Borderline personality disorder (BPD) is a particularly devastating (Bleiberg, 2001) form of psychopathology. This disorder is a complex mental illness typically characterized by numerous deficits in cognitive, emotional, and behavioral functioning including affective instability; stormy, enmeshed relationships; unrealistic fears of abandonment by loved ones; dissociative thoughts; chronic feelings of emptiness; impulsivity; paranoid thoughts; and suicidal ideation or gestures (American Psychiatric Association [APA], 1994). Even though borderline pathology is commonly encountered by clinicians, it has been infrequently considered by researchers and almost never by developmental researchers.

Developmental psychologists' lack of attention to borderline pathology is likely due to the fact that, for many years, researchers, clinicians, and other mental health professionals have viewed personality as lacking in cohesiveness and durability prior to the age of 18. This perspective was (and for some personality disorders still is) supported by the DSM (APA, 1994) criteria for diagnosis of personality disorders. For instance, currently the identification of BPD is strongly discouraged during childhood and adolescence. One unfortunate implication of this approach has been that the study of precursors of personality disorders such as BPD has focused primarily on adults for many years. As a result, relatively little empirical understanding of the etiology and development of these disorders has been achieved (Bleiberg, 2001; Geiger & Crick, 2001; Kernberg, Weiner, & Bardenstein, 2000; Paris, 2003). In fact, much of the existing knowledge has been obtained through retrospective studies of adults (e.g., Links, Boiago, Huxley, Steiner, & Mitton, 1990), an approach with obvious, serious limitations.

The lack of attention to the development of personality disorders, as assessed prospectively in childhood, is surprising given that personality has been posited to begin forming early in life, substrates of which may be present at birth (Hartup & van Lieshout, 1995). Just as personality does not simply “appear” at age 18, it is highly unlikely that the symptoms or patterns of attributes characteristic of a personality disorder just appear at age 18. Thus, it is important to be attentive to individual differences in the way that personality develops as this developmental process may include variations that, for some children and adolescents, represent vulnerabilities for the development of personality disorders. Investigation of these vulnerabilities using prospective designs, psychometrically sound measures, and samples of children and adolescents is sorely needed to build a systematic knowledge base of the etiology and development of personality disorders, and to inform the generation and implementation of empirically supported intervention and prevention efforts.

The debate regarding the appropriateness of discussing borderline pathology in childhood has raged among psychologists and psychiatrists (Paris, 2003). Although some individuals argue that borderline pathology cannot and should not be assessed prior to adulthood, others have proposed that it is important to identify diagnostic criteria for assessment of borderline pathology in children (Bemporad, Smith, Hanson, & Cicchetti, 1982). Children identified as displaying borderline pathology exhibit similar symptoms and risk factors (e.g., histories of abuse, neuropsychological deficits) as adults with BPD (Zelkowitz, Paris, Guzder, & Feldman, 2001a), suggesting that borderline pathology may exist in childhood. However, one challenge to the study of borderline pathology in childhood has been the lack of clarity in the definitions of borderline pathology used by researchers (e.g., Paris, Zelkowitz, Guzder, Joseph, & Feldman, 1999). We believe that this problem may be partially ameliorated by considering possible heterotypic continuity in the manifestation of borderline pathology across development. That is, perhaps symptoms of borderline personality (e.g., impulsivity, overly close relationships) are manifested differently at different development periods. Despite differences in outward behavior or symptoms, however, we propose that the underlying meaning remains the same across development. Attentiveness to issues of heterotypic continuity across development is necessary in the measurements we employ to assess borderline personality features at various developmental stages.

The idea that pathological behavior in childhood may be associated with increased risk of personality disorder in adulthood has previously been acknowledged in the study of antisocial personality disorder. In particular, one of the criteria in the DSM-IV for diagnosis of antisocial personality disorder is a history of childhood conduct disorder (APA, 1994). In a similar vein, we propose that there may be a constellation of symptoms demonstrated in childhood that reflect the development of borderline personality problems. These symptoms, in turn, would increase the likelihood of a future diagnosis of BPD.

In the present study, we propose that although BPD is not clearly defined in childhood, some children do exhibit features characteristic of adult borderline pathology, reflecting the emergence of borderline pathology across development. Moreover, we believe that studying these features and their trajectories throughout development may provide us with a better understanding of the development of BPD. In effect, because personality is relatively malleable in childhood (Geiger & Crick, 2001), we do not believe that it is appropriate to classify children as exhibiting BPD. Nonetheless, we believe that some children do exhibit rather stable cognitive, affective, and behavioral tendencies, termed borderline personality features in the present paper, that resemble borderline pathology in adulthood. Additionally, we believe that certain types of information-processing patterns, relationship orientations, and behavioral characteristics exhibited in childhood may reflect maladaptations in personality development that place children at risk for a diagnosis of a personality disorder in adulthood. Thus, a primary goal of the present study was to examine whether borderline personality features could be reliably and validly measured in childhood.

Indeed, one major hindrance to the study of the etiology and development of borderline personality has been the lack of assessment tools that reliably and validly measure borderline personality pathology in childhood. To date, assessment has been hallmarked by two key features, both of which may be problematic for initial studies of borderline pathology in childhood. The first is that assessment has relied primarily on the categorical, psychiatric diagnostic scheme of BPD (i.e., as defined by the DSM) that was developed for adults (for an exception, see Bemporad et al., 1982). The degree to which this scheme is appropriate for use with children is unclear. For example, this categorical approach focuses narrowly on clinically relevant symptoms, and does not allow for the study of the entire range of borderline symptoms. Assessing a broad range of symptoms allows identification of not only those individuals who are demonstrating “clinically significant” levels of symptomatology, but also those who may be considered “at risk.” Indeed, the identification of at-risk individuals may be particularly important when studying the etiology of a personality disorder among an age group in which personality is believed to still be forming. When studying issues of development, the current categorical focus seriously limits our ability to identify and study all possible trajectories toward or away from psychopathology over time. For instance, biological, emotional, cognitive, and behavioral indexes of psychopathology may wax and wane with maturation. Thus, longitudinal studies including dimensional assessments that take into account heterotypic continuity in the manifestation of psychopathology are necessary. Moreover, subclinical levels of psychopathology or even characteristics that do not appear to be psychopathological at one point in development may portend future clinical levels of psychopathology.

A second key feature of past childhood studies is that clinical samples have been the primary focus. This issue is particularly salient when considering issues of development and etiology because clinical samples are unlikely to be representative of childhood populations exhibiting high rates of borderline pathology. For example, clinical samples are more likely to suffer from comorbid disorders, to experience difficulties that are relatively extreme in nature, and to have been the recipients of past treatment (Trull, 1995). Additionally, reliance on clinical samples is likely to introduce gender bias into our understanding of borderline personality among children and adolescents (e.g., Guzder, Paris, Zelkowitz, & Marchessault, 1996) because boys are far more likely than girls to appear in treatment facilities during childhood. For example, in one study of children recruited through day treatment centers (see Paris, 2003), girls comprised less than 20% of the sample. The composition of this sample and others like it contrasts sharply with adult studies of gender differences that demonstrate that BPD is more prevalent among females than males (e.g., Block, Westen, Ludolph, Widon, & Jackson, 1991). In fact, some research has shown that 70–75% of individuals diagnosed with BPD are female (Gunderson, Zanarini, & Kiesel, 1991; Swartz et al., 1990). Thus, existing studies of borderline personality pathology during childhood may be most relevant for the gender that is least likely to suffer from the disorder (i.e., boys). Biases such as these have led some researchers to conclude that “mental health care professionals are missing troubled girls” (Paris, 2003, p. 40; see also Crick & Zahn–Waxler, 2003).

To address these issues, the first goal of the present research was to develop a reliable and valid measure of borderline personality features for use with children that (a) allows for dimensional assessment of borderline pathology so that a wide range of severity levels can be captured and (b) is appropriate for use with normative (i.e., nonclinical) samples of children and adolescents. Toward this goal, we modified a widely used adult measure of borderline pathology, the BPD Scale (BOR) of the Personality Assessment Inventory (PAI; Morey, 1991) to create a dimensional measure of borderline personality features appropriate for use with children. The original PAI has demonstrated reliability and validity for use with subjects aged 18 years and older, and has been shown to be particularly useful for the assessment of psychopathology (including borderline personality) in nonclinical populations (e.g., Kurtz, Morey, & Tomarken, 1993; Morey, 1991). The instrument developed for use in the present study, the Borderline Personality Features Scale for Children (BPFS-C), was designed for use with children aged 9 years and older.

In the absence of other established, dimensional measures of borderline pathology appropriate for use with children that could be used to establish the validity of the newly developed BPFS-C, we relied on developmental psychopathology theory to guide our assessment of the construct validity of this instrument. This theory has been instrumental in guiding research on psychopathology during childhood and adolescence (Cicchetti & Cohen, 1995; Sroufe, 1997). The primary tenet of the developmental psychopathology approach is that deviant development can only be understood within the context of normative development. That is, at each phase of development, children utilize internal and external resources to master developmentally appropriate tasks (Price & Lento, 2001; e.g., learning to regulate emotion, developing successful and satisfying relationships with peers and friends). Multiple domains (e.g., social, emotional, biological) interact and influence one another over time as children negotiate these developmental tasks, a process that contributes to adaptive or maladaptive patterns of behavior. Children who successfully negotiate these challenges (adaptive pattern of functioning) are likely to develop in a normative manner, whereas those who do not (e.g., young children who experience difficulties developing secure attachments to caregivers or adolescents who fail to build appropriately intimate friendships) may be at risk for compromised adaptation or maladaptive patterns of functioning and the development of psychopathology (Cicchetti & Cohen, 1995; Sroufe & Rutter, 1984).

Based on the developmental psychopathology model, childhood features associated with atypical personality development may be best identified through consideration of difficulties in negotiating developmentally appropriate, normative tasks that have relevance for the particular disorder of interest. Moreover, there may be heterotypic continuity across development in the manifestation of various forms of psychopathology, which is influenced by one's developmental stage and the salient issues of that time. Consistent with the developmental psychopathology approach (e.g., Sroufe & Rutter, 1984), we recently conducted a content analysis of current DSM criteria for diagnosis of personality disorders (Geiger & Crick, 2001), which was guided by existing knowledge of normative development and of ways in which failure to successfully negotiate developmentally salient tasks in childhood might result in pathways toward personality pathology. In this theoretical model, five childhood indicators of BPD were identified: hostile, paranoid world view; intense unstable, inappropriate emotion; overly close relationships; impulsivity; and lack of sense of self. Each childhood indicator was theorized to capture a unique facet of BPD and to reflect difficulty in negotiating important developmental tasks in childhood.

With respect to the first childhood indicator, research and theory has shown that adults who exhibit borderline pathology tend to possess a hostile, paranoid world view that leads them to be preoccupied with negative information in the environment and to be suspicious of the intent of others (e.g., Kernberg, 1967; Morey, 1988a, 1988b). This characteristic may represent a failure to accomplish a key developmental task of childhood, which is to acquire the ability to recognize that others may have different intent and perspectives than the self, and to correctly identify the nature of those perspectives (Selman, 1980). Social information processing research has shown that children who fail to develop this ability at appropriate ages and who tend to incorrectly infer hostility in others (i.e., who exhibit hostile attributional biases) are at risk for serious concurrent and future adjustment problems such as aggressive behavior patterns (for a review, see Crick & Dodge, 1994). According to the Geiger and Crick (2001) model, this cognitive sensitivity may represent childhood vulnerability for borderline personality pathology.

The second proposed indicator of borderline pathology is intense, unstable, inappropriate emotion, particularly anger (APA, 1994, p. 654). Individuals with borderline pathology typically experience emotions intensely and to a degree that is often considered extreme relative to the situation with which they are faced (Morey, 1991). Their moods are often unpredictable, and may fluctuate widely within the space of a few hours (e.g., from anger to depression to anxiety). This pattern represents emotion regulation difficulties that may reflect a deviation from typical developmental trajectories. According to developmental theory, acquisition of emotion regulation skills typically begins in infancy (Mangelsdorf, Shapiro, & Marzolf, 1995) and continues throughout childhood and adolescence (Denham, 1998). The failure to master these abilities likely represents significant pathology that has relevance for borderline personality (Geiger & Crick, 2001). Among elementary school children, the expression of intense, inappropriate emotion may be assessed by examining children's emotional sensitivity to potentially benign situations (e.g., by evaluating whether they respond with anger to a provocation situation in which the intent of the other person is unclear; Geiger & Crick, 2001). Emotional sensitivity to an ambiguous social environment represents a failure to master emotion regulation skills, and may in turn, place children at risk for the expression of borderline features. Thus, emotional sensitivity may be an age-appropriate indicator of borderline features in childhood.

The third proposed indicator of borderline features in childhood is overly close relationships. The relationships of individuals exhibiting BPD have often been described as intense, stormy, dependent, and enmeshed (Block et al., 1991; Morey, 1991). Consistent with this pattern, borderline pathology has been shown to be significantly associated with a preoccupation with relationship concerns and heightened emotionality about relationships (e.g., fears of rejection or abandonment by loved ones; heightened emotional sensitivity to perceived relational difficulties or slights). These problems may reflect prior difficulty acquiring appropriate levels of autonomy and trust within close relationships, a key developmental process that begins in early childhood in the context of the parent–child relationship (Erikson, 1963, 1982; for a review, see Bartholomew, Kwong, & Hart, 2001). For example, young children who develop secure attachments with their caregivers (an optimal developmental outcome) learn to trust that their caregivers are responsive, predictable, and loving and this view of relationships is carried over into those established outside the family (e.g., with friends, teachers, romantic partners; Shaffer, 2000). In contrast, children who develop insecure attachments (e.g., those who develop resistant attachments) often have parents who are inconsistent in their parenting (Ainsworth, 1979; Isabella & Belsky, 1991). Faced with this situation, children commonly respond with desperate and continued attempts to gain the parents' affection, attention, and support, strategies that are followed by anger and resentment when the desired parental response is not forthcoming. This set of experiences often results in a view of relationships that is characterized by mistrust, heightened sensitivity to relationship difficulties, and fears and doubts about being truly cared for and supported. This pattern may represent a childhood precursor to the relationship problems that have been shown to be associated with BPD (Geiger & Crick, 2001). Among elementary school children, establishing and maintaining friendships is an important developmental task (Hartup, 1992); thus, friendships may provide a particularly salient context for overly enmeshed relationships in middle childhood. In particular, children who develop overly exclusive friendships with others may be at risk for the development of borderline features, and friendship exclusivity may be an age-appropriate indicator of borderline features in childhood.

The fourth indicator of borderline features is impulsivity, a key characteristic of borderline personality (Kernberg et al., 2000; Paris, 2003); one that typically manifests in two ways: the inability to inhibit excesses in behavior (e.g., lack of self restraint) and aggressive outbursts toward self and others (Geiger & Crick, 2001). These tendencies may have their roots in individual differences in the attainment of self-control (i.e., the ability to control impulses and actions that are inappropriate or that interfere with a goal), which is an important feature of normative development (Bandura, 1986; Kopp, 1987; Mischel, 1986). Early in life, children are unable to self-regulate, and must rely on external resources (e.g., parents) to help them control their behavior. As children mature, they gradually develop the capacity to internalize self-control strategies and societal norms, a skill that allows them to regulate their impulses independently. Children who fail to master these skills are at risk for the development of impulse-related difficulties such as aggressive and risk-taking behaviors (Coie & Dodge, 1998). This pattern may represent a developmental precursor to the impulsive characteristics associated with borderline pathology.

As previously stated, one of the key features of impulsivity is engagement in aggressive behavior (Geiger & Crick, 2001; Schmidt, 2003). Studies in which the association between aggression and borderline pathology has been examined have focused almost exclusively on forms of aggression that are most typical of males (i.e., physical aggression; e.g., McManus, Alessi, Grapentine, & Brickman, 1984). However, recent studies have identified a relational form of aggression that is more salient for females than the physical forms of aggression that have captured the majority of previous empirical attention (Crick et al., 1999). In contrast to physical aggression, in which physical damage is the agent of harm, relational aggression involves behaviors in which damage to relationships serves as the vehicle of harm. Given that BPD is more prevalent among females than males (i.e., at least among adults; Gunderson et al., 1991), assessment of indicators of impulsivity that are most typical of females, such as relational aggression, is particularly important for increasing our understanding of borderline pathology. Consistent with this idea, initial evidence indicates that, for adults, relational aggression is significantly associated with borderline personality features (Werner & Crick, 1999). Consequently, in the present study, both physical and relational forms of aggression were assessed as indicators of impulsivity.

In the present study, the association between these four developmentally appropriate indicators of borderline pathology (i.e., cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, and aggression) and children's performance on the newly developed BPFS-C was examined to demonstrate the construct validity of the BPFS-C.¹

After establishing the favorable psychometric properties of the BPFS-C, we had several additional objectives for this research. Our second goal was to use the BPFS-C to evaluate the stability of borderline personality features in childhood. Given that numerous aspects of personality are likely to change and evolve to some degree during the childhood years, we hypothesized that individual differences in borderline personality features would be moderately, rather than highly, stable across the course of a year.

The third objective of this investigation was to evaluate gender differences in borderline personality features. Little information is available regarding the role of gender in borderline pathology during childhood. Further, the knowledge that does exist has been biased by the focus on clinical samples. In these studies, a larger percentage of boys than girls have been identified as exhibiting borderline pathology (Paris, 2003). However, as discussed previously, this pattern may be due to the fact that clinical samples include substantially greater numbers of boys than girls and/or that clinical samples differ from normative populations on a number of dimensions (e.g., severity of symptoms). In the present study of a nonclinical sample that included approximately equal numbers of boys and girls, we hypothesized that, if gender differences were apparent, they would follow the pattern shown to be typical in adult samples (i.e., girls would report higher levels of borderline personality features than boys at the first assessment period and/or girls would exhibit greater growth in borderline features over time than boys; Block et al., 1991).

Our fourth goal was to provide the first examination of the utility of the BPFS-C for distinguishing borderline personality from other forms of psychopathology (i.e., the specificity of the measure). Past studies have demonstrated that borderline pathology is often comorbid with depressive symptoms and disorders (Gunderson & Elliott, 1985; McGlashan, 1983). As a first step in evaluating the specificity of the BPFS-C for measuring borderline personality features, we tested the degree to which the BPFS-C provided unique information, relative to a widely used measure of depressive symptomatology (i.e., the Children's Depression Inventory [CDI]; Kovacs, 1985), in the concurrent and future prediction of four of the five childhood indicators of borderline pathology identified by Geiger and Crick (2001).

To address these four objectives, we recruited a large sample of fourth, fifth, and sixth graders (n = 400; 215 girls) through their elementary schools. Measures of borderline personality features (BPFS-C), depressive symptoms (CDI), and the four indicators of borderline pathology (i.e., hostile world view; inappropriate emotion; overly close, enmeshed relationships; and impulsivity) were group administered in children's classrooms during the Fall of Year 1, the Spring of Year 1, and the Fall of Year 2.

Results

Analyses were conducted to address the four main objectives of the present study: (a) to examine the construct validity of the BPFC-S by investigating whether scores on the BPFC-S tracked over time with the theoretically related, developmentally appropriate indicators of borderline features; (b) to examine the stability of borderline features over time; (c) to explore gender differences in borderline features in childhood; and (d) to investigate the specificity of the BPFS-C for measuring borderline features in childhood.

Construct validation of the BPFS-C

To address the first objective of the present study, LMM (with SAS Proc Mixed 8.2) was used to examine whether scores on the BPFS-C were dynamically associated with theoretically related, age-appropriate indicators of borderline features (i.e., cognitive sensitivity, emotional sensitivity, friend exclusivity, and aggression). LMM was chosen for two reasons. First, unlike traditional methods of repeated measures analysis (e.g., repeated measures analysis of variance), LMM can accommodate missing data (see Long & Pellegrini, in press). Second, LMM can be used to explore the dynamic association among variables over time (e.g., can explore whether changes in a set of variables “track” with changes in an outcome variable over time; see Long & Pellegrini, in press). LMMs can be expressed as multilevel models (i.e., as a set of equations) to provide an interpretive framework for the parameters. Alternatively, LMMs can be conceptualized as a single equation including only Level 2 parameters. In the present study, an LMM approach was chosen as the framework for all analyses. Descriptive information regarding children's scores on each measure (i.e., means and standard deviations) at the three assessment periods are presented in Table 1.

Sample means and standard deviations across time

To address the first objective of the present study, participants' cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, relational aggression, and physical aggression served as dynamic predictors of borderline personality features over time. Testing parameters requires the specification of an appropriate variance–covariance structure (Diggle, 1988). Model comparisons among covariance structures (e.g., autocorrelation, unstructured, random effects plus measurement error) using the most complex model tested in the present paper (Equation 5) revealed that the unstructured covariance structure yielded a superior fit with the data.³

In the analysis including all five predictors of borderline features and depressive symptoms (Equation 5), model fit comparisons indicated that the unstructured covariance structure was a better fit than the autocorrelation covariance structure. The model with the random effects plus measurement error covariance structure did not converge. Thus, the unstructured covariance structure was adopted.

Thus, the unstructured covariance structure was adopted in all analyses presented in this paper.

In the first set of analyses, five separate linear mixed models were run to examine whether cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, relational aggression, and physical aggression, respectively, each significantly tracked with borderline features over time. For each model, the equation used to estimate the fixed effects was

where μ_j is the mean level of borderline scores at time j, μ_predictorj represents the mean score of the predictor at time j (predictor = cognitive sensitivity in Model 1, emotional sensitivity in Model 2, exclusivity with a best friend in Model 3, relational aggression in Model 4, and physical aggression in model 5), and γ₀ indicates the strength of the dynamic association between borderline personality features and the predictor. As expected, the results revealed that each predictor significantly and positively tracked with borderline personality features (see Table 2). In other words, increases in each predictor (i.e., cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, relational aggression, and physical aggression) were associated with time-dependent increases in borderline personality features.

Longitudinal association between indicators of borderline features and scores on the BPFS-C

In addition to examining whether each theoretically related predictor significantly tracked with borderline features over time, the present study also explored the unique longitudinal association between each predictor and borderline features. Indeed, given that the content analysis of borderline pathology conducted by Geiger and Crick (2001) indicated that impulsivity, hostile, paranoid world view, overly close relationships, and intense, unstable, inappropriate emotion were each important, unique facets of borderline features, we expected that each predictor would be uniquely associated with borderline features over time, even when controlling for the other four predictors. Thus, an additional linear mixed model with all five predictors entered simultaneously was conducted so that the unique longitudinal association between each predictor and borderline features could be assessed. The equation used to estimate the fixed effects was

where μ_j is the mean level of borderline scores at time j; μ_{cog j}, μ_emoj, μ_excj, μ_{ragg j}, and μ_{pagg j} represent the mean cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, relational aggression, and physical aggression at time j, respectively; and γ₀, γ₁, γ₂, γ₃, and γ₄ indicate the strength of the dynamic association between borderline personality features and cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, relational aggression, and physical aggression, respectively. In other words, this analysis allowed us to examine whether borderline personality features uniquely tracked with the five predictors over time (i.e., whether changes in the predictors were associated with similar time-dependent changes in borderline features, when controlling for the longitudinal association between borderline features and the other four predictors).

We expected that borderline personality features, cognitive sensitivity, emotional sensitivity, friend exclusivity, relational aggression, and physical aggression would be related such that increases in children's borderline features over time would be uniquely associated with increases in their scores on each of the five predictors. The results of this analysis indicated that change in children's borderline personality scores were positively associated with change in their cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, and relational aggression. However, change in borderline features was not associated with change in physical aggression over time (see Table 3). That is, as predicted, changes in cognitive sensitivity, emotional sensitivity, friend exclusivity, and relational aggression were each uniquely associated with changes in borderline personality features over time. In effect, each of these four predictors “tracked” with borderline features across time, even when controlling for the other predictors. In contrast, physical aggression did not uniquely track with borderline features, above and beyond the longitudinal association between borderline and the other four predictors.

Unique longitudinal association between indicators of borderline features and scores on the BPFS-C

Stability of borderline features over time

To assess the stability of borderline symptoms over time, correlations between scores on the BPFS-C at Time 1, Time 2, and Time 3 were conducted. As predicted, the results indicated that children's borderline features were moderately stable over the course of the study (see Table 4).

Stability of borderline personality features across three assessments (Fall, Winter, Spring)

Gender differences in scores on the BPFS-C in childhood

To examine gender differences in borderline personality features in childhood, a LMM was conducted exploring children's trajectories of borderline features over time. In addition, gender differences in these trajectories were assessed. In the present analysis, the equation used to estimate mean change in borderline features over time, and to explore whether this change was conditional on gender was

where μ_j is the mean borderline personality features as assessed with the BPFS-C at time j; l_j represents the linear term at time j (l_j = 0, 1, and 2 at Time 1, Time 2, and Time 3, respectively); g is the participant's gender (0 = male, 1 = female); γ₀ and γ₂ represent the intercept (i.e., mean borderline features at Time 1) and slope (i.e., linear change in borderline features over the course of the study) for boys, respectively; and γ₁ and γ₃ represent the interaction of gender with the intercept and the linear term, respectively. This analysis thus permitted the investigation of whether males and females differed in their borderline features at the first assessment (i.e., intercept difference), and whether they exhibited different trajectories of borderline features over the course of the study (i.e., slope difference).

Based on findings that, among adults, women report higher levels of borderline features than men (e.g., Block et al., 1991), we expected that girls would exhibit greater mean levels of borderline features at the first assessment and/or girls would display greater growth in borderline features over the course of the study. The results revealed that the Gender × Intercept interaction was statistically significant, suggesting that girls were exhibiting greater levels of borderline features at Time 1 (see Table 5). In addition, the interaction between gender and linear growth approached statistical significance (Table 5). It is surprising that, whereas boys did not exhibit changes in borderline features over time, girls displayed a decrease in their symptoms across the course of the study. However, it is important to note that this interaction did not reach conventional levels of statistical significance.

Trajectories of borderline personality features conditional on gender

Specificity of the BPFS-C

The fourth and final goal of the present investigation was to examine whether borderline personality features, as assessed with the BPFS-C, were associated with the proposed indicators of borderline personality features in particular, and not psychopathology in general. To address this objective, analyses were run to explore whether each of the proposed five indicators longitudinally tracked with borderline features, controlling for children's depressive symptoms. Depressive symptoms were chosen based on their high degree of comorbidity with borderline personality features (Gunderson & Elliott, 1985; McGlashan, 1983) and the high degree of overlap between borderline features and depressive symptoms observed in the present sample (correlations ranged from .52 to .58 across the three assessment periods). In the first set of analyses, five separate linear mixed models were conducted to explore whether cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, relational aggression, and physical aggression, respectively, each significantly tracked with borderline features over time, controlling for the longitudinal association between borderline features and depressive symptoms. For each model, the equation used to estimate the fixed effects was

where μ_j is the mean level of borderline scores at time j; μ_cdij represents the mean depressive symptoms at time j; μ_predictorj is the mean score of the predictor at time j (predictor = cognitive sensitivity in Model 1, emotional sensitivity in Model 2, exclusivity with a best friend in Model 3, relational aggression in Model 4, and physical aggression in Model 5); and γ₀ and γ₁ indicate the strength of the dynamic association between borderline personality features and depressive symptoms and each predictor, respectively. The results presented in Table 6 indicate that depressive symptoms and borderline personality features significantly tracked together over time. In addition, as expected, each predictor significantly and positively tracked with borderline personality features, even when controlling for the longitudinal association between borderline features and depressive symptoms (see Table 6). In other words, increases in each predictor were associated with time-dependent increases in borderline personality features, even when controlling for depressive symptoms over time.

Specificity of the association between indicators of borderline features and scores on the BPFS-C

The results presented in Table 6 suggest that each indicator of borderline features is associated with borderline features in particular, and not psychopathology in general. In other words, each indicator tracked significantly with borderline features over time, even when controlling for children's depressive symptoms. However, given that each predictor is theorized to be an important and unique facet of borderline pathology, a second important question concerns the specificity of each predictor to borderline features, controlling for the longitudinal association between the other four predictors and depressive symptoms. To accomplish this goal, an LMM, with all five predictors and depressive symptoms entered simultaneously, was conducted using the following equation to estimate the fixed effects

where μ_j is the mean level of borderline scores at time j; μ_cdij, μ_{cog j}, μ_emoj, μ_excj, μ_{ragg j}, and μ_paggj represent the mean depressive symptoms, cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, relational aggression, and physical aggression at time j, respectively; and γ₀, γ₁, γ₂, γ₃, γ₄, and γ₅ indicate the strength of the longitudinal association between borderline personality features and depressive symptoms, cognitive sensitivity, emotional sensitivity, exclusivity with a best friend, relational aggression, and physical aggression, respectively. Thus, this analysis allowed us to examine which predictors were uniquely associated with borderline features over time, controlling for the longitudinal association between borderline features, depressive symptoms, and the other four predictors.

The results of this analysis (Table 7) indicated that cognitive sensitivity, friend exclusivity, and relational aggression tracked with borderline symptoms, even when controlling for depressive symptoms and the other four predictors (although note that the significance of cognitive sensitivity was p < .07). In other words, three of the predictors found to exhibit specificity regarding borderline features in Table 6 remained significant longitudinal predictors of borderline features, even when controlling for depressive symptoms and the other four predictors. Overall, these results bolster the conclusion that children's scores on the BPFS-C are specific to borderline pathology in particular (as assessed via the four childhood indicators), and not psychopathology in general (i.e., as indexed by depressive symptoms).

Unique longitudinal association between indicators of borderline features and scores on the BPFS-C, controlling for depressive symptoms