Peer problems have been extensively studied in children with ADHD (Hinshaw and Melnick, 1995; Hoza et al., 2005; Hoza, 2007; Lee et al., 2008; McQuade and Hoza, 2008; for a thorough review see Ros and Graziano, 2017). Difficulties with peers have been associated with impaired functioning across a wide range of domains (Wheeler and Carlson, 1994; Mrug et al., 2012). Research suggests that both IA and HI symptoms may influence peer relations, but in a different manner. Specifically, children with a combined presentation of ADHD (ADHD-C; i.e., those with a high number of inattentive and hyperactive-impulsive symptoms) are described as more aggressive with their peers than children in comparison groups, and therefore more likely to receive peer rejection, whereas children with primarily IA symptoms (ADHD-I) are often withdrawn during social interactions, probably related to a poor ability to recall the content of conversations with their peers along with other social skills deficits (Mikami et al., 2007). Furthermore, a cascading effect may exist (see Hinshaw, 2017), through which ADHD symptoms and social problems magnify each other over time (Tseng et al., 2014). Community studies show that these findings are not restricted to children with a formal ADHD diagnosis (e.g., Andrade and Tannock, 2014). However, questions remain regarding pathways between the IA and HI dimensions and future peer-related outcomes.

Intellectual function is a multidimensional concept that has strong predictive power to important life outcomes in the general population (Deary, 2012). Tests of intellectual function, like the third edition of the Wechsler Intelligence Scale for Children (WISC-III; Wechsler, 1991), include subtests assessing partially independent indices of cognitive function (Kamphaus et al., 1994; Keith and Witta, 1997; Roid and Worrall, 1997). The WISC-III generates indices for Verbal Comprehension Index (VCI), Perceptual Organization Index (POI), Freedom from Distractibility Index (FFDI), and Processing Speed Index (PSI). Although a meta-analysis by Frazier et al. (2004) found a general pattern of impairment across these four indices, other studies have shown that IA may be particularly tied to impaired performance in specific aspects of intellectual function. For example, children with ADHD-I show slower processing speed compared to individuals with other presentations of ADHD (Chhabildas et al., 2001; Calhoun and Mayes, 2005; Riccio et al., 2006; Mayes et al., 2009; Thaler et al., 2013). Similarly, slow processing speed has been linked with teacher reports of hypoactivity, a behavior closely related to IA (Lundervold et al., 2011). This pattern suggests that there is a unique link between symptoms of IA and processing speed, which might aid the understanding of future problems with peers.

Several factors may act as predictors of peer problems. In a community study, Bellanti and Bierman (2000) showed that both IA and intellectual function in kindergarten were independent predictors of social functioning in first grade. Interestingly, the association with poor social function was stronger for IA than for low intellectual ability. The importance of IA was also supported by Huang-Pollock et al. (2009), who showed that effects on social function (i.e., the ability to pick up subtle verbal cues and remember a conversation) were primarily driven by IA rather than HI symptoms. Others, like Bunford et al. (2015), have instead described IA as a mediator between specific aspects of intellectual/executive function (i.e., response inhibition) and social adjustment. Finally, as noted above, compelling findings show that children with ADHD-I show slow processing speed, and that this slowness can help explain why inattentive children struggle socially (Calhoun and Mayes, 2005; Thaler et al., 2013).

Taken together, previous research has documented a wide range of factors predicting problems with peers. Investigations including performance on a test of intellectual function point to the importance of processing speed, but there are still few population-based studies that investigate how different aspects of intellectual function might contribute to explain the link between IA and HI on future peer problems.

In a clinically diverse sample of children participating in a population-based study, we first investigated the independent contributions of IA and HI in predicting future peer problems. IA and HI were reported by parents and teachers when the children were 7–9 years (baseline), and peer problems were reported by parents at baseline as well as when the children were in early adolescence (ages 11–13 years). Key demographic variables (age and sex), the presence of an ADHD diagnosis, and baseline peer problems reported by parents were adjusted for in the statistical analyses. Second, we investigated whether the association between IA/HI and peer problems was mediated by the child's performance on the key indices of the WISC-III at ages 8–10. Based on previous findings, we hypothesized that (i) IA and HI in primary school children would predict later peer problems over and above demographic variables and baseline peer problems; (ii) indices from WISC-III would partially mediate the direct effect of IA and/or HI on peer problems, with the strongest mediating effect from processing speed, given its prior linkages with IA; and (iii) that these findings would be retained with the presence of an ADHD diagnosis included as a covariate.

The Bergen Child Study (BCS) is a longitudinal, multi-wave, population-based study on childhood mental health and development. The present study includes data from three sequential time points of data collection. The first study wave was launched in 2002, organized into three phases. In the first phase, a questionnaire including the Strength and Difficulties Questionnaire (SDQ, Goodman, 1999) and the Swanson Nolan and Pelham-IV Questionnaire (SNAP-IV, Swanson, 1992) was sent to parents and teachers of all children attending 2nd−4th grade (7–9 years of age) in any school in the city of Bergen, Norway (9,439) (see Heiervang et al., 2007 for details). This constitutes the baseline measure of peer problems, as well as IA and HI symptoms, for the present study.

In a second phase, parents of children categorized as screen positives (for whom parents or teachers reported an SDQ total score that exceeded the 90th percentile and severe impairment on the impact section, or the score of one of the other included questionnaires exceeded the 98th percentile) and parents of a subset of screen negative children were interviewed according to the Development and Well-Being Assessment (DAWBA; Goodman et al., 2000). About 1 year later, a subgroup (n = 421) of the children and their parents were invited to a third clinical phase, including all children with any diagnosis according to DAWBA (n = 139). A total of 329 children and their parents participated in this phase, consisting of a short physical examination, a neuropsychological assessment including a test of intellectual function (WISC-III, Wechsler, 1991), and a clinical interview according to the Schedule for Affective Disorders and Schizophrenia for School Aged Children, Present and Lifetime Version (KSADS-PL; Kaufman et al., 1997) (see Lundervold et al., 2011 for details). We include this latter portion of the third phase as the intermediate time point, in between documentation of ADHD symptoms and later ascertainment of peer functioning, at which measures of intellectual functioning were ascertained.

Finally, when the children were 11–13 years old, parents and teachers completed a similar questionnaire as in the first wave, including the SDQ and SNAP-IV. This data collection represents the third time point.

The present study included the subsample of 127 children with complete data for the variables of interest (i.e., across all the three described time points). Approximately half of the mothers and fathers of the participating children had completed education at a college or university level (see Supplementary Table 1 for details). Information about education was not available for four fathers. Ethnicity was not assessed, given the homogeneity of the Norwegian population. That is, the vast majority of the population-level BCS sample spoke Norwegian at home, with only 5% reporting another primary language (Heiervang et al., 2007). The included subsample was not significantly different from those without complete data in terms of IA and HI scores, scores on the WISC-III, age, or sex (all ps > 0.05). Parents gave written consent for participation, and the study was approved by the Regional Committee for Medical and Health Research Ethics in Western Norway.

Peer problems were assessed by the total sum score of parent reports on the peer problems subscale of the SDQ (Goodman, 1999), both at the first and third data points. Although cross-informant information from both parents and teachers is recommended (e.g., Renk and Phares, 2004), parent reports were selected as the outcome measure because teacher reports were available for only 66 participants (52%) at age 11–13. The subscale consists of five items rated on a Likert scale from 0 (“not true”) to 2 (“certainly true”), assessing whether the child is a victim of bullying, has at least one good friend, is solitary, is generally liked by other children and if he/she gets along better with adults than with other children. The items assessing the presence of at least one good friend and being liked by other children were reverse scored, so that a higher total score reflects more peer problems. The peer problems subscale has adequate psychometric properties and has been validated in several countries, including Nordic countries (Obel et al., 2004). The peer problems subscale of the SDQ had acceptable internal consistency at both data points (α = 0.72 at age 7–9; α = 0.79 at age 11–13).

IA and HI were assessed using the mean parent and teacher scores as reported on the SNAP-IV (Swanson, 1992) at the first data point (i.e., when the children were between 7 and 9 years old). The scale has been shown to have acceptable internal consistency (Bussing et al., 2008). It includes the 18 items used to define the IA and HI symptoms associated with an ADHD diagnosis. In the original SNAP-IV, each item is evaluated according to four levels. In the present study, parents and teachers evaluated the responses on a 3-level Likert-type scale (not true, somewhat true, or certainly true) to follow the response pattern of the remaining scales included in BCS questionnaires, with a higher score revealing higher symptom severity. Parent and teacher ratings were available for both included subscales. These different-source ratings were strongly correlated both for IA, r₍₁₂₅₎ = 0.64, p < 0.01, and HI, r₍₁₂₅₎ = 0.58, p < 0.01. Therefore, the mean scores reported by the parents and teachers were summed to construct cross-informant composite scores of IA and HI. To run an extra check for cross-informant discrepancy in predicting parent-rated peer problems, separate reports by source were included in supplementary analyses.

Intellectual function was assessed by the four indices generated from the Norwegian version (Eilertsen and Johnsen, 2003) of the WISC-III (Wechsler, 1991): the VCI, POI, FFDI, and PSI. The WISC-III was the most recent version available when the study was conducted, and the test has been validated in both typically developing and clinical samples (Burton et al., 2001). The difference between performance on similar indexes in WISC-III and WISC-IV is shown to be modest (Mayes and Calhoun, 2006). In the present study, the subtests of WISC-III were scored according to Swedish norms (Sonnander et al., 1998). The index scores were defined according to Kaufman (1994). VCI includes a set of verbal tasks assessing different aspects of the child's verbal reasoning abilities; POI includes timed tests of visual construction and analysis assessing the child's non-verbal reasoning abilities; FFDI comprises tasks assessing different aspects of working memory, and PSI includes tests assessing the speed with which the child is able to process information without errors.

ADHD diagnosis, along with fulfillment of criteria for any other DSM-IV disorder, was assessed using the K-SADS (Kaufman et al., 1997). The K-SADS is a structured clinical interview, validated for children between the ages of 6 and 18. Good agreement between K-SADS and other instruments used to identify ADHD was shown in a previous study from the BCS group (Posserud et al., 2014). Here, a group of psychologists and a medical doctor interviewed the parents and children to determine the presence or absence of any DSM-IV diagnoses. ADHD status was defined as positive if a child presently had a possible or definite diagnosis of ADHD. All other children were defined as ADHD negatives.

Pearson correlations were calculated to assess covariation between all included variables. We then performed a series of t-tests to investigate group differences between children with positive and negative ADHD status and between boys and girls. Hierarchical linear regression analyses were then performed to investigate the contribution of each predictor variable, as well as each step's incremental explained variance. Age, gender, and ADHD status comprised in the first step, baseline peer problems the second step, and the parent/teacher composite scores of IA and HI symptoms the third step. The unique contribution of IA and HI was also tested by including one before the other in two different steps (see Pedhazur, 1982). Before investigating the indirect effect of the WISC indices on the direct effect between the two symptom dimensions (IA and HI) and later peer problems, we computed the least-angle regression procedure (LARS) in Stata 20 to reduce the number of variables (Efron et al., 2004). This restricted model selection approach introduces less noise and has greater power compared to a model including all variables. All variables from the previously described analyses were subjected to LARS, together with VCI, POI, FFDI, and PSI from the WISC-III. The selected variables were then included in a second set of regression analyses to investigate the significance of the WISC-III indices in this optimized model.

To confirm the effects revealed by these analyses, the variables from the restricted model were included in a parallel mediation analysis using the PROCESS macro (Hayes, 2013). This analysis evaluated the best-fitting WISC-indices (measured at age 8–12) as potential parallel mediators of the relation between childhood ADHD symptoms (at age 7–9) and preadolescent peer problems (at age 11–13), with baseline peer problems included as a covariate. The PROCESS analyses applied bias-corrected and accelerated (BcA) bootstrapping with 10,000 resamples, as this method has been shown to produce valid confidence intervals and reduce the influence of bias from non-normality and heterogeneity (Efron, 1987; Wilcox, 2012). All analyses except LARS were performed using IBM SPSS Statistics 23.

The total sample included 127 children, with 32 (25%) meeting criteria for an ADHD diagnosis (n = 5 females, 15%). Twenty-eight children with ADHD also met criteria for another diagnosis (n = 11 females, 39%). Ninety-five children did not meet criteria for ADHD (n = 40 females, 42%), but 29 of them met the criteria for one or more other mental disorders according to the KSADS-PL (Supplementary Figure 1).

The total sample included 82 boys (65%) and 45 girls, and boys were more likely to meet criteria for an ADHD diagnosis (χ = 7.34, p = 0.01). Boys showed higher scores on the IA and HI subscales, and scored lower on the FFDI and PSI indices, compared to girls (all ps < 0.05). Children with ADHD were older, obtained higher scores on the IA and HI scales, had higher peer problems scores, and obtained lower scores on VCI, POI, FFDI and PSI indices, compared to children without ADHD (all ps < 0.05; see Table 1).

The IA and HI subscales correlated moderately with peer problems and WISC-III indices, and the correlation between the peer relations scores in childhood and preadolescence was strong. Smaller, but statistically significant correlations were found for age, parent and teacher ratings of IA and FFDI (Table 2).

Hierarchical regression analyses revealed significant contributions of age, gender, and ADHD diagnosis [ΔR² = 0.082, ΔF_{(3, 123)} = 3.67, p = 0.01] to future peer problems at age 11 – 13. Not surprisingly, information about baseline peer problems added 18.5% to the explained variance [ΔF_{(1, 122)} = 30.83, p < 0.001]. Next, the IA and HI subscales added 4.5% to the explained variance [ΔF_{(2, 120)} = 4.17, p = 0.02], with the total model explaining 31.5% of the variance [F_{(6, 120)} = 9.19, p < 0.001].

To ascertain the relative contributions of IA and HI, we added each subscale in separate analyses (Pedhazur, 1982). When included in a third step beyond age, gender, ADHD status, and baseline peer problems, HI did not make a significant contribution [ΔR² = 0.007, ΔF_{(1, 121)} = 1.21, p = 0.27], whereas IA did, increasing the model's explained variance by 4% [ΔR² = 0.040, ΔF_{(1, 120)} = 7.07, p = 0.01]. Entering IA and HI scores rated by either parents or teachers separately did not alter the results. Inspection of the single predictors in the total model showed that only IA (b = 0.218, SE = 0.092, p = 0.017) and baseline peer problems (b = 0.405, SE = 0.124, p = 0.002) were significant in predicting future peer problems, when covarying the other variables.

Next, the contributions of age, gender, ADHD diagnosis, baseline peer problems, HI, IA, and the four WISC-III indices (VCI, POI, FFDI, and PSI) on future peer problems were analyzed using LARS, providing a reduced model including variables for the PROCESS analysis. LARS showed that the combination of baseline peer problems and IA in one step [ΔR² = 0.308, ΔF_{(2, 124)} = 27.59, p < 0.001), and VCI and PSI in a later step [ΔR² = 0.038, ΔF_{(2, 122)} = 3.54, p = 0.03] best explained variance in future peer problems, in total explaining 34.6% of variance [F_{(4, 122)} = 16.13, p < 0.001]. At the level of the individual variables, baseline peer problems, IA, and PSI (but not VCI) were significant predictors (Table 3).

The results from the LARS restricted the PROCESS analysis to include baseline peer problems as a covariate, IA as a direct predictor, and VCI and PSI as potential parallel mediators, with peer problems in preadolescence as the criterion variable. The PROCESS analysis showed a significant total effect for the baseline peer problems, IA, VCI, and PSI on peer problems in preadolescence [b = 0.155, p = 0.001, 95 CI (0.065, 0.245)], with a significant direct effect of IA [b = 0.132, p = 0.009, 95 CI (0.033, 0.232)]. Mediation analyses showed that PSI was a significant mediator of the effect [b = 0.053, 95 CI (0.013, 0.110)], whereas VCI was non-significant [b = −0.031, 95 CI (−0.013, 0.075)] (see Figure 1). In short, the relation between childhood IA and future peer problems was partly explained through its impact on PSI, such that future peer problems were best explained by high IA scores in conjunction with low PSI performance. Importantly, these results remained after adjusting for the presence of an ADHD diagnosis.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.