Introduction

Life course criminology was slow to respond to calls to include females in both theory and research [1]. This was due, in large part, to the low base rate of female offending compared to male offending as well as the lack of empirically rich longitudinal data on females [2, 3]. In recent years, however, the life course literature on female offending has grown substantially. In fact, because of the increased rate of female offending and more recent longitudinal data collection efforts focused on females, we have witnessed a number of corresponding advances in our understanding of the theoretical mechanisms that shape the course of female offending from onset to desistance.

As exciting as these advances are, the theoretical and empirical literature on female offending pathways is at a distinct disadvantage compared to the literature on male offending pathways. The female offending literature lacks a historical context against which we can assess the scope and breadth of empirical findings and their related theoretical implications. This is important for at least two key reasons. First, it is unclear whether the theoretical mechanisms documented in contemporary samples are unique to the current socio-cultural context of females or are generalizable over time. Second, even if the mechanisms extend beyond the current socio-cultural context, there are bound to be historical contingencies that are only evident when comparing findings across historical contexts. This has obvious retrospective and contemporaneous implications, but more important, it has implications for the robustness of our theories moving forward.

Understanding how historical contexts influence relevant life course events and offending patterns is important for establishing the scope and generalizability of core theoretical mechanisms. To address this concern, we use data culled from the comprehensive life histories of 500 female offenders originally collected by Sheldon and Eleanor Glueck in the early twentieth century. These data provide a historical backdrop against which we compare current theoretical and empirical findings on female offending. Specifically, we focus on desistance, an area that has produced a theoretically and empirically rich life course literature on females over the past 15 years [4].

Before exploring the historical context of female offending pathways, it is important to define desistance. Desistance is the process by which individuals transition from offenders to non-offenders [5]. Contemporary thinking around desistance implicates both external and internal change mechanisms in this process. More specifically, theoretical and empirical accounts suggest that, for both females and males (though with important differences in degree and kind), desistance occurs via external change mechanisms realized through ties to social institutions that generate social capital and social control [6–8]. More recent theorizing suggests that prosocial identity shifts both facilitate and are enhanced by these institutional ties and the related opportunities for change that they introduce [9–11]. By using historical data on female offenders, we offer two important contributions to this body of work. First, focusing on the key external change mechanisms identified in contemporary samples of women and men, we assess the degree to which similar institutional ties contributed to the likelihood of desistance among female offenders nearly 100 years ago. In this way, we assess the robustness and generalizability of these mechanisms over time. At the same time, historical data also allow us to illuminate the specific ways in which historical context matters. Elder (1998) argued that individual lives are embedded in unique historical contexts, shaping the opportunities and constraints that influence the way these lives play out. Applied to desistance, this reminds us that even where broad theoretical mechanisms might apply across historical contexts, the socio-historical experiences of the women we study likely influence the sources and salience of various mechanisms and the specific ways in which these mechanisms shape desistance outcomes. This is important for further establishing the scope of our theories and articulating the socio-cultural conditions under which specific theoretical mechanisms are more or less salient. If our theories cannot adapt in ways that account for the dynamic cultural and structural conditions that characterize the social world and the changing behavioral and social norms that continually redefine the boundaries of crime and deviance, their utility is short lived at best.

The Historical Context of the Glueck’s Data

Sheldon and Eleanor Glueck’s data documenting the life and offending trajectories of 500 female offenders incarcerated in the Massachusetts Reformatory for Women in the early 1920s [12] were collected during a unique and well-documented period in women’s history. The early 1900s was a period marked by important economic and social shifts that were particularly dramatic for women. For example, industrial capitalism firmly replaced the domestic economy and working class women assumed an undeniable presence in the labor market as domestic and factory workers [13]. As a result, the traditional public/private divide that had been institutionalized in the nineteenth century became increasingly difficult to sustain, especially among working class families who relied on the labor of their older daughters and wives to supplement the family income [14–16]. In this context, women’s sexuality became a matter of central concern. Indeed, some criminologists explicitly linked female sexuality and criminality [17]. Young working class women not only worked in the public sphere, but also increasingly flocked to the city’s public venues for leisure [16, 18]. Here, they interacted with young men and engaged in sexual behavior outside of marriage. Such behaviors stood in sharp contrast to predominant middle class values that only recognized women’s sexual desires under the pretense of matrimony and motherhood. Collecting their data during this era, it is hardly surprising that, like Pollack, the Glueck’s also implicated women’s sexuality (among other individual, familial, contextual, and structural influences) in their offending behavior [12].

In response to the increasingly overt sexuality of working class girls, middle class reformers launched a variety of social campaigns designed to curb immoral behavior and preserve women’s purity. While many of these campaigns relied on informal social controls, corresponding efforts on the part of formal social control agents also intensified, criminalizing the sexual behavior of young, and largely poor, women [19]. The women’s reformatory system emerged in this context, employing the formal authority of the state to influence the “suspect” behavior of working class girls [15]. As Freedman [20] and Rafter [21] both document, this system was designed specifically to reform the “fallen woman” who had strayed too far from normative middle class ideals of femininity. Reformatories focused their efforts on young women convicted for vice or moral offenses, reflecting a concern with uncontrolled female sexuality and behaviors that might encourage promiscuity (e.g., smoking, drinking, vagrancy). While their violations were often petty, reformatory sentences were lengthy, with indeterminate sentences of 2 to 5 years being common [21]. Notably, these sentences were imposed for behaviors that were ignored or leniently sanctioned among males [22]. Reformatory personnel justified these lengthy commitments by arguing that the women referred to the reformatory system were in need of intense moral training that could only be realized through a substantial time commitment. The goal of the reformatory, then, was to instill in working class women, middle class norms that hailed marriage as the only proper venue under which sexual behavior could be legitimately fulfilling, and motherhood as its prime goal [15]. It is within this twentieth century context that we examine desistance outcomes for a cohort of predominantly working-class women released from the Massachusetts Reformatory for Women.

External Change Processes and Gender

There is a growing body of scholarship that investigates how external change processes, particularly those realized through social control, influence desistance for females [4]. This work builds directly from Sampson and Laub’s [8] account of age-graded social control processes in the life course offending patterns of males. Their research pointed to the social controls realized through adult institutions like marriage, work, and the military, as central drivers of desistance in adulthood. However, given their exclusively male sample, subsequent scholarship questions whether these same broad processes influence desistance among women. This body of work suggests that the social control mechanisms that facilitate men’s desistance may also have salience for women. However, there appear to be some important differences in the types of institutional ties that motivate desistance for women compared to men. Specifically, research documents some notable gender variation in the roles of marriage, parenthood, and work for the desistance process.

Marriage is the social institution that has garnered the most research attention with respect to gendered desistance patterns. Sampson et al. [23] speculate that the effects of marriage on desistance may be stronger for males. Male offenders, they argue, tend to “marry up” while female offenders tend to “marry down.” So, husbands may not provide the kinds of informal social control and direct monitoring that Laub and Sampson [7] find wives direct towards their husbands. Contemporary research examining potential gender differences in the effect of marriage on desistance generally finds stronger effects for males, but still reports a significant marriage effect among females [24–26]. There are exceptions. Zoutewelle-Terovan et al. [27] find a significant marriage effect for males and no effect for females in a high-risk sample of men and women in the Netherlands, followed for 21 years. Doherty and Ensminger [28] find strong marriage effects for African American men, but weaker and more inconsistent effects among African American women, with marriage actually increasing their odds of a drug arrest. Craig and Foster [29] find marriage to have similar effects on desistance among males and females. In seeming contrast, Giordano et al.’s [6] quantitative results show no significant marriage effect for either males or females, but their qualitative analysis does suggest that, for women, marriage to a prosocial spouse can hasten desistance. Other research reinforces the suggestion that an intimate partner’s behavioral profile is a key predictor of the influence that relationship will have on women’s desistance outcomes. This body of work finds that females are more vulnerable to the influence of a romantic partner than males [30] and that females who persist in criminal offending are more likely to be in relationships with romantic partners who exert an antisocial influence on them [3]. In sum, contemporary research finds that marriage may matter for males and females but its salience in the desistance process is less consistent among women, and at least partly dependent on the behavior of the spouse. The more tenuous effect of marriage more females compared to males may also reflect the contemporary status of marriage as both increasingly less common and more tenuous than was the case in the 1920s when the women in the Glueck sample were incarcerated.

In many ways, marital norms of the era were central to women’s incarceration in the 1920s. The women in the sample that we draw on here were institutionalized largely for breaking moral codes regarding women’s behavior and particularly their sexual behavior. For many, this meant adultery or sex outside of marriage, both of which violated middle class gender norms. For women who came of age in the 1920s, marriage was a virtually universal goal and entry into marriage signaled support for the middle class norms and values thought to guard against sexual waywardness [13, 15]. According to the 1920 census, 75 % of adult women “functioned exclusively as wives, mothers, and housekeepers in their own home” ([13]: 48). For this sample then, marriage may, almost by default, reduce women’s involvement in the crimes against morality for which many of them were incarcerated and also limit their interaction with formal social control agents. However, even in this context, we hypothesize that the strength of the marriage effect will be contingent on the husband’s prosocial influence and on the related quality of the marriage, as is the case today, despite significant shifts in rates of and social mandates for marriage among women.

On balance, contemporary research indicates that marriage can influence women’s desistance but that it is more salient for males. Indeed, Opsal [31] reports that none of the parolees in her qualitative study of female re-entry invoke marriage or romantic partnerships as central to their successful reintegration. Kraeger et al. [32] suggest that it is motherhood not marriage that is the more salient driver of desistance among females. Analyzing data from a sample of females residing in high-risk Denver neighborhoods, Kreager and colleagues find that the effect of marriage is insignificant once they account for the influence of motherhood on women’s desistance. This is consistent with scholarship showing that women continue to prioritize childcare over paid work, but they no longer prioritize other household duties long associated with the “good wife” role [33]. Indeed, marital equality declines in the transition to parenthood as women’s priorities shift from paid work to childcare [34].

Qualitative accounts, including Giordano et al.’s [6], reinforce the centrality of motherhood in women’s desistance narratives. Edin and Kefalas’ [35] interviews with single mothers in disadvantaged Philadelphia neighborhoods strongly implicate motherhood in these women’s efforts to turn their lives around. This finding is echoed in Hunt et al.’s [36] qualitative study of female gang members, who report restricting their alcohol consumption and spending less time on the street once they have children. Bachman et al. [37] find that motherhood does not directly influence desistance, but can help solidify prosocial identities following successful desistance from crime and drugs. One likely reason for the tenuous relationship between motherhood and desistance is that motherhood is a double edged sword—societal expectations of appropriate mothering often place unrealistic demands on lower class women that can both criminalize their “improper” enactment of the mothering role and compel them to crime in an effort to manage the demands of mothering [22, 38]. It is precisely these push and pull factors that help explain why motherhood often does not emerge as a significant predictor of desistance in quantitative studies [32].

Motherhood, then, may be linked to women’s entry into the system, but it can also compel women to prioritize desistance for the sake of their children. Indeed, qualitative studies have found that separation from children is among the chief sources of anxiety for incarcerated women [39–41] and some quantitative evidence suggests that women with children are more likely to desist than those without [42, 43]. Benda’s [44] study of boot camp participants implicates the number of children women have with more successful re-entry, but this is not the case for the men. This comports with Kraeger et al. [32] findings that convincingly position motherhood as central to women’s desistance process. Given the strong normative expectations around motherhood in women’s lives during the early twentieth century, we expect motherhood to also play prominently in the desistance process for women in the Glueck’s database. Links between motherhood and desistance reinforce the argument that, as is the case for men, external mechanisms can facilitate women’s pathways out of offending.

In addition to external influences tied to the family, for men, ties to work can also activate key desistance mechanisms, especially where those ties represent stable employment opportunities [7, 8, 45–47]. It is not clear whether the same would hold for women. Work has long been central to men’s identities and self worth, but this is not necessarily the case for women. Even as women make inroads towards pay and status equity, progress is slow and women have yet to tie their identities to their careers to the extent that men do [48]. Moreover, gender norms continue to reinforce women’s role as the primary parent responsible for keeping the family and home in order. Consistent with this, research indicates that women still prioritize care giving over paid work [33]. In this sense, work may be less salient for women and less likely to trigger the kind of social control processes that link it to desistance for males. Moreover, even if work were central to women’s desistance, it is a particularly difficult resource for disadvantaged women to access. For example, Giordano et al. [6] report that the women in their sample have difficulty finding work and, when they do find work, it is generally in low wage, unstable, service sector jobs.

Research examining the influence of work on women’s desistance is relatively sparse. Benda [44] finds that job satisfaction has a strong influence on men’s re-entry success following boot camp intervention, but no discernible influence on women’s success. Using data from a sample of drug-abusing female probationers, Griffin and Armstrong [49] found that employment reduced their drug dealing behavior, but not their involvement in other types of offending. Still, some studies do report that work can help women transition out of offending [43, 50–52]. The divergent and inconclusive findings in this domain may reflect the often insurmountable obstacles that restrict women’s employment opportunities post-release [6]. While work may, in principle, offer a pathway towards desistance for women, in practice, opportunities are limited and so women may not attach much salience to this desistance pathway. Still, Opsal’s [31] qualitative study of 43 female parolees in Denver provides evidence to the contrary. The women in her sample view work as a primary driver of desistance, providing both material resources and a sense of purpose that can help them craft a non-criminal identity. At the same time, like the women in Giordano et al.’s [6] study, stable work often proved illusive for these women and their unstable work experiences tended to compromise their successful re-entry. Opsal concludes that “because conventional roles like mother and wife are significantly more accessible to women than ‘worker,’ these may become more compelling ways to continue to do identity work and craft replacement selves” (2012: 399). In this way, our historical sample is not all that different from contemporary ones. For example, even when work may have been the only viable avenue for desistance, moral codes of the era would have clearly preferenced domesticity over work in the public domain. Indeed, it is housekeeping and related domestic tasks that women were explicitly trained for in the Reformatory. So, whereas Sampson and Laub report stable, quality employment to be central to the desistance process for males, it is less clear how work might affect women’s desistance trajectories given the ambiguous role of work for women generally, and especially during the Progressive Era when these data were collected.

The Current Study

Here, we assess the influence of key external change mechanisms (marriage, motherhood, and employment) on desistance for a sample of women transitioning out of criminal justice system supervision in the early 1900s. Our aim is to assess how well these mechanisms apply in a unique historical context. Building from Farrall et al. [9], we argue that understanding women’s desistance patterns requires a nuanced assessment of their social positions and related opportunities. The specific features of women’s social worlds (and indeed men’s) are not historically stable and both theory and policy need to be able to account for the dynamic features of the social worlds women and men navigate. Using the Glueck data, we can assess the role and nature of key desistance mechanisms while also illuminating how varying structural and cultural contexts influence opportunities and expectations for women and shape the specific avenues through which they can access desistance mechanisms. Our research, then, speaks to both broad theoretical desistance mechanisms and the specific ways in which their manifestation reflects gendered and historically contingent social structural contexts.

Methods

Sample

Data for the current investigation come from the Sheldon and Eleanor Glueck’s Women’s Reformatory Study. Details of the original study and related findings are presented in the book 500 Delinquent Women [12]. In brief, the data contain retrospective information on the childhood, adolescent, and early adult backgrounds of 500 female offenders, focusing on physical, psychological, and social characteristics. In addition, they include detailed information on the women’s Reformatory experiences and post-Reformatory adjustment to work and family as well as their offending behavior during a 5-year post-parole follow-up period. Through their research efforts, the Gluecks sought to identify the key factors influencing female offending and to develop a typological schema for classifying female offenders.

While the Gluecks experienced some sample attrition, complete retrospective life history and prospective follow-up data is available for 424 women. Of the 76 women for whom complete data is unavailable (15 % of the total sample), the majority (53 %; N = 40) were excluded from the follow-up as a result of death (24 %; N = 19), institutionalization (24 %; N = 18), or deportation (4 %; N = 3), while the remainder (47 %; N = 36) were never located during follow-up (35 %; N = 27) or had limited/inadequate follow-up data (12 %; N = 9).

Focusing on the 424 women for whom we have complete data (see Table 1 for sample descriptives and Appendix for variable correlation matrix), the average age at the time of their current commitment was 25 years old. For most, this was their first commitment but not their first arrest. Additionally, though the majority are not arrested again once released, 38 % have subsequent interactions with law enforcement, allowing us to split the sample into desisters and persisters to address our central research questions and to compare our findings with contemporary research on the desistance mechanisms that facilitate women’s successful transitions out of prison. Though the sample is overwhelmingly white, most have immigrant roots, with either their mother or father born outside the USA. In addition, most come from homes in which one or both patents lacked formal education and, in many instances were illiterate. Clearly, these women come from challenging circumstances, as is the case with contemporary samples of incarcerated women. That said their offending profiles are distinct from those of contemporary samples of incarcerated women, and reflect prevailing social norms and the related criminalization of women’s sexuality that characterized this era. Not surprisingly, sexual offenses (prostitution, sex/fornication, adultery, lewd and lascivious behavior, or running a “house of ill fame”) rather than drug offenses feature most prominently in the criminal profiles of the women in this sample. Still, property offenses, charges linked to drugs and alcohol, and charges of child abuse or neglect are also evident, suggesting that these incarcerated women are not entirely different from contemporary samples of women in prison. Importantly, this unique context and the historically contingent set of offenses for which this sample of women were incarcerated offers an opportunity to assess contemporary theories against a distinct set of moral norms that play out in a particular social and cultural landscape.

Table 1 Summary statistics for variables used in analyses
Full size table

Procedure

According to the Gluecks’ documentation, the women in this study were selected for participation if their parole from the Women’s Reformatory expired between the years 1921 and 1925. The Gluecks selected cases in consecutive order, beginning in December 1924 and working backward until they reached the target number of cases (N = 500), with the majority of the women originally committed between 1918 and 1922. For each woman in the sample, the Gluecks collected information from a number of sources (e.g., Reformatory files, the offender, family members, employers, social workers, etc.) and then recorded key pieces of data on five standardized data collection instruments documenting each offender’s personal, family, reformatory, parole, and post-parole experiences. Since much of the data used to fill out each assessment instrument was based on offenders’ retrospective accounts of their experiences as documented by Reformatory staff in case records, the validity of the data is an important concern. The Gluecks went to considerable lengths to verify the data they collected from these records, and a detailed account of their verification procedures is documented in their book [12].

To conduct our analyses, we accessed case files for each of the 500 women from microfiche records held at the Henry A. Murray Research Center of Radcliffe College. We converted the data to an automated format using a computerized data entry system. All automated data were validated and, once the database was complete, frequencies in our computerized database were checked against those reported by Glueck and Glueck [12]. Our ability to recreate the summary statistics reported by the Gluecks provides confidence in the reliability of the computerized data.

Dependent Variable—Desistance from Crime

Women in our sample were classified as either desisters (N = 264, 62 %) or persisters (N = 160, 38 %). Specifically, if a woman was arrested during the post-parole 5-year follow-up period, she was classified as a persister. Those who remained arrest-free during this period were considered desisters. While this measure does not capture the processual nature of desistance, it does allow us to assess the influence of key external change mechanisms on women’s odds of desistance post-release. These change mechanisms are implicated in contemporary accounts of women’s desistance processes and as such, assessing their relation to women’s likelihood of re-arrest post-release among this sample helps to establish their theoretical utility across historical contexts.

The 160 persisters accumulated 235 arrests post-parole. The majority of these arrests were for non-sexual moral offenses (47 %), especially alcohol-related offenses as they exited parole during the prohibition. Sex-related offenses were also common (35 %), with the remaining arrests reflecting person/property offenses (18 %), primarily larceny/theft. Table 1 provides summary statistics for all variables in our analyses and shows their distribution by desistance status. Clearly, the arrests that characterize the offending histories of these women reflect the prevailing legal codes and social norms of the early to mid-1900s. Many of the behaviors for which these women were incarcerated would not be considered either criminal or immoral in a contemporary setting. This is important because legal codes and social norms are fluid and our theories need to be able to account for this fluidity by explaining the when and why individuals are likely to violate legal codes and social norms, as well as the conditions under which such violations will be subject to informal and formal sanctions. Consider, for example, the dynamic landscape around marijuana laws and sentences in contemporary settings and its impact on arrest and incarceration trends. Despite these shifts in the scope of drug laws, our theories of offending behaviors, patterns, and processes should remain relevant. Indeed, one of our core contributions is assessing how well the mechanisms implicated in desistance for contemporary samples of female offenders apply in the distinct context that characterized the social and cultural environments of the women in the Glueck’s sample.

Independent Variables

Marriage

The measure of marital status reflect whether the offender was married and living with her husband during the post-parole period (1 = married, 0 = not married). To more fully assess whether and how marriage might matter in the post-parole period, we also examine whether the character of a woman’s spouse in the post-parole period matters, particularly his offending behavior and employment patterns. Women are categorized as having a “pro-social” husband if he has a steady job and no arrest history (1 = prosocial, 0 otherwise). We also assess the influence of marital quality in the post-parole period with a four-item measure that includes: the attitudes of the offender towards marital responsibility, as well as those of her husband (both items coded as 1 = neglects responsibility; 2 = accepts responsibility); the quality of conjugal relations; and the offender’s competence as a homemaker (both items coded as 1 = poor; 2 = fair; 3 = good). We combine these indicators into a factor-weighted marital quality scale (alpha = 0.84). Our focus is on the influence of marriage in the post-parole period but we also control for whether or not women were married and living with their husbands prior to commitment (1 = married pre-commitment; 0 otherwise).

Motherhood

To assess whether having children influences the odds of desistance, we include a measure of motherhood status at commitment (1 = mother prior to commitment; 0 otherwise) and another that documents whether she became a mother in the post-parole period, irrespective of her motherhood status at commitment (1 = mother post-parole; 0 otherwise). Both measures are important since those who were mothers going into the Reformatory will still be mothers on release while others will become mothers after release. As suggested in contemporary research [6, 35, 36], both influences may be important as motherhood may not have kept these women out of the Reformatory; it may nonetheless have given them motivation not to return.

Work

In addition to employment status post-parole and a control for prior employment status (working at commitment and working post-parole: 1 = yes; 0 = no), we also examine women’s commitment to work post-parole for those who were employed during this period (N = 243). The measure is a factor-weighted summed index (alpha = 0.79) based on two items reflecting the offenders work habits (1 = poor; 2 = fair; 3 = good) and the steadiness of her employment (1 = irregular, 2 = fairly irregular; 3 = fairly regular; 4 = regular). In addition, we also include a measure that reflects whether women who are not working alternatively identify as housewives post-parole. In this era, domesticity was viewed as a legitimate alternative to employment for women. Indeed, to many social reformers, this was seen as preferable to employment and an explicit endorsement of the dominant middle class values of the era. Overall, 57 % of the women had jobs in the paid labor force, 23 % identified as housewives, and 20 % had no legitimate employment (the reference group in our analyses).

Control Variables

In addition to controls for marriage and employment prior to commitment, all models include controls for other key measures thought to influence desistance. We control for number of prior arrests. The average number of priors is 3, but 16 % of the sample have no prior arrests and 17 % have 5 or more. Table 1 reports descriptive data based on the full range of priors (0 to 41), but, the regression models use a measure that ranges from 0 to 5 or more to preserve a more normal distribution for this variable. We also control for age at current commitment and early delinquent involvement (first reported delinquency prior to age 14) as well as immigrant background (mother and/or father immigrants). We also include a control for education level that reflects the highest grade attained by the women. We use a dummy variable because despite generally low education levels among the women, there is a clear split between those with an elementary school education or less and those who attended middle school or high school (6th grade or more = 1; 5th grade or less = 0). None have any college education and only 2 % graduated high school, with 8 % even starting high school. Forty-seven percent of the sample left school sometime between 6th and 9th grade.

Missing Data

Overall, the data fields are well populated. However, some of the measures do have enough missing data to warrant data imputation in the interest of preserving cases. We impute values for any variables with over 2 % missing (data missing for 10 or more women) using multiple imputation by chained regression (MICE in Stata 13.0). This technique uses switching regression, an iterative multivariable regression technique for imputing missing data. For analyses with the full sample and for relevant subsample analyses, we imputed 20 datasets and account for the uncertainty this introduces by adjusting the model standard errors for all analyses (Rubin, 1976). For the full sample, we impute data for the whether women have kids post-parole (6 % missing), whether they were working at commitment (2 % missing), and whether they were delinquent prior to age 14 (4 % missing), education (18 % missing). For subsample analyses, we ran separate subsample imputations. In addition to the variables we impute for the full sample, we also impute values for marital quality among those married in the post-parole period (7 % missing; N = 5) and work commitment among those who were working post-parole (23 % missing; N = 57).

Plan of Analysis

Our analytic strategy proceeds in four steps. First, using the full sample, we estimate logistic regression models that examine the influence of each external change mechanism separately on desistance. We then assess how they compare by estimating a full model to see which effects hold, controlling for the others. Our second set of analyses looks more closely at marriage by assessing the influence of marital quality, among the other potential external change mechanisms, on desistance for the subsample of women who are married and living with their husbands in the post-parole period (N = 196). Third, we consider the importance of job quality for desistance among those working in the post-parole period (N = 243). Finally, in an effort to better assess which combination of external change mechanisms best facilitate desistance, we perform a conjunctive analysis that illustrates and compares how these mechanisms overlap for desisters compared to persisters.

Results

The summary statistics reported in Table 1 suggest some important differences between desisters and persisters. To assess the significance of these differences for desistance, we begin by modeling the influence of specific external change mechanisms first separately and then in a full model that considers the independent influence of each mechanism on the odds of desistance controlling for the others. Starting with marriage (Table 2: model 1), we find that in itself, marriage is immaterial to the odds of desistance. Rather, what matters in the desistance process is the type of man that these women marry. Specifically, women who desist are more likely to be married to a prosocial husband who has a steady job and no arrest history. This is telling, particularly when we consider that, though 34 % of persisters are married post-parole, only 5 % have a prosocial spouse (Table 1). In contrast, 55 % of desisters are married in the post-parole period, and 25 % have a prosocial spouse. Still, this suggests that, for the majority of desisters who do not have a prosocial spouse (75 %), something aside from marriage is driving the shift in their offending behavior post-release.

Table 2 Logistic regression models predicting desistance (N = 418)
Full size table

Model 2 (Table 2) considers the influence of motherhood on desistance. Table 1 indicates that having children (whether they were born before or after release) is more common among desisters than persisters. Regression analyses reveal that, regardless of whether these women had children before the current incarceration or upon release, motherhood appears to act as an important external change mechanism. Specifically, 71 % of desisters were mothers. Among them, 56 % had a child before the current incarceration and another 34 % had a child after release from supervision (see Table 1), either their first child (15 %) or an additional child (19 %). And, separate regression models (not shown) indicate that simply being a mother (irrespective of timing relative to incarceration) is also a significant predictor of desistance. Still, a substantial number of persisters (53 %) were also mothers at commitment (45 %) and/or following release (13 %). So, the motherhood effect, though significant, does not universally activate desistance processes.

In addition to a quality marriage and motherhood, we also find that working in the post-parole period increases the odds of desistance (model 3; Table 2). This is despite the fact that similar percentages of desisters and persisters have jobs post-parole (58 and 56 %, respectively; Table 1). Of note, though, is the influence of the homemaker role on desistance. Those women who are engaged as homemakers and not working in the paid labor force are more likely to deist (model 3; Table 2), and this role is notably more common among desisters compared to persisters (32 and 9 %, respectively; Table 1). So, while working can facilitate desistance for these women, a less common but more discerning influence is unpaid labor in the domestic sphere in lieu of paid work in the public sphere. Notably, this option is substantially more common among women with a prosocial spouse, 46 % of whom are homemakers, compared to 19 % for the rest of the women in the sample. These influences, though, appear to operate independently (no significant interaction effect—model not shown).

In the final analysis with the full sample, we examine the influence of each of these domains (marriage, motherhood and work) controlling for the others. Results change little. All three domains exhibit some influence on the desistance process, controlling for the others. The only substantive change is that becoming a mother post-parole is no longer significant once all three domains are modeled jointly. Though, having had a child prior to commitment remains significant, suggesting some role for motherhood in facilitating the behavioral changes that help women avoid re-offending once paroled.

Our next set of analyses further evaluates why and how marriage and work might motivate desistance among the subsamples of women who access these potential change mechanisms in the post-parole period. As we noted above, though these statuses do increase the odds of desistance, not all who access them are able to desist. Table 1 shows notable differences between desisters and persisters where marital and job quality scores are concerned and subsample analyses with those who access these institutions in the post-parole period (Tables 3 and 4) confirm the positive influence of marital and job quality on desistance. Focusing on the population who is married and living with their husband in the post-parole period, marital quality increases the odds of desistance (model 1; Table 3) and this effect holds controlling for motherhood status (model 2; Table 3) and remains strong even with controls for work status (model 3; Table 3). Also notable, this effect trumps the effect of having a prosocial husband. And, though there is overlap between marital quality and marriage to a prosocial husband, it is far from complete (b = 0.28). In other words, even where women marry men with unsteady work histories or a criminal past, a quality relationship can augment their pathway to desistance. Also of note is that, for women who are married post-parole, work has no influence on their odds of desistance. Rather, a quality marriage and having had a child before commitment are the most important drivers of desistance for this group. The positive effect of quality ties to adult institutions is similarly evident when we focus on the population of women who access paid work in the post-parole period (Table 4). Here, we find that commitment to work in the form of exhibiting good work habits and maintaining a steady job is positively associated with desistance (model 1; Table 4). These effects hold with controls for other key external change mechanisms (model 2; Table 4).

Table 3 Logistic regression models predicting desistance for women married post-parole (N = 194)
Full size table
Table 4 Logistic regression models predicting desistance for women working post-parole (N = 243)
Full size table

The descriptive statistics and the regression models suggest that marriage, motherhood, and work all influence the likelihood of desistance for these women, but do not tease out whether and how these mechanisms might overlap to best facilitate desistance. To offer some insight into this question, we conducted a conjunctive analysis of case configuration (CACC: [53]) focusing on the key external change mechanisms implicated in the regression models. Building on qualitative approaches such as qualitative comparative analysis (QCA), this method is largely descriptive in its application [53]. It allows researchers to explore patterns in the data that characterize profiles evident across cases (in this instance across the women in the sample) by aggregating individual characteristics into groups of unique case configurations. With the resultant aggregations, researchers can identify dominant case configurations that characterize specific outcomes. Here, we present the dominant case configurations that characterize desisters and persisters as well as those that do not reliably discern across the two groups (shared profiles). The cutoff for identifying dominant profiles is somewhat subjective, though Hart [53] suggests a cutoff of 5 cases per profile for samples of under 1000. We use a slightly more rigorous cutoff of 8 cases, since using this cutoff yielded 16 case configurations or profiles (out of a possible 64, based on 6 variables with 2 response categories each: 26) that accounted for 74 % of desisters and 70 % of persisters. These 16 profiles are presented in Table 5.

Table 5 Prominent profiles emergent from conjunctive analysis comparing patterns of access to external change mechanisms for desisters and persisters
Full size table

A number of interesting findings are evident from an inspection of the dominant case configurations that emerge from the CACC. First, it is clear that desistance is most likely when a number of external change mechanisms converge. When examining the dominant profiles for the desisters, all but one have accessed three or more external change mechanisms. This group of case configurations characterizes 34 % of desisters, but is only evident among 8 % of persisters. Conversely, the dominant profiles for persisters are ones in which access to external change mechanisms is virtually absent. This characterizes 21 % of persisters and only 6 % of desisters. The least discerning profiles are those that fall somewhere in the middle—with anywhere from one to three external change mechanisms available (34 % of desisters and 41 % of persisters).

Second, the distribution of dominant profiles highlights certain clusters of external change mechanisms as particularly important for desistance. The overlap between marriage post-parole and a prosocial spouse is prominent in the profiles of desisters. At the same time, prosocial spouses are entirely absent from the dominant case profiles for persisters and from the shared profiles. Also of note is the fact that having children in the post-parole period and being a homemaker figure prominently in the dominant case configurations of desisters but are both entirely absent from the profiles that most commonly characterize the other two groups. The influence of work post-parole also features prominently among desisters—indeed, they are either homemakers or working in the paid labor force. On the other hand, the dominant profiles characterizing the persisters include neither of these activities, increasing the likelihood that persisters are engaged in illicit money making activities. Also prominent in desister profiles, and consistent with regression models, is having had children prior to commitment. However, these dominant profiles indicate that this is among the least discerning features of the desister profile since it is also evident in the other two profile groups.

Discussion

Our aim in using historical data to assess the key external change mechanisms that frame women’s pathways to desistance was to test the historical generalizability and scope of contemporary accounts of women’s desistance. Specifically, we wanted to see whether the external change mechanisms implicated in contemporary accounts of women’s desistance also influenced desistance among women exiting the criminal justice nearly 100 years ago. We also wanted to assess how the historically contingent gender norms that both framed their opportunities and constrained their behaviors influenced the source and salience of these desistance mechanisms. Our results reinforce the generalizability of external change mechanisms as central drivers of desistance but also indicate that socio-historical contexts frame the specific ways in which these processes unfold among women.

Farrall et al. [9] propose a model of desistance that positions the process within macro-level structural contexts and highlights the role of community and social structures for both opening and constraining the opportunities and choices that facilitate change. Individual agency and related internal change mechanisms, they argue, reflect how individuals navigate and interpret the macro- and meso-level contexts that orient and define their opportunities and choices. In effect, these contexts frame offenders’ access to external change mechanisms and narrow the agentic choices available for crafting a new non-criminal self. In this way, women’s pathways to desistance reflect not just gendered opportunities, but the gendered norms and expectations that frame these opportunities; norms that are dynamic and historically contingent.

Our findings fall neatly in line with this theoretical frame. Desistance, for the women during this historical time point, is clearly a reflection of women’s access to the opportunities that allow them to conform to mainstream notions of women’s “place” in the social structure. In the early part of the twentieth century, this meant that the women who were the least likely to return to the criminal justice system on release were those were in a committed marriage and those acting as homemakers or mothers. Further, married women with a prosocial spouse or a high quality marriage were particularly likely to desist and among those in the paid workforce, those with good work habits and a steady job fared best. These findings remind us that external change mechanisms are critical, but, for women, the salience of specific change mechanisms reflects the unique historical context in which they unfold and related gender norms. In studying desistance, we need to be cognizant, not only of these central change mechanisms, but of the broader contexts that frame how these mechanisms can manifest for particular social groups in specific temporal and spatial contexts.

Our regression results coupled with the CACC results regarding the overlapping external change mechanisms that characterize desister profiles highlight this nicely. With respect to marriage, for instance our regression models indicate that being married, in itself does not promote desistance. Instead desistance among the women in this sample is more strongly linked to marital quality and a prosocial husband. The CACC results reinforce this, with the majority of dominant desister profiles including both marriage and a prosocial spouse. In none of these profiles is marriage alone sufficient. More often than not, it is coupled with motherhood and or homemaking. In short, a marriage effect is clearly evident but it is contingent on spousal characteristics, marital quality, and its overlap with other key change mechanism.

Interestingly, Giordano et al. [6] find no marriage effect among males or females in their contemporary sample, and argue that contemporary demographic and social trends (postponement of childbirth and marriage, rising divorce rates, and increased prevalence of generally unstable cohabitation arrangements), have significantly reduced the likelihood of marriage, particularly among lower-status minority populations that are disproportionately over-represented in the criminal justice system [6]. Whereas only 27 % of the women in their sample married, 47 % of the women in this sample were married following their release. Our findings comport with Giordano et al.’s suggestion that the centrality of marriage, and more specifically marital quality, in relation to desistance may be historically contingent, such that as marriage becomes less accessible, its role in the desistance process becomes less prominent.

Despite our finding that much of the marriage effect can be attributed to martial quality and a prosocial spouse, in practice this is of limited relevance. We say this because the data indicate that few of the women in this sample actually marry prosocial men. Focusing on post-parole marriages, only 17 % of the sample married prosocial partners with stable work histories and no criminal history. Even among those who desisted, only 24 % were married to prosocial spouses post-parole. If we look only at those desisters who are married, less than half have a prosocial spouse. It seems unlikely that husbands who are themselves criminal and/or have difficulty holding a job could provide the kinds of social control the wives that Laub and Sampson [7] talk of provide for the men they studied.

What then explains the marriage effect? Sampson et al. [23] suggest that aside from social bonds and social control marriage also provides a change in leisure routines that limit criminal opportunities and a change in identity towards a more conventional self. This explanation may be particularly important for the women in our sample. Among these women, marriage itself may represent an endorsement of conventional social values and a shift towards a more conventional lifestyle. In this context, marriage was the only legitimate venue within which women’s sexual behavior was socially sanctioned. Moreover, as the CACC results evidence, for desisters, marriage commonly pairs with motherhood and homemaking. In other words, it is not just marriage, but the related conventional family lifestyle that often accompanies it that likely explains its link to desistance for these women.

Related to marriage and commitments to conventional gender norms, our findings also suggest that work, while implicated in the desistance process, operates in gendered and historically contingent ways. To begin, among the key predictors of desistance in regression models is assuming the role of homemaker in lieu of paid employment. Indeed, the majority of dominant desister profiles in the CACC evidence this as well. This is a clearly gendered and historically contingent finding. Exiting the labor market is not a likely path to desistance for males or contemporary samples of disadvantaged female offenders. Even among those contemporary female offenders who marry, work likely remains a necessity, albeit an often-difficult one to come by. Research by Holtfretter et al. [51] suggests that, among contemporary samples of female offenders, recidivism is strongly linked to poverty, and state-sponsored programs that aim to help women combat poverty, especially via education and vocational training, reduce recidivism. Moreover, this again reinforces the link between women’s compliance with social expectation and their desistance. Work was a necessity for many poor women in the early 1900s, but it was not, broadly speaking, normative and many social reformers linked women’s work to a host of social problems.

However, not all of the women in our sample exit the public sphere and assume a homemaker role, and, for these women, desistance is still possible, as both the regression and CACC analyses suggest. Here, our findings again mirror Sampson and Laub’s and, more recently Opsal’s [31], indicating that, among women who work, it is not having a job that matters, rather it is job quality. Interestingly though, while marriage to a prosocial partner seems to facilitate desistance regardless of work status, paid labor is immaterial to desistance for the subsample of women who are married post-parole. So, while both are important, in this era, exiting the labor force was particularly salient to women’s pathways out of crime. While contemporary studies still do not position work as central to women’s desistance, the moral proscriptions against female employment have shifted substantially.

Along with marriage and homemaking, our findings indicate motherhood is associated with desistance, though not as strongly as is the case for homemaking or for quality marriage and employment. The desisters in this sample are more likely to have children—both before and after commitment. However, once we take into account other external change mechanisms, only having been a mother at commitment retains significance. This is also the case when we focus on married women and control for marital quality. Moreover, among the employed subsample, once we control for work quality, the influence of children loses its significance entirely. The relatively weaker influence of motherhood in comparison to marriage, homemaking, and employment may have to do with its more tenuous normative status. Motherhood, in the early 1900s (and still today, though the boundary is perhaps more flexible) is normative in the context of marriage, while single motherhood signals a blatant violation of normative sexual standards condemning sex outside of marriage. Indeed, of the nine dominant desister profiles from the CACC that include motherhood before commitment and/or after release, only two do not also include marriage. Also of note is that a significant portion of the sample entered the system as mothers (52 %), reinforcing the fact that parenthood itself does not keep women from offending. At the same time, having children may be a “hook for change” such that those who have children are more motivated to stay out of the system in the future and having children once released, in some instances, also helps to keep women from re-offending. This is perhaps our least historically contingent finding. Contemporary work indicates that up to 80 % of women prisoners are mothers [54] and that parenthood is implicated in their eventual desistance [42, 43]. However, this finding is clearly gendered, with parenthood not similarly implicated in men’s desistance [6, 8].

In sum, we would argue that external change mechanisms are important to women’s desistance. However, we would point out that embracing conventional adult roles, particularly the role of wife, homemaker, and mother, is associated with desistance not just because these roles tie women to social networks and institutions that monitor their behavior, but because they are a reflection of a broader commitment to conventional social norms regarding women’s place in the social world. We suspect this to be the case among contemporary samples as well. As early as 1950, Pollak argued that female offenders are deemed particularly immoral—guilty of “double deviance”—because their offending behavior violates both legal/moral codes but also normative expectations regarding women’s behavior. Though difficult to test explicitly, gender norms that portray women as less aggressive, more empathetic, and more law abiding remain and it is still the case that “we tend to think of the female offender as an anomaly” ([55]: 1). As such, one marker of female desistance is the extent to which behavioral shifts are consistent with behavioral expectations for women. In the early twentieth century, this meant embracing marriage, motherhood, and the homemaker role and, when work was a necessity, being a reliable worker in a steady job. Giordano et al.’s [6] finding that social capital is not a common route to desistance for disadvantaged women in a more contemporary setting is not necessarily inconsistent with our findings. For their sample, external change mechanisms proved relatively difficult to access. Signaling their commitment to normative behavior protocols required that these women take advantage of whatever personal hooks for change they could access to reframe their cognitive identities and redirect their behavioral trajectories to fit the normative expectations associated with their identities as both women and non-offenders. Indeed, though external change mechanisms held limited statistical significance in their models, the qualitative data indicate that “many of the women who were more successful as desisters crafted highly traditional replacement selves (e.g., child of God, the good wife, involved mother) that they associated with their successful exits from criminal activities” ([6]: 1053).

Limitations

Our findings are not without caveat. Our interest is in desistance, but our measure of desistance is restricted to official data covering a 5-year post-parole follow-up period. Moreover, it is a simple bivariate measure of any arrests during this period. In practice, this means that some of the women we consider desisters may have engaged in antisocial behavior after release but avoided criminal justice system contact. Others may have exhibited temporary desistance, and were re-arrested after the 5-year follow-up. Nonetheless, our results speak to an important transitional period—re-entry—and document the mechanisms that help women disengage from the criminal justice system on release. Other measures are similarly imperfect, but most operate as expected, suggesting they are reasonable proxies for the key mechanisms we assess. Further, these data do not offer any reliable measures of the internal change mechanisms that feature prominently in contemporary accounts of desistance, especially for women. We suspect that many of the external change mechanism we document as important are influential precisely because they activate internal change mechanisms so important for behavioral change. So, while not ideal, we believe the data offer a unique opportunity to assess how external change mechanisms influence women’s desistance and to situate these influences in a particular historical context.

Conclusions

On balance, our results comport with mounting evidence that external change mechanisms help drive desistance. Most importantly, our historical vantage point introduces important contingencies relevant to theories of female desistance. Namely, our findings suggest that these change mechanisms not only activate informal social controls, they also implicate compliance with gendered social norms. As these norms change, the salience of key external change mechanisms for women’s desistance should also change. Of course, this is also the case for males, and we would encourage desistance researchers to consider the importance of gender as a social institution and the ways in which it conditions the desistance mechanisms available to offenders.