Mental disorders have a marked negative impact on our society due to both substantial health and socio-economic costs (Trautmann et al., 2016). A better understanding of the determinants of the most prevalent mental disorders in adolescence, such as personality, could have important implications for developing prevention and treatment intervention programs. Various personality trait structure models exist, but the Five-Factor Model (FFM) of personality (Costa & McCrae, 2010) offers a useful descriptive taxonomy according to many personality psychologists (John et al., 2008). When exploring the associations between the FFM and single mental disorders or scales of symptoms, various meta-analyses reveal that neuroticism is the most closely related trait to anxiety and mood disorders (Jeronimus et al., 2016; Kotov et al., 2010), while low agreeableness and low conscientiousness are associated mainly with drug use, behavioral and oppositional defiant disorders (Herzhoff et al., 2017; Kotov et al., 2010; Malouff et al., 2007; Ruiz et al., 2008).

There are also a few studies that explore the associations between personality traits and a correlated model of psychopathology in which an internalizing factor, composed mainly of anxiety and depression symptoms; and an externalizing factor, composed mainly of substance use and antisocial behavior, are specified and correlated with each other. These studies showed that neuroticism is mainly related to the internalizing factor, while low agreeableness and low conscientiousness are related to the externalizing factor (Carragher et al., 2015; Cosgrove et al., 2011; Hengartner, 2018; Krueger & Markon, 2006; Mezquita et al., 2015; Watts et al., 2019). Recently, a bi-factor model of psychopathology has also been tested, in which in addition to the internalizing and externalizing factors or even a psychotic factor, a general factor of psychopathology (or p factor) arose (Caspi et al., 2014, Lahey et al., 2012; Laceulle et al., 2014; Gomez et al., 2019; Murray et al., 2016). While the correlated model of psychopathology explains the high comorbidity among the commonest mental disorders within each spectrum, the bi-factor model emphasizes the general predisposition to psychopathology. Studies that relate the p factor with personality showed that the p factor linked mostly with high neuroticism, followed by low conscientiousness and low agreeableness in adolescents (Castellanos-Ryan et al., 2016; Etkin et al., 2020) and adults (Caspi et al., 2014).

Considering the close association between personality and psychopathology, and in order to better understand the meaning of the p factor, previous studies have also explored whether there is overlap between the p factor and the General Factor of Personality (GFP) (Etkin et al., 2020; Oltmanns et al., 2018; Linden et al., 2017). Some authors have suggested that the overlap between both factors may represent the extent of impairment or dysfunction associated with a certain personality configuration and the presence of psychopathological symptoms (Oltmanns et al., 2018). In the adult population, a correlation of 0.72 (Oltmanns et al., 2018) among both factors has been found, while in adolescents previous studies found beta indices from 0.42 to 0.47 (p < 0.001) (Etkin et al., 2020).

Although all of these studies evidence that personality and psychopathology are associated with each other, they do not clarify the mechanisms by which they are related. To explain the functional relationship between both variables, four major models have been proposed (for reviews, see Fruyt et al., 2017a, b; South et al., 2010). The first model, predisposition/vulnerability, proposes that pre-existing personality traits predispose people to develop certain mental illnesses; for instance, high neuroticism may lead to the development of depressive disorders in children (Nigg, 2006; Tackett, 2006) and adults (Hengartner et al., 2016). The second model, complication/scar, suggests that experiencing a certain form of psychopathology causes some changes in personality. For instance, childhood antisocial behavior problems predict increased neuroticism in adulthood (Shiner et al., 2002). The third model, pathoplasty/exacerbation, indicates that premorbid personality is considered to have an effect on the expression, course, and severity of disorders, and also on treatment response, but they might have independent causes. For example, one study indicated that the changes in childhood personality and over-reactive parenting were associated with adjustment problems later on in adolescence (van den Akker et al., 2010). Finally, the fourth model, continuity/spectrum, suggests that personality traits and psychopathology are both part of one continuous latent dimension (Durbin, 2019). For instance, personality disorders are understood as extreme versions of general personality traits (Samuel et al., 2010). Regarding this fourth model, it is possible to theoretically differentiate between a general continuity model that claims a trait and disorder exist on a continuum from normal traits to psychopathology, and a spectrum model that also assumes common causes and a variability mechanism on such a continuum (De Fruyt et al., 2017b).

Different studies have investigated all four models, but very few have explored them simultaneously (De Bolle et al., 2012; Hengartner, 2018; Klimstra et al., 2010), which hinders broader conclusions (De Fruyt et al., 2017a). This is partly due to the difficulty to perform these kinds of studies, which required multiple assessments across time and to assess both variables, personality and psychopathology, in each wave of assessment. In one of these previous studies, Klimstra et al. (2010) addressed the longitudinal relations between the FFM and two problem behaviors, depression and aggression, performing cross-lagged panel models. They found that effects between personality and problem behavior were bidirectional. These effects were interpreted as vulnerability and complication effects. Specifically, neuroticism, extraversion and conscientiousness predicted depression, while agreeableness, conscientiousness and openness predicted aggression. Conversely, neuroticism, extraversion, agreeableness and conscientiousness were predicted by depression, while aggression predicted neuroticism and agreeableness. They studied cross-sectional correlations only at T1, and observed significant associations between low emotional stability, low extraversion and low agreeableness with problem behavior. These results have been interpreted in subsequent studies as evidence of the continuity model (De Bolle et al., 2012; Bolle et al., 2016).

As far as we know, only De Bolle et al. (2012) and De Bolle et al. (2016) have simultaneously studied the above-mentioned etiological models on the associations between personality and the correlated model of psychopathology (i.e., in which two factors of internalizing and externalizing symptoms without a p factor are specified) with a longitudinal-prospective design in children. The authors found evidence for the continuity model to explain the relations between psychopathology and personality when considering both the Five-Factor Model traits (De Bolle et al., 2012) and maladaptive traits (De Bolle et al., 2016). These associations were more robust for conceptually closer personality traits and psychopathology symptoms, such as the neuroticism/introversion–internalizing problems relation and the low agreeableness–externalizing problems association. Specific complication/scar effects were found from internalizing to neuroticism and conscientiousness, and from externalizing to extraversion and agreeableness traits. Pathoplasty effects were observed for agreeableness on internalizing and externalizing, and for extraversion on externalizing (De Bolle et al., 2012).

Despite these advances, longitudinal studies about the etiological models of personality-psychopathology associations are scarce, especially in adolescents (Durbin, 2019). Moreover, no previous studies with adolescents have longitudinally explored these associations between personality and the bi-factor model of psychopathology, nor between the GFP and the p factor.

The Present Study

The current work aims to empirically study the associations between the FFM of personality and the bifactor model of psychopathology, in which an internalizing, externalizing, hyperactivity and attention problems, and the general (p) factor are specified (see Etkin et al., 2020 Model 5 for a similar specification of the structural model) in a 3-year longitudinal design with Spanish adolescents. In order to achieve this aim, cross-lagged panel models would be performed. Additionally, the association between the p factor and the GFP would be longitudinally explored using the same methodology. The hypothesized associations were between neuroticism and all the symptoms, mainly the internalizing factor (De Bolle et al., 2012, 2016; Etkin et al., 2020) and the p factor (Brandes et al., 2019), whereas agreeableness and conscientiousness would be negatively associated with externalizing symptoms (De Bolle et al., 2012, 2016; Etkin et al., 2020; Klimstra et al., 2010) both within and across waves of measurement. We also expected to find a negative association between conscientiousness and hyperactivity/attention problems (Etkin et al, 2020), and we predicted to find negative associations with internalizing problems for extraversion (Caspi et al., 2014; Etkin et al., 2020). The trait openness would not be included in the analyses as most previous research suggests no significant associations with psychopathology (Kotov et al., 2010). For the trait-symptom association models, we expected to find general evidence for continuity, and to a lesser extent of complication and pathoplasty effects between the FFM and the different factors of psychopathology and between the GFP and the p factor (De Bolle et al., 2012, 2016). The links between previously assessed personality traits and subsequent symptoms have been considered suggestive of the vulnerability hypothesis in previous studies (Klimstra et al., 2010). However, as the effects in the present work are not restricted to participants without a history of mental disorders as in other studies (Laceulle et al., 2014), such a model cannot stringently be confirmed and it is, therefore, safer to attribute these effects to the pathoplasty model (Hengartner, 2018). Hence, vulnerability was not included in our hypotheses. This study is the first to explore the association hypotheses between the FFM and a bi-factor model of psychopathology, and between the GFP and the p factor in adolescents.

Method

Sample

High school students were assessed at three time points once a year. For the first wave (T1), the sample consisted of 831 Spanish adolescents, all aged between 12–18 years (M = 14.35, SD = 1.58; 50.6% girls); n = 619 for the second wave 1 year later (T2, 50.8% girls, mean age of 14.74 years; SD = 1.22); finally, n = 465 for the third wave 1 year later (T3, 49.9% girls, mean age 15.22; SD = 1.00). Of this total sample, analyses were performed on those participants that completed at least two of the three assessment time points for personality or psychopathological symptoms: n = 655; (M = 13.79, SD = 1.24; 49.8% girls). The age distribution in the final group was as follows: 43.3% between 12–13 years old, 55% between 14–16 years old and 1.7% between 17–18 years old. Although the sample was heterogeneous in nationality terms, most participants were born in Spain (82.3%).

Procedure

Participation was voluntary, during class hours, and after receiving informed consent from the school and parents or guardians. This study was previously approved by the Deontological Committee of the authors' university. Questionnaires were filled in on paper and safeguarding of personal data confidentiality was ensured. For the follow-up after the initial assessment, we continued assessing all the students available in their classroom on personality traits and psychopathological symptoms 1 year later, and 1 year after that. A numerical code was assigned to each participant and the correspondence to their identity was only accessible to the researcher in charge.

Measures

Psychopathological Symptoms

The Assessment System for Children and Adolescents (SENA; Fernández-Pinto et al., 2015). The scales included for this study were depression, anxiety, social anxiety, posttraumatic symptoms and somatic complaints (comprising an internalizing factor), aggression, antisocial behavior and defiant behavior (the externalizing factor) and hyperactivity and attention problems (comprising the Hyperactivity-attention problems factor). Participants answered a 5-point Likert-type scale and the score of each scale was obtained by summing all the corresponding items.

Personality Traits

The JS-NEO-A60 (Ortet-Walker et al., 2020) was used to assess neuroticism, extraversion, agreeableness and conscientiousness. The trait openness was not included in the analyses given evidence indicating its lack of association with psychopathology (Kotov et al., 2010; Levin-Aspenson et al., 2019). The inventory comprises 60 items that are answered on a 5-point Likert-type scale. The score of each scale was obtained by summing all the corresponding items.

Analyses

Using SPSS 24, descriptive analyses were conducted with the final group of participants who completed at least two of the three measure time points. Cronbach's alphas were performed to evaluate the internal consistency of the scales and Cohen’s d was used to compare the mean scores of personality traits and psychopathological domains among boys and girls, using the following online calculator: www.polyu.edu.hk.

Confirmatory Factor Analyses (CFA) were carried out for each wave of the four personality traits, loading the corresponding 12 items scores on each personality factor. Also, CFA were carried out for each wave of the GFP and the bifactor model of psychopathology. In the case of the GFP, a one-factor model where the total score of each of the five traits of the FFM loaded onto a single factor, was specified for each wave of assessment. In the case of the bi-factor model, the assessed symptoms scales loaded both on the corresponding factors of internalizing, externalizing, hyperactivity/attention problems and a general p factor (see Etkin et al., 2020, Model 5 for a similar specification of the bi-factor model). Next, longitudinal measurement invariance was tested for all these models by applying sequential restrictions (configural, metric and scalar invariance) to observe if the factor configuration held across waves. In case of invariance or reasonable partial invariance, the factor scores of the CFAs were saved to be later included in the cross-lagged models.

In order to study the different associations between the FFM traits and the psychopathology factors measured through the three waves and between the GFP and the p factor, three cross-lagged panel models were performed. The cross-lagged panel models allow us to simultaneously examine different association hypotheses, which are more accurate in predicting reciprocal associations than ordinary regression analyses (Klimstra et al., 2010). In the first one, the FFM personality traits (except Openness to the experience), and the internalizing factor, the externalizing factor and the hyperactivity/attention problems score assessed at the three time points were included. In the second one, we included the four traits assessed in the three waves of assessment and the p factor. In the third one, the GFP and the p factor were included. A simplified version of the models is presented in Fig. 1. The correlations between personality traits and psychopathology symptoms during the same measurement wave were interpreted as continuity effects or trait-disorder co-development (De Bolle et al., 2012; Hengartner, 2018). The pathways between the symptoms assessed in a previous wave with later personality traits were considered from a complication hypothesis (De Bolle et al., 2012, 2016; Hengartner, 2018; Klimstra et al., 2010). Finally, the associations of antecedent personality traits and subsequent wave symptoms were attributed to pathoplasty (De Bolle et al., 2012; Hengartner, 2018). All the structural equation models (i.e., CFA and cross-lagged panel models) were performed using the Mplus 7.4 software.

Fig. 1
figure 1

Simplified Figure of the Cross-Lagged Panel Model

Full size image

The models’ fit was assessed using the Comparative Fit Index (CFI) > 0.90 (acceptable), > 0.95 (optimal), the Root Mean Square Error of Approximation (RMSEA) < 0.06 and the Standardized Root Mean Square Residual (SRMR) < 0.08 (Hu & Bentler, 1999; Marsh et al., 2004). For both instruments, any questionnaires with more than 5% missing values were not included in the analyses. The remaining missing data were randomly distributed (less than 2% missing values per variable).

Results

Descriptive Data

The results of the descriptive analyses are displayed in Table 1. Girls scored significantly higher for neuroticism, agreeableness and conscientiousness than boys, albeit with small effect sizes. Regarding the symptom factors, girls also obtained high scores for internalizing problems such as anxiety, depression, somatic complaints and post-traumatic symptoms with a medium effect size, and boys for externalizing problems such as antisocial behavior and aggression, with a small effect size. Cronbach’s alphas are also shown in Table 1 and were all above the good cut-off point according to the criteria of George and Mallery (2003), with the exception of the defiant behavior scale, which is only composed of 3 items.

Table 1 Descriptive Results
Full size table

Confirmatory Factor Analyses

CFAs were conducted separately for each personality trait, the bi-factor model of psychopathology (i.e., Bi-MP) and the GFP in waves one, two and three (see Table 2). For each personality trait model, the results showed that a one-factor model composed of the 12 items of its scale fitted the data well after adding minor changes based on the modification indices. All the correlations were between items within the same facet. The Bi-MP also showed acceptable fit indices across assessment waves, while the GFP showed acceptable fit indices in wave 1 and 2 after including a correlation between N and O suggested by the modification indices. The factor loadings of each item/scale on their factor are presented in the Supplementary Material.

Table 2 CFA Models for Personality Traits and Symptom Factors at Waves 1, 2 and 3
Full size table

Measurement Invariance Across Waves

We tested the longitudinal measurement invariance of the personality traits and the Bi-MP across waves one, two and three (Table 3). In the case of the GFP, as the model fit was under the recommended cut-offs at Time 3, we only tested the measurement invariance across waves 1 and 2. All the measurement invariance levels were obtained (configural, metric, scalar) based on the fact that the fit of the more restrained models did not significantly worsen, as indicated by the ΔRMSEA values < 0.015 and ΔCFI values < 0.010. In the case of the personality traits, mostly partial (instead of full) measurement invariance was obtained based on the added modifications needed to obtain a good model fit. In step 1, for neuroticism, a good model fit was obtained for the configural model, which indicates that the same factor configuration holds across waves (i.e., configural invariance). Then in step 2 (additionally constraining the factor loadings to be equal across waves), we obtained partial metric invariance as it was necessary to release the invariance constraint in one of the factor’s loadings. Then, we also constrained the item intercepts across the three waves (i.e., step 3, scalar invariance) and obtained partial scalar invariance after releasing the invariance constraint in one of the intercepts. Concerning extraversion, two correlations were added to improve the fit indices for the configural model: one between two different items assessing positive emotions from different waves, and another between two different items assessing gregariousness in the same wave. We obtained partial metric invariance in step 2, as it was necessary to release the invariance constraint in one of the factor’s loadings. Then we obtained partial scalar invariance after releasing the invariance constraint in one of the intercepts. Regarding agreeableness, there was no need for extra correlations for the configural model. No modifications were needed to reach the cut-off point and full metric invariance was found. For step 3, we obtained partial scalar invariance after releasing the invariance constraint in one of the intercepts. Finally, for conscientiousness, no modifications were needed for the configural model, partial metric invariance was found after releasing the invariance constraint in one factor loading and partial scalar invariance was observed after releasing the invariance constraint of one intercept. Overall for metric invariance, less than 20% of parameters were freed, which is considered acceptable according to Dimitrov (2010). The bifactor model of psychopathology and the GFP showed full metric and full scalar invariance across the three and two waves of assessment, respectively (see Table 3).

Table 3 Measurement Invariance across Waves 1, 2 and 3
Full size table

Cross-Lagged Models

The first cross-lagged panel model, which included the four traits and the internalizing factor, the externalizing factor and the hyperactivity/attention problems score showed good fit indices, and all above the cut-off point (χ2 = 140.495; p < 0.001; d.f. = 58; CFI = 0.977; RMSEA = 0.047; SRMS = 0.032). Similarly, the second model, which included the p factor in addition to the four traits also showed good fit indices (χ2 = 55.954; p < 0.05; d.f. = 37; CFI = 993; RMSEA = 0.028; SRMS = 0.035). The associations between the personality and psychopathology factors of both models are presented in Table 4. The associations were significant between traits and psychopathological factors, mainly for the variables pertaining to the same assessment occasion.

Table 4 Standardized Estimates for the Cross-Lagged Panel Model, including Psychopathology Factors
Full size table

The third model in which only the GFP and the p factor at time 1 and 2 were included also showed adequate fit indices (χ2 = 2.011; p < 0.150; d.f. = 1; CFI = 0.998; RMSEA = 0.039; SRMS = 0.012). The correlations between the GFP and the p factor were 0.52 (p < 0.001) and 0.27 (p < 0.001) at time 1 and time 2, respectively. The path between the GFP at time 1 and the same factor at time 2, showed a standardized beta coefficient of 0.62 (p < 0.001), while the path from the p factor at time 1 to the same factor at time 2 was 0.55 (p < 0.001). Additionally, the p factor at time 1 was related to the GFP at time 1 (β = 0.10, p < 0.01), while the association between the GFP at time 1 with the p factor at time 2 was not significant (β = 0.02, p > 0.05).

Discussion

The current study used a cross-lagged panel analysis to tackle 3-year longitudinal associations between personality traits (neuroticism, extraversion, agreeableness and conscientiousness) and the bi-factor model of psychopathology in Spanish adolescents. Only a few previous studies have simultaneously explored different personality and psychopathology association models in adolescence. Klimstra et al. (2010) used cross-lagged panel models to study the bidirectional relation between the FFM and problem behavior, while De Bolle et al. (2012) explored association models between the FFM and the correlated model of psychopathology in which an internalizing and an externalizing factor were specified. Nonetheless, this is the first study to include three broad factors of psychopathology (i.e., internalizing, externalizing and hyperactivity-attention problems) and also the p factor in a cross-lagged model to study personality trait-psychopathology symptom associations prospectively in adolescents. In addition, this is also the first study that explores the functional associations between the GFP and the p factor.

In order to explore the reciprocal associations between personality and the different factors of psychopathology, previous CFA analyses were performed to test the structure and longitudinal invariance of each personality trait, the bifactor model of psychopathology, and the GFP. Similar to those found in previous studies, all the models for the four personality traits showed partial invariance (Hengartner, 2018). However, less than 20% of parameters were freed to reach the partial invariance, which is considered acceptable (Dimitrov, 2010), and allowed us to perform the cross-lagged panel models including the personality trait factors. In addition, the bifactor model for psychopathological symptoms showed a good data fit, and these structures appeared to be invariant over time (Hengartner, 2018; McElroy et al., 2017). In the case of the GFP, the structure was acceptable for times 1 and 2 but not for time 3, the reason for which GFP invariance was tested only in the first two points of assessment, and the associations of the GFP and the p factor were explored only in the first two waves of assessment as well. The lower sample size of wave 3 compared with wave 1 and 2, and the higher mean sample age in the third point of assessment, which reflected a different period of development, could be responsible for the differences in the model fit across time.

After that, a series of cross-lagged panel models were performed. Results showed that continuity (Oltmans et al., 2018) and to a lesser extent, complication effects were observed between the general factor of personality and the p factor. Neuroticism presented the most robust continuity effects with internalizing symptoms, the hyperactivity-attention problems (De Bolle et al., 2012, 2016; Du Rietz et al., 2018; Etkin et al., 2020; Watts et al., 2019) and the p factor (Brandes et al., 2019; Etkin et al., 2020). The strongest effects were observed for the associations pertaining to the first measurement occasion. Although with smaller effects and only from the first to the second wave, support was found for the pathoplasty model, with neuroticism prospectively predicting increases in the p factor, and to a lesser extent in the internalizing factor (Castellanos-Ryan et al., 2016; Klimstra et al., 2010; Kushner et al., 2012; van den Akker et al., 2010), and in the hyperactivity-attention problems (Gomez & Corr, 2014). In accordance with the complication model, neuroticism was, in turn, predicted by internalizing problems (De Bolle et al., 2012, 2016; Klimstra et al., 2010), the hyperactivity-attention problems and the p factor, but only for the associations between the first and second waves. Nonetheless, the complication effects of externalizing problems predicting neuroticism reported by Klimstra et al. (2010) were not found in the present study.

Extraversion presented (negative) continuity effects with internalizing symptoms across all the waves of assessment, but not externalizing behavior (De Bolle et al., 2012; Etkin et al., 2020). In line with the pathoplasty model, extraversion predicted hyperactivity-attention problems, externalizing symptoms (De Bolle et al., 2012) and the p factor, but effects were rather small. Extraversion also showed (negative) pathoplastic effects with internalizing symptoms, where lower levels of extraversion predicted higher internalizing factor scores (Klimstra et al., 2010; van den Akker et al., 2010). Moreover, we found no complication effects for extraversion with either the internalizing or externalizing factors, although some previous findings indicated that depression levels predict changes in extraversion (Klimstra et al., 2010). The association between extraversion and internalizing has not been replicated consistently in the field (Kotov et al., 2010), although specific symptoms within this psychopathological factor, specifically depression and social anxiety, appear to have relatively robust negative links with extraversion in adults (Kotov et al., 2010).

Agreeableness showed negative continuity associations with externalizing symptoms (De Bolle et al., 2012, 2016; Watts et al., 2019), the hyperactivity-attention problems (Etkin et al., 2020), and with the p factor (Etkin et al., 2020) for each wave, albeit with smaller effects for each successive wave. This trait also displayed significant bidirectional (both pathoplastic and complication) effects with externalizing problems. So, externalizing problems were found to negatively predict agreeableness (Klimstra et al., 2010), while low agreeableness predicted changes in externalizing symptoms (De Bolle et al., 2012, 2016; Hengartner, 2018), which consequently may lead to later adult antisocial behavior (Moffitt et al., 2011). These results fall in line with previous findings which suggest that externalizing pathology is predicted by declines in agreeableness (Hengartner, 2018; Mervielde & De Fruyt, 2002) and by increases in frustration, as well as by diminished effortful control (Laceulle et al., 2014) when considered from a temperament perspective. Thus, by considering the bidirectional effects (complication and pathoplastic effects), less agreeable individuals appear more likely to develop externalizing problems and, as they become less agreeable over time, they subsequently present more symptoms. To a lesser extent and unexpectedly, agreeableness was also positively linked with later internalizing problems, showing small pathoplastic effects. Future replication studies should clarify if this is a spurious or a robust effect.

Finally, low conscientiousness presented continuity effects with all the factors, but mostly with hyperactivity-attention problems (Etkin et al., 2020; Gomez & Corr, 2014; Nigg et al., 2002), and the p factor (Etkin et al., 2020), followed by externalizing problems (De Bolle et al., 2012, 2016; Etkin et al., 2020; Slobodskaya & Akhmetova, 2010; Watts et al., 2019). Moreover, complication effects were found with externalizing in line with other studies (De Bolle et al., 2012). In line with the pathoplasty hypothesis, lower levels of conscientiousness predicted increments of the p factor from wave 2 to wave 3 (Castellanos-Ryan et al., 2016). No pathoplasty effects emerged with the externalizing factor (Hengartner, 2018; Mervielde & De Fruyt, 2002) or hyperactivity-attention problems (Gomez & Corr, 2014) as in previous studies. Contrary to our expectations based on other studies (Klimstra et al., 2010), we also found a positive association between conscientiousness and the internalizing factor, cross-sectionally (i.e., continuity model) and prospectively (pathoplasty model). This might be due to the p factor capturing the nonspecific variance of the reported symptoms. However, as no previous studies have addressed the association between the FFM and the bi-factor model of psychopathology, additional replication studies are necessary to confirm if conscientiousness has a strong association with the resulting internalizing factor.

Overall, results of the performed cross-lagged models for personality trait-psychopathology symptom associations showed strong continuity effects as in previous studies (De Bolle et al., 2012, 2016). In addition, these cross-sectional correlations showed a high degree of specificity as in previous studies on the FFM and the bifactor model of psychopathology (Caspi et al., 2014; Castellanos-Ryan et al., 2016; Etkin et al., 2020). Specifically, the stronger cross-sectional correlations were found between the internalizing factor and neuroticism and introversion (Castellanos-Ryan et al., 2016; Etkin et al., 2020); the externalizing factor with low agreeableness and low conscientiousness (Caspi et al., 2014; Etkin et al., 2020); the hyperactivity and attention problems score with low conscientiousness and neuroticism (Etkin et al., 2020); and the p factor with neuroticism, low conscientiousness and low agreeableness (Caspi et al., 2014; Castellanos-Ryan et al., 2016; Etkin et al., 2020).

Moreover, we found some specific pathoplasty (mainly neuroticism and introversion predicting higher internalizing symptoms, neuroticism predicting higher p factor scores and lower agreeableness predicting higher externalizing symptoms) and complication effects (mostly internalizing predicting increments of neuroticism and externalizing predicting decrements in agreeableness), which agrees with De Bolle et al. (2012, 2016), and in a similar way to the bidirectional effects between personality and problem behavior reported by Klimstra et al., (2010). As in the cross-sectional associations, the prospective associations showed that the pathoplasty and the complication hypotheses are especially tenable for those personality-psychopathology combinations that are conceptually closer. These results confer evidence for both the relevance of personality characteristics in predicting symptomatology, and symptomatology possibly ‘scarring’ later personality in adolescents (Krueger & Tackett, 2003).

The findings of this study involve some clinical implications, on the one hand, that focusing treatment and prevention interventions on riskier personality profiles might prevent some symptoms from developing later (Jeronimus et al., 2014); on the other hand, treating psychopathological symptoms at early ages might change the course of some personality aspects and prevent dysfunctional personality development (Hengartner, 2018). Traditionally in clinical contexts, the utility of youth personality assessments for decision making has been largely ignored, although traits and symptoms appear closely interwoven and should, therefore, be considered in conjunction (De Bolle et al., 2012, 2016). In this line, our results support the notion of a continuity between personality and psychopathology, which is reflected in the similarities between the structures of both constructs (Krueger et al., 2018). Also, as suggested by previous research (Krueger & Markon, 2011), empirical evidence might help to develop an overarching model by grouping symptoms/disorders based on their empirical affinities along their shared trait vulnerability to hence promote the classification of personality and psychopathology within a unified framework. Therefore, at the different levels of this structure, diverse interventions could be relevant and unified intervention protocols could be useful for transdiagnostic spectra, such as internalizing problems (Barlow et al., 2017).

The present study also has some limitations. On the one hand, our findings on personality and psychopathological symptoms were based only on participants' self-reports, which could result in biased answers. Accordingly, data collection from multiple informants might improve our understanding of processes. On the other hand, our sample consisted solely of nonclinical participants, which makes it difficult to draw conclusions about the predictability of specifically diagnosed mental disorders. Therefore, more longitudinal studies are still needed as research should aim to elucidate developmental processes regarding personality and psychopathology (Durbin, 2019) and how they function in their full complexity. Finaly, while the factors of general, externalizing and internalizing symptoms help reduce the data to broad types of pathology, it will also be the case that the field of psychology and psychiatry will require finer grained consideration of personality conditions. Despite these limitations, this research work contributes to the scarce longitudinal studies on the associations between personality traits and psychopathology in youths, as it is the first to study the etiological association models between the FFM and the bifactor structure of psychopathology, including not only the classic internalizing and externalizing spectra, but a separate hyperactivity-attention problems factor along with a general (p) factor.