Compounding effects of domain-general cognitive weaknesses and word reading difficulties on anxiety symptoms in youth

Abstract

This study examined whether domain-general cognitive weaknesses in processing speed (PS) or executive functioning (EF) moderate the relation between word reading scores and anxiety, such that lower word reading scores in combination with lower cognitive scores are associated with higher anxiety symptoms. The sample included 755 youth ages 8–16 who were recruited as part of the Colorado Learning Disabilities Research Center twins study. Lower scores on PS (R² = .007, p = .014), EF (R² = .009, p = .006), and word reading (R² = .006-.008, p = .010-.032) were associated with higher anxiety scores. Additionally, the word reading × cognitive interactions were significant, such that lower scores on PS (R² = .010, p = .005) or EF (R² = .013, p = .010) combined with lower word reading were associated with higher-than-expected anxiety symptoms. Results suggest that weaknesses in PS, EF, and word reading are modestly associated with higher anxiety symptoms, and these anxiety symptoms may be compounded in youth with both PS or EF weaknesses and word reading difficulties. These findings can guide assessment approaches for identifying youth with word reading challenges who may be at increased risk for anxiety.

Up to 30% of children with reading difficulties have a comorbid anxiety disorder (Margari et al., 2013), a rate that is over twice as high as in children without reading difficulties (Carroll et al., 2005). Unfortunately, the co-occurrence of reading difficulties and anxiety is relatively under-studied compared to other comorbidities of reading difficulties (Hendren et al., 2018), and as a result, the factors contributing to the co-occurrence of these disorders are largely unknown (Haft et al., 2018). A lack of knowledge of the factors associated with comorbid anxiety in children with reading difficulties makes it difficult to determine appropriate screening and treatment approaches. The current study seeks to address this gap by assessing whether the association between word reading scores and anxiety symptoms is moderated by domain-general cognitive factors, such that lower word reading scores in combination with neurocognitive weaknesses may be associated with higher anxiety scores.

Reading Difficulties and Anxiety

Though there is a relative lack of research in this area, extant studies examining the comorbidity between reading difficulties and anxiety have been discussed in two narrative reviews and two meta-analyses that sought to clarify the precise magnitude of the prevalence rate of this comorbidity (Francis et al., 2019; Nelson & Harwood, 2011; Maughan & Carrroll, 2006; Mugnaini et al., 2009). While these amalgamations have concluded that reading difficulties are indeed associated with higher anxiety rates, the limited number of studies in this area has prevented quantitative evaluation of potential causes of heterogeneity across reported effect sizes (Francis et al., 2019). As such, it remains unclear to what extent comorbidity rates differ across age, demographic groups, or types of anxiety. Most importantly for the current study, little research has sought to examine the factors underlying this comorbidity, with most existing work focusing on establishing prevalence.

Though empirical evidence is somewhat sparse, three primary theories have been developed to explain the pathways leading to co-occurring reading difficulties and anxiety (Nelson & Hardwood, 2011). First, the correlated liabilities model postulates that anxiety and reading difficulties share genetic, neurobiological, or environmental risk factors that contribute to risk for both disorders. Behavioral genetic studies using twins have demonstrated that the genetic relationship between reading and anxiety is significant but modest (bivariate h² = .34, genetic correlation = .54; Willcutt, 2014), and therefore that shared risk factors may contribute to comorbid reading difficulties and anxiety.

The other two theories to explain the reading-anxiety relationship posit competing timelines of the emergence of anxiety and reading difficulties. Primary disorder theory proposes that anxiety can lead to word reading problems by using up space in the phonological loop of working memory (attentional control theory; Eysenck et al., 2007), while secondary reaction theory argues that anxiety develops as a consequence of reading difficulties. In this view, the experience of academic failure can lead to symptoms of anxiety that are initially centered on academics that then generalize to other domains. Longitudinal findings to date have most strongly supported secondary reaction theory by demonstrating that reading problems are observable before and predictive of later internalizing symptoms (e.g., Boetsch, 1997; Halonen et al., 2006; Lin et al., 2013). However, other evidence also indicates that reading and anxiety bidirectionally influence each other over time (Morgan et al., 2008).

Taken together, these theories and associated empirical evidence suggest several pathways that may lead to comorbid reading and anxiety. Based on the aforementioned longitudinal evidence, it seems likely that there is at least some causal connection between the experience of reading difficulties and later anxiety symptoms. As such, while the proposed study is cross-sectional and cannot establish temporal precedence of anxiety and reading problems (and therefore cannot differentiate between these theories, per se), results will be interpreted within the framework of secondary reaction theory, which is currently the most strongly supported by existing evidence. Further research is needed to better establish the development of this comorbidity over time.

Hypothesized Neurocognitive Risk Factors for Anxiety in Youth with Reading Difficulties

To date, only one study has examined the relation between neurocognitive factors and anxiety in the context of reading difficulties (Nelson et al., 2015). This study found that, amongst college students with reading disabilities, test anxiety was negatively associated with general intelligence, nonverbal ability, and working memory, but was not associated with reading skills, verbal ability and processing speed. Notably, this study focused on the correlations between cognitive factors, reading, and anxiety, and did not test interactions between reading and cognitive factors. Following these findings, the current study will use a school-aged sample to examine the relation between cognitive variables and word reading in their association with anxiety earlier in development. Additionally, this study will expand beyond test anxiety to examine symptoms from the major anxiety domains (i.e., separation, generalized, social, and somatic) and will also examine whether neurocognitive variables interact with word reading scores to statistically predict higher anxiety scores in youth with word reading difficulties compared to youth without word reading difficulties.

Processing speed (PS) and executive functioning (EF) were selected as the primary cognitive constructs for the current study, as they are implicated in both the word reading and anxiety literatures. Specifically, previous research has established the role of both PS and EF in word reading challenges (for reviews see Kudo et al., 2015; McGrath et al., 2011; Peterson & Pennington, 2015), though not all children with word reading challenges have these difficulties (Pennington et al., 2012). Even though PS and EF are known to be associated with word reading, this does not preclude the statistical examination of the interactions between these variables, as two variables may interact in their prediction of a third variable whether or not they are correlated (Grace-Martin, 2018).

Both PS and EF are also theoretically and empirically associated with anxiety, though the research is less definitive than in word reading. First, considering PS, theoretical explanations emerging from clinical observation have suggested that children with slower PS may feel as though they cannot keep up with the pace of their academic and social environments, making them more likely to experience anxiety symptoms (e.g., Braaten & Willoughby, 2014). Though empirical findings are somewhat mixed, a meta-analysis of the neuropsychological weaknesses found across developmental disorders documented that PS was one of the most consistent weaknesses across diagnostic categories, including anxiety (Willcutt et al., 2008). This evidence suggests that PS may be a generalized risk factor for multiple developmental disorders, and it is therefore possible that PS may magnify the risk for anxiety in the context of lower word reading.

Next, previous theorizing has implicated EF as a mechanism underlying emotion regulation (Zelazo & Cunningham, 2007), which is considered an etiological factor in anxiety disorders (Cisler et al., 2010). There are three core types of EF (inhibition, updating/working memory, and shifting), which share variance (i.e., common EF) but are also distinguishable from one another in the “unity and diversity model” of EFs advanced by Miyake et al. (2000). While previous research has suggested that EFs are related to anxiety, including in children (e.g., Kertz et al., 2015; Ursache & Raver, 2014; Zainal & Newman, 2018), it is not currently clear whether these findings are driven by the common or specific aspects of EF. Given that common EF is the most predictive of psychopathology (Snyder et al., 2015), the current research will use multiple measures of EFs to create a latent common EF variable in order to focus on examining the shared variance amongst EFs as a potential moderator of the association between word reading scores and anxiety symptoms.

The Current Study

It is currently unclear what factors may be implicated in the elevated rates of anxiety symptoms observed in youth with word reading challenges compared to youth without word reading challenges. To address this gap, the proposed study tests the hypothesis that domain-general neurocognitive weaknesses in PS and EF may interact with lower word reading scores to predict higher anxiety symptoms. The goal of this study is to help guide the development of screening and intervention efforts for comorbid word reading difficulties and anxiety. In terms of screening, should the association between word reading scores and anxiety symptoms depend on cognitive scores, lower performance on measures of PS and EF may be considered an indicator of potential risk for comorbidity. With respect to intervention, we note that some recommendations in schools are currently at odds for students with comorbid anxiety and word reading difficulties. Students with learning disabilities who have weaker PS and EF often receive accommodations such as extra time, while treatment for students with anxiety involves exposure to feared situations, such as enduring timed testing situations if that is an area of focus for treatment. Should cognitive scores moderate the association between word reading and anxiety, this would suggest avenues for resolving these discrepant recommendations. Specifically, it would suggest a need to balance appropriately accommodating students’ cognitive and academic challenges while also treating anxiety and preventing the perpetuation of avoidance behaviors through psychological intervention (i.e., preventing avoidance of feared situations, and instead helping children to developing skills to cope with distress and fear while enduring feared situations). Additionally, results from this study can provide avenues for future longitudinal work examining factors implicated in the development of comorbid word reading difficulties and anxiety.

Method

Participants

Participants were recruited from the Colorado Learning Disabilities Research Center (CLDRC) twin sample. Previous studies have documented the study design and recruitment practices of the CLDRC (e.g., DeFries, 1997; Gayán & Olson, 2001; McGrath et al., 2011; Willcutt et al., 2005; Willcutt et al., 2019). To briefly summarize, the CLDRC is an ongoing, community-based study examining twins recruited from the front range of Colorado. Twins living within 150 miles of metropolitan Denver were identified and recruited through 22 local school districts or through Colorado’s twin registry. Inclusion criteria for participation in the CLDRC required that children resided in primarily English-speaking homes, did not have a history of neurological conditions, genetic syndromes, or traumatic brain injuries, did not have uncorrected visual impairments, and were not deaf or hard of hearing.

For the present analyses, we included all participants in the CLDRC sample with complete data on the Revised Children’s Manifest Anxiety Scale (RCMAS), our primary outcome measure of interest. We also required participants in these analyses to have a Verbal IQ or Nonverbal IQ above 85 and a Full-scale IQ above 70 on the WISC-R or WISC-III (Wechsler Intelligence Scale for Children, Revised or 3^rd Edition; Weschler, 1974; Wechsler, 1991). One twin from each twin pair was randomly selected for inclusion in the present analyses to allow for the preservation of the assumption of independence required for statistical procedures. Participants were between the ages 8 and 16 (N = 755). Additional demographic characteristics of the sample are outlined in Table 1.

Table 1.

Sample Characteristics

Participant Demographics	Mean	Standard deviation	Range
Age (years)	11.3	2.3	8.1–16.5
Caregiver Education (years)1	15.4	2.2	7–20
Full Scale IQ Standard Score2	107.9	13.0	73–144
Verbal IQ Standard Score2	109.2	14.3	64–147
Performance IQ Standard Score2	105.1	12.9	67–147
PIAT Reading Recognition Standard Score	103.8	12.3	65–135
RCMAS Total Anxiety score (28 items)3	16.38	12.4	0–79
Parent-Report DSM-IV ADHD Rating Scale Total Score (18 items)4	12.27	10.66	0–50
Teacher-Report DSM-IV ADHD Rating Scale Total Score (18 items)4	10.02	11.36	0–50
Sex as Identified by Parent 5	Percentage
Female	50.60
Male	49.40
Race	Wave 1 (1997–2006) 7	Wave 2 (2006-current) 7
Asian	0.0%	0.0%
Black	<2.0%	<2.0%
Hispanic or Latino	2.0%	--
Multiple groups identified6	15.0%	10.5%
Native American/American Indian/Alaska Native/Indigenous	0.0%	<2.0%
White	82.1%	85.9%
Prefer to self-describe	0.0%	<2.0%
Ethnicity		Wave 2 (2006-current) 7
Hispanic or Latino		2.5%
Multiple ethnicities identified6		10.5%
Not Hispanic or Latino		87.0%

Note. IQ = Intelligence quotient; PIAT = Peabody Individual Achievement Test; RCMAS = Revised Children’s Manifest Anxiety Scale; DSM-IV = Diagnostic and Statistical Manual of Mental Disorders, fourth edition; ADHD = Attention-Deficit/Hyperactivity Disorder

¹In the cases in which we had two caregivers reporting years of education (95%), we averaged those values. In the cases in which we had only one caregiver reporting education (5%), we took that value.

²Either the WISC-R or WISC-III was used to calculate IQ standard scores because this is a long-standing study that prioritizes consistency in measures over time. Note that using older measures would be expected to result in inflated standard scores because of the Flynn effect (Flynn, 1984).

³Alternative scoring was used for RCMAS where responses were indicated using a four-point scale (0=Not at All, 1=Just a Little, 2=Pretty Much, 3=Very Much).

⁴Reported on a four-point scale from (0 =Not at all, 1=Just a Little, 2=Pretty Much to 3 =Very Much).

⁵Sex was assessed as binary and did not include intersex as an option. We did not assess self-reported gender.

⁶In earlier phases of data collection, parents self-reported race and ethnicity for themselves but not their children. We indicate here if parents endorsed multiple identifications. We want to note the limitations of this approach, however, as it does not capture the identification that families and children may choose for themselves. We have made revisions to the race and ethnicity data collection to align with current best practices for inclusiveness in research studies (e.g., Wadsworth et al., 2016).

⁷Our earliest waves of data collection included a single variable that combined race and ethnicity, consistent with federal guidance at the time. Since 2006, we have been collecting race and ethnicity information separately. For additional confidentiality protections for participants, if the percentage representation of a group is less than 2% we indicate <2%.

The recruitment procedures for the CLDRC involved over-sampling for reading and/or attention difficulties (Willcutt et al., 2019). A comparison group was also recruited and included twin pairs in which neither child met the screening criteria for a history of reading or attention difficulties. Notably, while participants are recruited for the CLDRC to be part of clinical or control groups, in reality the distributional properties of the academic and cognitive variables approximate a normal distribution in this sample. This is because co-twins of clinical probands exhibit a wide range of scores and we randomly selected whether the proband or co-twin would be included in this sample. Moreover, despite screening procedures, twins recruited for the control group sometimes had previously undetected reading and attention difficulties and twins who were recruited for the clinical group sometimes performed well on reading and attention measures, perhaps as a result of effective interventions. As such, the present analyses utilize the full sample, consistent with previous analyses in this sample (McGrath et al., 2011; Peterson et al., 2017), and analyses will consider continuous distributions on all measures, rather than conducting group comparisons between clinical and control groups. Histograms of key measures are available in Supplemental Figures 1–5.

Procedure

The Institutional Review Boards at the University of Colorado, Boulder (CU) and the University of Denver (DU) approved the study protocol. Informed consent from parents or legal guardians and assent from children was obtained at both institutions. Participants completed a 6-hour testing session at CU, followed by another 6-hour testing session at DU approximately 2 months later (median time between test dates = 69 days). The testing sessions at CU included the word reading and cognitive measures, such as IQ and PS. The testing sessions at DU included the socioemotional assessments (i.e., anxiety and depression) and the majority of the EF measures. Testing was conducted by trained research assistants with bachelor’s degrees or doctoral level graduate students. Examiners were trained to be sensitive to fatigue and to offer frequent small breaks (in addition to a longer lunch break) and behavioral support to maintain motivation (i.e., sticker charts). Examiners were blind to participants’ diagnostic status. Participants taking psychostimulant medication were asked to withhold the medication for 24 hours prior to testing.

Measures

Word Reading

Consistent with prior research from the CLDRC project (e.g., McGrath et al., 2011; Peterson et al., 2017), a single-word word reading composite was generated using scores from the PIAT Reading Recognition, PIAT Spelling, and Time-Limited Oral Reading tests (correlations between tasks: r = .84-.93). The PIAT Reading Recognition task involves reading words with increasing semantic and phonemic complexity (test-retest reliability = .85; Dunn & Markwardt, 1970). The PIAT Spelling task assesses orthographic word recognition by measuring participants’ selection of the correct spelling of an orally presented word from various phonologically similar words (test-retest reliability = .64; Dunn & Markwardt, 1970). While this task is referred to as “spelling,” it is not a spelling production task, but a word recognition task. Finally, Time-Limited Oral Reading involved reading words of increasing difficulty and required that subjects initiate a verbal response within a 2-second time-limit after the word was presented (test-retest reliability = .94; Olson et al., 1994; Olson et al., 1989). Though timed, the speed demands of this task are minimal because of the 2-second time window to initiate a response. Given that diagnoses of specific learning disabilities reflect an arbitrary cut-point on a continuous distribution of abilities (Peters & Ansari, 2019), analyses examined word reading as a continuous measure, rather than imposing arbitrary categorical distinctions of youth with and without word reading disabilities.

Anxiety

Children’s anxiety was indexed using the RCMAS (Revised Children’s Manifest Anxiety Scale; Reynolds & Richmond, 1978) self-report measure (test-retest reliability: .77-.88; Wisniewski et al., 1987). Self-reports of internalizing symptoms are considered preferable to parent-reports, based on the notion that youth are better reporters of their own internalizing experiences than are their parents, as the report reflects their own subjective internal experience (Salbach-Andrae et al., 2009). The RCMAS contains 28 items that are summed to create a Total Anxiety score. Nine items are also included in the measure to assess response biases (Reynolds & Richmond, 1978) but were not included in this study due to time constraints. In the present research, severity of anxiety symptomology was determined using continuous distributions of total scores on the RCMAS. While the original RCMAS requires participants to respond to items by indicating either Yes or No, in the current research responses were indicated using a four-point scale in order to capture symptom severity, rather than just the presence or absence of symptoms (0 = Not at all, 1 = Just a Little, 2 = Pretty Much, 3 = Very Much). This modified version of the measure demonstrates strong internal consistency in this sample (Cronbach’s alpha = .93).

Processing Speed

A PS composite was generating using Coding and Symbol Search from the WISC-III or WISC-R, the Colorado Perceptual Speed Test, and the Identical Pictures Task (correlations between tasks: r = .78-.83). Coding requires participants to copy symbols associated with numbers in a key as quickly as possible (test-retest reliability = .72; Wechsler, 1974), and Symbol Search requires participants to examine rows of non-namable symbols and mark whether any target symbols on the left side of the page are in the choices on the right side of the page (test-retest reliability = .81; Wechsler, 1974). The task contains a mix of trials with matches and trials without matches. The Colorado Perceptual Speed Test involves selecting matching unpronounceable strings of letters or letters and numbers from an array of options including three foils (test-retest reliability = .81; Decker, 1989). Finally, in the Identical Pictures Task participants are shown a target picture and have to select a matching picture from a set of options including four foils (reliability = .82; French et al., 1963).

Executive Functioning

EF was measured using a latent bifactor model consistent with the unity and diversity model of Miyake et al. (2001). Factor scores for common EF were derived from the latent model (Figure 1). The indicators used in this model are detailed below.

Figure 1.

Bifactor Model of Executive Functioning (EF)

Note. Black indicates factors and loadings relevant to the present analyses. Grey indicates factors that were not used in further analyses but were necessary for adequate model fit and are consistent with the unity and diversity model. EF = Executive functioning; RT = Response time; GDS = Gordon Diagnostic System; Vig = Vigilance; Dist = Distractibility; Com = Commission errors.

** = p < .01

Set Shifting was measured using the Trailmaking Test Part B completion time (Reitan & Wolfson, 1985). Part B requires participants to connect circles in ascending order, alternating between circles containing numbers and letters (i.e., 1, A, 2, B, 3, C…). In order to identify the latent factor for shifting which only had one indicator, Trailmaking Part B, we had to directly specify the error variance of this measure. We modelled the reliability of Trailmaking Test Part B time at r = .65, which is between the test-retest reliability values in youth (r = .65 - .71; Barr, 2003; McLeod et al., 2006) and the estimates derived from the monozygotic (MZ) twin correlations in the full sample (N = 428; r = .61). The MZ twin correlation can be used as a lower-bound estimate of test-retest reliability as MZ twins share genetic and shared environment features, and only differ in terms of unique environment (Plomin et al., 2001).

Inhibition was indexed using three measures: The Stop Signal Reaction Time (SSRT) task, and commission errors on the Vigilance and Distractibility tasks of the Gordon Diagnostic System (correlations between SSRT and Gordon measures = .39; correlation between Gordon Vigilance and Distractibility = .74). The SSRT task is a computer task that requires participants to press a key to a target but withhold pressing the key when they hear a beep which is presented at different stimulus onset asynchrony (SOA) times (test-retest reliability for SSRT in ADHD children = .72; Soreni et al., 2009; split half reliability = .95; Logan et al., 1997). The stop-signal reaction time refers to the average amount of time that it takes a participant to inhibit their prepotent response, with longer SSRTs reflecting weaker inhibitory control (Logan et al., 1997). Scores used here combine two comparable versions of the SSRT task which were standardized within-task before combining.

The next two measures of inhibition were derived from the Gordon Diagnostic System (Gordon & McClure, 1983), a continuous performance test. During the Vigilance Task participants see numbers flash on a screen and are required to press a button each time they see a specific sequence of digits (e.g., a 1 followed by a 9). The Distractibility Task is similar, with the addition of extraneous numbers that flash at random intervals beside the target numbers. Here, we used commission errors (the number of non-target button presses) during both tasks to index inhibitory control (test-retest reliability for Vigilance Commissions = .84-.94; test-retest reliability for Distractibility Commissions = .85; Gordon & Mettelman, 1988), consistent with previous studies using this sample (e.g., McGrath et al., 2011).

Verbal working memory was indexed using scores from WISC-III and WISC-R Digit Span Backwards, Sentence Span, and Counting Span (correlations between tasks: r = .46-.55). While the correlations between the tasks are modest, we have found the shared variance between these tasks to be a reliable predictor of word reading and ADHD symptoms in previous latent models (McGrath et al., 2011). In Digit Span Backwards, participants are presented with strings of numbers of increasing length and they are required to repeat them in the reverse order (test-retest reliability for Digit Span Forwards and Backwards = .73; Wechsler, 1974). Sentence Span requires participants to generate the final word to complete simple sentences, and to then repeat these last words in the correct order as the number of sentences increases across trials (test-retest reliability = .65-.71; Kuntsi et al., 2001; Siegel & Ryan, 1989). Finally, Counting Span involves having participants count the number of yellow dots on a page with both yellow and blue dots, and to then recall the number of dots that were presented on each page at the end of each set, which consist of an increasing number of trials (test-retest reliability = .55-.67; Case et al., 1982; Kuntsi et al., 2001).

Covariates

IQ.

In the current analyses, IQ was indexed using a composite of non-verbal subtests (Object Assembly, Picture Completion, Picture Arrangement, and Block Design) on the WISC-III (37.75% of the sample) or WISC-R (62.25% of the sample). Nonverbal subtests were selected because of the known relationship between dyslexia and verbal IQ. These subtests were also selected to reduce overlap with the focal cognitive predictors, as opposed to using Full-scale IQ since working memory and PS measures are also part of the Full-scale IQ estimate. However, since these subtests are timed, they will be overly conservative in models including PS. Notably, older versions of the WISC (WISC-R or WISC-III) were used because this is a long-standing study that prioritizes consistency in measures over time. Note that using older measures would be expected to result in inflated standard scores because of the Flynn effect (Flynn, 1984).

ADHD.

Given that ADHD is highly comorbid with both word reading (Willcutt & Pennington, 2000) and anxiety (Jarrett & Ollendick, 2008), a measure of ADHD symptoms was included as a covariate in the present analyses, to determine whether any observed effects were robust to controlling for ADHD. ADHD symptoms were determined using the Disruptive Behavior Rating Scale, which was completed by the child’s caregivers and teacher (test-retest reliability .66–94; Barkley & Murphy, 1998). This scale requires caregivers and teachers to rate DSM-IV symptoms of inattention and hyperactivity/impulsivity from 0 (not at all) to 3 (very often). ADHD was included in the current analyses as a continuous measure of symptom severity, calculated by taking the most severe rating from caregivers or teachers for each symptom (Lahey et al., 1994).

Depression.

Anxiety and depression are known to be comorbid (Costello et al., 2005; Garber & Weersing, 2010) and have some overlapping symptoms (Zbozinek et al., 2012). As such, a measure of depression was used as a dependent variable in secondary regression analyses in order to determine whether effects are specific to anxiety or are broadly applicable to internalizing symptoms. Depression was measured using the Children’s Depression Inventory self-report measure (CDI; Kovacs, 1981). The CDI consists of 27 items that ask the child to endorse one of three descriptions that best describes how they have been feeling during the past two-week period (Saylor, Finsh, Spirito, & Bennett, 1984; test-retest reliability: .66-.82; Finch et al., 1987). As has been previously discussed in the literature, there is a high degree of overlap between the CDI and RCMAS, with six very similar items on both measures (Seligman & Ollendick, 1998). In our sample, the correlation between these measures was strong, r = .68.

Analyses

Raw scores for word reading, EF and PS measures were standardized within our sample to create composite scores. Even though different versions (WISC-R/WISC-III) of Coding and Digit Span Backwards were administered, the tests were nearly identical and so it was appropriate to use raw scores. Only the WISC-III version of Symbol Search was administered, and so again raw scores were appropriate. Scaled scores were used to generate the nonverbal IQ composite, because we were combining across test versions that had different items (i.e., WISC-III and WISC-R). Outliers were winsorized to 4 standard deviations, and three variables (Trails B Completion Time and Gordon Vigilance and Distractibility Commissions) that exhibited extreme violations of normality (skew >3, kurtosis >3) were transformed using the negative inverse. Given that Bonferroni corrections are too conservative for use with analyses examining correlated variables (i.e., cognitive factors, comorbidities; Bland & Altman, 1995), we accounted for multiple testing effects by running hierarchical regressions to examine the significance of predictors as a group, prior to exploring the significance of individual variables.

Results

What Demographic, Comorbidity, and Cognitive Features are Associated with Anxiety?

As displayed in Table 2, hierarchical regressions revealed that word reading was modestly but significantly associated with anxiety above and beyond the demographic covariates in the first block, and over and above ADHD in the second block. In the third block containing interaction terms between word reading and the demographic and comorbidity variables, only the word reading × age interaction was individually significant. Further simple slopes analysis using the Stata 15.1. margins function indicated that word reading was only significantly related to anxiety in participants under 11.3 years (see Supplemental Figure 6 for simple slopes plot). However, we decided to proceed with analyses in the full sample (8–16 years). This decision was made to preserve the largest sample size and because we were interested in word reading × cognition interactions in predicting anxiety and we recognized that interactions are sometimes present without main effects (in this case, a lack of a word reading main effect on anxiety in older children). We conducted secondary analyses in the younger sample only (8–11.3 years) to test for consistency in overall pattern of results, which was largely replicated (see Supplemental Analyses).

Table 2.

Hierarchical Regressions Involving Demographic and Comorbidity Variables Predicting RCMAS Total (Shading Indicates the Variables Entered in Each Respective Block of the Analyses)

	Block 1: Demographics				Block 2: Comorbidities				Block 3: Interactions
Variable	B	SE	β	P	B	SE	β	P	B	SE	β	P
Constant	14.94	2.86		<.001	13.53	2.80		<.001	12.55	2.81		<.001
Age	−1.06	.66	−.09	.111	−1.42	.65	−.11	.029	−1.99	.67	−.16	.003
Sex	1.29	.99	.05	.192	3.13	1.01	.13	.002	3.21	1.00	.13	.001
Maternal Education	.07	.18	−.02	.706	.10	.17	.02	.567	.08	.17	.02	.651
Word Reading	−2.62	.67	−.21	<.001	−1.78	.67	−.14	.008	−5.87	3.46	−.47	.091
ADHD					3.09	.53	.24	<.001	3.04	.53	.24	<.001
Word Reading × Age									1.67	.53	.13	.002
Word Reading × Sex									−.50	.99	−.03	.613
Word Reading × Maternal Education									.29	.21	.37	.172
Word Reading × ADHD									−.20	.53	−.02	.968
N	600				599				599
R2 Change	.08				.05				.02
p-value	<.001				<.001				.005

Note. ADHD = Attention-Deficit/Hyperactivity Disorder. Bolding indicates the variables entered in each respective block of the analyses.

Do Neurocognitive Skills Moderate the Association Between Word Reading and Anxiety?

EF Bifactor Model

A bifactor model was constructed to allow us to represent both the relationships among all of the measures of EF (i.e., common EF) as well as the subdomains of EF (i.e., shifting-specific, inhibition-specific and working memory-specific variance). The model provided satisfactory fit to the data, Χ²(21) = 1853.528, p < .001, CFI = 1.0, RMSEA = .015 (90% CI 0-.047), SRMR = .012. All measures loaded significantly on the Common EF factor (p < .001). Factor scores for common EF were derived from this model for further analysis. Domain-specific factors were not used for further analyses given recent theoretical and empirical concerns focused on statistical irregularities with the group-specific factors (Bonifay et al., 2017; Eid et al., 2017).

We ran a hierarchical regression with word reading and demographic covariates that were significant in any of the previous models (i.e., age and sex) in block 1, cognitive predictors in block 2, and word reading × cognition interaction terms in block 3. Each block contributed significant variance to the prediction of anxiety (see Table 3). Consistent with our a priori strategy for post-hoc follow-up of significant blocks, further regressions were conducted to examine the main effects of individual cognitive terms, as well as to examine whether they moderated the relation between word reading and anxiety. Separate regressions were run in order to explore unique effects of individual cognitive predictors that may not have been evident given the degree of multicollinearity in the initial model where all cognitive predictors were entered simultaneously.

Table 3.

Hierarchical Regression Predicting Anxiety From Word Reading and Cognitive Factors

	Block 1: Demographic Covariates				Block 2: Cognitive Main Effects				Block 3: Word Reading × Cognitive Interactions
Variable	B	SE	β	P	B	SE	β	P	B	SE	β	P
Constant	15.69	.62		<.001	15.59	.63		<.001	14.46	.72		<.001
Age	−.94	.59	−.08	.111	.005	.78	.01	.995	.045	.77	.02	.953
Sex	1.30	.87	.05	.135	1.50	.90	.07	.094	1.80	.89	.09	.044
Word Reading	−2.38	.59	−.19	<.001	−1.30	.73	−.10	.073	−.98	.73	−.06	.181
PS					−.84	.99	−.07	.399	−1.33	1.03	−.12	.197
Common EF					−1.46	.75	−.07	.053	−1.48	.75	−.10	.049
Word Reading × PS									−.21	.75	.002	.781
Word Reading × Common EF									1.55	.72	.04	.031
N	755				755				755
R2 Change	.064				.009				.015
p-value	<.001				.032				.003

Note. PS = Processing speed; EF = Executive functioning. Bolding indicates the variables entered in each respective block of the analyses.

Results of the follow-up regression analyses are depicted in Table 4. We found small but significant main effects for PS (R² = .007, p = .014) and common EF (R² = .009, p = .006) after controlling for age, sex, and word reading scores. We also found significant interactions between word reading × PS (R² = .010, p = .005) and word reading × common EF (R² = .013, p = .001). Graphical depictions of these interactions can be seen in Figure 2. The interactions show a similar pattern, in which lower word reading and lower PS or lower common EF is associated with higher than expected anxiety symptoms based on either risk factor alone. We followed up these significant interactions by controlling for ADHD and nonverbal IQ to examine whether effects were robust to comorbidities and general cognitive functioning. Both interactions remain significant when controlling for ADHD (word reading × PS: R² change = .009, p = .006; word reading × common EF: R² change = .014, p = .001) and IQ (word reading × PS: R² change = .011, p = .003; word reading × common EF: R² change = .015, p < .001).

Table 4.

Multiple Regressions of Word Reading, EF, and PS Predicting RCMAS Total Scores

	Model 1: PS				Model 2: Common EF
Variable	B	SE	β	P	B	SE	β	P
Constant	14.55	.71		<.001	14.65	.69		<.001
Age	.00	.78	<.01	.999	−.59	.63	−.05	.349
Sex	1.79	.89	.07	.044	1.48	.86	.06	.087
Word Reading	−1.38	.69	−.11	.047	−1.27	.69	−.10	.064
PS	−2.31	.94	−.19	.014
Word Reading × PS	1.17	.42	.10	.005
Common EF					−1.87	.69	−.15	.006
Word Reading × Common EF					1.29	.39	.12	.001
N	755				755
Adjusted R2	.071				.078
p-value	<.001				<.001

Note. PS = Processing speed; EF = Executive functioning

Figure 2.

Simple Slopes Plot of the Word Reading × Processing Speed (PS) and Word Reading × Common Executive Functioning (EF) Interaction

Note. RCMAS = Revised Children’s Manifest Anxiety Scale.

Finally, to determine whether observed cognitive effects were specific to anxiety or were applicable to internalizing symptoms more broadly, follow-up regression analyses were run using the CDI as a dependent variable. Results mirrored those found using the RCMAS, suggesting the observed effects may be applicable to symptoms of both anxiety and depression. Analyses and further discussion can be found in the Supplemental Analyses.

Discussion

The literature has yet to specify what factors contribute to the higher than expected anxiety symptoms in youth with reading difficulties compared to youth without reading difficulties. The current research sought to address this gap by examining whether differences in cognitive profile may interact with reading scores to statistically predict anxiety symptoms. Specifically, we tested the hypothesis that weaknesses in PS and common EF would be associated with higher anxiety symptoms in youth with word reading difficulties as compared to peers who do not have word reading difficulties. Consistent with previous literature, there was a relation between lower word reading scores and higher anxiety scores. Similarly, results showed that weaknesses in PS and common EF were associated with increased anxiety symptoms across youth in the sample regardless of word reading ability. Consistent with our hypotheses, these cognitive abilities also interacted significantly with word reading, such that lower scores on PS and EF in the context of lower word reading was associated with compounded anxiety symptoms, albeit with small effect sizes. Below, we discuss the influence of demographic features and comorbidities on the relation between word reading and anxiety. Next, we interpret our neurocognitive findings. Finally, we discuss limitations of the current study as well as future directions for this research.

Demographic Features and Comorbidities

We first examined the extent to which word reading was associated with anxiety after controlling for demographic and comorbidity covariates. Word reading was significantly associated with anxiety above and beyond ADHD symptoms, age, sex, and caregiver education. That the relationship between word reading and anxiety holds when controlling for ADHD is important because children with ADHD are known to have high rates of comorbid anxiety (Jarrett & Ollendick, 2008). Additionally, ADHD has been shown to mediate the association between reading difficulties and externalizing problems as well as reading difficulties and depression, which are reduced to non-significance when inattention is controlled (Carroll et al., 2005; Willcutt & Pennington, 2000). In contrast, the relation between reading difficulties and anxiety remains significant when controlling for inattention, as demonstrated by the present findings as well as previous research (Carroll et al., 2005; Willcutt & Pennington, 2000), suggesting that there is true comorbidity between reading difficulties and anxiety.

Age and word reading were found to significantly interact in their association with anxiety. Plotting this interaction revealed that word reading difficulties were only significantly associated with anxiety in children under age 11.3 years. This finding corresponds with an unusual trend in the literature where younger children score higher on the RCMAS than older youth (Reynolds & Richmond, 1978; Reynolds & Paget, 1983; Turgeon & Chartand, 2003), and may suggest that the RCMAS is less sensitive to anxiety in older children (Seligman & Ollendick, 1998). It is somewhat surprising that this age-related pattern exists, given that the RCMAS measures anxiety symptoms that tend to increase with age (e.g., social anxiety) and does not measure anxiety symptoms that are typically higher in younger children (e.g., separation anxiety; Costello et al., 2003). Notably, however, the present age × word reading interaction does not seem to be entirely attributable to the RCMAS, as these results are also consistent with recent findings from Horbach et al. (2020) using parent reports on the Child Behavior Checklist. Specifically, this study found that children with learning disabilities showed increasing internalizing symptoms until a peak in fourth grade, prior to a significant decline in symptoms thereafter, lending some credence to these findings across samples and raters. Unfortunately, the lack of studies in this research area has prevented quantitative evaluation of moderators such as age on the reading-anxiety relationships (Francis et al., 2019). Thus far, narrative results suggest that lower reading is a risk factor for anxiety across age (Francis et al., 2019; Mugnaini et al., 2009). Longitudinal analyses on this topic are also limited, and therefore there is a dearth of evidence examining how the relation between reading and anxiety varies over time (e.g., Boetsch, 1997; Halonen et al., 2006; Lin et al., 2013).

If this developmental pattern continues to hold across measures where children with learning disabilities show elevated internalizing symptoms in early elementary school, it could have interesting applications to theories of reading development. For example, given that the present study addressed single word reading skills, it may be that difficulties in this domain are particularly salient during the “learning to read” stage of reading development during elementary school (Chall, 1983; 1996). In contrast, perhaps in later grades when there is relatively less emphasis on single word reading (i.e., “learning to read”) and more emphasis on comprehending what is read (i.e., “reading to learn”), single word reading skills are less strongly related to anxiety. Moreover, the onset of the pre-adolescent period brings additional dimensions to school success (i.e., peer relationships) such that word reading difficulties might diminish in importance as a stressor. Future research should examine whether this pattern of results differs when assessing the relation between reading comprehension and anxiety as opposed to single word reading and anxiety, as well as examining the relative weighting of stressors for students over time.

Neurocognitive Skills

After identifying important relations between demographic variables, word reading and anxiety, we explored the role of neurocognitive factors in the relation between word reading and anxiety. We found that weaknesses in PS and common EF were associated with increased anxiety symptoms, after controlling for word reading ability. These results are consistent with prior findings demonstrating that both of these cognitive factors are associated with anxiety (PS: Basso et al., 2007; Castaneda et al., 2011; EFs: Kertz et al., 2015; Ursache & Raver, 2014; Zainal & Newman, 2018). However, given our modest effects sizes as well as previous research suggesting that there is a relatively inconsistent neurocognitive profile in anxiety disorders compared to other psychological and learning disorders (Willcutt et al., 2008), more research is still needed in this area to determine the degree to which these cognitive factors are consistently implicated in anxiety disorders.

We next examined interactions between word reading and our cognitive predictors. PS and common EF were found to interact significantly with word reading, such that lower scores in these cognitive domains in the context of lower word reading was associated with compounded anxiety symptom elevations. These findings implicate cognition as one factor in the association between word reading and anxiety. In line with secondary reaction theory, the notion that word reading difficulties can contribute to later anxiety symptoms, we propose that children who have weaknesses in domain-general cognitive abilities like PS and EF may be at risk for developing anxiety due to the ways in which these weaknesses impact their experience in the classroom. Difficulties with both PS and EF may lead a child to feel overwhelmed in class, as they may feel that they cannot keep up with the pace and expectations of the learning environment. In turn, this may lead them to feel anxious. While this could impact all children with weaknesses in these domains, regardless of word reading ability (which would be consistent with our main effect results), it may compound risk for anxiety in children with word reading difficulties who may already be experiencing concerns and low self-esteem in the academic setting.

While we interpret these findings in the context of secondary reaction theory, our cross-sectional design does not allow us to comment on causality. We cannot rule out the possibility that anxiety also contributes to reading difficulties, as in primary disorder theory. In fact, research has suggested that reading and anxiety may also bidirectionally influence each other over time (Grills-Taquechel et al., 2012; Ramirez et al., 2019). As such, further longitudinal research is needed to more clearly delineate the relationship between reading, anxiety, and cognitive abilities over time.

Clinical Implications

These results suggest potential avenues for early screening, prevention, and intervention. Specifically, results in this area can contribute leads for developing early screening efforts to identify youth with word reading difficulties who are at risk for anxiety, as well as help to inform efforts to develop intervention practices. While these results suggest future directions, we acknowledge that the very small effect sizes preclude current action steps, and also that longitudinal research is needed to empirically examine the proposed temporal relations between cognitive, word reading, and anxiety variables.

Youth with learning and mental health challenges often receive interventions in the school setting that include academic accommodations. Currently, accommodations for students with learning disabilities who have weaker PS and EF often involve extra time on tests and assignments and modified assignments that might have fewer items but cover the same core principles. In contrast, the core evidence-based treatment technique for anxiety is exposure to feared situations (e.g., Compton et al., 2004; Whiteside et al., 2020). Thus, if relevant to the child’s anxiety symptoms, exposure could include enduring timed testing situations and being required to complete full assignments within expected timeframes. As such, intervention recommendations in schools are at odds for students with comorbid anxiety and reading difficulties. The present findings suggest that, if students with anxiety demonstrate reading difficulties and weaknesses in PS or EF, having them endure timed testing situations as an exposure, for example, may not be an efficacious treatment, as these students may not be able to complete work within typical time constraints, which may make exposures ineffective in terms of promoting self-efficacy and challenging beliefs. Instead, these findings suggest that such students may benefit more from other therapeutic techniques such as relaxation training and cognitive restructuring (i.e., modifying unrealistic and negative thought patterns) to address anxiety symptoms. As such, a skills-based approach to anxiety treatment, in addition to academic intervention and accommodations such as extended time on tests and assignments, may be the best course of action in treating comorbid reading difficulties and anxiety.

Limitations and Future Directions

The present findings must be interpreted in light of their limitations.

Measurement and Sample

In terms of measurement, we only used one measure of anxiety, the RCMAS. As discussed previously, given that the patterning of age-related differences in scores on the RCMAS is counter to typical developmental trends, this measure may not be ideally suited for drawing conclusions about age-related differences. Additionally, given that there is a high degree of overlap between the RCMAS and CDI, we were unable to disentangle anxiety and depression symptomatology in our analyses (see Supplemental Analyses for details). As such, future research should replicate the present findings using other anxiety measures. In addition, the present analysis focused only on global/total anxiety symptoms. However, there is some evidence to suggest that reading difficulties may be more likely to be comorbid with certain subtypes of anxiety (Francis et al., 2019; Mugnaini et al., 2009). Further, research examining anxiety that is restricted to the academic domain, such as reading-specific anxiety and test anxiety, and their potential generalization to other domains over time will be critical for understanding the reading-anxiety comorbidity (Nelson et al., 2015; Ramirez et al., 2019).

Next, the current study is also limited by the reading and cognitive measures used. First, we only included measures of single word reading in this study. It will be important for future research to assess other aspects of reading, such as reading comprehension, and their associations with anxiety. Additionally, this sample only included performance-based measures of executive functioning, which are not very highly correlated with caregiver- and teacher-rating scales and therefore may not be the best indicator of real-world executive skills (Toplak et al., 2013). Finally, given that this study utilized data from a longstanding research project that has emphasized continuity in measures over time, the CLDRC, some constructs included tasks from older measures, including the WISC-III, WISC-R, and PIAT. Though both processing speed and working memory included tasks from the WISC-III or WISC-R (i.e., Coding and Symbol Search for processing speed and Digit Span Backwards for working memory), in both cases these tasks were included along with several other measures of the same construct. Additionally, these specific measures (i.e., Coding, Symbol Search, and Digit Span Backwards) have changed relatively little over the various iterations of the WISC. Similarly, though two subtests from the PIAT are included in the reading composite, they are measures of single word reading and spelling where we might expect reasonable stability in item difficulty over time. Finally, the main analyses in this paper used raw scores from these older measures, not standard scores, which reduces the impact of the older norming procedures.

Another limitation of our current study is the relative homogeneity of the sample. The CLDRC sample is a population-based sample that resembles the front range of Colorado with respect to education levels and racial/ethnic diversity. However, the sample has less economic and racial/ethnic diversity than the broader United States population. As such, further research is needed to better understand how these findings would apply to children of diverse racial, ethnic, and socioeconomic groups from other geographic locations.

Modelling

We elected to use a bifactor model to index common EF, in accordance with the unity and diversity model of EF from Miyake et al. (2000). However, there have been recent theoretical and empirical concerns for the use of bifactor models generally, including that they may overfit data and that there are statistical irregularities with the group-specific factors (Bonifay et al., 2017; Eid et al., 2017). Nonetheless, we suggest that our results focused on the common EF factor are less susceptible to statistical critiques of the bifactor model.

Future Directions

While we speculate as to the nature of the relation between reading, anxiety and cognitive factors, the cross-sectional nature of our study precludes us from inferring causality from findings. As such, future longitudinal research is necessary to characterize the causal and temporal associations between reading, cognitive factors, and anxiety over time. Additionally, such research would be useful in better understanding whether the cognitive correlates of anxiety that we have are stable cognitive weaknesses over time or whether strategies/task approaches influence scores on these cognitive measures. Finally, given the modest effect sizes yielded by the present analysis, further research is needed to improve prediction of which children with reading difficulties are at risk for comorbid anxiety. Future analyses should explore other levels of analysis beyond the cognitive level, such as environmental factors like peer rejection and victimization (La Greca & Landoll, 2011).

Conclusion

In conclusion, the present study found that weaknesses in domain-general cognitive abilities (i.e., PS and common EF) was modestly associated with higher anxiety symptoms in youth with word reading difficulties compared to youth without word reading difficulties. These findings contribute to a growing literature on the comorbidity between word reading difficulties and anxiety and present the first attempt to understand what cognitive factors may contribute to this comorbidity. Results from this study can inform future research efforts to develop screening tools and intervention approaches for children with comorbid anxiety and word reading difficulties.

Availability of data and material:

Not applicable.