Introduction

Economic policies that are designed to improve the infrastructure of neighborhoods or communities may also negatively affect the health outcomes (mental and physical) of minoritized communities, such as Black individuals [1]. According to Freeman [2] gentrification is the process by which decline and disinvestments in inner-city neighborhoods are reversed. Of relevance to Miami-Dade is also “climate gentrification” where due to the threat of sea level rise (SLR), lenders will re-invest in neighborhoods of higher elevation that primarily consist of people of color and that have been long-neglected [1]. Gentrification also entails the gradual increase in wealth of a working-class neighborhood through the migration of people from higher socioeconomic classes [3, 4]. From an economic standpoint, gentrification results in new proximity to job opportunities, lower commute time, a larger tax base, and better public services [5].

However, there is no effort to invest in the long-term residents of these communities nor to ensure that they can stay or benefit from gentrification. Instead, current residents are displaced and forced to move because they can no longer afford to reside in the gentrifying neighborhoods with a rise in cost of living (e.g. food prices, rent, and property tax) [5, 6]. Further, their relocation may be to marginalized neighborhoods that are underinvested, provides lower quality of life (e.g. schools, commute, transportation, poverty, crime), and may have higher vulnerability to climate risks due to lower elevation [1, 7]. In addition, displacement results in disrupted social ties and networks and the perpetuation of geographically concentrated poverty, accompanied by increased exposure to chronic stress [8]. However, research on the health effects of both displacement and of living in gentrified or gentrifying neighborhoods are sparse in the fields of economics and public health [8]. Further, it seems that gentrification and displacement are conflated; however, gentrification is distinct from displacement [9]. Gentrification is such that investments are reshaping the communities at the census tract level (indirectly affecting the households), while displacement is directly affecting the household when new conditions push longtime residents out. Displacement is an artifact of gentrification and hence we postulate that gentrification is associated with displacement, and our theoretical and empirical analyses attempt to make the distinctions between the two concepts and its impact on the health outcomes of Black Women Living with HIV (BWLWH).

In Miami, Florida (FL), HIV is a major public health issue, with Florida ranking as number one for new HIV diagnoses in the U.S., and similar to gentrification, HIV disproportionately affects Black communities. For instance, 1 in 61 Black women are living with HIV compared to 1 in 1007 white women or 1 in 490 Latino women, 31% are not retained in care (e.g. attendance of needed health appointments), and 40% are not virally suppressed (viral suppression requires high medication adherence and prevents HIV transmission to others) [10]. These health disparities are likely driven by pervasive re-occurring stressors/adversities faced by BWLWH including racism, sexism, poverty, and unstable housing [11,12,13]. For instance, researchers have found among people living with HIV (PLWH) that housing instability (either lack of affordable or secure housing) serves as a barrier to engagement in care, adherence to antiretroviral medications, and to sustaining viral suppression [14,15,16]. Similarly, displacement has led to ruptures in ART adherence [17]. However, only a small body of literature has given attention to economic factors while employing novel econometric approaches. Wright and colleagues [18] found significant cross-sectional relationships, that is employment, education, crime, income, number of religious organizations, and low-income housing were significantly related to stigma, discrimination, mental health, HIV viral load, and medication adherence within neighborhood and in neighboring areas among BWLWH. This suggests that gentrification may be an important issue to examine particularly for BWLWH, who bear the brunt of the HIV epidemic among women. Gentrification may impose hardship in several dimensions such as elevated stress and depression, transportation barriers, and longer commute times to health facilities and pharmacies [19,20,21]. However, there has been no previous research on the impacts of displacement (associated with gentrification) on the mental and physical health outcomes of BWLWH. This paper will examine empirically how estimated displacement risk (related to gentrification) relates to mental and physical health outcomes among BWLWH in South Florida (focusing on Miami-Dade and Broward counties).

While the small literature giving attention to gentrification and displacement in the HIV field is growing, quantitative investigations are few and literature examining associations with physical and mental health outcomes among PLWH are even fewer. A recent publication used simulation (via a spatially dynamic agent-based model) to assess the impact of gentrification on HIV transmission among Black women and found minimal impact which they concluded may be due to displacement of the people most affected by HIV [22]. Another article, examining overlap in clusters of sexually transmitted infections (STI), gentrification, and rental cost in Atlanta, Georgia, found that among Black people (vs. White people) the overlap of STIs and gentrification was greater [23]. In a majority male sample of people with HIV in New Orleans, researchers found that among those living in gentrifying neighborhoods, people who were in redlined neighborhoods (compared to non-redlined neighborhoods) had a longer time to viral suppression [24]. Some researchers examining specific displacement events (e.g., evictions, forced relocations) among PLWH have highlighted the importance. For instance, Kennedy and colleagues found that among PLWH who use drugs in Canada, evictions increased the odds of having a detectable viral among both individuals who were unhoused and housed [25]. Similarly, among primarily indigenous women with HIV in Canada, scholars found that evictions were associated with depression and substance use [26]. However, we were unable to locate any literature that has examined the association between gentrification or displacement with mental health and HIV outcomes specifically among Black women living with HIV.

Due to intersectional stressors such as racism, HIV stigma, discrimination, and poverty faced by people living with HIV prior to diagnoses and after, they experience higher point and lifetime prevalence of mental health diagnoses (e.g., depression, posttraumatic stress disorder) compared to those not living with HIV [11, 12, 27,28,29]. The estimated risk of displacement may further exacerbate BWLWH’s mental health. Researchers have found that individuals living in a gentrifying neighborhood who were low-income, renters, and long-term residents had higher psychological distress compared to homeowners, individuals new to the community, or higher income residents [21]. The influx of white individuals may play a role in higher rates of distress in residents of color due to racist behaviors and forcing change to old norms (e.g., cookouts, block parties) for community gatherings, social support, and stress reduction [30]. Furthermore, longtime residents of gentrifying neighborhoods may experience a decline in social support and ties as friends, family, and neighbors are displaced, which may have negative effects on their mental health [19].

To date there has not been any study examining the relationship between displacement (associated with gentrification) and mental and physical health outcomes of Black women living with HIV. Given that Black women are both disproportionately impacted by HIV and gentrification our findings will have implications for economic and public health policies. Fig. 1 presents the hypothesized pathways linking neighborhood change, displacement risk, and health outcomes among Black women living with HIV. Using non-linear regression models for a sample of 151 BWLWH while accounting for spatial dependence, this paper examines how the risk of displacement may influence the health outcomes for BWLWH in South Florida.

Fig. 1
Fig. 1
Full size image

Hypothesized Pathways Linking Neighborhood Change, Displacement Risk, and Health Outcomes among Black Women Living with HIV

Note: The first two stages typicarelly serve as precursors for gentrification as documented in existing literatu

The remaining sections of this paper are organized as follows: Sect.  2 defines gentrification and displacement, Sect.  3 describes the methodology and data, Sect.  4 discusses and presents the empirical findings, and Sect.  5 concludes.

Gentrification and Displacement Defined

The Encyclopedia of Housing [31] defines gentrification as “the process by which central urban neighborhoods that have undergone disinvestments and economic decline experience a reversal, reinvestment, and the in-migration of a relatively well-off, middle- and upper middle-class population.” This generally leads to rising home values and rents [32]. According to scholars [5] gentrification has been viewed by some as a solution to many of the problems facing older central cities. However, there is concern about the potential role of displacing disadvantaged residents.

According to Johnson et al. [33] while most researchers agree that gentrification leads to an influx of residents of higher socio-economic status, they disagree about the degree to which it displaces long-term, low-income residents. Accordingly, these disagreements stem from the difficulty of measuring displacement. In some cases, gentrification can directly lead to displacement caused by rising rents, formal evictions, or eminent domain. In others, displacement can indirectly result from gentrification, occurring through deliberate landlord neglect, the closure of small businesses that cater to low-income populations, or cultural shifts [9]. Displacement is forced or involuntary household movement from places of residence [9]. According to Butler et al. [34], residential displacement can happen before gentrification when there is disinvestment in neighborhoods, which creates vacant land, position a neighborhood for eventual reinvestment, and creates conditions that are viable for gentrification. Also, this type of displacement can result from additional disruptions such as natural disasters, rent increases, and highway or transit construction and expansion [34, 35]. For these reasons, gentrification scholars typically agree that while gentrification may cause displacement, the two phenomena should be considered as distinct from one another. In this paper, we will focus on how the drivers of gentrification may increase the estimated risk of displacement and the associations between displacement risk and health outcomes for BWLWH.Footnote 1

Economic Impact of Gentrification

Gentrification improves communities by creating job opportunities, a larger tax base, and better public services while lowering crime rates [5]. However, while gentrification may increase employment opportunities, discriminatory practices by local businesses cause a reduction in employment opportunities for minoritized individuals [36, 37]. Note that simultaneously, the attractiveness of these neighborhoods increases the cost of living (landlords increase the rents and the prices of food and other basic amenities rise), which impose significant hardship on the residents and sometimes force relocation [37]. However, according to Freeman [2] and Ellen & O’Regan [38] there is conflicting literature suggesting that disadvantaged households do not necessarily leave more quickly and that gentrified neighborhoods do not have relatively high exit rates, implying that there are both negative and positive benefits of remaining in gentrified neighborhoods for disadvantaged households. Note though that economic policies that ignore health implications may have long-term harmful implications that outweigh the short-term benefits.

Empirical Measures of Gentrification

According to McKinnish et al., [6] gentrification is present if the bottom quintile of neighborhoods has median income growth substantially above that experienced by neighborhoods in the four richer quintiles, and the 90th percentile of income growth is only higher in the top quintile. Martin & Beck [39] uses two other measures (i.e. low and high threshold) to identify gentrifying tracts. A low threshold is if the neighborhood experiences income growth that is 25% higher than the metropolitan area’s income growth and a high threshold if the neighborhood’s income growth exceeds the metropolitan area’s income growth by at least 50%. The impact of gentrification is non-linear and very localized and hence these metrics vary substantially across the various neighborhoods and may not reflect accurately what is reshaping the respective neighborhoods.

Over the past decade the number of Airbnb investments across various communities has been growing and are competing with the hotels and resorts since they represent home-like set-up with laundry, more space, home-cooked meals, safety etc. The rise of short-term rentals for Airbnb has been found to be a key factor in addressing the concept of tourism gentrification or touristification within urban, coastal, and rural spaces. Through the influx of tourists into neighborhoods, characteristics such as the market price of housing and availability of long-terms rentals drastically shift the nature of neighborhoods to accommodate this new market of individuals. Additionally, these shifts create a deficit for locals who are looking for long-term and affordable housing, potentially pushing them out from their communities to find affordable housing. Cocola-Grant [40] states that the attraction of visitors accelerates the pressure of gentrification - that is the increase in utilization of existing land that are already at full capacity drives up the value of commercial and residential properties. Thereby leading to place-based displacement and a loss of experience by residents due to higher costs from tourism, which displace them from places they belong. The implication is that tourism and gentrification are not mutually exclusive and have consequential impacts on residential displacement. Research completed in Toronto supports these claims in which the use of short-term rentals through Airbnb created a significant shift in long-term rental availability to local renters [41]. According to Amore et al. [42], in three different cities, Athens, Lisbon, and Milan there were high rates of renter displacement due to tourism gentrification.

According to Rabiei-Dastjerdi, McArdle, and William [43], gentrification has temporal and spatial dimensions, which can be traced in time and space at the neighborhood level. The rental market (both short and long-term) is experiencing a significant change due to Airbnb activities which is a key driver of gentrification, housing shortage, and high rental cost. Hoffman and Heisler [44] states that Airbnb and short-term rentals have contributed to rising housing costs, displacement, and gentrification. In fact, studies [43, 45] have found positive correlations between the density of Airbnb listings and increases in rent and house prices, and makes the link between short-term rentals, displacement, and gentrification.

Marcuse [46] explained that when a household vacates a housing unit voluntarily and that unit is then gentrified such that a similar household is prevented from moving in, the number of units available to the second household in that housing market is reduced. The second household, therefore, is excluded from living where it would otherwise have lived. This is called “exclusionary displacement” which is different from “direct displacement”. According to Wachsmuth [47] and Weisler both forms of displacement are taking place due to Airbnb activities. Wachsmuth and Weisler [47] state that Airbnb facilitates short-term rental housing that systematically drives gentrification and displacement. The authors used data for 188,000 Airbnb listings which were active in the New York metropolitan region between September 2014 and August 2017 and the conclusion was that gentrification induced displacement due to increase growth in Airbnb activities.

The presence of Airbnb properties in general is indicative of higher local economic activities in restaurants, bars, museums, and other venues which are significant measures of gentrification and often lead to higher costs of local goods and services incurred by existing residents. In our analysis we will use the changes in Airbnb fees for the respective census tracts as a proxy to measure how the characteristics of the neighborhoods are changing.

Data and Methodology

Participants and Procedures

Participant data was collected from the NIMH-funded parent project, Monitoring Microaggressions and Adversities to Generate Interventions for Change among Black women living with HIV (MMAGIC]), which began enrolling women (see Table 1 on sociodemographic characteristics) in the Fall of 2019 and assess mental and physical health outcomes every three months (see Table 2 for selected outcomes applicable to this project). This study uses the baseline data from the parent study consisting of 151 BWLWH. Recruitment of participants included posting flyers in health clinics and centers, hospitals, and at community-based events. Individuals who were interested completed a phone screening to see if they were potentially eligible for participation. In order to be included women had to meet the following criteria: (1) Living with HIV, (2) Identify as Black and/or African American, (3) Age 18 or older, (4) Cisgender woman, (5) English speaking, (6) Capable of completing and fully understanding the informed consent process and study procedures and (7) Own a cell phone with text and internet capacity. Women were excluded if any of the following two exclusion criteria applied: (1) Unable to completely and fully understand the informed consent process and the study procedures and (2) Significantly interfering mental health symptoms such as active psychosis. Participants completed informed consent, blood draws, questionnaires, and a semi-structured clinical mental health interview. Upon finishing the study visit individuals received a $75 stipend. All procedures were approved by the Institutional Review Board of the University of Miami.

Table 1 Sociodemographics and Characteristics among 151 Black Women Living with HIV
Full size table
Table 2 Descriptive Statistics of Trauma, Mental Health, and HIV Behaviors and Outcomes
Full size table

Data

Predictors

Neighborhood Variables from Census Block and Zip Code Data: We collected data from Airbnb, American Community Survey (ACS), and the Census, that serve as proxy changes in the characteristics of the various communities. We computed the percentage change for the period 2015 to 2019 for the Airbnb data and 2014 to 2019 for the ACS and Census data for variables that are used in the analysis to measure the impact of gentrification.

ACS and Census. Census tract level data was collected for Miami-Dade and Broward counties on the following: Sex by age which consists of White alone (Whites), not Hispanic or Latino; Sex by age which consists of Black or African American alone (Blacks); Total bachelor’s degree (Degrees), place of birth by educational attainment in the United States; Owner occupied housing (OOH), total householder (residence one year ago by tenure in the United States) lived in owner-occupied housing units; and Median gross rent (Rent) which is the median gross rent by year for the householder that moved into the unit.

Airbnb. We calculated the median Airbnb fees for apartments (APT), condominiums (CONDO) and houses. These are used for proxies to measure significant changes in the characteristics of respective census tracts.

Demographics Variables: To examine how individual level variables affect health outcomes we also include age, income and self-report crime.

Crime Victimization. The National Crime Victimization Survey was used to collect information from participants on neighborhood crime in the past 12 months [48]. The survey collects information on 7 types of crime: assault, burglary, robbery, identity theft, motor vehicle theft, rape, and sexual assault. If individuals indicated that they had experienced a crime, they were also asked about the frequency of the occurrences.

Address/residency: At each visit women are asked for the complete address of where they reside. To capture the neighboring effects, we use the addresses to geocode the longitude and latitude values to compute an inverse distance measure that is used to capture geographical dependence.Footnote 2

Outcomes

Our main outcomes of interest are mental health, health behavior (i.e., HIV medication adherence), and physical health (i.e., HIV viral load) described below. These were obtained via self-report measures given through REDCap, clinician administered semi-structured interviews, and blood specimen collections.

Mental and Physical Health Outcome Measures

Center for Epidemiologic Studies Depression Scale [49]. A self-report measure (20 items) of depressive symptoms with good reliability/validity.

The Life Events Checklist for DSM-5 (LEC-5) [50, 51]. A 17-item self-report measure capturing lifetime exposure to traumatic events (e.g., physical assault or abuse, sexual assault or abuse, serious accidents, life threatening illness, tragic death, war, and natural disasters). It has demonstrated good validity in the general literature and among BWLWH.

PTSD Checklist for DSM-5 (PCL-5) [52]. A 20-item measure of post-traumatic stress disorder symptoms (PTSD) with excellent reliability and validity in prior literature [52].

Post Traumatic Cognitions Inventory [53]. A measure of trauma-related thoughts with excellent internal consistency (α = 0.86-0.97).

Physical Health and Health Behaviors

HIV Medication Adherence. We included a 3-item self-report medication adherence measure that was previously validated against electronic drug monitoring in an observational study [54]. This self-reported HIV adherence measure included one question asking participants to rate their adherence from 1 being “very poor’ to 6 being “excellent” for the past week 4 weeks.

HIV Viral Load. Blood specimen was collected to run assay for HIV viral load based on the Roche COBAS AmpliPrep/COBAS TaqMan HIV-1 Test, v2.0. Below 20 copies per millimeter was defined as undetectable and having less than 200 copies per millimeter was defined as viral suppression. In analyses we used viral load count, log transformation of viral load, and dichotomized variables for detectable viral load (1=undetectable. , 0=detectable) and viral suppression (1=suppressed and 0=unsuppressed) [55].

The Model

If one naively regresses the changes in the socioeconomic factors on the respective health outcomes and the results show that an increase in Airbnb fees elevate the stress levels of these individuals, then one would be inclined to ask why? We postulate that a primary driver that explains variation in the respective health outcomes is the likelihood of being displaced. The mechanism in which the change in the characteristics of the communities affect health outcomes is via the risk of households being displaced which is primarily due to the result of gentrified or gentrifying neighborhoods. We model two structural Eq. (1) health outcomes which are explained by individual’s traits (e.g., age) and (2) the risk of displacement which is explained by gentrification. The risk of displacement is a binary variable that takes the value one (risk of being displaced) and zero (not at the risk of being displaced) and is a right-hand side endogenous variable (it is influenced by several demographic and socioeconomic variables) in the main equation of interest. We use the percentage change for the respective socioeconomic and demographic factors as a proxy for gentrification. We then analyze how gentrification (via these proxy factors) influences the likelihood of being displaced (“the estimated risk of displacement”) and how this in turn affects individual’s health outcomes. Additionally, we modified the structural model to include neighborhood variables that are geographically correlated. We use spatial econometrics to accommodate these modifications. Spatial estimation is a technique that is used to model geographical dependencies, hence capturing cross-sectional correlation in this context. This technique is an appropriate fit to do the various analyses since neighboring factors such as percentage changes in Airbnb fees in neighboring communities are important in explaining variations in health outcomes. For a detailed discussion on spatial econometrics, see [56,57,58,59,60].

The general discrete choice model for displacement with spatial correlation in the observable is,

$$Di\:=\:\alpha\:+{\sum\:}_{k=1}^{K}{X}_{ik}{\beta\:}_{k}+{\sum\:}_{k=1}^{K}{\sum\:}_{j=1}^{n}{w}_{ij}{X}_{jK}{\theta\:}_{k}+{v}_{1i,}$$
(1)

for i = 1, …, n.

Note that Di = 1 if C ≤ \(\:\stackrel{-}{C}\) and 0 otherwise. \(\:\stackrel{-}{C}\) is a fixed threshold value or cutoff point. These thresholds are motivated by Eggers and Moumen [61] and Eggers, Moumen and Weicher [62] who delineated ratio of household income to median income in the area in which an household lives to taxonomize, “extremely low income less than 30%”, “low income”,….to “extremely high income, that is household income is higher than 120% of local area median income”.

The health outcome for individual i is given as,

$$H_i=\;\gamma0+\gamma1\;D_i+\gamma2Zi\;+v_{2i,}\;for\;i=1,\;\dots,\;n$$
(2)

Note that \(\:{H}_{i}\) may be a continuous or binary health outcome. For the binary outcome Hi*, it is assumed to stem from an unobservable latent variable in which there is a critical level for some health outcome index, Hi** such that if Hi* does not exceed Hi** then Hi = 1 and zero otherwise. For example, the viral load for individual i becomes undetectable below a certain threshold and is modeled as a discrete outcome. In this case, the probability (Hi = 1|x, z) =probability (Hi*≤ Hi**) has a normal cumulative distribution.

It is assumed that:

$$\left(v_1|x, z \quad v_2|x, z\right) \sim N \left[ \left(0 \quad 0\right), \left(\sigma_{v1}^2, \quad \rho \sigma_{v1} \sigma_{v2} \quad \rho \sigma_{v1} \sigma_{v2}, \quad \sigma_{v2}^2\right) \right]$$
(3)

Note, also that the probability that \(\theta^{*} = [{\theta}\quad {\beta}\quad {\gamma}]^{\prime} (\text{D=1}|\text{x, z}) = \varvec{\Phi} (\varvec{x}^{*}\varvec{,\theta}^{*})\) where \(\:{\Phi\:}\) is the normal cumulative distribution function where, \(\:x\)* \(\:=[\text{x},z,\:\)wx], \(\:\theta\:\)*= [ \(\:\theta\:\)\(\:\beta\:\:\gamma\:]{\prime\:}\) and \(\:{\sigma\:}_{v1}^{2}\) may normalize to one (1).

The variable Di (risk of displacement) is a dummy variable that represents 1 if the individual or household is at the risk of being displaced and zero otherwise. The cut-off points are based on the ratio of the individuals’ household incomes to the median income for the respective census tract. For example, if the ratio is less or equal to 0.3 then these households will be at risk of being displaced hence Di = 1. We analyze several thresholds in which the households are at the risk of being displaced, these are 0.3, 0.4, 0.5, 0.6 and 0.7 for the respective census tracts. Participant frequency at each threshold is presented in Table 3.

Table 3 Frequency of Participants at each Displacement Risk Threshold
Full size table

The estimate of gamma (\(\:\gamma\:\)1) will measure how displacement affects individual health outcomes. The Xs are assumed to be exogenous variables while \(\:{\sum\:}_{k=1}^{K}{\sum\:}_{j=1}^{n}{w}_{ij}{X}_{jK}\) is a weighted average of neighboring variables that is used to explain variations in health outcomes that would otherwise not be included in the model. Note that the parameter \(\:\theta\:\) measures the strength of neighboring variables that are spatially correlated – it measures the spatial spillovers of the neighboring communities and their impacts. If these variables are omitted from the model, then we will have the classical “omitted variable bias” problem that will typically lead to biased and inconsistent estimates. The wij is an element of the weighted matrix W which is non-negative. These pre-specified weights are used to capture interdependencies in the covariates. The weights are calculated using an inverse distance measure in which closer neighbors have larger impacts but dissipate as the distance increases. The weight (wij) connects or links neighboring community i to j for i ≠ j. According to Kelejian & Prucha [58, 59] and Elhorst [57] we need these assumptions for the weighted matrix to guarantee that the estimates are consistent. (1) All the diagonal elements of the weighted matrix equal to zeros. This is to ensure that there is no interaction or dependence between unit i and itself. (2) The row and column sums of the matrix “W” before normalization should be uniformly bounded in absolute value as n goes to infinity that is the correlation between two spatial units should converge to zero as the distance separating them increases to infinity. In our empirical analyses we use a non-normalized matrix, see [63].

The variable Di enters Eq. 2 (our main equation of interest) as a right-hand side endogenous variable. The classical two stage least squares or instrumental variables are not applicable since a linear projection of the square is not the square of the linear projection. That is, we cannot replace the nonlinear function of an endogenous explanatory variable with the same nonlinear function of the fitted values from the first stage regression. In this case, the method of two stage least squares (2SLS) will produce inconsistent estimates [64]. The variable Zi is exogenous and represents individual or household traits such as age that explain variations in health outcomes. Note that \(\:{v}_{1i}\) and \(\:{v}_{2i}\) are two random error terms that capture random noise or idiosyncratic shocks. We allow non-zero correlation (\(\:\rho\:)\) between the two error terms and assume that they are jointly normally distributed with mean zeros and the variances \(\:{\sigma\:}_{v1}^{2}\) and \(\:{\sigma\:}_{v2}^{2}\), respectively. The changes in the socioeconomics variables for the respective neighborhoods are not uniform and do not affect households the same, that is the rate at which households are displaced varies as such we will use a non-linear model to examine the likelihood of being displaced and its impact on health outcomes. The risk of displacement and health outcomes are jointly determined and therefore all the models are estimated using the method of maximum likelihood while accounting for robust standard errors [65,66,67].

Results and Discussion

Presented and discussed below are findings from (a) analyzing associations between outcomes (e.g., mental health, HIV medication adherence, viral load) and covariates of age and crime, (b) assessing associations between displacement and the various outcomes, and (c) examining how neighborhood variables collected from the census, ACS and Airbnb databases relate to the likelihood of being displaced for the outcomes. For (a) and (b) exact coefficients, effect sizes, t statistics, p values, and confidence intervals are depicted in Tables 4, 5 and 6, and 7. While for (c) statistics can be found in a supplemental online file (Appendix A Tables A1, A2, and A3).

Table 4 Displacement, Depressive Symptoms, Traumas, and Posttraumatic Stress Disorder Symptoms
Full size table
Table 5 Displacement, Viral Load Linear, and HIV Medication Adherence
Full size table
Table 6 Displacement, Undetectable Viral Load, and Viral Suppression
Full size table
Table 7 Effect Size/ Marginal Effects for Probit Models Examining Associations between Displacement Risk, Undetectable Viral Load, and Viral Suppression
Full size table

Associations between Depressive Symptoms, Covariates (Age and Crime), and Displacement Risk

The Center for Epidemiological Studies Depression Scale’s (CESD) scores are negatively associated with age across the various range for the estimated displacement risk at (-0.2724, -0.3015, -0.2847, -0.2171, -0.2919; p < 0.01, p < 0.01, p < 0.05, p < 0.05, p < 0.01, respectively) suggesting that as women become older they may experience less depressive symptoms and perhaps are better equipped to deal with life stressors that may impact their moods, see Table 4. An additional year (age) is associated with a reduction in depressive symptoms by 0.2724 at the threshold level 0.3 holding all the other variables constant. However, exposure to crime (0.0273, 0.0227, 0.0188, 0.0205, 0.0240; p < 0.01, p < 0.05, p < 0.1, p < 0.05, p < 0.05), respectively, is significantly associated with increases in depressive symptoms across all the thresholds of estimated displacement risk (at threshold level 0.6, significant at the 10%) which is consistent with priori expectation [19,20,21]. The estimated risk of displacement at the threshold of 0.3 (household income is 30% of the median income in that specific census track) is associated with higher depressive symptoms (15.95, p < 0.01). Specifically, Black women who are at the estimated risk of being displaced at the threshold level 0.3 may experience 15.95 higher CESD score when compared to women who are not at risk of displacement. With a CESD score of >/=16 (score ranges from 0 to 60) indicating probable clinical depression [49], the risk of displacement could have significant implications for women’s mental health. This significant association is also true for threshold levels 0.4 and 0.6 (22.15, p < 0.01; 27.01, p < 0.01), respectively. This may suggest that individuals who are living in communities that are gentrifying or gentrified may anticipate a higher likelihood of “force move” which may elevate depressive symptoms.

Factors Associated with the Risk of Displacement for Depressive Symptoms

For the threshold point 0.3, we note that as the percentage of White individuals/households (Whites; 0.0003, p < 0.05), OOH (0.0128, p < 0.01) and Airbnb fees for APT ( 0.00001, p < 0.01) increases, they are associated with increases in the likelihood of being displaced for individuals experiencing depressive symptoms. Higher fees are indicative that the cost of living in those neighborhoods is going up and hence households’ stress levels may be elevated. The spatial distribution of the neighboring variables also influences the likelihood of displacement. The change in the percentage of degrees (W_Degrees; 0.00002, p < 0.01) in neighboring communities (census tracts) is positively associated with the likelihood of displacement. As the neighboring communities attract more residents with higher level of education it may have negative spillovers – it is associated with increases in the likelihood of displacement. Note though, as the percentage change in educational degrees within the respective communities increases it is associated with a reduction in the likelihood of displacement (Degrees; -0.0006, p < 0.01). Having a higher education may serve as a buffer against displacement, see Table A1 in the online appendix. The risk of displacement is positively associated with Black residents (0.0009, p < 0.1) in neighboring communities. However, as the number of people at the risk of displacement increases the sign switches to negative, that is Black residents in neighboring communities (-0.0008, p < 0.1; -0.0005, p < 0.05), respectively, for cut-off points 0.4 and 0.6 decreases the risk of displacement. The results for whites, OOH, and APT fees are consistent for thresholds 0.4, that is they are positively associated with the likelihood of displacement (0.0001, 0.0077 and 0.000001; p < 0.01, p < 0.05 and p < 0.01), respectively. For cut-off points 0.4, OOH (-0.0004; p < 0.01;) in neighboring communities is negatively associated with the estimated risk of displacement while the Airbnb fees for houses in neighboring communities is positively related to the risk of displacement at the 10% level, see Table A1 in the appendix.

Associations between Traumatic Life Events, Covariates (Age and Crime), and Displacement Risk

The results for life event checklist (LEC) which captures traumatic life events are similar to CESD for the cut-off points at 0.3, 0.4, and 0.7 (see Table 4) that is higher crime is related to higher reported traumas (0.0067, 0.0085, 0.0081; p < 0.1, p < 0.05, p < 0.1), respectively. While age (-0.0425, -0.0402; p < 0.05, p < 0.1) is negatively related to trauma at the threshold levels 0.4 and 0.5, respectively. The average household who is likely to be displaced experience more traumatic events at the thresholds of 0.4, 0.5, 0.6 and 0.7 (6.5194, 5.7483, 5.7779 and 4.5386; p < 0.01, p < 0.01, p < 0.01, p < 0.05), respectively. For instance, at the 0.4 threshold, women who are at risk of being displaced have 6.5194 higher traumatic life events compared to women not at displacement risk. Each point on the LEC represents a different traumatic life event (e.g., rape, physical assault) [51] as such at the 0.4 threshold displacement risk is associated with six to seven more traumas. Note that as the proportion of people that are likely to be displaced increases, it corresponds with higher levels of traumas at the respective threshold levels. These findings are consistent with research indicating that historically underinvested neighborhoods (usually targets for gentrification) experience a higher rate of violent penetrating trauma [68] and gentrifying neighborhoods have higher firearm injury [69].

Factors Associated with the Risk of Displacement for Traumatic Experiences

The results for factors influencing estimated displacement risk for traumatic experiences (LEC) at the cut-off point 0.4 are similar for whites (0.0002; p < 0.01) and educational degrees of neighboring communities (0.000006, p < 0.01) and are positively associated with the likelihood of displacement. As the percentage of individuals with educational degrees in neighboring communities increases, it is related to increases in the likelihood of displacement. While as the percentage of degrees within neighborhood increases (-0.0002; p < 0.05) it is associated with decreases in the likelihood of being displaced. OOH across all the threshold levels (0.4, 0.5, 0.6 and 0.7) is significantly related to increases in the estimated risk of displacement. House fees for Airbnb (0.000002, 0.000002, 0.000001 and 0.000001; p < 0.05, p < 0.01, p < 0.05 and p < 0.05), respectively, for neighboring communities are positively correlated with the likelihood of displacement at threshold levels, 0.4, 0.5, 0.6 and 0.7. At the threshold levels 0.4 and 0.5, rent is positively correlated with the estimated risk of displacement albeit at the 10% level of significance. We should note that at the cut-off points 0.4 and 0.5, Black residents (-0.0050 and − 0.0072), respectively, are negatively associated with the risk of displacement at the 10% level. While at threshold levels 0.6 and 0.7, Black residents are positively associated with the risk of displacements at the p < 0.01 level of significance. For the threshold levels 0.4, 0.5, 0.6 and 0.7, Airbnb fees for houses (-0.00005, -0.0001, -0.00003 and − 0.00002; p < 0.01, p < 0.01, p < 0.1 and p < 0.1, respectively) are negatively correlated with the estimated risk of displacement, see Table A1 in the online appendix.

Associations between PTSD symptoms, Covariates (Age and Crime), and Displacement Risk

The estimated risk of displacement related to higher PTSD symptoms (trauma-related feelings, thoughts and behaviors) measured by the PCL (33.7415, p < 0.01; 31.1992, p < 0.01), for threshold levels 0.6 and 0.7, respectively. Therefore, at the 0.6 threshold women at the risk of displacement may experience 33.7415 higher PTSD symptom score when compared to women not at displacement risk. Note that a score of 31 to 33 on the PCL (score ranges 0 to 80) is the cutoff for probable PTSD [52, 70], therefore at this threshold the estimated risk of displacement may be linked with clinically distressing and interfering PTSD. PTSD symptoms for the cut-off points 0.3, 0.4, and 0.6 is negatively correlated with age (-0.3006, -0.0.2852 and − 0.2603; for all p < 0.1), respectively, suggesting that as age increases, women may experience lower PTSD symptoms which may be linked to having enhanced coping strategies and/or receiving therapy over the years. This is consistent with prior literature indicating that older individuals experience posttraumatic growth and less PTSD symptoms later in life due to the utilization of therapy, coping strategies, and social support [71,72,73]. We should note that for PTSD symptoms at the cut-off of 0.3 the estimated risk of displacement has a negative sign and is insignificant while cut-off point at 0.4 is negative and significant at the 10% level, however as the number of households that are likely to be displaced increases, this is associated with households being more likely to experience higher PTSD symptoms indicated by threshold levels 0.6 and 0.7, respectively, see Table 4.

Traumatic cognitions/thoughts (measured by the PTCI) is also negatively associated with age (-0.8721, -0.8163, -0.8969, -0.8437, -0.9524; p < 0.1, p < 0.05, p < 0.05, p < 0.05 and p < 0.01), but positively associated to self-reported crime (0.0961, 0.0972, 0.0936, 0.0979 and 0.0943 for all p < 0.01) for all cut-off points, respectively. The estimated risk of displacement is positive at the 0.6 threshold and significant at the 10% level, that is an average household who is at the risk of being displaced may experience higher PTCI by 80.2389. These findings are consistent with the result noted with number of traumas as well as with a priori expectations based on existing literature [18].

Factors Associated with the Risk of Displacement for PTSD symptoms

For PTSD symptoms (measured by PCL) at the cut-off points 0.6 and 0.7, White residents and OOH are positively correlated with the estimated risk of displacement with OOH significant at p<0.10 at the 0.7 threshold. Educational degree is negatively correlated (p<0.10) with the estimated risk of displacement for the cut-off point 0.6. Black resident is positively correlated with the estimated risk of displacement for threshold 0.6 and 0.7. For posttraumatic cognitions (PTCI) at the cut-off point 0.6, increase in fees for APT is associated with increases in the estimated risk of displacement see Table A1 in the online appendix.

Associations between HIV Medication Adherence, Viral Load, Covariates (Age and Crime), and Displacement Risk

For HIV medication adherence, the estimated risk of displacement is related to a significant decrease in medication adherence (-1.6226, p < 0.01, -1.8574, p < 0.01) across these thresholds, 0.4 and 0.6, respectively, see Table 5. At the 0.6 threshold women at risk of being displaced have 1.857 lower medication adherence score when compared to women not at displacement risk. A near two-point decrease in adherence score (ranges from 1 to 6) could be the difference between adherence ratings of “fair” vs. “very good” and “very poor” vs. “fair” [54]. This suggests that the estimated risk of being displaced may adversely impact women’s ability to adhere to their HIV medication as needed, which may ultimately impact their HIV health outcomes (e.g., viral load).

In addition, households who are likely be displaced also have relatively higher viral load log (VLOG-the logarithm of viral load; 1.5358; p < 0.01) and VL _C_ML (continuous measure of viral load; 25428.9987); p < 0.1) at the threshold level 0.7. Crime is also positively associated with viral load log and linear outcomes at the 0.7 threshold level albeit at the 10% level of significance. Consistent with this, self-reported crime is negatively associated with CD4 count across all the threshold levels (0.3,0.4 0.5 and 0.60). In addition, at the 0.6 threshold, women at displacement risk have a 24.1971 (p < 0.01) lower CD4 count (i.e., lowers immune function) compared to women not at risk of displacement. These empirical results are consistent with the literature [30, 74] highlighting that within Black communities undergoing gentrification or housing insecurity people experience adverse physical health outcomes. Similarly, researchers Gibbons & Barton [74] found that Black individuals in actively gentrifying neighborhoods are nearly 75% more likely to self-report having negative health.

Factors Associated with the Risk of Displacement for HIV Medication Adherence and Viral Load

We note that white individuals (White individuals − 0.0001; p < 0.1), OOH (0.0051; p < 0.1) and house Airbnb fees for neighboring communities (W_House [0.000003; p < 0.05]) are positively correlated with the estimated risk of displacement relating to medication adherence, while fees for house (-0.00004; p < 0.05) within neighborhood is related to a decrease in the estimated risk of displacement at threshold level 0.4, see Table A3 in the online appendix. As the percentage of Black individuals ([0.0016; p < 0.01]) for the threshold 0.6 increases, it is associated with an increase in the estimated risk of displacement relating to medication adherence. For CD4 (threshold 0.6), White individuals (-0.0002; p < 0.05), educational degrees (-0.0001; p < 0.05) and fees for houses within neighborhood (-00003; p < 0.01) are negatively correlated with the risk of displacement while fees for houses in neighboring communities (0.000001; p < 0.05) and OOH (0.0067; p < 0.05) is associated with an increase in the likelihood of being displaced. For the threshold level 0.7, a rise in the percentage of degrees in neighboring communities (W_Degrees) [0.00001; p < 0.1] is associated with an increase in the estimated risk of displacement for viral load while as the percentage increase in rent (-0.0092; p-value < 0.05) it is associated with a decrease in the likelihood of displacement relating to viral load, see Table A3 in the online appendix.

Associations between Undetectable Viral Load, Viral Suppression, Covariates (Age and Crime), and Displacement Risk

The technical difference between undetectable and unsuppressed viral load is that detectable uses a cutoff of > 20 viral load count, while unsuppressed uses a cutoff of > 200. In terms of the health implications, women only need to be unsuppressed to prevent HIV transmission, however being undetectable indicates that there is less of the virus present (not detectable in lab assays) which is good for women’s immune system. Self-reported crime (-0.0016; p < 0.05) for the cut-off point 0.7 is negatively associated with viral suppression. That is an increase in crime may negatively impact women’s HIV health, perhaps via medication adherence or biological mechanisms (e.g., inflammation, immune functions) [75]. The estimated risk of displacement is associated with lower undetectability (positive indicator of good HIV health) for the cut-off point of 0.4 (-1.5435; p < 0.01), see Table (6). The computed marginal effect (-0.5511; p < 0.001) at this threshold suggests that the probability of estimated risk of displacement for an undetectable individual is 0.5511 lower when compared to an individual who is detectable, see Table (7). Similarly, for viral suppression, at threshold 0.7 (-1.6048, p < 0.01) the average household who is at the risk of being displaced experience a relatively lower viral suppression (negative HIV health). Computed marginal effects indicates that at this 0.7 threshold the probability of displacement risk for a woman with viral suppression is 0.4548 lower than a woman with viral non-suppression. The link between displacement, undetectable viral load, and viral suppression could be due to hardships (e.g., higher cost living) imposed by gentrification making it more difficult for women to engage in care (e.g., attend appointments, adhere to medication). These findings are consistent with existing literature noting that (a) gentrification plays a role in time to viral suppression [24] and (b) eviction (a displacement event) evictions increased the odds of having a detectable viral load [25].

While the above findings are aligned with our hypothesized associations between displacement and health outcomes, some findings are not aligned at other thresholds of estimated displacement risk. The estimated risk of being displaced is associated with an increase in the likelihood of having an undetectable viral load for the thresholds 0.3, 0.6 and 0.7 (1.3522, 1.0667, 1.0301; all the p < 0.01), see Table 6. The results are similar for viral suppression at the cut-off points of 0.3, 0.5 and 0.6 (1.1003, 1.3128, 1.1849; all the p < 0.01) with a higher estimated risk of displacement being associated with an increase in the likelihood of having suppressed viral load, see Table 6. Further, a similar contradiction is observed with CD4 where at the 0.3 threshold, women at displacement risk have a higher CD4 (15.7654, p < 0.05) compared to women not at risk of displacement.

Although contradictory to what one might postulate, these findings are consistent with a prior finding indicating that higher neighborhood median income was associated with lower likelihood of HIV viral suppression and undetectable viral load among BWLWH, see Wright et al. [18]. A plausible explanation of our current counterintuitive findings could be one of the benefits of gentrification - increase in tax revenue resulting in more nearby facilities such as pharmacies and local clinics which increase access to care. Further, factors unmeasured (e.g., actual displacement events) and unexamined here (e.g., changes in viral suppression, undetectable status overtime) may play a role [24, 25]. Overall, both intuitive and counterintuitive findings suggest that the paths by which risk of displacement may impact HIV outcomes among BWLWH are complex and requires ongoing investigation with attention to measurement, methods, and insights from BWLWH to detangle the complexity.

Factors Associated with the Risk of Displacement for Undetectable Viral Load and Viral Suppression

For detectable viral load, the likelihood of being displaced is positively associated with rent (0.0148; p < 0.05), OOH (0.0057; p < 0.05), neighboring house fees (0.000005; p < 0.01), and households with educational degrees (0.00002; p < 0.01) in neighboring communities at the 0.4 cut-off point. However, fees for houses within the communities (-0.0001; p < 0.01), neighboring community rent (-0.0020; p < 0.01), and neighboring OOH (-0.0008; p < 0.01) are negatively associated with the likelihood of being displaced at the same threshold level, see Table A2 in the appendix. The result is similar for HIV viral suppression at the threshold level 0.7 in which educational degrees in neighboring communities are positively correlated with risk of displacement albeit at the 10% level of significance, see Table A2 in the online appendix.

Conclusion

In this paper we examine if the estimated risk of displacement is important in explaining variation in mental health, physical health, and health behaviors among Black women living with HIV in South Florida. We specified two structural equations (outcomes and displacement) which are estimated jointly using the method of maximum likelihood while accounting for robust standard errors. Using non-linear methods, we model the estimated risk of displacement as a right-hand side endogenous variable while accounting for geographical dependencies, that is variation in outcomes may not only be associated with characteristics of women’s communities but also the characteristics of the neighboring communities. We incorporate these possible effects in the structural model providing a novel contribution to the literature. The empirical findings reveal that the risk of being displaced is associated with (a) higher mental health symptoms (e.g., depression, PTSD) (b) lower adherence to HIV medication, and (c) higher viral load linear and log. However, there were mixed findings for the risk of displacement in relation to viral load undetectability and suppression. Due to the dearth of information on the impact of displacement risk (related to gentrification) on the health and experiences of Black women living with HIV, the strength of the current research is evident and provides novel information that helps to begin to fill the gap in the existing literature and inform future research.

These empirical results are informative and provide insights to policy makers such as the need to address displacement risk and its association with health outcomes. Future research should focus on the creation and implementation of interventions that address displacement (associated with gentrification) and mitigate the impact it may have on both mental and physical health outcomes for BWLWH. Given that the sample is relatively small, longitudinally analyses in which a larger number of individuals are tracked overtime would provide additional inference. In addition, given the observational nature of the present data causal inferences are limited. Lastly, this study did not measure actual displacement events such as residential moves, evictions, or forced relocations nor disentangle all drivers of displacement such as sea level rise and climate change that may have adverse effects on health and force residents in South Florida to relocate. Therefore, these noted factors should be explored in future research.