Psychiatric disorders tend to co-occur in the same individual (Angold, Costello, & Erkanli, Reference Angold, Costello and Erkanli1999). Towards this end, recent efforts to understand comorbidity among psychiatric disorders have focused on examining the common and unique variation among psychiatric symptoms using continuous dimensions rather than categorical measures of psychopathology. These studies have revealed the presence of a general psychopathology factor, also called the “p factor,” which captures the shared variance among all psychopathology symptoms; residual variance is further clustered into orthogonal internalizing and externalizing factors (Caspi et al., Reference Caspi, Houts, Belsky, Goldman-Mellor, Harrington, Israel and Poulton2014). Although there has been considerable interest in utilizing this transdiagnostic approach to study psychopathology, relatively little longitudinal work has examined early predictors that may serve as transdiagnostic risk factors for general psychopathology (Caspi & Moffitt, Reference Caspi and Moffitt2018). In the current study, we examined whether infant temperament prospectively predicts general psychopathology as well as specific psychopathology dimensions (e.g., internalizing and externalizing) in childhood. Specifically, we employed a multimethod approach to examine whether dimensions of temperament, including individual differences in activity and affectivity (positive and negative affect), in early infancy are longitudinally associated with general or specific psychopathology factors.

In addition to better characterizing the structure of psychopathology, utilizing a transdiagnostic approach to measure psychopathology allows researchers to identify whether risk factors predict commonalities shared across all domains of psychopathology or more specific domains of psychopathology such as internalizing and externalizing problems. Importantly, the general psychopathology factor is unlikely to be the result of methodological artifacts (e.g., measurement bias) as it is associated with substantive and measurement-independent outcomes. For example, the general psychopathology factor exhibits significant genetic contributions (Allegrini et al., Reference Allegrini, Cheesman, Rimfeld, Selzam, Pingault, Eley and Plomin2020; Grotzinger, Cheung, Patterson, Harden, & Tucker-Drob, Reference Grotzinger, Cheung, Patterson, Harden and Tucker-Drob2019; Harden et al., Reference Harden, Engelhardt, Mann, Patterson, Grotzinger, Savicki and Tucker-Drob2020; Selzam, Coleman, Caspi, Moffitt, & Plomin, Reference Selzam, Coleman, Caspi, Moffitt and Plomin2018). Moreover, the general psychopathology factor has been related to familial psychiatric risk (Martel et al., Reference Martel, Pan, Hoffmann, Gadelha, do Rosário, Mari and Bressan2017), impaired school functioning (Lahey et al., Reference Lahey, Rathouz, Keenan, Stepp, Loeber and Hipwell2015), IQ (Caspi et al., Reference Caspi, Houts, Belsky, Goldman-Mellor, Harrington, Israel and Poulton2014), and lower executive functions (Huang-Pollock, Shapiro, Galloway-Long, & Weigard, Reference Huang-Pollock, Shapiro, Galloway-Long and Weigard2017; Martel et al., Reference Martel, Pan, Hoffmann, Gadelha, do Rosário, Mari and Bressan2017), as well as functional and structural differences in domain-general brain areas underlying the regulation of behavior, thoughts, and emotions (Snyder, Hankin, Sandman, Head, & Davis, Reference Snyder, Hankin, Sandman, Head and Davis2017). Moreover, the presence of a general psychopathology factor, along with specific factors for internalizing and externalizing, has been reported in studies across childhood, adolescence, and adulthood (Allegrini et al., Reference Allegrini, Cheesman, Rimfeld, Selzam, Pingault, Eley and Plomin2020; Caspi et al., Reference Caspi, Houts, Belsky, Goldman-Mellor, Harrington, Israel and Poulton2014; Hankin et al., Reference Hankin, Davis, Snyder, Young, Glynn and Sandman2017; McElroy, Belsky, Carragher, Fearon, & Patalay, Reference McElroy, Belsky, Carragher, Fearon and Patalay2018; Olino, Dougherty, Bufferd, Carlson, & Klein, Reference Olino, Dougherty, Bufferd, Carlson and Klein2014, Reference Olino, Bufferd, Dougherty, Dyson, Carlson and Klein2018). However, most of the work examining the characteristics associated with the general psychopathology factor in pediatric samples is based on cross-sectional data; much less is known about how early characteristics in infancy and early childhood longitudinally relate to either general psychopathology or specific internalizing and externalizing factors. An improved understanding of how early infant or childhood characteristics predict transdiagnostic risk could inform the etiology of the general psychopathology factor and aid efforts to identify children who might benefit most from interventions.

Temperament in infancy and early childhood is an important early predictor of later psychopathology (Kagan & Fox, Reference Kagan, Fox, Damon and Lerner2006; Rothbart & Bates, Reference Rothbart, Bates, Damon and Lerner2006). While theoretical perspectives on early temperament vary, there is general agreement that temperamental traits are constitutionally based individual differences in activity, affectivity, and self-regulation (or effortful control) that are relatively stable across time (Kagan & Fox, Reference Kagan, Fox, Damon and Lerner2006; Rothbart & Bates, Reference Rothbart, Bates, Damon and Lerner2006). Traditionally, specific temperament classifications have been associated with specific psychiatric disorders, such as attention-deficit/hyperactivity disorder (ADHD) (Miller, Degnan, Hane, Fox, & Chronis-Tuscano, Reference Miller, Degnan, Hane, Fox and Chronis-Tuscano2019), depression (Dougherty, Klein, Durbin, Hayden, & Olino, Reference Dougherty, Klein, Durbin, Hayden and Olino2010), and social anxiety disorder (Chronis-Tuscano et al., Reference Chronis-Tuscano, Degnan, Pine, Perez-Edgar, Henderson, Diaz and Fox2009). Taking an initial step towards a transdiagnostic approach to psychopathology, prior studies have also examined how temperament relates to the broader classifications of “externalizing” and “internalizing” psychopathology. In general, these studies have found that traits associated with withdrawal in infancy or toddlerhood, such as fear and sadness, are predictive of increased internalizing problems in later childhood (Eisenberg et al., Reference Eisenberg, Valiente, Spinrad, Cumberland, Liew, Reiser and Losoya2009; Morales, Beekman, Blandon, Stifter, & Buss, Reference Morales, Beekman, Blandon, Stifter and Buss2015; Stifter, Putnam, & Jahromi, Reference Stifter, Putnam and Jahromi2008). In contrast, higher levels of approach-related traits, such as activity level, positive affect, impulsivity, and low effortful control, are longitudinally associated with externalizing problems (Eisenberg et al., Reference Eisenberg, Valiente, Spinrad, Cumberland, Liew, Reiser and Losoya2009; Stifter et al., Reference Stifter, Putnam and Jahromi2008). Interestingly, some studies have also found that high activity level, high impulsivity, and low effortful control are not only related to externalizing but also to internalizing (Eisenberg et al., Reference Eisenberg, Valiente, Spinrad, Cumberland, Liew, Reiser and Losoya2009; Stifter et al., Reference Stifter, Putnam and Jahromi2008) or overall behavior problems (De Pauw, Mervielde, & Van Leeuwen, Reference De Pauw, Mervielde and Van Leeuwen2009). This suggests that these temperamental dimensions may act as early transdiagnostic risk factors for the development of general psychopathology, as opposed to internalizing or externalizing specifically.

To date, minimal work has directly studied the relations between childhood temperament and latent dimensions of psychopathology, including general psychopathology. Like most other research on general psychopathology in children, existing temperament–psychopathology work has mostly relied on cross-sectional designs and single-method assessments of temperament (most often parent reports). Existing studies find similar relations across childhood (3–17 years), such that general psychopathology is associated with lower effortful control, higher negative affect, and a composite measure of surgency that consists of high activity level, positive affect, impulsivity, and low shyness (Hankin et al., Reference Hankin, Davis, Snyder, Young, Glynn and Sandman2017; Olino et al., Reference Olino, Dougherty, Bufferd, Carlson and Klein2014). Moreover, these studies have also found specific links between temperament and internalizing or externalizing factors. Internalizing factors have been related to higher negative affect and lower surgency, whereas externalizing factors were related to lower effortful control and higher surgency. The only study examining longitudinal associations between socioemotional dispositions and general psychopathology found that parent-rated negative affect in childhood and adolescence (10–17 years) predicted the general psychopathology factor in early adulthood (23–31 years) (Class et al., Reference Class, Van Hulle, Rathouz, Applegate, Zald and Lahey2019). In contrast, the only study using behavioral observations of temperament for preschool-age children yielded null concurrent relations with general psychopathology (Olino et al., Reference Olino, Dougherty, Bufferd, Carlson and Klein2014). Thus, existing work investigating the role of early temperament on later transdiagnostic risk is inconsistent and relations to general psychopathology are limited to parent reports of temperament. This suggests that some of the relations linked to general psychopathology might be due to common-method variance. In the current study, we utilize a multimethod approach, including behavioral observations of temperament to avoid the issue of common-method variance.

Moreover, most previous studies linking temperament and general psychopathology have focused on broader temperament factors such as negative affectivity or surgency. Although these temperament factors are validated and widely used, the combination of multiple temperament dimensions might mask relations with psychopathology. For example, surgency and related temperament factors often include scales that index high levels of activity and impulsivity as well as positive affect and sociability (Degnan et al., Reference Degnan, Hane, Henderson, Moas, Reeb-Sutherland and Fox2011; Rothbart, Ahadi, Hershey, & Fisher, Reference Rothbart, Ahadi, Hershey and Fisher2001). These combined temperament factors might conceal the effects of specific dimensions that predict psychopathology or mask temperamental dimensions that relate to psychopathology in opposite directions (Buss, Kiel, Morales, & Robinson, Reference Buss, Kiel, Morales and Robinson2014). As such, in the current study, we examined relations with fine-grained dimensions of temperament (i.e., negative affect, positive affect, and activity levels) rather than broader temperament factors.

We examined whether infant temperament at 4 months of age prospectively predicted general psychopathology as well as specific psychopathology factors capturing internalizing and externalizing during mid to late childhood. We focused on the temperament of infants aged 4 months in an attempt to identify the earliest behavioral transdiagnostic risk factor to date. Longitudinal studies have found considerable developmental stability in the general psychopathology factor from early childhood to adulthood (Class et al., Reference Class, Van Hulle, Rathouz, Applegate, Zald and Lahey2019; Greene & Eaton, Reference Greene and Eaton2017; McElroy et al., Reference McElroy, Belsky, Carragher, Fearon and Patalay2018; Murray, Eisner, & Ribeaud, Reference Murray, Eisner and Ribeaud2016; Olino et al., Reference Olino, Bufferd, Dougherty, Dyson, Carlson and Klein2018; Snyder, Young, & Hankin, Reference Snyder, Young and Hankin2017), highlighting the importance of identifying early factors that may predict increased transdiagnostic risk across development.

The temperament dimensions of motor activity, positive affect, and negative affect can be reliably characterized in young infants via parent reports and behavioral observations (Fox, Henderson, Rubin, Calkins, & Schmidt, Reference Fox, Henderson, Rubin, Calkins and Schmidt2001; Kagan & Snidman, Reference Kagan and Snidman1991; Miller et al., Reference Miller, Degnan, Hane, Fox and Chronis-Tuscano2019; Rothbart, Reference Rothbart1981). Importantly, behavioral observations reduce recall and reporting biases and avoid common-method variance as they are independent measures from parent reports of psychopathology (Kagan & Fox, Reference Kagan, Fox, Damon and Lerner2006), whereas parent reports provide an efficient and scalable way to assess temperament, which also provides a broader perspective that includes a variety of naturally occurring situations (Rothbart & Bates, Reference Rothbart, Bates, Damon and Lerner2006). Therefore, in the current study, we combined these approaches to obtain a robust multimethod measure of temperament.

Moreover, we focused on psychopathology during school age, as opposed to earlier ages, since the behavioral manifestation of psychopathology may be most evident during school years, with an average age of onset around 7 years (Jones, Reference Jones2013; Kessler et al., Reference Kessler, Amminger, Aguilar-Gaxiola, Alonso, Lee and Ustun2007). For instance, many children display moderate to high levels of externalizing and internalizing problems during toddlerhood and preschool. Although many children “outgrow” these behaviors, for some, these problems persist into school age and adolescence, leading to socioemotional problems (Bufferd, Dougherty, Carlson, & Klein, Reference Bufferd, Dougherty, Carlson and Klein2011, Reference Bufferd, Dougherty, Carlson, Rose and Klein2012; Campbell, Spieker, Burchinal, & Poe, Reference Campbell, Spieker, Burchinal and Poe2006; Egger & Angold, Reference Egger and Angold2006).

Based on the above literature review, we predicted that higher infant motor activity, negative affect, or positive affect would all be associated with the general psychopathology factor later in life. Based on the findings of past studies, we further hypothesized that higher negative affect and lower positive affect would be associated with the internalizing-specific factor, whereas higher positive affect and motor activity (i.e., temperament dimensions falling under surgency) would be associated with the externalizing-specific factor. Finally, because there are important differences by gender in both the incidence of psychopathology (Seedat et al., Reference Seedat, Scott, Angermeyer, Berglund, Bromet, Brugha and Kessler2009) and the developmental pathways from early temperament to later psychopathology (Miller et al., Reference Miller, Degnan, Hane, Fox and Chronis-Tuscano2019; Morales et al., Reference Morales, Beekman, Blandon, Stifter and Buss2015, Reference Morales, Miller, Troller-Renfree, White, Degnan, Henderson and Fox2019; Rubin, Burgess, Dwyer, & Hastings, Reference Rubin, Burgess, Dwyer and Hastings2003), we also explored whether prospective relations between infant temperament and later psychopathology differed by gender. The reason we do not propose specific gender hypotheses is because none of the previous studies examining gender differences in the temperament–psychopathology relation have accounted for the shared variance across disorders.

Results

Preliminary analyses

Of the total sample (N = 291), 230 participants had at least one timepoint with a parent report of psychopathology; 191 had data at age 7 years, 186 had data at age 9 years, and 175 had data at age 12 years. Of these, 83.0% of children had data from at least two ages. As such, we utilized an analytic approach (i.e., FIML) that reduced the potential biases associated with missing data. Finally, children included in the bifactor model (n = 230) did not differ from the rest of the sample in gender (p = .82), maternal education (p = .33), or any of the temperamental dimensions (p values > .08). The only exception was maternal ethnicity (p = .05), such that children with data on these measures of psychopathology were more likely to have non-Hispanic Caucasian mothers than children with missing data. As such, this variable was included in the models (see below). The descriptive statistics and correlations among all the study variables are provided in Table S4 in the Supplementary Material.

Comparison of the bifactor model and correlated factors model

The correlated factors model and the bifactor model were fit as confirmatory factor analyses to examine the structure of psychopathology. The fit indices and factor loadings for both models are presented in the Supplementary Material (Table S5).

The correlated factors model had an adequate fit (comparative fit index (CFI) = 0.99, Tucker–Lewis index (TLI) = 0.98, root mean square error of approximation (RMSEA) = 0.03, standardized root mean square residual (SRMR) = 0.04). All symptoms loaded significantly onto the externalizing and internalizing factors. Moreover, the externalizing and internalizing factors were significantly related (b = 0.708, SE = 0.064, p < .001), implying considerable shared variance.

The bifactor model also demonstrated adequate fit (CFI = 1.00, TLI = 0.99, RMSEA = 0.02, SRMR = 0.04). All symptoms loaded significantly onto the general psychopathology factor. The remaining variance of all symptoms loaded significantly onto the specific internalizing and externalizing factors – the only exceptions were aggression at all ages and attention problems at age 12, which did not significantly load onto the externalizing factor while significantly loading onto the general psychopathology factor (Table S5). Of note, the general psychopathology factor demonstrated good reliability (omega = .90 and omega hierarchical = .71). Moreover, the general psychopathology factor explained the majority of the variance, as indicated by the explained common variance (ECV = .65). The specific factors, internalizing and externalizing, explained similar amounts of variance (ECV_{internalizing} = .19 and ECV_{externalizing} = .17) and exhibited lower reliability after accounting for the general psychopathology factor (omega_{internalizing} = .77 and omega_{externalizing} = .90; omega hierarchical_{internalizing} = .42 and omega hierarchical_{externalizing} = .18).

Although both the correlated factors model and the bifactor model demonstrated adequate fit, comparisons of the two models revealed that the bifactor model was a better reflection of the factor structure of psychopathology than the correlated factors model (Δχ² (11) = 25.34, p = .008). These results support previous literature and the a priori decision to examine temperament differences in psychopathology based on the three factors from the bifactor model (general psychopathology, internalizing, and externalizing). However, we examined the predictive relations of temperament and psychopathology factors using both the bifactor model and the correlated factors model because fit statistics can be biased in favor of bifactor models (Greene et al., Reference Greene, Eaton, Li, Forbes, Krueger, Markon and Docherty2019). Moreover, showing relations with the correlated factors model as in previous studies can aid the interpretation of novel relations with general psychopathology. Finally, we evaluated potential gender differences in the structure of psychopathology by testing if the models significantly differed by gender. To do this, we fitted a multigroup model where all the parameters were estimated separately for males and females and tested whether constraining the factor loadings to be equal across groups significantly worsened the model fit. We did this separately for each model of psychopathology (i.e., correlated factors model and bifactor model) and did not observe a significant difference across models (p values > .074), indicating that the factor structure did not differ between males and females.

The impact of infant temperament on the bifactor model

In order to examine the relations between the temperament dimensions and psychopathology, we added the three temperament dimensions (motor activity, positive affect, and negative affect) as predictors of the three psychopathology factors (general psychopathology, internalizing, and externalizing) within the bifactor model. Moreover, we also included child gender, maternal ethnicity, and maternal education as covariates. As shown in Table 1 and Figure 1a, motor activity levels significantly predicted the general psychopathology factor (β = .44, p < .001), such that increased motor activity levels in infancy predicted increased general psychopathology in childhood. Infant positive affect also significantly predicted the internalizing-specific factor (β = −.37, p = .009), such that increased positive affect in infancy predicted lower scores in the internalizing-specific factor. Importantly, these relations survived multiple comparison correction. No other relations between infant temperament and the latent psychopathology factors were statistically significant (Table 1 and Figure 1a). When examining for a potential moderation by gender in these relations, a chi-square difference test revealed a nonsignificant difference (Δχ² (9) = 13.83, p = .129), indicating the relations between temperament and psychopathology factors did not significantly differ between males and females.

Figure 1.

Standardized regression coefficients for the regression models for temperament dimensions predicting latent psychopathology factor scores in the bifactor model (a) and the correlated factors model (b), along with parent-reported predicting latent psychopathology factor scores in the bifactor model (c) and the correlated factors model (d). Black solid arrows indicate statistically significant relations after correction for multiple comparisons (false discovery rate; FDR). Gray dotted arrows represent nonsignificant regression paths. Albeit the analysis with the bifactor model was performed following the recommended methods from Koch et al. (Reference Koch, Holtmann, Bohn and Eid2018), the data are presented as traditional regression analyses to simplify the presentation. The effects of the covariates are not displayed in the figure, but are shown in Table 1.

Table 1.

Regression paths from the path model of temperament composites predicting psychopathology factors

Note: p factor = general psychopathology factor, β = standardized estimates, b = unstandardized estimates, LL = lower limit of 95% confidence interval, UL = upper limit of 95% confidence interval. Bold estimates and p values showed statistical significance with adjustment for multiple comparisons (q < .05; FDR). Albeit the analysis with the bifactor model was performed following the methods recommended by Koch et al. (Reference Koch, Holtmann, Bohn and Eid2018), the table is presented as a traditional regression analysis to simplify its presentation.

The effects of infant temperament on the correlated factors model

In order to examine the relations between the temperament dimensions and the more traditional factors of psychopathology, we added the three temperament dimensions (motor activity, positive affect, and negative affect) as predictors of the psychopathology factors (internalizing and externalizing) within the correlated factors model. We also included child gender, maternal ethnicity, and maternal education as covariates. As shown in Table 1 and Figure 1b, motor activity levels significantly predicted the externalizing and internalizing factors (β = .37, p < .001 and β = .27, p = .002, respectively), such that increased motor activity levels in infancy predicted increased externalizing and internalizing psychopathology in childhood. Moreover, positive affect negatively predicted internalizing problems (β = −.24, p = .016). Importantly, these relations with activity levels and positive affect survived multiple comparison correction. No other relations between infant temperament and the latent psychopathology factors were statistically significant (Table 1 and Figure 1b).

In exploratory analyses examining the potential moderation by gender in these relations, the results revealed a significant difference (Δχ² (6) = 20.76, p = .002), indicating that some of the relations between temperament and general psychopathology factors significantly differed between males and females. However, follow-up analyses comparing each regression path between males and females indicated that none of the paths significantly differed on their own – albeit several were marginally significant. Given the lack of a priori gender-specific hypotheses and absence of robust gender differences – particularly in the bifactor model, which is the focus of the current study – we did not explore gender differences further.