-induced displacement in Chengdu, China. Population Space and Place 30(8): e2825.

33.

Zhang

Qiao

Yeh

AG-O

(2022) Disamenity effects of displaced villagers’ resettlement community on housing price in China and implication for socio-spatial segregation. Applied Geography 142: 102681.

34.

Zuk

Bierbaum

Chapple

, et al. (2018) Gentrification, displacement, and the role of public investment. Journal of Planning Literature 33(1): 31–44.