Gender and Age Differences in the Development of ...

3 downloads 0 Views 415KB Size Report
often than girls while girls show relational/indirect forms of aggression more of- ..... Sticks and stones and sugar and spice: Girls' and boys' aggression in schools ...
Short Report

Gender and Age Differences in the Development of Relational/Indirect Aggression: First Results of a Meta-Analysis Herbert Scheithauer1, Nicole Haag1, Jana Mahlke1, Angela Ittel2 Freie Universität Berlin, Germany Universität der Bundeswehr, Munich, Germany 1

2

It has been suggested in many studies and reviews that the form of aggressive behaviour displayed differs between the two genders. Studies that take on a developmental focus argue that the prevalence of aggression in general and relational/indirect aggression in particular varies over the developmental course of boys and girls with regard to its different function in children’s and adolescents’ biopsychosocial development. A closer look at the results on gender-specific expression of aggression reveals an ambiguous picture, however. Thus, the aim of this present paper is to discuss first results of a meta-analysis on studies that deal with questions regarding gender and age differences in relational and indirect aggression. Our results on 30 effect sizes from 23 primary studies on relational and indirect aggression revealed that there is sparse evidence for gender differences and for age differences in relational/indirect aggression, supporting a gender similarities hypothesis as opposed to a gender differences hypothesis. Further analyses are needed to investigate whether these differences are expression of developmental variances or the results of an underlying moderator effect of information source or other moderators. Keywords: relational aggression, indirect aggression, meta-analysis, gender differences, age differences

Introduction

Studies that investigate the prevalence, course, and outcome as well as risk factors of relational and indirect forms of aggressive behaviour in children and adolescents have emerged in the literature since the beginning of 1980s and 1990s. While these forms of aggressive behaviours are similar, they are viewed as distinct concepts. Relational aggression is described as any behaviour that is intended to harm someone by damaging or manipulating someone’s relationships with others (Crick & Grotpeter, 1995). Indirect aggression is defined by social manipulation and by attacking the target in circuitous ways (Björkqvist, Lagerspetz, & Kaukiainen; 1992a; Lagerspetz, Björkqvist, & Peltonen, 1988). Thus, although these forms of aggressive behavior hold unique features, they have substantial similarities: they relate to harming a person indirectly or directly by damaging the person’s social group or dyadic relationships. Examples of such damaging behaviour are defamation, exclusion or spreading rumours about someone (e.g., Björkqvist et European Journal of Developmental Science [EJDS]. 2008, Vol. 2, No. 1/2, 176–189 © Vandenhoeck & Ruprecht 2008, ISSN 1863-3811

Gender Differences in Relational Aggression

177

al., 1992a; Cairns, Cairns, Neckerman, Ferguson, & Gariepy, 1989; Crick, 1995; Crick & Grotpeter, 1995; Galen & Underwood, 1997; Tomada & Schneider, 1997). It has been suggested in many studies and reviews that while males supposedly use physical aggression to a greater extent than females, women are believed to prefer indirect and relational aggressive behaviour (e.g., Côté, 2007). But a closer look at the results on gender-specific expression of aggression reveals an ambiguous picture: Some results provide evidence that boys exhibit physical forms of aggression more often than girls while girls show relational/indirect forms of aggression more often than boys (Baron & Richardson, 1994; Björkqvist, 1994; Björkqvist et al., 1992a; 1992b; Crick, 1997; Crick & Grotpeter, 1995; Crick, Casas, & Mosher, 1997; Eagly & Steffen, 1986; Lagerspetz & Björkqvist, 1994; Lagerspetz et al., 1988; MacDonald & O’Laughlin, 1997; Normark Olafsen & Viemerö, 2000; Österman et al., 1998; Owens, 1996; Pakaslahti & Keltikangas-Järvinen, 1998; Salmivalli, Lagerspetz, Björkqvist, Österman, & Kaukiainen, 1996). Other study results attest that boys exceed girls not only in physical but also in relational aggressive behaviour (David & Kistner, 2000; Goldstein et al., 2002; Henington, Hughes, Cavell, & Thompson, 1998, Ittel, Kuhn, & Werner, 2005) while other studies seem to suggest a “gender balance” in the expression of aggression (Adler & Adler, 1995; Craig, 1998; Crick & Grotpeter, 1995; Green, Richardson, & Lago, 1996; Österman et al., 1994). The association of age and the expression of relational/indirect aggression has lead to similar variations in results. Studies that take on a developmental focus argue that the prevalence of aggression in general and relational/indirect aggression in particular varies over the developmental course of boys and girls with regard to its different function in children’s and adolescents’ biopsychosocial development. For example, as boys and girls grow older, their peer groups change in structure (e.g., most girls hold few but very intense friendships once they have reached puberty, most boys are part of bigger cliques without the intense emotional relationships that girls report of) and function (e.g., for girls friendships function as a context of emotion regulation within exclusive dyads or triads of friendships, for boys dominance and self-assertion in larger peercliques are important) (Berndt, 1982; Daniels-Beirness, 1989; Fabes, Eisenberg, Smith, & Murphy, 1996; Maccoby, 1990; Thorne, 1993). Therefore, it is assumed that the form of aggressive behaviour exhibited by boys and girls varies according to the changes in function and structure of their peer-groups over the course of their development (Björkqvist et al., 2000; Block, 1983; Crick, Bigbee, & Howes, 1996; Galen & Underwood, 1997). For example, Crick and Grotpeter (1995) emphasize that aggression is characterized particularly by behaviour “that thwart or damage goals that are valued by their respective gender peer-groups” (p. 710). As this is the case, relational or indirect aggression may be of particular importance for girls. That is: Girls’ aggressive behaviour is predominantly related to issues concerning the relationship level, while boys’ aggressive behaviour is related to issues of social status and dominance within their peer-group.

178

H. Scheithauer et al.

Björkqvist and colleagues (Björkqvist et al., 1992a, b; Lagerspetz et al., 1988, Lagerspetz & Björkqvist, 1994) have reported that girls use indirect aggression significantly more often than boys from age 11 onwards. Overall, their results emphasize an age(development-) dependent display of different forms of aggression: while preschool children use both overt and physical forms of aggressive behaviour in similar ways, older children develop overt-verbal and indirect/relational forms of aggression and use physical forms to a lesser extent (Björkqvist et al., 1992a; b). Considering girls’ rapid biological, cognitive and psychosocial development during the preschool years (Keenan & Shaw, 1997), girls start using more subtle forms of aggressive behaviour (e.g. social manipulation, indirect forms of aggression) earlier than boys. After school entry, when most boys “catch up” and the gap in biological, cognitive, and psychosocial developmental status disappears (Keenan & Shaw, 1997), these gender differences in behavioural conduct disappear as well. In puberty, however, the gender gap in the exhibition of aggressive behaviour reemerges (Leschied et al., 2000). In early adolescence girls use relational/indirect aggressive behaviour with increasing frequency. Studies have shown that intimacy, closeness, and popularity among peers, are especially relevant during this age period (Fuligni, Eccles, Barber, & Clements, 2001). Because girls tend to have smaller peer groups and due to girls’ strong emphasis on intimacy and negotiating of values and norms within their friendships, relational/indirect aggression is assumed to be especially frequent and harmful among same-age girls. However, the results are not conclusive and studies have often not included boys in their sample. There are also controversial results concerning the prevalence of relational and indirect aggression for the period of late adolescence, indicating either a gender balance (Crick, Werner, & Schellin, 1998, in Crick et al., 1998) or significant gender differences in relational aggressive behaviour (Kaukiainen, Lagerspetz, Björkqvist, & Jokila, 1993; MacDonald & O’Laughlin, 1997). In sum, irrespective of the amount of attention that the constructs of relational and indirect aggression have received, this short review demonstrates that there are at least two questions that are discussed controversially in the existing literature: 1.) whether there are gender differences in the prevalence and function of relational and indirect aggression, and 2.) whether there are changes in prevalence and function of relational or indirect aggression over the life course. The ambivalent results concerning age and gender differences necessitate a thorough appraisal of the various results. Thus, the aim of this present paper is to discuss first results of a meta-analysis on studies that deal with questions regarding the exhibition of relational and indirect aggression throughout the course of childhood and adolescence. Method

The present analyses are part of an ongoing meta-analysis project on gender and age differences in relational, indirect and social aggression (see Scheithauer, 2003).

Gender Differences in Relational Aggression

179

Search Strategy

We searched online databases (e.g. PSYCHINFO, PBSC, ERIC) by relevant keywords (A complete list of databases and keywords can be requested by the first author.). We included all primary studies that met the selection criteria (see below) and that were published before the year 2002. In order to avoid a publication bias we decided to also include non-published literature (e.g. theses, presentations, unpublished manuscripts) that met the specified criteria. Non-published papers were selected by searching the internet, conference programmes and contacting researchers directly. Selection Criteria

We included all studies that (1) examined children/adolescents 18 years and younger, (2) operationalized relational or indirect aggression (results regarding perpetrators) consistent with our definitions of these constructs, (3) reported separate information on scores for boys and girls, (4) included information to estimate the children’s age as accurate as possible, (5) provided measures that would allow the calculation of an effect size. We excluded (1) intervention studies, (2) twin or adoption studies, (3) studies that examined aggression in a clinical or forensic context (e.g. studies that were conducted with samples not drawn from the general population), (4) studies using qualitative analysis only and (5) studies on other forms of aggression than relational and indirect. We obtained 169 references (with results for perpetrators and/or victims) after a first search, including many not yet published papers, dissertations and congress papers (“grey literature”). 92 of the papers did not fully meet our selection criteria and were excluded. Some of the primary studies constituting the remaining papers defined a cut-offscore of 1, 1.5 or 2 standard deviations above the mean aggression score to identify relationally or indirectly aggressive children and adolescents, and presented the number of boys and the number of girls exceeding this value in relation to the total number of boys and girls only (that is the number of cases exceeding a defined cutoff only). As we consider the extent of gender differences in relational/indirect aggression to differ among the two types of measurement (studies with a cut-off score and studies that report means for all boys and girls), we conducted separate analyses. For the present paper, we focussed on those studies reporting means for all boys and girls and results based on studies with a cut-off score are not presented here. Coding Strategy

A coding manual was developed to extract relevant information from the publications in a consistent and reliable manner. All studies that met our selection criteria were coded by a team of coders. To test for reliability, a random sample of 20% of obtained primary studies were coded twice in an interval of two weeks. Ambiguities in codings were corrected before starting the full coding procedure. Corrections

180

H. Scheithauer et al.

concerned primarily ratings that had to be inferred from the information given in the primary studies and could not be coded directly from the paper (high-inference-codings; Hall, Tickle-Degnen, Rosenthal, & Mosteller, 1994). For low-inference-codings that were retrieved directly from the studies reliability was high (e.g., κ>.90). As assumed, consistency of the high-inference-codings was low to moderate in the beginning of the coding procedure but could be increased to κ=.90 after repeated practice. Based on these respectable results, the coding process was completed and a final test for reliability with a subsample of 15% of all studies still showed a very high consistency. Effect Size Calculation

Hedges’ d was calculated for all studies included in this analysis. This index represents the standardized difference of two means, in our case girls’ and boys’ mean relational/indirect aggression. Effect size calculation was done in several steps: For means and standard deviations, t values, or F ratios (e.g., continuous data) effect size g, the standardized group mean difference was calculated, following procedures described in Hedges and Olkin (1985; see Hedges & Becker, 1986):

� �

��



��

� ������





(1),

with standard deviation

� ������ �

�� � � �� * � � �� � � ��� � �� * � � �� � �� � � �� � � � (2).

Because g tends to overestimate the true value for small samples, a simple correction gives an unbiased estimator (Hedges‘ di): �



�� � � � �� � � � * � � � � * � (3), where the values of cn are given to a very good approximation by

�� � � �

� �� � � � � � � � � �

(4),

Gender Differences in Relational Aggression

181

with variance

��

� �� �� � � � � � � * � � � �� � � � � �

(5).

In contrast to other measures of the standardized difference, Hedges’ d additionally includes a correction term for small sample sizes to adjust for the heightened standard error. A positive value of Hedges’ d indicates a higher prevalence of relational/indirect aggression among girls than among boys in the sample. Hedges’ g is normally distributed in adequately sized samples. Dependency of Effect sizes

Following the recommendation of Lipsey and Wilson (2001), we aimed to include only one effect size per set of dependent effect sizes. We defined effect sizes to be dependent when they represent several aggression measures for one sample (or two or more overlapping samples). There were two sources of dependencies: First, many studies examined one group of children by using different methods (e.g. reports of peers, self, teachers and parents for the same group of children), other studies reported longitudinal data. A second form of dependency occurs when the results of different publications were based on the same set of children. This type of dependency was difficult to determine at times as the sample description of the publications often was insufficiently detailed to determine the origin of the sample. If effect sizes were found to be dependent, the following strategies were used to reduce effect sizes to one measure (cf. Lipsey & Wilson, 2001): • exclusion of effect sizes based on samples > 1.500 for they would dominate other results • deletion of effect sizes based on one item only if there is a also a combined score including this item • preference of effect sizes based on larger sample sizes because they estimate the true effect size more precisely • random selection of one effect size if none of the other strategies was applicable. Application of these strategies resulted in a total of n=30 effect sizes from 23 primary studies1 included in the present analysis (see Tab. 1).

¹ Marked with an asterisk (*) in the references section.

182

H. Scheithauer et al.

Table 1. Effect sizes (di), variances of effect sizes (vi) and mean of age range of primary studies. Authors

dia

Relational aggression Bonica, Yershova, Arnold, Fisher, & Zeljo (2003) Campbell (1999) Campbell & Frabutt (1999) Crick (1995) Crick, Casas, & Mosher (1997) David & Kistner (2000) Goldstein, Tisak, & Boxer (2002) Hart, Nelson, Robinson, Olson, & McNeilly-Choque (1998) Henington, Hughes, Cavell, & Thompson (1998) MacDonald, D‘Amico, & O‘Laughlin (2000) McNeilly-Choque, Hart, Robinson, & Nelson (1996) Moretti, Holland, & McKay (2001) Pakaslahti & Keltikangas-Järvinen (1998) Prinstein, Boergers, & Vernberg (2001) Rys & Baer (1997) Rys & Baer (1997) Tiet, Wasserman, Loeber, McReynolds, & Miller (2001) Tiet, Wasserman, Loeber, McReynolds, & Miller (2001) Tiet, Wasserman, Loeber, McReynolds, & Miller (2001) Xie (1998) Indirect aggression Björkqvist, Lagerspetz, & Kaukiainen (1992) Björkqvist, Lagerspetz, & Kaukiainen (1992) Lagerspetz & Björkqvist (1994) Owens (1996) Owens (1996) Owens (1996) Owens (1996) Pulkkinen (1983) Salmivalli, Kaukiainen, & Lagerspetz (2000) Verlaan & Schwartzmann (1998)

vi

Mean age

0.66 0.05 0.05 0.33 0.71 -0.36 -0.39 -0.09 -0.27 -0.18 0.24 0.50 0.10 -0.11 0.19 0.20 -0.16 0.19 0.10 0.31

0.03 0.03 0.03 0.07 0.07 0.005 0.04 0.02 0.005 0.02 0.02 0.05 0.01 0.01 0.03 0.03 0.04 0.06 0.06 0.01

4.25 10.00 11.50 10.00 4.50 9.00 4.13 5.08 8.00 12.00 4.83 14.00 14.00 16.00 8.50 11.50 8.50 13.50 17.00 11.00

0.35 0.60 0.57 -0.08 0.16 0.58 0.95 -0.29 0.85 0.10

0.05 0.03 0.03 0.04 0.04 0.04 0.05 0.01 0.02 0.29

8.00 15.00 11.50 7.50 11.50 14.50 16.50 8.00 15.50 14.50

Note. Numbers are rounded to the third decimal. Mean age refers to the mean of the age range of participants in the respective primary study. aPositive values indicate higher aggression scores for girls.

Fixed or Random Effect Sizes Model

As we do not assume the effect sizes calculated on the basis of the different studies to be functionally equivalent because of differing samples, methods and moderators, a ran-

Gender Differences in Relational Aggression

183

dom effects model was used2 (Borenstein, Hedges, & Rothstein, 2007). Our decision to use this model was also based on our aim to make inferences that go beyond the observed studies and enable us to make generalizable inference (Hedges & Vevea, 1998). Publication Bias

To test for publication bias we calculated rank order correlations of the effect size versus the sample size (Rosenberg, Adams, & Gurevitch, 2000). Kendall’s τ as well as Spearman’s rs were non-significant indicating no publication bias (although these tests have low power considering the given number of effect sizes; Rosenberg, Adams, & Gurevitch, 2000). We additionally regarded a normal quantile plot which compares the standard normal distribution with the distribution of effect sizes. The plot neither showed “strange gaps” nor strong nonlinearity, showing no clear publication bias. Data Screening

Prior to analysis, effect sizes were examined for missing values, outliers, and normality of distribution following the recommendations by Tabachnik and Fidell (2007). There were no missing values and no univariate outliers. According to the Shapiro-WilkTest, the distribution of effect sizes did not deviate significantly from normality. Data Analysis

Data analysis addressed the following three issues: First we tested whether the mean effect size was significantly different from zero. Second, we performed tests for heterogeneity in order to study the variation in effect sizes. A significant test for heterogeneity indicates a variation in effect sizes that is so large that it is improbable that these effect sizes represent one single population effect size (Lipsey & Wilson, 2001). Therefore, for heterogeneous sets of studies a single mean effect size is of limited value. Third, we tested whether excess variability could be explained by associations with the a priori specified moderator variable “mean age”. The moderator test was conducted using a mixed effects model which allows for some portion of variability to be systematic and can be modeled. However, there are other sources of unsystematic between-study variability that cannot be modeled. The latter are captured by the random variance component (Lipsey & Wilson, 2001). All analyses were conducted using the MiMa-function by Viechtbauer (2006), which uses an iterative restricted maximum-likelihood (REML) procedure to estimate the random effects variance component τ2. This estimation procedure has been recom² Strictly speaking, we used a mixed model because we performed moderator analyses based on a

random effects model.

184

H. Scheithauer et al.

mended to be the most appropriate procedure by several authors (Thompson & Sharp, 1999; Viechtbauer, 2005). Results Overall Effect Size

The overall mean effect size for gender differences in relational/indirect aggression was significantly different from zero (d=.18, SD=.07, p=.05) with a random variance component of τ2=.11. However, the divergence in effect sizes was considerable, ranging from dmin=-1.04 to dmax=1.17. The effect size distribution was significantly heterogeneous, Q(29)=194.52, p