The Third Shift: Multiple Job Holding and the Incarceration of Women’s Partners

Abstract

A large body of research documents the sensitivity of women’s employment to changing family circumstances, but we know little about the relationship between partner incarceration—a common family transition in the lives of disadvantaged women—and employment. Despite reasons to suspect that changes in resources associated with incarceration have consequences for the employment of family members, previous research suggests that partner incarceration does not influence the number of hours women work at their main jobs. This paper uses data from the Fragile Families and Child Wellbeing Study (N=3,835) to examine how partner incarceration is associated with multiple job holding, an alternative strategy for increasing earnings. Results show that women with incarcerated partners are more likely to work multiple jobs than women in otherwise similar circumstances, suggesting partner incarceration is linked to a “third shift”—to additional employment on top of the paid work and caregiving women already do.

1. INTRODUCTION

The dramatic rise in women’s labor force participation has been one of the most substantial changes in the family over the past half-century. Women’s increasing responsibility for wage-earning and continued responsibility for housework and childcare means that many women work a double-day, with motherhood and household duties constituting a “second shift” (Hochschild, 2003). Although many working women struggle to balance their first and second shifts, work-family issues for women at the top and bottom of the income distribution are quite different. For professional-managerial women—who have often been at the center of work-family research—the central problem is how to spend time with their families when their jobs require long, rigid work hours (Bianchi, 2011; Blair-Loy, 2003). For women working low-wage jobs, the work-family dilemma centers on how to work enough hours to support their families financially while still providing sufficient care (Bianchi, 2011; Collins & Mayer, 2010). The dilemma for low-income women is compounded by the underemployment of the men in their lives. Low-income women often shoulder the burden of supporting their families without adequate contributions from their children’s fathers, who often face structural barriers to employment and involvement in their children’s lives (Christiansen & Palkovitz, 2001; Landale & Oropesa, 2001; Townsend, 2002). These circumstances characterize what race and poverty scholars have described (directly or indirectly) as the “the third burden:” socially disadvantaged women, particularly Black women, must contend with not only their own positioning within race or class and gender structures but also the fate of the men to whom they are connected (Bianchi, 2011; Bianchi & Milkie, 2010; Edin & Kefalas, 2005; Malveaux, 1990; Roberts, 2004).

The fate of disadvantaged men is not limited to underemployment. The expansion of mass incarceration over the past 50 years coupled with overrepresentation of racial/ethnic minorities in both jails and prisons means that incarceration has become a common experience for Black men and other racially and socioeconomically disadvantaged men (Carson, 2015; Glaze & Kaeble, 2015; Harlow, 2003; Pettit & Western, 2004). Consequently, the removal of men from family systems has become a common event for disadvantaged women (Lee, McCormick, Hicken, & Wildeman, 2015). If, as research suggests, the conditions of low-wage jobs make balancing work and family difficult (Henly & Lambert, 2014; Williams, 2010), then the predominance of family member incarceration in the same women’s lives may create additional complications. We know that women’s family obligations often structure their labor force participation. Marriage, child birth, husband unemployment and divorce—family events that alter caregiving responsibilities and economic resources—shape women’s employment, particularly for White and higher-income women (Bradbury & Katz, 2002; Brewster & Rindfuss, 2000; Cohen & Bianchi, 1999; Looze, 2017; Lundberg, 1985). Incarceration of women’s partners—which also alters caregiving responsibilities and economic resources (Arditti, Lambert-Shute, & Joest, 2003; Comfort, 2008; deVuono-powell, Schweidler, Walters, & Zohrabi, 2015; Geller, Garfinkel, & Western, 2011)—may be an important family event for low-income and racial/ethnic minority women and their employment.

In this paper, I use data from the Fragile Families and Child Wellbeing Study (FFCWS), a longitudinal study of mostly unmarried parents, to examine how men’s incarceration is associated with multiple job holding among their female partners. In an unstable low-wage labor market where employers often rely on part-time staff (Kalleberg, 2000; Tilly, 1996), multiple job holding may be one way that low-skill women with incarcerated partners work enough hours to support their families financially. Because the association between partner incarceration and multiple job holding may depend on the strength of family ties and the conditions of incarceration, this study also considers how the association varies by residential status and incarceration duration. Previous research has shown that the incarceration of a woman’s partner is not, on average, associated with the number of hours she works at her main job (Author, 2017). If, instead, partner incarceration increases the likelihood that a woman takes on a second job, it would suggest that these circumstances not only constitute a third burden but are literally linked to a “third shift”—to additional employment necessary for meeting the needs of family members both inside and outside of prison.

2. BACKGROUND

2.1. Linking Partner Incarceration and Women’s Employment

There are several reasons to expect that men’s incarceration will prompt their female partners to take on additional jobs. First, men’s removal from household wage earning can be financially destabilizing for families (deVuono-powell et al., 2015; Johnson, 2008; Schwartz-Soicher, Geller, & Garfinkel, 2011). Although their pre-incarceration wages can be modest, men often provide a primary source of income to their families prior to incarceration (Braman, 2004; Glaze & Maruschak, 2010; Travis, McBride, & Solomon, 2005). Even women who do not live with their children’s fathers lose important financial resources when their children’s fathers are incarcerated, as non-residential fathers often make contributions in the form of child support (Geller et al., 2011; Nepomnyaschy & Garfinkel, 2007). The incapacitation effect of incarceration on men’s employment can deprive families of income in the short- and long-term, as release from prison does not ensure a return to wage earning. Men with criminal records often struggle to find work and earn low wages when they do find work (Holzer, 2009; Kling, 2006; Pager, 2003; Western, 2006). Even an arrest or jail stay can jeopardize employment by interrupting work attendance (Fernandes, 2015; Grogger, 1995; Sullivan, 1989). The loss of one job, even temporarily, may compel a woman to work an extra job herself.

In addition to lost income, families also face a host of expenses associated with incarceration. Those who wish to maintain contact with individuals during their imprisonment must pay for phone calls, emails, packages, and travel to jails and prisons (Comfort, 2008; Grinstead, Faigeles, Bancroft, & Zack, 2001). In addition, families often serve as primary financial support for inmates and take responsibility for bail, attorney fees, fines, and legal debt associated with involvement in the criminal justice system (deVuono-powell et al., 2015; Harris, 2016). A participant in a study conducted by the Ella Baker Center for Human Rights and coordinating organizations (deVuono-powell et al., 2015) described the financial burden of having a family member involved in the criminal justice system:

Money you spent for lawyers, money you spent for trying to find investigators and whatever you need to try to help your loved ones, so they don’t have to do serious jail time. Then when they’re in jail you try to make sure you take care of the commissary and you take care of their children. You almost have to have another part time job. (p. 13)

There are several ways to use paid labor to generate additional income when a family member is incarcerated. Women may get a second job, but they also may work more hours at a single job or find a job that pays more per hour. If, as prior research suggests, partner incarceration does not lead to an increase in the number of hours women work at their primary jobs (Author, 2017), working a second job may be how women earn additional income when their partners are incarcerated. Indeed, multiple job holding may be a particularly important strategy given the constraints disadvantaged women often face in the labor market. Less-educated and racial/ethnic minority women—women for whom family member incarceration is most common (Lee et al., 2015)—often work in low-wage and service sector jobs (Malveaux, 1981; Presser, 2003; Reid, 2002; Reskin, 1999). The number of hours a woman can work at a low-wage job may be limited by her employer’s efforts to keep cost low by maintaining a part-time staff (Kalleberg, 2000; Tilly, 1996), and the low pay may mean that working even 40 hours per week does not result in enough earnings to support a family (Ehrenreich, 2001). Labor economists refer to insufficient work hours as the “hours constraint” motivation for multiple job holding (Averett, 2001; Renna, 2006). Individuals work second jobs because their primary jobs do not offer as many hours as they would like—or need—to work. Although economic theory suggests that some individuals work multiple jobs in order to participate in activities that interest them (Averett, 2001), survey data show that economic reasons predominate, particularly for racial/ethnic minorities and individuals with less education (Hipple, 2010; Martel, 2000). Additionally, research suggests that workers often choose to work second jobs because of family or economic circumstances that temporarily reduce economic resources (Guariglia & Kim, 2004; Krishnan, 1990). For example, Krishnan (1990) found that husbands often hold second jobs to substitute for their wives’ absence from labor force participation, perhaps in cases where women stay home to care for young children. It is possible that women do the same: work multiple jobs when their male partners are unable to participate in paid labor.

Working multiple jobs has the potential to alleviate financial burdens by allowing women to earn the money they need to support their families while also diversifying risk in an unstable labor market. Nonetheless, multiple job holding can have negative consequences for individuals and families. Studies have shown that multiple job holders sleep less than single job holders, have a heightened risk of work- and non-work-related injury, and experience significantly more work-family conflict and higher perceived stress (Australian Bureau of Statistics, 2009; Henly & Lambert, 2014; Marucci-Wellman, Lin, Willetts, Brennan, & Verma, 2014; Marucci-Wellman, Lombardi, & Willetts, 2016; Marucci-Wellman, Willetts, Lin, Brennan, & Verma, 2014). Working multiple jobs takes time and energy to not only work the second job but also commute from one job to another. This added expenditure of time and effort and associated strain limits time for self-care and makes it difficult to fulfill family responsibilities (Australian Bureau of Statistics, 2009; Henly & Lambert, 2014; Marucci-Wellman, Lin, et al., 2014; Marucci-Wellman et al., 2016; Zeytinoglu, Lillevik, Seaton, & Moruz, 2004). For women connected to incarcerated men, working multiple jobs may compound the stress they already experience—stress related to managing already limited time and resources, worrying about inmates’ well-being, tension among family members, and coping with shame and stigma (Braman, 2004; Daniel & Barrett, 1981; deVuono-powell et al., 2015; Fishman, 1990).

2.2. Differences by Residential Status and Incarceration Duration

The extent to which family life is disrupted by incarceration may differ based on the context in which the event occurs. Thus, partner incarceration may increase multiple job holding for some women while not altering (or decreasing) the jobs worked by other women. Prior research has found stronger adverse effects of incarceration among families living together before incarceration (Geller, Cooper, Garfinkel, Schwartz-Soicher, & Mincy, 2012; Geller & Franklin, 2014; Turney, 2014). It is plausible that women who have partners removed from their homes feel the financial impact of incarceration most acutely and adjust their employment accordingly. We would expect to see a weaker (or null) association between multiple job holding and incarceration for women who do not live with their partners—women who are connected to incarcerated men primarily through their shared children. Women in non-residential and co-parenting relationships will likely feel the financial impact too because they often receive formal and informal child support (Geller et al., 2011; Nepomnyaschy & Garfinkel, 2007); however, these women may have structures in place to support fluctuation in the contributions of their children’s fathers. They may have re-partnered or rely more heavily on female kin and others in their networks (Edin & Lein, 1997; Patillo-McCoy, 1999; Stack, 1974), making the economic shock less profound.

Additionally, the association between partner incarceration and multiple job holding may vary by conditions of the incarceration itself. For instance, the amount of time a woman’s partner spends in jail or prison may shape the financial impact it has on families as well as women’s employment decisions. On the one hand, incarceration of a few months or less may compel women to maintain their levels of employment because their partners’ financial contributions are missing for only a short period of time. Longer incarceration ensures greater lengths of time without men’s contributions and a potentially heightened need to take on a second job to make up for the lost income. Evidence suggests that longer jail and prison stays are more detrimental to families than shorter stays, reducing the chances of nuclear living arrangements, fathers’ involvement with their children, and children’s contact with paternal grandparents (London & Parker, 2009; Turney, 2014; Turney & Wildeman, 2013). Thus, long spells of incarceration may diminish family resources by incapacitating men for longer periods of time and by weakening family ties. On the other hand, economic theory suggests that women alter their employment in response to temporary, rather than long-term, losses of household income. Long-term losses are expected to lead to an adjustment in consumption (Lundberg, 1985). For women balancing work and family, multiple job holding may be a more feasible short-term response to partner incarceration than long-term response. Prior research suggests that short-term incarceration, compared to long-term incarceration, is more strongly associated with women’s work hours at their main jobs (Author, 2017).

3. DATA, MEASURES, AND ANALYTIC STRATEGY

3.1. Data

To examine the relationship between partner incarceration and multiple job holding, I use data from the Fragile Families and Child Wellbeing Study (FFCWS). This longitudinal study follows a sample of nearly 5,000 couples living in 20 large U.S. cities with children born between the years 1998 and 2000 (Reichman, Teitler, Garfinkel, & McLanahan, 2001). Initial surveys were conducted with both mothers and fathers shortly following the birth of their child, and subsequent interviews were conducted one, three, five, nine, and fifteen years later. The FFCWS oversampled unmarried parents, resulting in a racially/ethnically diverse sample of socioeconomically disadvantaged families, many of whom have experienced incarceration. Nearly half of fathers have spent time in jail or prison by the five-year survey. Although men who have been incarcerated are a highly disadvantaged group, FFCWS sample members with no incarceration history have relatively low incomes and educational attainment as well, making them and their families a valuable comparison group in the examination of incarceration’s unique impact on family life. These data are well-suited to study the consequences of partner incarceration for at least two other reasons. First, the study interviews both mothers and fathers; information from both incarcerated men and their partners is a rare feature of datasets used to study the consequences of incarceration for individuals. Second, the survey asks respondents to report on a variety of individual and family characteristics, such as social and material well-being and father involvement, which allows for the inclusion of a wide range of control variables that may affect the likelihood of both partner incarceration and employment.

Men’s incarceration is most accurately measured between the three- and five-year surveys (for details, see Section 3.2). Thus, I rely primarily on data from these two surveys, although some covariates are measured using baseline data. I restrict the sample to women who completed both the three- and five-year surveys and provided information on multiple job holding. This results in a final analytic sample of 3,835 women. In total, 1,051 (21%) women were dropped because they did not participate in either the three- or five-year surveys.¹ An additional 12 (0.3%) women were dropped because they had missing data on the outcome variable. In the final sample, about 8 percent of observations had missing values for incarceration, and between 0.1 percent and 13 percent had missing values for other covariates (most control variables were missing for 2 percent or fewer of the observations). It is unlikely that the propensity for missing data is independent of both the observed and unobserved data (the only circumstance under which analysis using complete cases produces unbiased results). Thus, to strengthen confidence in the results I used multiple imputation by chained equations (White, Royston, & Wood, 2009) to preserve missing values (implemented in Stata 14). Multiple imputation is preferable to listwise deletion, as it assumes the propensity for missing data is independent only of unobserved data (Allison, 2002; Schafer & Graham, 2002). Although we cannot be certain that missing data is independent of unobserved data, I did ensure that the imputation model fit the observed data well and imputed values were reasonable (Eddings and Marchenko, 2012) (results available upon request). Five imputed data sets were created, and regression results were combined across the five data sets using Rubin’s rule (Allison, 2002; Schafer & Graham, 2002).² I also estimated models using complete cases, which show results nearly identical to those presented (results available upon request).

Analyses of respondents who attrited from the study suggested that they were more disadvantaged than the analytic sample. Women who attrited were more likely to be Hispanic and immigrants. They were less likely than the analytic sample to have received a high school diploma, and more likely to have incomes below the poverty line. I discuss the implications of attrition for the interpretation of the results in the limitations section (Section 5.1).

3.2. Measures

The key outcome variable, multiple job holding, was derived from women’s responses to the question, asked at the five-year survey, “was there a time in the past 12 months that you worked more than one regular job at the same time.” When reporting information on “regular jobs” FFCWS asked respondents to consider any work, including self-employment, for which they received a regular paycheck. Using responses to this question, I created a categorical measure that indicates whether a woman “worked no jobs,” “worked only one job at a time” or “worked multiple jobs at the same time” in the 12 months prior to the five-year survey. “Worked no jobs” and “worked only one job at a time” are alternately set as the reference category in the models in order to compare multiple job holding to both outcome categories. This particular measure of multiple job holding provides an important piece of information about women’s employment. However, we do not know precisely when in the past 12 months women worked multiple jobs or for how long.³ Some models include a lagged measure of the dependent variable, or, more specifically, a measure of women’s multiple job holding status at the three-year survey.

The primary explanatory variable is recent partner incarceration. A woman experienced recent partner incarceration if her partner, the father of the child at the center of the FFCWS survey, was incarcerated between the three- and five-year surveys, including incarceration at the five-year survey. Incarceration during this time period was captured in three ways: direct reports from both women and their partners, indirect reports (e.g., respondent cited incarceration as the reason a man was separated from his child), and administration of men’s five-year survey in prison or jail. Recent partner incarceration is measured between the three- and five-year surveys because it is during this time period that the FFCWS collected the most accurate information regarding incarceration. For instance, the five-year survey asked women not only if the father of the focal child was currently in jail or prison but also if he had been incarcerated since the three-year survey.⁴ The latter question, which was not asked at prior waves, allows the construction of an incarceration measure without any time gaps. Identifying between-survey incarceration is important because some prison and jail stays last less than two years but still have a relevant impact on women and families.

Although the FFCWS provides the best measure of partner incarceration available from survey data, the measure is subject to limitations. The data do not include complete information about the conditions of incarceration, such as whether the partner was incarcerated in jail or prison and the length of sentence. Variation along these lines may result in differential impacts of incarceration on women’s employment and other individual and family outcomes. That said, in a recent article, Wildeman, Turney and Yi (2016) used the incomplete FFCWS information to compare women whose partners’ facility type was known and unknown on a range of outcomes (e.g., relationship dissolution, parenting engagement, parenting stress, depression). They found little evidence of variation across jails, prisons and unknown facilities, which suggests the impact of partner incarceration on women varies little by the type of facility in which their partner is housed (or whether the facility is known or unknown). Although information about the duration of incarceration is available for only 81 percent of incarcerated men in the FFCWS data, research using this information does indicate that family outcomes vary by length of sentence (Turney, 2014; Turney & Wildeman, 2013). Because the length of stay may shape the economic consequences of incarceration for families and thus, women’s employment response, I use this information in analyses that I consider preliminary.

Women connected to incarcerated men presumably differ from other women in ways that influence their employment outcomes. Thus, the analyses adjust for several individual-level characteristics—such as human capital, family structure and economic resources—that may confound the relationship between partner incarceration and multiple job holding. Control variables were chosen both for their theoretical relevance and the explanatory power they added to the model. These variables were measured at or before the three-year survey, prior to the measurement of recent partner incarceration. Demographic controls include: race/ethnicity (non-Hispanic white (reference category), non-Hispanic Black, Hispanic, and other), foreign-born, age, and lived with both parents at age 15. The analyses also control for several measures of human capital, a key determinant of women’s employment. These include educational attainment (less than high school (reference category), high school diploma or GED, some college, and bachelor’s degree or higher), hourly wage at current or most recent job, cognitive scores (derived from the Similarities Subtest of the Weschler Adult Intelligence Scale – Revised [WAIS-R]), and impulsivity (derived from an abbreviated form of Dickman (1990) impulsivity scale). Models also adjust for a categorical measure of weekly work hours at women’s primary formal sector jobs (part-time: 1–34 hours (reference category), full-time: 35+ hours, and did not work), and whether they participated in the informal labor market in the 12 months leading up to the survey. The number of hours available to women at their main jobs and the availability of informal work may influence women’s decisions to work second, formal sector jobs. Self-reported fair or poor health and drug use, factors that may impede women’s ability to hold more than one steady job, are also included. The measure of drug use indicates whether a woman used a nonprescription drug or misused a prescription drug in the past 12 months. The series of survey items concerning drug use follows the Composite International Diagnostic Interview - Short Form (CIDI-SF) list of substances (Kessler, Andrews, Mroczek, Ustun, & Wittchen, 1998).

Several variables measuring family economic resources are also included. Receipt of public assistance indicates whether a woman received TANF, Food Stamps, Medicaid, or Social Security Income (SSI) in the year prior to interview. Models also adjust for whether the partner was employed and perceived financial support, or whether the respondent thought she could count on someone to loan her $1000 or co-sign a $1000 bank loan. Other household and family characteristics include: number of children under the age of five in the household, number of other adults (excluding partners) living in the household, and residential status (living with partner, living with a new partner, and living with no partner (reference category)). Finally, a measure of partner’s prior incarceration indicates whether a woman’s partner was incarcerated at or before the three-year survey. This variable is distinct from recent incarceration; each measure refers to a separate time period in which incarceration occurred. However, the two are not mutually exclusive; in fact, there is considerable overlap between the two variables. About 85 percent of partners who were recently incarcerated were also incarcerated at some point in the past.

3.3. Analytic Strategy

In the first analytic stage, multinomial logistic regression models estimate women’s multiple job holding as a function of their partners’ incarceration. I begin the analysis with an “unadjusted” model, which includes recent partner incarceration, the key explanatory variable, and an indicator of partner’s prior incarceration (Model 1). Model 2 adds control variables in order to isolate the predictive role of recent partner incarceration from other factors influencing multiple job holding. Model 3 adds a lagged measure of the dependent variable, which explicitly assesses change in multiple job holding over time (Finkel, 1995). In Model 4, I restrict the sample to women whose partners were incarcerated at or prior to the three-year survey. In doing so, I diminish some concern about unobserved heterogeneity because the women who remain are connected to men who have a high risk of experiencing incarceration between the three- and five- year surveys. Although this strategy does not eliminate the possibility that any association between partner incarceration and multiple job holding is spurious, it does provide more conservative estimates of the association and is consistent with the literature (Turney, Schnittker, & Wildeman, 2012; Turney & Wildeman, 2013).⁵ All models include city fixed-effects to account for the clustering of respondents within cities. In tables displaying results, I present log-odds coefficients, but I convert these coefficients to odds ratios for interpretation. I also discuss marginal effects, or the difference in the predicted probabilities of multiple job holding for women with incarcerated partners and those without, as they are even more easily interpretable than odds ratios. I use Stata’s -mimrgns- command to estimate marginal effects.

In the second analytic stage, I consider how the association between recent partner incarceration and multiple job holding varies by two characteristics: residential status and incarceration duration. I investigate the moderating role of residential status by including an interaction between women’s year-three residential status (i.e., living with their partner, alone, or with a new partner) and subsequent partner incarceration. I then calculate the marginal effect, or the difference in predicted probability of multiple job holding for women with incarcerated partners and those without, for each residential group. In addition, I consider the relationship between duration of partner incarceration and multiple job holding by using an alternative specification of the explanatory variable. This variable consists of four categories: recent incarceration less than three months (5 percent), recent incarceration three months or greater (10 percent), duration missing (7 percent), and no recent incarceration (77 percent).

3.4. Sample Description

The first column of Table 1 shows sample characteristics for the full sample. Ten percent of women reported working multiple jobs at the five-year survey. Multiple job holders worked about 60 hours per week across all jobs, and 61 percent of multiple job holders worked primary jobs in service or administration (not shown). With respect to the key independent variable, 22 percent of women experienced the incarceration of their partners at some point between the three- and five-year surveys. As we would expect based on FFCWS’s strategy of sampling mainly births to urban, unwed parents, the sample as a whole is relatively disadvantaged across a wide range of other characteristics. Nearly half of the sample is non-Hispanic Black, and 26 percent is Hispanic. Two-thirds of women had no education beyond a high school diploma or GED when their child was born, and more than one-third were living without a partner in the household at the three-year interview. More than 60 percent of women received some form of public assistance, and 44 percent have partners who had been incarcerated by the time the three-year survey was administered.

Table 1.

Sample Characteristics by Partner Incarceration

	Full Sample		Recent Incarceration		No Recent Incarceration
	Mean or Percent	SD	Mean or Percent	SD	Mean or Percent	SD	Significantly Different?
Multiple job holding (y5)
Worked no jobs	24.7		22.8		25.2		*
Worked only one job	65.3		62.0		66.2		*
Worked multiple jobs	10.0		15.2		8.6		*
Recent partner incarceration (y5)	22.0
Race/Ethnicity (b)
Non-Hispanic White	21.6		14.1		23.7		*
Non-Hispanic Black	49.1		64.5		44.8		*
Hispanic	25.8		19.0		27.6		*
Non-Hispanic other race/ethnicity	3.5		2.4		3.8		*
Age (y3)	28.2	6.0	25.9	5.3	28.8	6.1	*
Education (b)
Less than high school	32.8		44.2		29.5		*
High school diploma or GED	30.9		34.1		30.1
Some college	25.2		20.8		26.4		*
Bachelors degree or higher	11.1		.9		14.0		*
Lived with both parents at age 15 (b)	42.8		29.1		46.6		*
Foreign-born (b)	14.3		4.9		17.0		*
Hourly wage at most recent job (y3)	11.5	8.3	9.6	5.3	12.1	8.9	*
Cognitive score (y3)	6.8	2.6	6.5	2.5	6.8	2.7	*
Impulsivity (y3)	2.0	0.6	2.2	0.6	2.0	0.6	*
Weekly work hours at primary job (y3)	20.7
1 to 34 hours	15.1		13.9		15.4		*
35 hour or more	41.3		40.7		41.5		*
Did not work	43.6		45.5		43.1
Informal work (y3)	15.5		18.8		14.6		*
Fair or poor health (y3)	13.3		16.1		12.5		*
Drug use (y3)	6.9		12.2		5.3		*
No. of children <age 5 in household (y3)	1.6	0.8	1.7	0.8	1.6	0.8	*
No. of adults other than partners in HH (y3)	0.5	0.9	0.6	1.0	0.4	0.9	*
Residential status (y3)
Living with no partner	36.3		58.9		29.9		*
Living with partner (child’s father)	54.0		24.5		62.3		*
Living with new partner	9.7		16.7		7.8		*
Public assistance (y3)	62.7		82.2		57.1		*
Perceived financial support (y3)	67.1		52.2		71.3		*
Partner’s employment status (y3)	75.6		49.4		83.0		*
Partner prior incarceration (y3)	43.7		87.3		31.5		*
N	3,835		831–851		2,984–3,004

Note: Timing of variable measurement in parentheses; b = baseline survey, y3 = three-year survey, y5 = five year survey. HH = household. The size of the recent partner incarceration subsample varies by imputed data set; the range is shown. Significant differences between recent incarceration and no recent incarceration are indicated by an asterisk (p < .05)

Table 1 also shows differences in sample characteristic by recent partner incarceration. Women whose partners have been incarcerated are more disadvantaged than those whose partners have not been incarceration across nearly all characteristics. For instance, they are more likely to be Black, have less education, and are less likely to have grown up with both biological parents. Additionally, women with incarcerated partners earned less per hour at their most recent job and were more likely to have worked in the informal labor market. Their health was worse and their likelihood of drug use was higher. They were also less likely to have lived with their partners at the three-year survey, more likely to have received some form of public assistance and less likely to have a family member or friend to rely on for financial support.

4. RESULTS

4.1. Estimating Women’s Multiple Job Holding

Table 2 shows results from multinomial logistic regression models estimating multiple job holding as a function of recent partner incarceration. In Model 1, which adjusts only for partners’ prior incarceration, there is a strong, significant association between recent partner incarceration and multiple job holding, whether compared to working one job or no jobs. For the comparison between multiple and single job holding, the magnitude of the association decreases but remains statistically significant when control variables (Model 2) and a lagged measure of the dependent variable (Model 3) are added to the model. In Model 3, the coefficient of 0.431 for working multiple jobs versus working one job corresponds to an odds ratio of 1.54 (e^.431); thus, recent partner incarceration is associated with a 54 percent higher odds of working multiple jobs, net of women’s prior multiple job holding status. For the comparison between multiple job holding and not working, the incarceration coefficient remains statistically significant when controls are added (Model 2) but is reduced to non-significance once the lagged dependent variable is included (Model 3). The marginal effect for Model 3 indicates that the predicted probability of multiple job holding is 3.4 percentage points higher for women whose partners are incarcerated, compared to those whose partners are not. Model 4, which includes all covariates from Model 3, restricts the sample to women whose partners were previously incarcerated. In this restricted sample, recent partner incarceration is again associated with higher odds of working multiple jobs (vs. working one job).

Table 2.

Multinomial Logistic Regression Model Estimating Multiple Job Holding, All Coefficients

	Model 1		Model 2		Model 3		Model 4
	Unadjusted		+ Controls		+ Lagged DV		Restricted Sample
	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job
Recent partner incarceration	0.604***	0.577**	0.465**	0.443*	0.431*	0.346	0.459*	0.357
	(0.149)	(0.188)	(0.165)	(0.218)	(0.169)	(0.225)	(0.211)	(0.282)
Partner prior incarceration	0.041	0.104	−0.076	0.105	−0.052	0.130
	(0.132)	(0.152)	(0.143)	(0.172)	(0.144)	(0.176)
Race/ethnicity (ref. non-Hispanic White)
Non-Hispanic Black			0.123	0.524*	0.115	0.369	−0.126	0.026
			(0.173)	(0.204)	(0.177)	(0.212)	(0.261)	(0.314)
Hispanic			0.246	0.457	0.212	0.384	−0.101	−0.177
			(0.211)	(0.244)	(0.214)	(0.252)	(0.318)	(0.380)
Non-Hispanic other race/ethnicity			0.131	0.087	0.015	−0.109	0.050	−0.050
			(0.380)	(0.427)	(0.385)	(0.439)	(0.624)	(0.741)
Age			−0.022	−0.068***	−0.017	−0.052***	−0.004	−0.043*
			(0.012)	(0.014)	(0.012)	(0.014)	(0.019)	(0.022)
Education (ref. Less than high school)
High school diploma or GED			−0.026	0.259	−0.081	0.201	−0.130	0.066
			(0.152)	(0.173)	(0.156)	(0.179)	(0.208)	(0.241)
Some college			0.151	0.666**	0.061	0.558**	−0.014	0.309
			(0.167)	(0.199)	(0.171)	(0.206)	(0.244)	(0.297)
Bachelors degree or higher			0.245	0.522	0.198	0.474	−0.185	0.946
			(0.276)	(0.319)	(0.279)	(0.329)	(0.659)	(1.012)
Lived with both parents at age 15			−0.307*	−0.415**	−0.319*	−0.373*	−0.327	−0.476*
			(0.130)	(0.151)	(0.132)	(0.156)	(0.197)	(0.231)
Foreign-born			−0.269	−0.332	−0.145	−0.160	−0.274	−0.233
			(0.242)	(0.268)	(0.245)	(0.275)	(0.490)	(0.544)
Hourly wage at most recent job			−0.007	0.019	−0.008	0.013	−0.001	−0.003
			(0.009)	(0.011)	(0.009)	(0.011)	(0.016)	(0.017)
Cognitive score			0.056*	0.069*	0.058*	0.062*	0.110**	0.120**
			(0.025)	(0.028)	(0.025)	(0.029)	(0.037)	(0.043)
Impulsivity			0.116	−0.012	0.105	−0.047	−0.004	−0.158
			(0.096)	(0.112)	(0.098)	(0.116)	(0.139)	(0.164)
Weekly work hours (ref. 1 to 34 hrs)
Did not work			−0.355*	−1.698***	−0.097	−0.718**	0.080	−0.269
			(0.175)	(0.203)	(0.190)	(0.221)	(0.274)	(0.319)
35 hours or more			0.058	0.916***	0.036	0.871***	0.137	1.114**
			(0.156)	(0.207)	(0.159)	(0.209)	(0.245)	(0.321)
Informal work			0.585***	0.607***	0.439**	0.487**	0.399	0.257
			(0.139)	(0.168)	(0.144)	(0.175)	(0.205)	(0.248)
Fair or poor health			0.076	−0.331	−0.030	−0.535**	0.029	−0.345
			(0.175)	(0.194)	(0.181)	(0.202)	(0.249)	(0.275)
Drug use			0.505**	0.578*	0.452*	0.457	0.101	0.264
			(0.194)	(0.235)	(0.199)	(0.245)	(0.270)	(0.333)
No. of children <age 5 in household			0.045	0.014	0.057	0.022	0.088	0.090
			(0.073)	(0.083)	(0.074)	(0.085)	(0.101)	(0.115)
No. of adults other than partners in HH			−0.154*	−0.221**	−0.136	−0.227**	−0.080	−0.204
			(0.075)	(0.084)	(0.076)	(0.086)	(0.105)	(0.117)
Residential status (ref. living with no partner)
Living with partner (child’s father)			−0.245	−0.396*	−0.177	−0.313	0.165	0.009
			(0.138)	(0.165)	(0.139)	(0.169)	(0.201)	(0.239)
Living with new partner			0.074	−0.063	0.124	−0.039	0.312	0.073
			(0.186)	(0.225)	(0.190)	(0.233)	(0.233)	(0.278)
Public assistance			−0.164	0.026	−0.202	−0.053	−0.295	−0.541
			(0.144)	(0.173)	(0.147)	(0.178)	(0.218)	(0.285)
Perceived financial support			−0.208	−0.121	−0.184	−0.131	−0.355*	−0.268
			(0.127)	(0.149)	(0.129)	(0.151)	(0.180)	(0.208)
Partner’s employment status			0.025	0.332	−0.052	0.242	−0.062	0.152
			(0.150)	(0.177)	(0.153)	(0.184)	(0.191)	(0.234)
Multiple job holding, y3 (ref. no jobs)
Worked only one job					0.308	1.701***	0.418	1.632***
					(0.237)	(0.243)	(0.330)	(0.339)
Worked multiple jobs					1.615***	2.922***	1.896***	2.958***
					(0.266)	(0.309)	(0.375)	(0.435)
Constant	−2.547***	−1.800***	−2.144**	−0.132	−2.861***	−2.193**	−3.473***	−1.853
	(0.280)	(0.295)	(0.620)	(0.711)	(0.681)	(0.776)	(0.961)	(1.109)
AIC	6532.65–6541.688		5752.298–5761.575		5496.837–5504.087		2608.004–2,627.009
Observations	3,835		3,835		3,835		1,673–1,680

Note: ref. = reference, DV = dependent variable. HH = household. Standard errors in parentheses. All models include city fixed-effects. All covariates are measured at the three-year survey with the exception of recent partner incarceration (measured at year five), race/ethnicity, education, lived with both parents at age 15, and immigrant status (measured at baseline). AIC values vary across the five imputed data set and cannot be averaged using Rubin’s rules; the range for each model is shown. The sample for Model 4 is restricted to women whose partners have a history of incarceration; sample size varies by imputed data set.

^*p<.05

^**p<.01

^***p<.001 (two-sided tests).

A number of control variables are also associated with multiple job holding. Because of limited research on multiple job holding, I summarize results from Model 3 here. Age is negatively associated with multiple holding (vs. not working). Women with some college education are more likely than those with no high school diploma to work multiple jobs (vs. no job). Higher cognitive scores, weekly work hours and participation in the informal labor market at the three-year survey, are also positively associated with multiple job holding at the five-year survey. Women who report poor health are less likely to work multiple jobs than to work no jobs, but drug use is positively associated with multiple job holding (vs. single job holding). Women who lived with both parents at age 15, a common measure of childhood advantage, are less likely to work multiple jobs. Finally, more adults (i.e., extended family, roommates) living in the household reduces the likelihood that a woman works multiple jobs (vs. no jobs).

4.2. Differences by Residential Status and Incarceration Duration

In the second analytic stage, presented in Figure 1 and Table 3, I consider variation in the association between recent partner incarceration and multiple job holding. I turn first to the moderating role of residential status. Figure 1 shows the difference in predicted probability of working multiple jobs for women with incarcerated partners versus women without incarcerated partners, across three residential statuses.⁶ Differences in probabilities shown in Figure 1 are based on a multinomial logistic regression model estimating multiple job holding that includes an interaction between incarceration and residential status along with all controls, city fixed effects, and a lagged measure of multiple job holding. Table A1 in the Appendix presents log odds coefficients from these models. The results indicate that there is a significant difference among only one group: women living with their partners at the three-year survey. Among these residential couples, women with incarcerated partners have a 7.1 percentage point higher probability of working multiple jobs, compared to women whose partners have not been recently incarcerated.

Figure 1.

Difference in Probability of Working Multiple Jobs for Women with Incarcerated Partners versus Women without Incarcerated Partners, by Residential Status

Note: 95% confidence intervals are shown.

Table 3.

Multinomial Logistic Regression Model Estimating Multiple Job Holding with Alternative Specification of Recent Partner Incarceration: Duration

	Model 1		Model 2		Model 3
	Baseline		+ Controls		+ Lagged DV
	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job
Recent partner incarceration (ref. no incarceration)
Less than 3 months	0.907***	1.058***	0.770***	0.905**	0.668**	0.709*
	(0.206)	(0.254)	(0.214)	(0.283)	(0.220)	(0.290)
3 months or greater	0.499**	0.555*	0.335	0.396	0.324	0.314
	(0.183)	(0.217)	(0.200)	(0.248)	(0.203)	(0.249)
Incarcerated, missing duration	0.432	0.185	0.295	0.058	0.503	−0.084
	(0.247)	(0.320)	(0.271)	(0.366)	(0.301)	(0.345)
Constant	−2.544	−1.795	−2.086	−0.058	−2.579	−2.014
AIC	6006.281–6007.059		5719.464–5725.842		5056.163–5063.185
N	3,835		8353,835		3,835

Note: DV = dependent variable. Standard errors in parentheses. Models include all controls and city fixed- effects. AIC values vary across the five imputed data set and cannot be averaged using Rubin’s rules; the range for each model is shown.

^*p<.05

^**p<.01

^***p<.001 (two-sided tests).

In Table 3, I present results from the analysis that considers how incarceration duration is related to women’s multiple job holding. To conserve space, coefficients for control variables are not included in Table 3 but are available upon request. Model 1 controls only for partners’ prior incarceration. Model 2 adds the full set of controls. Model 3 further controls for women’s previous multiple job holding status. These results show that incarceration lasting less than three months (compared to no incarceration) is positively associated with multiple job holding. Longer spells, those lasting three months or more, are not significantly associated with multiple job holding once controls are included in the model. In Model 3, the magnitude of the coefficients suggests that short-term incarceration is more strongly associated with multiple job holding than long-term incarceration; however, directly comparing the two duration categories indicates that this difference is not statistically significant (p = 0.19).

4.3. Supplemental Analyses

To test the strength of the association between incarceration and multiple job holding, I run a series of additional analyses. Because multiple job holding could be considered an ordinal rather than a nominal variable, I re-estimate the relationship using a generalized ordinal logistic regression model (Williams, 2006).⁷ The substantive results are unchanged. Results show that women with incarcerated partners are more likely to work multiple jobs (the highest category) than to work either a single job or no jobs (any lower category) (b = 0.454, p < .01).

Additional supplemental analyses restrict the sample in two ways: 1) to women employed at both the three- and five-year surveys and 2) to women not living with a new partner (i.e., a partner other than the child’s father) at the five-year survey. Studies of the causes and consequences of multiple job holding tend to restrict the analytic sample to employed individuals (Husain, 2014; Marucci-Wellman, Lin, et al., 2014), comparing multiple job holders solely to single job holders. This specification produces nearly identical results (b = 0.436, p < .05). When women who live with a new partner—women who have an alternative source of support—are removed from the analytic sample, the association between recent incarceration and multiple job holding is somewhat magnified (b = 0.493, p < .05). This is in line with expectations; women who do not have a new partner to rely on for financial support may be more in need of additional work.

Finally, I consider how the estimated impact of partner incarceration differs from the impact of two other shocks to household income: partner unemployment and loss of financial support from the partner. Although there is certainly overlap between partners who become unemployed and those who stop providing support, I compare incarceration to both experiences for two reasons. First, the comparison to partner unemployment is motivated by a long-standing interest among labor economist in the “added worker effect,” or the notion that a woman will increase her labor supply when her husband becomes unemployed in order to offset a loss of household income (Lundberg, 1985; Mattingly & Smith, 2010; Moehling, 2001). Theoretically, partner unemployment and incarceration may operate on women’s multiple job holding in a similar way. Second, the comparison to loss of financial support broadens this theoretical framework to consider instances in which partners remain employed but discontinue support for other reasons. This scenario may be particularly relevant for non-residential partners, or men and women who share children but do not live together.

To implement these analyses, I follow a strategy similar to that used by Geller and colleagues (2012). I first re-estimate Model 3 from Table 2, described above, and include a categorical variable that indicates whether women experienced partner incarceration, partner unemployment, or neither event between the three- and five-year surveys (reference category: partner unemployment). In a separate model, I replace this variable with one indicating whether women experienced partner incarceration, loss of financial support or neither event (reference category: loss of financial support). The incarceration coefficients shown in Table 4 indicate that the association between partner incarceration and multiple job holding is not significantly different than either the association between partner unemployment and multiple job holding (Table 4, Panel A) or the association between loss of financial support and multiple job holding (Table 4, Panel B). These results suggest that the relationship with multiple job holding is similar across events that can strain household budgets. These shocks may operate in a similar manner by increasing the need for additional income and, consequently, women’s tendency to seek out additional jobs.

Table 4.

Comparing Incarceration to Partner Unemployment and Loss of Partner’s Financial Support

	Model 1
	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job
Panel A. (ref. unemployment)
Recent partner incarceration	0.377	0.115
	(0.297)	(0.265)
Neither incarceration nor unemployment	−0.060	0.033
	(0.276)	(0.219)
AIC	5500.738–5506.909
N	3,835
Panel B. (ref. loss of financial support)
Recent partner incarceration	0.037	−0.436
	(0.233)	(0.234)
Neither incarceration nor loss of partner financial support	−0.471*	−0.018
	(0.198)	(0.311)
AIC	5496.886–5504.213
N	3,835

Note: Standard errors in parentheses. Models include all controls and city fixed-effects. AIC values vary across the five imputed data set and cannot be averaged using Rubin’s rules; the range for each model is shown.

^*p<.05

^**p<.01

^***p<.001 (two-sided tests).

5. DISCUSSION

This study contributes to a large body of research documenting the sensitivity of women’s employment to changing family circumstances. It advances our knowledge by focusing on an overlooked but common family transition in the lives of disadvantaged women: partner incarceration. As such, this study also adds to mounting evidence that the consequences of incarceration spill over into family life, impacting the employment of not only currently and formerly incarcerated men but also their female partners. Specifically, the findings show that partner incarceration is associated with working multiple jobs. Across a variety of model specifications, I find a robust, positive relationship between partner incarceration and multiple job holding, compared to single job holding. This finding is consistent with expectations that changes in income and expenses associated with incarceration (Geller et al., 2011; Grinstead et al., 2001; Johnson, 2008) can lead to shortfalls in family budgets and increase the need for additional income from employment. This finding stands in contrast to a previous study which showed that the incarceration of a woman’s partner does not influence the number of hours she works at her main job (Author, 2017). That women’s employment response to the incarceration of their partners is constrained, by and large, to multiple job holding suggests the economic and employment situations of this population are especially precarious. Additionally, the lack of significant association between incarceration and multiple job holding when compared to not working suggests that incarceration does not necessarily push women into the labor force, but it may push those who are already working into multiple jobs.

There is further evidence that the relationship between partner incarceration and multiple job holding is concentrated among women who had been living with their partners. This finding is consistent with previous research which has shown stronger adverse effects of paternal incarceration on several outcomes for children with co-residential parents (Geller et al., 2012; Turney, 2014). It stands to reason that partner incarceration is particularly disruptive for women who have partners removed from their homes, as they have the greatest potential of losing a primary source of income and accruing costs associated with incarceration. Women living apart from their children’s fathers may have already adapted to his absence and have other resources to rely on when his contributions are inconsistent. The insignificance of incarceration for women living with a new partner supports this conclusion.

In addition, I find that incarceration lasting less than three months is associated with multiple job holding, but incarceration lasting three months or more is not. This suggests that short-term incarceration can be destabilizing for families. The cost of bail and other fees accompanied by their partners’ interrupted work attendance and suspended family contributions, even if for a short time, may prompt women to pick up additional work. Women may find other strategies, perhaps reducing consumption (Lundberg, 1985), for dealing with strained budgets when incarceration spells are longer. This finding is consistent with evidence that short- versus long-term partner incarceration is more strongly associated with the hours women work at their main jobs (Author, 2017). That said, in this study, comparing short-term incarceration to long-term incarceration, rather than to no incarceration, shows that the two duration categories are not differentially associated with multiple job holding. This casts some uncertainty on the importance of incarceration duration for conditioning the strength of the relationship with multiple job holding. These results should be considered preliminary, as a large number of observations had missing information on duration, and the nature of missingness was not random. Better data and additional research are necessary to untangle any relationship that may exist between the length of jail and prison stays and multiple job holding.

5.1. Study Limitations

This study has some limitations. First, the FFCWS data provide few details about the circumstances of incarceration. Information about whether the partner was incarcerated in prison or jail, the duration and precise timing of incarceration, the distance of the prison/jail from home, and the frequency of visits is incomplete or unavailable but may have consequences for women’s time, energy, and financial resources and influence their decisions about whether to work multiple jobs. The measure of multiple job holding is subject to similar uncertainties. We do not know precisely when women worked multiple jobs, for how long, or if they considered working more than one job to be a burden. Because the precise timing of both incarceration and multiple job holding are unknown, it possible that some women worked multiple jobs before the incarceration began. Thus, the results may reflect a relationship not between incarceration and multiple job holding but between criminal justice system involvement more generally (e.g., warrant for arrest), or even engagement in criminal activity, and multiple job holding. Such relationships would be consistent with evidence that second jobs are sometimes used as a precautionary strategy when unemployment risk is high and family earnings are uncertain (Boheim & Taylor, 2004; Guariglia & Kim, 2004). A woman may need to take on a second job long before jail or prison incarceration occurs. To be sure, additional information regarding the timing of multiple job holding would lead to a better understanding of how women prepare for or respond to changes in family circumstance using their employment.

Although the study design incorporates several features that increase confidence in the results—such as adjustment for a wide range of potential confounders, the inclusion of a lagged measure of multiple job holding, use of a reference group with a heightened risk of partner incarceration, and a series of robustness checks—threats to causal inference do remain. I cannot rule out the possibility that unmeasured variables increase the likelihood of both incarceration and multiple job holding resulting in a spurious relationship between the two. For instance, the rising costs of raising children may lead mothers to work more and fathers to engage in criminal activity in order to earn additional income.

Finally, the FFCWS data remain the most suitable data for answering questions regarding the impact of partner incarceration on women, as most longitudinal data sets used for incarceration research (e.g., National Longitudinal Survey of Youth, National Longitudinal Study of Adolescent to Adult Health) do not include information about partners of incarcerated individuals. However, the results are not generalizable to the full U.S. population. The data represents only families who gave birth in large U.S. cities between 1998 and 2000. Attrition may also influence the findings. Women who attrited from the study are more economically disadvantaged and more likely to be Hispanic. This could impact the findings in two ways. On the one hand, the results may underestimate the association between partner incarceration and multiple job holding because incarceration is more common among low-income and racial/ethnic minority families. On the other hand, results may be biased upward because multiple job holding is slightly less common among less-educated, Black and Hispanic women (Hipple, 2010; Husain, 2014). Although I cannot be certain, conflicting evidence regarding the direction in which attrition could influence results suggests that substantive results would not change if all women had remained in the study.

5.2. Conclusion

Despite these limitations, this study provides evidence that having an incarcerated partner is linked to a “third shift.” This has implications for family well-being. When women respond to family transitions and changes in household budgets by taking on additional jobs, it may be good for their families in that it helps to cover basic expenses. However, additional jobs add to the work responsibilities and expanded “second shift” women with incarcerated partners already face. Staying in touch and supporting an inmate—responding to his requests for food, clothing and books, preparing packages to the correctional institution’s specifications, coordinating family member visits, and keeping up with legal cases and appeals—can feel like a second job in and of itself (Comfort, 2008). In addition to caring for inmates, women often bear the responsibility of caring for their children. Sole responsibility for childcare, particularly when children are struggling with the absence of their fathers (Geller et al., 2012; Haskins, 2015), may make working a second job challenging. If multiple job holding induces stress, as prior research indicates (Australian Bureau of Statistics, 2009; Henly & Lambert, 2014), the results of this study suggest that multiple job holding could be one pathway through which incarceration impacts stress-related illnesses (Lee, Wildeman, Wang, Matusko, & Jackson, 2014). Future research should examine the consequences of multiple job holding for health and well-being and the possibility that increased employment or role conflict mediates the relationship between family member incarceration and adverse health outcomes.

Furthermore, it is unlikely the multiple jobs worked by partners of incarcerated men are “good jobs,” or jobs that build human capital. It is possible that multiple job holding among low-skill and other disadvantaged groups puts them further behind, despite any short-term benefits for monthly income. Indeed, balancing multiple work roles in addition to family member incarceration may keep women from going to school or participating in other activities that improve their socioeconomic standing over the long-term. By further constraining the time and energy of the most disadvantaged women, multiple job holding may be one of many ways that partner incarceration inhibits women’s ability to reach their full social and economic potential, thereby perpetuating and deepening inequalities between women. We know little about the short- and long-term consequences of multiple job holding for families or about how the consequences vary across the population. These results suggest a need for future research that more directly examines how both multiple job holding and mass incarceration impact economic well-being and socioeconomic inequality more broadly.

Appendix

Table A1.

Multinomial Logistic Regression Model Estimating Multiple Job Holding with Interactions

	Model 1		Model 2
	All controls and lagged DV		+ Interaction with residential status
	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job	Worked Multiple v. 1 Job	Worked Multiple v. 0 Job
Recent partner incarceration	0.431*	0.346	0.231	0.174
	(0.169)	(0.225)	(0.223)	(0.273)
Residential Status (ref. living with no partner)
Living with partner (child’s father)	−0.177	−0.313	−0.338*	−0.477*
	(0.139)	(0.169)	(0.164)	(0.197)
Living with new partner	0.124	−0.039	0.080	0.044
	(0.190)	(0.233)	(0.254)	(0.326)
Residential Status x recent incarceration
Living with partner x recent incarceration			0.567	0.634
			(0.302)	(0.365)
Living with new partner x recent incarceration			0.115	−0.198
			(0.416)	(0.536)
Constant	−2.861***	−2.193**	2.731**	−2.050***
	(0.681)	(0.776)	(0.685)	(0.780)
AIC	5496.837–5504.087		5500.318–5506.022
N	3,835		3,835

Note: DV = dependent variable. Standard errors in parentheses. Models include all controls and city fixed-effects. AIC values vary across the five imputed data set and cannot be averaged using Rubin’s rules; the range for each model is shown.

^*p<.05

^**p<.01

^***p<.001 (two-sided tests).