Fathers’ Mental Health and Children’s Aggressive Behaviour A Study Based on Data from the Norwegian Mother, Father and Child Cohort Study (MoBa)

The present study is based on data from the Norwegian Mother, Father and Child Cohort Study (MoBa) [15, 17]. The establishment of MoBa and initial data collection was based on a licence from the Norwegian Data Protection Agency and approved by the Regional Committee for Medical and Health Research Ethics. The MoBa cohort is currently regulated by the Norwegian Health Registry Act. The present study was approved by the Regional Committee of Medical and Health Research Ethics (ref. no. 2010-3204). MoBa is a prospective population-based pregnancy cohort study. The participants were recruited from all over Norway from 1999 to 2008. Pregnant women consented to participate in 40.6% of all available pregnancies. The cohort now includes 114,500 children, 95,200 mothers, and 75,200 fathers. The present study is based on version V, April 2010, of the quality-assured data files released for research on fathers’ mental health and child development. To be included in the study, each family was required to complete the following questionnaires: mothers’ questionnaire at 17 or 18 weeks of gestation, fathers’ questionnaire at 17 or 18 weeks of gestation, and mothers’ reports concerning their children at 18 months, 3 years and 5 years of age. There were 20,155 completed questionnaires for (triads of) fathers, mothers and children for these measurement points.

Fathers’ mental health in pregnancy weeks 17 or 18 was assessed by the 5-item Symptom Checklist (SCL-5) [18, 19]. The SCL-5 is a screening measure of psychological distress that is used as an indicator of global mental distress. The SCL-5 consists of five items: (1) ‘Feeling fearful’, (2) ‘Nervousness or shakiness inside’, (3) ‘Feeling hopeless about the future’, (4) ‘Feeling blue’ and (5) ‘Worrying too much about things’. Each of the five items is scored on a scale from 1 to 4, depending on how bothered the respondent has been in that area during the 14 days prior to the time of self-report: 1 = ’Not bothered’, 2 = ’A little bothered’, 3 = ’Quite bothered’, and 4 = ’Very bothered’. The checklist mainly screens for symptoms of anxiety and depression [20]. Descriptive characteristics of the mean scale are shown in Table 1. Intra-scale consistency as estimated by Cronbach’s alpha was 0.79. The customary cut-off of 2.00 [19] for case-level psychological distress was used.

At children’s age 18 months and 3 and 5 years, mothers were asked to respond to several of the items from the Child Behavioural Checklist Revised (CBCL-R) [20‐22], including the child hitting others. The item was rated on a 3-point scale: 0 = ’Not true’, 1 = ’Somewhat/sometimes true’ and 2 = ’Very true/often true’. The responses were dichotomised with children’s hitting others reported as ‘Sometimes true’ and ‘Often true’ in one category (= 1) and ‘Not true’ (= 0) in the other.

Dagitty is a graphical tool for analysing causal diagrams, and it is also known as directed acyclic graphs (DAGs). To identify the variables that the statistical models should be adjusted for, the following relevant constructs were entered into a dagitty diagram: fathers’ age, education, marital status, somatic condition, cigarette smoking, prenatal psychological distress, marital satisfaction and whether they lived with the family; mothers’ age, prenatal psychological distress, marital satisfaction, daily care of the children (kindergarten), number of children in household and whether the pregnancies were planned; children’s temperament, language development and whether they hit others. According to the resulting dagitty diagram, the models should be adjusted for children’s temperament, children’s language development, number of children in household, daily care, whether fathers were living with family, fathers’ age, mothers’ age, and mothers’ prenatal psychological distress. Descriptive statistics of these variables are shown in Fig. 1 and Tables 1 and 2.

The MoBa included 12 of the original 20 items constituting the Emotionality Activity Sociability (EAS) Temperament Survey [21] at children’s age 18 months, 3 years and 5 years. The 12 items represented the original ‘Emotionality’ (three items), ‘Activity’ (three items), ‘Shyness’ (three items) and ‘Sociability’ (three items) subscales of the EAS. In the Norwegian version of the instrument, the item ‘Makes friends easily’ was rephrased into ‘Enjoys being with other children’ [21]. The original MoBa items were scaled from 1 to 5, with 1 representing ‘Very characteristic or typical of your child’, and 5 meaning ‘Not characteristic or typical of your child’. Nine of the 12 items were reverse-scaled so that for all items, 1 corresponded to low temperament and 5 to high temperament. Figure 1 shows the EAS item frequencies at 18 months and 3 and 5 years of age.

The original EAS subscales each included five items that added up to produce scales ranging from 5 to 25. In the present study, four averaged subscales representing children’s activity, emotionality, sociability and skewness were computed at each point of measurement by adding up the three items from each of these EAS dimensions available in MoBa and then dividing the sum of these items by the number of available items within each subscale. The means and standard deviations (SDs) of these subscales representing children’s temperament are shown in the lower part of Fig. 1. Inter-item reliability as assessed by Cronbach’s alpha ranged from 0.64 to 0.66 for the activity, 0.63 to 0.75 for the emotionality, 0.32 to 0.71 for the sociability and 0.65 to 0.69 for the shyness subscales across children’s age of 18 months, 3 years and 5 years.

The MoBa questionnaire included the 10-item MSS [22]. Examples of items are ‘My partner and I have problems in our relationship’, ‘I am very happy in my relationship’, ‘I am satisfied with my relationship with my partner’; and ‘We agree about how children should be raised’. Each of the items was scored on a 6-point scale: 6 = ‘Totally agree’, 5 = ‘Agree’, 4 = ‘Slightly agree’, 3 = ‘Slightly disagree’, 2 = ‘Disagree’ and 1 = ‘Totally disagree’. Negatively worded items were reverse-scaled. A summary scale was created by adding up the item scores (Table 1). Higher scores on the summary scale indicated a more positive relationship with the partner. Internal scale consistency, as assessed by Cronbach’s alpha in pregnancy, was 0.88 for fathers’ and 0.91 for mothers’ MSS.

Descriptive statistics of mothers’ SCL-5 mean scale in pregnancy are shown in Table 1. Intra-scale consistency in pregnancy as estimated by Cronbach’s alpha 0.57 for mothers’ SCL-5 total scale. Mothers’ mean SCL-5 scores were M = 1.3, SD = 0.39, range 1–4, M = 1.3, SD = 0.41, range = 1–4, and M = 1.2, SD = 0.37, range 1–4 at children’s age 18 months, 3 years and 5 years, respectively.

We included two items from the Ages and Stages Questionnaire (ASQ) [23]. The ASQ is a screening instrument for child development from four to 60 months of age based on parents’ reports [24]. It contains 30 items, with each item scored on a 3-point scale: 1 = ‘Yes’, 2 = ‘A few times’ and 3 = ‘Not yet’. Two items are used in the present analysis (1) ‘Without giving him/her help by pointing or using gestures, ask your child to put the shoe on the table and put the book under the chair. Does your child carry out both of these directions correctly?’ (impressive language skills) and (2) ‘Can your child tell you at least two things about an object he/she is familiar with? If you say, for example, tell me about your ball, will your child answer by saying something like, It is round, I can throw it, it is big? ‘ (expressive language skills).

Repeated measures data have a hierarchical structure which can be analysed using multilevel models. The effect of children’s temperament on the association between fathers’ prenatal psychological distress and children’s hitting was therefore investigated using linear mixed effects models (LMEs). Analyses were done in three steps. First, we assessed the association without children’s temperament by using the unadjusted LME model for children’s hitting depending on fathers’ SCL-5 in pregnancy and children’s age, as well as their interaction, with a random individual intercept and simple contrasts over time. Second, we adjusted the first model for each of the four EAS temperament scales to study the effect of these adjustments on the model parameters. Finally, we adjusted each of the four EAS-adjusted models successively for the number of children in the household, daily care (kindergarten), children’s language skills and further characteristics. Further characteristics were entered as a block consisting of fathers’ and mothers’ age, mothers’ SCL-5 and whether fathers were living with the family. These adjustment variables were selected using DAGs. The models were tested in three steps and compared using Analysis of Variance (ANOVA). This required the same number of observations in each step. This is why we used the cases available in all steps to compare the models and all available observations to interpret the models. All analyses were done both for the total sample of all children and stratified for children’s gender.

The computation was done in the nlme package of R 3.6 (R Core team [25]) and IBM SPSS 23 (IBM Corp., Armonk, NY), and graphics were created using Matlab 9.0 (Mathworks, Natick, MA). Tests were two-tailed with a significance level of 0.05.

The associations of children’s age and hitting for different levels of fathers’ prenatal psychological distress as measured by SCL-5 in pregnancy are shown in Fig. 2. Without adjusting for children’s temperament, we observed that for fathers without prenatal psychological distress (i.e., SCL-5 = 1), there was no change in children’s hitting from children’s age 18 months to 3 years, while children’s hitting decreased significantly from 18 months to 5 years (p < 0.001) (Table 3). With increasing SCL-5 scores in fathers, children’s hitting increased significantly from children’s age 18 months to 3 years (B = 0.046, 95% confidence interval (CI) [0.014, 0.079], p = 0.005) and decreased to the same level as for fathers without prenatal psychological distress (i.e., SCL-5 = 1) again at 5 years (Fig. 2a).

As shown in Table 3, Fig. 2 and Online Resource 1, the association of fathers’ prenatal psychological distress and children’s age with children’s hitting was unaffected by children’s temperament, with one exception. We observed a significant increase in children’s hitting also for fathers without psychological distress (B = 0.029, 95% CI [0.019, 0.039], p < 0.001) when adjusting for the EAS ‘Activity’ subscale. However, the associations between the interaction terms of fathers’ SCL-5 with children’s age were not affected by adjusting for the ‘Emotionality’, ‘Shyness’ and ‘Sociability’ subscales. The coefficients for the association of fathers’ prenatal psychological distress and change in children’s hitting from children’s age 18 months to 3 years (0.042–0.048) varied within the 95% CI of the unadjusted model, while the associations of fathers’ prenatal psychological distress and change in children’s hitting at children’s age 18 months and changes from 18 months to 5 years remained nonsignificant after adjusting for children’s temperament. Adjusting for other family- and parent-related variables did not lead to an essential change in the coefficients for the associations of fathers’ prenatal psychological distress and children’s age with children’s hitting.

The LME model, as stratified by children’s gender, is shown in Table 3. For boys, we observed an increase in children’s hitting from children’s age 18 months to 3 years (B = 0.041, 95% CI [0.027, 0.055], p < 0.001). This increase was independent of fathers’ prenatal SCL-5. Otherwise, the same pattern as described for all children was found in boys. For girls, the hitting decreased significantly, both from 18 months to 3 years (B =  − 0.033, 95% CI [− 0.046, − 0.019], p < 0.001) and 5 years (B =  − 0.171, 95% CI [− 0.185, − 0.158], p < 0.001).

For increasing prenatal psychological distress in fathers, we observed an estimated increase of hitting from children’s age 18 months to 3 years in boys (B = 0.047, 95% CI [0.001, 0.093], p = 0.045). With increasing prenatal psychological distress in fathers there was also a descriptive, but nonsignificant, increase in hitting for girls from 18 months to 3 years of age (B = 0.043, 95% CI [− 0.004, 0.089], p = 0.070).

The present study was population-based with a large sample size and, as such, has the statistical power to detect even weak associations between fathers’ prenatal psychological distress and children’s, age, temperament and hitting. By testing these aspects of the association between fathers’ mental health and children’s aggressive behaviour in a population sample, the serious selection biases that are commonly found in clinical studies have probably been avoided. However, we cannot exclude inclusion biases as limitations. The study participants may differ from nonparticipants in ways that are not random with respect to the exposure and outcome [30, 31]. However, Nilsen et al. [30] previously concluded that only the prevalence estimates of exposure and outcome, not the exposure–outcome associations, were biased because of self-selection in MoBa [30]. Further, the prospective design of the study is a strength because it makes a child-to-parent effects less likely as the cause of the associations found. Rather, the associations found represent evidence of parent-to-child directionality. Some limitations regarding the information gathered and operationalisation of the measures need to be mentioned. First, all data were self-reported by fathers or mothers and, thus, are subject to reporting biases, possibly limiting the reliability of the findings. Second, the reliability of the findings could have been higher if complete valid clinical instruments were used. Unfortunately, only 12 of the original 20 items of the EAS were available in the MoBa and the construct reliability of the measure may be reduced. We chose to measure hitting as outcome variable because it is a concrete type of physically aggressive behaviour, which makes it relatively easy for the parents to discern and quantify. However, the outcome variable ‘hitting’ had only three response categories and, therefore, a limited variance. The restriction of variance on the outcome variable could also lead to the weakening of the associations studied. Finally, the lack of postnatal information about fathers’ psychological distress in the MoBa study limits the ability to draw causal inferences about the effects of fathers’ postnatal psychological distress on their children’s hitting.