Confirmatory factor analysis
Mothers
The ASQ was first fit using a diagonally weighted least square confirmatory factor analysis for the 39 items administered. Based on the five pre-established fit criteria, the 39-item model showed good fit to the data (χ2 = 11110.333, df = 692, and p<0.001, CFI = 0.918, TLI = 0.912, RMSEA = 0.065 and AGFI = 0.939). Despite criteria showing a good fit, perusal of the loadings showed six low item loadings (<0.3). These items were removed without substantially reducing measurement model fit (χ2 = 5038.197, df = 485, and p<0.001, CFI = 0.961, TLI = 0.957, RMSEA = 0.051 and AGFI = 0.968). The final 33-item measure produced a KMO of 0.94, and Bartlett’s sphericity test was significant (χ2 = 45676.37, df = 528, and p <0.001). The 39-item models and correlation matrix are shown in Supplementary Figures 2 – 4. Cronbach’s α ranged from 0.72 – 0.80 across each factor in the mother cohort (Supplementary Table 2).
Offspring
Similarly for the offspring, the fit criteria based on 39-items showed good fit to the data (χ2 = 12719.393, df = 692, p<0.001, CFI = 0.921, TLI = 0.916, RMSEA = 0.068, and AGFI = 0.940). The same six items that were removed from the CFA in the mothers cohort also had low item loadings (<0.3) in the offspring cohort and were again removed. The final 33-item model resulted in an adequate fit (χ2 = 6272.182, df = 485, and p<0.001, CFI = 0.959, TLI = 0.955, RMSEA = 0.057, and AGFI = 0.967). The 33-item measure produced a KMO of 0.94 and significant Bartlett’s sphericity test (χ2 = 52698.13, df = 528, and p <0.001). The 33-item model for both mothers and offspring cohort are shown in Figure 1. The 39-item models and correlation matrix are shown in Supplementary Figures 2 – 4.
Cronbach’s α ranged from 0.77 – 0.80 in the offspring cohort (Supplementary Table 3).
Sex invariance
For the 33-item ASQ, the unconstrained multi-sex model showed good fit (χ2 =144047.632, df = 1056, CFI = 0.963, TLI = 0.960, RMSEA = 0.054, SRMR = 0.062) showing that across sexes, the same factorial structure exists. Next, metric testing was conducted where sexes had equal factors loadings but varying intercepts. The Chi-square statistic was significant (p = 0.045) and a substantial change in ΔCFI was observed (0.045) although, group differences between individual item loadings were minor. However, as conditions were not met for indicating metric invariance, testing for scalar invariance was not carried out. A summary of invariance testing is provided in Table 3. Additionally, the 33-item models for male and female offspring are shown in Supplementary Figure 3.
Convergent validity
Convergent validity of the ASQ and measures of depression/anxiety symptoms in the mothers and offspring are presented in Table 1. There was a similar pattern of correlations between the ASQ-33 and depression/anxiety symptoms for both mothers and offspring. There were strong positive correlations for discomfort with closeness (DWC), need for approval (NFA) and preoccupation with relationships (PWR) subscales (0.54 to 0.72) and moderate positive correlations for relationships as secondary (RAS) (0.38 to 0.42) with depression/anxiety symptoms, for both mothers and offspring. There were strong negative correlations for the confidence (CON) subscale (-0.54 to -0.66) for depression/anxiety symptoms in both mothers and offspring.
Differences in mean ASQ subscale scores for mothers and offspring scoring in the clinical range for anxiety and depression compared with the remaining 90% are presented in Table 2. The largest differences were observed on the NFA and PWR subscales which both represent anxious-attachment. For the mother cohort, mean differences (MD) of 1.2 (2.4 vs. 3.6) and 0.9 (2.8 vs. 3.7) respectively, were observed, with the higher mean scores being for those with clinical symptoms of anxiety and depression. For offspring, similar differences were observed (NFA MD = 1.5 [2.8 vs. 4.3], PWR MD = 1.1 [3.1 vs. 4.2]), with the clinical range anxiety and depression group having higher anxious-attachment scores.
Latent class analysis
Mothers
A latent class analysis was used to assess each of the five ASQ subscales against Bartholomew and Horowitz’s four-category model of adult attachment. Contrary to the four classifications based on high versus low attachment avoidance and attachment anxiety in this model, the analysis identified four classes of attachment security based on simultaneous changes in avoidance and anxiety. These classes of attachment security were labelled as ‘high’ (19%), ‘moderate’ (42%), ‘low’ (29%) and ‘very low’ (10%). Almost three quarters (73.5%) of mothers classified as ‘high secure’ were in the strongly agree category of the confidence subscale, and more than 90% of mothers were categorised as totally disagreeing with items on the relationships as secondary and need for approval subscales (91.8% and 92.7% respectively). The overwhelming majority of ‘high secure’ mothers were categorised as either totally or strongly disagreeing with items on the discomfort with closeness and preoccupation with relationships scales (88.3% and 96.6% respectively). Mothers in the ‘very low secure’ group were most likely to be categorised as slightly disagreeing with items on the confidence subscale and most likely to be strongly disagreeing with relationships as secondary (AAv). They were most likely categorised as slightly agreeing with items on the need for approval (AAx) and preoccupation with relationships (AAx) subscales. Item response probabilities are shown in Table 4.
Offspring
The four offspring classes were labelled in parallel with the mother sample based on similar patterns of item response probabilities. Population percentages differed, the top two attachment security groups had lower population proportions than the mothers (‘high secure’ 13% vs 19% and ‘moderate secure’ 37% vs 42%), while the ‘low secure’ and ‘very low secure’ groups had a higher population percentage (37% vs. 29% and 14% vs. 10%, respectively). Offspring classified as ‘high secure’ were most likely to be categorised as strongly agreeing (70.0%) on the confidence scale, and totally disagreeing on the four remaining subscales (DWC 50.1%, RAS 65.9%, NFA 80.5%, PWR 57.1%). ‘Very low secure’ offspring were more likely to slightly disagree on the confidence (53.9%) and relationships as secondary (40.7%) subscales and slightly agree on each of the other three subscales (DWC 58.2%, NFA 64.1%, PWR 57.2%). Item response probabilities are shown in Table 4.