Procedure and participants
All respondents included in this study participated in the prospective cohort study JOnG! [13], which followed the development of mental health, family and healthcare of a Flemish cohort of twelve-year olds. All parents of twelve-year old adolescents living in eight districts (both urban and rural areas) of Flanders (N = 9861) were invited to participate in this study. This sample represents 15.2 % of all twelve-year olds in Flanders [13]. The JOnG!-study is commissioned, financed and steered by the Ministry of the Flemish Community (Department of Economics, Science and Innovation; Department of Welfare, Public Health and the Family). The work was performed by the Policy Research Centre for Welfare, Public Health and the Family and in assistance of a collaboration between two Flemish universities.
The JOnG! study was approved by the Ethics Committee of both universities cooperating in the JOnG! project. All participating adolescents and parents gave informed consent. In addition, parents gave informed consent for the adolescent reports.
In total, 1499 families provided informed consent and agreed to participate in this longitudinal study. Of all 1499 families who provided informed consent, we received 93.20 % (N = 1397) valid adolescent reports (age 12) and 95.93 % (N = 1438) valid parent reports (i.e., valid reports are those questionnaires with less than 10 % missing data). Parent reports were completed by 88.70 % mothers, 4.30 % fathers, 1.20 % step, adoptive or foster parents. Adolescent reports consisted of 54.70 % girls and 45.30 % boys. At time 2, 1132 adolescents (age 13) and their parents participated in this study resulting in 827 valid adolescent reports (73.57 %) and 936 parent reports (82.68 %). At time 3 (one and a halve years after time 2), in total 839 adolescents (age 14) and their parents participated, resulting in 754 valid adolescent reports (89.86 %) and 790 valid parent reports (94.50 %). The adolescent-reported data were described elsewhere [2]; whereas the present study focuses on the parent-reported data.
Participants (both adolescents and parents) with and without complete data were compared in terms of gender, district, presence of psychiatric disorder, and psychological complaints. The Missing Completely At Random (MCAR) [14] test resulted in a non-significant Chi-square value, χ
2 (177) = 175.70, which suggests that missing data are completely at random. Missing data are handled using full information maximum likelihood (FIML). Data analyses were conducted with Mplus using a robust mean- and variance-adjusted chi square estimator (WLSMV), which is appropriate for binary variables, in this study NSSI (present/absent) [15]. NSSI is a binary predictor (0/1), which only changes from 0 to 1 and not by a standard deviation. Using two different methods of standardization in one figure would be confusing, therefore only the unstandardized pathways are shown in the figure.
Measures
Engagement in NSSI was assessed by means of a single-item screening question in both adolescent and parent questionnaires. Adolescents were asked at time 1 ‘Have you ever intentionally injured yourself (e.g., cut, burned, scratched), without the intent to die? (Yes/No)’, and at time 2 and time 3 they were asked “Did you intentionally injure yourself since the previous survey? (Yes/No)’. Parent reported NSSI at age 13 and age 14 was examined as follows: ‘Has your son/daughter ever intentionally injured him/herself – e.g., by cutting, burning, scratching – without the intent to die?’ (Yes/No). According to Muehlenkamp and colleagues [5], the use of a single-item measure of NSSI renders consistent estimates of NSSI prevalence.
Parenting behavior was measured with the parent-reported Parental Behavior Scale, short version (PBS) [16] (time 1, 2 and 3). The subscales ‘positive parenting’, ‘parental rules setting’, ‘punishing’, and ‘harsh punishing’ were used in this study. The PBS subscales ‘positive parenting’ and ‘parental rules setting’ (including both limit setting and learning rules) were used as indicators of parental support. The PBS subscales ‘punishment’ and ‘harsh punishment’ were used as indicators of parental control. The results of confirmatory factor analyses confirmed this model in previous studies [2]. At time 1, the Cronbach’s alpha coefficient was .85 for positive parenting and .70 for parental control. The Cronbach’s alpha coefficients for positive parenting and parental control were respectively.86 and .78 at time 2. At time 3, the Cronbach’s alpha coefficients were.87 for positive parenting and .77 for parental control.
Analyses
First, means, standard deviations, correlations, and reliability coefficients were calculated (see Table 1). All continuous subscales were rescaled to Percent of Maximum Possible Scores (POMP) [17]. A POMP score is the percentage of the distance (0-100 %) from the minimum to the maximum of a scale, which allowed us to examine both the magnitude and impact of the observed relationships between variables even when the underlying units of metric are different.
Cross-lagged path analyses were conducted using Mplus 7.3 [15]. The four subscales of parent-reported PBS [16] at all three time points, next to adolescent reported NSSI behaviors at three time points were entered in a cross-lagged model to examine reciprocal effects. The full model is shown in Fig. 1. Model fit was estimated by means of the Comparative Fit Index (CFI) and the Root Mean Square Error of Approximation (RMSEA). The CFI should exceed .90 for a reasonable fit and .95 for a good fit to the data, and the RMSEA should be less than .05 for a close approximate fit, or between .05 and .08 for a reasonable fit to the data [18].