The Effect of Family Disruption on Children’s Personality Development: Evidence from British Longitudinal Data

Abstract

This research documents the effects of different forms of family disruptions—measured by separation, divorce and death—on personality development of British children included in the 1970 British Cohort Study. There are statistically significant correlations between family disruptions prior to the age of 16 and personality development in early childhood. Parental divorce has the largest negative effect on a child’s personality development. Family disruptions have smaller effects on personality development when children are older and patterns differ by gender. The relationship between personality development and family disruption is partially driven by selection. Placebo regressions reveal significant correlations between family disruption and personality development before disruption. The omitted variable bias is mitigated by investigating mechanisms through which the selection operates.

Appendix: Rutter, Self-Esteem, and Locus of Control

Three measures are used to capture personality traits: self-esteem, internal locus of control, and a Rutter index for behavioural problems. Self-esteem measures an individual’s sense of self-worth, internal locus of control measures a child’s perceived achievement control, and the Rutter index gives an indication of behavioural problems.

Self-esteem assesses respondents’ self-esteem with reference to teachers, peers and parents and consists of 12 items, while locus of control measures children’s perceived achievement control with a list of 15 items. Items are scaled 0 “yes” 1 “don’t know” 2 “no”, except the first self-esteem item “Parent like to hear about ideas” and the locus of control item “Believe in planning ahead”, which are reversely coded. Higher scores thus indicate higher self-esteem and higher internal locus of control. Table 7 shows the inter-item correlations or covariances for these sets of items, and the Cronbach’s alpha statistic for the scale formed from them, both for ages 10 and 16. The self-esteem and locus of control scores are constructed by summing the scores on the set of 12/15 items, and then standardizing the resulting score to have mean zero and standard deviation one.

Table 7 Inter-item correlations and covariances of self-esteem and locus of control items

At ages 5, 10, and 16, the respondents’ parents, usually the mother, are asked to indicate to what extent a certain description applies to the respondent. At ages 5 and 16, the items are scaled 0 “does not apply” 1 “applies somewhat” 2 “certainly applies”. The age-10 items are scaled 1 through 100, where 1 indicates “does not apply” and 100 indicates “certainly applies”. Higher scores indicate more severe behavioural problems. Table 8 shows the inter-item correlations or covariances for the 19 items, and the Cronbach’s alpha statistic for the scale formed from them. The three Rutter scores, at ages 5, 10, and 16, are constructed by summing the scores on the 19 mother-rated items, and then standardizing the resulting score to have mean zero and standard deviation one. The correlation coefficients between the resulting measures for self-esteem, locus of control, and behavioural problems (Rutter) are reported in Table 9.

Table 8 Inter-item correlations and covariances of Rutter items

Table 9 Correlation coefficients between personality measures

For each set of of items, we apply multiple imputation through chained equations to impute missing values, using the remaining items as predictors for the item of interest. By doing so, we increase the number of observations with valid personality measures. For each item i within each set of Rutter/self-esteem/locus of control items, we estimate the following model to predict missing values:

$$\begin{aligned} {\textit{Item}}_{i}=\alpha _{i}+\sum _{j \ne i} \beta _{j} {\textit{Item}}_{j} + \epsilon _{i} \end{aligned}$$

This process starts with the item with least amount of missing observations. Missing observations are then imputed, and the process continues to the item with next fewest missing observations. For the age-10 Rutter score we apply the above model using linear regression, since the items are scaled from 1 through 100. Imputed values below 1 are set to 1, and those above 100 are set to 100. For the age-5 and age-16 Rutter scores, and the age-10 and age-16 self-esteem and locus of control scores, we use multinomial logit regression to impute missing values.