Housing precarity and mental health among shoebox house residents in Hong Kong: latent profile analysis of perceived social support and its associations with loneliness, depression, anxiety and stress

Abstract

Introduction

Perceived social support is vital for mental health, yet person-centred configurations of support remain understudied in housing-precarious populations. This study investigated perceived social support profiles among residents of subdivided units (SDU) in Hong Kong and their associations with sociodemographic characteristics and mental health.

Methods

In a cross-sectional sample of 1194 SDU residents recruited via local non-governmental organisations and project promotion activities, perceived social support was measured using the Chinese version of the Multidimensional Scale of Perceived Social Support (12 items; 7-point Likert scale; a=0.937 for family, a=0.941 for friends and a=0.935 for significant others, such as spouse and romantic partners). Latent profile analysis was conducted to identify distinct support profiles, with profile selection guided by the Akaike information criterion, Bayesian information criterion (BIC), sample-size adjusted BIC, entropy, Lo-Mendell-Rubin test and bootstrap likelihood ratio test. We also examined sociodemographic predictors, including age, sex, education, years living in an SDU, household income and residency status as well as mental health outcomes, including loneliness, depression, anxiety and stress.

Results

Six social support profiles were identified: Strongly supported (32.2%), kin anchored (4.5%), more supported (23.7%), less supported (17.1%), socially strained (11.1%) and minimally supported (11.5%). Age, educational level, years living in an SDU and household income were found to predict social support profiles. Strongly supported residents scored lower on loneliness, depression, anxiety and stress. Notably, those with the highest family support but low friends and significant others’ support exhibited similarly good mental health outcomes as those with high support from all three sources.

Conclusion

The findings highlight the need for systematic assessment of social isolation and targeted interventions strengthening both kin and community support (eg, integrating family support groups in transitional houses) for mental health initiatives aimed at housing-precarious populations.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• High levels of perceived social support are associated with better mental health.

• Previous latent profile analyses have identified distinct profiles of perceived social support among various populations; however, there is a notable lack of research focusing on housing precarious populations including residents of subdivided units (SDUs).

WHAT THIS STUDY ADDS

• Among a sample of 1194 SDU residents in Hong Kong, we found six distinct social support profiles: Strongly supported, kin anchored, more supported, less supported, socially strained and minimally supported.

• Linking the six social support profiles to different levels of loneliness, depression, anxiety and stress offers nuanced insights into the mental health among SDU residents in Hong Kong.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Clinicians should systematically assess the composition of social support in housing-precarious populations when approaching mental health conditions, rather than assuming uniform social isolation.

• Tailored interventions should be given to individuals’ support networks through family-inclusive mental health programmes and community integration initiatives.

Introduction

Housing precarity represents a major public health challenge affecting 1.6 billion people worldwide.¹ Cramped, inappropriate home environments are strongly associated with compromised mental health and well-being; while deprived, isolated individuals are more likely to end up in poor quality housing.2,5 Mechanisms linking housing precarity to mental health problems include chronic stress from the threat of losing shelter, disruptions of social networks, inadequate access to support services and poor living environments that diminish personal safety and comfort.⁶ This vicious cycle of comorbidities to problems renders residents living in inadequate housing particularly prone to social isolation, mental health disorders including depression, anxiety and suicidal ideation, as well as barriers to accessing timely mental health services.^{7 8} Notably, housing precarity operates not only as a strong predictor of compromised mental health, but also as a driver of social network degradation. For instance, the hyperconfined physical environment of subdivided units (SDUs) restricts residents’ capacity to host friends or relatives, cultivating a ‘spatial stigma’ that contributes to external social withdrawal.^{9 10}

In Hong Kong, a highly-developed metropolitan area of 7.53 million people in Asia, income inequality has put more than 215 700 low-income individuals to live in small, overcrowded SDUs,¹¹ a type of precarious housing in the urban sociology and public health literature.12,14 These living spaces are partitioned from larger living areas and often have cramped conditions, with a median floor area of merely 6 m² per person.¹¹ Local evidence documents that residents in confined residences are experiencing elevated levels of stress and a greater risk of mental health issues compared to the general population.^{14 15} Yet, current investigations often limit themselves to built environments (eg, densely populated housing, poor ventilation),^{14 15} lacking research in the psychosocial factors on compromised mental health in this vulnerable population.

Perceived social support is a critical modifiable determinant of mental health, with higher social support linked to fewer mental disorders.^{16 17} Perceived social support (‘social support’) refers to ‘an individual’s perceptions of general support or specific supportive behaviours from people in their social network, which enhances their functioning or may buffer them from adverse outcomes’,¹⁸ which often includes those from their significant others, families and friends. Two prominent models explain the benefits of social support: the stress-buffering model, which suggests that social support reduces the effects of stressful life events on mental health, and the main effect model, which posits that support enhances well-being regardless of adversity.¹⁹ Empirical evidence supports these models, showing that higher levels of social support are linked to better mental health, whereas a lack of social support is associated with heightened risks of mental health issues, such as depression and anxiety.20,22

However, contrasting results have also been documented. For instance, although increased support from friends is typically associated with enhanced mental health, other research has shown that peer support could elevate depressive symptoms in Japanese university students. This counterintuitive relationship is primarily driven by the mechanism of co-rumination, where peers repeatedly dwell on negative emotions and focus excessively on shared problems.²³ Consequently, while intended as a form of social support, co-rumination can inadvertently exacerbate internalising symptoms such as depression and anxiety.²³ In context of residents of SDUs, when individuals facing identical hardships and severe housing precarity rely on one another, their interactions risk transforming into repetitive loops of co-rumination rather than active problem-solving.

Methodologically, the inconsistency in the effect of support from friends may stem from the limitation of variable-centred approaches (eg, correlational analyses), which assume population homogeneity and aggregate scores from the entire sample.²⁴ While these methods offer insights into the associations between constructs and developing general interventions, they do not capture the complexity of people’s social support network, thereby overlooking naturally occurring subgroups within the population. Notably, heterogeneity remains in the quantity and quality of social support people perceive, even when they are from the same community.25,27 This heterogeneity in perception highlights that shared community membership does not guarantee a uniform sense of support, as factors such as gender, culture, network structure and personal history contribute to diverse interpretations of similar social environments. As such, some members may have high support from family and friends, while others may experience high family support but low friend support, or vice versa, shaping different mental health outcomes.²⁸ As Patino-Fernandez et al²⁸ showed, even within the same community, adults report vastly different support perceptions, with subjective experiences of support networks strongly predicting emotional well-being and underscoring the need for personalised mental health interventions.

Person-centred approaches such as latent profile analysis (LPA) therefore emerge in social support investigations to fill this void.^{29 30} LPA identifies distinct subgroups of individuals based on their patterns of social support characteristics, rather than examining relationships between variables across the entire population. Unlike variable-centred analysis that focuses on how social support dimensions relate to outcomes for the average person, LPA recognises that individuals may experience qualitatively different configurations of social support.³¹ For instance, Li et al’s³² research with 23 192 participants has demonstrated that LPA can identify distinct profiles in different sources of social support, including family, friends or small groups, communities, organisations/institutions and society as a whole. They identified five profiles: high social support, predominantly proximal support (ie, high family support but low support from other sources), moderate social support, predominantly remote support (ie, low family support yet high support from other sources) and low social support. Results further indicated that individuals in the high social support profile reported better mental health outcomes. Additionally, those in the moderate and predominantly proximal support profiles had similar mental health outcomes, both significantly better than individuals in the predominantly remote and low social support profiles. This person-centred approach is particularly valuable in understanding social support in socially vulnerable contexts, as it grants a more nuanced understanding of the diverse needs and lived experiences within these populations, enabling targeted interventions that are sensitive to the specific support configurations and challenges individuals face.

Objectives

Using a person-centred approach, our study explored social support profiles and their associations with sociodemographic characteristics and mental health among 1194 socially vulnerable, housing precarious individuals in Hong Kong. The study investigated the following research questions: (1) How many social support profiles can be identified among SDU residents in Hong Kong, and what are their characteristics? (2) How do these social support profiles differ in relation to the sociodemographic characteristics of SDU residents? (3) How do the social support profiles differ in their associations with the mental health of SDU residents? Specifically, we hypothesised that (1) there are different profiles of perceived social support among residents of SDUs in Hong Kong, (2) younger, female, higher household income and a shorter duration of SDU residency predict a higher likelihood of membership into profiles characterised by higher perceived social support, and (3) in line with the Stress-Buffering Model, profiles with higher overall perceived support will report significantly lower levels of loneliness, depression, anxiety and stress. This study could provide a unique perspective in understanding social support and mental health against the backdrop of housing precarity in an urban setting, serving as the necessary cornerstone to find tailored care solutions for communities living in inappropriate housing.

Methods

Participants

This is a cross-sectional investigation of a territory-wide cohort of housing precarious residents in Hong Kong. Participants were recruited through multiple channels, including community outreach, collaborations with local non-governmental organisations (NGOs), social media promotion and word of mouth. Community outreach was conducted across all districts in Hong Kong Island, Kowloon and the New Territories, reaching a population of over 6000 individuals with no age limitation. Eligible participants were Chinese adults capable of providing informed consent who were living in precarious housing conditions at the time of recruitment, including SDUs, transitional housing or experiencing homelessness. Initial screening was conducted at the NGO level and verified on-site to ensure eligibility. All data were collected via interviewer-administered questionnaires to support participants with lower literacy or language needs. The details of participant recruitments are published elsewhere.^{14 33 34} The data used in this study include participants who are (1) Hans Chinese; (2) aged 18 years or older at recruitment; (3) recruited between 1 July 2021 and 12 November 2024; (4) living in SDUs at the time of recruitment; and (5) able to give consent and are literate. Figure 1 illustrates the workflow of this study. On recruitment, participants and their cohabitants were invited to a 1-hour survey to provide information of their demographic backgrounds, housing environments and migration experience, social interaction patterns, mental health conditions and health-related quality of life. Data collection followed an interviewer-assisted self-report format. Although the measures are designed for self-reporting, trained interviewers were present to provide clarification of survey items on request and to assist participants with functional difficulties (eg, decreased visual acuity). This approach ensured data quality and participant comprehension while maintaining privacy and reducing social desirability bias. Ethics approval was obtained at the authors’ institutional review board.

Figure 1. CONSORT diagram of this study.

Measures

Multidimensional scale of perceived social support Chinese version

Multidimensional Scale of Perceived Social Support-Chinese Version (MSPSS-C) is a valid and reliable 12-item self-report questionnaire that measures perceived social support from three sources: family, friends and significant others, with each subscale comprising four items. Subscale scores were calculated by summing the relevant items for each source, with higher scores reflecting higher levels of perceived support. Standard cut-off recommendations classify mean scores below 3 as low, 3–5 as moderate and above 5 as high perceived support. Studies in Hong Kong Chinese populations have consistently supported the psychometric robustness of the MSPSS.^{35 36} The Cronbach α values in the present study were 0.937 for family, 0.941 for friends and 0.935 for significant others.

Depression, anxiety, stress scales

Depression, Anxiety and Stress Scale-21 (DASS-21) is a brief, 21-item self-report instrument designed to assess the severity of depression, anxiety and stress symptoms through three distinct subscales, each comprising seven items.³⁷ Respondents indicate the extent they experienced each symptom over the past week on a 4-point Likert scale (0–3), with higher scores reflecting greater psychological distress. DASS-21 is valid and reliable in Hong Kong Chinese populations.³⁷ The Cronbach α values in the present study were 0.874 for depression, 0.814 for anxiety, and 0.878 for stress.

UCLA 3-item loneliness scale

The UCLA 3-Item Loneliness Scale is a self-report screening instrument designed to assess perceived feelings of loneliness by asking respondents how often they feel (1) they lack companionship, (2) left out and (3) isolated from others, with each item rated on a 3-point Likert scale.³⁸ In Hong Kong Chinese populations, including older adults and adolescents, the Chinese version of the UCLA 3-Item Loneliness Scale has demonstrated strong psychometric properties.³⁹ The Cronbach α value was 0.855 in this study.

Demographic information

Age, sex, employment, education level, income, living district, household member composition, household food insecurity and social security assistance information and self-reported medical histories were collected by questionnaire and triangulated using official documents (eg, tenancy contract, community centre documents). Housing environment information including floor area, availability of natural artificial ventilations, years of residence in the current housings were also collected through interviews and triangulated with home visits.

Data analysis

SDU residents’ descriptive statistics were produced using SPSS V.26 and their social support profiles were identified using LPA with MPlus V.8.3.⁴⁰ Missing data at the variable level accounted for 0.0–0.9% of all included participants’ responses, and were handled using maximum likelihood estimation. The three scale scores for social support from family, friends and significant others were used as indicators for the LPA. To investigate social support profiles, eight LPA solutions were generated. To determine the optimal solution, the following fit indices were consulted to: the Akaike information criterion (AIC), Bayesian information criterion (BIC), sample-size adjusted BIC (SSA-BIC), Lo-Mendell-Rubin likelihood ratio test (LMR), bootstrapped likelihood ratio test (BLRT), average latent class posterior probabilities and entropy.^{41 42} Lower values of AIC, BIC and SSA-BIC indicate better profile solutions.⁴³ The LMR and BLRT compare k profile solutions to k-1 profiles, with significant p-values indicating better fit for the k profiles. Average latent profile posterior probabilities indicate the accuracy of participants’ classification into their most likely profile, with values above 70% considered satisfactory.⁴³ Higher entropy values reflect more accurate classifications. After determining the optimal profile solution, the R3STEP method was performed to examine associations between sociodemographic characteristics and social support profiles. Results are considered significant if the 95% CI of the OR does not include one. Furthermore, we adopted the standard for minimum class size that requires each profile to contain at least 25 participants.⁴⁴ Finally, the associations between identified profiles and distal outcomes were examined using the Bolck, Croon, and Hagenaars (BCH) method, with all mental health outcomes treated as continuous variables. To account for the risk of Type I errors due to multiple comparisons across the identified profiles, Bonferroni corrections were applied.

Patient and public involvement

Patients and the public were not involved in the design, conduct or reporting of this study. Although the data collection was conducted by trained laypeople from the community, no formal patient or public involvement process was undertaken in setting the research agenda or developing the study methodology. Findings will be disseminated through peer-reviewed publication and shared with community partners involved in the study.

Results

Descriptive statistics

Data from 1194 residents living in SDU in Hong Kong were analysed. Among these participants, 912 (74.6%) self-identified as female, and 805 (72.7%) were permanent Hong Kong residents. The average age of the participants was 48.6 years, and they had lived in SDUs for an average of 6.0 years. Additionally, 137 (11.6%) had received higher education, and nearly all (98.6%) had a household income below the median monthly household income of HK$30 000 (HK$1=US$0.13). Table 1 shows the descriptive statistics of interested variables.

Table 1. Sociodemographic and research variables.

Variable	Frequency	Per cent (%)	Mean	SD
Gender
Female	912	76.4
Hong Kong permanent identity
Yes	805	67.4
Education
Elementary school or below	218	18.3
Junior high school	495	41.5
High school	333	27.9
College degree or above	137	11.5
Household income*
HK$0–1999	281	23.5
HK$2000–3999	21	1.8
HK$4000–5999	20	1.7
HK$6000–7999	33	2.8
HK$8000–9999	54	4.5
HK$10 000–14 999	183	15.3
HK$15 000–19 999	188	15.7
HK$20 000–24 999	119	10.0
HK$25 000–29 999	31	2.6
HK$30 000–39 999	12	1.0
HK$40 000–59 999	1.0	0.1
Age			48.26	13.05
Years living in subdivided units			5.96	6.93
Loneliness			2.08	2.36
Family support			19.65	7.66
Friend support			18.26	7.86
Significant others support			18.78	7.83
Depression			6.12	7.54
Anxiety			6.20	6.64
Stress			9.07	8.64

^*Household income was measured in Hong Kong dollars (HK$1=US$0.13).

Social support profiles

Online supplemental appendix table 1 demonstrates the fit statistics of eight LPA solutions. Although the AIC, BIC and SSA-BIC values decreased as the number of profiles increased, the selection of the optimal solution relies on a holistic evaluation of all fit indices employed. The LMR p values for the seven-profile solutions were non-significant (p>0.05). This suggests that this more complex solution did not significantly improve model fit over the parsimonious ones. Among the remaining six solutions, the six-profile solution was deemed optimal. It showed significant LMR and BLRT results, suggesting that it was superior to the five-profile solution. Moreover, it demonstrated satisfactory results for average latent profile posterior probabilities, ranging from 87.8% to 94.5%, suggesting accurate profile classification. Although the smallest group comprised only 4.5% of the total sample, it satisfies a widely accepted standard requiring more than 25 participants per profile.⁴⁴ More importantly, this family-centric support pattern is a well-documented phenomenon in social support literature. This is particularly true within Confucian-influenced cultures where familial ties often serve as the primary source of support.⁴⁵

Profile one comprised one-third participants (32.2%) and was labelled as the strongly supported profile because these individuals reported high levels of support from all three sources. Profile 2, which represented 4.5% of participants, was characterised by the highest family support but the lowest friend support and low significant others support, indicating kin anchored profile. Profile 3 accounted for 23.7% of participants, showing slightly above-average levels of support from all three sources, suggesting a more supported profile. In contrast, Profile 4 consisted of 17.1% of participants who reported slightly below-average levels of support from all sources, indicating a less supported profile. Profile 5 included 11.1% of participants with relatively low levels of support across all three sources and was labelled as the socially strained profile. Finally, profile 6, representing 11.5% of participants, was labelled as the minimally supported profile due to its lowest levels of support from family and significant others, along with the second-lowest friend support. Figure 2 illustrates the standardised estimates (ie, z-scores) from the analysis, while online supplemental appendix table 2 presents the means and SD of the profiles.

Figure 2. Standardised estimates of profiles identified.

Associations between SDU residents' sociodemographic characteristics and social support profiles

Significant associations were found for age, educational level, years living in an SDU and household income (see table 2), while sex and residency status showed no significant correlation with social support profiles (see online supplemental appendix table 3). In terms of age, older SDU residents were more likely to be classified into Profiles 1 (the strongly supported), 2 (the kin anchored) and 4 (the less supported) compared with Profile 5 (the socially strained). Residents attained a higher educational level were more likely to belong to Profile 4 (the less supported) than to Profiles 1 (the strongly supported) and 5 (the socially strained). Additionally, those who had lived in SDUs for a longer duration were more likely to be in Profile 5 (the socially strained) compared with Profile 1 (the strongly supported). Lastly, SDU residents with higher household income were more likely to be in Profile 1 (the strongly supported) compared with Profiles 3 through 6 (the more supported, the less supported, the social strained and the minimally supported), and they were also more likely to be in Profile 3 (the more supported) compared with Profile 6 (the minimally supported).

Table 2. Associations between demographic characteristics and perceived social support profiles.

Variables	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
	Strongly supported versus More supported (Re)		Strongly supported versus Less supported (Re)		Strongly supported versus Socially strained (Re)		Strongly supported versus Minimally supported (Re)
Age	1.024	(0.999 to 1.050)	1.007	(0.985 to 1.030)	1.055	(1.026 to 1.085)	1.023	(0.996 to 1.050)
Sex*	1.893	(0.993 to 3.610)	0.982	(0.471 to 2.047)	1.195	(0.473 to 3.019)	1.776	(0.833 to 3.786)
Years in SDU	0.983	(0.950 to 1.017)	0.979	(0.947 to 1.012)	0.955	(0.924 to 0.987)	0.995	(0.899 to 1.102)
Household Income	1.145	(1.035 to 1.268)	1.256	(1.144 to 1.380)	1.227	(1.103 to 1.364)	1.335	(1.210 to 1.473)
Education	1.124	(0.854 to 1.479)	0.737	(0.561 to 0.968)	1.145	(0.814 to 1.611)	0.940	(0.681 to 1.298)
HKPID†	1.402	(0.816 to 2.410)	1.638	(0.940 to 2.854)	1.377	(0.735 to 2.578)	1.091	(0.543 to 2.195)
Variables	Kin anchored versus Socially strained (Re)		More supported versus Less supported (Re)		More supported versus Minimally supported (Re)		Less supported versus Socially strained (Re)
Age	1.076	(1.019 to 1.136)	0.984	(0.957 to 1.010)	0.999	(0.970 to 1.028)	1.047	(1.016 to 1.079)
Sex*	0.785	(0.202 to 3.049)	0.519	(0.225 to 1.197)	0.938	(0.424 to 2.076)	1.216	(0.397 to 3.724)
Years in SDU	0.959	(0.895 to 1.028)	0.996	(0.963 to 1.030)	1.012	(0.912 to 1.124)	0.976	(0.942 to 1.010)
Household Income	1.119	(0.912 to 1.375)	1.097	(0.980 to 1.228)	1.165	(1.049 to 1.295)	0.976	(0.864 to 1.103)
Education	1.124	(0.603 to 2.095)	0.656	(0.467 to 0.922)	0.837	(0.588 to 1.191)	1.553	(1.032 to 2.339)
HKPID†	1.167	(0.371 to 3.677)	1.168	(0.599 to 2.277)	0.778	(0.369 to 1.640)	0.841	(0.394 to 1.795)

Significant results are indicated in bold.

^*Sex was coded as: 0=male, 1=female.

^†HKPID=Hong Kong permanent identity was coded as: 0=no, 1=yes. Re=reference group.

SDU, subdivided unit.

Associations between SDU residents’ social support profiles and mental health

Comparisons between profiles revealed significant differences for all mental health outcomes (table 3). In terms of loneliness, residents in Profiles 1 (the strongly supported), 2 (the kin anchored) and 3 (the more supported) reported lower scores compared with those in Profiles 4 (the less supported), 5 (the socially strained) and 6 (the minimally supported). Profile 1 residents (the strongly supported) had the lowest depression scores, followed by those in Profiles 3 (the more supported), 2 (the kin anchored), 4 (the less supported) and 5 (the socially strained); with Profile 6 residents (the minimally supported) reporting the highest depression scores. Profile 6 residents (the minimally supported) also reported the highest anxiety and stress scores, while those in Profiles 1 through 5 did not show significant differences among each other in anxiety nor stress risks.

Table 3. Associations between social support profiles and mental health outcomes.

Outcomes	Strongly supported	Kin anchored	More supported	Less supported	Socially strained	Minimally supported	χ 2	Cramer’s V
Loneliness	0.894a (0.089)	1.209a (0.344)	1.983a (0.144)	2.854b (0.187)	3.007b (0.223)	3.890b (0.289)	235.745***	0.45
Depression	2.999a (0.333)	6.346b (1.245)	5.366b (0.430)	7.952b (0.633)	7.958b (0.809)	12.426c (0.966)	144.010***	0.16
Anxiety	3.566a (0.237)	5.231a (0.963)	6.034a (0.421)	7.064a (0.551)	7.675a (0.659)	11.574b (0.851)	143.649***	0.16
Stress	5.635a (0.365)	7.921a (1.192)	8.821a (0.536)	10.802a (0.752)	11.178a (0.872)	15.118b (0.968)	134.818***	0.15

Means with different subscript letters were significantly differed at p<0.005 (Bonferroni adjusted alpha level). Based on the degree of freedom in the model, Cramer’s V=0.06 indicates a small effect size, 0.17 a moderate effect size and 0.29 a large effect size (Cohen, 1988).

***p<0.001.

Discussion

Using an urban cohort of SDU residents in Hong Kong, our research highlights six different social support profiles among housing-precarious populations, with mental health implications that align with previous literature.^{46 47} Notably, some families facing housing precarity could rely heavily on social supports from kins, but not from other sources. Results further indicate that the six profiles correspond differently to SDU residents’ sociodemographic characteristics (ie, age, educational level, years living in an SDU and household income) and mental health outcomes (ie, loneliness, depression, anxiety and stress). This kin-centric pattern reflects a configuration where robust familial ties co-occur alongside deficits in broader community support, a finding echoed across high-cost housing markets. This understanding can inform targeted interventions for those at higher risk of experiencing insufficient social support, ensuring that resources are effectively allocated to those who need them most.

Social support profiles

We found a mix of social support profiles in this cohort, representing a significant advancement in understanding social dynamics within housing-precarious populations. Despite the common perception that housing-precarious populations lack social support,48,53 this study further highlights remarkable heterogeneity, not only in the levels but also in the composition of their support systems. Consistent with previous studies of other populations including adults, single mothers and bullied adolescents,^{32 54 55} our study identified high, moderate and low social support profiles. This suggests that such profiles emerge consistently across different populations. Moreover, these profiles can be considered converging profiles, as different sources of support go hand in hand, such as all high or all low levels of support. This pattern aligns with the ‘rich get richer’ hypothesis, which suggests that those with strong support networks tend to possess effective social skills and abundant resources, allowing them to further expand their networks and access even greater support.⁵⁶ Conversely, individuals with limited support often lack these advantages, resulting in a cycle of restricted social support networks, shedding light on their additional service needs that might occur.

However, the emergence of a unique ‘kin-anchored’ profile in our cohort, characterised by high family support but limited connections with friends and significant others, underscores the particular reliance on familial networks in this group. This pattern distinguishes SDU residents from other populations who might rely on family support equally with other sources^{32 54 55}, or those who centre their support on peers, neighbours or the broader communities when their family functions fail including vulnerable youth⁵⁷ and older adults.⁵⁸ Housing-precarious groups could rely on familial support as a coping strategy in the face of unstable housing. Literature shows that kin networks often function as a primary safety net in the absence of community resources such as emergency shelter, childcare⁵⁹ and financial aid through intragenerational and intergenerational support. In countries with high housing unaffordability including New Zealand⁶⁰ and the UK, young adults facing precarious housing conditions commonly rely on financial assistance from parents or extended family to access private rentals or achieve homeownership, as market solutions fall short. Longitudinal investigations from the USA⁵⁹ also show that parents in homeless shelters or temporary accommodation would rely on extended familial networks, for example, grandparents, to preserve parenting routines and emotional support to the children to help buffer the impacts of housing financial stress. Our findings echo these observations and highlight the important yet exclusive role of familial support in housing precarious populations.

The minimally supported profile, characterised by low perceived support across all three sources, warrants attention in the context of the gender composition of the present sample. Given that males typically report lower levels of perceived social support and are less likely to engage in support-seeking behaviours than females,⁶¹ the predominance of female participants (76.4%) in this study suggests that the prevalence and severity of this profile may be conservatively estimated relative to what would be observed in a gender-balanced sample. Future research may seek to replicate these profiles in samples with greater gender diversity.

Associations between sociodemographic characteristics and social support profiles

Socio-economic factors such as age, education and income predict the social support profiles of housing-precarious populations, whereas biological sex and duration of housing precarity do not appear to play a significant role. Specifically, older residents are more likely to perceive higher levels of social support than younger residents. This may be because older adults focus their time and energy on intimate relationships, which enhances their perception of support.⁶² Younger people, on the other hand, often maintain a broader social support network that includes not only family and friends but also additional sources, such as classmates, colleagues and online friends.^{63 64} Future studies are advised to assess additional sources of support to depict a fuller picture of young people’s social support networks. Residents with higher education levels tend to perceive less social support (profile 4). This suggests that they have some support but not much, likely because they dedicate substantial time to their careers, resulting in limited interactions with family, friends and significant others.⁶⁵ Aligning with prior studies, residents who have lived longer in SDU and those with lower household incomes tend to perceive lower levels of support.^{51 52} However, sex did not relate to the social support profiles of SDU residents, suggesting that social support could be more of a result of gender differences in social roles rather than of biological sex.⁴⁹ Residency status also showed no association with social support profile in our sample. It may indicate that having a permanent identity does not impact SDU residents’ perceived support, at least from family, friends and significant others. Future research could consider additional sources of social support, such as community support, to further explore this relationship.

Associations between social support profiles and mental health outcomes

Among the six social support profiles, SDU residents who reported strongest support from all three sources (profile 1) experienced the best mental health outcomes, while those lacked support from all three sources (profile 6) exhibited the poorest outcomes. Generally, residents with higher levels of social support reported lower levels of loneliness, depression, anxiety and stress compared with those with lower support. Even residents with just one source of support (profile 2) or those with relatively low support (profiles 3–5) were less likely to experience depression and anxiety than those without any support (profile 6). The benefits of social support on mental health are supported by empirical studies^{20 21 62} and align with two key social support models¹⁹: the stress-buffering model, which suggests that social support can mitigate the effects of living in SDU on residents’ mental health, and the main effect model, which posits that social support enhances residents’ mental health regardless of exposure to adverse situations.

Interestingly, SDU residents in the kin anchored profile (profile 2) showed similar outcomes in loneliness, anxiety and stress as those in the strongly supported profile (profile 1). This finding suggests that strong family support can fill gaps in support from friends and significant others. Indeed, prior studies showed that high family support can supplement insufficient support from supervisors for employees,⁶⁶ lack of public support for the elderly⁶⁷ and inadequate instructor support for postgraduate students.⁶⁸ Our findings suggest that this beneficial effect of strong family support also applies to residents living in shoebox houses.

This study has several limitations despite its contribution. First, the cross-sectional design prevents causal inferences. For example, while a lack of support may contribute to mental health issues, mental health problems could also reduce an individual’s perceived social support.^{69 70} Future research could employ a longitudinal design, allowing repeated observations over time to reveal the temporal associations between social support and mental health among SDU residents. Second, our study assessed only three sources of support. Future research could consider including additional sources of support. For instance, younger SDU residents in our sample reported lower levels of support from these sources compared with older individuals. They may rely more on support from classmates, colleagues or online friends.⁶³ Expanding the range of support sources in future studies would provide a more comprehensive understanding of SDU residents’ social support networks. Third, the requirement for participants to be literate may introduce a selection bias, as literacy is closely related to socio-economic status and thus our findings may not fully represent the experiences of illiterate individuals within this population. Fourth, since this study focused on non-Han residents exclusively, the generalisability of our findings to other ethnic groups may be limited. Future research is suggested to incorporate more diverse ethnic cohorts to validate whether these findings extend to the wider population. Finally, the current study did not account for physical health comorbidities or prior mental health history, which are critical factors that shape an individual’s willingness and capacity to reach out for and social support.⁷¹ For instance, individuals with mental health disorders are often more susceptible to experiencing heightened self-stigma and social dysfunction, which can fundamentally alter both their willingness to seek help and their perception of social support. Future studies are advised to integrate comprehensive health histories when examining the perceptive social support among residents of SDUs.

Implications

For clinicians, our findings highlight the need to systematically assess the composition of social support in housing-precarious population rather than presuming uniform social isolation. Tailored interventions, such as family-inclusive mental health programmes and community integration initiatives, can leverage existing support structures and promote mental health resilience. Integrated housing and mental health interventions including Housing First and supported accommodation models have shown efficacy in strengthening social ties among highly isolated individuals.⁷² Incorporating regular social network assessments into clinical practice and addressing barriers to both kin-based and community-based support are essential strategies to mitigate mental health risks among those experiencing housing precarity globally.

For public health policymakers, our findings underscore the vital role of family support in the mental health among housing-precarious population. Rather than relying on individual-centric interventions, authorities could prioritise the implementation of family-inclusive programming. For example, the Multiple Family Group Therapy programme has been documented to facilitate positive attachment among family members and foster mutual-aid networks across SDU households.⁷³ Through such programmes, within- and cross-family support can be leveraged as a critical stress-buffering mechanism for SDU residents.

Data availability statement

Data are available upon reasonable request.