In daily life, most of us depend on our global executive functioning (EF) capacity to effectively accomplish tasks, communicate, handle emotions, make choices, prioritize goals, and solve problems [Reference Nigg1,Reference McClelland and Cameron2]. Global EF is defined as a group of multidomain cognitive control systems entwined with attention, information processing, and other cognitive abilities [Reference Karr, Areshenkoff, Rast, Hofer, Iverson and Garcia-Barrera3,Reference Baggetta and Alexander4]. Our global EF systems comprise facets of inhibition (capacity to abstain from autopilot actions), working memory (WM; ability to alter cognitive representations with incoming data in real-time), shifting (adeptness to flexibly switch from one mental set to another) [Reference Friedman, Miyake, Young, DeFries, Corley and Hewitt5], and verbal fluency [Reference Aita, Beach, Taylor, Borgogna, Harrell and Hill6]. Relatedly, evidence has shown consistently that language-based, temporal lobe-mediated verbal fluency ability (marked by scores on diverse time-limited word generation on animal- and phonemic-cued tests) had strong and unique relations with common EF variance (i.e., global EF capacity) in diverse youth and adult samples [Reference Gustavson, Panizzon, Franz, Reynolds, Corley and Hewitt7–Reference Gustavson, Elman, Panizzon, Franz, Zuber and Sanderson-Cimino10]. Given its importance, EF problems have been linked to issues with career, social relationships, diet, nutrition, and health [Reference Favieri, Forte and Casagrande11,Reference Pineda-Alhucema, Aristizabal, Escudero-Cabarcas, Acosta-López and Vélez12]. Executive dysfunction-related health problems include cardiorespiratory, metabolic, neuroendocrine, and psychiatric disorders [Reference Berryman, Stanton, Bowering, Tabor, McFarlane and Moseley13,Reference Hofmann, Schmeichel and Baddeley14]. Thus, understanding the risk factors and consequences of EF decrements is essential.

Scar theories propose that increases in psychiatric symptoms can precede and predict future EF decline. Specifically, scar models posit that chronic increased depression and anxiety may build up oxidative and inflammatory-stress, thereby adversely impacting EF-related brain regions over protracted durations [Reference Gałecki, Talarowska, Anderson, Berk and Maes15–Reference Copeland, Shanahan, Worthman, Angold and Costello17]. Relatedly, scar models such as the vascular- [Reference Alexopoulos, Meyers, Young, Campbell, Silbersweig and Charlson18] and executive dysfunction syndrome-depression [Reference Alexopoulos19] hypotheses assert that increased depression and anxiety could impair future EF via buildup of tissue injury (e.g., lacunes, microinfarcts, and white matter hyperintensities) in cardiovascular systems, cognitive control-, and reward processing-related brain regions, over long timescales [Reference Alexopoulos20]. These brain areas might include frontal–striatal pathways (e.g., dorsolateral prefrontal cortex, basal ganglia, thalamus, and anterior cingulate cortex) [Reference Alexopoulos, Hoptman, Yuen, Kanellopoulos, Seirup and Lim21,Reference Alexopoulos, Hoptman, Kanellopoulos, Murphy, Lim and Gunning22].

Thus far, 47 longitudinal studies have offered support for scar models. For instance, higher depression severity during adolescence was associated with lower vocabulary score in early adulthood 8 years later [Reference Beaujean, Parker and Qiu23]; however, whether such pattern applied to various stages in adulthood could not be inferred from that study. Other studies suggested such a possibility. Swedish and American adults with (vs. without) major depression displayed worsened episodic memory, EF, or verbal fluency after 6 months to 5 years despite symptom remission [Reference Hammar, Sørensen, Årdal, Oedegaard, Kroken and Roness24,Reference Wilson, Mendes de Leon, Bennett, Bienias and Evans25]. Likewise, among mid-life and older community adults, increased anxiety was related to reduced immediate and delayed auditory memory abilities following 12 years [Reference Gulpers, Oude Voshaar, van Boxtel, Verhey and Köhler26]. Similarly, 2 meta-analyses of 43 studies showed that heightened anxiety and depression dovetailed with larger EF decline and incidence of major neurocognitive disorders in diverse community and clinical samples across 1–17 years [Reference Gulpers, Ramakers, Hamel, Köhler, Oude Voshaar and Verhey27,Reference Diniz, Butters, Albert, Dew and Reynolds28].

Simultaneously, vulnerability models argue that EF decline can function as a precursor of later heightened depression and anxiety. Vulnerability models assert that poorer EF may forecast future anxiety and depression across prolonged periods due to chronic problems with disengaging from negative self-referential perseverative thinking (e.g., worry and rumination) [Reference Koster, De Lissnyder, Derakshan and De Raedt29,Reference Aldao, Nolen-Hoeksema and Schweizer30]. Likewise, EF deficits can make it perpetually hard to detach from threats, leading to excessive focus on anxiety-inducing factors in one’s surroundings and risk for increased anxiety [Reference Cabrera, Brugos and Montorio31,Reference Barry, Vervliet and Hermans32]. Moreover, it has been thought that poorer EF, especially WM, can predict increased depression and anxiety across long durations, in part due to difficulties with adjusting to various changing emotion-eliciting contexts in versatile and optimal ways [Reference Hantke, Gyurak, Van Moorleghem, Waring, Adamson and O’Hara33]. In sum, vulnerability theories argue that worse EF may forecast increased depression and anxiety over long durations.

To date, 31 prospective investigations have empirically supported vulnerability theories. For example, an earlier study demonstrated that poorer WM was related to future chronic course of increased depression [Reference Harvey, Le Bastard, Pochon, Levy, Allilaire and Dubois34]. Likewise, reduced inhibition, WM, shifting, verbal fluency, and other cognitive functioning indices were connected with pathological worry dimensionally and categorically 9 years later in community adults [Reference Zainal and Newman35]. More recently, meta-analytic data on 29 studies (n = 121,749) showed that cognitive deficits were associated with increased major depression severity following several months to 45 years in diverse clinical and community-dwelling samples [Reference Scult, Paulli, Mazure, Moffitt, Hariri and Strauman36].

However, the mostly two-time-point, between-person, regression studies testing the prospective relations between mental health symptoms and global EF to date introduce shortcomings to clinical science. Such methods do not account for the nesting of repeated assessments within persons to capture change-to-future change trajectories across time [Reference Molenaar, Huizenga, Nesselroade, Staudinger and Lindenberger37]. Mounting global pressures related to neuropsychiatric illnesses, increasing life expectancy, and aging [Reference Lang, Clifford, Wei, Zhang, Leung and Augustine38–Reference Cohen, Edmondson and Kronish40] make it crucial to explore whether change in global EF over long durations may be related to future change in mental illness during adulthood development. Further, between-person differences across time may be due to stable variations observed across the lifespan [Reference Gustavson, Panizzon, Elman, Franz, Reynolds and Jacobson41], or to individual differences in aging-associated rate of EF decrements [Reference Braun, Schmukle and Kunzmann42–Reference Yu, Yang, Geldhof, Tyler, Gansert and Chase45]. The latter possibility can only be captured by using within-person methods that also capture change. Moreover, the foregoing scar and vulnerability theories posit that EF-symptom relations unfold within persons across long durations [Reference Hülür, Ram, Willis, Schaie and Gerstorf46–Reference Allerhand, Gale and Deary51]. Awareness of within-person prolonged trajectories of increased depression or anxiety, EF decrements, and their covariation may guide the design of personalized prevention, diagnostic, and treatment efforts that rely on idiographic (or within-person) more than between-person data, as part of precision psychiatry [Reference Lenze, Nicol, Barbour, Kannampallil, Wong and Piccirillo52–Reference Zuidersma, Riese, Snippe, Booij, Wichers and Bos54]. It is also important to note that observations of between- and within-person differences in EF and psychopathological symptoms do not always align with each other [Reference Moore, Campbell, Delgadillo, Paolillo, Sundermann and Holden55–Reference Zainal and Newman57]. To broaden and deepen comprehension of EF and mental health in mid-life and older adulthood, within-person (co)variations and change must be considered. Tethering within-person data analytic approaches with longitudinal study designs is thus important to comprehend the bidirectional within-person changes in EF and subsequent changes in symptoms (and vice versa).

Two cutting-edge techniques that attain these aims are random-intercept cross-lagged panel models (RI-CLPM) [Reference Hamaker, Kuiper and Grasman58] and bivariate dual latent change score models (BLCS) [Reference Grimm and Ram59]; two forms of longitudinal structural equation modeling (SEM). These longitudinal SEM approaches benefit researchers by accounting for prior lagged relations and regression to the mean, minimizing measurement unreliability, and using all available values instead of listwise deletion [Reference Zainal and Newman57]. Further, by adjusting for temporally stable between-person differences and autoregressive effects, these models can test if change in one variable across a previous time-period or time-lag is associated with change in another variable at the next time-period or time-lag within persons. Accordingly, by evaluating lead–lag change-to-future change connections, RI-CLPM and BLCS models move us toward the ability to draw causal inferences [Reference Mund and Nestler60]; inquiries essential to clinical science.

Thus far, three studies of adult participants have tested the longitudinal, dynamic, within-person relations between EF and anxiety, depression, or pertinent concepts with BLCS. Using BLCS, increase in anxiety was related to cognitive functioning decline in older adults [Reference Zainal and Newman57]; despite that, the two-wave study prevented understanding of how symptom change predicted subsequent EF change (and conversely). Relatedly, BLCS models across three waves showed that 9-year growth in excessive worry dovetailed with future 9-year decline in global and unique EF facets [Reference Zainal and Newman49]; however, whether change in EF forecasted later change in worry was not examined. Another five time-point study demonstrated that rise in trait neuroticism at one time-lag preceded and linked to reductions in spatial processing, WM, and processing speed at the next time-lag [Reference Zainal and Newman57]; nonetheless, one-item assessments of cognitive functioning were used in the study. To our knowledge, no studies have tested EF-psychological symptom relations in older adults with RI-CLPM. However, a recent study in youths that utilized RI-CLPM [Reference Oh, Greenberg, Willoughby, Vernon-Feagans, Greenberg and Blair61] suggested the possibility of EF problems serving as risk factors for later increased depression and anxiety.

Building on this literature, this study aimed to examine the within-person associations between a global EF composite (formed via a latent composite of five measures) and depression or anxiety severity using RI-CLPM and BLCS in older adults. Based on scar theories, we hypothesized that within persons, higher anxiety or depression severity would reliably precede and relate to greater future EF decline at the next time-point and time-lag. Moreover, based on vulnerability models, we hypothesized that within persons, lower EF would forecast subsequent rise or increased depression or anxiety severity at the next time-point or time-lag. Last, using a SEM-based model comparison approach [Reference Widaman, Ferrer and Conger62], we aimed to directly juxtapose the effect sizes indicating the scar (vs. vulnerability) hypothesis to determine if any differences emerged.

Method

Participants

The present study was a secondary analysis of the Aging, Demographics, and Memory Study (ADAMS) publicly available and restricted-use datasets [Reference Steven, Gwenith, Michael, Kenneth, Mary Beth and Brenda63]. Ethical approval was provided by the University of Michigan and Duke University Medical Center, and all participants voluntarily consented to enroll. Participants (n = 856) averaged 81.59 years of age (SD = 7.10, range = 70–110), 58.53% were female, and 76.87% identified as White, compared to the other 23.13% who identified African American or other ethnicities. In addition, 28.62% (n = 245) of the participants needed support for dressing, feeding, or bathing based on caregiver report, or were diagnosed with Diagnostic and Statistical Manual-Fourth Edition-Text Revised (DSM-IV-TR) [64,65]—defined dementia, major depressive disorder, stroke, or other neurological condition. These dementia syndromes included probable and possible Alzheimer’s disease, cardiovascular, and other causes. All DSM-IV-TR diagnoses were attained through expert consensus with a multidisciplinary team of neurologists, psychiatrists, geriatric psychologists, and other healthcare professionals [Reference Langa, Plassman, Wallace, Herzog, Heeringa and Ofstedal66]. The online supplementary material (OSM) offers more details on the sample characteristics.

Procedures

Participants completed performance-based EF measures and had a significant other caregiver (e.g., spouse and children) who could reliably report on their behavioral symptoms across multiple time-points. Data were collected across four waves in 2004 (Time 1; T1), 2006 (Time 2; T2), 2008 (Time 3; T3), and 2010 (Time 4; T4) [Reference Langa, Plassman, Wallace, Herzog, Heeringa and Ofstedal66]. The following caregiver-rated symptom assessment and EF tests were administered.

Measures

Mental health symptoms

The widely used caregiver-rated Neuropsychiatric Inventory (NPI) [Reference Cummings, Mega, Gray, Rosenberg-Thompson, Carusi and Gornbein67]—depression and anxiety domains were utilized to assess past-month depression and anxiety severity in the form of a structured interview. Caregiver ratings were used in this study as self-reported mental health symptom severity data was only available at T1 and T2 [Reference Steven, Gwenith, Michael, Kenneth, Mary Beth and Brenda63,Reference Mehta, Stewart, Langa, Yaffe, Moody-Ayers and Williams68,Reference Fisher, Franks, Plassman, Brown, Potter and Llewellyn69], and due to the reliable nature of caregiver-reported data that tends to align with self-rated symptom measures [Reference Steffens, Fisher, Langa, Potter and Plassman70]. To measure depression, caregivers were asked about the presence and duration of depression symptoms (e.g., sadness, irritability, feels worthless, and suicidal thoughts) the participant may have exhibited. To assess for anxiety, caregivers were inquired on the presence of any anxiety symptoms (e.g., excessive worry, breathlessness, and behavioral avoidance). Also, for each participant’s symptom domain, the informants reported on the degree of the following four facets: severity (3-point scale; 1 = mild to 3 = marked); change from past typical behaviors (3-point scale; 0 = no; 1 = yes; 2 = exaggeration of previous problems); distress (6-point scale; 0 = not at all to 5 = very severely or extremely). Supplementary Tables S1 and S2 in the OSM show that these four manifest indicators for the depression and anxiety scales had excellent model fit using a series of confirmatory factor analysis (CFA) at distinct time-points. Further, the NPI has reliably shown strong internal consistency, as well as convergent and discriminant validity [Reference Cummings, Mega, Gray, Rosenberg-Thompson, Carusi and Gornbein67,Reference de Medeiros, Robert, Gauthier, Stella, Politis and Leoutsakos71]. In this study, internal consistencies were high for the depression severity (Cronbach’s αs = 0.93, 0.94, 0.92, and 0.93 at T1, T2, T3, and T4, respectively) and anxiety severity scales (αs = 0.93, 0.96, 0.93, and 0.90 at T1, T2, T3, and T4, respectively).

Executive functioning

The following five measures of EF were used to create a composite latent global EF composite: (a) controlled oral word association (a verbal fluency assessment that captures unplanned generation of words within a time limit that start with some assigned letter) [Reference Borkowski, Benton and Spreen72]; (b) animal fluency (another time-limited verbal fluency test based on the animal category) [Reference Ardila, Ostrosky‐Solís and Bernal73]; (c) serial 7 subtraction (extent of accuracy of counting down from 100 by 7 within a time limit) [Reference Manning74]; (d) backward digit span (degree of accuracy of recall in reverse order of integer strings of increasing length) [Reference Wechsler75]; and (e) symbol digit modality test (level of accuracy of replacing a single-digit integer for randomized displays of geometric patterns) [Reference Smith76]. These EF assessments have been shown to have good internal consistency, strong convergent, and discriminant validity [Reference Oliveira, Mograbi, Gabrig and Charchat-Fichman77–Reference Hinton-Bayre and Geffen79]. In this study, the αs for the global EF composite were strong across all time-points (αs = 0.92, 0.83, 0.87, and 0.88 at T1, T2, T3, and T4, respectively). Moreover, at each time-point, a composite global EF index was created by standardizing each EF measure and averaging the scores across EF measures. Further, exploratory factor analysis and a series of CFA demonstrated that a one-factor latent global EF composite had good model fit across waves of assessment (refer to page 4 and Supplementary Table S3 of the OSM). In addition, these global EF scores have been normed based on age and education, and appropriate adjustments were made for participants with hearing impairments [Reference Fong, Fearing, Jones, Shi, Marcantonio and Rudolph80–Reference Welsh, Breitner and Magruder-Habib83]. Also, Supplementary Table S4 in the OSM shows the descriptive statistics of the study variables based on SEM analyses.

Data analyses

All longitudinal SEM analyses were performed with the lavaan package [Reference Rosseel84] in R Version 3.6.3. Model fit was assessed utilizing practical fit indices and heuristic cut-offs: confirmatory fit index (CFI; CFI ≥ 0.90) [Reference Bentler85] and root mean square error of approximation (RMSEA; RMSEA ≤ 0.09) [Reference Steiger86]. To maximize all available data points, we used full information maximum likelihood, the gold standard [Reference Graham87], to manage missing data. In total, 16.29% of the data were missing. Further, the data were missing completely at random (χ²[df = 113] = 134.32, p = 0.084).

Next, we established longitudinal measurement invariance; a prerequisite for longitudinal SEM [Reference McArdle88]. We progressively evaluated for configural invariance (equivalence of factor structure), metric invariance (equal factor structure and item loadings [λs], freely estimated item intercepts [τs], and item error variances [εs] across the time-points), scalar invariance (equal factor structure, λs, and τs, but freely estimated εs across time-points), and strict invariance (equal factor structure, λs, τs, and εs, across time-points) [Reference Widaman, Ferrer and Conger62]. To test for measurement invariance, we conducted a Δχ² difference test. A statistically significant Δχ² meant that the more (vs. less) restricted model had poorer fit [Reference Cheung and Rensvold89]. However, as Δχ² is affected by sample size despite negligible misfit changes, the following change in practical fit indices, ΔCFI ≤ −0.01 or ΔRMSEA < +0.015 [Reference Chen90], from the less restricted to more restricted models signaled measurement nonequivalence.

The RI-CLPM was used to manage interdependent repeated assessments nested within persons, and to distinguish between within-person (dynamic state) variance and between-person (trait) variance [Reference Mulder and Hamaker91]. RI-CLPM procedures permitted us to test these within-person reciprocal cross-lagged relations (γs) accounting for within-person autoregressive effects (βs; level of one variable forecasting its subsequent level), trait variances (αs), and regression to the mean [Reference Falkenström, Finkel, Sandell, Rubel and Holmqvist92,Reference Usami93]. Of primary interest were the within-person cross-lagged associations between level of EF at a prior time-point (T − 1) and level of depression or anxiety symptom severity (SYM) at the next adjacent time-point (T) following about 2 years (γs) (and vice versa), as shown in Equations (1) and (2).

(1) $$ \unicode{x03BC} {\left(\mathrm{EF}\right)}_{\left[ T\right]}={\unicode{x03B1}}_{\mathrm{Y}}\times {\left(\mathrm{EF}\right)}_{\left[ T\hbox{-} 1\right]}+{\unicode{x03B2}}_{\mathrm{Y}}\times {\left(\mathrm{EF}\right)}_{\left[ T\hbox{-} 1\right]}+{\unicode{x03B3}}_{\mathrm{X}}\times {\left(\mathrm{SYM}\right)}_{\left[ T\hbox{-} 1\right]} $$

(2) $$ \unicode{x03BC} {\left(\mathrm{SYM}\right)}_{\left[ T\right]}={\unicode{x03B1}}_{\mathrm{X}}\times {\left(\mathrm{SYM}\right)}_{\left[ T\hbox{-} 1\right]}+{\unicode{x03B2}}_{\mathrm{X}}\times {\left(\mathrm{SYM}\right)}_{\left[ T\hbox{-} 1\right]}+{\unicode{x03B3}}_{\mathrm{Y}}\times {\left(\mathrm{EF}\right)}_{\left[ T\hbox{-} 1\right]} $$

Concurrently, BLCS approaches were utilized to test if within-person change in depression or anxiety symptom severity at a previous time-lag (ΔT − 1) would be related to change in EF at the next successive time-lag (ΔT) (and vice versa). BLCS is a cutting-edge method that empowers researchers to test within-person change-to-future change associations (coupling effects; δs) after accounting for trait-level initial status, trait-level constant change parameters (αs), and within-person autoregressive paths (proportional effects; change in a variable predicting subsequent change in itself; βs) [Reference Grimm and Ram59]. The BLCS models relevant to our research question can be denoted in Equations (3) and (4) as follows.

(3) $$ \Delta {\left(\mathrm{EF}\right)}_{\left[\varDelta T\right]}={\unicode{x03B1}}_{\mathrm{Y}}\times {\left(\mathrm{EF}\right)}_{\left[\varDelta T- 1\right]}+{\unicode{x03B2}}_{\mathrm{Y}}\times {\left(\mathrm{EF}\right)}_{\left[\varDelta T- 1\right]}+{\unicode{x03B4}}_{\mathrm{X}}\times {\left(\mathrm{SYM}\right)}_{\left[\varDelta T- 1\right]} $$

(4) $$ \Delta {\left(\mathrm{SYM}\right)}_{\left[\varDelta T\right]}={\unicode{x03B1}}_{\mathrm{X}}\times {\left(\mathrm{SYM}\right)}_{\left[\varDelta T\hbox{-} 1\right]}+{\unicode{x03B2}}_{\mathrm{X}}\times {\left(\mathrm{SYM}\right)}_{\left[\varDelta T\hbox{-} 1\right]}+{\unicode{x03B4}}_{\mathrm{Y}}\times {\left(\mathrm{EF}\right)}_{\left[\varDelta T\hbox{-} 1\right]} $$

As recommended, the within-person cross-lagged associations (γs) in the RI-CLPM and within-person coupling effects in the BLCS (δs) were constrained to be equal across waves of assessments to reduce SEs in parameter estimates (refer to Supplementary Figure S1 shows a BLCS model in [Reference McArdle94]). Also, baseline psychopathology and EF were controlled for in all models.

As we aimed to directly compare the scar and vulnerability hypotheses, we contrasted a model that freely estimated the cross-lagged or coupling effects (EF predicting future SYM and conversely) to a model that constrained the cross-lagged or coupling effects to equality. A statistically significant change (Δ) in χ² value in comparing these two models would indicate notable differences in the strength of effect sizes for one hypothesis versus the other [Reference Widaman, Ferrer and Conger62]. If the Δχ² test was not significant, the more parsimonious model with equality constraints on cross-lagged or coupling effects was chosen as the final model. Effect sizes were calculated using the formula, Cohen’s $ d=\frac{\beta }{SD\left(\beta \right)}\times \sqrt{\left(\frac{2}{N}\right)} $ [Reference Dunlap, Cortina, Vaslow and Burke95], where β is the unstandardized regression estimate, SD(β) its standard deviation, and N is the sample size. Cohen’s d values of 0.2, 0.5, and 0.8 indicated small, moderate, and large effects, respectively.

Power analysis

Following best practices [Reference Arend and Schafer96], an a priori Monte Carlo power analysis based on a conservative effect size of d = 0.20 for the cross-lagged effects in the RI-CLPM and coupling effects (bidirectional change-to-future change EF-symptom relations) in the BLCS was performed using the RAMpath R package [Reference Zhang, Hamagami, Grimm and McArdle97]. After 1,000 replications per condition, we observed 90.6–100.0% power to detect significant within-person cross-lagged or coupling effects. Further, there was 90.0–100.0% power to identify other significant parameter estimates.

Results

Longitudinal measurement invariance

Supplementary Table S5 in the OSM shows the longitudinal measurement invariance analyses for the constructs of interest. Overall, strict level of equivalence (equal λs, τs, εs) was observed for the constructs of depression severity, anxiety severity, and EF. Therefore, conducting analyses using RI-CLPM and BLCS approaches were appropriate.

Lagged relations between depression severity and executive function

Table 1 displays all of the parameter estimates for the RI-CLPM testing the cross-lagged relations between depression severity and EF. The model with equality constraints on the cross-lagged effects did not significantly differ from the model that freely estimated those parameters (Δχ²[df = 1] = 0.017, p = 0.895). The parsimonious model with equality constraints showed good model fit (χ²[df = 24] = 45.160, p = 0.006, CFI = 0.984, RMSEA = 0.032). Within persons, higher prior depression severity substantially predicted lower EF at the subsequent time-point (β = −0.073, 95% CI [−0.119, −0.026], d = −0.292). Likewise, lower previous EF significantly predicted greater depression severity at the next time-point within persons (β = −0.073, 95% CI [−0.119, −0.026], d = −0.292). Also, between persons, higher random intercept depression severity was significantly correlated with lower random intercept EF (β = −0.055, 95% CI [−0.094, −0.016], d = −0.264).

Table 1. Random-intercepts cross-lagged panel model of DEP and EF across four time-points.

Note: Model fit indices: χ²(df = 24) = 45.160, p = 0.006, CFI = 0.984, RMSEA = 0.032, 95% CI [0.017, 0.046]. Within-person cross-lagged effects refer to level in DEP at a prior time-point (T-1) predicting (➔) future Δ in EF at the next adjacent time-point (T) (and vice versa). Within-person coupling effects and proportional effects, residual covariances between DEP and EF, as well as variances of DEP and EF were each uniquely fixed to be equal across all three time-lags.

Abbreviations: CI, confidence interval; DEP, depression severity; EF, executive functioning; RI, random intercept.

* p < 0.05.

** p < 0.01.

*** p < 0.001.

Table 2 presents the parameter estimates for the BLCS models examining the change-to-future change associations between depression severity and EF. The freely estimated (vs. constrained) models were not significantly different from each other (Δχ²[df = 1] = 0.235, p = 0.628). The final model with equality constraints on the coupling effects showed acceptable model fit (χ²[df = 25] = 47.000, p = 0.005, CFI = 0.974, RMSEA = 0.039, 95% CI [0.021, 0.057]). Within persons, greater growth in depression severity at a prior time-lag significantly predicted EF decrement at the next time-lag (β = −0.540, 95% CI [−0.955, −0.124], d = −0.245). Likewise, within persons, EF decline at a previous time-lag was significantly associated with larger increase in depression severity at the subsequent time-lag (β = −0.540, 95% CI [−0.955, −0.124], d = −0.245). Figures 1 and 2 summarize the analyses of the lagged relations between depression severity and EF.

Table 2. Bivariate dual latent change score model of DEP and EF across four time-points.

Note: Model fit indices: χ ²(df = 25) = 47.000, p = 0.005, CFI = 0.974, RMSEA = 0.039, 95% CI [0.021, 0.057]. Within-person coupling effects refer to change (Δ) in DEP at a prior time-lag (ΔT-1) predicting (➔) future Δ in EF at the next adjacent time-lag (ΔT; and vice versa). Within-person coupling effects and proportional effects, residual covariances between DEP and EF, as well as variances of DEP and EF were each uniquely fixed to be equal across all three time-lags.

Abbreviations: CI, confidence interval; DEP, depression severity; EF, executive functioning.

* p < 0.05;

** p < 0.01;

*** p < 0.001.

Figure 1. Random-Intercept Cross-Lagged Panel Models Between EF and Depression Severity.

Note. **p < .01; ***p < .001. Δ = within-person change in construct from a time-lag to the next adjacent time-lag; DEP = depression severity; EF = executive functioning.

Figure 2. Bivariate Dual Latent Change Score Models Between EF and Depression Severity.

Note. **p < .01; ***p < .001. Δ = within-person change in construct from a time-lag to the next adjacent time-lag; DEP = depression severity; EF = executive functioning.

Lagged relations between anxiety severity and executive function

Table 3 shows the model parameter estimates for the RI-CLPM evaluating the cross-lagged relations between anxiety severity and EF. The freely estimated model was not significantly different from the constrained model (Δχ²[df = 1] = 0.069, p = 0.792). The final model with equality constraints on the cross-lagged effects demonstrated good model fit (χ²[df = 23] = 86.84, p < 0.001, CFI = 0.952, RMSEA = 0.057). Within persons, no cross-lagged relations were observed between prior anxiety severity and EF at the subsequent time-point (β = −0.025, 95% CI [−0.101, 0.051], d = −0.062). Likewise, no within-person cross-lagged relations were found between previous EF and anxiety severity at the next time-point (β = −0.025, 95% CI [−0.101, 0.051], d = −0.062). However, between persons, higher random intercept anxiety severity was significantly related to lower random intercept EF (β = −0.070, 95% CI [−0.104, −0.036], d = −0.395).

Table 3. Random-intercepts cross-lagged panel model of ANX and EF across four time-points.

Note: Model fit indices: χ²(df = 24) = 33.102, p = 0.102, CFI = 0.982, RMSEA = 0.021, 95% CI [0.008, 0.031]. Within-person cross-lagged effects refer to level in ANX at a prior time-point (T-1) predicting (➔) future Δ in EF at the next adjacent time-point (T) (and vice versa). Within-person coupling effects and proportional effects, residual covariances between ANX and EF, as well as variances of ANX and EF were each uniquely fixed to be equal across all three time-lags.

Abbreviations: ANX, anxiety severity; CI, confidence interval; EF, executive functioning; RI, random intercept.

* p < 0.05.

** p < 0.01.

*** p < 0.001.

Table 4 shows the parameter estimates for the BLCS models testing the change-to-future change relations between anxiety severity and EF. The constrained (vs. freely estimated) models were not significantly different (Δχ²[df = 1] = 0.005, p = 0.943). The final model with equality constraints on the coupling parameters showed acceptable model fit (χ²[df = 25] = 46.996, p < 0.001, CFI = 0.966, RMSEA = 0.057). Within persons, prior change in anxiety severity at a previous time-lag was not significantly associated with change in EF at the subsequent time-lag (β = −0.254, 95% CI [−0.951, 0.444], d = −0.068) and vice versa (β = −0.254, 95% CI [−0.951, 0.444], d = −0.068).Footnote ¹

Table 4. Bivariate dual latent change score model of ANX and EF across four time-points.

Note: Model fit indices: χ²(df = 25) = 46.996, p < 0.001, CFI = 0.966, RMSEA = 0.057, 95% CI [0.044, 0.071]. Within-person coupling effects refer to change (Δ) in ANX at a prior time-lag (ΔT-1) predicting (➔) future Δ in EF at the next adjacent time-lag (ΔT; and vice versa). Within-person coupling effects and proportional effects, residual covariances between ANX and EF, as well as variances of ANX and EF were each uniquely fixed to be equal across all three time-lags.

Abbreviations: ANX, anxiety severity; CI, confidence interval; EF, executive functioning.

* p < 0.05.

** p < 0.01.

*** p < 0.001.