You are on page 1of 15

Article

Acta Sociologica
1–15
Does housing instability matter ª The Author(s) 2020
Article reuse guidelines:
sagepub.com/journals-permissions
for youths’ educational DOI: 10.1177/0001699320939629
journals.sagepub.com/home/asj
attainment? Findings from
Swedish longitudinal register data.

Anna Kahlmeter
Swedish Institute for Social Research, Stockholm University, Sweden

Abstract
There is an ample body of research demonstrating the link between housing instability and
adverse outcomes. The bulk of this research, however, largely relies on broad operationaliza-
tions, generally not considering different types of housing instability. This study extends previous
research by focusing on adolescents facing a variety of residential events, including moves,
imminent threats of eviction and forced relocations, while also considering the significance of
distance. Adopting a counterfactual approach, and drawing on unique data on evictions in
Sweden alongside a link to longitudinal registers, this study examines the association between
housing instability and educational attainment, operationalized as graduation from upper sec-
ondary school. Theoretically, the study draws on the family stress model and theory on social
capital, the findings providing support for both approaches. Single relocation was found to have a
small impact on educational attainment, but forced relocations, repeated relocations and long-
distance relocations are of particular significance for understanding the link between housing
instability and educational outcomes. The study contributes to an understanding of the roles that
different types of residential events play in youths’ educational attainment, and policy implica-
tions are discussed.

Keywords
Educational attainment, eviction, forced relocation, housing mobility, housing instability,
residential events, youth

Introduction
Housing and educational attainment are both salient aspects of youths’ wellbeing and development.
There is an ample body of research demonstrating the link between unstable housing and mobility in

Corresponding Author:
Anna Kahlmeter, Swedish Institute for Social Research, Stockholm University, Stockholm 106 91, Sweden.
Email: anna.kahlmeter@sofi.su.se
2 Acta Sociologica XX(X)

childhood, and poor outcomes later in the life course (see Jelleyman and Spencer, 2008; Scanlon and
Devine, 2001). The bulk of this research, however, conflates different types of residential events, largely
relying on broad operationalizations of housing instability, this being likely to obscure important varia-
tion. It seems that moving is to a large extent a stressful event that compounds negative outcomes, yet we
have little information about the mechanisms at play and the significance of different types of residential
strain. The purpose of the present study is to further our understanding of the association between
housing instability during youth and educational attainment, conceptualized as graduation from upper
secondary school. Housing instability is operationalized as four types of residential events: single
relocation (moving once during upper secondary school, without involvement of the Enforcement
Agency), repeated relocations (moving at least twice during upper secondary school, without involve-
ment of the Enforcement Agency), imminent threat of eviction (eviction threat was registered with the
Enforcement Agency, but was not followed by a move), and forced relocation (eviction threat was
registered with the Enforcement Agency, and the youth moved). In addition, based on the supposition
that long-distance moves imply greater loss of social resources, the study also explores whether the
distance moved is of significance.
Sweden, the context of this study, makes an interesting case for a number of reasons. The Swedish
housing system has traditionally been a bearing pillar in the welfare state. With a national model of
public housing, the aim has been to provide rental housing of high quality for all, rather than means-
tested social housing targeted at low-income families (Grander, 2018). Over the past decades, however,
Sweden has pivoted toward a deregulated, market-based system and a reduction of the rental sector
(Magnusson and Turner, 2008). Moreover, inequalities in educational outcomes are rising in Sweden
with an increasing gap between both the highest and the lowest performers and socioeconomically
advantaged and disadvantaged students (OECD, 2017). At the same time, education is high on the
agenda and the reduction of early school leaving is a primary objective in the Europe 2020 strategy
(European Commission, 2020). Educational attainment is vital for future labor market attachment
(Brunello and Paola, 2014) and, by extension, to the prospects of obtaining and retaining a home,
highlighting its relevance as a measure of success in young adulthood.
Theoretically, the study adopts both the family stress model (FSM) and theories on social capital to
understand the outcomes. Drawing on unique data from the Swedish Enforcement Agency, and linking
to comprehensive longitudinal register data that allow adjustment for a multitude of covariates, the
present study is the first to inform on a broad range of residential events. Furthermore, by means of
inverse probability weighting with regression adjustment (IPWRA), the study adopts a counterfactual
approach that reduces selection bias when randomized trials are not viable, providing a good basis for
analyzing housing instability. Finally, by scrutinizing the assumptions upon which the analyses rest, the
robustness of the estimates is also evaluated.

Theory and previous research


Links between housing instability and educational outcomes. While knowledge about the link between evic-
tion and educational attainment is scarce, previous research has demonstrated that eviction is associated
with a host of other adverse outcomes, demonstrating serious implications, such as decreased chances of
acquiring decent and affordable housing (Desmond et al., 2015), increased economic hardship (Kahl-
meter et al., 2017), elevated risks of criminal convictions (Alm, 2018), and of placement in out-of-home
care for children (Berg and Brännström, 2018). Furthermore, there is an ample body of research that
links housing mobility in general to reduced academic achievement (Cutuli et al., 2013; Haelermans and
De Witte, 2015). Metzger et al. (2015) investigated the impact of housing instability in adolescence on
the likelihood of graduating from high school. Controlling for a range of predictors of housing mobility,
they found that the likelihood of graduating from high school by the age of 25 was 50% lower among
mobile adolescents, regardless of whether the student moved to a poorer or less poor neighborhood. In a
Swedish context, Chen (2013) investigated the impact of both housing tenure and residential mobility on
Kahlmeter: Does housing instability matter for youths’ educational attainment? 3

adolescents’ academic performance, finding that residential mobility has a harmful effect whereas there
is a positive impact for homeownership. Astone and McLanahan (1994) investigated if high levels of
residential mobility account for part of the association between living in a non-intact family and early
school leaving, finding that children from single-parent families and stepfamilies are highly mobile, and
that 30% of the risk difference can be explained by differences in housing mobility.
Haveman et al. (1991) examined the effect of a range of family events during childhood on the like-
lihood of high school completion, finding a strong negative impact of residential moves on the probability
of graduating from high school. Similar results were found by Hagan et al. (1996) in a study of students in
Toronto. Students who moved during the three years prior to baseline were significantly less likely to
graduate from high school or college. The authors also found that these effects were significantly more
pronounced in families where the parents were uninvolved or unsupportive, suggesting a compensatory
role of parental involvement. In contrast, South et al. (2007) found that the elevated risks of early school
leaving for mobile youths could be attributed to changes in the structure and composition of adolescents’
peer networks rather than to differences in parent–child relationships. Moreover, scholars have predicted
that long-distance moves are more harmful, since they are associated with greater disruptions in relation-
ships with pro-social peers and teachers and with diminished sources of social capital, compared with
short-distance moves (Haynie et al., 2006; South and Haynie, 2004).
While most research finds a negative association between residential mobility and educational
achievement, there are studies suggesting that this relationship is mainly attributable to selection (Pettit
and McLanahan, 2003; Pribesh and Downey, 1999), or even that mobility has a positive impact on
educational attainment (Hango, 2006). It has also been suggested that the effect of residential mobility
varies according to socioeconomic status (Hofferth et al., 1998) or ethnicity (Perkins, 2017).
It is well established that those at risk of eviction are an economically disadvantaged group (Tsai and
Huang, 2019; von Otter et al., 2017), and low-income families have been found to be more mobile than
high-income families (Metzger et al., 2015; Pribesh and Downey, 1999). There is also evidence, however,
that the relationship between income and mobility is U-shaped, with those at the lower and upper ends of
the income distribution moving the most, but for different reasons (Desmond et al., 2015). This demon-
strates the significance of considering different types of residential event, a perspective that previous
research has largely overlooked. Furthermore, the impact of residential events is likely to be context-
specific (Ginsburg et al., 2011), and while there are numerous examples of research from the USA, there
are few studies conducted in a Nordic welfare state context. These issues motivate further investigation.

Social capital and family stress. Housing has gained increasing attention in research, and the link between
housing mobility and adverse outcomes is largely explained by a loss of social capital. The concept of
social capital refers to relationships and interactions both within and outside the family (Coleman, 1988).
Within the family, social capital is constituted by positive parent–child interactions, which allow for the
transfer of human capital to children. Outside the family, social capital is represented by the adolescent’s
peer network and by social relationships between parents, the school, and other institutions, providing
the adolescent with consistent expectations and norms. Residential relocations can be expected to be
harmful for youths because of diminished social capital in terms of the breakage of ties with significant
adults outside the household as well as disruptions in peer relationships (Coleman, 1988; Hagan et al.,
1996; Pribesh and Downey, 1999; Teachman et al., 1996).
The current study investigates a variety of residential events, including threats of eviction that were not
followed by an actual relocation. These may nevertheless induce a sense of insecurity and have conse-
quences for youths (Vásquez-Vera et al., 2017), whereby the study draws on both the concept of social
capital and the FSM. The FSM takes its starting point in economic hardship, but can also be applied to other
environmental stressors (Masarik and Conger, 2017). The most common reason for receiving an eviction
threat is rent arrears, and within the framework of the FSM, an eviction threat is hypothesized to generate
additional psychological pressure, as well as economic pressure to repay rent arrears in order to regain
one’s protected tenancy. This may lead to parental distress, disrupted parenting with increased family
4 Acta Sociologica XX(X)

conflicts, and reduced support for the adolescent, which may negatively affect school performance. While
the threat of eviction may be a stressful life event in its own right, the act of moving may induce additional
stress, especially if the relocation is forced by authorities. In addition, during stressful events when intra-
family relations are strained, relationships outside the family may be particularly important. Accordingly,
the two theoretical approaches are treated as complementary rather than conflicting.
With reference to the FSM, it is predicted that there is an independent impact of eviction threats on
educational attainment. Based on theories of social capital, however, moving, even without involvement
of the Enforcement Agency, is also expected to have an impact on educational attainment, particularly in
cases of long-distance moves, since this is expected to diminish social capital resources. It is further
hypothesized that the impact is greater if the relocation is forced, or in cases of repeated relocations.

The Swedish context


Housing and eviction. The Swedish rental market has historically been regarded as the role model for an
integrated rental system outlined by Kemeny (1995) and characterized by an extensive public rental
sector and state intervention in the market, as opposed to a dual housing system with a nonprofit sector
targeted at the poor. With extensive housing policies, state subsidized construction, strong tenant
protection, collectively negotiated rent levels and strong regulation of the public housing companies,
housing has been an important element in the Swedish welfare state. However, during the past decades,
the housing system has adjusted toward market orientation, more cooperatively owned residences, and a
reduction of the rental sector (Magnusson and Turner, 2008). Moreover, Sweden witnesses a wave of
renovations of the so-called Million program, which was a national housing program resulting in roughly
one million housing estates being built in Sweden during the 1960s and 1970s. The renovations are
associated with displacement as many tenants face a mounting burden of rental costs (Baeten et al.,
2016; Boverket, 2014). At the same time, the number of evictions decreased over the past decades
(Swedish Enforcement Agency [Kronofogden], 2018), the root of this development being uncertain. A
prerequisite for coming under threat of eviction is, however, to have a rental agreement. At present there
is a shortage of rental housing, particularly in the urban areas, and a rental agreement may be further out
of reach for unemployed or low-income households.
The eviction process is similar to most other European countries, but the process is more prompt in
Sweden (Djankov et al., 2003). The most vital steps in the eviction process are a notice to quit from the
landlord, followed by the summary or court procedure, and then an execution of the eviction ruling. These
procedures are the same whether the estate is owned by a private actor or by the municipality. In 2009, only
around 6% of the cases registered with the Enforcement Agency resulted in an eviction, that is, a dis-
placement executed by a bailiff. However, this formal definition of eviction is believed to underestimate
the actual number of forced relocations, since some households are assumed to leave their homes earlier in
the process, whereupon it is never registered as an eviction (Kjellbom, 2014). In the case of indebted
homeowners, there is a judicial foreclosure of real estate and if the Enforcement Agency executes the
displacement, it is also registered as an eviction. However, few foreclosures appear in the eviction registers
as most homeowners move by own force (for an extensive description of the eviction process, see von
Otter et al., 2017).

The educational system. All youths in Sweden who have completed compulsory school are entitled to enroll
in education at the upper secondary level. The compulsory education is nine years and it is generally
completed by the age of 16. Moving on to the upper secondary level is free of charge and accessible for
youths between 16 and 20 years. However, it is still possible to undergo upper secondary adult education
after the age of 20. Although not mandatory, 98% enroll in upper secondary school, which consists of
national programs, either vocational or preparatory for university, and an individual program aimed at
students who do not qualify for the national programmes.1 Of those who enrolled in upper secondary
Kahlmeter: Does housing instability matter for youths’ educational attainment? 5

school in 2009, 69% graduated within three years, the expected time of duration (Swedish National
Agency for Education [Skolverket], 2020).

Method
Data and analytical sample
The study uses comprehensive data from the Dynamics of Evictions in Sweden (DEVS) database, which
contains all eviction cases in Sweden for the period 2009–2012. The DEVS database is linked with several
national administrative registers, such as Statistics Sweden’s (SCB, by Swedish acronym) Longitudinal
Integration Database for Health Insurance and Labour Market Studies (1990–2013); SCB’s Geography
and Domestic Residential Mobility Database (1990–2013); student registers from the National Agency for
Education (1987–2013); the Swedish National Council for Crime Prevention’s data on criminal convic-
tions (1990–2013); and the National Board of Health and Welfare’s patient register (1990–2013). The
database also includes a 10% representative sample of the adult population (age 16þ in December 2008,
excluding those found in the eviction data) as a reference population. Information about other household
members was also collected for all individuals. The full database contains 3,200,000 individuals, of which
youths in upper secondary school were drawn, to analyze the association between housing instability and
educational attainment. The analytical sample thus consists of children in households either in the eviction
data or in the reference population. The inclusion criteria was defined as youths enrolled in the first year of
upper secondary school in 2009 and who had not moved away from home to start a household of their own
at that time. Those who left school before they experienced housing instability were not included to ensure
that there was no reversed order of events. Listwise deletion was applied to handle missing values. In total,
2084 observations (4.9%) were omitted from the analyses due to missing values, resulting in a sample of
40,622 individuals.
Information on sex, year of birth, and country of birth was retrieved from the Total Population
Register. Data were linked through use of the personal ten-digit identification number assigned to all
individuals at birth or upon immigration (PIN). A random reference number then replaced the PIN and
all data were analyzed anonymously.2

Exposure: Housing instability


The rationale for operationalizing housing instability as various types of exposure is that the experiences
of different residential events are expected to vary in several respects, not least since moves following a
ruling from the Enforcement Agency, by definition, are driven by imperative legal reasons and not by
opportunity seeking. The exposed groups are mutually exclusive and consist of individuals who expe-
rienced (a) single relocation (moving once during upper secondary school, without involvement of the
Enforcement Agency, n ¼ 7,280); (b) repeated relocations (moving at least twice during upper second-
ary school, without involvement of the Enforcement Agency, n ¼ 2,195); (c) imminent threat of eviction
(eviction threat was registered with the Enforcement Agency, but was not followed by any relocation,
n ¼ 1,182); and (d) forced relocation (eviction threat was registered with the Enforcement Agency, and
the youth moved, n ¼ 783). The unexposed comparison group consist of the youths who did not
experience any of the residential events described above (n ¼ 29,182). For the analysis of distance, the
sub-sample exposed to single relocation was used.
Relocation was measured using information on change of address between years. This implies that
moves from all types of housing are considered, while the data do not, however, reveal the core reason why
the relocation came about. The categories “single” and “repeated” relocation were, however, not forced in
the sense that the enforcement authorities were not involved, but individuals in these categories could still
have moved more or less of their own accord. As for the forced relocation category, the present study does
not distinguish displacements executed by a bailiff from cases where the individual move before the
6 Acta Sociologica XX(X)

eviction ruling is executed. The rationale for this is that the formal definition of eviction, in which a bailiff
executes the displacement, is believed to underestimate the actual number of forced relocations. Hence,
the forced relocation category could include individuals who have experienced different degrees of
coercion related to the displacement. The eviction threat category, on the other hand, contains only
individuals exposed to threat of eviction, but who did not experience any relocation whatsoever.

Outcome
The dependent variable is dichotomous, measuring whether or not an individual has graduated from upper
secondary school within four years, thus within one year from the expected time of graduation. Graduation
was coded as a noncase if the individual was either not present in the graduation data or had attained zero
grades. Those still enrolled in upper secondary school in autumn 2013 were not included in the sample
since data on graduation were not available after the end of 2013.

Covariates
The impact of housing instability on the probability of attaining an upper secondary degree is estimated by
means of IPWRA. The IPWRA estimator combines an outcome model with a selection model for pre-
dicting assignment to exposure (see below). In the selection model, the propensity of exposure to the
different residential events was conditioned on a set of covariates, guided by previous research. Since
eviction threats are directed toward the parent and not, in legal terms, toward the youth, and moves are
likely driven by decisions made by adult members of the household, the covariates in the selection model
are primarily related to parental characteristics. Those are: highest parental educational level (lower
secondary, upper secondary, or tertiary/doctoral); parental psychiatric care (at least one parent received
inpatient psychiatric care, including alcohol and substance abuse, 1990–2008); parental criminal convic-
tion (at least one parent was convicted 1990–2008, comprising a wide range of convictions); parental
receipt of means-tested social assistance (parents received social assistance in 2008); household dispo-
sable income (equivalized by household size); immigration (youth born outside Sweden or both parents
born outside Sweden); family type (two-parent, step-parent, or single-parent household); number of
moves during a three-year period prior to the youth starting upper secondary school (0, 1, 2þ); and type
of municipality of residence (see Table 1). The classification of municipalities was made according to the
Swedish Association of Local Authorities. The smaller categories—commuter/sparsely populated/man-
ufacturing municipalities—were merged. For parental education, the highest level of education was used
in case of a two-parent family, while in the case of a single-parent family, the education of the co-residing
parent was used.
In the outcome model, covariates more directly related to the youth and early school leaving were
added: sex, age, youth’s criminal convictions (0, 1, 2þ), grades from lower secondary school (> 160 merit
points, indicating that the student has passed all courses) and youth’s psychiatric care (received inpatient
psychiatric care, including alcohol and substance abuse, at some point 1990–2008).

Estimating the impact of housing instability


The study draws on a counterfactual approach (Rubin, 1974), in which a central aspect is what would have
happened to a group of individuals who received a treatment, had they not been treated. The estimate is
obtained by comparing the outcomes of the treated individuals to the potential outcomes of the same group
when not treated. In the present study, treatment refers to exposure to the residential events and the terms
“treatment” and “exposure” are used interchangeably.
Since, outside randomized trials, assignment to exposure is not random, the counterfactual outcome is
estimated by adopting a modeling strategy that addresses the selection to exposure. The predicted prob-
ability of being exposed to the different residential events was obtained by fitting a binary logistic model
Table 1. Sample characteristics (prevalence and means); N ¼ 40,622.
No Single Repeated Eviction threat, Forced
Variable Definition Range exposure relocation relocations no relocation relocation

Sex Girl 0–1 46.4 49.7 54.8 46.4 50.1


Agea Age in 2009 16–20 16.26 (0.59) 16.34 (0.64) 16.38 (0.65) 16.37 (0.65) 16.45 (0.70)
Educational attainment Graduated from upper secondary school within a year 0–1 76.7 67.5 57.6 56.3 46.6
of expected time of graduation.
Previous relocations, 0 moves 0–1 83.0 63.5 50.6 75.0 49.6
2006–2008 1 move 0–1 14.3 28.2 34.1 20.1 34.0
2þ moves 0–1 2.7 8.3 15.3 5.0 16.5
Grades from lower Merits > 160 in 9th grade 0–1 80.3 71.7 62.8 57.6 48.4
secondary school
Criminal convictions 0 0–1 92.9 90.5 86.9 87.6 83.8
1 0–1 5.6 7.6 9.8 10.1 12.6
2þ 0–1 1.5 1.9 3.3 2.3 3.6
Psychiatric care Received psychiatric care at some point 1990–2008 0–1 2.0 2.9 4.1 3.2 3.4
Immigration Born outside Sweden or both parents born outside Sweden 0–1 15.1 17.7 14.1 34.8 22.3
Family type Two-parent household 0–1 62.5 30.7 15.7 38.2 19.4
Step-parent family 0–1 11.1 16.9 19.3 12.6 14.0
Single-parent family 0–1 25.8 50.6 60.9 48.9 66.2
Other 0–1 0.6 1.8 4.1 0.3 0.4
Highest parental Lower secondary or less 0–1 8.4 11.6 14.0 22.3 23.8
educational level Upper secondary 0–1 47.4 51.5 53.1 56.1 61.4
Tertiary or doctoral 0–1 44.3 36.9 32.8 21.6 14.8
Parental criminal At least one parent convicted 1990–2008 0–1 19.9 23.2 27.8 40.1 46.9
conviction
Parental psychiatric care At least one parent received psychiatric care 1990–2008 0–1 5.7 8.0 11.4 13.8 20.6
Household disposable First quartile 0–1 18.3 29.8 36.1 47.0 52.7
incomeb Second quartile 0–1 23.2 27.5 26.6 31.6 29.8
Third quartile 0–1 28.0 18.6 14.9 14.0 11.5
Fourth quartile 0–1 30.5 24.2 22.4 7.4 6.0
Parental social assistance Parent received social assistance in 2008 0–1 6.2 11.3 15.2 31.3 35.0
Type of municipality Metropolitan 0–1 13.3 12.8 8.7 18.4 8.8
of residence 2008 Suburban 0–1 17.7 15.3 11.6 13.6 9.5
Large city 0–1 26.4 28.5 29.3 33.1 30.4
Commuter/sparsely populated/manufacturing 0–1 17.6 17.6 19.6 13.7 22.3
Other > 25,000 inhabitants 0–1 14.0 15.7 15.9 14.3 19.2
Other < 25,000 inhabitants 0–1 11.0 10.2 14.9 6.9 9.8
N 29,182 7,280 2,195 1,182 783

a. Mean age (SD).

7
b. Equivalized by household size.
8 Acta Sociologica XX(X)

for each exposure (online supplementary Table S1). The rationale for estimating binary models for each
exposure respectively rather than a multinomial model is, first, that this allows the average treatment effect
on the treated (ATT) to be estimated for each exposure group respectively and, second, that it produces
better balance, that is, weighted samples that are more similar.3 The probabilities from the selection
models were then used to weight the observations, the weights being the inverse of the probability that
an individual receives a treatment (Funk et al., 2011). Observations in the exposed groups are weighted by
1/p so that weights are large when the probability of being exposed is low. The unexposed observations are
weighted by 1/(1-p) so that weights are large when the probability of not being exposed is low. The
weighted samples were then used in the binary logistic outcome regressions (online supplementary Table
S2) and the IPWRA estimator is thus doubly robust, since it models both the outcome and the exposure to
account for the nonrandom assignment (Wooldridge, 2010). The outcome estimate is the ATT, which is
the average difference between the outcome for the exposed group and the outcome for the unexposed
group in the weighted samples (Austin, 2011). In the present study, this is represented by the average
difference in the probability of graduating from upper secondary school.

Key assumptions when estimating treatment effects


The validity of the results rests on a set of assumptions. A key qualification is that there is balance across
the weighted groups so that there are no systematic differences between the exposed and the unexposed
groups in the weighted samples. This was investigated by means of over-identification tests and by
comparison of standardized differences and variance ratios in the weighted samples. Another crucial
premise is the conditional independence assumption (CIA), implying that no selection on unobserved
covariates would bias the estimated impact of housing instability. This was assessed by estimating the
correlation between the residuals of the selection model and the outcome model (Wooldridge, 2010).
Finally, a central assumption is that of overlapping, implying that the predicted probabilities should have
similar patterns for the exposed group and the unexposed group. When estimating ATT, the overlap
assumption requires that each individual in the exposed sub-populations has a positive probability of
being unexposed. This was explored by plotting the estimated densities of the propensity to be unexposed.
For the sake of brevity, results from these analyses are presented in the online supplement.

Results
Descriptive results
In the unexposed group, around 77% of the individuals graduated within four years, compared with 68% in
the group exposed to single relocation and 58% in the repeated relocations group. Of those exposed to
eviction threat, 56% graduated, while in the forced relocation group, roughly 47% graduated. All the
exposed groups were disadvantaged in several respects prior to starting upper secondary school (Table 1).
For example, they were more likely to have faced criminal convictions and to have poor grades from lower
secondary school, compared with the unexposed group, and their parents generally have attained lower
educational levels and have more often received psychiatric care. However, the individuals exposed to
forced relocation are a particularly vulnerable group. To point out some examples, they are substantially
more likely to have parents who received means-tested social assistance, to have a parent who received
psychiatric care and to live in a single-parent household. This underscores the importance of addressing
these differences in the modeling of the selection to exposure.

Impact of housing instability


Table 2 shows the results from the IPWRA analyses. It reports both crude and adjusted differences (ATT)
in the probability of having graduated from upper secondary school for each type of residential event. The
crude risk differences range from roughly 9 percentage points for the group who moved once to 30
Kahlmeter: Does housing instability matter for youths’ educational attainment? 9

Table 2. Estimates from inverse probability weighting with regression adjustment (IPWRA). Average impact
of residential instability on graduation from upper secondary school; N ¼ 40,622.

Potential outcome mean (POM)


Average treatment effect
Exposure Analysis Exposed Unexposed on the treated (ATT) 95% CI

Single Crude 0.675 0.767 0.092 0.104; 0.080


relocationa IPWRA 0.675 0.699 0.024 0.035; 0.013

Repeated Crude 0.576 0.767 0.191 0.212; 0.170


relocationsb IPWRA 0.576 0.644 0.068 0.089; 0.046

Eviction threat, Crude 0.563 0.767 0.205 0.233; 0.176


no relocationc IPWRA 0.563 0.616 0.053 0.080; 0.026

Forced Crude 0.466 0.767 0.301 0.336; 0.266


relocationd IPWRA 0.466 0.559 0.093 0.129; 0.057

Notes: CI ¼ 95% confidence interval.


a. Exposed: raw n ¼ 7,280; weighted n ¼ 18,193. Unexposed: raw n ¼ 29,182; weighted n ¼ 18,269.
b. Exposed: raw n ¼ 2,195; weighted n ¼ 15,653. Unexposed: raw n ¼ 29,182; weighted n ¼ 15,724.
c. Exposed: raw n ¼ 1,182; weighted n ¼ 15,111. Unexposed: raw n ¼ 29,182; weighted n ¼ 15,254.
d. Exposed: raw n ¼ 783; weighted n ¼ 14,887. Unexposed: raw n ¼ 29,182; weighted n ¼ 15,078.

percentage points for the group exposed to forced relocation. The second row for each exposure (Table 2)
shows that, when controlling for background factors, the crude risk differences are considerably reduced
and the observed differences can largely be attributed to selection.4 There remain statistically significant
effects, however. Moving once during upper secondary school is associated with a small (2.4 percentage
points) difference in the probability of graduating from upper secondary school. Exposure to repeated
relocations is associated with a risk difference of 7 percentage points and for threat of eviction with no
relocation the adjusted risk difference is 5 percentage points. Forced relocation is associated with a
difference of 9 percentage points in the probability of graduating from upper secondary school, as
compared with the unexposed group. This supports the hypothesis that there is an independent impact
of moving, but the risk difference is larger if exposed to forced relocation or repeated relocations. There is
also, as hypothesized, a negative association between threat of eviction and educational attainment, even
when the threat is not followed by an actual relocation.
In the next step, the significance of distance was investigated. This analysis was performed separately
on the sub-sample who moved once during the observed period, which was divided into two groups: short-
distance relocation ( 40 km, n ¼ 6,387) and long-distance relocation (> 40 km, n ¼ 893). The rationale
for restricting this analysis to the single relocation sub-sample is that the other exposed groups may have
moved a short distance at one point and a longer distance at another point, making it difficult to disentangle
the significance of distance, and those exposed to threat of eviction have not moved at all. The thresholds
for short and long-distance moves were set so that the long-distance category represents a substantial fetch
while keeping the sizes of the groups in mind, since a vast majority of individuals only move a short
distance.5 Table 3 reports both the crude and weighted differences from the IPWRA analysis. When
considering distance, there is a crude risk difference of around 8 percentage points for short-distance
relocation and of 16 percentage points for long-distance relocation. In the weighted analysis, when
accounting for selection, the risk difference for short-distance relocation is very small and barely statis-
tically significant. Long-distance relocation, on the other hand, is still associated with a 7 percentage point
decrease in the probability of graduating from upper secondary school, suggesting that long-distance
relocation poses a bigger strain on youths.
10 Acta Sociologica XX(X)

Table 3. Estimates from inverse probability weighting with regression adjustment (IPWRA). Average impact
of single relocations by distance on graduation from upper secondary school; N ¼ 36,462.

Potential outcome mean (POM)


Average treatment effect
Exposure Analysis Exposed Unexposed on the treated (ATT) 95% CI

Relocation  40 kma Crude 0.685 0.767 0.082 0.095; 0.070


IPWRA 0.685 0.704 0.019 0.030; 0.007
Relocation > 40 kmb Crude 0.605 0.767 0.163 0.195; 0.130
IPWRA 0.605 0.670 0.065 0.096; 0.034

Notes: CI ¼ 95% confidence interval.


a. Exposed: raw n ¼ 6,387; weighted n ¼ 17,754. Unexposed: raw n ¼ 29,182; weighted n ¼ 17,815.
b. Exposed: raw n ¼ 893; weighted n ¼ 14,996. Unexposed: raw n ¼ 29,182; weighted n ¼ 15,079.

Validity of the estimated effects


Covariate balance was assessed through the comparison of standardized differences and variance ratios
in the weighted samples and through over-identification tests. The null hypothesis posed for the over-
identification test is that there is no difference in the covariate distribution between the exposed and the
unexposed groups. These tests showed that the null hypothesis can be rejected (p < 0.05), indicating that
the samples are not balanced on all observed covariates. However, the weighted standardized differences
are all close to zero (online supplementary Table S3) and no variable has a standardized difference that
exceeds the recommended level of 0.1 (Austin, 2009). Worth noting is that, in large samples, even small
differences in the covariate distribution may be statistically significant.
The overlap assumption was assessed by plotting the estimated densities of the propensity to be
unexposed (online supplementary Figures S1–S4). For all the exposed sub-populations, the masses of the
estimated densities are in regions where they overlap, indicating that the overlap assumption is not
violated.
Finally, to assess the necessary condition for CIA, the correlation between the residuals of the selection
model and the outcome model was estimated, using a control-function approach of endogenous treatment-
effect estimators and then running a Wald test (Wooldridge, 2010). The Wald tests were not statistically
significant for the exposures single relocation, repeated relocations or forced relocation, indicating that
there is no correlation between the residuals in the selection model and the outcome model (single
relocation: p ¼ 0.16, repeated relocations: p ¼ 0.26, forced relocation: p ¼ 0.56). For the exposure threat
of eviction with no relocation, however, the Wald test showed a correlation between the residuals
(p ¼ 0.0002), indicating that the estimates for threat of eviction are less robust. A possible explanation
for this is that it was not possible to control for eviction threats prior to 2009. Whereas previous relocations
largely predict the assignment to the other exposures, it does so to a lesser extent for eviction threat, this
being a less mobile group (see Table 1 and supplementary Table S1).

Discussion
Leaving school without qualifications in a labor market with shrinking opportunities for those without
education poses a major disadvantage (e.g. Bäckman, 2017; Townsend et al., 2007), which underscores
educational attainment as an indicator of success in young adulthood. Using comprehensive longitudinal
Swedish register data, this study investigated if different types of housing instability matter for the
probability of graduating from upper secondary school within one year of the expected time of graduation.
With reference to the FSM, it was hypothesized that eviction threat, without any relocation, has a negative
impact on educational attainment. With reference to social capital theory, it was further surmised that all
types of moves are negatively associated with educational attainment, but that the impact will be greater if
Kahlmeter: Does housing instability matter for youths’ educational attainment? 11

the relocation is forced by authorities or in case of repeated relocations. Finally, it was hypothesized that
long-distance relocations will have a greater impact than short-distance relocations, since social capital is
largely connected to the spatial context.
The groups exposed to the different residential events were all disadvantaged in several respects
compared with the unexposed group. For example, the exposed groups were more likely to have faced
criminal convictions, to have poor grades from lower secondary school, and to have parents with lower
levels of education. This reveals that youths facing housing instability are a vulnerable population, which
may motivate concern in its own right. This was further substantiated by findings indicating that all the
residential events were negatively associated with educational attainment. However, when accounting for
selection, the risk differences are not as pronounced as suggested in some other studies (see e.g. Metzger
et al., 2015) and regarding the exposure to eviction threat, the estimates are uncertain. Moreover, the
finding that single short-distance relocation has a negligible impact on educational attainment to some
extent challenges much of the literature on housing mobility.
Taken together, the results of the present study suggest that the instability and stress following forced
relocations, repeated relocations and long-distance relocations are of particular significance for the link
between housing instability and educational outcomes. This implies that both the notion of loss of social
capital and the FSM are relevant for understanding the mechanisms at play. In the cases of long-distance
moves, the link to poor educational achievement can be understood as a response to greater disruptions
in important relationships due to changed spatial context and difficulties in sustaining pro-social net-
works. The impact of repeated relocations can be linked to social capital theory, both through the loss of
social resources and through reduced opportunities to establish new social ties when mobility is high.
The association between forced relocations and poor educational attainment can be seen as the result of a
combination of psychological and economic pressure in the family and disruptions in important relation-
ships when changing residential context.
The present study has the advantage of national and longitudinal data, providing a good opportunity to
analyze housing instability. The study is not without limitations, however. One manifest weakness is the
relatively short follow-up period. The available data contain information on eviction cases between 2009
and 2012 and on graduation until 2013, with the consequence that the study focuses on residential events
experienced during upper secondary school and that no conclusions on long-term effects can be drawn. In
addition, data did not contain information on which school the youths attended and, accordingly, change of
school could not be analyzed. Though this may be picked up to some extent in the analysis on distance, it is
an interesting query for follow-up studies. Another limitation is that there is only information about moves
between years and, thus, the number of moves might be underestimated in some cases.
As with most register data, generally not assembled for research purposes, the information is limited.
Since a randomized controlled trial would not be viable to investigate housing instability, the IPWRA
estimator was used to compensate for the nonrandom assignment to the exposure. However, as in all
observational studies, it cannot be ruled out that there still exists unobserved heterogeneity. For example,
the data do not hold information about the quality of parent–child relationships or about peer networks.
Likewise, the data do not reveal the core reason for moving and people could move for a number of reasons
and more or less of own accord. On the other hand, the data are a national sample, suffering less from
nonresponse problems compared to survey data and include objective measures on a range of covariates,
while also allowing more detailed operationalizations of housing instability than has been done in previous
research. Finally, it was beyond the scope of the present study to analyze whether the different types of
housing events were associated with upward or downward moves in terms of neighborhood quality, which
could also inspire future research questions. Other suggestions for future studies are to investigate effects
on educational achievement post the upper secondary education, while also exploring if resources attained
in young adulthood mediates the impact of housing instability in youth.
The limitations notwithstanding, the present study provides further evidence that youths experiencing
housing instability are a disadvantaged group, particularly those facing repeated relocations or when
coercion is involved. The findings add to previous knowledge that having a stable home is of great
12 Acta Sociologica XX(X)

importance for wellbeing and suggest that severe housing instability should be regarded as a disruptive life
event associated with elevated risks of poor educational attainment, and this applies even in the context of
a relatively comprehensive welfare state. Moreover, the results bring to attention the significance of
considering various residential events when investigating housing instability, since we might otherwise
overrate the importance of single and short-distance relocations. Key implications of the findings are the
significance of a housing market where moves are not driven by economic, social, or legal imperatives,
and the need for measures to decrease the incidence of repeated and involuntary relocations. This also
underscores the importance of combating the underlying economic and social disadvantages faced by
mobile youths. An implication is the necessity of a functioning cooperation between the social services
and landlords, and established procedures to address back rent at an early stage in order to preclude forced
relocations following the accumulation of rent arrears. Finally, the results call for support directed toward
mobile youths, in particular to those who are highly mobile or who face residential events where coercion
is involved. Teachers and other school professionals need to be attentive to the fact that these youths may
be at risk of poor educational attainment.

Acknowledgements
The author wishes to thank Olof Bäckman, Susanne Alm and Lars Brännström for methodological advice, and Sten-
Åke Stenberg for providing the opportunity to work within the DEVS project. Four anonymous reviewers are also
gratefully acknowledged, for their valuable comments on an earlier version of the article.

Funding
The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this
article: The work was supported by Swedish Research Council for Health, Working Life and Welfare (Forte). Grant
number 2013–0452.

ORCID iD
Anna Kahlmeter https://orcid.org/0000-0002-7078-3421

Notes
1. Changes to the school system were made in 2011 and the description refers to the system in force
2009, the year the individuals in the sample enrolled in upper secondary school.
2. The research project was approved by the ethics committee in the Stockholm region (no. 2014/24–31/5)
3. When estimating a multinomial treatment model and alternating the treatment level, the estimates
are virtually the same as when estimating binary logistic selection models for each exposure
respectively.
4. The reason that the weighted potential outcome means for the unexposed group differ between the
four IPWRA analyses is that a binary logistic regression was conducted for each exposure respec-
tively and the weights from the selection models are thus based on each separate binary analysis.
5. Different thresholds, ranging between 30 and 50 km, were applied and the estimates are similar. Also,
separate analyses for rural and metropolitan areas were tried, but without any substantial differences
in the point estimates.

References
Alm S (2018) Isolating the effect of eviction on criminal convictions. Acta Sociologica 61(3): 263–282.
Astone NM and McLanahan SS (1994) Family structure, residential mobility, and school dropout: A
research note. Demography 31(4): 575–584.
Kahlmeter: Does housing instability matter for youths’ educational attainment? 13

Austin PC (2009) Using the standardized difference to compare the prevalence of a binary variable
between two groups in observational research. Communications in Statistics – Simulation and Com-
putation 38(6): 1228–1234.
Austin PC (2011) An introduction to propensity score methods for reducing the effects of confounding in
observational studies. Multivariate Behavioral Research 46(3): 399–424.
Bäckman O (2017) High school dropout, resource attainment, and criminal convictions. Journal of
Research in Crime and Delinquency 54(5): 715–749.
Baeten G, Westin S, Pull E, et al. (2016) Pressure and violence: Housing renovation and displacement in
Sweden. Environment and Planning A: Economy and Space 49(3): 631–651.
Berg L and Brännström L (2018) Evicted children and subsequent placement in out-of-home care: A
cohort study. PLoS ONE 13(4): e0195295.
Boverket (2014) Flyttmönster till följd av omfattande renoveringar [Displacement patterns due to
extensive renovations]. Karlskrona: Swedish Board of Housing, Building and Planning.
Brunello G and Paola MD (2014) The costs of early school leaving in Europe. IZA Journal of Labor
Policy 3(22). Available at: http://www.izajolp.com/content/3/1/22
Chen J (2013) Housing tenure, residential mobility and adolescents’ education achievement: Evidence
from Sweden. The Annals of Regional Science 50(1): 275–294.
Coleman JS (1988) Social capital in the creation of human capital. American Journal of Sociology 94:
95–120.
Cutuli JJ, Desjardins CD, Herbers JE, et al. (2013) Academic achievement trajectories of homeless and
highly mobile students: Resilience in the context of chronic and acute risk. Child Development 84(3):
841–857.
Desmond M, Gershenson C and Kiviat B (2015) Forced relocation and residential instability among
urban renters. Social Service Review 89(2): 227–262.
Djankov S, La Porta R, Lopez-de-Silanes F, et al. (2003) Courts. The Quarterly Journal of Economics
118(2): 453–517.
European Commission (2020) Education and training. Available at: https://ec.europa.eu/education/pol
icies/european-policy-cooperation/et2020-framework_en (accessed 15 May 2020).
Funk MJ, Westreich D, Wiesen C, et al. (2011) Doubly robust estimation of causal effects. American
Journal of Epidemiology 173(7): 761–767.
Ginsburg C, Richter LM, Fleisch B, et al. (2011) An analysis of associations between residential and
school mobility and educational outcomes in South African urban children: The birth to twenty
cohort. International Journal of Educational Development 31(3): 213–222.
Grander M (2018) For the benefit of everyone?: Explaining the Significance of Swedish public housing
for urban housing inequality. PhD Thesis, Malmö University, Sweden.
Haelermans C and De Witte K (2015) Does residential mobility improve educational outcomes? Evi-
dence from the Netherlands. Social Science Research 52: 351–369.
Hagan J, MacMillan R and Wheaton B (1996) New kid in town: Social capital and the life course effects
of family migration on children. American Sociological Review 61(3): 368–385.
Hango DW (2006) The long-term effect of childhood residential mobility on educational attainment.
Sociological Quarterly 47(4): 631–664.
Haveman R, Wolfe B and Spaulding J (1991) Childhood events and circumstances influencing high
school completion. Demography 28(1): 133–157.
Haynie DL, South SJ and Bose S (2006) The company you keep: Adolescent mobility and peer behavior.
Sociological Inquiry 76(3): 397–426.
14 Acta Sociologica XX(X)

Hofferth SL, Boisjoly J and Duncan GJ (1998) Parents’ extrafamilial resources and children’s school
attainment. Sociology of Education 71(3): 246–268.
Jelleyman T and Spencer N (2008) Residential mobility in childhood and health outcomes: A systematic
review. Journal of Epidemiology and Community Health 62(7): 584–592.
Kahlmeter A, Bäckman O and Brännström L (2017) Housing evictions and economic hardship. A
prospective study. European Sociological Review 34(1): 106–119.
Kemeny J (1995) From Public Housing to the Social Market: Rental Policy Strategies in Comparative
Perspective. London: Routledge.
Kjellbom P (2014) Påtvingad avflyttning från bostad. En rättssociologisk studie av socialtjänstens roll i
teori och praktik. PhD Thesis, Department of Social Work, Stockholm University, Sweden.
Magnusson L and Turner B (2008) Municipal housing companies in Sweden – social by default.
Housing, Theory and Society 25(4): 275–296.
Masarik AS and Conger RD (2017) Stress and child development: A review of the Family Stress Model.
Current Opinion in Psychology 13: 85–90.
Metzger MW, Fowler PJ, Anderson CL, et al. (2015) Residential mobility during adolescence: Do even
“upward” moves predict dropout risk? Social Science Research 53: 218–230.
OECD (2017) Education policy outlook: Sweden. Report OECD, Education Policy Outlook series.
Available at: http://www.oecd.org/education/Education-Policy-Outlook-Country-Profile-Sweden.pdf
(accessed 15 May 2020).
Perkins KL (2017) Reconsidering residential mobility: Differential effects on child wellbeing by race
and ethnicity. Social Science Research 63: 124–137.
Pettit B and McLanahan S (2003) Residential mobility and children’s social capital: Evidence from an
experiment. Social Science Quarterly 84(3): 632–649.
Pribesh S and Downey DB (1999) Why are residential and school moves associated with poor school
performance? Demography 36(4): 521–534.
Rubin DB (1974) Estimating causal effects of treatments in randomized and nonrandomized studies.
Journal of Educational Psychology 66(5): 688–701.
Scanlon E and Devine K (2001) Residential mobility and youth well-being: Research, policy, and
practice issues. Journal of Sociology and Social Welfare 28(1): 119–138.
South SJ and Haynie DL (2004) Friendship networks of mobile adolescents. Social Forces 83(1):
315–350.
South SJ, Haynie DL and Bose S (2007) Student mobility and school dropout. Social Science Research
36(1): 68–94.
Swedish Enforcement Agency [Kronofogden] (2018) Avhysningsstatistik 1982–2017. Available at:
https://kronofogden.se/download/18.77420b161614b6f20243592/1519201469392/Avhysning%
201982–2017.pdf. (accessed 5 December 2018).
Swedish National Agency for Education [Skolverket] (2020) Available at: https://www.skolverket.se/
skolutveckling/statistik/sok-statistik-om-forskola-skola-och-vuxenutbildning?sok¼SokC&verk
form¼Gymnasieskolan&omrade¼Betyg%20och%20studieresultat&lasar¼2012%2F13&run¼1
(accessed 15 May 2020).
Teachman JD, Paasch K and Carver K (1996) Social capital and dropping out of school early. Journal of
Marriage and Family 58(3): 773–783.
Townsend L, Flisher AJ and King G (2007) A systematic review of the relationship between high school
dropout and substance use. Clinical Child and Family Psychology Review 10(4): 295–317.
Tsai J and Huang M (2019) Systematic review of psychosocial factors associated with evictions. Health
& Social Care in the Community 27(3): e1–e9.
Kahlmeter: Does housing instability matter for youths’ educational attainment? 15

Vásquez-Vera H, Palència L, Magna I, et al. (2017) The threat of home eviction and its effects on health
through the equity lens: A systematic review. Social Science & Medicine 175: 199–208.
von Otter C, Bäckman O, Stenberg S-Å, et al. (2017) Dynamics of evictions: Results from a Swedish
database. European Journal of Homelessness 11(1): 1–23.
Wooldridge JM (2010) Econometric Analysis of Cross Section and Panel Data, 2nd ed. Cambridge, MA:
MIT Press.

Author biography
Anna Kahlmeter is a PhD candidate in sociology at the Swedish Institute for Social Research, Stock-
holm University. Her research interests primarily concern poverty, inequality of opportunity and pro-
cesses of social exclusion, with a particular focus on youth. Her PhD project investigates stressful life
events during the youth-to-adulthood transition. Prior to her doctoral studies, she has several years of
experience as a social worker practicing mostly with young adults with substance abuse and precarious
housing conditions.

You might also like