Method

Study selection

A comprehensive search of the literature published until 2017 was conducted to identify research on the relation between parenting style and empathy in children and adolescents up to age 23. The included studies were found in three consecutive steps. The first step was to identify studies through keyword searches in electronic databases, including Google-Scholar, ERIC, PubMed, PiCarta, Scien- ceDirect. Proquest, Whiley, Narcis, and Web of Knowledge. Keywords included parenting and empathy, and several keywords related to parenting (e.g. maternal responsiveness, harsh discipline) and empathy (e.g. victim-based orientation, emotional understanding). In the second step, studies were searched by using the snowball method. This entailed inspection of the reference sections of relevant (already retrieved) articles, narrative reviews, and book chapters. In the third step, all authors of the included studies were emailed to ask whether they were working on or knew other relevant studies that were not yet found and retrieved. Finally, 14 studies examining the association between parenting style and their children’s empathy were included (see Table 4.1 for an overview of included smdies).

Study coding procedures

Potential moderators of the association between children's empathy and parenting were grouped into publication characteristics, sample characteristics, study characteristics, and assessment characteristics. Publication year and impact factor of the journal in which the study was published were coded as publication characteristics. As for the sample characteristics, the gender of the children in a sample (males, females, or both), the children’s ethnicity (in terms of the percentage Caucasian/White children in a sample), age of the children (early childhood, middle childhood, or adolescence), and the socio-economic status of families in the sample (low, medium, or high SES) were coded. Further, as an assessment characteristic, the type of instrument that was used to assess parenting style and empathy (questionnaires or observations), and the country where the research was conducted was coded as a study characteristic. Finally, the parenting style - with the categories authoritative, authoritarian, permissive, and neglectful parenting - and the dimension of empathy - with the categories cognitive empathy, affective empathy, and empathy without further specification - were coded.

Measures of empathy in the selected studies

None of the selected studies in this meta-analysis used the Basic Empathy Scale (BES; Jolliffe & Farrington, 2006) as an empathy measure. A total of seven studies were conducted before the development of the BES in 2006. Some studies used observational measures, such as facial expressions, to investigate empathy, especially in younger children (Kiang et al., 2004; van der Mark et ah, 2002; Zhou et ah, 2002). Some studies (Farrant et ah, 2011; McGrath & Zook, 2011;

Table 4.1 Included studies

Note. # Studies = number of independent studies; # ES = number' of effect sizes: Intercept/Mean r = mean association between parenting and empathy; 95% Cl = 95% confidence interval

*p<05, **p< 01; ***p< .001

Schaffer et al., 2009; Strayer & Roberts, 2004) used questionnaires distinguishing between cognitive and affective empathy, which is in line with the two- dimensional structure of the BES.

Statistical analyses

To correct for dependency in effect size caused by the extraction of multiple effect sizes from individual primary studies, a three-level random effects model was used for estimating an overall association between parenting style and empathy in youth and conducting moderator analyses. In this three-level approach to rneta- analysis, three different sources of variance are modelled (van den Noortgate, Lopez-Lopez, Marin-Martinez, & Sanchez-Meca, 2015). The sampling variance of each effect size estimating a population effect size is modelled at the first level. At the second level, variance in effect sizes within studies is modelled (i.e. within- study variance), and at the third level, variance in effect sizes retrieved from different studies (i.e. between-study variance) is modelled. We applied the procedure and R script described by Assink and Wibbelink (2016).

Parameters were estimated using a Restricted Maximum Likelihood method. The first step was to calculate an overall estimate of the effect size. Next, whether the variance at level 2 (within studies) and the variance at level 3 (between studies) significantly deviated from zero was investigated by performing two onesided log-likelihood ratio tests. This was done by testing whether constraining either the level 2 or level 3 variance deteriorated the model fit compared to the unconstrained model in which both variance components were estimated. In case of significant level 2 or 3 variance, univariate moderator analyses were performed to investigate which moderators significantly influenced the association between parenting style and empathy. Prior to performing analyses, a check for outliers in all Fischer r scores was performed by determining whether standardised values exceed (—)3.29. but no outliers were identified.

To test for publication bias, a trim and fill analysis was performed to investigate whether the distribution of effect sizes was asymmetrical and, if that was the case, what effect sizes should be imputed to the right or the left side of the funnel to restore the symmetry in the effect size distribution (Duval & Tweedie, 2000a, 2000b). With this procedure whether the estimated overall effect may be an underestimation or an overestimation of the tme effect can be checked. If effect sizes are missing on the left side of the funnel, there is indication of publication bias in the effect sizes that are analysed.