Introduction

Over the past few decades, international migrant labor has increased and is predicted to grow further (Martin and Zürcher 2008). Despite original intentions to return to the home country, many migrant laborers stay in the host country and establish families, frequently without legal status (Castels 2004). Of the many unauthorized migrants worldwide, approximately 50 % are minors, who too are undocumented (Lindert et al. 2008). This phenomenon is causing concern worldwide on many levels including implications for economic, health care, social, and child development policies.

The social-ecological environments of undocumented children of migrant workers include varying levels of risk factors. Multiple variables at the community, neighborhood, and family levels influence these children’s development adversely (Meir and Slone 2013). Migrant workers are disproportionately likely to live in poverty (Rector 2006) and neighborhoods are frequently segregated across class, race, ethnicity, and language (Orfield and Lee 2006). Distrust of public authorities may lead to housing problems, avoidance of reporting crimes and violations, and lower use of resources. The combination of financial need, fear of authority and indebtedness forces many unauthorized parents to work long hours and can lead to absence from the home. In addition, discrimination and the constant threat of deportation further complicate the lives of families without legal status (Suárez-Orozco et al. 2011).

Growing up in these conditions compromises children’s development on all levels including cognitive, emotional and social development (Meir and Slone 2013). Cognitive skills of undocumented children are at risk relative to their peers in authorized families, possibly related, in part, to attending segregated under-resourced schools and receiving limited parental educational support (Crosnoe 2006; Han 2006; Ortega et al. 2009; Yoshikawa 2011). Multiple stress factors influencing these families may cause a variety of socio-emotional difficulties among children and include a high incidence of reported depressive, anxiety, and post-traumatic stress disorder (PTSD) symptoms (Suárez-Orozco et al. 2011). There is limited research on the psychosocial implications of growing up unauthorized, but a recent study conducted in Israel has shown that high exposure to stressful life events, such as neighborhood violence, deportation of parents and parental absence from the home, was correlated with children’s severe emotional and behavioral difficulties (Meir et al. 2012).

In the 1990’s, Israel initiated a policy of migrant labor as a result of political and economic changes in the region (Kampf and Raichman 2003). This policy led to entry of masses of migrant laborers from different countries over the years such that in the year 2009, the number of foreign workers staying illegally in Israel reached 208,000 (Weisblay 2009). By law, children born to migrant workers are not granted legal status in Israel. There is no accurate estimate of the size of the migrant workers’ children population due to lack of reporting of births but it has been estimated at approximately 2000 (Natan 2010). In Israel, most migrant workers live in South Tel Aviv in low socio-economic conditions which are characterized by living in shared and cramped apartments and a violent neighborhood. In addition, parents tend to work long hours, frequently leaving the child in informal crowded day-care facilities (Ben Rabi and Hassin 2004). Although these conditions place children of migrant workers at risk for developmental impairment, families have limited access to mental health and developmental services. Many children are in need of psychotherapy, however, due to limited resources, only a few children in extreme circumstances receive specialized intervention (Mesila 2009).

Although children of foreign workers represent the most vulnerable sectors in society, only few published studies have documented the effectiveness of specialized interventions aimed to enhance their mental health. The FRIENDS program is one example of a stress-resilience program applied to culturally diverse migrant groups residing in Australia and was found to be effective in lowering anxiety and promoting positive future outlook (Barrett et al. 2001). Another drama workshop program conducted in the US was effective in facilitating the adjustment of newly arrived migrant teens, although assessed using only qualitative methods (Rousseau et al. 2004). A study conducted in Israel testing a program for young children of migrant workers during a crisis situation of threat of deportation, was effective in lowering anxiety and depression and in teaching new coping strategies (Meir et al. 2012a, b). Since no interventions for school children of migrant workers have been conducted or examined empirically, the present research undertook to construct and examine the effectiveness of a group intervention program to enhance children’s mental health.

The present study examined the effectiveness of an intervention program aimed at enhancing self-efficacy and mental health of migrant workers’ school-aged children. Self-efficacy is commonly defined as one’s belief in the ability to succeed in a particular situation (Bandura 1982). These beliefs determine ways in which people think, feel and behave. A major assumption of self-efficacy is that individuals with a strong belief in their ability to succeed tend to cope better in different life situations (Bandura 1992; Eden 1996). High self-efficacy has been found to be one of the main protective factors for mental health in adults, adolescents and children (Rutter 2000) and to be positively related to mental health and negatively related to depression and anxiety (Luszczynska et al. 2009; Vieno et al. 2007). Enhancing self-efficacy may be a key factor in promoting mental health, especially in those situations when social support is minimal, as in the present population (Meir and Slone 2013).

The group intervention program enabled reaching out to large groups of children in their natural setting and together with their peer group. The intervention used multi-modal techniques such as art therapy, psychodrama, play therapy and psycho-education. The repeated measures design represents an empirical effectiveness study that evaluated the use of this intervention as opposed to a control group.

Research Hypotheses

The study advanced three hypotheses. The first predicted a greater improvement in self-efficacy between the pre-test and post-test for children in the intervention as opposed to control group. The second hypothesis predicted a greater reduction in the self- and teacher-reports of emotional, social and behavioral difficulties as measured by the SDQ child-report and by the two teacher-report scales (SDQ and CBCL) between the pre-test and post-test for children in the intervention as opposed to the control group. The third hypothesis examined the interaction between self-efficacy and intervention over time on the dependent variables and predicted an interaction between group (intervention or control), time (pre-test and post-test), and self-efficacy on pre-test (high or low) on emotional, social and behavioral difficulties as measured by children’s child-report and teacher-reports.

Method

Participants

Participants were 70 children born in Israel to illegal migrant workers from the Philippines, Africa and other countries (e.g. Romania, Turkey and South American) between the ages 8–12 (M age = 9.77, SD = 1.33), cared for in an after-school program. The final sample was composed of all children in the afternoon care facility that met the study criteria for age, fluency in Hebrew, and no diagnosis of pathology or diagnosed special needs. Sample size was limited according to children participating in the after-school programs in the academic year. All 70 children were eligible to participate in the study. After commencement of the trial, no changes were made in eligibility criteria. All parents complied with informed consent requirements. Children were randomly divided into intervention and control groups by the project manager from lists provided by the teachers. 77 children were allocated to intervention or control groups; however seven of them did not complete the study, four due to absence and three due to refusal to complete questionnaires. In total, the intervention group comprised 35 and the control 35 children. A participant flow diagram based on the consort model (Schulz et al. 2010) is presented in Fig. 1.

Fig. 1
figure 1

Participant flow diagram

Full size image

Comparison between intervention and control groups on demographic data showed no age differences between intervention (M age = 9.85, SD = 1.36) and control (M age = 9.70, SD = 1.31) groups, t(68) = −.49, ns. In addition, no socio-economic status differences (calculated as the ratio between the number of people living in the house and the number of rooms), were found between intervention (M = 1.64, SD = .86) and control (M = 1.83, SD = .81) groups, t(68) = .95, ns. The intervention group included 14 boys and 21 girls. The control group included 21 boys and 14 girls. These frequencies were not significantly different χ2(1, N = 70) = 2.8, ns. The intervention group included 16 children of Philippine origin, 8 of African origin and 11 children of other origins. The control group included 16 children of Philippine origin, 12 of African origin and 7 children of other origins. These frequencies were not significantly different, χ2(2, N = 70) = 1.70, ns. The intervention group included 19 children living in single-parent families and 16 living with both parents. The control group included 12 children living in single-parent families and 23 living with both parents. These frequencies were not significantly different χ2 (1, N = 70) = 2.83, ns.

Instruments

Demographic Questionnaire

Demographic data was collected using a short demographic questionnaire given to the children requesting details of age, gender, origin of parents, number of rooms and people living in the house. Age and origin were validated with teachers. This questionnaire has been used in other studies with similar populations of children (Meir et al. 2012a, b).

General Self-efficacy Scale (GSES)

The 17-item GSES (Sherer et al. 1982) measures a general set of expectations that the individual carries into new situations and includes items such as: “I give up easily”. Children were asked to mark their agreement with each item on a 1–5 scale. This measure has good content and criterion validity (Sherer et al. 1982). The reported internal consistency coefficients (Cronbach’s α) of the GSES in youth samples range from .83 to .86 (Sandor et al. 1994). In this study, Cronbach’s α coefficients ranged between .63 and .67. The reduction in reliability (as compared to Sandor et al. (1994) where the age range of participants was 14–18) may be attributed to the fact that participants in this sample were younger (8–12 years old). Generally, for younger participants, it is more complicated to obtain reliable self-reports (Mellor 2004).

Strengths and Difficulties Questionnaire (SDQ)

Children’s emotional, social and behavioral difficulties were assessed using the 25-item SDQ (Goodman et al. 1998). In this study, the teacher-report of teachers in the after-school program and the child-report versions were used. The questionnaire yields a total difficulties score and five scales comprised of five items each—Emotional symptoms, Conduct problems, Hyperactivity, Peer problems and Pro-social behavior. In this study, the total difficulties score, calculated as a sum of the four difficulties scales, was used. The instrument reports excellent criterion validity for community and clinical samples and high cross-informant correlations for child-report to parent-report and teacher-report (Goodman et al. 1998). In community samples, the internal consistency coefficients (Cronbach’s α) of the different subscales ranged between .70 and .88 for the teacher-report version, and between .41 and .67 for the self-report version (Goodman 2001). In addition, the SDQ has demonstrated good test–retest stability of .73 (mean of subscale correlations) in teacher-reports and a lower test–retest stability of .51 (mean of subscale correlations) in child-reports. The SDQ has good internal and criterion validity (Goodman 2001). In this sample, Cronbach’s α coefficients were acceptable for the total difficulties score teacher-report in the pre- and post-test (.61–.63) and for the child-report in the pre- and post-tests (.71–.72). Correlations between teacher and child-report were not significant in the pre- and post-tests.

The Child Behavior Checklist (CBCL)

The CBCL (Achenbach 1991; Achenbach and McConaughy 1987) comprises 113 items assessing symptoms rated by parents or adult informant familiar with the child on a 0 (not true) to 2 (very often) Likert scale. In this study, the CBCL was completed by the teachers in the afternoon care facilities where the children spend many hours a day, who were well-acquainted with the children. In community samples, the internal consistency coefficients (Cronbach’s α) of the different subscales ranged between .63 and .97 and test–retest reliabilities of the different subscales ranges between .73 and 0.94 (Achenbach 1991). In addition, authors report evidence of discriminative, concurrent, convergent, and predictive validities (Achenbach 1991). The scale has a Hebrew version and reports good construct validity, discriminating between normative and psychiatric admission children groups (Ofer 1983). In the current study, internal reliability was excellent with Cronbach’s α coefficients of .92 in the pre- and post-tests. For the purpose of the study, a total raw score was calculated, that is, the sum of all item values.

Intervention Program Materials

The program included ten weekly sessions of one and a half hour each, conducted in small groups of seven to eight children. Each group was led by two group facilitators, child clinical psychology graduate students, who went through 3 days training with a child clinical psychologist who specializes in the fields of group therapy and therapeutic work with children of migrant workers. On the first day of training, the group facilitators received a general introduction to the children’s background and their unique needs. In order to understand children’s life circumstances, the group facilitators visited relevant community sites such as kindergardens, churches and community gathering locations. In these sites, they meet with community workers who specialize in working with children of migrant workers. On the second day, the group facilitators learned about group interventions with children and about self-efficacy and its importance to psychological well being. On the third day, the group facilitators practiced the intervention program sessions under supervision and prepared all group materials for the intervention. In addition, after each session of the program, the group facilitators received supervision on their work from a child clinical psychologist. The intervention program was administrated in a quiet room allocated to the research in the afternoon care facility. The program called “A Journey to the Island Named Me”Footnote 1 aimed to enhance self-efficacy. The name of the program reflected the idea that children were invited into a process in which they discover themselves and strengthen their belief in themselves. In order to build a safe setting, each session had the same general structure, as presented in Table 1. The different sessions of the program are presented in Table 2.

Table 1 The structure of each session in the intervention program
Full size table
Table 2 Description of the intervention program sequence
Full size table

Procedure

After receiving authorization for the research from the University Helsinki Ethics Committee, informed consent was obtained from afternoon care facility teachers, parents and children. Teachers, parents and children were assured of confidentiality and were informed that all the data would be used solely for research purposes. Participating children were informed that they could terminate their participation at any time. All ethics procedures were adhered to including legal requirements in Israel. In addition, there was no conflict of interests between any of the organizations and researchers involved in the present study.

In this study, a randomized controlled design was used. The research took place at the beginning of the academic year. Children were randomly allocated by the project manager into intervention or control groups from a list provided by the teachers. Each child was assigned a number and the numbers were randomly distributed across the two groups without using any special methods such as blocks. Two weeks before the beginning of the program, eight experimenters, psychology students, who were blind to children’s group allocations, administered questionnaires to children and teachers. In order to keep data confidential, each child was given a personal code, which was written on teacher-report and child-report questionnaires of pre- and post-tests in order to match the data for each child. A week after termination of the program, the same experimenters administered the post-test battery consisting of the same questionnaires. After commencement of the trial, no changes were made in the outcomes. While children in the intervention group participated in the program, children in the control condition continued their regular activities, which included various social activities in the afternoon care facility. Children in the control group did not engage in any activity which was similar to the intervention program. All control group children received the intervention program immediately after termination of the study, therefore avoiding unequal opportunities.

Results

Four t tests for independent samples were conducted in order to test differences in baseline levels of the variables. Table 3 reports means and standard deviations for the variables divided between control and intervention group and between pre and post-test. No significant differences were found between the groups (intervention or control group) in baseline levels of all dependent variables; however a significant difference was found in baseline level of self-efficacy, such that the intervention group was lower in self-efficacy than the control group.

Table 3 Descriptive statistics and t tests for model variables—intervention versus control groups, pre and post-tests
Full size table

The research hypotheses were tested in a statistical model integrating the pre-test and post-test data. In order to test interactions, participants were divided into high or low self-efficacy groups. According to Preacher et al. (2006), the cut off points were around the means, that is, the cut off for high self-efficacy is above the mean plus one standard deviation, and the cut off for low self-efficacy is below the mean minus one standard deviation. However, in the regression runs, self-efficacy was treated as a continuous variable. Pre- and post-test values for self-efficacy, SDQ child-report and teacher-report and the CBCL were examined in a two-wave longitudinal analysis which belongs to the category of multi-level models. A full longitudinal analysis requires at least three waves to generate the intercept and slope for individual growth but, in this case, data were collected from only two time points. Thus, the two-wave model still considers a meaningful correlation between the different scales measured over time, yet is limited in describing a growth curve (Singer and Willett 2003). In contrast, conducting an ANOVA model with repeated measures is constrained by prior assumptions which are relaxed by the model suggested here. The resulting estimates are shown in Tables 4, 5, 6 and 7. A standard methodology for multi-level models is to run an unconditional model first which does not include any fixed effects and thus compares the variation of the dependent around its grand mean with respect to the random within-subjects and between-subjects (level one and level two, respectively). The unconditional model is set to define the percent variance explained by the within-subjects versus the percent variance explained by the between-subjects to intra-class correlation (ICC). The next models estimate direct fixed effects and interactions between levels. All sources of the interactions were calculated using Preacher’s calculator (Preacher et al. 2006). Observations with missing values were deleted (list-wise deletion). Only the SDQ reported by teachers showed higher level of missing values (24 % missing) while the percentage missing in the other models was 13 % for the CBCL and 1.4 % for SDQ child-report. A missing value analysis proved that missing data were completely at random (Little’s MCAR test: χ2 = 28.82, df = 24, p = .23), (Little 2013).

Table 4 Results for self-efficacy for fixed and random effects in a multi-level regression
Full size table

Table 4 presents analysis results for the first hypothesis showing estimates for self-efficacy change over time. As shown, 51 % of the total variance is explained by the level-two variation, the variation between children. When time is added as a fixed effect in the second model, it shows that self-efficacy increased over time (b = 4.43, p < 0.001). The model provides additional information to the unconditional model (∆χ2 = 15.27, p < 0.001). The third model provides additional explanation (∆χ2 = 4.21, p < 0.05) showing a significant two-way interaction between group type and time (b = 4.57, p < 0.05). This interaction is explained by an increase in self-efficacy levels for the intervention group (b = 6.71, t(70) = 4.33, p < .001) but not for the control group (b = 2.14, t(70) = 1.38, p = .17). These results provide support for the first hypothesis.

Tables 5 to 7 present the examination of the second and the third hypotheses. Each table presents data for a different dependent variable. The interaction between time and group type is set to test the second hypothesis, and the three-way interaction between self-efficacy, time, and group type is set to test the third hypothesis. As seen in Table 5 model 3, for the SDQ child-report an interaction effect was found between time and group type (b = −2.35, p < 0.05). The power of each regression model was calculated using the GPower calculator v3.1.3 (Erdfelder et al. 1996). Calculations consider medium Cohen’s f squared effect size at 0.15 (Cohen 1988). The power of this model is 1 − β = .82. This interaction derived from the reduction in child-report difficulties over time for the intervention group (b = −2.01, t(69.6) = 2.44, p = .02), but not for the control group (b = .33, t(69.6) = .43, p = .67). This result supports the second hypothesis. In addition a three-way interaction was found between self-efficacy, time and group (b = −.32, p < 0.01). This interaction derived from the moderation effect of self-efficacy on the interaction between group type and time (b = −3.29, t(71.9) = 4.77, p < .001), such that among children with high self-efficacy those who participated in the intervention group experienced reduction in their difficulties but this reduction was not experienced by children in the control group. This pattern was not found for children with low self-efficacy. Other possible interactions were insignificant. Figure 2 illustrates this interaction. These results support the third hypothesis suggesting that higher self-efficacy was related to reduction in children’s difficulties over time in the intervention group.

Table 5 Results for SDQ child report for fixed and random effects in a multi-level regression
Full size table
Fig. 2
figure 2

Interaction between group, time and self-efficacy on SDQ child-report

Full size image

The power of the SDQ teacher-report model is 1 − β = .86. A main effect for time was found. Teacher-report of children’s difficulties decreased over time (b = −2.51, p < .01). In addition, an interaction between time and group type was found (b = −3.50, p < 0.05), as shown in Table 6, model 3. This interaction derived from the reduction in teacher-report of children’s difficulties over time for the intervention group (b = −3.98, t(53.3) = 3.85, p < 0.001), but not for the control group (b = −0.48, t(53.3) = 0.41, p = .69). This result supports the second hypothesis. Interaction between self-efficacy, time and group type on the SDQ teacher-report was insignificant.

Table 6 Results for SDQ teacher-report for fixed and random effects in a multi-level regression
Full size table

The power of the CBCL model is 1 − β = .79. For the CBCL measure, an interaction between time and group type was found (b = −13.93, p < 0.05), as shown in Table 7, model 3. CBCL symptoms decreased over time for the intervention group (b = −6.93, t(67.8) = 1.83, p = .08), and increased for the control group (b = 7.01, t(67.8) = 1.78, p = .08). This result supports the second hypothesis. The interaction between self-efficacy, time and group type on the CBCL was insignificant.

Table 7 Results for CBCL measure for Fixed and Random Effects in a Multi-Level Regression
Full size table

Discussion

The study presented a novel intervention program aimed at enhancing mental health of a community sample of underprivileged children from one of the most deprived populations in Israeli society. Post-test improvement in measures of self-efficacy in the intervention but not the control group, confirmed the first hypothesis. Nevertheless, prior differences in baseline levels of self-efficacy may limit the interpretation of this finding. In line with the second hypothesis, the intervention program led to significant reductions in children’s emotional, social and behavioral difficulties as assessed by both child-reports and teacher-reports. A main effect emerged in which self-efficacy increased and teacher-report general difficulties decreased in both groups between the pre- and post-tests. A possible explanation for this effect might be attributed to the timing of conducting the intervention study at the beginning of the academic year. Possibly the results reflect the process of group change from the initial adjustment period at the beginning of the year to a more cohesive, settled group climate that occurs in many class settings. Since these children are exposed to multiple risk factors, that have been shown to negatively affect children’s development (Meir et al. 2012), the importance of increasing children’s sense of their own self-efficacy and their psychological adjustment in such an uncontrollable, uncertain environment cannot be over-emphasized (Rutter 2000). The intervention program presented here substantiates that this can be achieved in a short-term intervention.

Although impossible to separate the many elements that may be active in an intervention program, this program had a specific aim of enhancing self-efficacy. It was therefore necessary to examine whether this aspect of the program was indeed related to the reduction in symptoms. Findings revealed an interaction between group type, self-efficacy and time from the pre- to post-test only on the SDQ child-report measure. The results indicated that irrespective of group type, children with low self-efficacy reported no significant reduction in their difficulties. However, in the case of high self-efficacy, only children participating in the intervention group reported significant reduction in their difficulties. This interaction can imply that there was a specific combination of sense of high self-efficacy and engagement in the intervention group that led to an experience of decreased difficulties. It is possible that the particular focus on self-efficacy in the intervention program that encouraged its application as a coping strategy in stressful situations led to a sense of greater coping and decreased difficulties.

This relation between self-efficacy and difficulty reduction among intervention group participants did not occur for the teacher-report measures. Although teachers reported reduced difficulties among intervention group children from the pre- to post-test, this did not seem to be linked to the children’s own sense of their efficacy. Although at the level of the child’s experience, there was a relation between sense of efficacy and sense of having difficulties, it may have been difficult for the teachers to perceive the internal process of changes in the child’s sense of self-efficacy. This fortifies the importance of using multi-informants, since teacher reports cannot always reflect personal experience (Meir and Slone 2013). For instance, teachers and parents may report decreased levels of emotional distress than reported by children (Meir et al. 2012). This is further validated by the finding in the present study that no correlations were found between teacher-report and self-report of the SDQ. These discrepancies in parent–child reporting of emotional and behavioral problems have been found in other large samples (e.g. Van Roy et al. 2010). The differences in these reports may be due to different perspectives on well being, actual changes in child behavior and differences in thresholds of reporting difficulties. This validates the need for different informants when assessing children’s mental health, especially in multi-cultural environments (Stevens and Vollebergh 2008).

Applying community interventions with children from underrepresented populations is no small feat. Access to the research population was dependent on trust and cooperation both with official agents and with the participants themselves. Logistic difficulties in accessing this unique group of children and in maintaining a rigorous control and intervention program in a lively and vibrant environment limited possibilities for including further necessary measures. Both teachers and children completed the maximum questionnaires possible to administer in this setting. However, it would have been useful to examine additional variables including perceived support and meaningful child–adult relationships as moderators, and well-being and performance measures as outcomes. Further, addition of an active control group would have allowed for clearer separation of the significant elements of the intervention program. Another limitation of the study is the prior difference in base-line levels of self-efficacy, such that intervention group showed lower level of self-efficacy than the control group. On one hand, it could be that children with lower self-efficacy may be more interested in getting help to enhance their self confidence, therefore adhering to any kind of intervention. On the other hand, since intervention group children had lower self-efficacy, the increase in their self-efficacy after the intervention program may suggest its effectiveness. Finally, generalization of the study results should be viewed cautiously due to the small study sample.

Nonetheless, findings provide important evidence for the effectiveness of the intervention program, especially for this group of disadvantaged children. For mental health professionals, this study emphasizes that this short term playful intervention program proved to be effective in enhancing the psychological adjustment of extremely disadvantaged children, suggesting its application to other at-risk groups of children.