Introduction

Immigrant families are economically vulnerable for a variety of reasons, including lower educational attainment, poor host-country language skills, and lower returns to foreign education and work experience (Sullivan and Ziegert 2008). One factor that remains underexplored, however, is the effect of female spouses’ employment on immigrant economic mobility. Although qualitative studies highlight the significant roles of employed immigrant women in the early settlement stages, the quantitative assessment of whether their employment helps lift their families out of poverty is limited. The earning contributions of working female spouses to family incomes may seem obvious, but we know little about whether their earnings are high enough to lift their families out of poverty. Moreover, while previous research has examined ethnic variations in female immigrant employment, research bridging this topic and immigrant poverty is limited (Baker 2004; Creese et al. 2008; Grahame 2003). Nor do we know whether there are differences linked to immigrant ethnicity. In this study, I address these gaps by asking two questions. First, for recent immigrant families experiencing poverty, does the employment of female spouses help them exit poverty? Second, to what extent does the ethnic variation in female immigrant employment explain the ethnic gaps in escaping poverty? I answer these questions using data from the Longitudinal Survey of Immigrants to Canada (LSIC), a nationally representative survey of new arrivals.

While this study builds on past studies on immigration, it also improves upon them in three ways. First, it links two research areas: immigrant poverty, and gender and migration. Second, it looks at the post-migration factor of women’s employment in the host country. While previous research has analyzed immigrant poverty dynamics, including exits from and reentries into poverty, it has considered only time-constant pre-migration characteristics, including region of origin, age at immigration, and foreign education (Fleury 2007; Picot et al. 2008). Third, it uses two analytical innovations: the bivariate probit model and Fairlie decomposition technique. The former allows me to consider the possibility of low-income immigrant women’s selection into employment and its bias associated with their unobserved characteristics. The latter helps specify the extent to which ethnic poverty exit gaps can be explained by ethnic variations in female immigrant employment.

Immigrant Poverty and Ethnic Variations

In many Western immigrant destinations (e.g., Denmark, Spain, Sweden, the United States), immigrants have higher poverty rates than the native-born (Blume et al. 2007; Jensen 1989; Muñoz de Bustillo and Antón 2011; Sullivan and Ziegert 2008). This is certainly true in Canada: in 2000, the poverty rate of immigrants aged 25–54 was 18 %—7 percentage points higher than that of the Canadian-born.Footnote 1 Poverty levels also vary widely by ethnicity: non-European groups have higher levels than their European counterparts (Lee 1994). While Arabs and West Asians have relatively high poverty rates (35 %), East/Southeast and South Asians have lower levels (22 % and 19 %, respectively).Footnote 2 Yet, the poverty levels of these non-European groups are higher than those of Europeans (11 %). Finally, recent immigrants are particularly vulnerable to poverty: during 2002–2003, 43 % of immigrants in Canada for two years were living in poverty.

These poverty patterns among immigrants in Canada are puzzling given that the majority of recent cohorts are highly educated, reflective of the country’s merit-based immigration policy (Picot and Hou 2003). It is imperative to understand what helps lift recent arrivals, especially non-Europeans, out of poverty quickly given that persistent poverty may limit the life chances of their offspring (Corcoran 2002). Although the discussion of solutions to immigrant poverty is limited, extensive research exists on factors contributing to the economic mobility of the poor in general, including employment, education/training, and government support (Cellini et al. 2008; Worts et al. 2010). Arguably, these findings should be applicable to immigrants.

Mechanisms Underlying the Relationship Between Immigrant Poverty and the Employment of Female Spouses

Does the employment of recently arrived immigrant women help lift their families out of poverty? If women from specific ethnic groups are less likely to work than others, how does this influence poverty exit? Although little research has considered these specific questions, the broader issue of whether and to what extent the employment of female spouses contributes to family income is widely addressed. Of the many theoretical approaches, the equal earner/female breadwinner and male breadwinner models are useful for the present discussion. I use these to derive two competing hypotheses about the contribution of immigrant women to the exit from poverty.

The Equal Earner/Female Breadwinner Model

The equal earner/female breadwinner model posits that female spouses earn as much as or more than their male counterparts. Such earning patterns are no longer uncommon in postindustrial Western countries (Drago et al. 2005; Harkness et al. 1997). In the United States, equal earner and female breadwinner couples (in which female spouses earn 40 % to 60 % and more than 60 % of the total income, respectively) made up 31 % of all couples in 2001, a 20 % increase from 1970 (Raley et al. 2006). A similar trend is observed in Canada: only 12 % of women in dual-earner couples outearned their male spouses in 1976, but this percentage rose to 29 % by 2008 (Williams 2010).

These statistics do not distinguish by the ethnoracial origin and nativity of families, but some evidence suggests that the earning contributions of female spouses are greater among ethnoracial minority and immigrant families. For example, in the United States, black wives make a greater contribution to the total family income than their white counterparts (45 % vs. 36 %), even though they are employed at similar rates (75 %) (Choi 1999). Meanwhile, U.S. Hispanic wives are far less likely to work (60 %) than blacks and non-Hispanic whites, but their earning contribution is similar to that of non-Hispanic whites (37 %) (Choi 1999). Qualitative works also indicate immigrant women’s noticeable earning contributions to their family income in the United States (Espiritu 1999). Despite their marginal positions in the host labor market, refugee women from Vietnam are found to make greater earning contributions to their family income in the United States than in their origin country because their male spouses experience drastic downward mobility after immigration (Kibria 1990, 1994). Men who held middle-class status in Vietnam are unemployed or working in low-paid, low-skilled jobs in the United States. These findings can be explained by the possibility that the male spouses of ethnoracial minority and immigrant families encounter racism, language barriers, and nonrecognition/underrecognition of overseas qualifications, making it more difficult for men to be the main or the sole breadwinner.

By contrast, the employability of immigrant and minority women has been rising in the female-intensive industries (e.g., garment and microelectronics industries) in the post-1970s economic restructuring in North America (Espiritu 1999). Another mechanism of the equal earner/female breadwinner model unique to immigrant families may be related to their settlement strategies to survive as families by combining skill upgrading and employment (Boyd 1989). Even if women’s paid work is less common in their origin countries, female immigrants may provide sizable financial support to their families while their husbands return to school for better job opportunities (Creese et al. 2008).

Based on the equal earner/female breadwinner model, I derive the following hypothesis about the returns to female immigrant employment on poverty exit:

  • Hypothesis 1 (H1): The employment of recently arrived female immigrant spouses has a significantly positive impact on family poverty exit even after the male spouse’s employment characteristics are adjusted.

Because the annual employment income of male immigrant spouses is unavailable in the data, I control for other employment characteristics—weekly earnings and weeks worked—as the closest proxies.

The Male Breadwinner Model

Gender research has characterized the division of work in industrial societies as following a male breadwinner model, whereby men seek paid work in the public sphere, and women are expected to engage in social reproduction in the private sphere. This model proliferated in Western societies in the late nineteenth and early twentieth centuries as a result of the development of the family wage ideology that male workers should earn family wages “sufficient for a man to support a wife and children” (Vosko and Zukewich 2006:73). The male breadwinner model suggests that female spouses earn “pin money,” the earnings that cover only incidental expenses for their family incomes if they work (DeRiviere 2008:234). The male breadwinner model also suggests that wives are less likely to work, leaving the majority of families heavily dependent on husbands’ earnings.

Scholars have noted a departure from the male breadwinner model in developed countries since the 1970s for reasons such as the rising female educational attainment, the increasing entry of highly educated women into the labor force, and the growing job insecurity of male workers (Blossfeld and Buchholz 2009; Charles and James 2005; Wilkie 1991). However, there is reason to believe that this model still holds strongly for immigrant families, given their gendered migration and host-country labor market experiences (Creese and Wiebe 2012; Pedraza 1991). For example, many immigrants come from countries where patriarchal gender relations persist. As part of this gendered understanding of their role, female immigrant spouses may have greater responsibilities for household and care work. They also may be responsible for maintaining and rebuilding family and community networks in their origin and host countries (Foner 1999; Purkayastha 2005). Such family responsibilities may lead to female immigrant spouses’ weaker labor market attachment and their smaller earning contributions when they work. In addition, the labor market of the host country is often gendered, with immigrant women channeled into lower-paid manual jobs (e.g., cleaning, light manufacturing) (Creese and Wiebe 2012).

If the male breadwinner model applies to recently arrived low-income immigrants, their female spouses will make no significant contribution to lifting their families out of poverty even if they work:

  • Hypothesis 2 (H2): The employment of recently arrived female immigrant spouses has no significant impact on the exit from family poverty if the male spouse’s employment characteristics are taken into account.

There may also be group differences in female employment rates. I test this possibility by using the Fairlie decomposition technique.

Does the Male Breadwinner Model Apply to Some Ethnic Groups More Than Others?

Just as immigrant poverty levels vary by ethnicity, so too there are ethnic variations in female immigrant employment. In the second step of this study, I examine the impacts of such variations on the ethnic gaps in immigrant poverty exit and consider whether the male breadwinner model applies to some groups more than others.

In many immigrant destinations (e.g., Australia, Israel, the Netherlands, Sweden, the United Kingdom, the United States), immigrant women, especially recent arrivals, are less likely to be employed than native-born women (Bevelander 2005; Bevelander and Groeneveld 2006; Foroutan 2008; Rendall et al. 2010) and immigrant men (Blau et al. 2011; Rebhun 2008).Footnote 3 There are also ethnic variations in their employment. In the United States, immigrant women of Chinese, Filipino, and Cuban origins have comparatively high employment rates; by contrast, immigrant women of other ethnic origins—including Arabs, Asian Indians, Japanese, Mexicans, and Puerto Ricans—are less likely to work (Duleep and Sanders 1993; Read 2004; Read and Cohen 2007; Stier and Tienda 1992). Equivalent quantitative research is sparse in Canada, with the exception of an analysis of the 2001 census by Preston and Giles (2004), which found notably higher labor force participation rates among Filipino women (more than 85 %) than among Arab, West Asian, Korean, and Japanese immigrant women (53 % to 65 %). Similar ethnic variations can be found among recently arrived women experiencing family poverty. While 70 % of low-income European immigrant women in Canada are employed four years after arrival, the employment rates of Arab and West Asian women are noticeably lower, at 21 % and 48 %, respectively. The lower female employment rates of these groups corroborate Read and Cohen’s study (2007) on immigrant women in the United States.Footnote 4

Why are immigrant women of some ethnic groups less likely to be employed? Research in the United States variously points to human capital, family condition, and cultural explanations (Duleep and Sanders 1993; Greenlees and Saenz 1999; Read and Cohen 2007; Stier 1991). The human capital explanation posits that women of higher education and host-country language proficiency and greater work experience are more likely to be employed. The ethnic groups known for higher levels of human capital (e.g., Asians) are thus expected to have higher female employment rates than those known for lower human capital (e.g., Latinos). The family condition explanation states that family characteristics and economic circumstances can influence women’s decision to work. For instance, women’s family responsibilities (e.g., taking care of small children or elderly relatives) may reduce their likelihood of employment, whereas budgetary constraints of the household may drive women to work, especially when they have extended family members who can look after their children. The cultural explanation argues that immigrant women are less likely to work if they come from cultures emphasizing domestic responsibilities, but exposure to more egalitarian cultures in the host country leads them to engage in paid work (Read and Cohen 2007).

These explanations developed in the United States may apply to the employment of immigrant women in Canada, given institutional similarities between the two countries, such as relatively unregulated labor markets (Doellgast et al. 2009). However, their immigrant populations may differ as a result of divergent immigration policies: Canada’s skill-based admission policy has led to an increased pool of highly skilled immigrants compared with the United States. Nevertheless, qualitative research suggests relatively low employment rates among highly skilled immigrant women in Canada (Man 1995; Salaff and Greve 2003).

If some ethnic groups have notably lower female employment rates, contributing to their low poverty exit rates, the male breadwinner model is more likely to hold for these groups. I use the Fairlie decomposition technique to assess the extent to which the differences in female immigrant employment rates between two ethnic groups contribute to their differences in exiting poverty. Given their relatively advantageous position in poverty exit and female employment, I use Europeans (a combined category of British, French, Western, Northern, Eastern, and Southern Europeans) as the reference group and compare them with Arabs/West Asians, South Asians, East/Southeast Asians, and others. Admittedly, aggregation of ethnic subgroups in this fashion may mask important within-group variations (Dale and Ahmed 2011; Evans and Lukic 1998). However, disaggregating each group would lead to substantially small unweighted cases and not produce reliable estimates.Footnote 5 I therefore analyze the aggregated groups based on the survey question, “To which ethnic or cultural groups do you belong?” (Statistics Canada 2001:9).

Data and Methods

Data

This study uses data from the Longitudinal Survey of Immigrants to Canada (LSIC). The LSIC targets immigrants aged 15 and older who landed in Canada as permanent residents between October 2000 and September 2001 (Grondin 2007). Face-to-face or telephone interviews were conducted approximately six months, two years, and four years after the respondents’ arrival. A total of 12,040 (Wave 1), 9,322 (Wave 2), and 7,700 (Wave 3) individuals participated in the interviews, producing response rates of 77 % and 64 % for Waves 2 and 3, respectively.

My sample includes female immigrants aged 25 to 54 in Wave 1 who were living with their male spouses (married or in common-law unions) throughout the three waves and whose total annual family income in Wave 2 was below Statistics Canada’s low-income cutoff (LICO). The analysis eliminates respondents whose spouse was born in Canada or immigrated more than six months before the respondents’ arrival because their immigration experiences may differ (Min and Kim 2009).

The results require careful interpretation given the period of the survey. During the early 2000s, the information technology (IT) sector faced a downturn, yet a growing number of entering immigrants had expertise in IT and engineering (Picot and Hou 2009). The immigrants interviewed in the LSIC are, therefore, considered to have experienced challenges entering the Canadian labor market.

This study uses the LICO as the poverty threshold to ensure that the research findings can be compared with other well-known studies on immigrant poverty in Canada (Kazemipur and Halli 2001; Picot and Hou 2003). The LICO is set at 20 % above the average percentage of family income spent on essentials (e.g., food, shelter, clothing); the latter is set at 44 % based on results from the 1992 Family Expenditures Survey. Therefore, if a family spends more than 64 % (=44 % + 20 %) of its income on essentials, all family members are considered to be “in strained circumstances” (Paquet 2002:11). This 64 % threshold is converted into 35 different cutoffs according to the family and community sizes.

The possibility of attrition bias is a concern in this study, given that the attrition rates are relatively high (22 % from Wave 1 to 2; 17 % from Wave 2 to 3). Besides, low-income immigrants may be more likely to drop out of subsequent surveys, as they may return to their home countries or move to other countries for better economic opportunities, thus biasing the results. However, my logistic regression analysis (results not shown here) suggests that the Wave 2 poor are not statistically different (p > .05) from their nonpoor counterparts in their probability of dropping out of the Wave 3 interview when other demographic and socioeconomic characteristics are controlled. This ensures that bias associated with attrition is minimal.

I use the longitudinal survey weight created by Statistics Canada, which is widely used in previous research (Adamuti-Trache et al. 2013; De Maio and Kemp 2010; Frank 2013; Fuller-Thomson et al. 2011; Roth et al. 2012). This weight makes the sample representative of the immigrant population remaining in Canada four years after arrival.Footnote 6

As Fig. 1 shows, 47 % of female spouses aged 25–54 (the sample) were in poverty in Wave 2. Two years later in Wave 3, 48 % of these women were no longer in poverty; they had exited poverty according to the study’s definition.

Fig. 1
figure 1

Transition into/out of low income between Waves 2 and 3: Female immigrant spouses aged 25–54, Canada, 2002–2005

Full size image

Measures

Poverty Exit

The dependent variable is an indicator of whether immigrants experiencing family poverty two years after arrival (Wave 2) exit the state of poverty two years later (Wave 3). This variable is coded as 1 if the respondent’s annual family income in Wave 3 is higher than the 2004–2005 LICO, and 0 otherwise.Footnote 7

Immigrant Women’s Employment

The independent variable indicates a female immigrant spouse’s employment status in Wave 3. The variable is coded as 1 if she is employed in Wave 3, and 0 otherwise.

Other Factors

The analysis controls for other factors expected to influence female immigrant spouses’ exit from family poverty in Wave 3: their own characteristics (age, ethnic origin, initial education, host-country language skills, and weekly earnings in Wave 2) and family characteristics (the number of children under age 18, city of residence, male spouses’ weekly earnings in Wave 2, changes in their weekly earnings (in 2001 dollars), and weeks worked from Wave 2 to 3) (Picot et al. 1999). As detailed in the next section, many of these control variables are also used to estimate the probability of employment for female spouses in Wave 3 in the bivariate probit models.

I control for age because younger immigrants may have better life chances, acquiring host-country language skills more quickly and finding more opportunities to interact with the native born (Martinovic et al. 2009). I therefore expect that immigrants who are younger in Wave 2 are more likely to exit family poverty.

The education variable is represented by the possession of postsecondary educational credentials before arrival in Canada.Footnote 8 The respondent’s self-reported proficiency in French (in Quebec) or English (in the rest of Canada) constitutes the measure of host-country language skills. Taking human capital theory into account, I predict that immigrants with higher educational attainment and English/French skills are more likely to exit poverty. Female immigrants’ ethnic origins may influence their chances of exit from poverty because the persisting ethnic stratification in the host-country institutions yields varied access to resources (Lian and Matthews 1998). I expect Europeans to have higher probabilities of exiting poverty than other ethnic groups.

As poverty is a family-level economic disadvantage, the respondent’s family characteristics may influence the probability of exiting poverty. I expect that immigrant women are more likely to exit poverty if their spouses are employed in Wave 3. I also expect that having more children in the household decreases the probability of poverty exit because of the greater financial costs and time associated with child care.Footnote 9

Finally, I control for the respondent’s place of residence in Wave 2, given that geographic variations in industries and economic circumstances may affect the availability of higher-wage jobs (Hiebert 1999). In fact, studies indicate diversity in economic opportunities and barriers in the three major immigrant destinations: Toronto (the reference group), Montreal, and Vancouver (Badets and Howatson-Leo 1999; Preston and Cox 1999).

Table 1 displays descriptive characteristics of the sample of female immigrant spouses experiencing poverty in Wave 2 (column 1). The majority have postsecondary credentials (73 %) before arrival. Despite their highly educated backgrounds, their host-country language skills are limited. Only 29 % speak English or French fluently or very well in Wave 2. Their employment rates in Waves 2 and 3 are about 55 %, 15–20 percentage points lower than their male spouses. However, the males’ employment rates in Wave 2 (71 %) are far lower than the average employment rates of new immigrant male spouses (84 %).Footnote 10 Possibly, the former have returned to school to obtain Canadian qualifications.Footnote 11

Table 1 Characteristics of immigrant women aged 25–54 who were living with a male spouse in Waves 1–3 and had low income in Wave 2, by ethnic origin, Canada, 2000–2005
Full size table

Analytical Techniques

The Bivariate Probit Model

To determine whether and to what extent immigrant women help their families exit poverty if they work, I use the bivariate probit model. This model is well suited to the present study: it handles selection bias associated with unobserved heterogeneity between employed and non-employed immigrant women. In non-experimental data like the LSIC, employment is not randomly assigned to the respondents. Unobserved characteristics, such as views on gender divisions of labor, household chores, and career, may strongly influence the employment of immigrant women, given that these characteristics are often influenced by the gender socialization and cultural norms of their origin countries (Blau et al. 2011; Duleep and Sanders 1993; Read and Cohen 2007; Stier 1991; Treas 1987).

The bivariate probit model consists of two equations. The first predicts the probability of receiving treatment (i.e., female immigrants’ employment); the second predicts the probability of its binary outcome (i.e., whether they exit poverty) as a function of the treatment and other observable characteristics. These two equations are presented in mathematical notation as follows:

$$ \begin{array}{l}{Y}_1=\updelta +\upgamma {Y}_2+{\upbeta}_1{X}_1+{\upvarepsilon}_1\hfill \\ {}{Y}_2=\upalpha +{\upbeta}_2{X}_2+{\upvarepsilon}_2,\hfill \end{array} $$

where Y 1 is the exit from poverty in Wave 3 for a low-income female immigrant spouse; Y 2 is her employment status in Wave 3; X 1 and X 2 are the control variables of Y 1 and Y 2, respectively; β1, β2, and γ are the coefficients of X 1, X 2, and Y 2; δ and α are intercepts; and ε1 and ε2 are error terms. As shown in Fig. 2, there is a substantial overlap in the variables included in X 1 (see the Other Factors subsection) and X 2. In other words, many of the immigrant women’s individual and family factors expected to influence the exit from poverty (Y 1) also affect employment status (Y 2). The overlapping variables (age, initial education, English/French skills, ethnic origin, number of children, and city of residence) are assumed to directly and indirectly (via female employment) influence poverty exit in Wave 3. The covariates (X 2) for immigrant women’s employment also include post-migration education, their employment status and that of their male spouse in Wave 2, and non-earned family income in Wave 2; these are expected to affect female employment but not directly affect poverty exit. The bivariate probit model assumes that the error terms of these two equations are normally distributed: that is, E(ε1) = E(ε2) = 0; and Var(ε1) = Var(ε2) = 1.

Fig. 2
figure 2

Relationships between direct and indirect marginal effects

Full size image

Using the bivariate probit model has two advantages. First, the correlation coefficient (ρ) between the error terms (ε1 and ε2) indicates the presence/absence of selection into treatment associated with unobserved heterogeneity. A ρ significantly different from 0 indicates the presence of such selection (Chiswick et al. 2004). Second, the model allows the elimination of unobserved heterogeneity by estimating two equations simultaneously using full-information maximum likelihood estimation (FIML) (Datar and Nicosia 2012; Greene 2008; Kimball 2006). Therefore, even if selection bias associated with unobserved heterogeneity is detected, the coefficient for the treatment effect is not influenced.Footnote 12

In the analysis, I assess whether immigrant women’s employment in Wave 3 (Y 2) has a positive effect on the exit from family poverty (Y 1) to test the hypotheses about the returns to female employment on family poverty exit (H1 and H2). If the positive effect of female employment remains when the employment characteristics of the male spouses and other factors are adjusted, the equal earner/female breadwinner model (H1) can be supported. To put the results into perspective, I also calculate the marginal effect of women’s employment in relation to their male spouse’s employment characteristics.

Decomposition Technique

To assess the extent to which the differences in female employment rates between the European group and one of the four non-European groups (Arabs/West Asians, South Asians, East/Southeast Asians, and others) contribute to the gap between the two groups in their exit from poverty, I use the Fairlie decomposition technique, a modified Blinder-Oaxaca decomposition technique (Fairlie 1999, 2006). It requires some modification because it is designed to explain group differences in the values of continuous dependent variables in ordinary least squares (OLS) regressions. In nonlinear regressions like probit models, the average value of an outcome Y does not necessarily equal the predicted probability that Y = 1 when the probit function is evaluated at the means of the Xs (covariates) (Van Hook et al. 2004:655). Nevertheless, the average of the predicted probabilities across all cases equals the average value of Y; therefore, the decomposition for a probit regression equation \( F\left(X\widehat{\upbeta}\right)=\Phi \left(X\widehat{\upbeta}\right) \) (where Φ(.) stands for the standard normal distribution function) between two groups can be described as follows:

$$ \overline{Y_E}-\overline{Y_A}=\left[{\displaystyle {\sum}_{i=1}^{N^E}\frac{F\left({\mathbf{X}}_i^E{\widehat{\upbeta}}^{*}\right)}{N^{*}}-{\displaystyle {\sum}_{i=1}^{N^A}\frac{F\left({\mathbf{X}}_i^A{\widehat{\upbeta}}^{*}\right)}{N^{*}}}}\right]+\left[{\displaystyle {\sum}_{i=1}^{N^A}\frac{F\left({\mathbf{X}}_i^A{\widehat{\upbeta}}^{*}\right)}{N^{*}}-{\displaystyle {\sum}_{i=1}^{N^E}\frac{F\left({\mathbf{X}}_i^A{\widehat{\upbeta}}^{*}\right)}{N^{*}}}}\right], $$
(1)

where \( \overline{Y_E}\left(\overline{Y_A}\right) \) and X E(X A) represent the average probability of exit from poverty and the vector of individual characteristics for the European (Arab/West Asian) group, respectively. The first term on the right side of Eq. (1) refers to the compositional differences weighted by the coefficient for the pooled sample of Europeans and Arabs/West Asians (\( {\widehat{\upbeta}}^{*} \)); the second term represents differences in returns weighted by the means for the Arab/West Asian group (X A).

Equation (1) indicates that the differences in dichotomous outcomes, such as poverty exit, are decomposed into compositional differences and differences in returns. The compositional differences are further divided into compositional differences in specific covariates, such as female immigrant spouses’ employment and education. In a model where only two dummy variables X 1 and X 2 are included as covariates, the effects of compositional differences in X 1 can be expressed as

$$ \frac{1}{N}{\displaystyle {\sum}_{i=1}^NF\left({X}_1^E{\widehat{\upbeta}}_1^{*}+{X}_2^A{\widehat{\upbeta}}_2^{*}\right)-}F\left({X}_1^A{\widehat{\upbeta}}_1^{*}+{X}_2^A{\widehat{\upbeta}}_2^{*}\right). $$
(2)

Equation (2) suggests that the contribution of each covariate to the overall gap in the probability of exit from poverty between European and Arab/West Asian groups is equal to differences in the average predicted probability when the Arab/West Asian value is switched with the European value of the covariate of interest while holding the values of the other covariates constant. Because the two groups differ in sample sizes, I randomly draw a subsample from the larger group (i.e., Europeans) that is equal in size to the smaller group (i.e., Arabs/West Asians). A subsample of Europeans and the full sample of Arabs/West Asians are sorted by the predicted probability of exiting poverty and matched by their ranking before comparison using Eq. (2).Footnote 13 Using STATA’s fairlie command, I repeat this process 1,000 times and report the averages of the 1,000 iteration results.Footnote 14

Results

Employment of Immigrant Women and Poverty Exit

I first examine the association between the focal independent variable (the employment of female immigrant spouses in Wave 3) and the outcome (the exit from family poverty in Wave 3) without adjusting for compositional differences in other characteristics (Table 2). While less than 30 % of non-employed low-income immigrant women exit poverty in Wave 3, 63 % of employed women do so. This percentage is even slightly higher than their male spouses who are employed at that time (57 %). These results suggest that among low-income new immigrant families, employed female spouses may be making notable contributions to their family income.

Table 2 Characteristics of immigrant women aged 25–54 who were living with a male spouse in Waves 1–3 and had low income in Wave 2, Canada, 2000–2005
Full size table

Table 2 also shows wide ethnic variations in poverty exit. Although more than 60 % of immigrant women of European origins who are in poverty in Wave 2 exit poverty in Wave 3, non-European women are less likely to do so. In fact, only 22 % of Arab women and 28 % of West Asian women exit poverty, more than 35 percentage points lower than their European counterparts.

These descriptive results suggest that employed immigrant women are more likely to move out of family poverty than their non-employed counterparts. But what about the relationship between immigrant women’s employment and family poverty exit, when selection into employment is taken into account? The bivariate probit model shows that their employment has a positive impact on exiting poverty four years after their arrival in Canada; the probit for immigrant women’s employment is positive and significantly different from 0 (p < .001) even after the employment characteristics of their male spouses are taken into account (Models 1 and 2, Table 3).

Table 3 Bivariate probit estimates of female employment and exit from poverty in Wave 3, low-income female immigrants aged 25–54 living with a male spouse in Waves 1–3, Canada, 2000–2005
Full size table

The rho coefficient (ρ) of Model 1 is significantly different from 0 (p < .001), suggesting the presence of selection associated with unobserved heterogeneity among low-income female immigrant spouses. Admittedly, the data do not specify which unobserved characteristics influence their employment and exit from poverty. Arguably, employed immigrant women experiencing poverty may be more likely to engage in lower-paying employment (e.g., low-wage service-sector jobs), thereby leading to decreased likelihood of exiting poverty (Zuberi and Ptashnick 2012).

Which theoretical perspective, the equal earner/female breadwinner or the male breadwinner model, better explains the association between female immigrant spouses’ employment and poverty exit? As Model 1 shows, the positive impact of the employment of immigrant women remains significant (p < .001) after the employment characteristics of their male spouses are adjusted, suggesting that female employment has a positive impact on poverty exit, independent of the male spouse’s employment. Moreover, the positive impact of female spouses’ employment is partly explained by the economic reward they receive, given that the probit is reduced from 1.413 to 0.991 (p < .001) when women’s employment characteristics are controlled (Model 2). Therefore, the employment of female immigrant spouses may contribute to the exit from family poverty regardless of their male counterparts’ earnings. These results reject the male breadwinner model (H2) and support H1: the ability of employed female immigrant spouses to help lift their families out of poverty is equal to or greater than that of their employed male spouses.

To evaluate the effect of female spouses’ employment in relation to their male counterparts, I calculate their total marginal effects in four scenarios using the Model 1 results (see Fig. 3): (1) neither the female nor male spouse is employed in Wave 2; (2) only the female spouse is employed; (3) only the male spouse is employed; and (4) both spouses are employed.Footnote 15 The marginal effects of other covariates in these four scenarios are shown in Tables S1–S4 in Online Resource 1.

Fig. 3
figure 3

Marginal effects of female employment on the probability of exit from poverty in Wave 3, by female and male spouses’ employment in Wave 2, Canada, 2000–2005

Full size image

In the first scenario, where neither the female nor male spouse works in Wave 2, the employment of female spouses in Wave 3 raises the predicted probability of poverty exit by 43 % (21 percentage points from 48.6 % to 69.5 %).Footnote 16 When either the female or the male spouse is employed in Wave 2 (scenarios 2 and 3), the effect of female spouse’s employment in Wave 3 increases. The total marginal effect stands at around 50 % (32 percentage points, from 62.6 % to 94.2 %) in scenario 2, and 28.8 percentage points (from 58.3 % to 87.1 %) in scenario 3. Further, if both spouses are employed in Wave 2, the marginal effect of Wave 3 female employment increases to 54 % (38.3 percentage points, from 71.7 % to 110.2 %). In short, employed immigrant women help in the exit from family poverty regardless of the male spouse’s employment status, but the positive effect is enhanced when either they or their male spouses were previously employed. If either the female or male spouse is building up an employment history in Canada, the employment income of the former will lift their families out of poverty more easily.

Ethnic Variations in the Employment of Immigrant Women and Poverty Exit

This section considers ethnic variations in the exit from poverty. First, I perform probit regression models for the European group and a non-European group (Arabs/West Asians, South Asians, East/Southeast Asians, or others) separately.Footnote 17 Given the groups’ small sample sizes and their relatively similar cultural characteristics and outcomes, I combine Arab and West Asian groups into one group (Read 2004). Using probit coefficients, means, and proportions of the variables in the models, I then perform the Fairlie decomposition technique.Footnote 18

Table 4 shows that the employment of immigrant women has a positive impact on the exit from family poverty for European, South Asian, and East/Southeast Asian groups when the male spouse’s employment characteristics and family characteristics are taken into account (p < .01). By contrast, the employment of Arab/West Asian and other origin women has no statistically significant impact.

Table 4 Probit estimates of exit from poverty in Wave 3 for European, Arab and West Asian, South Asian, East/Southeast Asian, and other ethnic origin groups, aged 25–54, Canada, 2000–2005
Full size table

Table 4 also indicates ethnic variations in demographic and socioeconomic characteristics. While East/Southeast Asian women have the highest percentage of postsecondary credentials (83 %), they are least fluent in English or French (16 %) (column 8). The English/French skill levels of Arab/West Asian and South Asian women are comparable with their European counterparts (around 35 %), yet their postsecondary credential rates are somewhat lower (57 % and 68 % vs. 71 %) (columns 4 and 6). These compositional differences may contribute to the differences in poverty exit among immigrant women from different ethnic origins.

To test this possibility, I decompose differences in the predicted probabilities of poverty exit between the European and four non-European groups using the Fairlie technique. As Table 5 shows, the predicted poverty exit rate gap between the European and Arab/West Asian groups (net of control variables) is about 39 % (column 1). More than 70 % of this gap can be explained by the compositional differences in observable characteristics between the two groups (column 2). When the overall effect of compositional differences on the poverty exit gap is further decomposed into the effects of compositional differences among covariates, the results show the following. The lower female employment rates of the Arab/West Asian group explain 24 % of the ethnic poverty exit gap. This result is greater than the effect of compositional differences in the Wave 3 employment characteristics of their male spouses (19 % and 1 % for changes in weekly earnings and weeks worked from Wave 2 to 3, respectively). The analysis indicates a strong association between the lower employment rates among Arab and West Asian women and their lower poverty exit rates, which is consistent with expectations noted in the literature review. The male breadwinner model is thus more likely to apply to this group.

Table 5 Decompositions of ethnic origin gaps in poverty exit rates, for low-income immigrant women aged 25–54 living with a male spouse in Waves 1–3
Full size table

Although the female employment gaps between South Asians and East/Southeast Asians, and Europeans are narrower (14 % and 13 %, respectively; Table 4), the lower employment rates among these non-European groups also explain a notable portion of the differences in the exit from poverty. About one-third (31 %) of the difference in the probability of poverty exit is explained by the lower employment rates of South Asian and East/Southeast Asian women (columns 4 and 6). This is a fairly large contribution, but the male breadwinner model would apply less to these groups, as the male spouse’s Wave 3 employment characteristics (changes in weekly wages and weeks worked from Wave 2 to 3) make comparable contributions. These findings lead me to conclude that the male breadwinner model is supported for select non-European groups, namely Arab/West Asian groups, given their notably low female employment rates and the larger impact on the lower poverty exit rates when compared with their European counterparts.

Conclusion

Does the employment of recently arrived immigrant women help their families exit poverty? Although none of the existing theories addresses this specific question, two contrasting research hypotheses focus on the economic contributions of employed women in general. When extended to immigrant women, the equal earner/female breadwinner model suggests the employment of recently arrived immigrant women makes a significant contribution to family income, while the male breadwinner model underplays their role in their family’s economic well-being. My analysis of data from the LSIC supports the former model given that the employment of female spouses has, on the whole, a positive impact on the exit from family poverty four years after arrival in Canada, regardless of the male spouse’s employment. However, the results also suggest that the male breadwinner model is not necessarily a thing of the past. The employment rates among immigrant women of some non-European origins are notably low; thus, the male breadwinner model applies to some subpopulations of recent arrivals. Because immigrants experiencing poverty in early settlement stages are more vulnerable to persistent poverty, the lower female employment rates among low-income recent immigrants and the associated lower poverty exit rates may mean that their poverty will persist for years (Picot et al. 2008).

I also find a wide ethnic variation in the employment rates of immigrant women and ethnic variations in poverty exit. The findings corroborate Read and Cohen’s study (2007) of immigrant women in the United States in that female employment rates vary by ethnic origins. However, my finding of wider ethnic variations may be explained by the fact that the sample comprises recent immigrants in Canada for less than four years. Moreover, as the cultural explanation indicates, ethnic groups from strong patriarchal cultures may have significantly lower female employment rates. This pattern may change with increased time in the host country. My results suggest that the lower female employment rates do not necessarily deter specific ethnic minority groups from quickly recovering from their initial economic hardships. Instead, they broadly apply to new immigrants of non-European origins. Nevertheless, the wide ethnic variations in early settlement may contribute to the persistence of stratification along ethnoracial lines (Lian and Matthews 1998). Although an examination of why these women are less likely to work is beyond the scope of the article, the issue has social policy implications. Policy options may include removing barriers to the employment of non-European immigrant women (e.g., job training including Canadian work culture, affordable daycare for new immigrants, diversity training for employers).

Finally, while this study contributes to bridging the immigrant poverty and female immigrant employment literatures, it has implications for research on immigrant inequality in general. Its analyses of the impact of immigrant women’s employment on poverty exit and the decomposition of ethnic poverty exit gaps demonstrate that the study of immigrant inequality requires attention to both ethnic and gender inequalities. It notes the high poverty concentration among non-European groups and finds possible gender inequality (e.g., lower employment rates of women, lower earnings, lower hiring rates). Although a quantitative study like this one is unable to probe the mechanisms/sources of such persistent gender/ethnic inequalities due to data limitations, it highlights the inequalities.

Despite its contributions, the present work has some limitations. For example, it considers the economic effect of immigrant women’s employment in general. As the literature on precarious employment argues, individuals’ employment arrangements can significantly influence their earnings and, as a result, their family incomes (Kalleberg 2009). A question worth addressing is whether immigrant women’s various employment arrangement patterns (e.g., part-time work, self-employment, temporary work) lead to differential contributions to their family income. A Canadian study finds wide variations in immigrant women’s employment arrangement by source regions and ethnic origins (Cranford and Easton 2009). Immigrant women from Southeast Asia are more likely to engage in permanent full-time employment than Canadian-born European women, whereas immigrant women from East Asia have higher propensities for self-employment. To consider such varied patterns of employment arrangements and their impacts on poverty exits, future research will require multivariate probit models, which will allow an estimate of multiple outcomes (e.g., full-time work vs. part-time work vs. no work) and permit a more accurate prediction of the chances of exiting poverty as a function of these employment arrangements (Babalola et al. 2008).