This study is the first to test whether a latent processing speed factor is related to the p factor in youth. Results showed that the PS factor was significantly, negatively associated with the p factor, indicating that slower PS is associated with increased general psychopathology in youth. We found similar correlations (r ~ −0.42) between the p factor and PS across various modeling approaches encompassing different raters (mixed-reporter, caregiver only) and hierarchical models (second-order, bifactor), which speaks to the stability and generalizability of the correlation.
Processing Speed and Psychopathology
The significant correlation between the
p factor and PS expands the breadth of mental health symptoms that should be explored in relation to PS. PS is most frequently included in studies examining neurodevelopmental disorders (McGrath et al.,
2011; Peterson et al.,
2017), but this study, along with existing work, suggests that PS might have broader relationships with internalizing and externalizing disorders (Willcutt et al.,
2008; Nigg et al.,
2017; but see Calhoun & Mayes,
2005 for a different view). In the existing work showing associations between PS and specific mental health disorders, studies do not usually account for general mental health, making it difficult to know whether associations occur because PS is related to mental health generally, specific disorders, or both. The present study adds to the existing literature by suggesting that PS is related to mental health generally, providing some support that PS could be a pervasive correlate related to a wide range of mental health symptoms. These results suggest that PS may be a transdiagnostic mechanism with implications for prevention and early intervention for general psychopathology symptoms. Unfortunately, the cross-sectional nature of this dataset limits conclusions about causal directionality between PS and mental health. Future research should include longitudinal work to assess the developmental unfolding of this relationship. Furthermore, this study cannot speak to potential neurobiological mechanisms behind the PS/p-factor relationship, but one plausible hypothesis is that PS and various psychopathologies are related because they are both related to white matter connectivity (Nigg et al.,
2017; Thomason & Thompson,
2011). Future research across multiple levels of analysis (neurobiological, cognitive, and behavioral) is needed to develop a fuller understanding of the PS/
p-factor relationship.
Secondary, exploratory analyses examined the relationship between PS and specific psychopathology domains after accounting for the
p factor (i.e., bifactor model). These domain-specific relationships were not significant, suggesting that the association between PS and psychopathology is strongest for the general factor. We caution against overinterpretation of this result given the lack of reliability and stability of specific factors after extracting general variance (Eid,
2020; Forbes et al.,
2021). Due to the statistical limitations of bifactor models, we cannot fully rule out domain-specific associations between PS and specific psychopathology.
p Factor and Cognition
PS, EF, and IQ are overlapping, yet distinguishable cognitive constructs; thus, disentangling their general and specific relationships with mental health symptoms is important. The correlation between PS and the
p factor (
r = -0.42) was stronger than previously reported correlations of the
p factor with EF and IQ (
r ~ 0.1–0.3) (Caspi et al.,
2014; Grotzinger et al.,
2019), indicating that PS might account for more variance in general mental health than other aspects of cognition. Indeed, this pattern was found in our sample, as the correlation between the
p factor and our latent general cognition factor (
r = -0.24) was weaker than with our latent PS factor (
r = -0.42). When both PS and general cognition were included in the same model as predictors of the
p factor, PS contributed uniquely above and beyond general cognition, but general cognition did not contribute uniquely above and beyond PS. This analysis indicates that the relationship between general cognition and mental health previously found in the
p factor literature may be attributable to PS, aligning with both theoretical and empirical literature positing PS as a developmental driver of general cognitive skills, especially fluid reasoning (Fry & Hale,
1996; Kail,
2007). This study did not directly examine EF, so future research should examine all three cognitive constructs (PS, IQ, EF) to determine general and specific associations with the
p factor.
Modeling Approach
Strengths of our modeling approach included (1) latent measurement of both PS and psychopathology, (2) using a second-order model in primary analyses with convergence from bifactor models, and (3) including multiple raters, with child-report of internalizing symptoms and caregiver-report of externalizing symptoms. It was important that we used a latent modeling strategy for PS given confounding cognitive skills that influence PS measurement. Further, while second-order models have been used in some previous
p factor literature (Michelini et al.,
2019), most studies used bifactor models which have received scrutiny for model instability and inflated fit statistics (Eid et al.,
2017; Mansolf & Reise,
2017). Modeling methodology is evolving and each model (e.g., bifactor, second-order) has strengths and weaknesses (Eid,
2020; Heinrich et al.,
2021). The fact that we found similar correlations between PS and the
p factor using both second-order and bifactor models speaks to the robustness of the finding.
Our decision to use multiple raters was in line with best clinical practice and research (Kemper et al.,
2003; Smith,
2007), with the child reporting on internalizing symptoms and the caregiver reporting on externalizing symptoms. Previous
p factor literature has examined child-report or caregiver-report, but they have mostly been in separate models (for exceptions see Laceulle et al.,
2015; Lahey et al.,
2012). One common concern regarding the
p factor is that it may be overly influenced by individual differences in reporting style, such as a positive or negative skew when reporting symptoms (termed halo effect) (Caspi & Moffitt,
2018). Using two reporters of different symptoms domains in one model can help address this concern. We encourage consideration of this second-order, mixed-reporter model in future
p factor literature given the model strengths (e.g., raters aligning with best practice; removal of potential halo effect), as well as the convergence of results across models.
Unexpected Modeling Results
Across our primary and secondary
p factor models, we found a few unexpected results deserving further comment. We observed a difference in the loading of internalizing symptoms based on whether we used a single-reporter (
β = 0.81) or mixed-reporter (
β = 0.27) second-order model of the
p factor. There are a few potential explanations for why the loading may be lower in the mixed-reporter model. First, this discrepancy may reflect a lack of convergence between youth and caregiver reports (De Los Reyes et al.,
2015) coupled with the fact that internalizing symptoms were reported by the child, whereas both externalizing and attention were reported by the caregiver, potentially weighting the
p factor toward the caregiver. Alternatively, this result might indicate a true difference in the relationships between internalizing and externalizing symptoms with the
p factor. It is difficult to assess existing evidence for this hypothesis because there is a lack of convergence of internalizing loadings in previous
p factor literature (lower loadings ranging from
β = 0.13–0.46, higher loadings ranging from
β = 0.72–0.90) (Castellanos-Ryan et al.,
2016; Laceulle et al.,
2015; Michelini et al.,
2019). While we cannot resolve why the internalizing loading was lower in the mixed-reporter vs. single-reporter model, we report the loading discrepancy as an important consideration for future work, especially given the prevalent use of single-reporter models to date. For the purposes of our central question, we note that the correlation between PS and the
p factor was stable across mixed-rater versus single-rater models.
Consistent with previous factor analyses of psychopathology (Achenbach & Rescorla,
2001), model fit was better when attention was a distinct first-order domain than when included with externalizing symptoms. One interesting result is that when looking across all four
p factor models (second-order/bifactor; mixed-rater/single-rater), the loadings of attention measures on the
p factor were very high, to the degree of suggesting that the
p factor and attention may be synonymous constructs. This high loading, also found by Brikell et al. (
2020), warrants further investigation. It is consistent with both theory and scientific evidence suggesting that attention is a key transdiagnostic correlate that is relevant to a range of psychopathology symptoms (Aitken & Andrade,
2021; King et al.,
2013; Racer & Dishion,
2012). In response to the attention loading being so high, we completed a secondary analysis where we dropped the attention/hyperactivity measures from the
p factor to ensure that the correlation between PS and the
p factor was not solely due to these measures. The resulting correlation between PS and the modified
p factor (
r = -0.42,
p < 0.001) was similar to the primary result, providing evidence that PS is associated with general mental health symptoms apart from the known association with attention/hyperactivity.
Limitations and Future Directions
This study had several strengths, including a large sample size, multiple measures of psychopathology and PS, and use of latent modeling. However, our findings should be interpreted in the context of several limitations. First, our study is limited by the recruitment of participants in the study. This sample has less socioeconomic, racial, and ethnic diversity than the United States population, limiting the generalizability of the findings. Future research should include more diverse samples and consider the influence of self-reported gender. Additionally, the study does not include measures outside of internalizing, externalizing, and attention domains (e.g., psychosis, autism spectrum disorder, OCD). Further, we focus on cognitive speed tasks in this study, which represents one measurement tradition for PS. Because there are many different measurement approaches, we cannot be certain that these results would generalize to other forms of speed (e.g., EEA, decision time). However, there is extensive psychometric literature indicating that various types of speed load highly onto a general speed factor (McGrew & Evans,
2004; Salthouse,
1996), providing some evidence to expect generalization. Future research should examine the relationships between the
p factor and other types of PS measurement.
In addition to measurement limitations, this dataset is enriched for attention and reading difficulties, both of which have been shown to be related to PS (McGrath et al.,
2011). However, in practice, we have observed that it is a “soft selection” as those recruited for potential attention and reading challenges often do not meet criteria for a disorder, and those who were recruited in the control group often have undetected ADHD and reading difficulties, resulting in relatively normal distributions of these skills. In the current sample, 23% of children met symptom criteria (more liberal than full diagnostic criteria) for ADHD based on the OR rule from symptoms ratings (i.e., at least 6/9 ADHD symptoms from parent report OR teacher report [Lahey et al.,
1994]). Given the slight enrichment for ADHD, a question is whether this sample has higher-than-expected rates of other psychopathology which could artificially strengthen the covariance of mental health symptoms and therefore the
p factor. However, this does not seem to be the case, as clinically significant rates of symptoms are 9% for depressive symptoms (CDI ≥ 13 symptoms), 13% for internalizing symptoms broadly (child-report YSR T ≥ 70), and 4% for externalizing symptoms broadly (parent-report CBCL T ≥ 70). While some of these rates are higher than diagnostic rates in epidemiological studies (Merikangas et al.,
2009), we would expect this as these are symptom counts and not diagnostic interviews. Thus, the elevations do not seem to indicate much higher than expected rates of psychopathology. Secondly, while our attention loadings are higher than some previous studies, we note that another
p factor model in a population-based sample of youth showed similarly high loadings for attention symptoms (Brikell et al.,
2020). The high loading for attention is also consistent with other theoretical (Racer & Dishion,
2012) and empirical work (King et al.,
2013) showing that attention is a transdiagnostic correlate of internalizing and externalizing symptoms. Taken together, these studies provide some assurance that our
p factor is consistent with previous literature and our high attention loading is not entirely due to the sampling.
The sample consists of twins who are at higher risk for preterm birth, which has been linked to lower PS, lower cognitive scores, and increased mental health challenges (Beauchamp et al.,
2022). To understand whether this sampling approach biased our results, we compared our findings to those from non-twin samples. Our association between the
p factor and general cognition (
r = -0.24) mirrors previous findings in non-twin samples (
r = -0.19− -0.34) (Caspi et al.,
2014; Grotzinger et al.,
2019), providing some reassurance that a higher incidence of preterm birth did not exert a strong upward bias on the correlation between PS and the
p factor. Ultimately, however, it is important to replicate this finding in an independent, non-twin dataset.
Finally, there was a median of 66 days between testing sessions. At the first study visit, parents received child psychopathology questionnaires to complete at home and bring to their next study visit. Thus, it is possible that caregivers completed some ratings of child psychopathology up to two months, on average, before collection of youth-report measures. In practice, we found that many families filled out the questionnaires immediately before they needed to bring them to their next study visit, which was the visit at which children filled out their own psychopathology measures. However, to the extent that the measures could be separated in time, this would make finding a p factor more difficult and could attenuate the correlation with PS. We found a robust relationship between the p factor and PS, but we note that it could be stronger if data were consistently collected at the same timepoint.
In conclusion, this study was the first to examine the relationship between a latent PS factor and the p factor in a sample of youth. We found a significant, moderate association between PS and the p factor (r = -0.42) that was stable across different raters and different modeling techniques. This study expands the existing literature examining PS in relation to specific disorders by showing that PS is related to what is shared across psychopathology. The association with the p factor was stronger for PS than general cognition, both in this sample and when comparing to previous correlations with IQ, indicating that PS could be an especially important transdiagnostic construct that warrants further attention and investigation.