Pediatric Assessment of Research Domain Criteria Positive and Negative Valence Systems: Partial Genetic Mediation of Links to Problem Behaviors

Abstract

We use the highly structured Laboratory-Temperament Assessment Battery to measure behaviors that map onto the RDoC positive and negative valence systems. Using a birth record-based sample (N = 1374 individual twins; mean age 7.7 years), we created composites of observed behavior reflecting the RDoC constructs Reward Responsiveness, Frustrative Nonreward, Loss, and Fear. Next, we related the RDoC constructs concurrently and longitudinally to problem behaviors, measured using parent-report on the Health Behavior Questionnaire and symptom counts from the DISC-IV (reflecting DSM-IV). The four pediatric RDoC positive and negative valence system measures, especially Reward Responsiveness, Frustrative Nonreward, and Loss, were heritable and modestly but plausibly related to traditional DSM-based measures in a transdiagnostic manner. The modest predictions from RDoC measures to DSM-based measures were largely genetically mediated, although relationships with aggressive and oppositional behaviors were also influenced by common environmental factors

General Scientific Summary:

Many of the proposed behaviors within the RDoC matrix are not appropriate for use with children. This paper recommends the Laboratory-Temperament Assessment Battery as an observational measure of selected RDoC positive and negative valence system constructs. These RDoC constructs are modestly associated with symptom measures, both concurrently and longitudinally. Genetic differences play a role in both variability in RDoC measures and in the prediction from RDoC measures to symptoms.

The Research Domain Criteria (RDoC) identify six overarching domains of behavior within a matrix approach in which rows correspond to constructs and columns represent different units of analysis (Cuthbert, 2014). As research evolves, each cell of the matrix will identify construct-relevant units (e.g., neurological regions of interest) and measurement paradigms. Despite the early age of onset for many psychiatric conditions, RDoC contains few explicit recommendations for behavioral measures for children. However, the need for a child-focused perspective for RDoC is apparent (Casey et al., 2014; Franklin et al., 2015; Mittal & Wakschlag, 2017). Here, we examine the utility of a widely used child temperament assessment tool as a pediatric behavioral measure of the RDoC positive and negative valence systems. This utility is evaluated by examining links to concurrent and longitudinal behavioral symptoms as well as genetic mediation of those links.

Assessing RDoC-Relevant Behaviors in Pediatric Populations: Lab-TAB

Overview.

The Laboratory Temperament Assessment Battery (Lab-TAB) is one of the most commonly used behavioral assessments of childhood emotional behaviors (Planalp et al., 2016). Lab-TAB comprises standardized episodes designed to elicit specified behaviors in the presence of relatively realistic emotional incentives; coding and scoring schemes may be altered to address different research questions. Lab-TAB is an adaptable assessment tool grounded in observational research and the dimensions of individual differences, aligning with the RDoC framework. Lab-TAB taps the positive and negative valence systems, and indeed has been used in prior RDoC-inspired research with a developmental focus (Kessel et al., 2017).

Positive valence constructs.

Under RDoC, positive valence systems are indicated by responses to positive or reinforcing contexts. Here, we focus on reward responsiveness, which may correspond to temperamental measures of anticipatory positive affect, reward sensitivity, and/or approach motivation. Specifically, we focus on the subconstructs of reward anticipation and initial reward responsiveness (NIMH “Construct: Reward Anticipation” and “Construct: Initial Response to Reward,” 2020). The overarching reward responsiveness construct is defined primarily in terms of neural activity and the current RDoC matrix offers no behavioral measures for reward anticipation and only taste reactivity for initial response to reward.

Negative valence constructs.

We examine three of the RDoC negative valence systems, associated with responses to potentially or actually aversive situations. Frustrative Nonreward is reactivity to withdrawal or prevention of reward (NIMH “Construct: Frustrative Nonreward,” 2020) and is similar to temperament constructs such as anger proneness and distress to limitations. The Behavioral Assessment Methods for RDoC Constructs report (National Advisory Mental Health Council Workgroup on Tasks and Measures for Research Domain Criteria, 2016) lists developmentally appropriate behavioral measures for Frustrative Nonreward, including the Lab-TAB episodes Box Empty and Transparent Box. Loss involves acute or prolonged deprivation of an important person, object, or situation (NIMH “Construct: Loss,” 2020); it bears similarity to sad affect. Fear entails a pattern of responses aimed at defending the individual from perceived danger (NIMH “Construct: Acute Threat [‘Fear’], 2020). Suggested behavioral measures include Stranger Tests in childhood and the Trier Social Stress Test in adolescence; the latter shares similarities to the Lab-TAB episode Storytelling.

RDoC and Behavior Problems

Although explicitly designed to move away from traditional diagnostic categories RDoC constructs should nonetheless have clinical relevance, given the aim of creating a valid psychiatric classification system and ultimately more effective mental health interventions. A number of recent reviews have suggested potential linkages between RDoC constructs and clinical symptoms, both internalizing (Dillon et al., 2014; Olino, 2016; Woody & Gibb, 2015)and externalizing (Fonagy & Luyten, 2018; Meyers et al., 2017). Developmental psychopathology approaches are essential for advancing understanding of these relationships in a manner that informs understanding of etiology as well as early intervention efforts (Franklin et al., 2015).

RDoC and Genetics

Genetic studies explicitly focused on pediatric RDoC constructs have not yet appeared in the literature although others besides ours are in progress (Carneyet al., 2016; Cecilione et al., 2018). However, constructs that are part of RDoC have been studied extensively, including genetic investigations. For instance, genetic research on frustrative nonreward falls under broader concepts such as anger proneness in children. In a longitudinally followed sample of twin infants and toddlers, we found that Lab-TAB measures of anger expression were only modestly heritable at both ages, and the association of anger expression and inhibitory control (a regulatory feature) was mediated by environmental factors in toddlerhood (Gagne & Goldsmith, 2011). Most of the behavior-genetic research relevant to positive and negative valence systems falls under the rubric of temperament research. Clifford, Lemery-Chalfant, and Goldsmith (2015) review approximately 20 recent reports of heritability and gene-environment interplay for aspects of childhood temperament. They note the imbalance of few studies of positive affectivity versus substantial literature on the genetics of negative affectivity.

Study Aims

Our primary goal was to use Lab-TAB behavioral paradigms to derive constructs that fit the RDoC positive and negative valence systems, to contribute behaviors and paradigms appropriate to children to the RDoC matrix. Using multiple measures of each construct, we developed RDoC behavioral composites in a large, birth-record based twin sample. Second, we examined concurrent and longitudinal associations of the RDoC measures with parent-reported problem behaviors and psychiatric symptoms. Although RDoC was designed to address research limitations inherent in the DSM, we include measures of both behavior problems and DSM symptoms to help establish the relevance of RDoC constructs to our existing psychiatric nosology as well as to extant research on risk and resilience. Lastly, we leverage our twin sample for an examination of genetic influences on RDoC measures and on their association with behavior problems.

Methods

Participants

Participants were enrolled in the Wisconsin Twin Project, a set of longitudinal studies of temperament and psychopathology. Families were recruited from Wisconsin birth records (1989–2004) via letter; see detailed information on the design and sample in Schmidt et al. (2019). Protocols were approved by a University of Wisconsin–Madison IRB (protocols 2012-1145, 2012-0810, 2013-0961, 2014-1443, and 2014-1333), and parents completed consent forms. Families were paid for participation.

We used the middle childhood twin sample (N = 1374 [687 twin pairs]; mean age 7.75 years, SD = 8.27 months; 48.5% female). 82.1% of mothers identified their twins as White (3.3% Hispanic/Latino), 1.0% as Black, and 2.8% as other or >1 race. Median annual family income was $60,001 to $70,000. The most common education attainment for mothers was a college degree (28.2%), followed by tech/trade/some college (17.7%) and a high school degree (16.3%). For fathers, a high school degree (29.5%) was most frequent, followed by a college degree (23.2%) and trade/tech/some college (16.2%). 85.4% of mothers reported they were married to and living with the twins’ biological father at the time of assessment. 94.3% of mothers identified as the twins’ primary caregiver. The sample demographics are broadly representative of the region, and we recognize the limitations of a community-based, largely White and middle-class sample.

Longitudinal analyses used a subset of twins who also participated in an adolescent phase (N = 812 [406 twin pairs]; mean age 13.44 years [SD = 18.68 months]; 52.5% female). At the adolescent phase, families had higher incomes (median: $70,001 to $80,000 annually) and mothers and fathers both reported more years of education. Children who participated at both ages were less likely to be Hispanic/Latino than children in the subset of families who only participated at age 7. They also had moderately lower Loss (Mean=.−.05 vs. Mean=.06; t(1370) = 4.46, p<.001) and Frustrative Nonreward (Mean=−.06 vs. Mean=.08; t(1370) = 4.69, p<.001), and higher Fear (Mean=.16 vs. Mean=−.25; t(1370) = −13.74, p<.001).

Procedures

At mean twin age 7 years, mothers completed a child behavioral screen using the MacArthur Health and Behavior Questionnaire (HBQ; Armstrong et al., 2003). Families were selected for additional assessment if one twin scored >1.5 SD on any symptom scale or low on all symptom scales. We created RDoC composites using coded behaviors from video recordings during middle childhood home visits; inclusion in data analyses required participation in the home visit at age 7. Of the resulting 1374 twins in 687 families, 14 families were excluded due to physical, developmental, or intellectual disorders. The four-hour, in-home assessment involved additional parent questionnaires and interviews, child interviews, observer ratings, and the Lab-TAB. Data collection for this middle childhood phase took place across more than five years. Adolescent data were collected from families who were re-contacted at twin mean age 13 years and invited to participate in another set of telephone-based questionnaires and home visits.

Measures

Laboratory Temperament Assessment Battery.

Lab-TAB (Goldsmith et al., 1993) is a laboratory-based behavioral assessment that comprises multiple episodes designed to tap observable elements of temperament dimensions. Lab-TAB episodes are specifically recommended in the RDoC matrix as measures of Frustrative Nonreward, and other Lab-TAB episodes contain the affective incentives described for other RDoC constructs. Lab-TAB was administered during the childhood home visit and was modified slightly for use in homes (Gagne, Van Hulle, Aksan, Essex, & Goldsmith, 2011). During Lab-TAB administration, children’s behavior was videotaped and later coded by individuals blind to other information about the child. Individual raters did not rate both twins from the same family. 10% of the videos were rated by a master coder, and agreement between master coder and the other coder (Kappa ≥ .70) was required.

We identified response-level parameters, attempting to maximize concordance between selected variables and RDoC construct definitions. Each Lab-TAB episode (3–10 minutes duration), provided multiple responses scored in 5–30 second epochs or in discrete trials. Parameters included latency to first response, occurrence of a target response within an epoch or trial (mean response), and the magnitude or intensity of a target response (peak response). In general, positivity was coded as absence/presence (0/1) while facial, bodily, and vocal angry, sad, and fearful responses were coded on a 0–2 or 0–3 scale.

For detailed descriptions of each episode and of scoring procedures see the Lab-TAB manual (Goldsmith et al., 2010); here, we offer brief descriptions of selected episodes. In Balloon Bop, the child tester and child try to keep a balloon in the air. During Free Play, the child is provided four toys and may engage in self-directed play. For Hungry Hippos, the child tester and child play the children’s game three times, with the examiner losing each time. In Impossibly Perfect Stars, the child is asked to draw a star while the child tester offers critiques of each effort. In Not Sharing, the child tester divides candy between herself and the child in gradually more unequal fashion. During Scary Mask, one of the child testers enters the room wearing a scary mask, first in silence and then interacting with the child. In Storytelling, the child stands in front of multiple child testers and is asked to talk about what they did the prior day, with least one prompt given by the child tester. In Transparent Box, the child is encouraged to open a locked transparent box filled with several toys but is given a set of keys that cannot open the box. For Wrong Gift, the child is promised a gift he selected but is given the gift he previously ranked as his least favorite.

Post-visit observer ratings.

Two child testers from each middle childhood home visit independently completed post-visit ratings for each twin on 28 items related to child behavior, where “1” indicates the absence of the characteristic or behavior and “5” describes an extreme reaction. Behavior was observed throughout the visit, including times before, between, and after administration of Lab-TAB episodes. Some items include modified content from the Behavior Rating Scales (BRS) from the Bayley Scales of Infant Development (Bayley, 1969). Child tester ratings were averaged for each item; item-specific correlations between raters ranged from .38 to .49.

MacArthur Health and Behavior Questionnaire.

We used relevant symptom scales of the MacArthur Health & Behavior Questionnaire (HBQ; Armstrong, Goldstein, & The MacArthur Working Group on Outcome Assessment, 2003; Essex et al., 2002). The adolescent version of the HBQ overlaps substantially with the child version, with age appropriate items added and inappropriate items dropped.

Parents rated their child’s behavior over the past six months using a 3-point scale (0 = rarely, 2 = certainly applies). Internal consistency reliability (alpha) for age 7 HBQ subscales ranged from .67 to .84 for mother-report and from .62 to .85 for father-report. Internal consistency estimates for age 13 HBQ subscales ranged from .80 to .88 for mother-report and from .79 to .89 for father-report. Mother and father scores were moderately and significantly correlated (age 7 rs ranged from .27 to .53, all ps < .001; age 13 rs ranged from .31 to .56, all ps < .001) and were mean-averaged into a single parent-report score at each age.

Diagnostic Interview Schedule for Children – IV.

Primary caregivers completed the Diagnostic Interview Schedule for Children, Version IV (DISC-IV), a computer-based diagnostic instrument with moderate to good diagnostic reliability (Shaffer et al., 2000). At age 7 assessment, 59.9% of children with DISC-IV data did not meet diagnostic criteria for any disorder, 28.1% met criteria for only one disorder, and 12.0% met criteria for multiple disorders. At age 13, 76.4% of participants with DISC-IV data did not meet diagnostic criteria for any disorder, 17.9% met criteria for only one disorder, and 5.8% met criteria for multiple disorders. The most common diagnoses at age 7 and at age 13 were specific phobia (20.9% and 13%, respectively), followed by oppositional defiant (10.0% and 5.9%) and ADHD (8.1% and 5.5%).

Zygosity.

Primary caregivers completed of the Zygosity Questionnaire for Young Twins (Goldsmith, 1991). If necessary, we also used information from hospital placenta(e) reports; follow-up questionnaires completed by parents, twins, and research assistants; genotyping; and/or photographs and video. Zygosity could not be unambiguously determined for 27 pairs who were not genotyped. Of those with zygosity scores at age 7, 38.0% of twin pairs were monozygotic (MZ), 29.0% same-sex dizygotic (ssDZ), and 29.1% opposite-sex dizygotic (osDZ).

Socioeconomic status.

Family socioeconomic status (SES) was indexed at both waves of data collection using a standardized composite score of mother years of education, father years of education, and annual family income category. Higher values indicate higher SES.

Data Analysis Plan

Behavior composite formation.

Our analytic approach follows the procedures of previous Lab-TAB research (Gagne et al., 2012; Planalp et al., 2017). We adapted our analysis of the raw behavioral coding to explicitly tap RDoC constructs by examining positive and negative responses across episodes regardless of target affective response. In other words, rather than including presence of smiling only from designated Lab-TAB exuberance episodes, we incorporated positivity to potential reward from all pertinent episodes. This approach has been used (Dyson et al., 2012), although conceptualizing behaviors displayed across Lab-TAB episodes with different affective incentives as manifestations of the RDoC positive and negative valence systems represents an extension of Lab-TAB work and is consistent with other recent publications (Kessel et al., 2017).

Using raw data from each episode, we first calculated response-level parameters (mean, peak, and speed). We used a reciprocal square root function to transform latency scores into speed measures, calculated mean and peak intensity scores of target responses, and calculated percent of epochs with the target response for presence/absence coded responses. For episodes with clear shifts in affective context (e.g., anticipatory positive affect prior to receiving a less desirable gift in Wrong Gift), only behaviors from the relevant time frame were included. Variables with significantly non-normal distributions (skew > +2), were transformed with a square root function. Behaviors with low frequency or variance were dropped.

We then created episode-level composites for relevant RDoC constructs. Response-level parameters were z-transformed and included in a unit-weighted mean average for each episode. This yielded 17 episode-level composites, which are presented along with their response level parameters and internal consistency estimates in Table 1. Episode-level composites were averaged with observer ratings to create cross-episode composites that aligned with four RDoC constructs: Reward Responsiveness, Frustrative Nonreward, Loss, and Fear. Initially, we created separate measures of reward anticipation and initial response to reward. These measures were highly correlated (r = .57, p < .001) and thus combined into a single Reward Responsiveness variable. Because children participated in Lab-TAB individually and responses were coded by raters naïve to twin status, each participant was treated as an individual in composite creation.

Table 1.

RDoC Construct Constituents Derived from Lab-TAB Episodes and Observer Ratings

RDoC Construct	Response Level Parameters	Internal Consistency (α)
	Balloon Bop: Speed to positivity, mean and peak smile intensity, mean and peak enthusiasm vigor, percent epochs with laughter present	.69
	Free Play: Speed to first touch, mean and peak play intensity, mean and peak toys played with	.81
Reward Responsiveness	Hungry Hippos (Expectation): Speed to positivity, mean and peak smile intensity (first 30 seconds of each game), mean and peak vigor of enthusiasm (first 30 seconds of each game)	.75
	Hungry Hippos (Reward): Mean and peak post-game positivity (following all three games)	(r = .86)
	Not Sharing: Mean and peak positivity first 3 epochs (prior to average first unequal share)	(r = .83)
	Transparent Box: Positivity in first epoch	n/a
	Wrong Gift: Speed to positivity, mean and peak anticipatory (pre-gift) positivity, percent gift-choosing epochs with positivity present	.77
	Observer Ratings: Anticipatory positive affect	(r = .49)
	Observer Ratings: Enthusiasm	(r = .49)
	Observer Ratings: Exuberance	(r = .47)
	Impossibly Perfect Stars: Speed to first anger, mean and peak facial anger, mean and peak bodily anger, speed to first anger vocalization, mean and peak anger vocalization	.80
Frustrative Nonreward	Not Sharing: Speed to first anger, mean and peak anger expression, speed to first anger vocalization, mean and peak anger vocalization, percent epochs with resistance	.84
	Transparent Box: Mean and peak facial anger, mean and peak bodily anger	.84
	Wrong Gift: Mean and peak facial anger, mean and peak bodily anger	.88
	Observer Ratings: Task frustration	(r = .50)
	Observer Ratings: Anger proneness	(r = .53)
	Impossibly Perfect Stars: Speed to first sadness, mean and peak facial sadness, mean and peak bodily sadness, speed to first sad vocalization, mean and peak sad vocalization	.79
Loss	Not Sharing: Speed to first sadness, mean and peak sadness expression, speed to first sad vocalization, mean and peak sad vocalization, percent epochs with resignation	.74
	Transparent Box: Mean and peak facial sadness, mean and peak bodily sadness	.84
	Wrong Gift: Mean and peak facial sadness, mean and peak bodily sadness	.84
	Observer Ratings: Sadness proneness	(r = .39)
	Scary Mask: Speed to fear, mean and peak facial fear, mean and peak bodily fear	.81
Fear	Storytelling: Speed to fear, mean and peak bodily fear, percent of episode silent	.77
	Observer Ratings: Fear	(r = .38)

Note. Internal consistency measured using Cronbach’s alpha for measures with 3+ items and simple correlations for 2 items.

Relation of RDoC constructs to psychopathology.

To analyze the association of RDoC composites with parent reported child behavior problems, we used a multilevel model (MLM using SAS proc glimmix) with correlated residuals for each individual twin to account for non-independence within families. Models were run separately for each problem behavior (HBQ and DISC), and all RDoC constructs were used as predictors in each model. These multilevel models test whether RDoC constructs relate to problem behaviors in theoretically consistent ways while also controlling for standing on other RDoC behaviors. The HBQ subscales were all normally distributed, so we imposed such a distribution on the MLM. For analyses examining DISC symptom counts we used a Poisson distribution (Agresti, 2003). In addition, to control for age, sex, and SES on RDoC behaviors, we regressed each out of the RDoC constructs before conducting the MLMs. Significance was determined by examining FDR corrected p-values (Benjamini & Hochberg, 1995) to control for the proportion of tests expected to be significant. Raw and adjusted p-values calculated using R’s p.adjust are presented (R Core Team, 2015).

Biometric modeling.

We used a structural equation modeling (SEM) approach in Mplus version 7.3 (Muthén, Muthén, & Asparouhov, 2016) to examine biometric models of RDoC constructs. Mplus uses robust maximum likelihood estimation to account for non-normal and missing data. Univariate biometrical models partition the variance in RDoC constructs into latent additive genetic (A), shared environmental (C), and nonshared environmental (E) influences. Higher intrapair similarity among MZ pairs compared with DZ pairs signifies genetic influences; shared environmental influences are non-genetic influences that make siblings reared together more similar than expected based on genetic factors; and nonshared environmental influences are non-genetic influences that contribute to sibling differences (including measurement error) (Neale & Cardon, 2013). Estimates of A, C, and E effects, which sum to 1.0 when expressed as variance components, reflect the cumulative effect of genes and environment without implicating specific genes or experiences.

The bivariate Cholesky (Figure 1) decomposes the covariation between RDoC constructs and behavior problem measures into genetic and environmental covariation. Six latent factors are estimated in this model (A₁, C₁, E₁, A₂, C₂, E₂). Latent variables A₁, C₁, and E₁ account for all the variation in the RDoC construct (a₁₁, c₁₁, and e₁₁) as well as part of the variation in concurrent or later behavior problems, a₂₁, c₂₁, and e₂₁ (i.e., the variance in behavior problems that can be accounted for by RDoC constructs). A₂, C₂, and E₂ account for the residual variation in behavior problems via paths a₂₂, c₂₂, and e₂₂.

Figure 1.

Bivariate Biometric Model of Genetic and Environmental Influences Relating RDoC Constructs to Problem Behaviors, Either Concurrently or Longitudinally

Key to latent variables and paths:

A₁, A₂ = genetic latent variables

C₁, C₂ = shared environmental latent variables

E₁, E₂ = nonshared environmental latent variables

a_11, c_11, e₁₁ = effects of A_1, C_1, and E_1, respectively, on RDoC construct

a_21, c_21, e₂₁ = effects of A_1, C_1, and E_1, respectively, on problem behavior

a_22, c_22, e₂₂ = effects of A_2, C_2, and E_2, respectively, residual problem behavior variance

Sex and age variation within our two phases were related to some RDoC constructs and behavior problems. Failing to adjust for age and sex can lead to an overestimation of twin similarity (McGue & Bouchard, 1984). Because a full exploration of sex-linked genetic differences requires larger sample sizes, we regressed sex and age from both the RDoC constructs and the behavior problem measures prior to fitting the biometric models.

Results

RDoC Construct Creation

Correlations for coded behaviors across episode and within RDoC construct are in Table 2. Reliability can be calculated with different levels of aggregation of the constituent measures. That is, we can use the parameter-level elements from each episode (e.g., 27 elements for Reward Responsiveness), or the more molar episode-level composites and observer ratings (e.g., 10 elements for Reward Responsiveness). Reliability estimates calculated in these two ways were as follows: Reward Responsiveness, α (27 parameter-level elements) = .85; α (10 episode-level elements) = .77; Frustrative Nonreward, α (25 parameter-level elements) = .86; α (6 episode-level elements) = .71; Loss, α (24 parameter-level elements) =. 78; α (5 episode-level elements) = .43; and Fear, α (10 parameter-level elements) = . 77; α (3 episode-level elements) = .37. The estimates based on episode-level elements are very low for Loss and Fear, which was due to both fewer items (especially for Fear) and weaker convergence of the observer ratings with the elicited behaviors for Loss and Fear.

Table 2.

Correlations Across Episodes Within RDoC Composites for Coded Behavior

Reward Responsiveness	1.	2.	3.	4.	5.	6.	7.	8.	9.
1. Balloon Bop
2. Free Play	.13***
3. Hungry Hippos (Anticipation)	.30***	.10***
4. Hungry Hippos (Response)	.32***	.13***	.35***
5. Not Sharing	.20***	.12***	.19***	.31***
6. Transparent Box	.07**	.08**	.15***	.12***	.23***
7. Wrong Gift	.16***	.18***	.20***	.31***	.29***	.11***
8. Observer Anticipatory PA	.22***	.22***	.20***	.30***	.32***	.11***	.42***
9. Observer Enthusiasm	.28***	.26***	.17***	.34***	.34***	.13***	.25***	.52***
10. Observer Exuberance	.38***	.30***	.30***	.39***	.28***	.13***	.25***	.50***	.62***
Frustrative Nonreward	1.	2.	3.	4.	5.
1. Not Sharing
2. Impossibly Perfect Stars	.34***
3. Transparent Box	.19***	.23***
4. Wrong Gift	.20***	.12***	.20***
5. Observer Task Frustration	.35***	.34***	.31***	.30***
6. Observer Anger Proneness	.31***	.28***	.22***	.30***	.63***
Loss	1.	2.	3.	4.
1. Not Sharing
2. Impossibly Perfect Stars	.18***
3. Transparent Box	.07*	.16***
4. Wrong Gift	.10**	.13***	.08**
5. Observer Sad Proneness	.08**	.11***	.18***	.25***
Fear	1.	2.
1. Scary Mask
2. Storytelling	.34***
3. Observer Fear	.12***	.08**

Note.

^*p < .05,

^**p < .01,

^***p <.001.

The four constructs were not fully independent, with significant and small to moderate cross-construct correlations. The strongest relationship was between Frustrative Nonreward and Loss (r = .42, p < .001). Reward Responsiveness was moderately correlated with Frustrative Nonreward (r = .32, p < .001) and had a small positive correlation with Loss (r = .06, p < .05). Fear was negatively associated with Reward Responsiveness (r = −.17, p < .001) and Frustrative Nonreward (r = −.09, p < .01) and positively correlated with Loss (r = .08, p < .01).

Table 3 provides descriptive statistics for RDoC constructs, including relations with age, SES, and sex. All constructs except Reward Responsiveness were negatively but modestly correlated with twin age. Higher SES individuals showed more Frustrative Nonreward and less Fear. Males showed higher levels of Reward Responsiveness and Frustrative Nonreward, and females showed higher levels of Fear.

Table 3.

Descriptive Statistics for RDoC Constructs

	r Age	r SES	Male Mean (SD)	Female Mean (SD)	t-test (df)
Reward Responsiveness	−.06*	.04	.04 (.53)	−.04 (.48)	t(1369.32) = −2.97**
Frustrative Nonreward	−.19***	.07*	.10 (.57)	−.11 (.50)	t(1363.10) = −7.20***
Loss	−.26***	.01	.01 (.45)	−.01 (.45)	t(1370) = −.83
Fear	−.10***	−.11***	−.08 (.58)	.06 (.58)	t(1370) = 4.53***

Note. t-test accounts for unequal group variances.

^*p < .05,

^**p < .01,

^***p <.001.

RDoC Construct Associations with Problem Behaviors and DSM Symptoms

Zero-order correlations for relations between observed RDoC behaviors and parent-report on the HBQ and DISC are in Supplemental Table 1, with results for multilevel models (MLMs) in Tables 4 and 5. Multilevel parameter estimates in Tables 4 and 5 are not standardized as in traditional linear regression. Instead, estimates are pooled based on the specific predictors and outcomes within each model and thus estimates are not directly comparable across models (Raudenbush & Bryk, 2002).

Table 4.

Multilevel Models: Concurrent Associations of RDoC Constructs with HBQ and DISC at Age 7

Outcome HBQ	Predictors	Estimate	s.e.	p-value	FDR p-value	Outcome DISC	Predictors	Estimate	s.e.	p-value	FDR p-value
Depression	Reward Responsiveness	−0.001	0.012	0.9161	0.9457	Depression	Reward Responsiveness	0.114	0.064	0.0766	0.1572
	Frustrative Nonreward	0.020	0.013	0.1147	0.1913		Frustrative Nonreward	0.153	0.065	0.0184	0.0595
	Loss	0.034	0.014	0.0186	0.0595		Loss	0.251	0.072	0.0005	0.0036**
	Fear	0.024	0.010	0.0197	0.0600		Fear	0.089	0.056	0.1104	0.1910
Overanxiousness	Reward Responsiveness	−0.001	0.016	0.9630	0.9630	Generalized Anxiety	Reward Responsiveness	0.012	0.066	0.8560	0.8981
	Frustrative Nonreward	0.014	0.016	0.3857	0.5610		Frustrative Nonreward	0.075	0.068	0.2721	0.4050
	Loss	0.049	0.019	0.0095	0.0380*		Loss	0.149	0.075	0.0483	0.1104
	Fear	0.030	0.013	0.0231	0.0672		Fear	0.081	0.056	0.1502	0.2403
Separation Anxiety	Reward Responsiveness	−0.011	0.017	0.5074	0.6622	Separation Anxiety	Reward Responsiveness	−0.131	0.070	0.0618	0.1318
	Frustrative Nonreward	0.010	0.018	0.5618	0.6622		Frustrative Nonreward	0.113	0.072	0.1166	0.1913
	Loss	0.013	0.021	0.5231	0.6622		Loss	0.105	0.079	0.1842	0.2807
	Fear	0.031	0.014	0.0309	0.0824		Fear	0.050	0.060	0.4010	0.5703
Inhibition	Reward Responsiveness	−0.115	0.022	<.0001	<.0001***	Social Anxiety	Reward Responsiveness	−0.261	0.110	0.0174	0.0595
	Frustrative Nonreward	−0.109	0.023	<.0001	<.0001***		Frustrative Nonreward	−0.231	0.120	0.0541	0.1194
	Loss	0.012	0.026	0.6363	0.7144		Loss	0.328	0.124	0.0084	0.0358*
	Fear	0.063	0.018	0.0007	0.0045**		Fear	0.294	0.096	0.0022	0.0128*
Impulsivity	Reward Responsiveness	0.088	0.022	<.0001	<.0001***	Specific Phobia	Reward Responsiveness	0.027	0.066	0.6797	0.7373
	Frustrative Nonreward	0.108	0.023	<.0001	<.0001***		Frustrative Nonreward	0.040	0.068	0.5578	0.6622
	Loss	0.060	0.027	0.0257	0.0715		Loss	0.151	0.075	0.0445	0.1055
	Fear	0.015	0.019	0.4179	0.5814		Fear	0.094	0.056	0.0951	0.1739
Opposition	Reward Responsiveness	−0.009	0.021	0.6614	0.7298	ADHD	Reward Responsiveness	0.227	0.056	<.0001	<.0001***
	Frustrative Nonreward	0.076	0.022	0.0005	0.0036**		Frustrative Nonreward	0.277	0.055	<.0001	<.0001***
	Loss	0.053	0.025	0.0357	0.0914		Loss	0.190	0.062	0.0024	0.0128*
	Fear	0.013	0.018	0.4669	0.6225		Fear	0.029	0.049	0.5557	0.6622
Overt Aggression	Reward Responsiveness	0.025	0.019	0.1783	0.2783	Conduct Disorder	Reward Responsiveness	0.056	0.088	0.5281	0.6622
	Frustrative Nonreward	0.033	0.020	0.0921	0.1734		Frustrative Nonreward	0.307	0.086	0.0004	0.0036**
	Loss	0.054	0.023	0.0176	0.0595		Loss	0.172	0.098	0.0786	0.1572
	Fear	−0.005	0.016	0.7736	0.8252		Fear	0.058	0.077	0.4488	0.6111
Relational Aggression	Reward Responsiveness	0.027	0.017	0.1059	0.1883	Oppositional Defiant	Reward Responsiveness	0.016	0.033	0.6352	0.7144
	Frustrative Nonreward	0.030	0.017	0.0818	0.1586		Frustrative Nonreward	0.097	0.034	0.0043	0.0212*
	Loss	0.054	0.020	0.0069	0.0315*		Loss	0.078	0.039	0.0430	0.1055
	Fear	0.001	0.014	0.9564	0.9630		Fear	0.016	0.028	0.5691	0.6622

Note. HBQ: MacArthur Health and Behavior Questionnaire; average of mother and father report. DISC: Diagnostic Interview Schedule for Children-IV symptom counts; primary caregiver report. FDR: False Discovery Rate.

^*p < .05,

^**p <.01,

^***p < .001.

Significance reported only for FDR p-values.

Table 5.

Multi-level Models: Longitudinal Associations of RDoC Constructs at Age 7 with HBQ and DISC at Age 13

Outcome HBQ	Predictors	Estimate	s.e.	p-value	FDR p-value	Outcome DISC	Predictors	Estimate	s.e.	p-value	FDR p-value
Depression	Reward Responsiveness	0.016	0.017	0.3572	0.5384	Depression	Reward Responsiveness	0.055	0.084	0.5140	0.6976
	Frustrative Nonreward	0.021	0.018	0.2594	0.4808		Frustrative Nonreward	0.171	0.084	0.0412	0.1648
	Loss	0.045	0.021	0.0317	0.1417		Loss	0.220	0.091	0.0161	0.0941
	Fear	0.003	0.016	0.8392	0.9315		Fear	−0.033	0.077	0.6722	0.8240
Overanxiousness	Reward Responsiveness	−0.002	0.020	0.9058	0.9809	Generalized Anxiety	Reward Responsiveness	−0.022	0.092	0.8145	0.9280
	Frustrative Nonreward	0.026	0.022	0.2396	0.4808		Frustrative Nonreward	−0.009	0.097	0.9293	0.9809
	Loss	0.037	0.025	0.1431	0.3468		Loss	0.192	0.104	0.0650	0.2352
	Fear	0.016	0.019	0.4158	0.5951		Fear	0.057	0.085	0.5023	0.6941
Separation Anxiety	Reward Responsiveness	0.017	0.017	0.3086	0.5099	Separation Anxiety	Reward Responsiveness	0.116	0.141	0.4122	0.5951
	Frustrative Nonreward	0.019	0.018	0.2875	0.5081		Frustrative Nonreward	0.131	0.143	0.3613	0.5384
	Loss	0.027	0.020	0.1739	0.3737		Loss	0.255	0.152	0.0930	0.3213
	Fear	0.012	0.016	0.4228	0.5951		Fear	−0.030	0.129	0.8181	0.9280
Social Anxiety	Reward Responsiveness	−0.064	0.025	0.0110	0.0697	Social Anxiety	Reward Responsiveness	−0.311	0.140	0.0261	0.1322
	Frustrative Nonreward	−0.041	0.027	0.1228	0.3369		Frustrative Nonreward	−0.209	0.155	0.1770	0.3737
	Loss	0.046	0.031	0.1315	0.3369		Loss	0.235	0.161	0.1460	0.3468
	Fear	0.032	0.023	0.1718	0.3737		Fear	0.121	0.132	0.3613	0.5384
Inhibition	Reward Responsiveness	−0.072	0.026	0.0051	0.0388*	Specific Phobia	Reward Responsiveness	−0.125	0.116	0.2821	0.5081
	Frustrative Nonreward	−0.100	0.028	0.0003	0.0076**		Frustrative Nonreward	0.113	0.119	0.3407	0.5384
	Loss	0.029	0.031	0.3527	0.5384		Loss	0.354	0.124	0.0043	0.0363*
	Fear	0.025	0.024	0.3066	0.5099		Fear	0.002	0.109	0.9853	0.9940
Impulsivity	Reward Responsiveness	0.038	0.024	0.1180	0.3369	ADHD	Reward Responsiveness	0.146	0.096	0.1281	0.3369
	Frustrative Nonreward	0.116	0.026	<.0001	<.0001***		Frustrative Nonreward	0.208	0.094	0.0281	0.1335
	Loss	0.069	0.030	0.0209	0.1135		Loss	0.294	0.102	0.0041	0.0363*
	Fear	0.014	0.023	0.5526	0.7241		Fear	−0.005	0.090	0.9537	0.9929
Inattention	Reward Responsiveness	0.032	0.029	0.2594	0.4808	Conduct Disorder	Reward Responsiveness	0.194	0.126	0.1235	0.3369
	Frustrative Nonreward	0.093	0.031	0.0026	0.0363*		Frustrative Nonreward	0.147	0.127	0.2491	0.4808
	Loss	0.110	0.035	0.0018	0.0342*		Loss	0.211	0.141	0.1330	0.3369
	Fear	0.000	0.027	0.9940	0.9940		Fear	0.023	0.120	0.8457	0.9315
Conduct Disorder	Reward Responsiveness	0.004	0.009	0.6074	0.7568	Oppositional Defiant	Reward Responsiveness	0.073	0.045	0.1096	0.3369
	Frustrative Nonreward	0.027	0.009	0.0034	0.0363*		Frustrative Nonreward	0.067	0.047	0.1526	0.3514
	Loss	0.020	0.011	0.0590	0.2242		Loss	0.106	0.052	0.0409	0.1648
	Fear	0.004	0.008	0.5862	0.7425		Fear	0.068	0.042	0.1041	0.3369
Opposition	Reward Responsiveness	0.029	0.026	0.2574	0.4808
	Frustrative Nonreward	0.080	0.028	0.0041	0.0363*
	Loss	0.033	0.032	0.3007	0.5099
	Fear	0.001	0.024	0.9799	0.9940
Overt Aggression	Reward Responsiveness	0.006	0.015	0.6988	0.8430
	Frustrative Nonreward	0.061	0.017	0.0003	0.0076**
	Loss	0.011	0.019	0.5648	0.7275
	Fear	−0.001	0.015	0.9229	0.9809
Relational Aggression	Reward Responsiveness	0.003	0.014	0.8121	0.9280
	Frustrative Nonreward	0.041	0.015	0.0076	0.0525
	Loss	0.011	0.017	0.5363	0.7151
	Fear	−0.004	0.013	0.7435	0.8829

Note. HBQ: MacArthur Health and Behavior Questionnaire; average of mother and father report. DISC: Diagnostic Interview Schedule for Children-IV symptom counts; primary caregiver report. FDR: False Discovery Rate.

^*p < .05,

^**p <.01,

^***p < .001.

Significance reported only for FDR p-values.

Concurrent associations from MLMs.

At age 7, higher scores on observed RDoC Reward Responsiveness were associated with lower parent-reported inhibition on the HBQ and higher HBQ impulsivity and greater DISC ADHD symptoms. Higher RDoC Frustrative Nonreward was associated with lower parent-reported HBQ inhibition but higher HBQ impulsivity and opposition as well as higher symptoms of ADHD, conduct disorder, and oppositional defiant disorder on the DISC. Higher observed Loss was associated with higher HBQ overanxiousness and relational aggression and greater symptoms of depression, social phobia, and ADHD on the DISC. Finally, higher RDoC Fear was related to higher HBQ inhibition and higher DISC social anxiety. In summary, these four RDoC observational constructs predicted the concurrent HBQ and DISC parentally reported psychopathology scales in a manner that transcends traditional dimensions of non-comorbid psychopathology.

Longitudinal associations from MLMs.

Some of the concurrent relationships between observed RDoC constructs and parent reported measures were replicated in longitudinal analyses. Specifically, higher scores on observed RDoC Reward Responsiveness at age 7 were related to lower levels of parent-reported inhibition on the HBQ at age 13. Higher RDoC Frustrative Nonreward at age 7 was associated with lower HBQ inhibition but higher HBQ impulsivity, inattention, opposition, aggression, and conduct problems at age 13. Loss predicted higher levels of HBQ inattention as well as greater symptoms of specific phobia and ADHD on the DISC. RDoC Fear observed at age 7 showed no significant longitudinal associations with parent-report.

Genetic Analyses

Twin Similarity.

The intraclass correlations (ICCs) that allow simple genetic analyses are in Table 6. In all cases except RDoC Fear, the MZ ICC is approximately twice the magnitude of the DZ ICC. This pattern implies additive genetic contributions to the phenotypic variance and lack of contribution of shared environmental effects. For RDoC Fear, however, the DZ ICC (.403) is only modestly lower than the MZ value (.511); this pattern implies shared environmental variance.

Table 6.

Monozygotic and Dizygotic Twin Intraclass Correlations and Univariate Genetic and Environmental Path Estimates for the Four RDoC Constructs

RDoC constructs	MZ ICC	DZ ICC	Parameters from fitting ACE model
	(261 pairs)	(399 pairs)	a	c	e
Reward Responsiveness	.760	.316	.856	.000	.517
Frustrative Nonreward	.620	.341	.779	.173	.602
Loss	.373	.203	.539	.256	.803
Fear	.511	.403	.444	.553	.705

Note. ACE values are standardized, unsquared estimates, which can be squared for variance estimates. Bolded estimates are significantly different from 0. Sex and age were regressed out of scores before ICCs were calculated; thus, both same-sex and opposite-sex pairs are included in the DZ column.

Genetic and environmental variance estimates from univariate models of RDoC constructs are shown on the right side of Table 6. The heritabilities, calculated by squaring the a estimates, are high for Reward Responsiveness (73%) and Frustrative Nonreward (61%) and modest for Loss (29%) and Fear (20%). Fear is notable for the significant contribution of shared environmental effects. Parallel genetic analyses (ICCs and univariate ACE models) for problem behaviors are in Supplemental Table 3.

Bivariate biometric models.

Table 7 provides results of the bivariate analyses. Each significant bivariate phenotypic prediction of symptoms from RDoC constructs from Tables 4 and 5 was analyzed with biometric Cholesky decompositions.

Table 7.

Model Fit and Estimates for Bivariate Cholesky Decompositions of the Prediction of Symptoms from RDoC Constructs

Significant Relation from Multilevel Models	RMSEA (90%CI)	CFI	a11	c11	e11	a21	c21	e21	a22	c22	e22
HBQ Age 7 (concurrent)
Reward Responsiveness → Inhibition (−)	.057 (.031 – .083)	.963	.821	.241	.517	−.264	.130	−.070	.794	--	.528
Reward Responsiveness → Impulsivity (+)	.050 (.021 – .077)	.968	.856	.000	.517	.169	.000	.012	.807	--	.566
Frustrative Nonreward → Inhibition (−)	.045 (.011 – .072)	.971	.777	.180	.603	−.212	−.032	−.042	.818	--	.532
Frustrative Nonreward → Impulsivity (+)	.037 (.000 – .066)	.979	.778	.179	.602	.215	.013	.042	.797	--	.564
Frustrative Nonreward → Opposition (+)	.028 (.000 – .060)	.986	.773	.194	.604	.151	.069	−.014	.786	--	.595
Loss → Overanxiousness (+)	.015 (.000 – .052)	.993	.531	.272	.802	.281	−.160	−.008	.706	--	.630
Loss → Relational Aggression (+)	.048 (.018 – .075)	.945	.606	.049	.794	.179	.628	−.032	.325	--	.683
Fear → Inhibition (+)	.028 (.000 – .060)	.981	.541	.253	.802	−.097	−.002	.008	.841	--	.532
DISC Age 7 (concurrent)
Reward Responsiveness → ADHD (+)	.081 (.058 – .105)	.922	.857	.000	.516	.155	.000	.043	.837	--	.524
Frustrative Nonreward → ADHD (+)	.071 (.047 – .096)	.932	.768	.217	.603	.300	.001	−.012	.795	--	.527
Frustrative Nonreward → Conduct Disorder (+)	.017 (.000 – .053)	.997	.794	.100	.599	.123	.583	−.039	.635	--	.490
Frustrative Nonreward → Oppositional Defiant (+)	.000 (.000 – .000)	1.000	.799	.042	.600	.159	.372	−.046	.780	--	.475
Loss → Depression (+).54	.071 (.047 – .096)	.842	.544	.247	.802	.165	.062	.051	.750	--	.636
Loss → Social Anxiety (+)	.071 (.047 – .095)	.696	.516	.293	.805	.110	−.060	.022	.574	--	.809
Loss → ADHD (+)	.079 (.056 – .103)	.879	.532	.265	.804	.261	.057	.029	.806	--	.527
Fear → Social Anxiety (+)	.065 (.040 – .090)	.875	.442	.556	.704	.264	−.009	.025	.525	--	.808
HBQ Age 13 (longitudinal)
Reward Responsiveness → Inhibition (−)	.055 (.028 – .081)	.948	.856	.000	.517	−.134	.000	−.096	.730	--	.663
Frustrative Nonreward → Inhibition (−)	.037 (.000 – .066)	.968	.781	.174	.600	−.251	−.003	.012	.703	--	.666
Frustrative Nonreward → Impulsivity (+)	.072 (.048 – .096)	.898	.769	.207	.604	.302	.005	−.022	.782	--	.544
Frustrative Nonreward → Inattention (+)	.070 (.046 – .095)	.915	.855	.000	.519	.118	.000	−.060	.768	--	.626
Frustrative Nonreward → Opposition (+)	.064 (.040 – .090)	.923	.850	.085	.520	.094	.197	−.086	.673	--	.702
Frustrative Nonreward → Overt Aggression (+)	.081 (.058 – .106)	.877	.852	.073	.519	.051	.547	−.085	.282	--	.782
Frustrative Nonreward → Conduct Disorder (+)	.070 (.045 – .094)	.906	.845	.130	.519	.004	.397	−.051	.487	--	.777
DISC Age 13 (longitudinal)
Loss → Specific Phobia (+)	.052 (.24 – .079)	.838	.539	.257	.802	.135	.102	.003	.684	--	.709
Loss →ADHD (+)	.056 (.030 – .082)	.867	.537	.260	.802	.278	−.004	−.026	.725	--	.629

Note. Refer to Figure 1 for graphic depiction of the paths. For residual symptom variance, the shared environment latent variable (c₂₂) was set to zero, as suggested by the results for univariate symptom models in Supplemental Table 3. +/− in parentheses denote direction of relation.

Bolded estimates are significant.

The results of these bivariate models allow four types of inferences. The first type is the phenotypic estimates that can be derived from the model. Phenotypic prediction from RDoC to symptoms is derived by squaring—converting to variance estimates—each of the standardized cross-measure parameters (a₂₁, c₂₁, and e₂₁) and adding the three squares to yield r². Thus, for the first row in Table 7 (Reward Responsiveness→Inhibition), r² = (.264² + .130² + .070²) = .091, which corresponds to a standardized bivariate regression of −.302. This estimate is similar to the corresponding estimate of −.27 from the MLMs in Table 4, which was calculated with different covariates and different estimation techniques.

We next turn to the three behavior-genetic inferences. The second type of results—A, C, and E effects on the RDoC constructs—reiterates the results of the univariate models for RDoC constructs from Table 6. The third type, represented by the a₂₂ and e₂₂ parameter estimates on the right side of Table 7, indicates that each of symptom measures shows residual variance (independent of the RDoC construct) that has both genetic and nonshared environmental components.

The fourth type, which is our main focus here, consists of the a₂₁, c₂₁, and e₂₁ estimates, the biometric decomposition of the regression of symptoms on RDoC constructs. These parameter estimates are low because the phenotypic regressions are modest. As an example, the 7^th row from the bottom of Table 7 shows the biometric decomposition of the prediction of adolescent impulsivity (HBQ) from childhood RDoC Frustrative Nonreward. The cross-time coefficients (a₂₁, c₂₁, and e₂₁) show that the genetic effect (.302, or 9% of the variance in HBQ impulsivity) accounts for essentially all of the longitudinal prediction. Overall, the set of longitudinal results shows that genetic components of the RDoC constructs most frequently accounted for most, or nearly all, of the predictive power for concurrent and subsequent symptoms. However, in 5 cases when the symptom outcome involved the related problems of defiance, oppositional behavior, aggressiveness, or conduct disorder, then shared environmental factors accounted for the prediction. In 4 of these 5 cases, the predictor was Frustrative Nonreward; in the other case, the predictor was Loss.

Discussion

We describe the derivation of behavioral measures of RDoC constructs for children in the Positive and Negative Valence domains using Lab-TAB. We provide psychometric information about reliability and independence of the measures. To address validity and clinical relevance, we note several plausible associations of RDoC measures with more traditional childhood psychopathology measures. We show that RDoC measures, especially Reward Responsiveness and Frustrative Nonreward, are heritable and that predictions from RDoC measures to symptom measures are genetically mediated in most cases. Mediation findings relating Loss to problem behaviors are mixed, indicating both genetic and environmental mediation, and outcomes in the defiance/conduct/aggressiveness domains showed shared environmental mediation from RDoC predictors.

Motivational Tendencies and Context are Crucial

One implication of our findings, also noted by Franklin and colleagues (2015), is that motivational tendencies matter. Reward Responsiveness comes from the positive valence system and Frustrative Nonreward from the negative valence system; however, these two RDoC constructs are moderately positively correlated and share approach motivation (Carver & Harmon-Jones, 2009). Thus, though the RDoC matrices imply strong delineation between positive and negative valence systems, affective valence alone is insufficient to account for the patterning of expression of RDoC measures. An alternative interpretation is that the correlation between Reward Responsiveness and Frustrative Nonreward reflects variance due to partially overlapping contexts of assessment, although this cannot explain the lack of a similar finding with Loss, which also draws on the same episodes.

The importance of context is underscored by our methodological approach. Lab-TAB provides objective assessment of behavior, yet objective measures are relatively thin slices of behavior with limited reliability. Additionally, how broad or narrow each RDoC construct might be is unknown. A broad construct organizes reactions to a wide range of contexts with related affective incentives, as Reward Responsiveness and Frustrative Nonreward apparently do. In contrast, correlations across episodes for Loss were low and observer ratings of Fear were weakly correlated with episode-level parameters of fear scored from videotapes.

Lab-TAB episodes are designed to elicit a range of emotional responses to everyday stimuli. Whether the components of a construct measured by Lab-TAB form a scale versus an index is not always apparent. We generally assume that latent constructs are the source of correlation among their indicators (as in factor models); thus, the indicators form a scale and can be subjected to, for instance, reliability estimation via Cronbach’s alpha. However, different indicators could be uncorrelated but still contribute to a common outcome, such as an RDoC construct. For instance, though both indicators of fear, children cannot physically both withdraw and freeze at the same time; there may also be contingencies that would reduce co-occurrence among other fear indicators that are less readily apparent. Thus, we cannot exclude the possibility that composites derived from behavioral indicators might have a mixture of scale-like and index-like features. Cronbach’s alpha cannot be used to evaluate the reliability of indices.

Clinical Relevance of RDoC Behavioral Measures in Children

The RDoC behavioral measures were not designed to align with DSM symptoms. Our clinical measures, the HBQ and the DISC-IV, do tap symptoms, only some of which are affective in nature. Because Lab-TAB video was scored independently by experimenters and parents reported on the HBQ and DISC, no methodological assessment overlap confounds the interpretation of our clinically focused findings. Moreover, both concurrent and longitudinal associations emerged although the interval between assessments is substantial and includes part of the puberty transition. Given the modest stability seen even in longitudinal analyses utilizing the same methods across time points, we suggest that our significant findings are noteworthy and here highlight two patterns of potential clinical relevance.

Inhibition.

Our RDoC measure of Fear is conceptually different from behavioral inhibition, as defined by Kagan et al. (1987); however, HBQ Inhibition scores align rather precisely with Kagan’s construct. Nevertheless, children with high levels of observed RDoC Fear were rated by their parents as higher in inhibition and social anxiety. Children higher in Reward Responsiveness or Frustrative Nonreward were rated lower in inhibition. These findings may appear at first glance to represent merely two sides to the same coin, such that fearful and anxious children are less likely to be impulsive while impulsive children are less likely perceived as highly inhibited. This interpretation, however, is not supported by the relatively low correlation between Fear and other RDoC constructs, consistent with prior temperament findings in this (Clifford et al., 2015) and other samples (Putnam et al., 2008). Thus, fear alone does not confer vulnerability for anxiety problems; indeed, there was no significant genetic or environmental mediation relating fear to problem behaviors. Instead, both approach and avoidance tendencies are associated with anxiety risk and resilience, consistent with equifinality models (Cicchetti & Rogosch, 1996).

Impulsivity.

A similar equifinality may explain findings with ADHD and impulsivity as outcomes. Children higher in Reward Responsiveness or Frustrative Nonreward were rated higher in impulsivity and ADHD, with varying patterns of associations depending on age. Observed Loss at age 7 was also associated with greater symptoms of ADHD both concurrently and longitudinally. The relationship of impulsivity and ADHD to both positive and negative valence constructs may reflect the multiple pathways models proposed by Nigg, Goldsmith, and Sachek (2004) and others, in which some children with ADHD show heightened positive approach behaviors whereas other children show negative approach, hostility, and impaired emotion regulation.

Implications of Genetics Findings

Abundant evidence supports the heritability of temperament traits (Gagne, Vendlinski, & Goldsmith, 2009), so genetic influences on variability in RDoC constructs, which are akin to dimensions of emotional individuality, are unsurprising. However, the observation that Reward Responsiveness and Frustrative Nonreward are more heritable than Loss and especially than Fear is notable and perhaps implicative for future RDoC research across multiple units of analyses. As Hettema (2016) notes, finding heritable phenotypes is a first step in determining which traits may be most amenable to molecular genetic investigation. The lower heritability of Loss and Fear may also be partially accounted for by their somewhat lower reliability, which would serve to increase the non-shared environment estimate and correspondingly decrease the additive genetic and shared environmental estimates.

Future investigations should explore whether the higher heritability of Reward Responsiveness and Frustrative Nonreward—and to some extent Loss—is part of the reason these constructs were more strongly related to symptoms in our analyses than Fear was. When Reward Responsiveness and Frustrative Nonreward predicted symptoms, that prediction was usually genetically mediated, as our biometric bivariate models showed. Thus, the three characteristics of being more heritable, being more predictive of symptoms, and showing genetic mediation of this prediction (in most but not all cases) generally characterized Reward Responsiveness and Frustrative Nonreward in our results whereas the opposite pattern was more characteristic of Fear. Loss showed a more complex pattern, with significant but modest heritability coupled with predictive power for symptoms outcomes, which was mediated by genetic or shared environmental factors, depending on the type of symptom outcome.

Options for Researchers

Lab-TAB provides options for RDoC researchers who investigate children’s mental health. Here, we used the middle childhood version of Lab-TAB, but Lab-TAB also has pre-locomotor, locomotor, and preschool versions for younger children. Lab-TAB is a direct behavioral measure, unfiltered through parent or teacher perceptions. The objectivity provided by Lab-TAB is consistent with many performance measures provided in the RDoC matrix, and it can be altered to fit specific research aims. For example, researchers could explore the role of context-inappropriate emotions (as in Locke et al., 2009) or choose to model emotional reactivity based on each episode’s anticipated salience (as recommended by Durbin, 2010), among other options. Such approaches may improve the estimation of cross-episode consistency and help clarify to what extent associations differ by context.

Conclusions

This study, which included both positive and negative valence constructs, is broadly conceived, as seems appropriate in this nascent stage of RDoC pediatric research. While Lab-TAB was designed and our data were collected before the advent of RDoC, we easily mapped many of the component measures from Lab-TAB onto the RDoC matrix. The resulting four RDoC positive and negative valence system measures were heritable and modestly associated with DSM-based measures in a transdiagnostic manner. These modest associations were often genetically mediated (but not in all cases), suggesting avenues for future research. We acknowledge that scientific judgment about the fit of Lab-TAB measures to RDoC constructs must accompany any proposed strategy but believe that this work demonstrates the feasibility and initial validation of Lab-TAB as a tool for researchers interested in pediatric behavioral paradigms within the RDoC framework.