The phenomenology of attention-deficit/hyperactivity disorder (ADHD) has traditionally been emphasized in children, but there is increasing recognition of its relevance in adults. Epidemiological and intervention studies converge to suggest that adult ADHD is prevalent, impairing, and benefits from pharmacological and psychosocial interventions (Dodson 2005; Solanto et al. 2010). Parents are one subgroup of adults for whom ADHD may be particularly consequential. In the family context, parental ADHD is associated with elevated parenting- and family-related stress, as well as marital dissatisfaction and marital dissolution (Ersoy and Ersoy 2015; Johnston et al. 2012; Michielsen et al. 2013). Beyond these considerable impairments, an additional burden for parents with ADHD extends to potential consequences for offspring development. Reflecting genetically- and environmentally-mediated effects, parental ADHD is associated with poor youth outcomes (Minde et al. 2003). Humphreys et al. (2012) reported that parental ADHD was uniquely associated with a latent dimension of child psychopathology derived from ADHD, oppositional defiant disorder (ODD), and internalizing problems (e.g., anxiety, depression). Similarly, Agha et al. (2013) reported that parental ADHD was positively associated with severe offspring ADHD and CD symptoms. Thus, there is persuasive evidence that parental ADHD is linked to clinically significant problems.

To adequately test the predictive validity of ADHD in parents with respect to offspring outcomes, several important limitations must be addressed. First, parental ADHD frequently co-occurs with depression: 9–16 % of adults with depression were diagnosed with ADHD and an even larger proportion (up to 50 %) of adults with ADHD experience significant depression (Kessler et al. 2006; McIntosh et al. 2009). Maternal depression reliably predicts significant child externalizing problems (e.g., hyperactivity, conduct problems; Chronis et al. 2007; Gross et al. 2008; Luoma et al. 2001). Because comorbid parental psychopathology is particularly predictive of poor youth outcomes (Loeber et al. 2009), accurate predictive models must evaluate parental ADHD and depression simultaneously to discern unique associations.

Second, an important quasi-experimental test of a potential causal risk factor is testing its time-varying association with outcomes (Shadish et al. 2002). Thus, prospective change in parental ADHD predicting worse trajectories of child outcomes would suggest that parental ADHD is a potential independent causal risk factor. Despite their unique advantages, time-varying predictive models are infrequently employed in developmental psychopathology. A key exception is a study where yearly increases in youth CD symptoms uniquely predicted escalating ADHD, ODD, and depression symptoms, thus elucidating elegant temporal covariation among psychopathology dimensions (Lahey et al. 2002). Given that adult ADHD symptoms vary across time (Faraone et al. 2006; Hill and Schoener 1996; Todd et al. 2008), even within a 1-year period (Sitholey et al. 2010), parental ADHD symptoms may constitute a dynamic risk factor for offspring psychopathology. We contend that innovations in family-based interventions will follow from tests of dynamic associations between parental ADHD symptoms, with differentiated measures of offspring ADHD, conduct problems, and substance use. For example, directly targeting parental ADHD symptoms improves parental engagement and implementation of key ingredients of behavioral training interventions (Wang et al. 2014).

Although developmental psychopathology prioritizes understanding explanatory factors and processes underlying both typical and atypical development (Hinshaw 2002), the field continues to identify risk factors and their outcomes without commensurate effort to gain traction on risk processes (Rutter 2012). Even among the most well-studied interventions, in which mechanistic explanations of treatment outcomes are emphasized, there is persistent uncertainty over what mediates intervention effects (Kazdin 2007). Next, despite the important distinction between statistical (Baron and Kenny 1986) and causal mediation, as well as consensus that temporally-ordered predictors, mediators, and outcomes are necessary to infer causal mediation (Kraemer et al. 1997; MacKinnon 2008), these criteria are rarely met. In a cross-sectional study, paternal and maternal ADHD were each associated with dimensions of youth neuropsychological functioning (e.g., verbal IQ, working memory), but these neuropsychological dimensions did not significantly mediate associations between parental ADHD and offspring ADHD (Thissen et al. 2014). Overall, despite consistent evidence that parental ADHD is a risk factor for negative child outcomes, mediators of these associations remain elusive.

There is considerable evidence that parent-child interactions are highly conflictual in families of children with ADHD; moreover, negative interactions are evident early in development and persist through adolescence (Barkley 2003). Although multiple factors contribute to elevated dysfunction in families with ADHD, including parenting stress and parental psychopathology (Wiener et al. 2016), one compelling factor is expressed emotion (EE). EE consists of the broad, emotional climate within the family, including criticism and emotional over-involvement (EOI), and reflects the attitudes and emotions of other family members toward the proband (Hooley and Richters 1995). In the family context of ADHD, among school-age girls with and without ADHD, high EE from both mothers and fathers was each positively associated with categorical and dimensional measures of youth oppositional behavior, aggression, and particularly ADHD (Peris and Hinshaw 2003). Similarly, negative EE expressed towards sons was also associated with more severe youth ADHD symptoms (Psychogiou et al. 2007). However, both studies were cross-sectional and given that child-level factors predict parent factors (e.g., parenting) as much as, if not more than, the reverse (Burke et al. 2008), prospective designs are necessary to improve directional inferences. Recently, EE showed moderate cross-sectional associations with youth ADHD but there were no prospective associations in either direction (i.e., baseline EE with later youth psychopathology nor baseline psychopathology with later EE; Richards et al. 2014). However, in a longitudinal follow-up of youth with ADHD, latent classes of ADHD and ODD were sensitive to early EE factors (e.g., high criticism; Musser et al. 2016). Thus, it is unclear whether high EOI or criticism prospectively predicts child externalizing symptoms. Next, as reviewed by Johnston et al. (2012), deficits associated with ADHD may contribute to poor self-regulation in parents, including neurocognitive dysfunction (e.g., working memory; Deater-Deckard et al. 2009) and affect regulation (Dix 1991). By extension, disruptions in parental self-regulation, including emotional responsiveness secondary to parental ADHD (Sonuga-Barke et al. 2013), may critically mediate predictions of offspring psychopathology. Thus, whereas studies of parental EE have focused mostly on their prediction of child outcomes, we evaluated parental EE facets (i.e., criticism, EOI) as temporally-ordered mediators of predictions of youth ADHD and ODD from parental ADHD symptoms.

A second plausible mediator of emergent youth psychopathology is disrupted parenting behavior. Spanning diverse risk factors (e.g., socioeconomic disadvantage, parental alcohol problems, marital conflict, parental depression) and outcomes (e.g., alcohol problems, academic achievement), individual differences in positive and negative parenting behavior may represent a common mechanism through which risk factors predict offspring psychopathology. Controlled, family-focused intervention studies, for example, suggest that improved negative/ineffective parenting behavior partially mediated predictions of youth outcomes in the Multimodal Treatment Study of ADHD (Hinshaw et al. 2000). Similarly, improved observed parenting behavior secondary to a parental ADHD intervention significantly mediated predictions of offspring ADHD (Chronis-Tuscano et al. 2011). Thus, correlational, longitudinal, and experimental designs converge to suggest that parental ADHD symptoms are consistently associated with disrupted parenting behavior (Johnston et al. 2012). Tung et al. (2015) employed multiple mediation to test multi-dimensional (e.g., positive, negative) and multi-method (e.g., observed, self-reported) measures of parenting behavior as mediators of the inter-generational association of parental and offspring ADHD symptoms. Controlling for parental depression and baseline child ADHD and ODD, negative parenting uniquely mediated the cross-time association of parental and youth ADHD. Thus, mediation by individual differences in parenting behavior is plausible, although we note that extension to dimensions of child psychopathology beyond ADHD is also critically needed (i.e., ODD, CD).

There is increasing recognition of the importance of ADHD in parents, both with respect to adult functioning as well as implications for family interaction and youth psychopathology. However, existing designs and methods provide little clarity on whether parental ADHD symptoms constitute a causal risk factor for differentiated measures of offspring externalizing problems. Moreover, explanatory factors (i.e., mediators) underlying predictive models are even more elusive. To elucidate risk processes and to facilitate innovations in intervention and prevention for children with externalizing behavior, our goals were two-fold: (1) To test parental ADHD symptoms over 6 years (across three waves) as a time-varying predictor of 6-year change in youth ADHD symptoms as well as ODD, CD, and early substance use (controlling for youth ADHD and parental depression symptoms); (2) For significant predictive models, we conducted post-hoc tests of multiple mediation by EE (criticism, EOI) and then separately for negative and positive self-reported parenting behavior.

Methods

Participants

At baseline (i.e., Wave 1), 230 ethnically-diverse (55 % Caucasian; 7 % African American; 9 % Hispanic; 3 % Asian; 23 % mixed/other) 5–10-year-old youth (M = 7.4 years, SD = 1.1) with (n = 120) and without (n = 110) ADHD (68 % male) were recruited for a laboratory-based study in a large metropolitan city in the Western United States. Participants were recruited from mental health centers and pediatric offices, as well as through flyers posted in local elementary schools and other public locations. Inclusion criteria for participating youth consisted of English fluency, residing with at least one biological parent at least half of the time, and full-time enrollment in school. Exclusionary criteria for all youth were an IQ below 70, or a neurological, pervasive developmental, or seizure disorder. Youth ADHD proband status was based on a positive diagnosis from a fully structured DSM-IV interview with the parent (see below). Children who met diagnostic criteria for any disorder other than ADHD were placed into the non-ADHD group to avoid recruiting an improbably high functioning comparison sample that may have exaggerated diagnostic group differences. 86 %, 87 %, and 89 % of parents were mothers at Wave 1, Wave 2, and Wave 3, respectively. All families were recruited, screened, and assessed using identical procedures. There were no diagnostic group differences with respect to age, sex, race-ethnicity, or socioeconomic status among youth with and without ADHD.

2 years after Wave 1, all families were asked to participate in a laboratory-based follow-up visit (i.e., Wave 2). At Wave 2, n = 205 of the original 230 families (89 % retention) participated in the follow-up assessment (M = 10.2 years, SD = 1.3). Based on Wave 1 ADHD diagnostic status, the Wave 2 participants consisted of n = 111 (ADHD) and n = 94 (non-ADHD) youth. Relative to non-participants, families retained for Wave 2 were comparable across multiple clinical and demographic measures at Wave 1 (e.g., age, sex, race-ethnicity, parental ADHD symptoms). However, Wave 2 participating families had significantly higher levels of offspring ADHD symptoms relative to non-participating families, t(226) = −2.08, p = 0.04. The third follow-up assessment (i.e., Wave 3) occurred 2–3 years after Wave 2; once again, all families were invited to participate in this assessment. At Wave 3, n = 192 of the 205 Wave 2 participants were assessed (M age = 12.0 years, SD = 1.5). For Wave 3, there were no differences found between participants and non-participants with respect to key clinical and demographic measures (e.g., age, sex, race-ethnicity, parental ADHD symptoms). Clinical and demographic characteristics of the sample, at each wave, are described in Table 1.Footnote 1

Table 1 Participant characteristics
Full size table

Measures

Predictors: Parental ADHD and Depression Symptoms

At each of the three waves, parents self-reported their ADHD symptoms using the Adult ADHD Self-Report Scale (ASRS-v1.1; Kessler et al. 2005). The ASRS is an 18-item, self-report measure of ADHD severity. Rated on a five-point Likert scale (i.e., never, rarely, sometimes, often, very often), the 18 items were summed to provide a composite measure of adult ADHD symptom severity. Mean ASRS scores across waves appear in Table 1. Utilizing thresholds developed in the National Comorbidity Survey (2005), in which a score of 29 on the ASRS indicated the 90th percentile of ADHD symptoms, 37 %, 29 %, and 24 % of parents in our study met or exceeded this threshold at Wave 1, Wave 2, and Wave 3, respectively. Although the ASRS is not a diagnostic measure of adult ADHD and therefore does not provide a formal clinical cut-off score, the percentage of parents meeting the NCS criterion suggests that a high proportion of parents was experiencing significant levels of ADHD symptoms. The ASRS is a reliable and valid scale with excellent test-retest reliability and internal consistency (Kessler et al. 2005; Kessler et al. 2007). The Cronbach’s alpha was 0.94 at Wave 1, 0.99 at Wave 2, and 0.93 at Wave 3. Parents also self-reported their depression symptoms at Waves 1, 2, and 3 on the Beck Depression Inventory, First Edition (BDI-I; Beck et al. 1996). The 21 items are scored on a four-point Likert Scale of depression severity over the past two weeks. Given that depression is best represented dimensionally (Ruscio and Ruscio 2000; Slade and Andrews 2005), all items were summed for total depression (mean scores given in Table 1). The BDI-I is extensively-validated and psychometrically sound, with mean internal consistency estimates of .86 in psychiatric patients and .81 in non-psychiatric patients (Beck et al. 1988). The BDI also has good positive predictive power relative to current and past-year depression (Shean and Baldwin 2008). The BDI-I had excellent internal consistency in this study, with a Cronbach’s alpha of 0.85 at Wave 1, 0.86 at Wave 2, and 0.87 at Wave 3.

Wave 2 Mediators: Expressed Emotion and Parenting Behavior

At Wave 2, parental expressed emotion (EE) was assessed using the Five-Minute Speech Sample (FMSS), a brief instrument used to measure a person’s feelings about a family member and their perception of the quality of their relationship (Magaña et al. 1986). Parents were asked to speak, without interruption, for five minutes about what type of person their child is, their relationship with their child, and how they get along. Responses were recorded and coded to evaluate tone and content of their speech sample. The two global dimensions of EE, Criticism and Emotional Over-Involvement (EOI), were assessed separately and coded ordinally (low, borderline, high; 0–2). Parents were coded high on Criticism if any of the following were present: a negative initial statement (e.g., “my child is very irritating”), a negative parent-child relationship described (e.g., “my child and I do not get along”), or one or more critical comments about their child (i.e., harsh or angry statements about the child). Critical statements had to be extreme in tone and content, and parents who only expressed mild statements of dissatisfaction (e.g., frustration, being “bothered”) were coded as “borderline” on Criticism. Notably, purely descriptive statements that did not contain additional negative judgment (e.g., “he has trouble paying attention at school” or “she argues with adults a lot”) about the child were not coded as Criticism. If none of these criteria were present in a parent’s statement, the Criticism dimension was rated as Low. For the EOI dimension, parents were rated high if they engaged in any of the following: indicated extreme levels of self-sacrificing behavior (e.g., “I never do anything for myself because I need to help my child”), overprotective behavior (e.g., “I don’t let her leave the house because it is unsafe”), lack of objectivity (e.g., “she is failing all her classes, but that’s because her teacher is out to get her”), or engaged in extreme emotional display during the audio recording (i.e., breaks down crying). In addition, a high EOI score was coded if parents displayed at least two of the following: excessive detail about the past (i.e., descriptions of infancy), extreme statements of positive attitude (e.g., “he is the center of my universe”), or excessive praise (i.e., five or more positive remarks), and parents who only exhibited one of these criteria were coded as “borderline” on EOI. Parents whose statements did not exhibit any of these criteria were given a Low score for EOI. The second author (IT) and four bachelor-level coders were trained in coding the FMSS through weekly meetings until all coders met at least 80 % agreement with codes of an original Family Project Lab member and achieved an inter-rater reliability of ICC > .80. Coders met weekly as a team to resolve discrepancies. Each recording was coded separately by two raters, with strong reliability for Criticism (ICC = 0.77) and EOI (ICC = 0.83). The coding procedure was highly similar to that used in several recent studies of EE using the FMSS, including Musser et al. (2016); Peris and Baker (2000); and Peris and Hinshaw (2003).

At Wave 2, positive and negative parenting behaviors were measured using the Alabama Parenting Questionnaire (APQ; Shelton et al. 1996). The APQ is a 42-item self-report of parenting behavior; parents rate the frequency of each behavior on a 5-point Likert scale, ranging from 1 (never) to 5 (always). Identified negative factors include corporal punishment (e.g., “You spank your child with your hand when he/she has done something wrong”), inconsistent discipline (e.g., “You threaten to punish your child and then do not actually punish him/her”), and poor monitoring (e.g., “You get so busy that you forget where your child is and what he/she is doing”), and positive behaviors include involvement (e.g., “You help your child with his/her homework”) and positive reinforcement (“You praise your child if he/she behaves well”) (Chronis-Tuscano et al. 2008; Hawes and Dadds 2006). Following previous studies (Prevatt 2003), we examined a composite score of positive (i.e., involvement, positive reinforcement) and negative (i.e., corporal punishment, inconsistent discipline, poor monitoring) parenting behaviors as mediators between parental ADHD symptoms and child ADHD and ODD symptoms. For the positive parenting composite, the internal consistency was 0.83 and for the negative composite, the value was 0.65.

Youth Outcomes: ADHD, ODD, CD, and Alcohol/Substance Use

At Waves 1, 2, and 3, parents completed the Disruptive Behavior Disorder (DBD) Rating Scale, a 45-item DSM-IV rating scale of ADHD, ODD, and CD symptoms (Pelham et al. 1992). Behaviors were rated on a four-point scale, ranging from 0 (not at all) to 3 (very much) and a total score was calculated separately for ADHD, ODD, and CD symptoms. The DBD has excellent psychometric properties, including superior predictive validity (Lahey et al. 2004). For this sample, the Cronbach alphas were 0.96 for ADHD, 0.89 for ODD, and 0.67 for CD at Wave 1; 0.96 for ADHD, 0.86 for ODD, and 0.59 for CD at Wave 2; and 0.95 for ADHD, 0.88 for ODD, and 0.55 for CD at Wave 3.

At Waves 2 and 3, we administered the Self-Report of Antisocial Behavior Scale (SRA), an interview assessment of youth antisocial behavior and substance use in the previous six months (Loeber et al. 1989). Youth were asked about their participation in the behavior, with responses for each item ranging from 0 (never), 1 (once), 2 (twice), or 3 (more often). We calculated the average score of the five items that evaluated use of beer, wine, liquor, tobacco, and marijuana use. Thus, scores represented the average frequency of alcohol/substance use in the past 6 months. The SRA demonstrated validity in community samples (Loeber et al. 1989) and was sensitive to early individual differences in psychopathic traits (Jezior et al. 2015).

Procedures

At Wave 1, study eligibility was based on a telephone screening with the parent. Eligible families were invited for a laboratory-based assessment and they were mailed rating scales to be completed prior to the assessment. All families completed identical procedures. After parental consent and youth assent was obtained, parents and children were interviewed simultaneously in different rooms by separate interviewers. All interviewers were well-trained clinical psychology Ph.D. students or well-qualified B.A.-level staff. Assessment methods included structured interviews, parent self-report measures of psychopathology and parenting, as well as rating scales of child psychopathology. Parents were asked to report on their child’s unmedicated behavior. The Wave 2 and Wave 3 assessments consisted of highly parallel procedures to Wave 1, including laboratory-based assessments of parent and child psychopathology, family functioning, and peer relationships; we also expanded into developmentally-appropriate domains, such as delinquency and substance use. The IRB approved all study procedures.

Data Analytic Procedures

To evaluate parental ADHD symptoms as a dynamic predictor (i.e., change from Waves 1–3) of prospective change in child ADHD, ODD, and CD symptoms (Waves 1–3), as well as alcohol/substance use (Wave 2–3 only), we employed Generalized Estimating Equations (GEE) in Stata (Version 13.1). GEE is an extension of the general linear model, but uses robust variance estimation in a repeated measures design (Hanley et al. 2003). GEE minimizes Type I error and increases statistical power by adjusting for correlated observations across waves. We specified a Poisson distribution and an exchangeable working correlation matrix; all tests were based on the z-statistic and B parameters are in logs. Thus, parental ADHD and depression symptoms were treated as independent, time-varying predictors (Hardin and Hilbe 2003). To enhance specificity, we included youth sex, race-ethnicity, and age as covariates in all models; predictions of ODD and CD symptoms, as well as alcohol/substance use also controlled for Wave 1 ADHD diagnostic status (i.e., ADHD versus non-ADHD control). GEE accommodates missing data using the “all available pairs” method, in which all non-missing pairs of data are used in estimating the working correlation parameters. Thus, only the observation for that subject is missing (not all variables for that subject).

Based on significant predictive models outlined above, our second goal was to examine Wave 2 positive and negative parenting behavior and Wave 2 EE as separate mediators of significant predictions of Wave 3 externalizing problems from Wave 1 parental ADHD symptoms. We employed multiple mediation to discern collective and individual indirect effects by separable parenting behavior (i.e., positive, negative); we then reproduced the same model but tested EE facets (i.e., criticism, EOI). Multiple mediation with bootstrapping is a powerful, nonparametric resampling procedure that evaluates multiple individual mediators simultaneously and adjusts for potential covariates (MacKinnon et al. 2000; Preacher and Hayes 2008). Unlike traditional mediation, evidence of causal mediation with bootstrapping does not require a significant direct effect of the predictor on the outcome (MacKinnon et al. 2000; Preacher and Hayes 2008; Zhao et al. 2010). In addition to evaluating the total mediation effect in the model, multiple mediation controls for intercorrelation among the mediators, thus discerning the unique role of each individual mediator. Finally, bootstrapping-based multiple mediation is more powerful than traditional mediation methods (i.e., Sobel test; Zhao et al. 2010). We conducted mediation analyses using the PROCESS macro in SPSS 23.0. Parameter estimates and 95 % bias-corrected and accelerated confidence intervals for total and specific indirect effects were generated based on 5000 bootstrap simulation samples (Preacher and Hayes 2008).

Results

Prospective Change in Youth ADHD, ODD, CD, and Alcohol/Substance Use: Dynamic Prediction by Parental ADHD Symptoms

To review, we examined six-year change (across three waves) in parental ADHD symptoms as a time-varying predictor of 6-year change in youth ADHD, ODD, and CD symptoms. Predictions of change in early alcohol/substance use spanned Wave 2 to Wave 3 only. First, correlations among predictors, covariates, and outcomes appear in Table 2. Next, controlling for child age, sex, and race-ethnicity, change in parental ADHD symptoms was significantly and positively associated with change in offspring ADHD symptoms (β = 0.01, z = 2.81, p < 0.01) across the time period, but change in parental depression symptoms was not (β < 0.01, z = 0.01, p = 0.99).Footnote 2 We then constructed a nearly identical model in predictions of ODD symptoms, but included ADHD symptoms as a time-varying covariate given its relevance to early conduct problems (Hinshaw et al. 1993). As expected, change in child ADHD symptoms positively predicted change in ODD symptoms (β = 0.08, z = 11.00, p < 0.001). As with predictions of child ADHD symptoms, controlling for child age, sex, race-ethnicity, change in parental ADHD symptoms significantly and positively predicted change in ODD symptoms (β = 0.01, z = 2.82, p < 0.01) across the three waves, whereas change in parental depression symptoms did not (β = 0.01, z = 1.46, p = 0.14). Model summaries appear in Table 3.

Table 2 Correlations among study variables
Full size table
Table 3 Time-varying predictions of child ADHD and ODD symptoms from parental ADHD symptoms
Full size table

With respect to predictions of 6-year change in CD symptoms, we reproduced the identical model employed above in predictions of ODD symptoms. Expectedly, escalating youth ADHD symptoms positively predicted escalating youth CD symptoms (β = 0.12, z = 9.44, p < 0.01); however, neither change in parental ADHD symptoms (β < 0.01, z = 1.33, p = 0.19) nor change in parental depression symptoms (β = 0.02, z = 1.66, p = .10) were significant time-varying predictors of change in CD symptoms. Similarly, controlling for child age, sex, race-ethnicity, and change in youth ADHD symptoms, change in parental ADHD symptoms (β = −0.02, z = −1.30, p = 0.20) and change in parental depression symptoms (β = .04, z = 1.27, p = 0.20) were each unrelated to change in youth self-reported alcohol/substance from Wave 2 to Wave 3.

Predictions of Wave 3 ADHD and ODD Symptoms from Wave 1 Parental ADHD Symptoms: Mediation by Wave 2 Parenting Behavior and Expressed Emotion

To elucidate explanatory mechanisms underlying the significant predictions of youth ADHD and ODD symptoms from parental ADHD symptoms described above, we tested positive and negative parenting behavior as mediators. All mediation models controlled for child sex, age, race, and Wave 1 parental depression symptoms. Next, each multiple mediation model estimated the following parameters (Preacher and Hayes 2008): (a) the total effect of Wave 1 parental ADHD symptoms on Wave 3 child ADHD symptoms (i.e., excluding the mediators), (b) specific effect of Wave 1 parental ADHD symptoms on each Wave 2 parenting behavior mediator, (c) specific effects of each Wave 2 mediator on Wave 3 child ADHD, and (d) the direct effect of Wave 1 parental ADHD symptoms with respect to Wave 3 child ADHD through each proposed mediator. This model was then reproduced with EE factors (i.e., Criticism, EOI) as simultaneous and independent mediators. Finally, controlling for Wave 1 ADHD symptoms, these two mediation models were reproduced with Wave 3 ODD symptoms as the outcome.

First, results based on positive and negative parenting behavior in predictions of offspring ADHD symptoms from parental ADHD symptoms appear in Fig. 1A. Controlling for child sex, age, race, and Wave 1 parental depression symptoms, there was a significant total effect of Wave 1 parental ADHD symptoms on Wave 3 child ADHD symptoms. Wave 1 parental ADHD symptoms inversely predicted Wave 2 positive parenting and positively predicted Wave 2 negative parenting. However, negative parenting was positively associated with Wave 3 child ADHD symptoms; there was no association with positive parenting. Next, controlling for all covariates and mediators, the direct effect between parental ADHD symptoms and child ADHD symptoms remained significant. Total and specific indirect effects of parental ADHD symptoms and child ADHD through positive and negative parenting behaviors are presented in Table 4. Although the total indirect effect (i.e., difference between the total effect and direct effect through all the mediators) was non-significant, a significant indirect effect of negative parenting was observed: the prediction of Wave 3 offspring ADHD symptoms from Wave 1 parental ADHD symptoms was significantly and independently mediated by Wave 2 negative parenting.

Fig. 1
figure 1

Multiple mediation: Prediction of child ADHD symptoms from parental ADHD symptoms, through (a) positive and negative parenting behavior, and (b) expressed emotion measures

Full size image
Table 4 Multiple mediation by parenting behavior and expressed emotion on parental ADHD symptoms and child ADHD and ODD symptoms
Full size table

Next, we examined Wave 2 EE facets (Criticism, EOI) as mediators of the association of Wave 1 parental ADHD symptoms on Wave 3 offspring ADHD symptoms (Fig. 1B). Again, there was a significant total effect of parental ADHD symptoms on child ADHD symptoms, although parental ADHD symptoms were unrelated to Wave 2 Criticism and EOI; EOI was similarly unrelated to Wave 3 child ADHD symptoms. However, Wave 2 Criticism uniquely predicted Wave 3 child ADHD symptoms. Controlling for all covariates and EE facets as mediators, the direct effect of parental ADHD symptoms on child ADHD remained significant. As shown in Table 4, all total and specific indirect effects for this model were non-significant, indicating that EE facets neither independently nor cumulatively mediated the association of parental and offspring ADHD symptoms.

Following these analyses, we reproduced these same meditational models (i.e., parenting behavior, EE) with Wave 3 child ODD symptoms as the outcome. First, controlling for Wave 1 child ADHD symptoms and all other covariates, we examined positive and negative parenting behavior as mediators of the association between parental ADHD symptoms and offspring ODD symptoms (Fig. 2A). There was no significant initial total effect of Wave 1 parental ADHD symptoms on Wave 3 child ODD symptoms. Parental ADHD symptoms were significantly and positively associated with Wave 2 negative parenting behaviors, which positively predicted Wave 3 child ODD symptoms. In contrast, parental ADHD symptoms were unrelated to positive parenting behavior, which was similarly unrelated to child ODD symptoms. Controlling for all covariates and mediators, there was no significant direct effect of Wave 1 parental ADHD symptoms on Wave 3 ODD symptoms. Although the total indirect effect was not significant, Wave 2 negative parenting marginally mediated predictions of Wave 3 ODD symptoms from Wave 1 parental ADHD symptoms, above and beyond parental depression symptoms and offspring ADHD symptoms. In contrast, there was no significant indirect effect of positive parenting behavior predicting Wave 3 ODD symptoms from parental ADHD symptoms.

Fig. 2
figure 2

Multiple mediation: Prediction of child ODD symptoms from parent ADHD symptoms through (a) positive and negative parenting behavior and (b) expressed emotion

Full size image

Finally, we examined whether EE facets mediated the association of parental ADHD symptoms with offspring ODD symptoms (Fig. 2B). Wave 1 parental ADHD symptoms were unrelated to all Wave 2 EE facets. Wave 2 EOI at Wave 2 did not predict Wave 3 child ODD symptoms, but Wave 2 Criticism positively predicted Wave 3 ODD symptoms, controlling for all covariates and other facets of EE. All total and specific indirect effects for this model were non-significant, indicating that EE facets did not independently or cumulatively mediate the association of parental ADHD symptoms and offspring ODD symptoms (Table 4).

Discussion

To improve the empirical basis for parental ADHD symptoms as a potential causal influence on offspring externalizing problems, we prospectively followed an ethnically diverse sample of children with and without ADHD for 6 years across three waves. Controlling for demographic factors and parental depression symptoms, we examined parental ADHD symptoms as a time-varying predictor of prospective change in youth ADHD, ODD, and CD symptoms, as well as early alcohol/substance use. Our second goal was to test positive and negative parenting behavior, as well as EE facets (i.e., criticism, EOI), as temporally-ordered mediators underlying these predictions. Several important patterns were observed: (1) 6-year change in parental ADHD symptoms was unrelated to change in CD symptoms and youth self-reported alcohol/substance use, but change in parental ADHD symptoms uniquely predicted more persistent youth ADHD and ODD symptoms, even with control of parental depression symptoms and youth ADHD symptoms (for ODD models only). (2) Although EE facets did not mediate predictions of Wave 3 youth ADHD and ODD symptoms from Wave 1 parental ADHD symptoms, Wave 2 self-reported negative parenting behavior uniquely mediated predictions of ADHD symptoms, beyond parental depression symptoms and positive parenting behavior; negative parenting behavior also marginally mediated predictions of youth ODD symptoms, with identical control of parental depression symptoms and positive parenting behavior, as well as youth ADHD symptoms.

These preliminary, quasi-experimental data suggest that parental ADHD symptoms are a causal risk factor for change in youth ADHD and ODD from childhood to early adolescence. This pattern is consistent with evidence that parental ADHD is associated with offspring externalizing problems, including ODD and CD symptoms (Agha et al. 2013; Chronis-Tuscano et al. 2011; Ellis and Nigg 2009). In the current study, parental ADHD symptoms were unrelated to prospective change in CD symptoms and alcohol/substance use, however. Several key processes may be relevant herein: reflecting potentially important differences based on cognitive factors and executive functioning (Kamradt et al. 2014; Mostert et al. 2015), adult ADHD is clinically and etiologically heterogeneous. Thus, parental ADHD may include distinct subgroups that are divergently associated with child outcomes (e.g., ADHD but not for CD and alcohol/substance use). Second, individual differences in CD symptoms and alcohol/substance use are highly sensitive to developmental influences, including pubertal onset, as well as social (e.g., peer) and neural (e.g., reward sensitivity) changes (Bjork and Pardini 2015; Dandreaux and Frick 2009). Because early onset alcohol/substance use (i.e., prior to age 15) is empirically distinct from later onset alcohol/substance use and is likely causally related to poor outcomes (Odgers et al. 2008), the relatively young age of the sample at Wave 3 (M = 11.96 years) may have yielded insufficient variation in these domains. Continued follow-up through adolescence, as well as careful consideration of sampling (e.g., avoiding age-censored samples) in future studies, will be crucial to adequately characterize parental ADHD symptoms and their prediction of delinquency and emergent alcohol/substance use.

To improve traction on how risk factors produce negative outcomes, explanatory mechanisms must be identified. Given that the family context of ADHD is often conflictual (Barkley 2003; Ersoy and Ersoy 2015; Michielsen et al. 2013), we explored two family-based mediators of change in child externalizing problems: EE and positive/negative parenting behavior. First, with control of parental depression symptoms, Wave 1 parental ADHD symptoms were consistently unrelated to each facet of Wave 2 EE (i.e., criticism, EOI). In the most authoritative study to date on parental ADHD and EE, parental psychopathology was unrelated to EE facets, although the latter was concurrently associated with youth psychopathology (Richards et al. 2014). One contributing factor may be the instability of EE over time, as replicated in studies of children with disruptive behavior problems (Daley et al. 2003; Richards et al. 2014). For example, in a small community sample, the stability of EE from kindergarten to first grade was modest (Peris and Baker 2000). Although EE facets did not mediate predictions from parental ADHD symptoms in the current study, Wave 2 Criticism uniquely predicted Wave 3 child ADHD and ODD symptoms, controlling for EOI as well as parental ADHD and depression symptoms. This is consistent with Peris and Baker (2000) who found that EE criticism was the only consistent predictor of change in externalizing behavior. Thus, although considerable work remains on understanding the determinants of individual differences in EE, and perhaps its mediation of different risk factors and outcomes, predictions from EE reinforce its importance in the family and its potential role in shaping offspring development.

Consistent with prevailing theory and evidence, negative parenting behavior significantly mediated predictions of offspring ADHD symptoms (and marginally for offspring ODD symptoms) from parental ADHD symptoms. Among school-age children, controlling for pre-treatment ADHD, maternal ADHD symptoms predicted post-treatment child ODD and CD symptoms; this association was mediated by change in observed negative parenting (Chronis-Tuscano et al. 2011). Similarly, negative parenting behavior partially mediated the intergenerational association of parental and offspring ADHD (Tung et al. 2015). Given that deficits central to adult ADHD, including cognitive biases, executive dysfunction, and emotional responsiveness, may interfere with effective parenting behavior (Johnston et al. 2012), mediation by deficient parenting is plausible. For example, deficient emotion regulation significantly mediated the association of maternal ADHD and harsh parenting behavior, controlling for ADHD and disruptive behavior (Mazursky-Horowitz et al. 2015). A primary implication of negative parenting behavior as a causal mediator is that current interventions may need to assess and/or remediate parental ADHD symptoms. Despite consensus that behavioral interventions are efficacious in the treatment of youth ADHD (Fabiano et al. 2009), undiagnosed and untreated parental ADHD may negatively affect implementation of these strategies. Modular sessions targeting parental ADHD may improve the effectiveness of existing parenting-based interventions (Jans et al. 2015; Wang et al. 2014); alternatively, untreated parental ADHD may hinder parenting interventions, with potentially significant consequences for offspring development.

Despite the three-wave, 6-year prospective follow-up design, consisting of innovative, temporally-ordered mediational constructs, there are critical limitations to the study that should be considered. First, although parental depression often co-occurs with parental ADHD, other dimensions of psychopathology are also relevant, including anxiety and alcohol/substance use disorders (Chronis et al. 2003; Margari et al. 2013). Second, several key constructs, including parenting behavior and childhood ADHD, ODD, and CD symptoms, were assessed via parent rating scales; additional methods (e.g., observational, structured interviews) and informants (e.g., teachers) would reduce concerns over potential biases (e.g., shared method variance), especially with self-report data (Podsakoff et al. 2003). Third, our hypothesized family-level mediators do not discount the role of other plausible family factors (e.g., marital discord, stressful life experiences). Next, although statistically significant, mediation by negative parenting behavior was a modestly-sized effect, underscoring the need to identify other mediating factors. Additionally, internal consistency values for the CD items on the DBD in our sample were quite low; alternative measures of CD or broader constructs, such as delinquency, may have been more sensitive to these early predictors. The internal consistency value for the APQ negative parenting composite score was also modest; however, this may reflect the small number of items (three) used in the Corporal Punishment subscale. Also, this study was under-powered to investigate child sex differences in the effect of parental ADHD symptoms on child ADHD symptoms; however, this is an important question for future work to examine. Lastly, because this sample consisted mostly of mothers, and given important differences between mothers and fathers (Lewis and Lamb 2003; Psychogiou et al. 2010), we await future work to highlight the important role of fathers in the development of children with and without ADHD.

We observed persuasive evidence that prospective change in offspring ADHD and ODD symptoms were significantly predicted by variation in parental ADHD symptoms across a 6-year period, even with control of child age, child sex, and child race-ethnicity. Moreover, these effects were partially mediated by negative parenting behavior, beyond the contribution of EE facets and positive parenting behavior. To catalyze innovations in intervention and prevention, experimental designs and novel sampling strategies (e.g., fathers with ADHD) will be necessary to improve traction on the dynamic covariation of parental psychopathology and offspring development.