Observed Emotional Reactivity in Response to Frustration Tasks in Psychiatrically Hospitalized Youth with ASD

Emotion regulation deficits have been proposed as a transdiagnostic phenomenon underlying affective reactivity and behavioral challenges seen in multiple mental health conditions, including autism spectrum disorder (ASD; Mazefsky et al., 2013). Although not part of the core diagnostic criteria, individuals with ASD often present with emotion dysregulation, including tantrums, outbursts, and sustained negative responses (Samson et al., 2014). These problems can exacerbate social difficulties (Blair et al., 2015), increase parental stress (Davis & Carter, 2008), and contribute to high rates of co-occurring psychiatric and behavioral problems (Khor, Melvin, Reid, & Gray, 2014; Mazefsky, Borue, Day, & Minshew, 2014; Samson, Hardan, Lee, Phillips, & Gross, 2015) that necessitate psychotropic medication prescription (Fung et al., 2016; Houghton, Ong, & Bolognani, 2017) and psychiatric hospitalizations in this population (Mandell, 2008).

Research on emotional reactivity in ASD has increased in recent years, but has primarily relied on questionnaire data from more cognitively able individuals or observational measures during structured tasks in very young children (Cai, Richdale, Uljarević, Dissanayake, & Samson, 2018). As a result, relationships between emotional reactivity and other aspects of child functioning, including ASD severity, adaptive behavior, and verbal ability, are not well understood. In the current study, we describe emotional reactivity and recovery in response to a series of standardized frustration tasks in a large, well characterized, heterogenous sample of psychiatrically hospitalized youth with ASD. We explore relationships between emotional reactivity and several demographic features and functional domains in order to better understand characteristics and correlates of emotional reactivity in this population.

Negative Emotional Reactivity in Individuals with ASD

Research utilizing parent-report questionnaires suggests that, on average, individuals with ASD exhibit higher levels of negative emotionality and lower levels of positive emotionality than non-ASD peers from infancy (Garon et al., 2009) through childhood and adolescence (Capps, Kasari, Yirmiya, & Sigman, 1993; De Pauw, Mervielde, Van Leeuwen, & De Clercq, 2011; Joseph & Tager-Flusberg, 1997). A few recent studies have utilized tasks from the Laboratory Temperament Assessment Battery (Lab-TAB; Goldsmith, Reilly, Lemery, Longley, & Prescott, 1993)—a series of structured tasks designed to elicit specific emotional reactions such as frustration, fear, or joy—to examine emotional reactions and emotion regulation strategies in toddlers or pre-school aged children with ASD. Results have been somewhat mixed, but overall findings suggest that young children with ASD use fewer adaptive or developmentally-advanced regulation strategies and display more intense and longer negative emotional reactions than their non-ASD peers (Costa, Steffgen, & Samson, 2017; Jahromi, Meek, & Ober-Reynolds, 2012; Macari et al., 2018; Nuske et al., 2017). For example, Jahromi et al. (2012) reported that preschool children with ASD displayed longer and more intense ‘resignation’ than typically developing children in response to a Lab-TAB frustration task in which a toy was locked in a transparent box, making it inaccessible. However, this same study found no differences between children with ASD and non-ASD peers in intensity or duration of facial/bodily negativity during frustrating tasks. Macari et al. (2018), on the other hand, reported that toddlers with ASD displayed higher levels of anger/frustration in response to Lab-TAB tasks than developmentally delayed, but not typically developing, peers. Importantly, this study differed from Jahromi et al., (2012) in that it examined only valence and peak intensity of facial and vocal expressions (rather than average intensity across the entire episode or total duration of negativity). Costa et al. (2017) coded valence of facial expressions and behaviors in 10-second intervals during a toy-removal task and reported more negative behaviors and more negative facial expressions in school-aged children with ASD compared to typically developing peers. Another more recent study by Costa, Steffgen, & Vögele (2019) examined emotional reactivity (frequency of negative, neutral, and positive facial expressions) in children (age 3–13) with and without ASD during a toy-removal task with their parent and found higher frequency of negative and neutral, and lower frequency of positive, facial expressions among ASD youth. Taken together, this research generally suggests increased negativity in response to frustration in ASD youth, but also points to the importance of considering multiple methods for quantifying negative response in order to better understand what this increased negativity looks like.

A limitation of the extant observational research on emotional reactivity in youth with ASD is that it focuses on participants’ emotions and behavior only during distressing tasks, not in comparison to baseline levels of emotion or recovery from tasks. Given parental reports of general increased negative emotionality and lower “soothability” in children with ASD, examining change in emotion from baseline to reaction and through recovery may be particularly important in this population. Furthermore, as mentioned above, the majority of existing research focuses on either young children or questionnaire data from older individuals (predominantly those who are verbal and without intellectual disability) and examines only overall group differences between individuals with and without ASD. There is limited understanding of the heterogeneity in negative emotional reactivity within groups of individuals with ASD.

Correlates of Emotional Reactivity in ASD

Although difficulties with emotion regulation and reactivity are prevalent in ASD, substantial variability exists across individuals. Understanding the correlates of this variability has important implications for intervention. Knowing what characteristics of individual functioning relate to emotional reactivity and recovery can provide insight into possible causes and consequences of dysregulation and could help determine who is most in need of interventions focused on facilitating emotion regulation, and potentially guide the type of intervention. In typically developing individuals, chronological age, biological sex, and language ability have all been linked to emotion competence, regulation, and/or reactivity. Specifically, older children are found to be less emotionally reactive and utilize more adaptive emotion regulation strategies than younger children (Kubicek & Emde, 2012; Murphy, Eisenberg, Fabes, Shepard, & Guthrie, 1999); boys express more anger, but less sadness and fear, than girls (Chaplin & Aldao, 2013); and children’s emotional competence and emotion regulation are positively related to their language skills (Beck, Kumschick, Eid, & Klann-Delius, 2012; Kubicek & Emde, 2012).

As far as we are aware, relatively few studies have examined correlates of emotion regulation or reactivity in individuals with ASD, and results have been mixed. Berkovits, Eisenhower, & Blacher (2017) examined relationships between parent-reported emotion regulation/dysregulation and IQ, verbal ability, autism severity (ADOS-2 comparison score), social/communicative ability (Social Responsiveness Scale; SRS; Constantino, 2002), and socioemotional/behavioral problems (Child Behavior Checklist; CBCL; Achenbach & Rescorla, 2001, 2000). Emotion dysregulation was found to be unrelated to IQ, verbal ability, and autism severity score, but significantly related to social/communicative ability (SRS score) and socioemotional/behavioral problems (CBCL Total Problems). Similarly, Samson et al. (2014) found that parent-reported emotion regulation was unrelated to their child’s IQ and adaptive behavior, but significantly related to SRS score, restricted and repetitive behaviors, and sensory difficulties. Jahromi, Bryce, & Swanson, (2013) did not find relations between parent-reported ER and measures of mental age, expressive language age, or receptive language age, but did find relations between ER and measures of executive function and peer engagement. Macari et al., (2018) examined the relationship between toddlers’ observed emotional expression during the Lab-TAB and scores on the Autism Diagnostic Observation Schedule-2 (ADOS-2; Lord et al., 2012) and did not find associations between expressions of anger/frustration or fear and severity of autism symptoms. Finally, Costa et. al. (2019) did not find differences between children with ASD with and without cognitive impairment in observed emotional reactivity.

Together, this small body of research suggests that emotional reactivity is largely unrelated to IQ and overall ASD severity (as measured by ADOS-2 comparison scores) but may be related to specific ASD symptom domains, such as social communication and repetitive behaviors, as well as social-emotional difficulties. As with most research in ASD, more severely affected older children and adolescents with ASD have largely been left out of studies examining emotion regulation and reactivity (Stedman, Taylor, Erard, Peura, & Siegel, 2019). This is arguably problematic, as this population may be particularly impacted by problems with emotion regulation. A recent review by Cai et al. (2018) describes a body of research that associates increased maladaptive (and decreased adaptive) emotion regulation strategy use in individuals with ASD with internalizing disorders such as anxiety and depression. Furthermore, more severely affected individuals with ASD often engage in self-injurious behaviors (SIB) and other problem behaviors such as physical aggression more often than less severely affected individuals (Farmer et al., 2015; Minshawi et al., 2014), difficulties that may be associated with poor emotion regulation.

The Present Study

Data gathered as part of the Autism Inpatient Collection (AIC; Siegel et al., 2015)—a six-site study of children, adolescents, and young adults admitted to specialized inpatient psychiatric units for youth with ASD and other developmental disorders—provides a unique opportunity to examine correlates of emotional reactivity in a heterogenous population spanning the spectrum of autism severity and functioning. The sample consists of youth with ASD who have required psychiatric hospitalization due to serious emotional or behavioral challenges. In the present study, 150 participants from the AIC (ranging in age from 5 to 20 yrs) were administered several tasks adapted from the Lab-TAB designed to elicit frustration. The study had two primary aims. First, to describe the emotional response (i.e., amount, intensity, and duration of negative emotion; total range of emotion displayed) of these youth before, during, and after frustrating tasks in order to examine change in emotional reactivity from baseline to frustration and to a recovery period. The second aim of the study was to examine relationships between emotional response to frustration and recovery and individual domains of functioning (i.e., age, sex, autism severity, adaptive behavior, non-verbal IQ, verbal ability, and psychiatric/behavioral problems). Based on previous research, we hypothesized that participants who were younger, male, displayed less adaptive behaviors, had lower verbal ability, and presented with higher reported psychiatric/behavioral problems would have greater increasing emotional reactivity from baseline to frustration tasks, and less recovery from frustration tasks. We did not expect non-verbal IQ or autism severity to relate to reactivity (Berkovits, Eisenhower, & Blacher, 2017; Macari et al., 2018).

Methods

Participants

Participants included 150 youth (ages 5 to 20 yrs; M = 12.76; SD = 3.69) with a research-reliable diagnosis of ASD from the AIC. All assessments and procedures took place during the individual’s stay in one of six specialized psychiatric inpatient units in six different states¹. Participants in the study were administered an Autism Diagnostic Observation Schedule-2 (ADOS-2; Lord et al., 2012) by a research-reliable examiner to confirm ASD diagnosis². The research-reliable ADOS-2 examiner provided a final diagnosis based on the child’s ADOS-2, developmental history, parent report, the social communication questionnaire (SCQ; Rutter, Bailey, & Lord, 2003), and ASD criteria from the DSM-5. All children included in this study received a final diagnosis of ASD. Exclusion criteria included not having a parent available who was proficient in English or the individual with ASD being a prisoner or ward of the state. The full methods of the AIC have been published previously (Siegel et al., 2015).

Table 1 displays demographic information, participant length of stay on the inpatient psychiatric unit, and characterization of participant functioning (i.e., age, sex, non-verbal IQ, adaptive behavior, autism severity, verbal ability, and psychiatric/behavioral problems; see “Measures” section below for details). Participants were classified into verbal ability groups based on their required ADOS-2 module. ADOS-2 module determinations were made by research-reliable ADOS-2 administrators after observing the child for a language sample and receiving input from clinical staff familiar with the individual, in accordance with ADOS-2 guidelines. Briefly, ADOS-2 Module 1 is intended for children 31 months and older who do not consistently use phrase speech; Module 2 is for children who use phrase speech but are not fluent; and Modules 3 and 4 are for fluently verbal children/adolescents and adolescents/adults, respectively.

Table 1.

Participant Characteristics.

Variable (Available Data)
Age (N = 150)	M = 12.76	SD = 3.69	Range: 5.98–20.80
Sex, Male (N = 150)	78% (N = 117)
Racial or Ethnic Minority (N = 134)	19.4% (N = 26)
Length of Stay on Inpatient Unit (days; N = 147)	M = 31.01	SD = 31.56	Range: 4–183
Total Household Income (N = 127)
<$20k	19.7% (N = 25)
21k–50k	37.8% (N = 48)
51k–80k	17.3% (N = 22)
81k–130k	18.1% (N = 23)
>131k	7.1% (N = 9)
Parent Education (Highest Level; N = 130)
Some HS or HS degree	23.1% (N = 30)
Some college or Associates degree	39.2% (N = 51)
Bachelors Degree	20.8% (N = 27)
Post-Graduate Degree	16.9% (N = 22)
Non-Verbal IQ (N = 144)	M = 74.28	SD = 26.36	Range: 31–135
Vineland Adaptive Behavior Composite (N = 113)	M = 60.78	SD = 12.89	Range: 31–110
ADOS Severity Score (N = 133)	M = 7.68	SD = 1.69	Range: 4–10
ADOS Module (verbal ability; N = 149)
Module 1 (non-verbal/single-words)	27.5% (N = 41)
Module 2 (phrase speech)	12.8% (N = 19)
Module 3 (fluent speech, child)	43.6% (N = 65)
Module 4 (fluent speech, adolescent/adult)	16.1% (N = 24_
CBCL Internalizing t-score	M = 65.18	SD = 8.57	Range = 41–87
CBCL Externalizing t-score	M = 69.69	SD = 7.32	Range = 44–86

Measures

The following demographic information was collected through parent report and chart review.

The Autism Diagnostic Observation Schedule, Second Edition (ADOS-2; Lord et al., 2012)

The ADOS-2 is a semi-structured, standardized measure used to assess and diagnose ASD. Raw scores from the ADOS administration were used to calculate comparison scores (Gotham, Pickles, & Lord, 2009), a standardized metric of severity of ASD-specific features ranging from 1 to 10 (with 1 = no ASD features and 10 = severe ASD symptoms). Comparison scores could not be calculated for children who were beyond the age cut off for their respective module (e.g., participant over the age of 14 receiving Module 1; N = 16).

The Leiter International Performance Scale, Third Edition (Leiter-3; Roid, Miller, Pomplun, & Koch, 2013) The Leiter-3 is a non-verbal test of intellectual functioning that does not require any spoken language by the examiner or the participant (both instructions and responses are given non-verbally). The Leiter was administered to children by trained research staff to determine nonverbal IQ score.

The Vineland Adaptive Behavior Scales – Second Edition (Vineland-II; Sparrow, Cicchetti, & Balla, 2005) The Vineland-II is a standardized measure of adaptive functioning for individuals of any age. A primary parent/caregiver with knowledge of the child’s everyday routines and skills was asked to complete the parent/caregiver rating form. The Adaptive Behavior Composite score combines results from communication, daily living skills, socialization, and motor skills domains to provide an overall score of the child’s functioning level. Lower scores indicate greater impairment in adaptive functioning.

Child Behavior Checklist (CBCL; Achenbach & Rescorla, 2000, 2001) The CBCL is a 113-item parent report measure of children’s psychiatric and behavioral functioning. A primary parent/caregiver was asked to complete the CBCL 6–18. The CBCL provides composite scale t-scores for Internalizing and Externalizing Problems. Higher scores indicate more problems in these areas.

Procedure

Emotion elicitation tasks

Participants engaged in a series of tasks adapted from the Laboratory Temperament Assessment Battery – Preschool Version (Lab-TAB; Goldsmith, Reilly, Lemery, Longley, & Prescott, 1993). Although these tasks were originally designed for use with younger children, care was taken to select materials that were likely to appeal to a broad developmental range (e.g., engaging toys with sensory aspects). Additional adaptations consisted of simplifying the language the examiner used with the tasks, selecting materials safe for use in an inpatient setting, and shortening tasks. Aspects of the original tasks meant to be completed by parents were replaced by the research assistant to improve standardization. Finally, we modified portions of tasks where participants were meant to be left alone. Instead, a research assistant and an inpatient staff member remained with the participant at all times, but were neutral and unengaged (e.g., looking at a clipboard). Both the research assistant and the staff member were known to the child prior to engaging in the tasks.

The procedure took place in the psychiatric inpatient unit in a room familiar to the participant. Prior to beginning the tasks, participants were introduced to the protocol through a “social story” that included pictures of the procedures and simple descriptions of what they would be doing. Participants were also shown a visual schedule of activities (similar to that depicted in Figure 1) and had the opportunity to check off activities as they completed them. Participants completed five tasks meant to provoke various emotions, both positive and negative, in a fixed order. Between each task participants watched a video with neutral valence and arousal for one minute. Figure 1 displays a visual depiction of the entire procedure. As can be seen in the Figure, the procedure began and ended with a positive task. Because we were interested in characterizing negative reactivity, only the video periods (baseline; recovery) and frustration tasks are included in the present analysis. Tasks are described in detail below.

Figure 1.

Emotion Elicitation Procedure

Baseline/Recovery (~1 minute)

Between each task, participants watched a one-minute video clip with neutral valence. The videos did not have sound and displayed drops of colored ink moving through water on a white background. For the purposes of the present paper, the video shown immediately prior to the onset of frustration tasks (see Figure 1) was considered the child’s “Baseline” (i.e., reactivity immediately prior to participating in frustration tasks). Videos shown immediately following each frustration task were considered “Recovery” periods.

Tower of Patience (~5 minutes)

The participant was asked to take turns with the examiner stacking toy blocks to build a tall tower. Instructions were explained verbally and visually (i.e., the examiner demonstrated taking turns). Once turn-taking began, the examiner waited increasingly longer intervals before putting on a block, making the child wait their turn for longer periods of time (5, 10, 15, 20, and 30 second waits). In the period between the 15 and 20 second wait times, the examiner also removed three blocks. This procedure was repeated twice.

Transparent Box (~3 minutes)

The examiner allowed the child to choose which toy he or she wanted to play with (pin art toy, light-up balls, etc.), then immediately placed the chosen toy in a transparent locking box. The examiner locked the box with the toy inside and presented a different set of keys to the child that would not open the box. The examiner completed work for two minutes while the child attempted to open the box. If the child asked for help, the examiner encouraged the child to keep trying. After two minutes, the examiner gave the child the correct keys to access the toy and apologized for giving them the wrong set when the task began.

End of the Line (EOTL; ~2 minutes)

The examiner presented the participant with a ramp toy, two small cars, and a ball. Once the child was engaged in play with the examiner, the examiner abruptly took the toy away and exclaimed that they did not want them to play with the toy anymore. The examiner did not respond to attempts by the child to regain access to the toy. If the child asked why the examiner took the toy away, the examiner would answer “Because that’s the way it is.” The examiner kept the toy for 30 seconds before allowing the participant to resume playing.

Coding

To assess participants’ reactions to the emotion elicitation procedures, an observational behavioral coding system was adapted from the Facial Expression Coding System (FACES; Kring & Sloan, 1991) and original Lab-TAB codes. Utilizing Elan Coding Software (Version 5.1; Nijmegen, 2018), four trained RA’s, naïve to study hypotheses, watched video-recorded emotion elicitation sessions and assigned numerical codes to corresponding expressions of affect observed through facial, bodily, and vocal expressions and actions.

Level of engagement (LOE) was coded for each task on a binary scale (1 = participant was engaged; 0 = participant was not engaged). Engagement did not correspond to enjoyment but rather the child’s focus and involvement in the task. Beginning with a possible one-minute coding interval for affect, successively shorter intervals were trialed until a length of time was found that balanced efficiency (longest possible) with loss of information (likelihood of multiple affect codes within the period). A 10-second interval was selected. Valence and intensity of affect were coded for each 10-second interval on a scale of −3 (highly negative) to +3 (highly positive). A code of 0 indicated neutral or flat affect; a code of −1 or +1 indicated subtle, fleeting, ambiguous negative/positive affect; a code of −2 or +2 indicated an obvious expression of negative/positive affect (e.g., a clear frown or smile); and a code of −3 or +3 indicated a high intensity expression of negative/positive affect typically involving multiple indicators of emotion (e.g., smile + laughing). Coders used facial expressions, vocalizations, and body movement to help calculate valence and intensity of reactions. All available contextual information (e.g., facial expression, tone of voice/sounds/words, gestures, and actions) was considered to assign the most accurate coding designations possible, and the most clear and obvious indicators of emotion were weighted most heavily.

Training and Reliability

When coding was initiated, group coder meetings were held over a nine-month period to establish understanding of the codes and reliability while viewing videos of participants with a range of ages, functioning levels, and reactions. All videos were consensus coded until reliability of at least .80 was achieved for independent coding. A subsample of 18 videos (12%) were double coded for reliability purposes. Kappa for coding of Level of Engagement (0 or 1) was .72 (91% agreement across 110 tasks). Interclass Correlation Coefficient (ICC) for coding of affect was .75. ICCs for affect variables (see descriptions in “Data Reduction” section below) were .48 for proportion of intervals with negative affect, .54 for range of affect, .77 for mean intensity of negative affect, and .66 for mean length of negative affect, indicating adequate to good reliability.

Data Reduction

Due to natural variations in length of study procedures (see descriptions above), all calculated variables were independent of time. The following variables were calculated based on coded affect data: Proportion of intervals with negative affect (NA) was calculated by dividing the number of 10-second intervals coded −1 or below by the total number of intervals in the task(s). Mean intensity of NA was calculated by taking the absolute value of the mean of all intervals where NA was present (Affect < 0). This variable provides a sense of the intensity of NA when NA was displayed. Note that individuals who displayed no NA during a given task were given a score of 0 for mean intensity of NA. Mean length of NA was calculated based on the mean length of “runs” (i.e., consecutive intervals) of NA (including instances where NA occurred in only a single interval). Finally, Range of affect was calculated by taking the difference between the most negative and most positive intensity affect displayed [from −3 (highly negative) to +3 (highly positive)].

Out of 1,050 total possible tasks (150 participants, 7 tasks), data were excluded or missing from a total of 40 tasks (3.8%) due to equipment malfunction (N = 28 tasks), safety concerns (N = 5 tasks; all from the same participant), or participant refusal/lack of cooperation (N = 7 tasks). In addition, data from Frustration Tasks were only included in analyses if the participant was coded as “engaged” during the task. Table 2 displays a summary of missing/included data. As can be seen in the table, the vast majority of children were engaged with the tasks, and only 10 total tasks (<1%) were excluded from analyses due to lack of engagement. Note that the Transparent Box task was deemed invalid for 20 children because they were able to get the locked box open themselves despite not having the correct key (e.g., lock malfunctioned, box broke). One participant was not able to watch any of the videos due to a problem with the DVD player, one participant was not engaged during any of the frustration tasks, and one participant discontinued the protocol after the Tower of Patience due to safety concerns. Thus, 149 children had data for Baseline, 149 children had data for Frustration tasks, and 148 children had data for Recovery tasks.

Table 2.

Included and excluded task data

Task	Total Included	Missing/Excluded	LOE = 0
Baseline	149	1	-
Tower of Patience	147	0	3
Recovery 1	147	3	-
Transparent Box	123	24	3
Recovery 2	148	2	-
EOTL	142	4	4
Recovery 3	144	6	-

Note. LOE = Level of Engagement

Data Analysis

Hierarchical linear modeling

Hierarchical linear modeling (HLM) was used to examine change in affect variables from baseline to frustration tasks and recovery, as well as the moderating effect of demographic (i.e., age, biological sex) and characteristics (i.e., non-verbal IQ, adaptive behavior, autism severity, verbal ability, internalizing and externalizing problems) on reactivity. HLM permits analysis of nested, hierarchically structured data (i.e., multiple tasks nested within individuals) to assess both change within individuals across tasks (Level 1) and variation between individuals in that change (Level 2). HLM can accommodate missing data, thus all available data can be used without the need for listwise deletion (Huttenlocher, Haight, Bryk, Seltzer, & Lyons, 1991; Willett, Singer, & Martin, 1998).

Level 1 included data from each task with coded data from the Tower of Patience task, Transparent Box task, and EOTL task dummy coded as “Frustration Tasks,” and coded data from the videos shown immediately following each frustration tasks (video 3, video 4, and video 5, see Figure 1) coded as “Recovery” tasks. The “Baseline” task served as the reference category. In a first set of analyses, HLM was used to estimate individual change in affect (i.e., proportion NA, mean intensity NA, mean length NA, total range affect) as a function of task type (i.e., baseline, frustration, recovery). The basic model is shown in Eq. 1:

$$Y_{ti}={\beta }_{00}+{\beta }_{10}\left(FrustTsk{s}_{ti}\right)+{\beta }_{20}\left(Re{c}_{ti}\right)+r_{0i}+r_{1i}*FrustTsk{s}_{ti}+r_{2i}*Re{c}_{ti}+e_{ti}$$

Here, β₀₀ represents baseline affect for child i, β₁₀ represents change in affect from baseline to frustration tasks, and β₂₀ represents change in affect from baseline to recovery tasks. Coefficients from the frustration task and recovery terms were modeled as random effects in all models (r_0i ,r_1i , r_2i ). Follow up analyses with frustration tasks entered as the intercept examined change from frustration tasks to recovery tasks.

A second set of analyses examined whether demographic (i.e., age, biological sex) and other characteristics (i.e., non-verbal IQ, adaptive behavior, autism severity, verbal ability, internalizing and externalizing problems) moderated change in affect from baseline to frustration tasks and/or recovery. A series of analyses were run with each demographic/characterization variable included at Level 2. Verbal ability (based on ADOS module) was dummy coded with non-verbal/single word (Module 1) serving as the reference group. To increase statistical power and precision, individuals with fluent speech (receiving Module 3 or Module 4) were combined into a single group. Due to significant differences between verbal ability groups in age (F(3,145) = 16.109, p<.001), age was included as a covariate at Level 2 in analyses of the moderating effect of verbal ability.

Results

Overall Change in Affect During Frustration Tasks and Recovery

Table 3 and Figure 2 display descriptive statistics for each affect variable during baseline, frustration tasks, and recovery periods. Table 4 displays the results of HLM models examining change in each variable from Baseline to Frustration Tasks and from Baseline to Recovery.

Table 3.

Descriptive Statistics for Affective Variables during Baseline, Frustration Tasks, and Recovery.

		Proportion Negative Affect		Intensity Negative Affect		Range Affect		Mean Length Negative Affect
	N	Mean	(SD)	Mean	(SD)	Mean	(SD)	Mean	(SD)
Baseline	150	0.16	(0.24)	0.51	(0.62)	1.3	(1.15)	0.81	(1.31)
Tower	147	0.18	(0.2)	1	(0.6)	3.16	(1.27)	1.55	(2.08)
Recovery 1	150	0.18	(0.27)	0.56	(0.66)	1.38	(1.15)	0.9	(1.45)
Lockbox	147	0.22	(0.25)	0.81	(0.71)	2.04	(1.32)	1.72	(2.71)
Recovery 2	150	0.18	(0.28)	0.53	(0.68)	1.39	(1.25)	0.94	(1.6)
End of the Line	146	0.21	(0.21)	1.04	(0.58)	2.69	(1.24)	2.02	(1.92)
Recovery 3	150	0.18	(0.28)	0.5	(0.65)	1.27	(1.26)	0.84	(1.39)

Figure 2.

Change in Affect Variables Across Tasks.

Table 4.

HLM Model Results

		β	SE	t (df)	P
Proportion Negative Affect	Baseline, β00	0.16	0.02	7.96 (149.79)	<.001
	Frustration Tasks, β10	0.04	0.02	2.12 (148.61)	0.035
	Recovery, β20	0.01	0.02	0.58 (162.73)	0.575
Mean Negative Affect Intensity	Baseline, β00	0.5	0.05	9.75 (154.06)	<.001
	Frustration Tasks, β10	0.46	0.05	8.38 (148.86)	<.001
	Recovery, β20	0.02	0.05	0.36 (198.32)	0.723
Range Affect	Baseline, β00	1.31	0.1	13.55 (249.67)	<.001
	Frustration Tasks, β10	1.35	0.1	13.38 (353.83)	<.001
	Recovery, β20	0.03	0.1	0.3 (680.26)	0.767
Mean Run Length	Baseline, β00	0.8	0.13	6.43 (297.14)	<.001
	Frustration Tasks, β10	0.96	0.16	5.87 (187.43)	<.001
	Recovery, β20	0.07	0.14	0.47 (459.7)	0.637

As can be seen in Table 4, results for all four variables showed the same pattern. Models revealed significant increases in proportion NA, mean intensity NA, mean length NA, and total affect range from baseline to frustration tasks, and no significant change from baseline to recovery tasks. Follow up analyses rotating the intercept to examine change from frustration tasks to recovery revealed significant change from frustration tasks to recovery for all four variables (proportion NA: β = −.03, SE = .01, t(322) = −2.48, p = .014; intensity NA: β = −.44, SE = .04, t(516.2) = −12.20, p < .001; length NA: β = −.90, SE = .11, t(193) = −7.90, p < .001; affect range: β = −1.32, SE = .07, t(313.77) = −18.41, p < .001).

Relationship between demographic/characterization variables and frustration/recovery

The next set of analyses examined the moderating effect of demographic (i.e., age, biological sex) and characterization variables (i.e., non-verbal IQ, adaptive functioning, autism severity, social-emotional/behavioral problems, and verbal ability) on affect during frustration and recovery. Results of analyses examining each variable are reported in turn below.

Age

Age was a significant predictor of NA during baseline for proportion NA (β = −.016, SE = .005, t(148.7) = −2.97, p = .004), mean intensity NA (β = −.047 SE = .013, t(155.8) = −3.52, p < .001), and mean length NA (β = −.075, SE = .03, t(318.4.7) = −2.23, p = .03), but did not predict change in these variables from baseline to frustration tasks or baseline to recovery. These results indicate that younger children displayed more, more intense, and longer periods of NA overall, but younger age was not a significant predictor of change in affect.

Given the use of standardized tasks across a large range of ages, we ran two sets of additional analyses to a) verify whether change in affect from baseline to frustration tasks was evident in both younger and older children and b) determine whether younger and older children differed in their response to the three frustration tasks (Tower, Lockbox, EOTL). In order to establish whether both younger and older children displayed change in affect from baseline to frustration, we split the data by median age (13 years) and reran our initial HLM analyses including only children 13 years or younger and only children older than 13 years, respectively. For all four variables, rated affect during the frustration tasks were greater than affect during baseline for both the younger and older age groups.

Next, we compared younger and older children’s responses to the three frustration tasks. Figure 3 displays the mean affective responses to each task split by age. As can be seen in the figure, and consistent with the results of our initial HLM analyses, younger children had larger reactions to frustration tasks overall. However, this was generally consistent across tasks. HLM analyses predicting each of the four affect variables with frustration task type (Tower, Lockbox, EOTL), age group (<13 years or >13 years), and the interaction between these two variables, confirmed no interaction effects between frustration task and age for any of our dependent variables.

Figure 3.

Response to Frustration Tasks by Age Group

Biological sex

Biological sex significantly moderated change in mean intensity of NA from baseline to recovery, with male participants showing a significantly greater increase from baseline to recovery (decreased recovery) than female participants (β = −.26, SE = .13, t(203.6) = −2.04, p = .04). Sex was not a significant predictor of baseline or change in any other variable.

Non-Verbal IQ

Non-verbal IQ was not a significant predictor or moderator of change for any of the affect variables.

Adaptive Behavior

Adaptive behavior level moderated change from baseline to frustration tasks and from baseline to recovery for proportion NA (frustration tasks: β = −.004, SE = .002, t(112.5) = −2.47, p = .015; recovery: β = −.003, SE = .002, t(126.4) = −2.21, p = .029) and mean intensity NA (frustration tasks: β = −.009, SE = .005, t(112.1) = −2.03, p = .04; recovery: β = −.01, SE = .005, t(157.84) = −2.29, p = .02). Individuals with lower adaptive functioning displayed a significantly greater increase in amount and intensity of NA from baseline to frustration tasks and from baseline to recovery. In addition, adaptive behavior significantly moderated change from baseline to recovery for total range of affect (β = −.02, SE = .009, t(628.3) = −1.99, p = .047) such that individuals with lower adaptive functioning displayed significantly greater increase in NA from baseline to recovery.

Autism Severity

Autism Severity score was not a significant predictor of baseline NA or moderator of change in any of the affective variables.

Psychiatric and Behavioral Problems

Two separate HLM models examined the moderating effect of CBCL Internalizing and Externalizing subscales respectively. Neither the CBCL Internalizing Subscale nor the Externalizing Subscale predicted baseline NA or change in affective variables from baseline to frustration tasks and recovery.

Verbal Ability (ADOS Module)

Significant moderating effects of verbal ability were found for proportion NA, intensity NA, and length NA. Figure 4 displays change from baseline to frustration tasks to recovery for individuals in each of the verbal ability categories for these three variables. Controlling for participant age, individuals with phrase speech (ADOS module 2) and individuals with fluent speech (ADOS modules 3/4) displayed significantly less increase in proportion of NA from baseline to frustration tasks (module 2: β = −.15, SE = .068, t(145.6) = −2.27, p = .02; module 3/4: β = −.13, SE = .045, t(147.7) = −2.77, p = .006) and from frustration tasks to recovery (module 2: β = −.18, SE = .066, t(160.7) = −2.84, p = .005; module 3/4: β = −.13, SE = .045, t(161.5) = −2.95, p = .004) than non-verbal individuals (ADOS module 1). Individuals with phrase speech also displayed significantly less increase in mean intensity of NA from baseline to frustration tasks than non-verbal participants (β = −.49, SE = .18, t(144.6) = −2.66, p = .009), and individuals with phrase speech and individuals with fluent speech both showed significantly less increase in mean intensity of NA from frustration tasks to recovery than non-verbal individuals (module 2: β = −.46, SE = .18, t(213) = −2.57, p = .01; module 3/4: β = −.30, SE = .12, t(214) = −2.44, p = .015). Finally, individuals with fluent speech showed significantly less increase in mean length of NA from baseline to frustration tasks than non-verbal individuals (β = −.96, SE = .38, t(190.3) = −2.52, p = .013).

Figure 4.

Moderating effect of verbal ability on affect during baseline, frustration, and recovery.

Discussion

This study was designed to explore relationships between individual characteristics and observable emotional reactivity during baseline, frustration tasks, and recovery in a heterogenous sample of psychiatrically hospitalized youth with ASD. Our results support the feasibility and effectiveness of modified Lab-TAB tasks at eliciting emotional responses in a heterogenous sample of individuals with ASD across a wide age range. Further, while successful in allowing for the observation of negative affect, the paradigms were found to be safe, in that, even among an inpatient sample, only 1 of 150 sessions needed to be terminated early due to safety concerns (physical aggression). In contrast to previous research that focuses only on emotional reactivity during frustration tasks (e.g., Jahromi et al., 2012; Macari et al., 2018), we coded both valence and intensity of observed affect in frequent intervals across the course of the entire testing episode, allowing us to describe emotional reactivity in terms of amount, intensity, duration, and range, and to examine change in affect across contexts (baseline to frustration tasks to recovery).

With regard to general reactivity, our participants increased the amount, intensity, and duration of negative affect from baseline to frustration tasks and displayed a subsequent reduction in negative affect during recovery. On the whole, participants’ affect did not differ between baseline (prior to the start of any frustration tasks) and recovery periods (occurring immediately following frustration tasks), suggesting negative affect associated with frustration tasks was transient. The largest changes in emotional reactivity were seen in the intensity and duration of negative affect and in the total range of affect displayed, and less so in the overall amount (i.e., proportion of intervals) of negative affect (see Figure 2). This may be due to the fact that mild negative affect was not uncommon during the baseline and recovery periods (occurring in about 16% of intervals), which is consistent with prior research suggesting overall increased negative emotionality in individuals with ASD (Capps et al., 1993; De Pauw et al., 2011; Garon et al., 2009; Joseph & Tager-Flusberg, 1997).

Another primary aim of the study was to identify factors related to within-ASD heterogeneity in emotional reactivity. As reported elsewhere (Berkovits et al., 2017; Macari et al., 2018), nonverbal IQ and autism severity (i.e., ADOS severity score) were also unrelated to affective reactivity in our sample. Consistent with previous literature on typically developing individuals (Murphy et al., 1999), several aspects of emotional reactivity did relate to age. Younger children in our sample displayed more frequent, more intense, and longer duration of negative affect overall; however, age did not moderate change in affect across tasks (i.e., from baseline to frustration tasks or recovery). Thus, while younger children were found to respond more negatively overall, they did not differ from older children in their reactivity from baseline to frustration or recovery. Although previous research has pointed to improved emotion regulation (and decreased reactivity) with age, to our knowledge no prior study has specifically examined change in affect from baseline to recovery during a regulatory challenge (frustration task).

The results reported here highlight the importance of examining affective responses across changing contexts in order to distinguish between general trait differences in negativity (observable at baseline) and differences in responsivity to frustration. Furthermore, previous observational work on response to frustration in individuals with ASD has focused largely on young children (toddler or preschool aged children), thus limiting opportunities to examine relationships to age across a large range. Importantly, however, our findings are cross-sectional and thus do not specify change in reactivity within individuals across development. Longitudinal studies are needed in order to better understand how negative emotionality, response to frustration, and recovery from frustration relate to age.

Previous research on non-ASD samples has reported that boys display more anger than girls overall (Chaplin & Aldao, 2013). In the present study, we did not find a difference between male and female participants in the amount of negative affect displayed during frustration tasks; however, male participants showed a significantly greater difference between baseline and recovery in intensity of negative affect than female participants. Specifically, our results suggest that rather than returning to baseline levels of negative affect, male participants remained at elevated levels of negative affect following a frustration task compared to baseline. We are unaware of any prior research examining sex differences in emotional reactivity in individuals with ASD using observational methods. Thus, this study provides new preliminary evidence that females with ASD may recover from frustration (at least in terms of observable displays of negative affect) more readily than their male counterparts.

Contrary to our expectations, parent-reported psychiatric and behavioral problems, as indexed by the Internalizing and Externalizing subscale scores from the CBCL, were not related to reactivity or recovery in our observational paradigm. This is in contrast to results reported by Berkovitz et al. (2017) demonstrating associations between the CBCL Total Problems score and parent-report measures of emotion regulation, as well as other reports linking poor emotion regulation to specific psychiatric symptoms (e.g., anxiety/depression) in ASD (see Cai et al, 2018 for review). Notably, a unique sample of psychiatrically hospitalized youth distinguishes our research from these prior studies. As evident in Table 1, scores on the CBCL were quite high in this sample (scores greater than 1.5 standard deviations above the normative sample mean) and therefore likely not reflective of the variability in psychiatric/behavioral problems seen in the general population. This restricted range of scores may have dampened our ability to detect an association. In addition, one strength of the present study is that emotional reactivity is measured observationally rather than relying solely on parental report, thus removing concerns about shared informant variance inflating correlations between measures. That being said, the lack of a relationship between observed emotional reactivity and psychiatric/behavioral problems is surprising and further research is needed to confirm this null result.

The two most important variables predicting reactivity and recovery in our sample were adaptive behavior and verbal ability. Participants with lower adaptive behavior levels displayed greater reactivity (with regards to amount and intensity of negative affect) and less recovery (with regards to amount, intensity, and range of affect) from frustration tasks than individuals with higher adaptive behavior levels. One possibility is that heightened emotional reactivity hinders adaptive behaviors, either directly or indirectly. For example, increased frustration may lead some children with ASD to give up on tasks more readily, thus limiting acquisition of adaptive behaviors in the first place. This interpretation is consistent with the findings of Jahromi et al. (2012) wherein preschool aged children with ASD displayed more resignation in response to frustration than typically developing peers. It is also consistent with findings from Jahromi, Bryce and Swanson (2013) that executive functioning skills predict emotion regulation. Difficulties with executive functioning are likely to impact both ER and adaptive behavior. Alternatively, or additionally, parents of children who engage in more negative reactive behaviors may reduce demands on their children over time, further decreasing opportunities to develop adaptive skills. It could also be the case that underdeveloped adaptive behaviors could lead to increased negative reactivity as individuals with lower adaptive behavior may have fewer strategies for dealing with frustration. This is consistent with literature showing a relationship between adaptive coping skills and severity of problem behaviors (i.e.. SIB, stereotyped behavior, and irritability) in youth with ASD (Williams et al., 2018).

With regard to language ability, participants who were non-verbal or had only single-word speech (ADOS-2, Module 1) were more reactive and showed less recovery from tasks than individuals with phrase speech and fluent speech. It is possible that a more nuanced assessment of verbal or communication ability would yield different information. However, prior research has not found a relationship between a continuous measure of language and parent-report of emotion regulation (Berkovits et al., 2017). The present study indicates a large difference between having minimal or no language and having even limited phrase speech, suggesting that the relationship between language and emotional reactivity may not be continuous. It may be that minimally verbal individuals do not have access to coping mechanisms that more verbal individuals have access to, such as self-talk, seeking support from others, ability to use functional communication supports or communicating needs/desires (Cole, Armstrong, & Pemberton, 2010). Indirect evidence of the importance of minimal verbal ability in relation to emotional reactivity and the challenging behaviors that often accompany it can be drawn from the overrepresentation of minimally verbal youth in inpatient psychiatric units (Siegel et al, 2015), for which aggression, SIB, and tantrums are the most common reason for admission (Siegel et al, 2011). Difficulties regulating emotions may also interfere with language learning and communication. This is consistent with theory and research from the typically developing literature suggesting a bidirectional relationship between language ability and emotion regulation skills (Cole 2010; Stansbury & Zimmerman, 1999; Dixon & Smith, 2000). Thus, our results highlight the importance of including minimally verbal individuals in research, their vulnerability to developing poor emotion regulation, and the need to develop effective interventions for this population, particularly interventions that directly address difficulties with emotion regulation.

Limitations

While this study has a number of strengths, including high density observational coding and a unique population of severely affect youth with ASD, a few limitations must be noted. First, the generalizability of our findings to a non-hospitalized sample of youth with ASD is unclear and requires further research. Relatedly, the lack of a comparison group limits our ability to determine whether the severity and correlates of reactivity reported here are specific to ASD individuals, specific to hospitalized individuals (with or without ASD), or consistent across non-ASD and non-hospitalized youth.

Second, choosing and implementing a set of frustration tasks that are equally appropriate and effective across a wide range of individuals (with regards to both age and cognitive/verbal ability) is challenging. The alternative would be to utilize different tasks and materials for different age ranges or developmental levels; however, this could introduce issues with comparability that would be difficult to quantify. As such, we selected tasks and materials that were most likely to be understood by our most impaired participants and yet still engaging to our more able or older participants. We were encouraged that both younger and older participants reacted to our selected frustration tasks and that non-verbal cognitive ability did not relate to reactivity. That being said, our observed effect sizes for reactivity may have been limited due to the use of frustration tasks initially designed for preschoolers that may not have elicited strong reactions in all participants.

Relatedly, establishing a true measure of “baseline” reactivity is complicated. For the purposes of this paper, we chose to use affect during a neutral activity (watching a one-minute video of colored ink on the screen) that occurred immediately prior to the start of frustration tasks as our “baseline” measure. We acknowledge that this method comes with potential problems, including possible carryover effects from the task immediately preceding this “baseline” period as well as the possibility that some children may have had a negative response to the video (e.g., boredom). Unfortunately, however, all potential means for measuring “baseline” are problematic, thus we felt a standardized task was the best option to improve comparability across children and sites. Regardless, we believe the changes in negative affect we report from this initial period to the task periods are reflective of reactivity to frustration, even if we remain agnostic to whether our “baseline” is truly the child’s baseline.

Finally, observational coding of affect from video recordings is a difficult and time-consuming task. It should be noted that some variables (i.e. range of affect, proportion intervals with negative affect) had lower inter-rater reliability and therefore some caution is warranted in interpreting results.

Conclusions and Future Directions

The present study provides new insight into the features and correlates of emotional reactivity in more severely affected individuals with ASD. We demonstrate the feasibility of using emotion eliciting tasks in individuals with ASD across a wide range of ages, severity, and verbal, adaptive, and cognitive abilities, as well as the ability to utilize these paradigms safely with individuals with emotional and behavioral challenges, such as those residing in a psychiatric inpatient unit. Relationships between emotional reactivity and adaptive behavior and verbal ability point to the importance of including individuals with ASD across a wide range of functioning levels in research studies. Further longitudinal research is needed to better delineate the causal and developmental mechanisms at play; however, our results highlight the need for interventions to consider the inter-dependence between these areas of functioning. For example, interventions that target both emotion regulation and communication strategies simultaneously may be most effective in improving functioning in both of these related domains. In addition to observational coding of affect, as done in the present study, there may be additional objective measures of distress, such as physiological responses, that can shed light on the relationships between individual characteristics within the ASD phenotype, distress, emotional reactivity, and observable behavior.