Developmental Differences in Child and Adolescent Reasoning about Anxiety Sensations

Abstract

Background:

The development of reasoning as to the potential negative consequences of emotional sensations is a critical aspect of emotion knowledge and central to cognitive risk for anxiety disorders. The purpose of this paper is to explore the reasoning children and adolescents give for negative interpretations of anxiety sensations, testing a priori hypotheses quantitatively and exploring the content of the reasons qualitatively.

Methods:

This study used a cross sectional design with interviews as well as cognitive and emotional assessments in a sample of 227 youth aged 6–17 years. Coding schemes to assess the logical validity, affective valence, and qualitative reasons that youth give to evaluate anxiety sensations and anxiety situations were developed.

Results:

Findings indicated diverse reasoning was used and responses could be reliably coded with developmental differences across age, cognitive, and verbal development. The logical sophistication of the reasoning used by youth increased across age in a non-linear manner and linearly with cognitive and verbal abilities. Child anxiety sensitivity and internalizing symptom levels moderated the main effect of age.

Conclusions:

The results add to the existing understanding of emotional development and are consistent with the idea that the process of cognitive-emotional understanding is not a simple linear one because various domains may show differential development.

The concept of emotion knowledge concerns the understanding of emotional expressions and initial work involved accurate labeling and identification of emotional expressions (Bender, Pons, Harris, Esbjørn, & Reinholdt-Dunne, 2015; Izard, 1971). Emotion knowledge is linked to a number of social and academic outcomes (Izard et al., 2001). The concept of emotion knowledge has expanded to include the understanding of the causes of emotion and understanding ways to regulate emotion (Southam-Gerow & Kendall, 2000). Learning how to regulate one’s emotions is similarly associated with adaptive social functioning and positive psychological adjustment (Eisenberg, Fabes, Guthrie, & Reiser, 2000; Izard et al., 2001). Alternatively, a lack of emotional regulation is associated with greater risk of developing internalizing and externalizing problems (Southam-Gerow & Kendall, 2002). Research implicates both biological and environmental factors in the development of children’s emotion knowledge (Calkins, 1994; Eisenberg & Fabes, 1994; Morris, Silk, Steinberg, Myers, & Robinson, 2007).

As Bender et al. (2015) note, emotion knowledge has received a large amount of empirical attention. They conclude from the extant literature that similar to Piaget’s theory about the development of cognition and reasoning broadly (see Piaget, 1950, 1983), human understanding of emotions also develops from a simple and limited understanding to a more logical and sophisticated understanding (i.e., cognitive epigenesis). It also involves developing a comprehension of the less visible and more cognitive aspects of emotions. Bender and colleagues (2015) also posit a process where youth begin to understand the difference between feeling emotions and the impact or consequences of emotions. A common way to study the development of emotion knowledge is the Test of Emotion Comprehension (TEC; Pons & Harris, 2000). The TEC contains scenarios of various emotional situations and youth are asked to attribute an emotion to the story’s central characters by identifying the most appropriate of four possible emotional outcomes of the story. While the TEC assesses multiple components of emotion knowledge including an understanding of the external causes of emotions (Pons et al., 2014) existing assessments do not focus on the reasoning children and youth use about the consequences of feeling emotions. The development of childhood reasoning as to why an individual might not like a particular emotional sensation (i.e., why an individual may assume a negative consequence of feeling an emotion) is not well understood but is a critical aspect of understanding the development of emotion knowledge. This study aimed to examine children’s reasoning about various potential negative consequences of having or showing anxiety sensations. In the following, the theoretical and practical importance of understanding the development of reasoning about anxiety sensations is presented followed by study goals and hypotheses.

Theoretical and Empirical Background

The study of children’s general reasoning has a long history in developmental psychology (see Piaget, 1950). As noted, the extant literature suggests that the understanding of emotions develops from a simple and limited understanding to a more logical and sophisticated understanding but that individual facets of reasoning may emerge at different times (Bender et al., 2015) as follows from the Piagetian concepts of epigenesis and decalage. These may also apply to children reasoning about the consequences of emotions. Cognitive models of emotional disturbances, such as anxiety disorders, propose that they can emerge from or be fostered by faulty and/or negative reasoning (e.g., Beck, 1976). For example, catastrophizing involves assuming extreme negative consequences of events (e.g., it’s raining so there will be a flood). Catastrophic interpretations of fear and anxiety sensations themselves are a form of negative reasoning associated with elevated anxiety and anxiety disorders. In particular, the concept of anxiety sensitivity (also known as “the fear of fear”) refers to beliefs that feeling or showing fear and anxiety related sensations can have severe negative consequences (Reiss, 1991; Schmidt, Lerew, & Jackson, 1997, 1999). The construct of anxiety sensitivity assumes a meta-cognitive ability; namely, that an individual both understands the consequences of the experience of feeling or showing fear or other anxious emotion, and exaggerates the potential negative impact for themselves (e.g., “My heart is beating fast therefore I may be having a heart attack,” versus, “My heart is beating fast therefore I must be excited.”).

Research on anxiety sensitivity in childhood suggests the reliability and validity of its measurement (e.g. Silverman, Fleisig, Rabian, & Peterson, 1991; Weems, Hammond-Laurence, Silverman, & Ferguson, 1997; Weems, Hammond-Laurence, Silverman, & Ginsburg, 1998). Similar to the adult literature anxiety sensitivity has shown links to and prediction of later anxiety problems (e.g. Hayward, Killen, Kraemer, & Taylor, 2000; Weems, Hayward, Killen, & Taylor, 2002). However, writers have also been critical of the assessment in youth with some suggesting the construct is too meta-cognitive to be valid in younger children (Chorpita & Lilienfeld, 1999). Anxiety sensitivity has been assessed through self-report measures. For instance, the Childhood Anxiety Sensitivity Index (CASI) was developed by Silverman, Fleisig, Rabian, and Peterson (1991) for use with school age children ages 6 to 17. Example questions are “It scares me when I feel shaky,” and “It scares me when I feel nervous.”

To date, it is unclear when and how children begin to employ the reasoning that is integral to anxiety sensitivity theory. Theoretically, the logical sophistication in the reasoning about what the consequences of anxiety sensations are - should evolve over the period of childhood into adolescence. A clearer understanding of the developmental progression of the logical reasoning that children use in interpreting anxiety sensations could help maximize the utility of anxiety sensitivity assessments and help to modify or actualize anxiety sensitivity theory for understanding childhood anxiety disorders. In addition, assessing the reasons children provide to interpret anxiety sensations could help determine if children use the reasoning posited by anxiety sensitivity theory. Specifically, the validity of assessing anxiety sensitivity in young children could be clarified by more directly assessing if children are capable of the logic in reasoning that is integral to anxiety sensitivity theory. That is, do the reasons children give as to why they might get scared if they felt shaky or why they would be bothered if they had feelings of nervousness - follow logically? And does the logical sophistication of these reasons increase at older ages? Do youth give similar reasons to those assumed to underlie anxiety sensitivity theory? (e.g., “because I might be sick” or “because people might laugh at me”). Some data suggest that the cognitive capacities for anxiety sensitivity beliefs begin after around 5 years of age (Muris, Vermeer, & Horselenberg, 2008). This is because while the full range of meta-cognitive abilities are still developing (Flavell, Green, & Flavell, 1993; Flavell, Green, Flavell, Harris, & Astington, 1995) children as young as 4 or 5 years of age beginning to understand beliefs and the role that these can have in directing action and emotion (Harris, 2006; Harris, Johnson, Hutton, Andrews, & Cooke, 1989; Muris et al., 2008).

While no studies have examined age differences in the logical sophistication or the types of reasons children spontaneously provide for the potential negative consequences of anxiety sensations, Mattis and Ollendick (1997) have investigated children’s cognitive responses to the physical symptoms of panic in a sample of non-referred third, sixth, and ninth graders. Specifically, children listened to a tape describing a panic attack and were told to imagine that they were experiencing the condition described on the tape. Children in third grade reported less sophisticated responses to questions regarding panic attacks than sixth and ninth graders. Mattis and Ollendick (1997) also found that children of all ages were more likely to make internal attributions relative to external attributions and speculated that “…it is possible that high levels of anxiety sensitivity and elevated internal attributions in response to negative outcomes set the stage for the development of panic attacks and subsequent panic disorder” (p. 55). The findings of Mattis and Ollendick (1997) suggest that even if anxiety sensitivity was developmentally dependent on the capacity for internal attributions, children in third grade [the mean age of third graders in Mattis and Ollendick (1997) was 8.59 years] would have the cognitive capacity articulate logical reasons about the consequences of anxiety sensations.

Summary and Hypotheses

In summary, the purpose of this paper is to explore the reasoning children and adolescents give for negative interpretations of anxiety sensations using a mixed methods approach. In this study, we test a priori hypotheses quantitatively and explore the content of the reasons qualitatively. Assessing the logical sophistication or logical validity (i.e., do the reasons make sense?), the affective valence (i.e., how catastrophic are the reasons?), and qualitative reasons that youth give to evaluate anxiety sensations would fill a gap in the emotional understanding literature by turning to understanding the development of the reasoning around the consequences of anxiety. Quantitatively, we hypothesized that the logical validity/cognitive sophistication (termed “logical validity” for short) and affective intensity/how catastrophic (termed “affective valence” for short) of the reasons youth give could be reliably coded (via inter-rater reliability). Drawing from the previous work on emotion understanding, we also predicted developmental differences across age, cognitive, and verbal development with the logical validity of the reasoning used increasing across age. We took a continuous approach to age and verbal ability and tested whether these effects were nonlinear in nature.

Drawing from the research on cognitive development and the concept of horizontal decalage, more complex or more logical cognitive reasoning may occur with regards to anxiety sensations in children who have more experience with anxiety sensitivity specifically or internalizing symptoms more generally. That is, children who experience anxiety sensitivity or experience various internalizing symptoms inherent in anxiety disorders and depression are more likely to have the opportunity to develop complex beliefs about the consequences of anxiety sensations through experience. On the other hand, a lack of emotion knowledge is related to symptoms of emotional disorders with evidence of deficits in emotion understanding evident among children with or at-risk for psychological disorder (Southam-Gerow & Kendall, 2000). Indeed, errors in the cognitive interpretation of events suggest that internalizing problems such as anxiety disorders and depression emerge from faulty thinking (Beck, 1976, 1985; Ellis, 1962). Thus, we also predicted that while older children will give more logical and sophisticated responses than younger children (i.e., age and verbal ability will be positively associated), that higher levels of anxiety sensitivity or internalizing symptoms would be negatively associated with the logical validity of the responses. We tested alternative hypotheses on the potential moderating effects of anxiety sensitivity and internalizing symptoms as well. Specifically, that child anxiety sensitivity or internalizing symptom levels would moderate the main effect of age. From a decalage perspective, for example, one might expect that although younger children, in general, will provide less logical and sophisticated responses, but that younger children with high levels of anxiety sensitivity (or internalizing symptoms of anxiety and depression more generally) will give more valid and congruent responses than young children with low anxiety sensitivity. From the cognitive model of psychopathology perspective, one might expect that anxiety would be associated with less sophistication in the logical validity of responses.

Finally, we also qualitatively explored if children would use physiological, psychological/mental incapacitation, emotional, and social reasons in describing the potential negative impact of experiencing or showing anxiety sensations. While there is a lack of complete agreement in the literature with regard to the number of factors, three or four has been most consistent (Francis, Noël, & Ryan, 2019). The factors typically center on physical concerns, social concerns, emotional concerns and mental incapacitation concerns (Silverman et al., 1999; see Francis et al., 2019). Descriptive statistics were used to examine the types and the frequency of different types of reasons and text the interrater reliability of coding the reasons and their fit into broader categories. In addition, we explored beyond the themes expected based on the extant anxiety sensitivity literature to determine if other broad themes emerged using qualitative analysis software.

Method

Participants

Data were collected from a socioeconomic and ethnically diverse sample of 227 youth aged 6–17 years (mean age = 11.23 years, SD = 3.4 years) 48% female. Participants were Euro-American (47%), African American (38%), Hispanic (7%), Asian (2%), and 6% were of other ethnic backgrounds, with a median family income of between $31,000 and $40,999 a year. Forty-eight percent of the sample was female. Power analyses (Faul, Erdfelder, Buchner, & Lang, 2009) using a medium- to small-sized effects indicated that the sample size for main analyses were powered above .85 for small and above .95 for medium-sized effects.

Measures

The Beliefs about the Impact of Anxiety (BAIA) was created for this study and is a structured interview designed for this study to assess children’s reasoning about anxiety sensations. The interview items employ anxiety sensations described in the CASI as the subject of reasoning about the consequences of experiencing anxiety sensations. Each question is a brief story of a child feeling an anxiety sensation and the child’s negative response to the sensation. Participants are asked why they think the child in the story would respond the way they did. For example, one of the questions on the interview states: “Mary was with a group of friends and suddenly she felt very scared and afraid; however she didn’t tell any of her friends that she was frightened.” And then asks: “Why do you think Mary would not want other people to know that she felt afraid?” The interview is read verbatim to the children and their verbatim responses to the questions about their reasons are recorded. Responses are then independently coded (more below). The interview has 15 vignettes and the last three questions asks the youth to reflect about their own reasons. A summary of the focus of the vignettes and the three self-reflective questions is in Table 1.

Table 1.

Beliefs about the Impact of Anxiety Interview Vignette Topic & Questions and Reason Broad Classification Percentages

Vignette/Question	Physical %	Social %	Psych %	Emotional %	Other/DN %	Kappa
1. Suddenly felt very scared and afraid; however she didn’t tell any of her friends that she was frightened.	1	86	4	6	3	.57
2. Worried and nervous because could not keep mind on schoolwork.	1	69	16	2	11	.20
3. Scared because began to feel shaky.	28	12	8	28	24	.63
4. Felt like going to faint and this made him very scared.	50	15	7	9	19	.55
5. Worried because she could not control how was feeling.	0	44	14	5	39	.57
6. Became very scared because his heart was beating so fast.	62	3	4	17	14	.55
7. Butterflies in her stomach and then it growled loudly and she became embarrassed.	12	74	4	5	5	.69
8. Became very scared because felt like going to throw up.	43	28	3	5	21	.60
9. Became nervous and scared because he could not catch his breath.	85	1	2	6	6	.52
10. Worried when her stomach started to hurt.	78	4	6	8	4	.58
11. Feel shaky and noticed that the other children were watching.	3	79	1	7	11	.46
12. Luke became very scared by unusual feelings in his body.	58	4	5	16	17	.49
13. Becomes afraid worries that might be crazy.	9	35	19	10	27	.65
14. When has nervous feelings he becomes very scared.	12	24	11	4	48	.53
15. Doesn’t like to let feelings show.	1	82	0	9	17	.60
16. Why might you be scared if your heart was beating fast or if you had nervous feelings?	43	6	10	11	30	.61
17. Why would you worry if you could not keep your mind on your schoolwork?	1	60	8	7	24	.62
18. Why might you not want to let you feelings or nervousness show?	1	82	2	2	14	.48

The coding allowed for qualitative evaluation of the types of reasons as well as quantification for statistical analysis (see below). Responses were quantified in terms of logical validity (e.g., does the reason make sense given the context of the question and how sophisticated?) and affective valence (e.g., how much does the response involve an extreme catastrophic reason?). For logical validity, responses were coded on a 1 completely illogical to 7 very logical and sophisticated scale. Two coders read and rated each of 18 responses with the same coders used to code for type, logical validity, and affective valence. Coders were trained to rate responses in terms of if the response made good logical sense and the level of sophistication. For example Item 7: Laura had butterflies in her stomach and then it growled loudly and she became embarrassed. Why might Laura be embarrassed when her stomach growls (or makes noise)? A prototypical high logical validity score answer: “Because she felt someone heard her and might laugh at her”. However, if the reason seems illogical, for example Item 7: Laura had butterflies in her stomach and then it growled loudly and she became embarrassed. Why might Laura be embarrassed when her stomach growls (or makes noise)? Answer: She might hit her head. Rank it a 1 or 2. For Affective Valence responses were again rated on a 1 “completely benign reason” to 7 “catastrophic/misinterpretation” scale. Again, two coders read and rated each of 18 responses. Coders were instructed: If the reason involves a seemingly catastrophic misinterpretation for example: They think they might die or have a heart attack and die, code 6 or 7 if totally benign a 1 or 2 if in between somewhere in between. Reliability of the coding for logical validity and affective valence is presented in the results section. Qualitative coding the types of responses is presented below.

The Childhood Anxiety Sensitivity Index (CASI; Silverman et al., 1991) was used to assess anxiety sensitivity. The CASI is an 18-item self-report measure designed to assess children’s fear of different symptoms of anxiety. Children rate each question by selecting one of three choices, None, Some, or A lot. Each item is scored with a one, two, or three. Example questions are “It scares me when I feel shaky” and “It scares me when I feel nervous”. The CASI has strong convergent validity estimates with other measures of anxiety sensitivity (e.g., r = .73; Weems, Berman, Silverman, & Rodriguez, 2002). Internal consistency estimates, for the CASI have been consistently reported as being above .80 (e.g., Silverman et al., 1991). For example, Weems et al., 1998 calculated alpha coefficients on the CASI for younger (6–11), older (12–17), and total sample of youth (N = 280) with anxiety disorders. The coefficient alphas found were .86, .89, and .87, respectively. Silverman et al. (1991) showed that CASI scores were relatively stable over a 2-week interval with a test-retest correlation of .76 in a clinic-referred sample (n = 33, aged 8 to 15 years; mean age = 10.6 years) and .79 in a non-clinic referred sample (n = 72; aged 11 to 16 years; mean age = 13.3 years). For the current study, the CASI total score demonstrated good internal consistency (α = .87).

The Revised Child Anxiety and Depression Scale (RCADS; Chorpita, Yim, Moffitt, Umemoto, & Francis, 2000; Spence, 1997) is a 47-item adaptation of the Spence Children’s Anxiety Scale and was used to assess symptoms of anxiety disorders and depression. Chorpita et al. (2000) modified the Spence scales for DSM-IV and evaluated the RCADS by examining the measure’s factorial validity in a school sample of 1,641 children and adolescents aged 6–18 and its reliability in a sample of 246 children and adolescents. The results suggest an item set and factor definitions that were consistent with DSM-IV anxiety disorders and depression, as well as good reliability and validity estimates. A total score was computed from all the items. For the current study, the RCADS total score demonstrated excellent internal consistency (α = .93).

Conservation of Liquid Task.

Drawing from Muris, Merckelbach, and Luijten (2002), the ‘conservation of liquid volume’ task was conducted on a subsample of (n = 177). The tester began by having the youth set the amounts of liquids (water) in the two glasses were the same (e.g., when the child did not think that this was the case, the amounts were adjusted until the child agreed that the two glasses contained exactly the same amounts). Then the contents of one glass was poured into a third glass that was taller and thinner and the question “are the amounts the same?” was asked. Conservation task performance was either scored as 0 = failed (e.g., children who said that the tall, thin glass contained more) or 1 = passed (i.e., children who said that the amounts remained the same). Power analyses on this sub sample (Faul, Erdfelder, Buchner, & Lang, 2009) using a medium- to small-sized effects indicated that the sample size for main analyses were powered above .80 for small and above .90 for medium-sized effects.

Peabody Picture Vocabulary Test (PPVT-III; Dunn & Dunn, 1997) was administered to sub-sample (n = 105) of the participants to examine the association between verbal ability and BAIS reasoning. A sub-sample was used to reduce time burden of the larger sample. The PPVT-III is a norm-referenced test of verbal intellectual ability designed for use in a wide age range of participants. The test was standardized on a nationally representative stratified sample and has good psychometric properties. Raw scores were used in addition to the standardized scores because we were interested in associations with age/cognitive development and standardized scores are age normed and thus would negate developmental effects. Power analyses on this subsample (Faul, Erdfelder, Buchner, & Lang, 2009) using medium effects indicated that the sample size for these analyses were still powered above .80 for the effects expected.

Procedures

Families were recruited through area schools and community outreach. The University Committee for the Protection of Human Subjects in Research IRB#: “03Mar11”, reviewed and approved the study. Children were excluded if parents indicated that the child had a history of one or more of the following diagnoses--all pervasive developmental disorders, mental retardation, selective mutism, organic mental disorders, schizophrenia, and other psychotic disorders, or were at risk for harm to self or others (only one child was excluded whose parent indicated the child had a diagnosis of pervasive developmental disorder NOS). Interested families were informed that we were conducting a study of youth behaviors and emotions.

Informed consent was obtained from the caregiver and informed assent was obtained from the child. Completion of the assessment took place in a quiet room and the child completed the assessment in a separate room from the parent. Both the youth and parent were greeted and given a general overview of the assessment procedures. Standardized specific instructions were then given to the parent and child separately. Youth completed the measures in random order and were assisted as necessary by trained research assistants [e.g., young participants were read the assessment battery by research assistants who closely monitored the child’s comprehension of the questions and fatigue as done in previous research (see Vernberg, La Greca, Silverman, & Prinstein, 1996)]. As noted above the PPVT and the conservation task, which were specifically administered to subsamples. The conservation task procedures were developed for time and consistency by piloting on initial participants so initial participant data was not used. The PPVT is a more elaborate test and was only given to a subsample to reduce participant time burden. Total assessment time was approximately one hour and was one hour and 30 mins for those completing the PPVT and Conservation task. At the conclusion of the study, participants were debriefed and given monetary compensation for their time.

Qualitative Analysis

To examine the types of reasons qualitatively, participants verbatim responses to each of the BAIA vignettes were entered into a spreadsheet and several initial passes were made through the responses and a coding system involving 64 different specific codes were made that fell into one of six different broad categories. In addition to the broad Physical, Social, Psychological responses expected, responses also clustered into Emotional response, a set of responses that were “other reasons” and a sixth code if youth said “they did not know”. These were used to compute interrater reliability. For example, specific code 1 was assigned to a response involving fear of a heart attack (i.e., specific code 1 was an asthma attack) then each code 1 would be a “physical concern”. Similarly, specific code 12 was assigned if the reason given involved being laughed at, code 15 if being teased/made fun of were the reasons, and broadly these were considered social concerns; codes 13 - “going crazy” and 14 - “losing control,” were assigned and broadly these fit into the psychological category. Additional responses emerged that did not fit clearly into one of the three broad categories hypothesized. Thus, reasoning that involved emotions such as “shyness” “being embarrassed” or “nervous feelings” were clustered into a fourth broad “emotional” reasons. A group of responses that did not fit into the other four broad categories were coded “other” reasons. Examples include, “Feel something bad may happen,” “Witness something scary/traumatic.” The final broad code was “unknown” and was used for the response “I don’t know.” The qualitative reasons were independently coded by two coders. Cohen’s kappa were computed for each of the 18 items on the broad codes and these are presented in Table 1. All but one BAIA vignette (vignette 2), had acceptable to good interrater reliability (i.e., in the .4 to .6 range, See Table 1).

Given the greater diversity than expected in the types of reasons, and the various “other” responses, we completed a second qualitative analyses of the responses using the qualitative software program MAXQDA (VERBI Software, 2017). Three coders, including the co-authors (RDC and ELN), initially familiarized themselves with the responses by taking several particular cases across several of the BAIA items, and then a thematic analysis was conducted with an initial coding scheme developed (i.e., transcripts were coded by the team and categories developed, refined and validated in collaboration).

Results

To examine interrater reliability for the logical validity and affective valence coding, a two-way mixed interclass correlation with absolute agreement was conducted to examine both consistency and level of agreement (both the logical validity ratings and affective valence ratings were averaged across the two raters for analysis). Results indicated that the intra-class correlation (ICC) was .84 (95% confidence interval = .79 to .88) for logical validity ratings and .85 (95% confidence interval = .67 to .91) for the affective valence ratings, consistent with the acceptable inter-rater reliability hypothesis. Differences between boys and girls and differences across ethnicity were examined and no significant differences were detected¹.

Examination of the data indicated that all missing data was missing at random (with the exception of PPVT and the conservation task, which were specifically administered to subsamples, as indicated above). All missing data was handled analysis-by-analysis such that all available cases were analyzed. Examination of the distributions and scatter plots of the study variables indicated acceptable distributions for all variables. Most were fairly normally distributed with one outlier on the RCADS (more than 4 standard deviations from the mean and almost two from the next closest score), negative skew on the logical validity ratings (more piled up on the upper tail), and positive skew for the CASI and RCADS (more piled up toward the lower tail). Because of this, parametric analyses were supplemented with non-parametric Spearman correlations (and curve analysis; see below) and the effect of the outlier was tested to determine if differential results were evident without the outlier.

Means, standard deviations, and correlations among the quantitative measures are presented in Table 2. Overall, the pattern of correlations were similar across Pearson and Spearman correlations². Consistent with expectations, age, PPVT scores, and being a conserver were positively associated with logical validity ratings. RCADS scores negatively correlated with logical validity ratings. Affective valence was positively correlated with logical validity ratings but not correlated with other variables.

Table 2.

Means, Standard Deviations, and Correlations among Quantitative Variables

	Logical Validity	Affective Valence	CASI	RCADS	Conserver	Raw PPVT	Standard PPVT	Age
Logical Validity	-	.30 **	−.08	−.17*	.25**	.40**	.24*	.37**
Affective Valence	.44**	-	.10	.06	.07	.07	−.01	.08
CASI	−.12	.06	-	.65**	−.22**	−.38**	−.34**	−.27**
RCADS	−.23**	.05	.61**	-	−.20**	−.31**	−.22*	−.26**
Conserver	.26**	.08	−.23**	−.24**	-	.50**	.35**	.38**
Raw PPVT	.46**	.15	−.39**	−.34**	.51**	-	.66**	.77**
Standard PPVT	.25*	.02	−.30**	−.24*	.35**	.67**	-	.10
Age	.42**	.09	−.29**	−.26**	.38**	.77**	.09	-
Mean (SD) or % n	5.06(.41) 223	3.41(.37) 223	28.96(6.82) 227	78.90(19.35) 220	48% 179	130.08(34.28) 105	100.90(16.75) 105	11.23(3.42) 226

Notes: Pearson below the diagonal, Spearman Rank order above. Conserver of fluid = 1 non-conserver = 0;

^*.Correlation is significant at the 0.05 level (2-tailed).

^**.Correlation is significant at the 0.01 level (2-tailed).

The relation between age and logical validity ratings are depicted in Figure 1. Curve estimation regression analyses tested if linear, quadratic (curve), or cubic (two curves) models best fit the relation. Results indicated a significant linear model (i.e., R² = .17, p < .001). However, a quadratic model (i.e., R² = .23, p < .001) was a significant improvement over the linear model (i.e., age squared coefficient was a significant addition to the model with t(216)=−4.28, p < .001). The cubic effect was not a significant improvement over the quadratic effect (i.e., age to the 3^rd power was not significant with t (216)= 1.51, p = .131). The relationship between age and affective valence ratings are also depicted in Figure 1. Curve estimation regression analyses tested if linear, quadratic (curve), or cubic (two curves) models best fit the relationship. Results indicated a non-significant linear model (i.e., R² = .01), but that a quadratic model (i.e., R² = .06, p < .001) was a significant predictor accounting for a relatively small amount of variance (i.e., age squared coefficient was a significant addition to the model with t(216)=−3.30, p < .001). The cubic effect was not a significant improvement over the quadratic effect (age to the 3^rd power was not significant with t (216) = 1.09, p = .278).

Figure 1.

The relation between logical validity (top) and Affective Valence (Bottom) ratings and age with linear, quadratic, and cubic fit lines.

To examine if levels of anxiety sensitivity moderated the association between age and the logical validity of the reasons we tested CASI total scores, age, and an interaction term (e.g., CASI Total by Age centered) to predict logical validity ratings. Predictors were centered via standardization and were entered simultaneously into the regression. In this analysis, the interaction term was also significant (model R² = .19, p < .001, interaction term t(214)= 2.18, β = .14, p = .031) and the nature of the interaction is again depicted in Figure 2. Post-hoc testing (simple slopes analysis at plus and minus one standard deviation) indicated the association between age and the logical validity of the ratings is positive and steeper at high anxiety sensitivity (+1 SD; β = .58, p < .001), and less steep but still significant at low levels of anxiety sensitivity (− 1 SD; β = .29, p = .001). We also tested anxiety sensitivity as a moderator of the curvilinear association with age and the age-squared by anxiety interaction term, but this relationship was not significant. Similar analyses predicting affective valence were not significant (age did not interact with either the CASI or RCADS in the prediction of affective valence of the codes).

Figure 2.

Levels of RCADS (Top Panel +1 SD; β = .54, p < .001; − 1 SD; β = .24, p = .008) and Anxiety Sensitivity (Bottom Panel +1 SD; β = .58, p < .001; − 1 SD; β = .29, p = .001) Moderate the Association between Age and the Logical Validity of the Reasons.

To examine if RCADS scores moderated the association between age and the logical validity ratings, we tested RCADS total scores, age, and an interaction term (e.g., RCADS Total by age centered) to predict logical validity ratings. Predictors were centered via standardization and were entered simultaneously into the regression. In this analysis, the interaction term was significant (model R² = .22, p < .001, interaction term t(206)= 2.33, β = .15, p = .021)³ and the nature of the interaction is depicted in Figure 2. Post-hoc testing (simple slopes analysis at plus and minus one standard deviation) indicated the association between age and the logical validity of the ratings is positive and steeper at high RCADS (+1 SD; β = .54, p < .001), less steep but still significant and low levels (− 1 SD; β = .24, p = .008). We tested RCADS as a moderator of the curvilinear association with age and the age-squared by anxiety interaction terms, but this relationship was not significant.

Subsample Analyses with PPVT and Conserver Status

Similar to the analyses with age, the relationship between PPVT standard scores and logical validity ratings tested if linear, quadratic (curve), or cubic (two curves) models best fit the relationship. Results indicated a significant linear model (i.e., R² = .06, p = .012), however quadratic and cubic effect models were not a significant improvements over the linear effect. The relation between PPVT standard scores and affective valence ratings with curve estimation regression analyses found no significant linear, quadratic or cubic effects. To examine if levels of anxiety moderated the association between PPVT standard scores and the logical validity of the reasons we tested RCADS total scores, PPVT standard scores and an interaction term (e.g., RCADS Total by PPVT centered) to predict logical validity ratings. Predictors were centered via standardization and were entered simultaneously into the regression. In this analysis, the interaction term did not reach two tailed .05 significance (model R² = .16, p < .001, interaction term t(99)= 1.80, β = .17, p = .076).⁴ The nature of the interaction was examined given that it was close to significant and the pattern was similar to those found with age. CASI scores did not significantly interact with PPVT scores. To examine if levels of anxiety moderated the association between being a conserver and the logical validity of the reasons, we tested RCADS total scores, conserver status, and an interaction term (e.g., RCADS Total by Conserver yes or no) to predict logical validity ratings. Predictors were centered via standardization and were entered simultaneously into the regression. In this analysis, the interaction term was significant (model R² = .13, p < .001, interaction term t(166)= 2.57, β = .24, p = .011). Post-hoc testing (simple slopes analysis) indicated the association between being a conserver and the logical validity of the ratings is significantly positive for those with high RCADS (+1 SD; β = .46, p < .001), but not significant at low RCADS total scores (− 1 SD; β = .02, p = .853). CASI scores did not significantly interact with being a conserver to predict logical validity. Similar analyses predicting affective valence were not significant (neither PPVT nor conserver status interacted with either the CASI or RCADS in the prediction of affective valence of the codes).

Qualitative Analysis

Frequencies of each of the responses in terms of the broad categories across all responses (N of responses = 4,122; i.e., summarizing responses across each vignette) were 39% Social, 27% Physical, 13% “other,” 8% Emotional, 7% Psychological, and 6% “I don’t know” responses. The top 6 most common specific responses across all responses (N of responses = 4,122; again, summarizing responses across each vignette) were 6.7% worried about others might think, 6.1% reasons involving failure, 5.7% being sick, 4.8% feeling embarrassed, 4.6% being teased, and 4.1% “trying to hide something” rounded out specific responses with greater than 4% frequency⁵.

Frequencies of each of the responses in terms of the broad categories for each of the 15 BAIA vignettes or three self-reflective questions is presented in Table 1. Overall, socially themed vignettes tended to have the highest frequency of social reasons and physically themes vignettes high levels of physical reasons given. There was more diversity in the psychological/cognitive themes vignettes with other, psychological, and emotional responses each fairly common reasons. A verbatim, prototypical physical reason, for example vignette 9 involving becoming nervous and scared because the protagonist could not catch his breath, is “If he had asthma, he could be having an asthma attack.” A verbatim, prototypical social concern for example vignette 1 involving suddenly feeling scared and afraid but not tell any of her friends that she was frightened is, “wants to hide her emotions, wants people to believe she is strong.” A verbatim prototypical psychological concern for example vignette 13 involving anxiety symptoms being indicative of going crazy is, “might not have control over her feelings.” This type of item shows complexity in answering with reasons involving both the psychological and social components such as this verbatim response, “because if she thought that she was crazy then no one would want to be her friend.” Emotional responses were common on vignette 3 involving being scared because the protagonist began to feel shaky. For example, this response, “the shakiness makes her feel scared.” Other responses were common on the rather meta-cognitive vignette 14 when the protagonist has nervous feelings, he becomes very scared. For example, this reason involving foreboding, “He probably thinks something bad is going to happen to him.”

Because of the complexity of responses transcending the a priori expected categories MAXQDA was used to further examine from an empirical approach. Responses were coded by the co-authors (RDC and ELN) and categories developed, refined, and validated in collaboration. Sixteen broad themes initially emerged. After review and discussion these were consolidated into 14 themes. Two of the initial themes were dropped because they comprised less than 1% of the responses or they were highly conceptually similar. That is, separate codes for “physical concerns – existential,” which included things like feeling like they could not breathe, pain in chest, etc., was combined with the more extreme “physical concerns – death,” which included responses such as death or feeling like they were going to die. In the first round of coding Saldana’s (2013) provisional coding approach was used - reviewing a small portion of the sample to generate a list of guiding codes. As the coders progressed through the data, Saldana’s (2013) axial coding approach was used to refine the list of codes and provide criteria for each code to ensure consistency. In the final round of coding, Saldana’s focused coding approach was employed to ensure the codes reflected specific words/phrases commonly used by the children to communicate the themes in a manner that directly represented their understanding of anxiety informed by our understanding of the types of anxiety. So some of the codes collapsed together, but others remained apart even if similar to provide additional detail as to how anxiety was experienced from the children’s lens. For example, the creation of the emotional category (being embarrassed, nervous feelings) might have been conceived of as falling into the existing social and physical categories but we found that there were responses that were relatively more emotional but not or less social or physical including emotion words. The 14 final themes are summarized in Table 3. In the table the overall percentage was calculated by taking the segments with that theme divided by the 4,808 total coded segments. The “Common in Items” column was determined for an item with 30 or more coded segments with that theme identified for a specific vignette answer. Three percent of data are represented with the “no response” category.

Table 3.

Broad Themes that Emerged from Qualitative Analysis with MAXQDA

Theme	%	Common in Items
1. Anticipation of negative peer social response: make fun of me, laugh at me, teasing	19%	1, 7, 8, 11, 13, 14, 15, 18
2. Feeling nervous: worried, scared, frightened, afraid	13%	1, 3, 4, 6, 11, 13, 14, 15, 16, 17, 18
3. Physical concerns involving sickness. examples: throw up, stomach-ache	10%	8, 10, 12, 16
4. Negative consequences: failing, bad grade, something bad will happen	8%	2, 3, 13, 14, 16, 17
5. Physical concerns - existential threat. examples: heart attack, can’t breathe, chest pain, suffocate, feels like die	7%	6, 11, 9, 16
6. Physical concerns- other: tired, hungry, cold, crying, puberty, running, gas	7%	4, 5, 7, 10
7. Control: controlling feelings, controlling actions, concentration, paying attention, emotional regulation	6%	2, 5, 15, 17, 18
8. Negative talk/thoughts: Generally referring to being different, weird, odd, crazy, weak, dumb, cowardly, stupid	6%	13, 14
9. Feeling embarrassed: embarrassed, feelings hurt, regret	5%	1, 11, 15
10. New/unknown situation: never happened before, don’t know what to do	5%	3, 4, 12
11. Physical concerns general. “something wrong/bad”	5%	5, 9, 12
12. Physical concerns involving passing out. examples: faint, dizzy, shaky, wobbly	3%	4, 9
13. Anticipation of negative adult social response: mom mad at me, teacher angry, getting punished	2%	2, 17
14. Monster: unknown bad guy, robber	1%
15. No Response	3%

Notes: Some initial codes dropped with less than 1%. Overall % was calculated by taking the segments with that code divided by the 4,808 total coded segment. The “Common in Items” was determined for an item with 30 or more coded segments with that theme identified.

Discussion

The results of this study add to the extant literature on the development of the reasoning children and adolescents give for negative interpretations of anxiety sensations. Consistent with our predictions, we were able to code reliably the logical validity (i.e., did the reason make sense) and affective intensity (i.e., how catastrophic) of the reasons youth provided, as demonstrated via the high inter-rater reliability among coders. Moreover, and consistent with previous work on emotion understanding in youth (Bender et al., 2015), we found developmental differences across age, cognitive, and verbal development with the logical sophistication of the reasoning used by youth increasing across age in a non-linear manner and linearly with cognitive and verbal abilities. Specifically, logical reasoning rose steadily with increased age for youth still in childhood and began to plateau among youth across early to late adolescence (see Figure 1), while those youth who demonstrated greater cognitive and verbal abilities provided more sophisticated, logical reasons for the negative interpretations of their anxiety sensations.

The findings also make an incremental contribution to our current knowledge about the development of emotion understanding in children and adolescents. This study was the first to explore the reasoning children employ to understand the consequences of anxiety sensations. Our findings showed a curvilinear pattern of the logical sophistication of reasoning for anxiety sensations as a function of age, such that logical reasoning rose steadily with increased age for youth still in childhood and began to plateau among youth across early to late adolescence (see Figure 1). Conversely, but also in line with previous work on emotion understanding in youth (Pons & Harris, 2005; Pons et al., 2003), we found that both cognitive and verbal abilities were linearly related to the logical reasoning of anxiety sensations, such that youth who demonstrated greater cognitive (conserver) and verbal abilities (PPVT scores) provided more sophisticated, logical reasons for anxiety sensations.

The present results fill a gap in the emotion understanding literature by suggesting that the reasoning youth use to understand the potential consequences of anxiety similarly develops from a simple and limited understanding first of the visible and less cognitive aspects of emotions, then to a deeper, more sophisticated understanding and comprehension of the less visible and more cognitive aspects of emotions. This point can be illustrated with the qualitative responses for vignette 12 (i.e., “Luke became very scared by unusual feelings in his body”). Younger youth tended to report developmentally common and concrete reasons such as this response by a six year old “Because if you see a monster in your closet you get scared.” Whereas as older youth were able to link feelings to more complicated and sophisticated reasons. Such as this response by a 16 year old: “He may be going through puberty and he’s not used to the strange thing he’s feeling” However, these main effects must be interpreted in the context of the interaction with anxiety that were found.

We also predicted that while older children or those with greater verbal ability will give more sophisticated responses than younger children, child anxiety levels would moderate the main effect of age. This result for age was found with both RCADS and CASI scores and was found with the RCADS for conserver status and with a similar trend on PPVT scores. These results thus add to an existing understanding of emotional development and are consistent with the idea that the process of cognitive-emotional understanding is not a simple linear one because various domains may develop differentially. However, our results suggest that the process of individual differential cognitive development termed “horizontal decalage” may apply to emotional understanding in the sense of speeding up the logical reasoning about anxiety. To draw the analogy to the basic explanation for horizontal decalage is that children who have substantial experience with anxiety sensations will begin to use more complex reasoning about anxiety - experience with anxiety trains the child’s cognitive ability in that specific domain. We found that more complex, or more “adult like,” cognitive processing may occur with regards to anxiety sensations in in younger children who report less anxiety but that there was a stronger association between age (and PPVT scores) and logical reasoning among those with higher anxiety. That is, younger youth who experience anxiety frequently were less likely to report more logical beliefs about the consequences of anxiety (see Figure 2) particularly for RCADS scores. However, there were stronger associations between age and logical reasoning for those with high anxiety.

Results also appear to support the psychopathology perspective that lack of emotion knowledge is related to symptoms of anxiety (Southam-Gerow & Kendall, 2000) and that errors in the cognitive interpretation of events foster the development of anxiety problems (Beck, 1976, 1985; Ellis, 1962). This was particularly true in that higher anxiety at younger ages was associated with lower logical validity. Given our cross-sectional design, a solid conclusion cannot be drawn but it may be that this disrupted logical reasoning early in development of reasoning skills results in later misinterpretations of anxiety sensations. That is, they are more likely to attribute their anxiety sensations to negative consequences which is strengthened through their continuing experience of anxiety, so even if their logical reasoning catches up with low anxious children the affective valence they attribute to those logical reasons is mostly negative and not always true.

The study also adds to the literature by qualitatively exploring if youth used physiological, psychological, and social reasons in describing the impact of anxiety sensations. Across vignettes, physiological and social reasons were most commonly reported (See Table 1) overall, and also tended to correspond to the content of the vignette in a conceptually consistent way (i.e., social vignettes with social reasons most common; physiological vignettes with physiological reasons most common). Emotional and psychological/cognitive reasons were generally less common. Psychological/cognitive vignettes tended to have a greater range of responses (i.e., social and physiological reasons were also commonly provided for these type of vignettes). However, psychological/cognitive reasons tended to be in the most reported reasons in those vignettes. Exceptions existed, for example, vignette #2 (in Table 1) involves being worried and nervous because the protagonist could not keep their mind on their schoolwork. In this vignette, the most common response was a social response and not a cognitive response, even though the vignette has a cognitive focus. This finding points to the potential utility of the additional themes identified by the qualitative coding with MAXQDA software.

While the themes in Table 3 also were broadly centered on physical, social, psychological, and emotional responses, several unique and common themes emerged and serve to flesh out findings, such as the one regarding vignette 2. Overall from Table 3, the anticipation of negative peer social response such as being made fun of, laughed at, or being teased was most common. Filling out the top most common reasons were those concerning the appraisal that feeling nervous in and of itself is a negative consequence, followed by physical concerns involving sickness such as not wanting to throw up or have a stomach-ache, and a category of negative personal consequences such as failing, getting a bad grade, or that something generally bad will happen. These themes help illustrate the types of social reasons, for example with regards to vignette 2, where the anticipation of negative adult social response was common such as “mom/dad will be mad at me,” “teacher would be angry,” getting punished in addition to the negative consequences of failing, or a bad grade generally, as well as controlling feelings, controlling actions, concentration, paying attention, and emotional regulation. As noted, anxiety sensitivity has been assessed through self-report measures and researchers have recently called for the expansion of items to improve the assessment of anxiety sensitivity in youth (Francis, Noël, & Ryan, 2019). The data from this study made serve to help guide the development of such additional items. Specifically, items can be developed to address all of the various consequences listed in Table 3.

This study’s contributions to the existing literature must be considered in light of its limitations. First, the cross-sectional and non-experimental design of the study precludes directional or causal interpretations of the findings. In terms of participants, the goal was to recruit a representative community sample across a wide age range. While the results of the study are informative with regard to a community sample, the findings may not generalize to other populations. One possibility for future research would be to replicate these findings with a clinic-referred sample of youth. Do youth with very high anxiety demonstrate similar findings or would the clinical nature of the sample negate the age-related findings? Similarly, this is a sample of only typically developing youth as those youth with developmental delays were excluded. Developmental or learning delays/disorders would likely influence conclusions. Learning more about emotional reasoning in such samples could be fruitful next steps. In addition, while the interrater reliability for the BAIA vignettes were acceptable, they were in the lower range. The conservation task and PPVT were only completed on a subsample so it is possible null findings with these variables are due to lower power. Finally, the cross-sectional nature of the study does not allow us to make truly developmental inferences. Thus, future research testing these models using prospective, longitudinal designs may detect effects not found in the current study.

Despite the limitations, this study added to the extant literature in a number of ways including identifying a possible nonlinear pattern of development and beginning to elucidate the emotional and cognitive interactions. The data presented here may also aid in modifying existing assessments of childhood anxiety sensitivity by helping to create more developmentally sensitive measures of anxiety sensitivity for young children or by helping to push the developmental window to earlier ages in the identification of a risk factor (i.e., anxiety sensitivity) for later anxiety problems. The information may also have future clinical utility by identifying the types of reasons children have for disliking feelings of anxiety and thus may suggest avenues for correcting unreasonable expectations about anxiety sensations.