Procedure and Participants
Data were used from the 1970 British Cohort Study (BCS70), which is an ongoing multi-disciplinary longitudinal study that monitors the development and lives of around 17,000 individuals in Britain. The mothers of 17,000 babies, born in a single week in April 1970, were asked if they would be willing to participate in the study with their newborn baby (Elliott & Shepherd,
2006). BCS70 has collected information about the health, educational, physical, and social development, and economic circumstances of participants. The initial BCS70 participants are referred to as Generation 1 (G1) in the present study and their offspring as Generation 2 (G2). For the analyses reported here, data from the following waves were used: 1975 (G1 age 5), 1980 (G1 age 10), 1986 (G1 age 16), and 2004 (G1 age 34). The overall participation rate for the waves used in this study was between 75.00% (G1 age 34) and 88.90% (G1 age 10). Detailed information on assessments and instruments used are provided in various cohort descriptions (Butler, Bynner, et al.,
2016; Butler, Dowling, et al.,
2016; Bynner et al.,
2019; University of London et al.,
2020).
During the first three waves, only data referring to G1 were collected. Information about G2 was collected when G1 participants were 34 years old, resulting in a wide age range of G2 because naturally G1 had started having children at different ages. To account for developmental differences among G2, different questionnaires were used to assess behavior of offspring aged 3 to 5 (G2 early childhood), 6 to 9 (G2 middle childhood), and 10 to 16 (G2 adolescence). By design, each G2 child belongs to only one age category.
In the first wave (G1 age 5), data were collected from 13,135 G1 participants. At G1 age 34, information about G2 was based on 5207 children from 2846 G1 participants. Only parent-child pairs were included in the analyses for whom information about peer aggression was available for both generations. Adopted children (n = 10) were excluded to account for genetic transmission. Some participants (G1) had more than one child (G2) in the same age category or had twins, which would introduce dependency of observations and violate modeling assumptions. Therefore, from these families, only the oldest G2 child was included in the analyses and one of the twins was selected randomly. Analyses are ultimately based on 2929 parent-child pairs.
The G1–G2 early childhood subsample contained 1132 G1 cohort members (59.8% female), mostly from the European ethnic group (40.5%). At age 34, most of G1 were married (79.4%), had a profession within the managerial-technical social class (30.1%), and had two children (54.9%). Children in the G1–G2 early childhood subsample had a mean age of 4.05 years.
The G1–G2 middle childhood subsample contained 1088 G1 cohort members (67.0% female), mostly from the European ethnic group (39.2%). At age 34, most of G1 in this subsample were married (72.2%), had a profession within the managerial-technical social class (25.1%) and had two children (56.7%). Children in the G1–G2 middle childhood subsample had a mean age of 7.52 years.
The G1–G2 adolescence subsample contained 709 G1 cohort members (78.3% female), mostly from the European ethnic group (35.4%). At age 34, most of G1 in this subsample were married (59.1%), had a profession within the skilled non-manual social class (19.6%) and had two children (46.4%). Children in the G1–G2 adolescence subsample had a mean age of 12.50.
Measures
Peer aggression
G1 peer aggression was assessed at multiple time points during childhood and adolescence. All assessments were included to use these data as well as possible by forming latent variables with peer aggression assessments as indicators. To account for developmental differences in peer aggression, separate latent variables for early and middle childhood were created. Adolescent peer aggression was included as manifest variable.
In detail, G1 early childhood peer aggression was assessed at age 5 when parents reported on whether their child fought with other children or bullied others (1 = doesn’t apply to 3 = certainly applies). The responses to these two items were used as indicators for the G1 early childhood aggression latent variable. G1 middle childhood peer aggression was assessed at age 10 from parents who reported on whether their child fought with or bullied others (continuous, ranged from 0 doesn’t apply to 100 certainly). The responses to these two items were used as G1 middle childhood aggression indicators. G1 adolescence peer aggression was assessed at age 16 from parents who reported on whether their child bullied others (1 = doesn’t apply to 3 = certainly applies). This item was used as manifest variable. For G1 child and adolescent peer aggression measurements, no time frame was given.
G2 peer aggression was assessed from parents at the G1 age 34 assessment by asking parents whether their child had fought with or bullied other children over the past 6 months (1 = not true to 3 = certainly true). No age differences were made in the assessments.
Harsh parenting
Harsh parenting was assessed from parents who reported how often in the last 3 months they had told their child off, had shouted at their child, and had smacked their child on a 5-point Likert scale (1 = never to 5 = daily) at age 34. The responses to these items were used as three indicators for a latent variable representing harsh parenting.
Socioeconomic status
Social class of G1 at age 34 was used as an indicator of SES in the home environment of G2 (1 =
professional, 2 =
Managerial-technical, 3.1 =
Skilled non-manual, 3.2 =
Skilled manual, 4 =
Partly skilled, 5 =
Unskilled, 6 =
Other). A previously published conceptualization by BCS70 researchers who use the British ‘Social Class based on occupation’ as indicator of SES was followed (see for example, Akasaki et al.,
2019).
G2 sex
Sex of child was assessed from parents during the G1 age 34 assessment (1 = male, 2 = female). In the G2 early and middle childhood samples 52% were girls and in the G2 adolescence sample 49% were girls.
Parent-child dyad composition
Parent-child dyads were coded to represent father-son, father-daughter, mother-son, and mother-daughter dyads. In the G2 early childhood sample, 20% were father-son dyads and equally many were father-daughter dyads, 28% were mother-son, and 32% were mother-daughter dyads. In the G2 middle childhood sample, 18% were father-son dyads, 15% were father-daughter dyads, 34% were mother-son dyads, and 33% were mother-daughter dyads. Finally, in the G2 adolescence sample, 11% were father-son dyads and equally many were father-daughter dyads, 40% were mother-son dyads, and 38% were mother-daughter dyads.
Analytic Strategy
R
psych package version 1.9.1.31 (Revelle,
2019) was used for descriptive analyses and Hmisc package version 4.6-0 (Harrell,
2021) was used to create correlation heat maps. Because of the skewed distribution of peer aggression and harsh parenting, Spearman’s correlations were computed.
Analytic models were estimated as structural equation models in Mplus 8. In most models, G1 peer aggression was constructed as latent variable with two indicators except for G1 peer aggression in adolescence, which was assessed with one item only and this item was entered as manifest variable in the model. Generally, latent variables can be identified if three indicators are present, but all latent variables were estimated in a larger model, thus information could be “borrowed” from other items, allowing for a latent variable based on two indicators only (G1 early and middle childhood aggression). Measurement models and structural models were estimated simultaneously, in this way making use of all available information of each path (Loehlin & Beaujean,
2017). Harsh parenting as latent variable consisted of three items in all models. SES and G2 peer aggression consisted of single items in all models. Nine models were computed in total, these represent peer aggression assessed at different developmental periods in G1 and G2 (but note that one model did not converge). Each model consisted of direct effects from G1 to G2 peer aggressions as well as indirect effects via harsh parenting and via SES. Harsh parenting and SES were allowed to correlate in the model. Sex of G2 and dyad composition were included as covariates. Indirect effects were bootstrapped using 5000 draws.
Bootstrapping required the use of maximum likelihood estimation, missing data were accounted for by using full information maximum likelihood estimation. For each model,
X2, Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Standardized Root Mean Square Residual (SRMR), and Root Mean Square Error of Approximation (RMSEA) were evaluated. Models with a nonsignificant
X2, i.e.,
p > 0.05 (Barrett,
2007), a CFI value larger than 0.90, a SRMR value lower than 0.10 (Kline,
2005), and a TLI value close to 1.0 (Loehlin & Beaujean,
2017) were considered well-fitting. For RMSEA, a value less than 0.06, was considered to constitute a good fit (Hu & Bentler,
1999).
To test whether effects differed by G2 sex or dyad composition, models were also computed in a multiple group framework, to evaluate whether model fit improved significantly (evaluated as Chi2 difference) when girls and boys, and different dyad compositions (mother-son, mother-daughter, father-son, father-daughter) were free to vary in substantive associations between variables. Given the large number of conducted comparisons (nine per potential moderator), a p value threshold of p < 0.01 was handled as cut-off for further interpretation.