Sage Journals: Discover world-class research

Abstract

Millions of individuals in the United States experience eviction each year, with low-income women being particularly at risk. As a result, scholarship has increasingly sought to understand what the implications of eviction are for families. In this article, we build on this work by presenting the first estimates of the impact of eviction on criminal justice involvement for mothers in the U.S. context and examining three pathways that may help to explain these associations. Using longitudinal data from the Fragile Families and Child Wellbeing Study, adjusted estimates suggest that mothers who have been evicted have more than two times higher odds of experiencing criminal justice involvement. When we differentiate by eviction timing, we find that both recent and less recent evictions are associated with criminal justice involvement. Last, we find that eviction indirectly affects criminal justice involvement through future financial hardship and substance use.

Keywords

eviction crime criminal justice involvement mothers

Introduction

Because of stagnating wages, increased rental prices, and inadequate public housing support, the United States is currently experiencing its worst affordable housing crisis in generations (Desmond and Bell 2015; Eggers and Moumen 2010). As a result, it is estimated that millions of individuals each year experience eviction, with low-income women being particularly at risk (Desmond 2016; Desmond and Kimbro 2015). In response to the growing prominence of eviction, scholarship has increasingly sought to understand what the implications of eviction are for families. In general, this scholarship has shown that eviction is not just caused by poverty but that it contributes to disadvantage in a number of ways that include increasing material hardship, reducing neighborhood quality, increasing employment instability, and increasing depression and other negative health indicators (Desmond and Gershenson 2016; Desmond, Gershenson, and Kiviat 2015; Desmond and Kimbro 2015; Desmond and Shollenberger 2015).

In fact, because of the negative consequences of eviction and its high prevalence, Desmond (2016) suggests that the impact of eviction on low-income women of color is quite similar to the impact that mass incarceration has had on men of color. While it is certainly accurate that men are more directly affected by mass incarceration than are women, it is also true that, in recent years, the female prison population has grown at a faster rate than the male prison population (Sawyer 2018; The Sentencing Project 2015). In this study, we draw on General Strain Theory (GST) and Rational Choice Theory (RCT) to argue that eviction and the criminal justice involvement of mothers are phenomena that may be connected.

In this article, we hone in on the impact of eviction on criminal justice involvement for urban mothers for two reasons: (1) This population is particularly likely to experience eviction (Desmond and Kimbro 2015), and 2) the criminal justice involvement of mothers also negatively affects their children (Dallaire, Ciccone, and Wilson 2010; Hagan and Foster 2012; Huebner and Gustafson 2007). Specifically, we seek to answer three questions: (1) Does experiencing eviction increase the likelihood that mothers experience criminal justice involvement? (2) Are both recent and less recent evictions consequential for criminal justice involvement? and (3) Does future financial hardship, depression, and substance use help to explain associations between eviction and criminal justice involvement?

To answer these questions, we draw on longitudinal data from the Fragile Families and Child Wellbeing Study (Fragile Families) and logistic regression models using the Karlson, Holm, and Breen (KHB) method (Karlson, Holm, and Breen 2012). Our results lead to two main conclusions: First, mothers who have been evicted are more likely to experience criminal justice involvement, even after including a wide range of controls that capture preexisting disadvantage, than mothers who have not been evicted. This finding holds for both recent and less recent evictions. Second, we find that eviction indirectly affects criminal justice involvement through future financial hardship and substance use.

Background

GST, Eviction, and Criminal Justice Involvement

GST is a criminological theory that builds on earlier iterations of strain theory and is unique in its emphasis on negative emotions and coping (Agnew 2006; Brezina 2017; Merton 1938). GST asserts that there are three types of strain that lead to criminal behavior: (1) failure (either anticipated or actual) to achieve desired goals, (2) removal of something an individual positively values, and (3) exposure to negative stimuli (Agnew 1992; Slocum, Simpson, and Smith 2005). From this perspective, exposure to any of these three types of strain produces negative emotions such as anger, anxiety, depression, and resentment (Agnew 1992; Slocum et al. 2005). In turn, these negative emotions often lead to criminal behavior or substance use as a means of reducing the strain or the negative emotions that accompany the strain (Agnew 2001; Brezina 2017). In cases where strain directly results in substance use, strain may lead indirectly to criminal behavior because both illicit drug use and heavy alcohol use are associated with future criminality (Bennett, Holloway, and Farrington 2008; Jennings et al. 2015).

Although Agnew did not discuss eviction specifically, he did argue that homelessness is a type of strain that is particularly likely to result in criminal behavior, suggesting that housing issues, more generally, may be a source of strain that greatly influences criminal offending (Agnew 2001; Brezina 2017). We anticipate that eviction is likely to cause strain in two ways. First, it will cause strain by removing something an individual positively values (i.e., taking away access to his or her home). Second, eviction also exposes individuals to negative stimuli, since it has been found to increase material hardship and lead to forced dislocation to more disadvantaged neighborhoods (Desmond and Gershenson 2016; Desmond and Kimbro 2015; Desmond and Shollenberger 2015; Desmond et al. 2015).

In response to this negative strain, eviction is likely to result in negative emotions, such as anger, resentment, anxiety, and depression (Agnew 2001; Brezina 2017; Desmond and Kimbro 2015). The negative emotions produced by eviction may result directly in mothers’ engaging in criminal behavior or may result in illicit drug and heavy alcohol use, which in turn will increase the likelihood that mothers engage in criminal behavior (Bennett et al. 2008; Brezina 2017; Jennings et al. 2015). Moreover, eviction may result in criminal behavior by leading to future economic strain (Desmond and Kimbro 2015); research suggests that when strain accumulates it has a particularly large impact on offending (Agnew 2001; Brezina 2017; Eitle and Turner 2002). By leading to an increased likelihood of engaging in criminal behavior, we anticipate that eviction also will increase the likelihood that a mother will become involved in the criminal justice system.

RCT, Eviction, and Criminal Justice Involvement

RCT is based on the view that individuals are rational actors. RCT argues that individuals make a rational choice about whether to engage in a criminal act by weighing the costs and benefits of committing the crime. When the benefits outweigh the costs, an individual will engage in criminal behavior to maximize utility; when the costs are greater than the benefits, the individual will not commit the criminal act (Becker 1968; Cornish and Clarke 1986; Hayward 2007; Loughran et al. 2016; Nagin and Paternoster 1993; Paternoster, Jaynes, and Wilson 2017). In Becker’s (1968) model, three inputs shape perceptions of costs and benefits: (1) the probability that an individual will get caught if he or she commits the criminal act, (2) the severity of the punishment if the individual is caught committing the criminal act, and (3) the utility gained if an individual commits a criminal act and is not caught.

From the standpoint of RCT, eviction has the potential to lead to criminal behavior by increasing the benefits (without affecting the costs) associated with criminal behavior (Becker 1968; Loughran et al. 2016). Specifically, research suggests that experiencing eviction is associated with an increase in financial hardship (Desmond and Kimbro 2015). In many instances, in addition to losing housing, the experience of eviction leaves an individual owing significant fees to his or her landlord and also results in the loss of material possessions (Desmond 2012). In cases wherein eviction produces material hardship, the benefits of financial crime, such as robbery and theft, are likely to increase (i.e., committing the crime may enable an individual to pay off debt owed to a landlord). As a result, eviction may make individuals more likely to engage in financial crime, which in turn may increase the likelihood that an individual will experience criminal justice involvement.

Existing Empirical Scholarship on Eviction and Criminal Justice Involvement

No studies have explored the implications of eviction for criminal justice involvement in the United States, and none have explored what mechanisms may explain the association. We are aware of only one study that has explored the effect of eviction on criminal justice involvement in any context. Using data from Sweden, Alm (2018) found that experiencing eviction is associated with an increased risk of criminal justice involvement for both men and women. It is unclear whether the results from Sweden will translate to the U.S. context for a number of reasons. First, eviction is far less common in Sweden than in the United States (Alm 2018; Desmond and Gershenson 2017). This suggests that the effects of eviction may be more prominent in Sweden than in the United States if eviction becomes more normalized in the United States and if the Swedish population who experiences eviction is more select than the U.S. population who experiences eviction (Johnson and Raphael 2012; Raphael and Stoll 2013). Second, Sweden tends to provide much more generous social supports to buffer disadvantage than does the United States, suggesting that the effects of eviction may be worse in the United States than in Sweden (Esping-Andersen 1990; Gottlieb 2017).

Although they do not provide a direct test, studies examining the implications of homelessness and housing instability on criminal justice involvement in the U.S. context provide some additional insight since eviction also constitutes a form of housing hardship. Specifically, scholarship has found that housing instability and homelessness increase the likelihood that men and women will engage in criminal behavior and experience prison and/or jail incarceration (Center for Poverty Solutions 2003; Gowan 2002; Greenberg and Rosenheck 2008; Hagan and McCarthy 1998; Heerde, Hemphill, and Scholes-Balog 2014; Weiser et al. 2009).

The Current Study

In the current study, we build on existing scholarship in three ways: (1) We present the first estimates of the association between maternal eviction and criminal justice involvement in the U.S. context, (2) we examine whether recent evictions and less recent evictions both place mothers at greater risk of criminal justice involvement, and (3) we provide the first estimates of the role played by three possible pathways in mediating the association between eviction and criminal justice involvement. Specifically, we examine the degree to which eviction affects criminal justice involvement by leading to future material hardship (i.e., by leading to the accumulation of strain or by increasing the benefits associated with criminal behavior), depression (a negative emotion), and substance use (a coping mechanism for the negative emotions induced by strain).

Based on the discussion above, we hypothesize the following: (1) Mothers who have been evicted will be more likely to experience criminal justice involvement than mothers who have not been evicted. (2) Mothers who experience eviction between baseline and year 1 and between the year 1 and year 3 survey waves both will be at a greater risk of criminal justice involvement; however, if the associations differ, the effects will be larger for more recent evictions. (3) Future financial hardship, negative emotions (depression), and coping responses to strain (substance use) will mediate the association between eviction and criminal justice involvement.

Data, Measures, and Analytic Strategy

Data

The data for our analyses come from Fragile Families. Fragile Families follows a cohort of 4,898 couples with children who were born between 1998 and 2000 in 20 large U.S. cities. To select families to study, researchers employed a stratified (stratification was based on policy environments and labor market conditions in cities) random sample of all U.S. cities with 200,000 or more people. Specifically, random sampling was conducted in three stages: (1) cities, (2) hospitals within cities, and (3) births within hospitals (Reichman et al. 2001). After the initial interview, follow-up interviews were conducted when the child was one, three, five, and nine years old.¹ Of the mothers who were first interviewed, 90 percent were interviewed again when the focal child was roughly one, 88 percent when the child was three, 87 percent when the child was five, and 76 percent when the child was nine.

Since the initial purpose of Fragile Families was to examine the conditions and capabilities of unmarried parents, by design, the study oversampled nonmarital births (approximately 75 percent of births occurred outside of marriage). As a result, the Fragile Families sample is disproportionately disadvantaged across a range of domains, such as family income. The data are ideal for the current study because (1) the disadvantaged nature of participants means that a significant number of mothers experienced eviction and (2) the study asks questions that capture maternal exposure to eviction and criminal justice involvement among mothers.

In this study, we restricted our analytic sample to individuals with complete data on our outcome measure that captures criminal justice involvement (being charged with a crime or booked by police between the year 5 and year 9 surveys). This restriction reduced our sample from 4,898 mothers to 3,515 mothers. For our other variables, we used multiple imputation to impute values for missing data (Allison 2002; Rubin 1976). As recommended by von Hippel (2007), we do not use imputed values for our outcome measure (criminal justice involvement) but did include it in our imputation model. This approach often leads to more efficient estimates than approaches that use imputed data for the outcome measure and guards against the possibility that problematic imputed outcome values will drive the results (von Hippel 2007). The analyses presented in this article are based on the average estimate from five imputed data sets.

Even though we use multiple imputation, because we restrict our sample to respondents with complete data on our outcome measure, it is possible that attrition affects the characteristics of our analytic sample and, in turn, our findings. In analyses comparing respondents in our analytic sample with those who attrited from our sample, we found that those who attrited were more likely to be foreign born, to have dropped out of high school, to be depressed at baseline, and to have performed poorly on cognitive assessments. Respondents who were not in our sample were also more likely to be Latina, less likely to be African American, and were slightly older. In the Supplemental Analyses section of this article, we examine the degree to which attrition affects our findings.

Measures

Dependent Variable

The dependent variable in each of our analyses, measured in the year 9 survey wave, measures criminal justice involvement. Specifically, it captures whether a mother was booked by the police or charged with a crime (anything other than a minor traffic violation) in the past four years (since the previous interview wave). This variable is measured dichotomously: Any mother who has been charged with a crime or booked by police during the past four years is coded as 1, and mothers are coded as 0 if they were not charged or booked during this time period.

Key Independent Variables

At the year 1 and year 3 survey waves, mothers are asked whether they were evicted in the past 12 months. We use these questions to create our key independent variables. Our first variable is a dichotomous measure indicating whether mothers experienced eviction by the time of the year 3 survey: Mothers are coded as 1 if they answered “yes” to the eviction question in at least one of the survey waves and 0 if they answered “no” in both survey waves. This measure undercounts the number of mothers who experienced eviction because it does not capture evictions that occurred after the year 1 survey and more than 12 months prior to the year 3 survey. Because we undercount eviction, our analyses likely lead to a conservative test of the hypothesis that ever experiencing eviction is associated with an increased risk of criminal justice involvement (Desmond and Kimbro 2015).

Our second set of measures captures eviction timing so that we can explore whether the effect of eviction on criminal justice involvement is dependent on the recency of the eviction. First, we create a dichotomous measure indicating whether mothers experienced eviction between the baseline and year 1 survey waves. Second, we create a measure indicating whether mothers experienced eviction between the year 1 and year 3 survey waves. Since the question this measure is based on asks only about the 12 months preceding the year 3 survey, this measure undercounts the prevalence of eviction between the two survey waves and likely results in a conservative estimate of eviction’s impact on criminal justice involvement (Desmond and Kimbro 2015).

For each of these variables, we do not include evictions that occurred prior to the baseline survey because eviction was not measured in that survey wave (when the child was born). Moreover, by focusing on evictions that occurred after the baseline survey, we are able to ensure that the control variables in our analyses capture characteristics that are not measured after (and therefore potentially caused by) the key independent variables. We do not include the eviction question from the year 5 survey to ensure that our key independent variables capture experiences that occurred prior to the measurement of the dependent variable (criminal justice involvement between the year 5 and year 9 surveys) and our mediating variables (measured in the year 5 survey wave).

Mediators

Our mediators capture three aspects of strain or responses to strain and are taken from the year 5 survey to ensure that they are measured after our key independent variables (which measure eviction) but prior to our outcome variable (criminal justice involvement between the year 5 and year 9 survey waves). To capture financial hardship during year 5, we create a continuous material hardship scale that indicates how many types of hardship a mother has experienced in the past 12 months (Zilanawala and Pilkauskas 2012). This scale uses nine questions to capture whether mothers experienced hardship across five domains: utilities (telephone was disconnected or gas/electricity was turned off), bills (did not pay full amount of rent/mortgage, gas, oil, or electricity bill), housing (homeless, evicted, or doubled up), food (received free food/meals), and medical (household member did not go to doctor because of cost).

To capture depression during year 5, we create a bivariate measure indicating whether the mother was depressed or not using the conservative threshold of the Composite International Diagnostic Interview Short-form (see Kessler et al. 1998 for more detail about the measure). Last, to capture substance use during year 5, we create a dichotomous measure: mothers who used an illicit drug without prescription (sedatives, tranquilizers, amphetamines or other stimulants, analgesics or prescription painkillers, inhalants, marijuana, cocaine or crack, hallucinogens, and heroin) or had four drinks or more in a day at least weekly were coded as 1, while mothers who did not engage in this behavior were coded as 0.

Control Variables

A key concern in determining whether maternal experiences with eviction are associated with criminal justice involvement is the fact that individuals who experience eviction and who have criminal justice involvement both tend to be disadvantaged. As a result, to guard against detecting spurious associations, we include a number of maternal characteristics that prior scholarship suggests may be associated with both criminal justice involvement and experiencing eviction as covariates. We control for prior criminal conviction (convicted by baseline survey or not),² maternal education (dropped out of high school or not), alcohol and drug use (whether the mother used illicit drugs or drank alcohol during her pregnancy), nativity status (whether a mother was born in the United States or not), depression (whether a mother was perceived by the interviewer as depressed or withdrawn in the hospital after her child’s birth), and family structure at age 15 (whether a mother lived with both biological parents when she was 15 years old), each of which is measured dichotomously.

We also control for race, marital status, income-to-needs ratio, health, and religiosity, each of which is measured categorically. Race is a variable with four categories (white, African American, Latino, or other). Marital status captures whether mothers were married, cohabiting, or single. The income-to-needs ratio accounts for family size and captures the degree to which a household’s income is above or below the poverty threshold. From this, we created a four-category measure that indicated whether a mother’s income was (1) 0 to 99 percent, (2) 100 to 199 percent, (3) 200 to 299 percent, or (4) at least 300 percent of the poverty threshold. To capture health, we used a self-reported measure. Mothers were categorized based on whether they considered themselves to be in great, very good, good, or fair or poor health. Our measure for religiosity grouped mothers into the following categories based on how frequently they attended religious services: once a week, several times a month, several times a year, hardly, or never.

Last, we included a number of control variables that were measured continuously: age, impulsivity, and cognitive ability. We measure impulsivity by employing an abbreviated six-item form of Dickman’s (1990) impulsivity scale. This scale is designed to capture self-control and includes items such as whether a mother often does or says things without considering the consequences. To measure cognitive ability, we use the sum score on the Wechsler Adult Intelligence Scale–Revised similarities subtest (Wechsler 1981). All covariates in our analyses are measured at baseline (when the child was born) except cognitive ability and impulsivity (which were measured in the year 3 survey but were designed to assess static characteristics). As mentioned previously, this ensures that our control variables capture characteristics that preceded (and were not caused by) the eviction and criminal justice involvement variables.

Analytic Strategy

In cases where an outcome variable is dichotomous (like criminal justice involvement), the traditional approach is to conduct logistic or probit regression (Cameron and Trivedi 2010; Long and Freese 2001). For our statistical analyses, we are interested in calculating total, direct, and indirect effects. To do so, we need to compare coefficients across nested models. Unlike in linear models, comparing coefficients across nested models in nonlinear models (like logistic or probit regression) is not straightforward (Karlson et al. 2012; Kohler, Karlson, and Holm 2011; Mood 2010). Essentially, in nested nonlinear models, coefficients for an independent variable (i.e., eviction) can differ when mediating variables are added not solely because of the effect of the mediating variables; when mediating variables are independently associated with an outcome variable, these differences also can result from rescaling (Karlson et al. 2012; Kohler et al. 2011; Mood 2010).

To address the issue of rescaling, we present estimates from logistic regression models using the KHB method (Karlson et al. 2012; Kohler et al. 2011).³ First, we provide estimates of the total, indirect, and direct effect of ever experiencing eviction by the year 3 survey wave on criminal justice involvement, net of a wide range of covariates. A key feature of the KHB approach is that it allows us to determine the independent contribution of each of our three mediators to the indirect effect, as well as calculate the total indirect effect (Karlson et al. 2012; Kohler et al. 2011). In addition to coefficient estimates, we provide estimates of the share of the total effect that is explained by the mediating variables in our model (Karlson et al. 2012; Kohler et al. 2011).

Our second analysis is the same as the first, except we examine the impact of eviction timing. To do so, we present KHB logistic regression estimates of the total, indirect, and direct effect of experiencing eviction between the baseline and year 1 and the year 1 and year 3 survey waves on criminal justice involvement, net of one another and a wide range of covariates. For both eviction measures, we also provide estimates of the share of the total effect that is explained by the mediating variables in our model (Karlson et al. 2012; Kohler et al. 2011).

Results

Descriptive Results

In panel 1 of Table 1, we provide descriptive information about the sample. As indicated in panel 1, approximately 5 percent of mothers in the study were involved in the criminal justice system between years 5 and 9 of the study (the key outcome measure). Roughly 4.5 percent of mothers experienced at least one eviction between the baseline and year 3 survey; nearly 3.0 percent and 2.2 percent of mothers experienced eviction between the baseline and year 1 survey and between the year 1 and year 3 survey, respectively.

Table 1.

Descriptive Statistics for Sample and by Eviction Status (N = 3,515).

	Full	Evicted	Not Evicted
	M / %	M / %	M / %
Criminal justice involvement (S)	4.98	14.94	4.48
Ever evicted by year 3 (S)	4.48	100.00	0.00
Evicted year 1 (S)	2.91	65.75	0.00
Evicted year 3 (S)	2.18	48.30	0.00
Hardship year 5 (S)	0.97 (1.43)	2.33 (2.08)	0.91 (1.36)
Depression year 5 (S)	12.02	25.90	11.46
Substance use year 5 (S)	6.65	16.55	6.24
Prior conviction (S)	2.81	6.36	2.80
Race
White	21.52	21.43	21.86
Black	49.94	51.95	49.85
Latino	25.06	23.38	24.85
Other	3.48	3.25	3.45
Marital status (S)
Single	40.00	45.45	39.57
Married	24.04	3.25	25.19
Cohabiting	35.96	51.30	35.24
Less than high school (S)	32.52	43.79	31.76
Poverty ratio (S)
0–99%	35.28	40.26	35.12
100–199%	26.43	35.71	25.83
200–299%	15.48	16.88	15.44
300%+	22.82	7.14	23.61
Age (S)	25.13 (6.02)	23.38 (5.02)	25.21 (6.05)
Cognitive ability	6.82 (2.67)	6.70 (2.76)	6.82 (2.67)
Impulsivity (S)	1.50 (1.71)	2.25 (2.09)	1.47 (1.68)
Depression at baseline	4.63	4.61	4.62
Substance use during pregnancy	13.07	16.99	12.83
Lived with both parents (S)	41.80	26.14	42.37
Born in United States (S)	85.88	94.81	85.86
Religiosity
Once a week	21.65	12.99	22.08
Several times a month	16.91	18.83	16.95
Several times a year	21.42	24.68	21.32
Hardly	25.47	26.62	25.29
Never	14.55	16.88	14.36
Health
Great	31.28	33.12	31.55
Very good	34.87	33.77	34.99
Good	26.30	22.08	26.09
Fair or poor	7.54	11.04	7.38

Note: S = difference between eviction and no eviction (p < .05). Descriptive statistics are percentages for categorical variables and means for continuous variables. Standard deviations are in parentheses for continuous variables. Chi-square tests are used for categorical variables, and T tests are used for continuous variables. Descriptive statistics are based on non-imputed data.

In panels 2 and 3 of Table 1, we describe differences by maternal eviction status. Three key themes stand out. First, mothers who have experienced eviction are significantly more likely to have experienced criminal justice involvement than mothers who have not. Second, mothers who had been evicted had more negative values on our mediating variables in year 5 than mothers who had not been evicted. Specifically, at year 5, mothers who had been evicted experienced higher levels of financial hardship and were more likely to be depressed and to use substances. Third, mothers who experienced eviction tended to be disadvantaged across a wide range of other characteristics (i.e., education and family income), necessitating the use of multivariate regression to isolate the association between maternal eviction and risk of criminal justice involvement.

Multivariate Regression Results

In Table 2, we present results from a KHB logistic regression analysis exploring the total, indirect, and direct effect of ever experiencing eviction between the baseline and year 3 surveys on criminal justice involvement between the year 5 and year 9 surveys. The coefficient in the “Total” column of the table indicates the cumulative effect (the sum of the indirect and direct effect) of ever experiencing eviction on criminal justice involvement, net of the covariates described in the Control Variables section of this article. The “Indirect” column identifies the degree to which the effect of experiencing eviction operates through our mediating variables. The “Direct” column describes the association between experiencing eviction and criminal justice involvement that is not explained by the mediators. Last, the “% Indirect Explains” column reports the percentage of the total effect that is explained by the indirect effect. All coefficients of effect estimates in this table, and moving forward, are reported as odds ratios for ease of interpretation.

Table 2.

Karlson, Holm, and Breen Logistic Regression Models Estimating Total, Direct, and Indirect Effect of Ever Experiencing an Eviction on Risk for Criminal Justice Involvement (N = 3,515).

	Total	Indirect	Direct	% Indirect Explains
Ever evicted	2.26*** (3.00)	1.44*** (5.06)	1.57 (1.62)	44.52
Hardship year 5		1.29*** (3.84)		31.13
Depression year 5		1.03 (0.95)		3.45
Substance use year 5		1.08*** (2.87)		9.94

Note: Coefficients are odds ratios. t values are in parentheses. Estimates are based on a model that controls for prior conviction, race, marital status, education, income-to-needs ratio, age, cognitive ability, impulsivity, depression, substance use, family structure at age 15, nativity, religiosity, and self-reported health. All variables are measured at baseline, with the exception of cognitive ability and impulsivity, which are measured at year 3 and are intended to capture unchanging characteristics.

***

p < .01.

The results in Table 2 lead to a number of conclusions. First, mothers who have experienced eviction by the year 3 survey wave have 2.26 times higher odds of criminal justice involvement (p < .01) than mothers who have not been evicted. Second, a substantial share of this total effect can be explained by indirect effects through our mediating variables. Specifically, we find a significant indirect effect for our mediating measures in combination: odds ratio = 1.44, p < .01; together our mediating variables explain approximately 45 percent of the total effect. Among mediating variables, we find significant indirect effects through financial hardship (odds ratio = 1.29, p < .01) and substance use (odds ratio = 1.08, p < .01). The indirect effect through financial hardship accounts for 31 percent of the total effect, while the indirect effect through substance use explains 10 percent of the total effect. Last, after including our mediating variables, we do not find a significant direct effect between experiencing eviction by year 3 and criminal justice involvement.

In Table 3, we build on the previous results by exploring whether the timing of eviction influences its impact on criminal justice involvement. Specifically, we present results from a KHB logistic regression analysis exploring the total, indirect, and direct associations of experiencing eviction between baseline and year 1 and the year 1 and year 3 survey waves on criminal justice involvement between the year 5 and year 9 surveys, net of one another and controlling for a wide range of covariates.

Table 3.

Karlson, Holm, and Breen Logistic Regression Models Estimating Total, Direct, and Indirect Effect of Eviction Timing on Risk for Criminal Justice Involvement (N = 3,515).

	Total	Indirect	Direct	% Indirect Explains
Evicted year 1	2.18** (2.27)	1.34*** (3.36)	1.63 (1.41)	37.33
Hardship year 5		1.23*** (3.37)		26.72
Depression year 5		1.03 (0.96)		3.39
Substance use year 5		1.06* (1.94)		7.23
Evicted year 3	2.55*** (2.66)	1.43*** (3.82)	1.78 (1.62)	38.12
Hardship year 5		1.30*** (3.56)		27.93
Depression year 5		1.03 (0.95)		2.69
Substance use year 5		1.07** (2.07)		7.51

Note: Coefficients are odds ratios. t values are in parentheses. Estimates are based on a model that includes both eviction at year 1 and eviction at year 3 and controls for prior conviction, race, marital status, education, income-to-needs ratio, age, cognitive ability, impulsivity, depression, substance use, family structure at age 15, nativity, religiosity, and self-reported health. All variables are measured at baseline, with the exception of cognitive ability and impulsivity, which are measured at year 3 and are intended to capture unchanging characteristics.

p < .10. **p < .05. ***p < .01.

The results presented in the first panel of Table 3 lead to a number of conclusions. First, mothers who experienced eviction between the baseline and year 1 survey had 2.18 times higher odds of criminal justice involvement than mothers who did not (p < .05). Second, our mediating variables explain a substantial share of this total effect. Specifically, we find a significant indirect effect for our mediating measures in combination (odds ratio = 1.34, p < .01); together the mediating variables explain 37 percent of the total effect. Of our mediating variables, we find significant indirect effects through financial hardship (odds ratio = 1.23, p < .01) and marginally significant indirect effects through substance use (odds ratio = 1.06, p < .10). The indirect effect through financial hardship accounts for nearly 27 percent of the total effect, while the indirect effect through substance use explains 7 percent of the total effect. Last, there is no significant direct effect of eviction between baseline and year 1 on criminal justice involvement once we include mediating variables.

The results from panel 2 of Table 3 suggest that patterns for more recent evictions are similar to those for evictions experienced between baseline and year 1. Mothers who experienced eviction between the year 1 and year 3 survey waves had 2.55 times higher odds of criminal justice involvement than mothers who did not experience eviction during this period. A significant portion of this association is explained by our mediating variables. Specifically, we find a significant indirect effect for our mediating measures in combination (odds ratio = 1.43, p < .01); together these measures explain 38 percent of the total effect. Among specific mediating variables, we find significant indirect effects through financial hardship (odds ratio = 1.30, p < .01) and substance use (odds ratio = 1.07, p < .05). The indirect effect through financial hardship explains 28 percent of the total effect, while the indirect effect through substance use accounts for 8 percent of the total effect. Last, after including our mediating variables, we do not find a significant direct effect between experiencing eviction within the 12 months leading up to the year 3 survey and criminal justice involvement.

Supplemental Analyses

In the Data section of this article, we discussed some differences in the characteristics of mothers in our analytic sample compared to mothers who attrited. In supplemental analyses, we tested the sensitivity of our results to attrition by conducting each analysis in this article using the full sample (4,898 mothers) and including imputed values for respondents missing data on our outcome measure. Results from these analyses (not shown) are generally similar to those presented in this article, suggesting that attrition does not greatly affect our findings. The one difference is that we find some evidence that significant direct effects remain after including our mediating variables (t = 2.33 for the ever evicted measure; t = 1.77 for the year 1 eviction measure; t = 1.82 for the year 3 eviction measure); by comparison in the results presented in this article, the direct effects were not statistically significant, but the coefficients were positive and nearly marginally significant (t = 1.62 for the ever evicted measure; t = 1.41 for the year 1 eviction measure; t = 1.62 for the year 3 eviction measure).

Discussion

In recent years, eviction has become an increasingly prominent part of American life, particularly for low-income women (Desmond 2016; Desmond and Kimbro 2015). As a result, scholars have begun to examine the implications of eviction for families (Desmond and Gershenson 2016; Desmond and Kimbro 2015; Desmond and Shollenberger 2015; Desmond et al. 2015). In this study, we build on this existing scholarship by (1) presenting the first estimates of the association between eviction and criminal justice involvement in the U.S. context, (2) examining whether recent evictions and less recent evictions both place mothers at greater risk of criminal justice involvement, and (3) providing the first test of which pathways help to explain associations between eviction and criminal justice involvement. For our analyses, we draw on data from Fragile Families—a longitudinal data source that is ideal because it oversamples disadvantaged mothers who are at risk of experiencing eviction—and employ logistic regression using the KHB method (Desmond and Kimbro 2015; Karlson et al. 2012).

Our results lead to two important conclusions. First, mothers who have been evicted are at greater risk of experiencing criminal justice involvement than mothers who have not been evicted; this is true for both recent and less recent evictions. These results are consistent with prior empirical work that has found that housing instability and homelessness are associated with an increased likelihood that men and women will engage in criminal behavior and experience prison and/or jail incarceration (Center for Poverty Solutions 2003; Gowan 2002; Greenberg and Rosenheck 2008; Hagan and McCarthy 1998; Heerde et al. 2014; Weiser et al. 2009). In addition, the fact that both recent and less recent evictions are associated with an increase in criminal justice involvement suggests that the impact of eviction is long lasting. Last, these results align with Alm’s (2018) findings in Sweden, a context far different than the context of the United States (Esping-Andersen 1990; Gottlieb 2017).

Our second key finding is that the GST and RCT frameworks both may be important for understanding why eviction increases the likelihood that mothers experience criminal justice involvement. Our mediating variables significantly mediated all associations between eviction and criminal justice involvement, explaining 45 percent of the association between experiencing eviction by the year 3 survey as well as approximately 37 percent and 38 percent of the associations for evictions occurring between baseline and year 1 and the year 1 and year 3 survey waves, respectively.

Among our mediating variables, the most impactful mediator was material hardship at year 5, which explained between 27 percent and 31 percent of the association depending on the eviction measure. This suggests that an important pathway through which eviction leads to criminal justice involvement is through the financial hardship that occurs as a result of eviction. This finding is consistent with both GST and RCT theories. In terms of GST, this finding is consistent with research suggesting that the accumulation of strain is particularly likely to result in criminal offending (Agnew 2001; Brezina 2017; Eitle and Turner 2002). From the perspective of RCT, this finding is consistent with the view that eviction may lead to criminal offending by increasing the benefits associated with engaging in criminal behavior, particularly financial crime (Becker 1968; Cornish and Clarke 1986; Hayward 2007; Loughran et al. 2016; Nagin and Paternoster 1993; Paternoster et al. 2017).

In addition to material hardship, we found that substance use at year 5 significantly mediated the association between eviction and criminal justice involvement, explaining between 7 percent and 10 percent of the total effect depending on the eviction measure. This is consistent with GST and prior research finding that substance use is often used to cope with strain (Agnew 2001; Brezina 2017) and scholarship finding that illicit drug use and heavy alcohol use are associated with future criminality (Bennett et al. 2008; Jennings et al. 2015). The one measure that did not significantly mediate associations between eviction and criminal justice involvement was depression at year 5. This was a surprising finding because GST argues that negative emotions play an important role in explaining strain’s impact on criminal offending (Agnew 1992, 2001, 2006). However, depression is only one of the negative emotions emphasized in GST; had we been able to include measures capturing other negative emotions, such as anger, anxiety, and resentment, we may have found that negative emotions mediate associations between eviction and criminal justice involvement (Agnew 1992, 2001, 2006).

When examining these findings, it is important to consider the limitations of this study. First, it is important to note that the findings of this study are associations rather than causal effects. Our results are based on observational data, so it is possible that some unobserved factor may explain the associations that we found. To better assess causality, future research should look to exploit natural policy experiments that either increased or reduced the prevalence of eviction, such as a substantial expansion or reduction in housing vouchers. Second, our study was not able to tease out the relative role played by GST and RCT in explaining the associations observed. The fact that material hardship has the strongest indirect effect is consistent with both the GST and RCT frameworks. Future research should attempt to adjudicate between these and other theoretical frameworks. To do so, researchers should incorporate data on perceived costs and benefits of crime into analyses. Researchers should also explore the association between eviction and different types of crime since some types of crime may be more consistent with a specific theoretical framework (i.e., assault and GST).

Despite these limitations, this study has potentially important implications for policy makers and social service providers. Given the substantial growth of the female incarceration rate in recent years and the implications of the incarceration of mothers for children, reducing the number of women involved in the criminal justice system is an important policy priority (Dallaire et al. 2010; Hagan and Foster 2012; Huebner and Gustafson 2007; Sawyer 2018; The Sentencing Project 2015). Since eviction is quite common among low-income women (Desmond 2016; Desmond and Kimbro 2015), the results from this study suggest that policies that aim to reduce eviction by making housing more affordable (such as increasing the supply and generosity of housing vouchers) or by increasing the material resources available to families (i.e., raising the minimum wage or increasing the generosity and accessibility of welfare benefits) may hold promise in making progress on this aim. In addition, policy makers and service providers should focus on creating interventions that mitigate the negative consequences of eviction once it occurs. The results from this study suggest that interventions that focus on improving the financial circumstances of families who have experienced eviction and that provide substance use treatment may be particularly effective.

Footnotes

Appendix

Table A.1.

Logistic Regression Models Estimating Effect of Eviction on Criminal Justice Involvement (N = 3,515).

	Model 1	Model 2
Ever evicted by year 3	2.31 (3.17)***
Evicted year 1		2.18 (2.30)**
Evicted year 3		2.46 (2.60)***
Prior conviction	4.32 (4.81)***	4.21 (4.68)***
Race (white = reference)
Black	0.60 (−2.27)**	0.61 (−2.23)**
Latino	0.62 (−1.78)*	0.63 (−1.71)*
Other	1.74 (1.27)	1.77 (1.31)
Marital status (single = reference)
Married	0.64 (−1.42)	0.64 (−1.39)
Cohabiting	0.74 (−1.69)*	0.74 (−1.70)*
Less than high school	1.89 (3.46)***	1.86 (3.37)***
Poverty ratio (300%+ = reference)
0–99%	1.56 (1.36)	1.54 (1.32)
100–199%	2.22 (2.49)**	2.20 (2.45)**
200–299%	1.74 (1.57)	1.72 (1.54)
Age	0.94 (−3.30)***	0.94 (−3.32)***
Cognitive ability	0.98 (−.0.45)	0.98 (−.0.57)
Impulsivity	1.08 (1.55)	1.07 (1.47)
Depression at baseline	1.55 (1.47)	1.58 (1.52)
Substance use during pregnancy	2.27 (4.14)***	2.25 (4.05)***
Lived with both parents	0.79 (−1.27)	0.79 (−1.24)
Born in United States	2.33 (2.16)**	2.34 (2.17)**
Religiosity (once a week = reference)
Several times a month	1.31 (0.92)	1.30 (0.89)
Several times a year	1.04 (0.14)	1.01 (0.05)
Hardly	1.14 (0.50)	1.14 (0.51)
Never	1.33 (1.02)	1.31 (0.95)
Health (great = reference)
Very good	1.83 (2.74)***	1.88 (2.85)***
Good	1.59 (1.96)**	1.62 (2.04)**
Fair or poor	1.28 (0.73)	1.33 (0.83)

Note: Coefficients are odds ratios. t values are in parentheses.

p < .10. **p < .05. ***p < .01.

Acknowledgements

We would like to thank Chang-ming Hsieh for helpful feedback on earlier versions of this manuscript.

1

We do not use data from the year 15 survey wave (the most recent wave) because, in that wave of data collection, mothers were not asked about their criminal justice involvement.

2

The first time criminal justice involvement (our outcome measure) is measured is at the year 5 survey wave, which is after our eviction variables (measured in years 1 and 3). Since criminal justice involvement is measured only after our eviction variables, we cannot include it as a control.

3

In the appendix, we present results from traditional logistic regression models. We include these models so individuals can see how eviction coefficients compare to coefficients for other covariates. These models include a full set of control variables, and the coefficients are quite similar to the total effect coefficient in our preferred specifications. The eviction coefficients are slightly different because of rescaling (Karlson, Holm, and Breen 2012).

Author Biographies

Aaron Gottlieb is an assistant professor at the Jane Addams College of Social Work, University of Illinois at Chicago. His research examines the causes and consequences of criminal justice involvement.

Jessica W. Moose received her MSW from the Jane Addams College of Social Work, University of Illinois at Chicago. Her research interests include improving access to education for individuals from low-income backgrounds as well as racial, ethnic, sexual, and gender minorities.

References

Agnew

Robert

. 1992. “Foundation for a General Strain Theory of Crime and Delinquency.” Criminology 30:47–87.

Agnew

Robert

. 2001. “Building on the Foundation of General Strain Theory: Specifying the Types of Strain Most Likely to Lead to Crime and Delinquency.” Journal of Research in Crime and Delinquency 38:319–61.

Agnew

Robert

. 2006. Pressured into Crime: An Overview of General Strain Theory. Los Angeles: Roxbury.

Allison

Paul D.

2002. Missing Data. Thousand Oaks, CA: Sage.

Alm

Susanne

. 2018. “Isolating the Effect of Eviction on Criminal Convictions: Results from a Swedish Study.” Acta Sociologica 61:263–82.

Becker

Gary

. 1968. “Crime and Punishment: An Economic Analysis.” Journal of Political Economy 78:169–217.

Bennett

Trevor

Holloway

Katy

Farrington

David

. 2008. “The Statistical Association between Drug Misuse and Crime: A Meta-analysis.” Aggression and Violent Behavior 13:107–18.

Brezina

Timothy

. 2017. “General Strain Theory.” Oxford Research Encyclopedia of Criminology and Criminal Justice. New York: Oxford University Press. doi:10.1093/acrefore/9780190264079.013.249.

Cameron

A. Colin

Trivedi

Pravin K.

2010. Microeconometrics Using Stata. Rev. ed. College Station, TX: STATA Press.

10.

Center for Poverty Solutions. 2003. Barriers to Stability: Homelessness and Incarceration’s Revolving Door in Baltimore City. Baltimore, MD: Center for Poverty Solutions.

11.

Cornish

Derek

Clarke

Ronald

1986. Introduction. Pp. 1–13 in The Reasoning Criminal, edited by Cornish

D. B.

Clarke

R. V.

New York: Springer-Verlag.

12.

Dallaire

Dannielle H.

Ciccone

Anne

Wilson

Laura C.

2010. “Teachers’ Experiences with and Expectations of Children with Incarcerated Parents.” Journal of Applied Development Psychology 31(4):281–90.

13.

Desmond

Matthew

. 2012. “Eviction and the Reproduction of Urban Poverty.” American Journal of Sociology 118:88–133.

14.

Desmond

Matthew

. 2016. Evicted: Poverty and Profit in the American City. New York: Crown Books.

15.

Desmond

Matthew

Bell

Monica C.

2015. “Housing, Poverty, and the Law.” Annual Review of Law and Social Science 11:15–35.

16.

Desmond

Matthew

Gershenson

Carl

. 2016. “Housing and Employment Insecurity among the Working Poor.” Social Problems 63:46–67.

17.

Desmond

Matthew

Gershenson

Carl

. 2017. “Who Gets Evicted? Assessing Individual, Neighborhood, and Network Factors.” Social Science Research 62:1–16.

18.

Desmond

Matthew

Gershenson

Carl

Kiviat

Barbara

. 2015. “Forced Relocation and Residential Instability among Urban Renters.” Social Service Review 89:227–62.

19.

Desmond

Matthew

Kimbro

Rachel T.

2015. “Eviction’s Fallout: Housing, Hardship, and Health.” Social Forces 94(1):295–324.

20.

Desmond

Matthew

Shollenberger

Tracey

. 2015. “Forced Displacement from Rental Housing: Prevalence and Neighborhood Consequences.” Demography 52(5):1752–72.

21.

Dickman Scott

1990. “Functional and Dysfunctional Impulsivity: Personality and Cognitive Correlates.” Journal of Personality and Social Psychology 58(1):95–102.

22.

Eggers

Frederick J.

Moumen

Fouad

. 2010. Investigating Very High Rent Ratings among Renters in the American Housing Survey. Washington, DC: U.S. Department of Housing and Urban Development.

23.

Eitle

David

Turner

R. Jay

. 2002. “Exposure to Community Violence and Young Adult Crime: The Effects of Witnessing Violence, Traumatic Victimization, and Other Stressful Life Events.” Journal of Research in Crime and Delinquency 39:214–37.

24.

Esping-Andersen

Gosta

. 1990. The Three Worlds of Welfare Capitalism. Princeton, NJ: Princeton University Press.

25.

Gottlieb

Aaron

. 2017. “Incarceration and Relative Poverty in Cross-national Perspective: The Moderating Roles of Female Employment and the Welfare State.” Social Service Review 91(2):293–318.

26.

Gowan

Teresa

. 2002. “The Nexus: Homelessness and Incarceration in Two American Cities.” Ethnography 3(4):500–34.

27.

Greenberg

Greg A.

Rosenheck

Robert A.

2008. “Jail Incarceration, Homelessness, and Mental Health: A National Study.” Psychiatric Services 59(2):170–77.

28.

Hagan

John

Foster

Holly

. 2012. “Children of the American Prison Generation: Student and School Spillover Effects of Incarcerating Mothers.” Law & Society Review 46(1):37–69.

29.

Hagan

John

McCarthy

Bill

. 1998. Mean Streets: Youth Crime and Homelessness. Cambridge, UK: Cambridge University Press.

30.

Hayward

Keith

. 2007. “Situational Crime Prevention and Its Discontents: Rational Choice Theory versus the ‘Culture of Now.’” Social Police & Administration 41:232–50.

31.

Heerde

Jessica A.

Hemphill

Sheryl A.

Scholes-Balog

Kirsty E.

2014. “‘Fighting’ for Survival: A Systematic Review of Physically Violent Behavior Perpetrated and Experienced by Homeless Young People.” Aggression and Violent Behavior 19(1):50–66.

32.

Huebner

Beth M.

Gustafson

Regan

. 2007. “The Effect of Maternal Incarceration on Adult Offspring Involvement in the Criminal Justice System.” Journal of Criminal Justice 35:283–96.

33.

Jennings

Wesley G.

Piquero

Alex R.

Rocque

Michael

Farrington

David P.

2015. “The Effects of Binge and Problem Drinking on Problem Behavior and Adjustment over the Life Course: Findings from the Cambridge Study in Delinquent Development.” Journal of Criminal Justice 43(6):453–63.

34.

Johnson

Rucker

Raphael

Steven

. 2012. “How Much Crime Reduction Does the Marginal Prisoner Buy?” Journal of Law and Economics 55(2):275–310.

35.

Karlson

Kristian B.

Holm

Anders

Breen

Richard

. 2012. “Comparing Regression Coefficients between Same-sample Nested Models Using Logit and Probit: A New Method.” Sociological Methodology 42:286–313.

36.

Kessler

Ronald C.

Andrews

Daniel Mroczek

Ustun

Bedirhan

Wittchen

Hans-Ulrich

. 1998. “The World Health Organization Composite International Diagnostic Interview Short-form (CIDI-SF).” International Journal of Methods in Psychiatric Research 7(4):171–85.

37.

Kohler

Ulrich

Karlson

Kristian B.

Holm

Anders

. 2011. “Comparing Coefficients of Nested Nonlinear Probability Models.” Stata Journal 11:420–38.

38.

Long

J. Scott

Freese

Jeremy

. 2001. Regression Models for Categorical Dependent Variables Using Stata. College Station, TX: Stata Press.

39.

Loughran

Thomas A.

Paternoster

Ray

Chalfin

Aaron

Wilson

Theodore

. 2016. “Can Rational Choice Be Considered a General Theory of Crime? Evidence from Individual Level Panel Data.” Criminology 54:86–112.

40.

Merton

Robert K.

1938. “Social Structure and Anomie.” American Sociological Review 3(5):672–82.

41.

Mood

Carina

. 2010. “Logistic Regression: Why We Cannot Do What We Think We Can Do and What We Can Do about It.” European Sociological Review 26:67–82.

42.

Nagin

Daniel S.

Paternoster

Raymond

. 1993. “Enduring Individual Differences and Rational Choice Theories of Crime.” Law & Society Review 27:467–96.

43.

Paternoster

Ray

Jaynes

Chae

Wilson

Theodore

. 2017. “Rational Choice Theory and Interest in the ‘Fortune of Others.’” Journal of Research in Crime and Delinquency 54:847–68.

44.

Raphael

Steven

Stoll

Michael A.

2013. Why Are So Many Americans in Prison? New York: Russell Sage.

45.

Reichman

Nancy. E.

Teitler

Julien O.

Garfinkel

Irwin

McLanahan

Sara S.

2001. “Fragile Families: Sample and Design.” Children and Youth Services Review 23:303–26.

46.

Rubin

Donald B.

1976. “Inference and Missing Data.” Biometrika 63:581–92.

47.

Sawyer

Wendy

. 2018. “The Gender Divide: Tracking Women’s State Prison Growth.” Prison Policy Initiative, January 9. Retrieved October 10, 2018 (https://www.prisonpolicy.org/reports/women_overtime.html).

48.

The Sentencing Project. 2015. Incarcerated Women and Girls. Retrieved October 10, 2018 (https://www.sentencingproject.org/publications/incarcerated-women-and-girls).

49.

Slocum

Lee A.

Simpson

Sally S.

Smith

Douglas A.

2005. “Strained Lives and Crime: Examining Intra-individual Variation in Strain and Offending in a Sample of Incarcerated Women.” Criminology 43(4):1067–1110.

50.

von Hippel

Paul

. 2007. “Regression with Missing Ys: An Improved Strategy for Analyzing Multiply Imputed Data.” Sociological Methodology 37:83–118.

51.

Wechsler

David

. 1981. Manual for the Wechsler Adult Intelligence Scale–Revised. New York: Psychological Corporation.

52.

Weiser

Sheri. D.

Neilands

Torsten B.

Comfort

Megan L.

Dilworth

Samantha E.

Cohen

Jennifer

Tulsky

Jacqueline P.

Riley

Elise D.

2009. “Gender-specific Correlates of Incarceration among Marginally Housed Individuals in San Francisco.” American Journal of Public Health 99(8):1459–63.

53.

Zilanawala

Afshin

Pilkauskas

Natasha V.

2012. “Material Hardship and Child Socioemotional Behaviors: Differences by Types of Hardship, Timing, and Duration.” Children and Youth Services Review 34(4):814–25.

The Effect of Eviction on Maternal Criminal Justice Involvement

Abstract

Keywords

Introduction

Background

GST, Eviction, and Criminal Justice Involvement

RCT, Eviction, and Criminal Justice Involvement

Existing Empirical Scholarship on Eviction and Criminal Justice Involvement

The Current Study

Data, Measures, and Analytic Strategy

Data

Measures

Dependent Variable

Key Independent Variables

Mediators

Control Variables

Analytic Strategy

Results

Descriptive Results

Multivariate Regression Results

Supplemental Analyses

Discussion

Footnotes

Appendix

Acknowledgements

1

2

3

Author Biographies

References