When talking goes awry: association between co-rumination and trait anxiety, test anxiety, and anxiety sensitivity in early and late adolescents

ABSTRACT

Background and Objectives

Friends are major sources of social support for adolescents. This support may sometimes lead to co-rumination when the problem is discussed exhaustively with a focus on negative feelings. Co-rumination has been associated with some forms of anxiety, including clinical symptoms. Further studies are needed to investigate whether this association extends to additional and non-clinical forms of anxiety in youth. This study aimed to explore the relationship between co-rumination and trait anxiety, test anxiety, and anxiety sensitivity using secondary data.

Design and Methods

In this 2019 cross-sectional study, 1204 (59% girls) Canadian 6th-grade early adolescents (ages 11-12) and 11th-grade late adolescents (ages 16-17) completed self-report questionnaires measuring co-rumination, trait anxiety, test anxiety, and anxiety sensitivity.

Results

Co-rumination was associated with anxiety sensitivity in early adolescents and with trait anxiety, test anxiety, and anxiety sensitivity in late adolescents.

Conclusions

Developmental factors may play a role in the association between co-rumination and different forms of anxiety. Anxiety sensitivity may appear alongside co-rumination in early adolescence and may broaden to trait and test anxiety in late adolescence. These results extend our understanding of the relationship between co-rumination and anxiety, as well as generate hypotheses for future longitudinal studies.

KEYWORDS:

• Co-rumination
• trait anxiety
• test anxiety
• anxiety sensitivity
• early adolescence
• late adolescence

Introduction

As adolescents rely on their friends as major sources of social support, co-rumination is often studied in this age group (Rose, 2021). When co-ruminating, a dyad discusses a problem repeatedly and exhaustively while focusing on the causes of the problem, its consequences, and the negative feelings that it generates (Rose, 2002). Although it has been described as its own unique construct, co-rumination lies at the crossroads of self-disclosure and rumination (Rose et al., 2007). Elements of the latter have been proposed to explain the link between co-rumination and internalizing symptoms (Rose, 2002). Indeed, rumination is a well-documented factor for the maintenance of anxiety, as it involves a perseverative and intensive focus on solicited negative feelings (McLaughlin & Nolen-Hoeksema, 2011). Thus, similar to aspects of rumination that can lead to internalizing symptoms (e.g., not engaging in other distracting activities; McLaughlin & Nolen-Hoeksema, 2011), co-rumination behaviors (e.g., revisiting the problem) may exacerbate negative affect generated by a problem (Rose et al., 2007). Most research has focused on the association between co-rumination and depression and less on that between co-rumination and anxiety (Spendelow et al., 2017). Nevertheless, cross-sectional and prospective studies found a positive association between co-rumination and anxiety in adolescents (Ohannessian et al., 2021; Rose, 2002; Rose et al., 2007; Tompkins et al., 2011; Waller & Rose, 2010) using measures of anxiety symptoms (e.g., Revised Children’s Manifest Anxiety Scale [Reynolds & Richmond, 1978], Youth Self-Report, anxious/depressed subscale [Achenbach & Rescorla, 2001]), and measures based on clinical forms of anxiety (e.g., Screen for Child Anxiety Related Disorders [Birmaher et al., 1997]). Given that different forms of anxiety may have varying and distinct associations with co-rumination (Starr & Davila, 2009), it is still unclear whether co-rumination in adolescence extends to many forms of anxiety. Also, given that co-rumination and anxiety increase across adolescence, especially among teenage girls (Beesdo et al., 2009; Hankin et al., 2010; Rose, 2002; Schwartz-Mette & Rose, 2012; Tompkins et al., 2011), sex and age should be considered when studying the relationship between co-rumination and anxiety.

In recent years, our research group developed a project to study various forms of non-clinical anxiety in youth. At two time points (May/June and October/November 2019), we collected anxiety data (e.g., trait anxiety, test anxiety, and anxiety sensitivity) from early and late adolescents in Quebec, Canada. Trait anxiety is a stable tendency of individuals predisposing them to experience anxiety when faced with a variety of threatening situations (Spielberger, 1972). Test anxiety is specific to evaluation situations and manifests through thoughts (worry during the evaluation), autonomic reactions (bodily responses such as sweating), and off-task behaviors (unrelated to the evaluation itself; Wren & Benson, 2004). Finally, anxiety sensitivity is a fear of one’s physiological sensations of anxiety (sweating, racing heart, etc.) and is characterized by the individual's belief that these sensations will have negative physical, psychological, and/or social consequences (McNally, 1990; Reiss et al., 1986).

We took advantage of this ongoing research project to explore the relationship between co-rumination and these different forms of anxiety (trait anxiety, test anxiety, and anxiety sensitivity) in early and late adolescent normative samples. Examining the interplay between co-rumination and various forms of anxiety may provide preliminary insight into which aspects of anxiety could be targeted by interventions, potentially mitigating the relationship between these factors during the transition from early to late adolescence.

Although some studies have been conducted on co-rumination and trait anxiety (Carlucci et al., 2018; Doyle, 2013), no research to our knowledge has investigated test anxiety and anxiety sensitivity. Despite the limited literature on the topic, we expected a positive relationship between co-rumination and trait anxiety, test anxiety, and anxiety sensitivity in both age groups. Given that co-rumination and anxiety tend to increase across adolescence, we also expected to see vaster (i.e., number of observed associations) and/or more pronounced (i.e., strength of the associations) differences in the associations in late versus early adolescents.

Methods

Methodological context

All data was obtained from Journault et al. (2022), who aimed to investigate the different individual susceptibilities that may trigger non-clinical anxiety in students. The addendum to integrate a co-rumination questionnaire into the larger study received ethics approval from the Research Ethics Board of the Centre intégré universitaire de santé et de services sociaux de l’Est-de-l’Île de Montréal on September 25^th, 2019. As the questionnaire was integrated into the larger study after the first data collection, co-rumination was only measured during the second time point.

Disclosure

This project was preregistered on the Open Science Framework on 16 January 2020. The preregistration and database associated with the results presented in this paper can be accessed using the following DOIs: 10.17605/OSF.IO/B2UJ7 and 10.17605/OSF.IO/CR8XT, respectively. For more information on the larger study, its preregistrations can be found at DOIs: 10.17605/OSF.IO/35UZ9 and 10.17605/OSF.IO/U3K7C.

Participants

The sample consisted of 1204 students from Grades 6 (ages 11–12; early adolescents) and 11 (ages 16–17; late adolescents; National Academies of Sciences et al., 2019) clustered in 58 classrooms from seven elementary and six high schools on the north shore of the island of Montreal in Quebec, Canada. All participants were fluent in French. Sample characteristics are displayed in Table 1. All students from each school were invited to participate in the study. Parental consent was obtained for early adolescents before the data collection itself and consent for late adolescents was obtained in person with the research team before data collection.

Table 1. Sociodemographic characteristics of the sample.

	Early adolescents	Late adolescents	Full sample
Participants	287	917	1204
Sex	278	914	1192
Girl	159 (55.4%)	550 (60.0%)	709 (58.9%)
Boy	119 (41.5%)	364 (39.7%)	483 (40.1%)
Missing	9 (3.1%)	3 (0.3%)	12 (1.0%)
Origins	204	205	409
White	148	172	320
Indigenous nations	3	2	5
Middle easterner	1	1	2
Asian	1	2	3
Black	2	0	2
Central and South America	4	6	10
Other	45	22	67
Missing	83	712	795
Socioeconomic status^a	189	199	388
0–25,000$	2	3	5
25,000–50,000$	15	14	29
50,000–75,000$	18	20	38
75,000–100,000$	31	35	66
100,000–125,000$	28	53	81
125,000–150,000$	33	7	40
150,000$ and over	62	67	129
Missing	98	718	816

^a Household family income in Canadian dollars (as reported by parents) was used as an indicator of participant socioeconomic status.

Measures

Co-rumination

To measure co-rumination with friends, we used the adapted 9-item version (Arroyo et al., 2017) of the original 27-item questionnaire (Rose, 2002). Each item (e.g., “We’ll talk about every part of the problem over and over”) was rated on a scale ranging from 1 (“Not at all true”) to 5 (“Really true”), where total scores varied from 9 to 45. Overall co-rumination scores were calculated via the mean ratings of the items in the questionnaire, where a higher score indicated greater co-rumination. Compared to the reliability of the 27-item version (α=.96; Rose, 2002), the short 9-item version had excellent reliability (α=.91–.94; Arroyo et al., 2017). Our research team used a double-blind translation technique to translate the questionnaire from English to French (Kristjansson et al., 2003). Within our sample, we obtained a Cronbach’s α of .87.

Trait anxiety

The French version of the State-Trait Anxiety Inventory for Children (STAI-C; Spielberger et al., 1983; translated and validated by Turgeon & Chartrand, 2003) was used to measure trait anxiety. For this study, only the trait subscale was used (20 items) to measure anxiety as an individual tendency and asked the participant how they feel in general. Items were rated on a scale ranging from 1 (“Hardly ever”), 2 (“Sometimes”), to 3 (“Often”). Participant scores varied from 20 to 60, where a higher score indicated higher trait anxiety. An example item is “Usually … I [hardly ever/sometimes/often] feel like crying.” The reported reliability for trait anxiety is .89 (Turgeon & Chartrand, 2003). Within our sample, we obtained a Cronbach’s α of .89 for trait anxiety.

Test anxiety

The 25-item Children’s Test Anxiety Scale (CTAS; Wren & Benson, 2004) was used to measure test anxiety. Participants were instructed to answer the questions while reflecting on how they behaved during an evaluation and the emotions they felt. Items (e.g., “When I do an exam … my heart beats fast”) were rated on a scale of 1 (“Almost never”) to 4 (“Almost always”). Total scores varied from 25 to 100, where a higher score indicated higher test anxiety. The reliability of the CTAS is .89 (Wren & Benson, 2004). Our research team used a double-blind technique to translate the original scale from English to French. Within our sample, we obtained a Cronbach’s α of .93.

Anxiety sensitivity

The French version of the Childhood Anxiety Sensitivity Index (CASI; Stassart & Etienne, 2014) was used to measure anxiety sensitivity. This index contained 18 items. Participants responded in one of three ways (“not at all”, “a little”, or “a lot”) to items such as “It scares me when I feel ‘shaky’”. Total scores varied from 18 to 54, with a higher overall score indicating greater anxiety sensitivity. The reliability of the French version of the CASI is .82 (Stassart & Etienne, 2014). Within our sample, we obtained a Cronbach’s α of .89.

Procedure

Data for this paper was collected in October and November 2019 (the second time point of the larger study). All testing occurred during class time with the aid of research assistants and/or the second author of this paper. In their classrooms, participants of all schools completed all self-report measures online via the Studies Web Automation Tool (SWAT), except for one school that used pencil and paper questionnaires due to technical issues.

Statistical analyses

Preliminary analyses

We examined the association between co-rumination and three different forms of anxiety in early and late adolescents, where the ranges of the anxiety scales vary greatly. Instead of standardizing using z-scores, scores for each anxiety measure were transformed to a common scale of 0–1 to more readily compare the dynamic changes between co-rumination and each unique anxiety measure, as in previous research comparing several different types of anxiety (Carey et al., 2017).

This was done by subtracting the minimal theoretical value (min) and multiplying by ${\displaystyle \frac{1}{(\max -\min )}}$.

Before any analyses were conducted, we ensured that the assumptions for linear mixed-effect models were not violated. To confirm the sex and age differences found in the co-rumination literature, preliminary analyses with mean comparisons using a linear mixed-effect model were performed to investigate whether co-rumination varied across sex (girls and boys) and age (early and late adolescents). A random intercept on classroom was added to account for the effect that anxiety inside the classroom may have on each individual’s anxiety (Charbonneau et al., 2022). If an effect of sex was found for co-rumination, this variable was adjusted for in our main analyses. Given that the number of late adolescents in our sample (n = 917) far outweighed the number of early adolescents (n = 287), the age groups were analyzed separately in the main analyses.

Main analyses

Two-level multilevel analyses allowed us to control for dependency between scores (e.g., students in the same class may respond similarly, a student may have similar answers across the four anxiety measures). These analyses separated the variance (of the outcome variables) into the variance within the classrooms (level 2; between-subjects variance), within the respondent (level 1; within-subjects variance), and the residual variance (baseline individual variance). The association between co-­rumination and anxiety was treated as a fixed effect. First, to test if co-rumination was associated with trait anxiety, test anxiety, and anxiety sensitivity, we used linear mixed-effect models, allowing us to add a random intercept by classroom. We examined the intraclass correlations (ICC) of the null base models (no predictors). The ICC values for trait anxiety, test anxiety, and anxiety sensitivity were as follows: early (0.046, 0.041, 0.033) and late adolescents (0.046, 0.038, 0.052). Multilevel analyses are necessary when ICC values exceed the established threshold value of 0.05 (Geiser, 2012; Heck & Thomas, 2015). Despite the majority of our ICC values falling below this threshold, authors recommend proceeding with these analyses to ensure the exactness of estimates (Bressoux, 2020). Thus, using multilevel analyses was justified and allowed us to control for classroom variance. We then tested the association between co-rumination and each form of anxiety. If statistically significant, we verified whether the association varied across sex. As multiple statistical comparisons can increase error rates (i.e., incorrectly rejecting the null hypothesis), a false discovery rate (Benjamini-Hochberg correction; α=0.05) was used to adjust p-values. Analyses were done using R (R Core Team, 2020) and package lme4 (Bates et al., 2015). Unadjusted ICC values were reported to describe the proportion of variance in anxiety forms explained by the fixed effects of the models (Lüdecke et al., 2021).

Results

Preliminary analyses

Preliminary results revealed that girls co-ruminated more than boys (β= 2.45, p < .001) and late adolescents co-ruminated more than early adolescents (β = 2.65, p < .001). Compared to early counterparts, late adolescents scored higher on trait anxiety (β = .08, p < .001) and test anxiety (β = .08, p < .001), while levels of anxiety sensitivity did not differ statistically across age groups (p = .24). Co-rumination and anxiety scores are presented in Table 2. Missing data values were low (ranging from 2% to 3% for anxiety measures and approximately 7% for co-rumination; see Table 2) and thus, our estimates are unlikely to be biased (Bennett, 2001). Of note, higher missingness for co-rumination was likely attributable to being the last completed questionnaire.

Table 2. Descriptive statistics for continuous variables.

	Early adolescents				Late adolescents
	n	M (SD)	Range	Scaled mean (SD)	n	M (SD)	Range	Scaled mean (SD)
Co-rumination	268	28.44 (7.82)	9–45		856	31.14 (6.68)	9–45
Trait anxiety	282	35.24 (8.74)	20–59	0.38 (0.22)	904	38.21 (8.13)	20–60	0.46 (0.20)
Test anxiety	285	50.64 (16.53)	25–97	0.34 (0.22)	900	56.51 (15.02)	25–100	0.42 (0.20)
Anxiety sensitivity	280	29.43 (7.73)	18–52	0.32 (0.21)	894	30.16 (7.30)	18–54	0.34 (0.20)

Association between co-rumination and anxiety in early adolescents

Co-rumination was significantly associated with anxiety sensitivity (β = .005, p = .009, ICC = .04) but not with trait anxiety (β = .002, p = .16) or test anxiety (β = .003, p = .16; see Table 3). A beta of .005 implies that when co-rumination increased by 10 units (on a scale ranging from 9-45), anxiety sensitivity increased by .05 units on the 0–1 unit scale. This indicated that between the minimal and maximal (36 units) values on co-rumination, anxiety sensitivity will increase by .18 units. The association between co-rumination and anxiety sensitivity did not vary across sex (p = .27).

Table 3. Linear mixed-effects models of co-rumination on forms of anxiety for early adolescents.

Form of anxiety	Effect	Estimate	SE	df	t	P(>|t|)
1. Anxiety sensitivity	βco-rumination	.005	.002	253.30	2.93	.009
2. Test anxiety	βco-rumination	.003	.002	254.40	1.44	.16
3. Trait anxiety	βco-rumination	.002	.002	254.60	1.42	.16

Note. Model was adjusted for sex. Bold rows identify a statistically significant slope for the association between co-rumination and anxiety (p-values were adjusted for a false discovery rate).

Association between co-rumination and anxiety in late adolescents

Co-rumination was significantly associated with trait anxiety (β = .003, p = .009, ICC = .02), test anxiety (β = .003, p = .022, ICC = .03), and anxiety sensitivity (β = .004, p < .001, ICC = .03) as shown in Table 4. A beta of .004 implies that when co-rumination increased by 10 units, anxiety sensitivity increased by .04 units on the 0–1 unit scale. This means that between the minimal and maximal (36 units) values on co-rumination, anxiety sensitivity will increase by .14 units. We found that the associations did not vary across sex (p = .17, p = .56, p = .12, respectively).

Table 4. Linear mixed-effects models of co-rumination on forms of anxiety for late adolescents.

Form of anxiety	Effect	Estimate	SE	df	t	P(>|t|)
1. Anxiety sensitivity	βco-rumination	.004	.001	794.60	4.22	<.001
2. Test anxiety	βco-rumination	.003	.001	798.40	2.44	.022
3. Trait anxiety	βco-rumination	.003	.001	800.40	2.85	.009

Note. Model was adjusted for sex. Bold rows identify a statistically significant slope for the association between co-rumination and anxiety (p-values were adjusted for a false discovery rate).

Discussion

Our objective was to investigate the association between co-rumination and trait anxiety, test anxiety, and anxiety sensitivity in early and late adolescents. Co-rumination was associated with anxiety sensitivity in early adolescents, though the effect was small. In contrast, co-rumination was associated with all tested forms of anxiety in late adolescents, where the effects were also small. Further, these associations were consistent across sexes and co-rumination explained a similar proportion of variance in anxiety sensitivity for both age groups (based on ICCs).

The association between co-rumination and anxiety seems to appear as early as in the first years of adolescence. Specifically, we found that co-rumination was associated solely with anxiety sensitivity in early adolescents. This aligns with previous research describing co-rumination as an avoidant emotion regulation strategy (Stone et al., 2017, 2019; Waller et al., 2014), which may intensify the saliency of problems (Rose et al., 2007) and foster anticipatory fears. Indeed, non-solution-focused exhaustive co-ruminative discussions may revolve around experienced anxiety manifestations and their potential negative repercussions. In consequence, these discussions may instill fear of these manifestations (i.e., anxiety sensitivity).

Conversely, in late adolescents, co-rumination was associated with all tested forms of anxiety (trait anxiety, test anxiety, and anxiety sensitivity). This suggests a shift in the relationship between co-rumination and forms of anxiety according to age. Specifically, the emerging association between co-rumination and anxiety sensitivity in early adolescence may broaden to other forms of anxiety (i.e., trait and test anxiety) with age. Various developmental factors may explain the relationships between co-rumination and additional forms of anxiety in late adolescence. One potential factor is the changing nature of social relationships during this period, including the increasing focus on peer relationships and reduced time spent with parents between ages 10–13 and 14–15 (Baxter, 2017). These changes may modify the impact of co-rumination on anxiety. This coincides with findings that co-ruminating with friends could entail greater emotional risks (i.e., anxiety, depression) compared to problem talk with parents (Waller & Rose, 2013). A second potential factor could be the expressive abilities of negative emotions that may vary between early and late adolescence. Indeed, the expression of negative emotions in co-rumination plays a key role in internalizing symptoms (Rose, 2002). As older adolescents tend to be more adept at expressing negative emotions (Nippold, 2006; Schwartz-Mette & Rose, 2012), this may explain the emergence of new associations between co-rumination and different forms of anxiety across adolescence. Taken together, the fluctuating role of parents and the evolution of expressive abilities throughout adolescence may explain the shift we found in co-rumination dynamics. Given the multitude of changes and experiences that an individual will navigate throughout adolescence, studies should identify the occurrences unique to different periods in adolescence that may nourish the relationship(s) between co-rumination and anxiety forms.

Though this study may lead to interesting future directions, it had several limitations. First, co-rumination explained between 2% to 4% of the variance in the three forms of anxiety. This could be explained by our use of an adapted 9-item version of the original 27-item co-rumination questionnaire. The latter may produce slightly larger effect sizes than shorter versions (Spendelow et al., 2017) and future research should replicate our findings using the original questionnaire or via laboratory-induced co-rumination (e.g., Rankin et al., 2018; Tudder et al., 2023). Nevertheless, our findings align with the small effect sizes documented in the literature (e.g., Calmes & Roberts, 2008; Rose, 2002) suggesting that co-rumination may co-occur with anxiety in youth. Second, as a small number of parents consented for their early adolescents to participate, we may have had a recruitment bias in this group. For instance, only parents of anxious adolescents (or vice versa) may have consented. In contrast, a large majority of the invited late adolescents accepted to participate and therefore, this recruitment bias was not present in this group. In addition, our sample was from a select few schools which may not be representative of individuals elsewhere. Similarly, the large majority of our sample identified as White. As such, this minimal diversity limits the generalizability of our findings to other racial and ethnic groups. Third, the discrepancy in the sample sizes across age groups prevented us from statistically investigating the association between co-rumination and forms of anxiety across ages within the same analyses. Specifically, the lower participation rate in early adolescents prevented us from performing moderation analyses on school level. Fourth, our data was collected within classrooms. As shown by the ICC values, classroom-related variance had a small impact on the associations between co-rumination and anxiety. It could be interesting for future studies to investigate whether non-academic grouping environments (e.g., sports teams) have a greater influence on the observed associations. Lastly, our study was correlational and therefore, no conclusions could be made regarding causality.

Conclusion

This study adds to the existing literature by simultaneously exploring the association between co-rumination and various forms of anxiety in two different normative adolescent cohorts. Our findings suggest that the relationship between co-rumination and anxiety follows a developmental trajectory that begins with anxiety sensitivity and broadens to additional forms of anxiety. That said, longitudinal studies are required to confirm the evolution of these relationships within the same individual. Positive results from these studies could allow for the development of early interventions. For example, these interventions could not only educate youth about how the relationship between co-rumination and anxiety starts with a fear of anxiety symptoms but also promote the use of healthier alternatives to co-rumination. By raising awareness, these interventions may prevent the escalation of co-rumination and anxiety throughout adolescence.

Acknowledgements

We wish to thank the youth who participated in this study.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data for this manuscript is available with the appropriate link listed in the materials and methods section.

Additional information

Funding

This work was supported by a Foundation Grant from the Canadian Institutes for Health Research to SL [grant number 143282] and a doctoral studentship from the Fonds de recherche en Santé du Québec to AAJ. The work of RC was supported by a master’s scholarship from the Canadian Institutes of Health Research. SJL holds a Canadian Research Chair on Human Stress.