With duplicates removed, the search process yielded 5,381 results screened for eligibility based on title and abstract first. Afterwards 297 full-texts were retrieved. Figure 1 depicts the whole process in more detail. In total, 41 studies were considered in the qualitative syntheses; 27 studies were included in the quantitative syntheses. Please refer to the Online Resource for a corresponding reference list.
------------------------- please insert Fig 1 here -------------------------
Table 1 provides brief information on study- and sample-level characteristics, for more details please refer to the Online Resource (Supplementary Tables S1, S2, and S3; extensive information on results of included studies appear in Supplementary Tables S4, S5, S6, and S7). In total, 34 studies focused on maternal employment status or working amount, whereas eight studies investigated employment duration, and six studies included analyses on timing of return to work. Child mental health was operationalized by BP in 15 studies, by EBP in 26 studies, by IBP in 15 studies, and by PB in nine studies. Analytical sample sizes varied from n = 49 to n ≈ 10,100. In 20 studies the association of interest was investigated longitudinally and nearly every study provided adjusted results. Most research stemmed from the US and relied on data of large cohort studies. Unfortunately, risk of bias had to be considered high for a huge majority (see Supplementary Table S3). Turning to sample characteristics, children were ten months [39] to 7.5 years old [40, 41] at outcome assessment. Studies investigated the association in both genders, with only few of them providing distinct estimates for girls and boys. Timing of maternal employment varied from the first year [e.g., 42] to seven years postpartum [43]. Most studies focused on samples of mixed ethnicity investigating (majority) coupled mothers.
------------------------- please insert Table 1 here -------------------------
Overall behavior problems
Comparing children of employed and unemployed mothers, four out of 10 studies [39, 44-52] pointed to employment being associated with less BP [44, 45, 48]. Subgroup analyses of Berger and colleagues [47] emphasised the importance of ethnicity. In Hispanic children maternal employment was linked to higher levels of BP whereas in Black children it was associated with lower levels of BP.
A meta-analysis including eight studies [44, 45, 47-52] indicated no association of maternal employment status and BP based on the random-effects model (OR = 0.717 [0.472 – 1.088], P = .118). As the hypothesis of homogeneity had to be rejected (Q (7) = 151.904, P < .001) moderator analyses followed (see Online Resource, Supplementary Table S8). At study-level, data source (Q (5) = 22.01, P < .001) was of importance. When compared to ECLS-B results MCS data were associated with a lower effect size (b = -1.104, P < .001). Here, being employed was linked to lower BP. When taking data source as dummy for country British data still were associated with a lower effect size (b = -0.821, P < .05). Focusing on sample characteristics, ethnicity emerged as moderator (Q (3) = 16.970, P < .001). In comparison to (majority) White samples both samples of mixed ethnicity (b = 0.924, P < .001) and samples without information on ethnicity (b = 0.786, P < .01) were associated with a higher effect size. Here, being employed was linked to more BP. Due to the small number of studies, joint effects of data source and ethnicity could not be assessed.
Focusing on full-time vs. part-time employment, results of six available studies [44, 49, 50, 53-55] did not yield any associations with BP. In a meta-analysis including four studies [44, 49, 50, 54] no summary effect emerged, neither based on the random-effects model (OR = 0.996 [0.761 – 1.304], P = .976) nor the fixed-effects model (OR = 1.020 [0.941 – 1.106], P = .631). As the hypothesis of homogeneity had to be rejected (Q (1) = 22.918, P < .001) moderator analyses followed (see Supplementary Table S8). At study-level, adjustment (Q (1) = 6.30, P < .05) was of importance with adjusted results being associated with a higher effect size (b = 0.454, P < .05). Here, full-time employment was linked to more BP. Focusing on sample characteristics, timing of maternal employment emerged as moderator (Q (1) = 6.30, p < .05). Measuring maternal employment later in children’s life was associated with a higher effect size (b = 0.121, P < .05). Here, full-time employment was linked to more BP. In an adjusted analysis including both characteristics, i.e., adjustment and timing of maternal employment, none of the single coefficients remained significant.
Further four studies [40, 41, 44, 48] focused on employment duration. In analyses of Kiernan and Mensah [48] and McMunn and colleagues [44], both using MCS data, single results indicated longer employment being linked to lower BP. A meta-analysis including two studies [41, 44] indeed indicated an association of maternal employment duration and BP based on both the random-effects model (r = -.072 [-.123 – -.020], P < .01) and the fixed-effects model (r = -.082 [-.109 – -.055], P < .001). Longer employment was associated with less BP (see Figure 2). As the hypothesis of homogeneity could not be rejected (Q (1) = 2.168, P = .141), no moderator analyses followed.
------------------------- please insert Fig 2 here -------------------------
Only a single study [50] included analyses on timing of return to work. Here, no interpretable results emerged.
Externalizing behavior problems
Overall externalizing behavior problems. Comparing children of employed and unemployed mothers, five out of 11 studies [6, 16, 49, 56-63] yielded significant but mixed results. Employment was linked to either less EBP [6, 61] or more EBP [6, 58, 59, 63], the latter especially in cases of full-time employment [58, 59]. A meta-analysis including seven studies [6, 16, 49, 58-61] indicated no association of maternal employment and EBP based on the random-effects model (OR = 1.031 [0.869 – 1.224], P = .724). As the hypothesis of homogeneity had to be rejected (Q (1) = 40.077, P < .001), moderator analyses followed. Here, none of the variables under investigation turned out to be of importance (see Supplementary Table S8).
Focusing on full-time vs. part-time employment, four studies [42, 61, 64, 65] did not yield any associations, whereas results of Hill and colleagues [58] and Haas [66] pointed to full-time employment or more maternal working hours being linked to more EBP. Analyses of Brooks-Gunn and colleagues [16] again emphasised the importance of ethnicity. In White children some evidence pointed to full-time employment being associated with higher levels of EBP, whereas this was shown for part-time employment in Black children instead. A meta-analysis including three studies [16, 49, 65] indicated no association of maternal employment and EBP based on the random-effects model (OR = 1.271 [0.816 – 1.980], P = .289). However, under the fixed-effects model working full-time was associated with more EBP (OR = 1.119 [1.010 – 1.239], P < .05; see Figure 2). As the hypothesis of homogeneity had to be rejected (Q (1) = 15.147, P < .01), moderator analyses followed. None of the variables under investigation turned out to be of importance (see Supplementary Table S8).
Further three studies [17, 57, 67] focused on employment duration with a single study finding more months of employment (within the same job) being associated with less EBP [17]. A meta-analysis based on all three studies indeed indicated an association of maternal employment duration and EBP based on the fixed-effects model (r = .038 [.018 – .058], P < .001), albeit pointing to longer employment being associated with more EBP instead (see Figure 2). Under the random-effects model no association emerged (r = .012 [-.071 – .095], P = .777). The hypothesis of homogeneity had to be rejected (Q (1) = 28.105, P < .001), but due to the small number of studies included, no moderator analyses followed.
Four studies [16, 42, 58, 59] included further analyses on timing of return to work often pointing to early return, i.e., within the first year postpartum, being associated with more EBP [16, 42, 59]. As these analyses were hardly comparable to each other, no meta-analysis followed.
Aggression. Comparing children of employed and unemployed mothers, three out of five studies [68-72] pointed to employment being associated with more aggression [68, 70, 71]. However, the opposite was found as well [70, 71], especially under higher levels of social support received by mothers. A meta-analysis including four studies [68-71] did not yield a summary effect, neither under the random-effects model (OR = 1.168, P = .548) nor under the fixed-effects model (OR = 0.865, P = .108). As the hypothesis of homogeneity had to be rejected (Q (1) = 15.251, P < .01), moderator analyses followed with none of the variables under investigation turning out to be of importance (see Supplementary Table S8).
Whitaker and colleagues [73] focused on the association of employment duration and aggression. Here, no associations were obtained.
Hyperactivity/inattention. Four studies [39, 59, 70, 74] compared children of employed and unemployed mothers. Lekfuangfu and colleagues [59] detected maternal full-time employment to be linked to more hyperactivity/inattention but analyses of Nomaguchi [70] yielded some evidence of part-time (part-year) employment being associated with less hyperactivity instead. Here, in a meta-analysis including three studies [59, 70, 74] no summary effect emerged, neither under the random-effects model (OR = 1.041, P = .261) nor under the fixed-effects model (OR = 0.956, P = .664) The hypothesis of homogeneity had to be rejected (Q (1) = 12.437, P > .01), but due to the small number of studies included, no moderator analyses followed.
Focusing on full-time vs. part-time employment, Hadzic and colleagues [75] found 35–40 maternal weekly working hours to be directly linked to more hyperactivity/inattention when compared to 16–34 working hours. Under consideration of maternal parenting practices as mediating variables further indirect effects emerged linking 1–15 and more than 40 working hours to less hyperactivity/inattention. A meta-analysis including two studies [70, 75] indeed indicated an association of maternal employment and hyperactivity/inattention based on the fixed-effects model (OR = 2.677 [2.222 – 3.226], P < .001). Here, working full-time was linked to more hyperactivity/inattention (see Figure 2). Under the random-effects model no summary effect emerged (OR = 3.003 [0.305 – 29.571], P = .346). The hypothesis of homogeneity had to be rejected (Q (1) = 149.022, P < .001), but due to the small number of studies included, no moderator analyses followed.
A further analysis on employment duration did not yield any associations [73]. The same holds true for studies focusing on timing of return to work [59, 74]. As these analyses were hardly comparable to each other, no meta-analysis followed.
Conduct problems. Comparing children of employed and unemployed mothers, one out of two studies [59, 74] found full-time employment to be linked to more conduct problems [59]. A meta-analysis including both studies indeed yielded a summary effect based on the random-effects model (OR = 1.180 [1.055 – 1.320], P < .01) and the fixed-effects model (OR = 1.168 [1.086 – 1.257], P < .001). Employment was linked to more conduct problems (see Figure 2). As the hypothesis of homogeneity could not be rejected (Q (1) = 2.147, P = .143), no moderator analyses followed.
Focusing on full-time vs. part-time employment, Hadzic and colleagues [75] found 1-15 and more than 40 weekly working hours to be indirectly linked to less conduct problems when compared to 16–34 working hours. In terms of return to work analyses yielded mixed results. Lekfuangfu and colleagues [59] found some evidence of early return, i.e., within months seven to 12 postpartum, to be associated with more conduct problems. However, results of Lombardi and Coley [74] pointed to an association with later return instead, i.e., between months 9 to 24 postpartum, especially when working part-time and with increasing household income. However, this link was no longer present when further potentially mediating variables were included into the model. As these analyses were hardly comparable to each other, no meta-analysis followed.
Internalizing behavior problems
Overall internalizing behavior problems. Of five studies comparing children of employed and unemployed mothers [56, 58, 59, 62, 63], only Im and Vanderweele [63] found results linking employment to more IBP (especially in cases of absent or low paternal involvement). Employment was associated with less IBP under high paternal involvement instead. A meta-analysis including two studies [58, 59] verified the impression of no association under the random-effects model (OR = 0.922 [0.798 – 1.065], P = .268). However, under the fixed-effects model a summary effect emerged (OR = 0.896 [0.836 – 0.962], P < .01), indicating that children of employed mothers exhibited less IBP (see Figure 2). As the hypothesis of homogeneity could not be rejected (Q (1) = 2.98, P = .08) no moderator analyses followed.
Focusing on full-time vs. part-time employment, in three studies no interpretable results emerged [58, 64, 65], replicated by a meta-analysis including two studies [58, 65] both under the random-effects model (OR = 1.068 [0.925 – 1.234], P = .369) and the fixed-effects model (OR = 1.068 [0.925 – 1.234], P = .369). As the hypothesis of homogeneity could not be rejected (Q (1) = 0.031, P = .860) no moderator analyses followed.
Further two studies focused on employment duration with limited evidence pointing to more IBP in children of mothers unemployed longer [17, 67]. A meta-analysis including these studies was conducted with the random-effects model yielding no summary effect (r = -.075 [-.220 – .073], P = .318). However, under the fixed-effects model being employed longer was linked to less IBP (r = -.072 [-.103 – -.040], P < .001; see Figure 2). Albeit the hypothesis of homogeneity had to be rejected (Q (1) = 21.764, P < .001), due to the small number of studies included, no moderator analyses followed.
In terms of return to work analyses of Lekfuangfu and colleagues [59] included some evidence that returning within months 13–18 postpartum may be associated with less IBP when compared to not having returned to work by 18 months postpartum. Hill and colleagues [58] could not obtain any associations instead. As these analyses were hardly comparable to each other, no meta-analysis followed.
Anxiety and depressive symptoms. In terms of employment status, only a single study [70] out of five in total [43, 57, 69, 70, 72] included some evidence of less anxiety among children of employed mothers. A meta-analysis based on two studies [69, 70] indeed yielded no summary effect, neither under the random-effects model (OR = 0.837 [0.694 – 1.009], P = .062) nor under the fixed-effects model (OR = 0.837 [0.697 – 1.004], P = .055). As the hypothesis of homogeneity could not be rejected (Q (1) = 1.057, P = .304), no moderator analyses followed.
Further two studies focused on employment duration [57, 73] with only Whitaker and colleagues [73] finding children of mothers employed shorter to have higher odds of exhibiting problematic anxious/depressed behavior. No meta-analysis was conducted as respective outcomes were considered to be not similar enough to be combined.
Other internalizing symptoms. Lekfuangfu and colleagues [59] investigated emotional and peer problems not finding any differences among children of employed and unemployed mothers. However, having returned to work by 7–12 months postpartum was linked to less emotional and peer problems when compared to not having returned to work by 18 months postpartum. Osborne and Knab [69] focused on withdrawn behavior also not finding any differences among children of employed and unemployed mothers.
Prosocial behavior
Comparing children of employed and unemployed mothers, four out of seven studies [39, 49, 57, 59, 70, 74, 75] yielded at least some evidence for more PB in children of employed mothers [39, 49, 70, 75]. However, in a meta-analysis including five studies [49, 57, 59, 70, 74] no summary effect emerged, neither under the random-effects model (OR = 0.997 [0.785 – 1.266], P = .979) nor under the fixed-effects model (OR = 1.020 [0.973 – 1.068], P = .413). As the hypothesis of homogeneity had to be rejected (Q (1) = 94.522, P < .001) moderator analyses followed (see Supplementary Table S8). At sample-level, marital status was of importance (Q (1) = 3.96, P < .05). When compared to (majority) coupled samples mixed marital status was associated with a smaller effect size (b = -0.179, P < .05). Here, being employed was linked to less PB.
Focusing on full-time vs. part-time employment, Hadzic and colleagues [75] found indirect links of 1–15 maternal weekly working hours and more PB as well as of more than 40 hours and less PB when compared to 16–34 working hours. In analyses of Gassman-Pines [64] more working hours during nights were linked to less positive behavior (especially on weekends). A meta-analysis including two studies [49, 70] did not yield a summary effect, neither under the random-effects model (OR = 0.979 [0.789 – 1.216], P = .850) nor under the fixed-effects model (OR = 1.027 [0.927 – 1.138], P = .610). As the hypothesis of homogeneity could not be rejected (Q (1) = 2.522, P = .112), no moderator analyses followed.
Further two studies [57, 67] focused on employment duration with limited evidence pointing to less PB in children of mothers being unemployed longer [67]. A meta-analysis including these studies yielded a summary effect under the fixed-effects model (r = .023 [.000 – .047], P < .05), linking longer employment to more PB (see Figure 2), whereas under the random-effects model no summary effect emerged (r = .047 [-.054 – .147], P = .362). The hypothesis of homogeneity had to be rejected (Q (1) = 14.391, P < .001), but due to the small number of studies included, no moderator analyses followed.
In terms of return to work analyses included some evidence of early return, i.e., within the first year postpartum, to be associated with less PB [59, 74]. As these analyses were hardly comparable to each other, no meta-analysis followed.