Smoke-Free Policies and Resident Turnover: An Evaluation in Massachusetts Public Housing From 2009–2018

Abstract

Introduction:

Smoke-free policies (SFP) in multi-unit housing are a promising tool for reducing exposure to tobacco smoke among residents. Concerns about increased housing instability due to voluntary or involuntary transitions induced by SFPs have been a primary barrier to greater widespread adoption. The impact of SFP implementation on transitions out of public housing in federally funded public housing authorities in Massachusetts was evaluated.

Methods:

Tenancy data from the Department of Housing and Urban Development were used to determine the time from admission to transitioning out of public housing based on a cohort study design. Periods of exposure to SFPs were defined based on dates of SFP implementation at each PHA. Multi-level Cox regression models were fit to estimate the effects of SFPs on the hazard of transitioning, adjusting for household- and PHA-level characteristics. Analyses were conducted in 2021–2022.

Results:

There were 44,705 households with a record of residence in Massachusetts PHAs over 2009–2018. Over this period, despite increasing adoption of SFPs among the PHAs, rates of transition remained steady at around 5–8 transitions per 1,000 household-months. There was no overall association between exposure to SFPs and transitions among the full sample (adjusted HR=0.99, 95% CI=0.95, 1.04, p=0.794). However, the association varied significantly by age group, race/ethnicity, timing of SFP adoption, and era of admission.

Conclusions:

Adoption of SFPs in public housing had a minimal overall impact on turnover for households in Massachusetts, though disparities in the impact were observed between different demographic and PHA-level subgroups.

INTRODUCTION

Smoking is the leading preventable cause of death in the U.S.¹ It has been estimated to cause over 480,000 deaths annually, among which 41,000 deaths are attributable to secondhand smoke (SHS). Smoke-free laws have been increasingly adopted to protect against exposure to tobacco smoke in public locations, including workplaces, restaurants, and bars.² However, there remain additional locations where individuals face a high risk of exposure.^2–4 Among such locations, multi-unit and public housing are especially concerning, as residents may be involuntarily exposed for prolonged durations. Residents of public housing are further vulnerable due to limited means of finding new housing, elevated rates of smoking, and poor underlying health conditions among low-income populations.^5–7

In recognition of the risk of SHS exposure among residents of public housing, the Department of Housing and Urban Development (HUD) required public housing authorities (PHAs) to implement a smoke-free policy (SFP) after July 2018.⁸ However, there has been limited empirical evidence on the unintended consequences of adopting SFPs in public housing. In particular, concerns over increased vacancies, increased turnovers, and reduced market share have been consistently reported by housing directors and residents of multi-unit housing as a primary barrier to broader adoption of SFPs.^8–13 Housing instability is also known to have adverse associations with health outcomes.¹⁴

Existing literature on the impact of SFPs on resident turnover have relied upon self-reported outcomes from surveys of owners, managers, and residents. Among early adopters at multi-unit housing in Portland,¹⁵ New York City,¹⁶ North Carolina,^7,17 Georgia,¹⁷ and South Dakota,⁹ most owners and managers reported no change in turnover rates following implementation of SFPs. Cases of voluntary transitions or evictions were also reported to be rare.^7,16,17 Although these studies provide timely accounts of the impact of SFPs, they are largely based on perceptions of housing operators and may be subject to bias. Objective evaluations of outcomes are needed.¹⁸

A majority of federally funded PHAs in Massachusetts (MA) adopted SFPs well before the federal ruling (Appendix Figure 1).¹⁹ This study leverages longitudinal HUD tenancy data for households in all federally funded PHAs in MA combined with data on dates of SFP implementation in MA PHAs to evaluate the impact of SFPs on transition rates out of public housing. This evidence from the early experience of MA PHAs can be used to inform the ongoing implementation of SFPs in public housing and future implementation at HUD-supported housing assistance programs and private multi-unit housing.

METHODS

The HUD Public Housing program administers federal grants and subsidies to state- and local-level PHAs to manage public housing for qualified low-income families, the elderly, and persons with disabilities.²⁰ HUD collects data from annual reexaminations on these households for eligibility, monitoring, and reporting purposes. These data include information on participant identities, demographics, contact information, dates of enrollment, recertification, termination, disability status, and the number of dependents in the household.

The dates of implementation of SFPs at MA PHAs were obtained from official PHA documents or publicly available meeting minutes, a survey conducted with directors and employees of MA PHAs in 2018, or from the Americans for Nonsmokers Rights Foundation²¹ (Appendix Table 1). This study was approved by the Mass General Brigham IRB.

Study Population

Using the HUD tenancy data, all households who had a record of participating in the Public Housing program in MA between January 1, 2009–May 31, 2018 were identified. Given limitations in follow-up of households in more recent years of the available tenancy data, the study period was restricted to exclude the effects of completely involuntary adoption, i.e., for PHAs that had not adopted until the implementation deadline in July 2018.⁸ Households who were admitted to a PHA after the PHA had adopted a SFP were excluded to focus on the impact among those who were in residence prior to the policy, although the impact of including these households was also assessed in sensitivity analyses. Households in one large PHA in eastern MA were excluded due to an administrative change at that PHA. Households who were admitted prior to 1980, terminated on the same day as admission, had a household head with age less than 18 years or greater than 110 years old at admission, or had missing covariates were also excluded.

Measures

The primary outcome for the time to event analyses was the time to transition out of the Public Housing program, where the transition date was defined by the date of lease termination, irrespective of the reason, as recorded in the HUD data. Time zero was the date of the beginning of tenancy. The time to transition was right-censored by the earliest of either the end of follow-up for the household, the end of the study period (May 31, 2018), or 2 years after the SFP at the household’s PHA, to examine the immediate impact following the SFP. The end of follow-up was defined as the date of the last observed interim or annual reexamination for households without a record of termination. Sensitivity analyses were also conducted to assess the impact of removing the restriction of the follow-up up to 2 years after the SFP. The time to transition was also left-truncated²² by the initiation of the study period (January 1, 2009) such that transitions occurring prior to 2009 are not observed (Appendix Figure 2).

Household-level characteristics including demographics (age, gender, race/ethnicity) and disability status for the head of household were obtained from the HUD data. The year of admission, duration of tenure, whether the household had children, and type of housing development where a household resided (family versus elderly/disabled housing development) were also identified. Both the presence of children in the household and type of housing development were time-varying covariates over the follow-up based on changes noted in reexamination records. PHA size and rurality were determined based on state records.^23,24

Statistical Analysis

The incidence of transitions was calculated as the number of transitions divided by the number of households under follow-up in each calendar month and plotted over the study period with CIs.²⁵ Multi-level Cox proportional hazards models were used to assess the association between exposure to SFPs and time to transition out of public housing.^26–28 The probability of transitioning out of public housing may vary over the course of a household’s tenure and can bias estimates of the policy effect since exposure to SFPs is more likely to occur after a longer tenure. The models adjust for this by using time since admission as the timescale, which allows the baseline hazard of transitions to be flexible over tenure length and enables comparison of the hazards between households with the same tenure.²⁹ The baseline hazard was stratified by 5-year intervals for the year of admission, starting from 1980, to account for cohort effects and secular trends over time. Exposure to SFP was encoded as a binary time-varying covariate that is zero prior to policy implementation at the PHA in which the household resided and one afterwards. The model also included PHA-level Gaussian random intercepts to account for correlated outcomes among households within a PHA and unobserved PHA-level effects.^30–31 Both a simple model without other fixed effect adjustments and a fully-adjusted model that additionally adjusted for household- and PHA-level fixed effects were fit.

A series of sensitivity analyses were conducted to assess the robustness of findings to varying data definitions. These included models in which the policy time was artificially set to be 1-year prior to the implementation date, censoring at 2 years after the policy date was not imposed, households admitted after the policy were included, and additional inferred transitions were included for households without a lease termination in Public Housing but had subsequent data in other HUD programs within 1–2 years of the last Public Housing record.

Heterogeneity in the policy effect across subgroups was assessed by fitting the main fully-adjusted model with an interaction term between the subgrouping factor and the time-varying effect for policy. A joint-Wald test based on the fitted model was applied to test for the presence of interaction effects. P-values in the main regression analyses and sensitivities are based on 2-sided Wald tests. The data were prepared using SAS 9.4 (SAS Institute Inc, Cary, NC) and analyzed using R 4.0.4 (R Foundation for Statistical Computing, Vienna, Austria). The analyses were conducted over 2021–2022.

RESULTS

There were 44,705 households with a record of residence during the study period (January 1, 2009–May 31, 2018) after applying the sample selection criteria (Appendix Table 2). Among the 68 federally funded PHAs in MA,²³ 7 PHAs (3,854 households) were excluded based on the selection criteria. The study sample includes 3,914 households from another 7 PHAs that adopted SFPs after the study period and contributed only pre-policy follow-up. Overall, heads of household were largely evenly distributed among the 20–40 (32.9%), 40–60 (32.0%), and 60–80 (29.4%) age groups at admission (Table 1). Household heads were more likely to be female (66.9%), and most were either White (41.8%), Hispanic (33.2%), or Black (17.9%). Most households were admitted between 2000–2014 (74.6%), having a tenure of over 2 years prior to study entry. A substantial number of household heads reported living with a disability at study period initiation (27.9%) or having children present during the study period (33.7%). The households were roughly evenly split between residing only in elderly/disabled (E/D; 50.4%) or family developments (46.0%). Most households were in large (68.7%) or medium-sized (26.4%) PHAs. There were some differences in characteristics among households at PHAs adopting SFPs after the study period (Appendix Table 3). During the study period, the monthly incidence of transitions was low, ranging around 5–8 transitions per 1,000 households each month, with no clear signal for an elevation of rates following implementation of a policy (Figure 1).

Table 1.

Summary of Household- and PHA-level Characteristics

Characteristics	Overall (n=44705)
Age at study entry, years, n (%)
18–20	1,278 (2.9)
20–40	14,718 (32.9)
40–60	14,309 (32.0)
60–80	13,140 (29.4)
80–110	1,260 (2.8)
Male, n (%)	14,800 (33.1)
Race/ethnicity, n (%)
Asian	2,894 (6.5)
Black	8,009 (17.9)
Hispanic	14,830 (33.2)
Multiple	147 (0.3)
Native American	144 (0.3)
White	18,681 (41.8)
Year of admission, n (%)
1980–1984	659 (1.5)
1985–1989	1,194 (2.7)
1990–1994	2,472 (5.5)
1995–1999	5,067 (11.3)
2000–2004	7,830 (17.5)
2005–2009	13,650 (30.5)
2010–2014	11,911 (26.6)
2015–2019	1,922 (4.3)
Tenure until study entry in years, median (IQR)	2.06 [0.00, 7.75]
Tenure until transition in years, median (IQR)	6.38 [3.03, 12.00]
Disabled head of household at study entry, n (%)	12,482 (27.9)
Children during study period, n (%)	15,085 (33.7)
E/D development, n (%)
Both during follow-up	1,607 (3.6)
E/D development throughout	22,548 (50.4)
Family development throughout	20,550 (46.0)
PHA size, n (%)
Large (≥1,000 units)	30,708 (68.7)
Medium (<1,000 units)	11,802 (26.4)
Small (<100 units)	2,195 (4.9)
Smoke-free policy adoption at PHA, n (%)
Early (2012 or earlier)	13,878 (31.0)
Middle (2013–2015)	15,066 (33.7)
Late (2016–2018)	15,761 (35.3)
Rural PHA, n (%)	359 (0.8)

^aAge, gender, and race refer to the head of each household.

^bTenure until study entry and transition are defined to be the time from admission to entry and transition, respectively, in years.

^cDisability status based on earliest record available for household, and children status based on having any children during the study period.

^dE/D development status is defined based on known status during follow-up of households.

E/D, elderly/disabled; PHA, public housing authority.

Figure 1.

Monthly incidence of transitions by calendar time.

^aMonthly transitions out of public housing per 1,000 households over January 2009 through May 2018.

^bShading represents “exact” 95% CIs for Poisson means.²⁵

^cDotted lines represent times of SFP adoption at a MA PHA.

SFP, smoke-free policy; MA, Massachusetts; PHA, public housing authority.

In the simple model, there was no significant difference in the rate of transitions between households exposed vs. unexposed to a SFP (HR=1.01, 95% CI=0.96, 1.06, p=0.783; Table 2). The lack of association persisted after adjustment for household- and PHA-level characteristics in the fully adjusted model (HR=0.99, 95% CI=0.95, 1.04, p=0.794). The lack of a policy effect on transitions was robust to alternative definitions for the policy timing, censoring and transitions, and the study population (Appendix Table 4).

Table 2.

Estimates of SFP Effect and Household- and PHA-level Characteristics

Household- or PHA-level characteristic	Simple model		Fully-adjusted model
	HR (95% CI)	p-value	HR (95% CI)	p-value
Smoke-free policy	1.01 (0.96, 1.06)	0.783	0.99 (0.95, 1.04)	0.794
Age at study entry, years
18–20			(ref)
20–40			0.74 (0.68, 0.81)	<0.001
40–60			0.63 (0.58, 0.7)	<0.001
60–80			0.82 (0.74, 0.91)	<0.001
80–110			1.57 (1.39, 1.78)	<0.001
Gender
Female			(ref)
Male			1.08 (1.05, 1.12)	<0.001
Race/ethnicity
Asian			0.58 (0.53, 0.63)	<0.001
Black			0.84 (0.8, 0.88)	<0.001
Hispanic			0.78 (0.75, 0.82)	<0.001
Multiple			0.88 (0.68, 1.15)	0.359
Native American			0.85 (0.64, 1.12)	0.242
White			(ref)
Disabled head of household			0.91 (0.87, 0.95)	<0.001
PHA size
Small (<100)			(ref)
Medium (<1,000)			0.95 (0.81, 1.11)	0.512
Large (≥1,000)			0.99 (0.81, 1.22)	0.955
Rural PHA			1.1 (0.73, 1.66)	0.652
Development type
Family housing			(ref)
E/D Housing			1.13 (1.08, 1.19)	<0.001
Children in household			1.13 (1.07, 1.19)	<0.001
PHA random intercept SD	0.27		0.25
Integrated AIC	265,967		263,693

Note: Boldface indicates statistical significance (p<0.05).

^aEstimates based on Cox proportional hazards models with a time-varying covariate for policy on the time-scale of time since admission.

^bTransitions ascertained based on records of lease termination within the Public Housing program.

^cFully-Adjusted model additionally adjusts for PHA-level random effects and stratifies the baseline hazard for years of admission by 5-year intervals starting from 1980.

^dPHA random intercept SD refers to estimated SD for Gaussian random intercept.

^eIntegrated AIC is the AIC calculated based on the log partial likelihood after integrating out the random effects.

AIC, Akaike information criteria; E/D, Elderly/disabled; HR, Hazard ratio; PHA, public housing authority; SFP, smoke-free policy.

In the subgroup analyses (Figure 2), there were significant differences in the policy effect by the age (p-interaction<0.001) and race/ethnicity (p-interaction=0.01) of the head of household. The effect of SFPs exhibited an increasing trend with age, in which households with younger household heads had a reduced rate of transitions under the policy (HR=0.76, 95% CI=0.61, 0.95 and HR=0.91, 95% CI=0.85, 0.98 for 18–20 and 21–40 age groups) and households with older heads had an increased rate (HR=1.23, 95% CI=1.03, 1.48 for 81–110 age group). Hispanic households experienced lower rates of transitions under the policy (HR=0.92, 95% CI=0.85, 0.99), whereas Native American households had higher rates under the policy (HR=1.77, 95% CI=0.97, 3.24). There was also heterogeneity in the SFP effect across PHAs with different timing of adoption (p-interaction<0.001). Households in early-adopting PHAs (2012 or earlier) experienced reduced rates of transitions under SFPs (HR=0.83, 95% CI=0.77, 0.90), while households in middle- and late-adopting PHAs experienced higher rates of transitions under SFPs (HR=1.10, 95% CI=1.03, 1.16 and HR=1.12, 95% CI=0.96, 1.29). The policy effect was also significantly higher among households that were admitted during the study period (HR=1.07, 95% CI=1.00, 1.15) vs prior to study period (HR=0.95, 95% CI=0.89, 1.01; p-interaction=0.005).

Figure 2.

Estimates of SFH policy effect by household- and PHA-level subgroups.

^aEstimates based on Cox proportional hazards model with a time-varying covariate for policy, main effects for each factor, and interaction effect between policy and the factor. The Cox model time-scale is the time since admission.

^bTransitions ascertained based on records of lease termination within the Public Housing program.

^cThe Pre-Policy Rate refers to number of transitions per 1,000 person-months. Follow-up refers to follow-up time in 1,000 person-months prior to the transition.

^dAge, gender, and race refer to the head of household.

^eCox models adjust for the same covariates as those reported in Table 2 as fixed effects and include PHA-level random effects stratifying by years of admission by 5-year intervals from 1980.

^fDashed line denotes the estimated HR of SFH policy based on the adjusted Cox model in the overall cohort.

^gp-value based on Wald-test for the joint-hypothesis that interaction terms.

Fup, follow-up; HR, hazard ratio; PH, Public Housing; PHA, public housing authority; p-val, p-value; SFP, smoke-free policy.

The elevated relative hazards in these subgroups translate to small absolute increases in rates of transition. Assuming constant hazards over time and interpreting the HR as an approximate incidence rate ratio,³² these estimates imply approximately 14.11 × (1.23–1) ≈ 3.25 excess monthly transitions per 1,000 households among those with heads of household older than age 80 years, for example (Figure 2). Similarly, the excess monthly transitions post-policy would be 0.55, 4.13, 3.30, 0.70, and 0.43 transitions per 1,000 households among White households, Native American households, rural households, households in middle adopter PHAs, and households admitted during the study period, respectively.

DISCUSSION

Along with the promise of reduced morbidity and mortality, SFPs in public housing have also raised concerns about unintended consequences. Recent studies evaluating SFPs through focus groups,^33–34 key informant interviews, and monitoring of nicotine and particulate matter levels have found limited effectiveness in smoke exposure reduction.^35–37 Findings in this paper complement these works by providing additional characterization of SFP effects. Previous studies evaluating associations between SFPs and turnover in public or multi-unit housing have largely relied upon cross-sectional, self-reported data collected from surveys administered to property owners/managers or residents.^7,9,15–18 The HUD administrative data in this study provide an objective accounting of turnovers before and after the adoption of SFPs to enable evaluation of SFPs across diverse demographic groups and PHAs in the state.

The rate of transitions out of public housing was low in MA, generally not exceeding more than 10 transitions per 1,000 household-months. Overall, no differences in transition rates between households exposed versus not exposed to SFPs were found. The associations between transitions and household-level characteristics in the fully adjusted model are generally consistent with previous research, in which households with older members, members living with disability, no children, female or racial or ethnic minority heads of household are more likely to have longer tenures.³⁸

Notwithstanding the lack of an overall effect, subgroup analyses identified subpopulations of households that may be affected by SFPs. Households with older heads were more likely to transition out of Public Housing after the implementation of SFPs. One possible explanation is that older people who smoke are less likely to quit smoking due to decreased interest in quitting, increased dependence, and generational differences in attitudes towards smoking.³⁹ Older people who smoke may also be more likely to believe that they are too old to change their behavior¹⁹ or to leave public housing to seek support at long-term care facilities. Native American and, to a lesser degree, White households were also more likely to transition under SFPs. These 2 racial/ethnic groups have the highest smoking rates in the U.S.⁴⁰ and would include more individuals who would have been directly affected. Households in rural PHAs also exhibited higher transition rates after SFP implementation than those in urban PHAs, though the differences were not significant. The high levels of demand for housing at urban PHAs may make it difficult for households with people who smoke to find alternative accommodations, in contrast to the greater housing mobility in rural settings. Due to the limited data available on rural and Native American households, these findings should be regarded as preliminary.

For some groups, SFPs are associated with reduced rates of housing transitions. Younger households, which likely include households with more children, and Hispanic households experienced greater stability under SFPs and may favor housing with SFP coverage. Households in PHAs adopting SFPs in later years experienced higher transition rates, while households among early-adopting PHAs experienced greater stability under SFPs. There may be selection effects such that early-adopting PHAs were those in communities where SFPs were more readily accepted. Additionally, implementation and enforcement strategies may develop and evolve over time, with early adopters having the least experience to inform their execution. The late adopters include 17 PHAs that implemented SFPs after the finalization of the federal rule in December 2016⁸ and may consist of PHAs that implemented, at least partially, to comply with the rule. There appears to be no clear departure in associations with SFPs among these PHAs relative to middle adopters. Households admitted prior to the study, which have tenure up to 10–20 years prior to the study period, were less likely to transition out of PHAs with SFPs, whereas households admitted during the study period exhibited more transitions. This may alleviate concerns expressed among PHAs about the possible forcing out of long-term residents who smoke.⁹

Limitations

Whether transitions observed were voluntary, involuntary, or known to be related to SFPs could not be verified since information on reason for termination was not available in the HUD tenancy data. This study leverages the natural variation in the time of SFP implementation relative to household admission and a rigorous set of statistical adjustments for household and PHA characteristics to isolate the implied effects of SFPs on all-cause turnovers. It remains unclear, however, whether the lack of excess turnovers could be due to willingness to comply with SFPs among residents or inadequate enforcement among PHAs. Other differences in implementation approaches across PHAs may also play an integral role on transitions. Similarly, data was not available on whether households included residents who smoke to enable estimation of effects specifically among households with people who smoke.

The analyses were restricted to assess the impact occurring at the level of households, though individual members may respond differently to SFPs within households. A more granular analysis would account for tenancy of individual members. A household-level analysis was adopted in this study, as data on member-level tenancy were less reliably recorded.

Lastly, the effect of SFPs estimated here are limited to households in federally funded public housing in MA between 2009–2018, prior to national HUD-wide SFPs for PHAs that took effect after July 2018. SFPs among PHAs that adopted at the implementation deadline in July 2018 may experience a different impact than those that adopted SFPs earlier on a voluntary or semi-voluntary basis. Nevertheless, these results provide a basis for further evaluations and would be especially pertinent to future adoption at multi-unit housing not mandated to be smoke-free.

CONCLUSIONS

The potential for SFPs to spur evictions, vacancies, and turnover has been among the top concerns about the policies among public health professionals and property managers in multi-unit housing.^9–12 Despite these concerns, overall there was no evidence that implementation of SFPs has led to increased rates of turnover in practice. This suggests that the perceived risk of increased turnover ought not to be a barrier for the adoption of SFPs in general. However, households among certain demographics may be more sensitive to the impact of SFPs. Though the absolute number of excess transitions is small, further work would be needed to evaluate the positive and negative impacts among these households and develop tailored approaches for implementing SFPs that minimize unintended consequences and maximize potential benefits.