Understanding the structure of children’s emotion-regulation strategies

Models of emotion regulation

Most current models of ER acknowledge the role of individuals, their goals, and their context in eliciting emotions and the appropriate deployment of regulation strategies (Aldao, 2013; Opitz, Cavanagh, & Urry, 2014; Shields & Cicchetti, 1998; Thompson, 1994). However, these models often do not emphasize the importance of goals or context in their structure. Gross (1998, 2002) proposed the process model of ER, the emphasis being the deployment of strategies in relation to the occurrence of emotion. Although this model details important, higher-order ER clusters, it fails to address critical factors that would be useful for guiding the selection of ER behaviors. As focus is placed on emotion-generation in relation to timing, the same observable strategy used at different stages in the process model is viewed as two differing strategies. Additionally, this model is silent on how context determines the effectiveness of ER strategies. The concepts of coping and ER have common characteristics, but not to the point of redundancy (Gross, 1998). Coping is often applied to non-emotional phenomena and implies stress (Skinner & Zimmer-Gembeck, 2007). However, emotion regulation does not automatically assume either a stressful event or a negative emotion to be managed. Emotion regulation often falls under the umbrella of coping, yet coping with heightened negative and positive emotions is but one part of ER (Kopp, 1989; Skinner & Zimmer-Gembeck, 2007). Coping focuses on managing distress and the coping context is usually experienced as negative (Gross & Thompson, 2007) whilst coping is usually considered to be a positive achievement. However, ER is neither a positive nor negative task in itself (Gross, 2002).

Callear and Harvey’s goal-directed model of ER (described in more detail in Callear, 2014) emphasizes the importance of goals and context in its structure. This model outlines how emotion and ER arise in the context of an emotion-eliciting event. The event elicits emotion by causing a shift in a person’s relationship with an individual goal. The quality and intensity of the emotion aroused depends on how great the perceived shift is and the individual characteristics of the person experiencing the shift. These individual characteristics include interpersonal, contextual and intrapersonal factors. Emotion regulation occurs to help the individual manage the emotions that occur with the relationship shift, and strategies can involve behavioral, interpersonal, experiential, cognitive, and/or physiological measures. Callear and Harvey’s model was used to filter items because it enabled the identification of specific observable ER strategies while emphasizing the objectives driving these and the child’s context in shaping the function and deployment of behavioral strategies.

The current research

The current research focuses on two key questions. First, what specific and observable ER behaviors do typically-developing children use? Second, how could these behaviors be examined so their relationships can be identified to inform a unified theoretical model? The goals of this research were to capture children’s ER in a way that a) represents the rich complexity found in children’s ER; b) samples a full range of ER strategies of middle childhood and; c) provides a data format that could have later applications, including potential use as a children’s ER measure. Rather than including private strategies, a key decision was made to develop an inventory of externally observable ER behaviors. The goal-directed model guided the appropriateness of strategy descriptors from various data sources, the language to frame items, and the decisions around item inclusion.

Two studies were undertaken to examine these research questions. The first uses focus group data, literature examination and an examination of the clinical and research measures to identify a domain of items that comprise an inventory of children’s ER strategies. The second validates domain coverage and uncovers the conceptual structure and inter-item relationships within the item domain. A multidimensional scaling (MDS) approach examines this by using similarity estimates of the inventory items (see the Supplementary material for more detail).

Study one

Method

An extensive review was undertaken of items of partially or wholly relevant standardized clinical and research measures (see Supplementary material; Table 1). A list of 404 children’s ER strategy descriptors was generated. Each descriptor item was assessed for inclusion based on whether it fit the definition of ER described in the goal-directed model of ER (Callear, 2014), whether it could be framed as a behavior independent of goals and/or context, and whether the item could be observed and reported by another.

Focus groups were held with professionals from a range of disciplines who work closely with children in the 6–12-year age group. This age range was chosen because of the importance this period has in the development of ER (Holodynski & Friedlmeier, 2006) and is an age well-suited for ER-focused interventions (Riggs, Greenberg, Kusché, & Pentz, 2006).

Participants

Four focus groups were run with 23 New Zealand-based professionals (19 female; 17 Pākehā/New Zealand European, 4 Māori, 2 from outside of New Zealand; Mean_age = 45.07) from a range of medical, psychological, educational, social work, and counselling/psychotherapy backgrounds, and who had at least 3 years’ experience working predominantly with children aged 6 to 12 years. Participants were located by means of a snowball method initiated through individuals and organizations known to work with children in this age group.

Procedure

Focus group discussions started with a transition question (Krueger & Casey, 2009) designed to initiate participants’ thinking about the purpose of ER. The remainder of each session was spent discussing four broad topic areas:

Strategies children use when dealing with emotions or emotional responses.

The kinds of ER strategies considered to be beneficial or useful.

The strategies considered to be counterproductive or unhelpful.

The signs that a child is having trouble regulating his/her emotions.

Data collection, transcription and analysis

Focus groups were audio recorded and transcribed. An iterative coding process was used to identify and extract specific words or phrases used to describe examples of ER in the transcripts (O’Brien, 1993). These descriptors were compiled with the items extracted from the instruments discussed above and, using a constant comparative analytic framework, sub-grouped in terms of their conceptual similarity (Krueger & Casey, 2009). The 226 groups derived from this process formed the first draft of the ER item list. After reviewing the location and context of items within the transcripts, the groups were then subcategorized using an iterative approach until no additional meaningful distinctions between the strategies could be found (Parkinson & Totterdell, 1999).

The draft inventory was reviewed by two senior independent collaborators, then sent to the original focus group participants for item review and feedback, including assessment of language use, overlapping items, domain coverage and item accuracy.

Study one: Results & discussion

An inventory of 103 other-report children’s ER strategies was generated (see Supplementary material; Table 3). The items resonate with prominent themes apparent in the ER literature. Gross’ process model focuses strongly on the stage of the emotion-generative process in which the ER occurs (Gross & Thompson, 2007). For instance, several items reflect behaviors aimed at selecting or modifying situations (e.g., “temporarily leaves a situation”) and attentional deployment (and situation modification) (e.g., “finds a distraction”). Cognitive change and response modulation could be linked to items like “gives self a pep-talk” and “maintains an expressionless face” respectively, although inference is required for both themes given their internal nature.

Each item’s location on the continuum from automatic to deliberate ER (Gross & Thompson, 2007; Mauss et al., 2007) depends entirely on the individual using it and the situation in which it is applied. Factors which might be significant include how well-practiced a particular strategy is, how early in the child’s life the strategy developed, timing and what socio-cultural norms are in play (Mauss et al., 2007). Thus, a child who “avoids showing certain emotions” may have developed this strategy early in life as a result of biopsychosocial factors, and this child may deploy the strategy unconsciously or automatically (even though it may have originally been a voluntary and controlled strategy). Conversely, the same strategy may be deployed deliberately and consciously by a different child in a different emotion eliciting context.

Many ER items found in other measures are structured based on their orientation towards a specific emotion or a specific context. Examples include: I have angry outbursts (emotion-specific; ERICA; MacDermott et al., 2010); When I felt happy, I could control or change how happy I felt (emotion-specific; How I Feel; Walden, Harris, & Catron, 2003). The items identified here exist independent to the emotion-eliciting event. Each event (and resulting emotion) is unique and completely reliant on the individual characteristics and goals of the person experiencing the emotion.

Having generated a list of 103 observable ER strategies used by children aged 6–12, we next examined inter-item relationships and uncovered any higher order conceptual themes within the inventory domain.

Study two

Method

Participants

To check the generality of results, items are presented to two different sample groups: people with no specific training or expertise in emotional concepts, and people with training in the area. Previous studies have reported that level of expertise in a psychological domain has little effect, with different participant groups drawing upon a similar collective implicit model (e.g., Harvey, Bimler, Evans, Kirkland, & Pechtel, 2012; Kirkland, Bimler, Drawneek, McKim, & Schölmerich, 2004). In the lay-person sort, 30 individuals (16 male, Mean_age = 32.4 years, SD_age = 12.9, Range_age = 21–67) participated. Participants were a sample of convenience and were recruited by means of snowball sampling. In the expert sort, 29 individuals with a postgraduate qualification in psychology and one with a postgraduate qualification in human development (25 female, Mean_age = 35 years, SD_age = 9.5, Range_age = 23–61) participated. Although the male-to-female ratio of this sample is asymmetrical, this gender balance is fairly representative of the population of psychologists in New Zealand (New Zealand Health Information Service 2007, 2006).

Procedure

Participants were volunteers and informed consent was obtained. The sorts were undertaken either individually or in groups of up to four. Participants were given the 103-item deck, a recording sheet, a pencil and a pen. Each item was printed in black ink on the center of a card measuring 35 × 75 mm.

The GOPA sorting task was performed in four phases (Bimler & Kirkland, 2007; see Supplementary material for details). Participants were asked to sort the 103 items into Groups according to similarity of meaning, with no constraints on the number of groups or items per group. In the O-phase, they chose pairs of items which were “most opposite groups of items” (i.e., the collective meanings of either group were antithetical to each other). In the P-phase, groups were partitioned into smaller more homogeneous subgroups. In the A-phase, they chose the most-similar pair of groups and merged these into a larger, more heterogeneous group, repeating this pairwise addition if possible. Participants recorded their responses on data entry forms which were entered into the Data Organiser program (Graybill, 2009). Their judgments were converted to a 103-by-103 matrix containing estimated similarities between each pair of items, for analysis with MDS (for similarity estimates, see Supplementary material; Table 4).

Results

Three-dimensional solutions were optimal for both the lay and the expert data sets, with a large improvement (lower stress₁) from unacceptable two-dimensional models, while proceeding to four and higher dimensions provided only small, progressive increments. The stress₁ indices for lay solutions were 0.245, 0.178, 0.132, and 0.104 for two, three, four, and five dimensions. The corresponding values for expert solutions were 0.235, 0.163, 0.122, and 0.098.

Three tests for comparing the two solutions showed them to be similar, indicating that the lay and expert sorters were basing their judgments on the same underlying mental structure. The Procrustes distance, which measures the residual distances between corresponding items after one map is superimposed on the other and rotated to maximize the overlap between them, was low (g₁ = 0.076). Pearson’s cophenetic correlation between corresponding inter-item distances was strong (r = 0.79). Finally, a Canonical correlations comparison of the two sets of coordinates indicated that each dimension (axis) in one model had a recognizable counterpart in the other, with all three canonical correlations reaching significance (R_c = 0.968; R₂ = 0.806; R₃ = 0.714; p < 0.005).

Given the strong resemblance between lay and expert models, the data were merged into a single similarity matrix (see the Supplementary material). Stress₁ values were 0.223, 0.159, 0.117, and 0.091 for two-, three-, four-, and five-dimensional solutions. Again, three dimensions provide the best trade-off between goodness-of-fit, ease of interpretation, and conceptual clarity. The combined similarities can be fitted with lower stress₁ values than either group in isolation because the increase in numbers reduces the statistical noise. However, the improvement in stress₁ was negligible, suggesting that little benefit would come from recruiting further participants.

The points comprising the combined MDS model, each point representing an item, were concentrated in a hollow spherical shell. Rotating and examining the model revealed a mid-sized void in the otherwise continuous distribution of points (see Supplementary material; Figures 1 and 3). This could indicate a genuine semantic discontinuity between incommensurate subdomains of meaning where items are simply not possible, or alternatively a gap in the inventory where items had been overlooked. Auxiliary research was undertaken in an effort to close the gap.

Auxiliary study

ER concepts of disobedience, chaos, messiness, and uncleanliness showed potential to fill the void. These conceptual areas were generated first by interpolating between items around the edges of the void, and second by inspecting items from the far side of the model and considering their antitheses. Apart from disobedience, these concepts were largely absent from existing instruments and focus-group data. Nine new items sampling these areas were constructed using the same iterative cycle as before (see Supplementary material; Table 3). Additional data were collected to “triangulate” these new items within the original model (Kirkland et al., 2004), while minimizing the impact on already-determined item locations (see Supplementary material; Table 3).

Method

From the original 103-item deck, 26 “landmark” items were extracted and combined with the nine new items, creating a new short 35-item deck. Those landmarks included items adjacent to the void and items diametrically opposite to it, with additional representatives from significant clusters. A mixture of 30 lay and expert individuals (18 females; Mean_age = 33.7 years, SD_age = 9.79, Range_age = 23–58) were recruited by snowball sampling as a sample of convenience. The task described earlier was followed, with some minor changes to reflect the smaller number of items.

Results

Integrating the new data with previous sorts, the three-dimensional solution remained optimal with a stress₁ indices of 0.165 (down from 0.242, for two dimensions), with smaller drops to 0.123 and 0.096) for four and five. Four of the new items fell around the edges of the original void, with the two items expressing aspects of disobedience being drawn into a conceptually-similar cluster, while the other five new items fitted well into the center of the void.

Final map interpretation

In consultation with eight of the original expert participants, the three-dimensional circumplex map was rotated to the most interpretable axes, while identifying and labelling its meaningful features (clusters of points with a common theme or content) (see Supplementary material for further details).

Figures 1 to 5 can be viewed in the Supplementary material. Supplementary Figures 1 and 2 each plot the “dependent” and “independent” hemispheres separately (i.e., items with negative and positive coordinates respectively on the first dimension), for the 103- and 112-item sets. For this purpose, each MDS solution is treated as a true spherical shell, treating the items as all equidistant from its center, then flattening their locations onto the plane with the Stereographic projection, as the two hemispheres of a globe are flattened in an atlas. Supplementary Figure 3 presents the same data in a single perspective view (to be imagined as three dimensional), with and without item labels, where the 103 items can be seen in their relationship with one another. Supplementary Figure 4 presents a similar perspective view of the final 85-item solution, to provide a clearer view of the items inserted by the auxiliary study into the previously-void area in the map. Finally, Supplementary Figure 5 presents two projections of the 85 items, showing the plane defined by the D1 and D3 axes (with D2 perpendicular to that plane), and the D2/D3 plane (with D1 perpendicular). Colors indicate 17 clusters of thematically-linked items.

The resulting dimensional concepts, characterized by key items at each extreme, are as follows (Davison, 1983; Davison & Skay, 1991), with the caveat that their explanatory value does not prove that they were the actual oppositions and gradients of meaning used by the subjects to assess the dissimilarity among the items.

Dependence dimension: dependence-related behavior. The strategies at one end of this dimension involve highly independent behaviors such as “Gets into fights.” Conversely, strategies at the negative end of this pole reflect highly dependent behaviors such as “Uses a comfort item.”

Connection dimension: forging a connection with others. At one extreme, dissociation and avoidance items are found such as “Avoids reminders of emotional experiences.” At the opposite end, items convey the notion of connecting with others (e.g., “Eager to please others”).

Attending dimension: using attention and focus to regulate emotions. At the positive pole, strategies focusing inward and attending deeply to the emotion can be seen (e.g., “Picks at skin or other parts of the body”). Conversely, the opposite end captures strategies reflecting outward-focused, distracting emotion regulating behaviors such as “Does artwork.”

Points in Figure 3 are coded as symbols to show the 21 clusters of conceptually-related, spatially-adjacent items. For this classification, inspection of the map was complemented by Hierarchical Clustering (unweighted mean linkage algorithm), which summarizes the similarity matrix as a tree-like structure or dendrogram (in which separate clusters become branches). Clusters are listed below. They are comparable to the proliferation of factors in a fine-grained Factor Analysis; conversely, the three dimensions are loosely comparable to broader, second-order factors.

a) Affection Seeking, b) Attacking, c) Avoidance, d) Cheerful, e) Compliance, f) Damaging, g) Disobedient, h) Disorganized, i) Displaced Control, j) Dissociation, k) Distraction, l) Dominate, m) Excess/Dramatic, n) Help Seeking, o) Make-believe, p) Self Soothe, q) Somatic, r) Unkempt, s) Verbalize Emotion, t) Solo Play, and u) Active (for details, see Supplementary material; Table 2).

Cluster descriptors were defined as a practical step to capture a conceptual linkage between cluster items but are not used in further MDS analyses. Despite some cluster descriptors superficially having an appearance of ER outcome rather than ER process and even appear to be exclusively used for down-regulation of unpleasant feelings, this is not the case. For example, the Self-Soothe cluster frames items as though they achieve self-soothing, presumably in order to reduce unpleasant feeling states. However, a child is not always experiencing an unpleasant feeling state before and during thumb sucking, and may, in fact, be attempting to maintain or up-regulate a more pleasant feeling state (such as contentment) or even down-regulate a pleasant but unhelpful feeling state (such as excitement).

Item reduction and construction of the ChERI (Children’s Emotion Regulation Inventory)

The final step in developing an inventory was to reduce the number of items for use in examining profiles of children’s ER strategies (Kirkland et al., 2004). Item selection was a three-phased, post-hoc process based on the procedures described by Kirkland et al. (2004). Spatial adjacency in the MDS map and close linkage in the dendrogram are indicators of redundancy, with two items capturing similar concepts. In such cases, the more general, observable item was retained, inspecting map and dendrogram to ensure that the clusters remained represented. After removing 22 redundant items using this process, a random selection of five additional items was removed. When the random sampling fell upon a key item, it was retained with the next item below in the dendrogram removed. The final selection of 85 items forms the ChERI. Sufficient items were retained to allow for a double deck of 40 items as future applications may require a second deck or alternate form to prevent test-retest bias.

Study two: Discussion

The results from Study Two revealed a unified underlying structure to children’s ER strategies. A 112 item three-dimensional, 21 cluster structure was used to create the 85 item Children’s Emotion Regulation Inventory (ChERI). Items are able to be organized as a 9-point scale or Q-Sort ranging from “clearly seen” to “rarely seen” emotion-regulation behavior. We retained three dimensions in the analysis, for ease of visual inspection, having found that they were interpretable, and replicated across data sets. It may be that future examination will reveal that a fourth dimension is also robust, although the diminishing return in terms of stress₁ values indicates that it would account for relatively little variance.

Links with other measures

Elements of ER strategies derived from factor analytic research are reflected in the ChERI. The “lability/negativity” factor in the Emotion Regulation Checklist (ERC; Cicchetti, 2009) for instance, is analogous to the excess/dramatic cluster and its “emotion regulation” factor is similar to the verbalize emotion cluster. Interestingly, the expressive reluctance factor identified by Penza-Clyve and Zeman (2002) is evident in items spread across the map. Importantly, corresponding items did not congregate together to form a dimension (factor) or cluster. Instead, the inverse of this factor appeared to be reflected in the verbalize emotion cluster. The first three of the four factors (active, distraction, avoidance, seeking social support) described in Ayers, Sandier, West, and Roosa’s (1996) model of coping, closely resemble three clusters found in this study. Placing these four factors on two-dimensional axes within the ChERI map, the avoidance (avoidant) cluster and items reflecting the support-seeking factor (“Asks for help or advice” and “Asks for verbal reassurance”) are positioned almost opposite to each other. The distraction and active clusters, however, are directly adjacent, suggesting these two strategy clusters are conceptually and/or tactically close.

Links with theory

The item configuration outlined in this study reveals the underlying conceptual structure of the relationships between the items and clusters of items within the inventory. This conceptual meaning was consistent across two distinct participant samples, demonstrating the shared and robust nature of the underlying structure. Much like Parkinson and Totterdell’s (1999) examination into deliberate emotion regulation strategies by adults, this research provides an objective representation of the structural relationship between items, providing the groundwork for a classification system for children’s observable emotion regulation.

The relationship between items appears to transcend theoretical perspectives and facilitated the emergence of links between these. Often, children’s problem behavior is divided into externalizing and internalizing behaviors (Calkins & Howse, 2004; Lahey et al., 2008). Although these two dimensions were historically described as forming two bipolar ends of the same concept (see, e.g., Achenbach & Edelbrock, 1978; Weintraub, 1973), recently, they have been depicted as two separate and occasionally co-existing dimensions (Lahey et al., 2008). When conceptualized as ER strategies, internalizing and externalizing behaviors were not found along a single dimension. Behaviors traditionally reflecting the internalizing dimension are not diametrically opposed to the externalizing behaviors in the MDS map, but are instead found primarily at the extreme disconnect end on a dimension of connection. At the opposite end of this dimension were ER strategies involving behaviors that attempt to forge connection with others. Moreover, instead of being conceptualized as internalizing, psychosomatic problem behaviors are found at the extreme inwards-focus end of an attending dimension. These results suggest that, although behaviors involving avoidance and withdrawal make up an important part of the internalizing dimension of ER, somatic behaviors are considered separate to both the internalizing and externalizing patterns.

Developmental psychologists describe the development of attentional skills as vital (Rothbart & Sheese, 2007). Attention is depicted here in the third dimension Attending. At one end of this dimension, a range of outwardly-focused distracting strategies are seen. At the other end of this dimension, strategies reflect an inwards focus. Importantly, some inwards-focused strategies appear both somatic in nature (e.g., “Chews fingernails”) and non-somatic (e.g., “Blames self”), suggesting the presence of cognitive strategies to regulate emotion.

Our aim was to develop a substantive, comprehensive inventory of children’s observable ER behaviors. Future applications include the provision of an assessment tool for professionals that augment children’s self-reported ER measures, particularly when self-report is either unsuitable or external validation is required. The ChERI provides a data format that could be used to generate profiles of ER behaviors, identify a range of ER strategies, provide an integrated, yet comprehensive picture of a child’s ER (Bariola et al., 2011), and guide interventions addressing problematic ER. Future research is required to validate the empirical relationships among items, evaluate how items perform in practice, and whether this corroborates with the present MDS solution. In conclusion, this research has successfully generated a coherent representation of children’s observable ER strategies. We hope it will be useful for professionals working with children as well as researchers investigating emotion regulation of children.