Transactional Relations between Attentional Biases for Affective Stimuli and Depressive Symptoms in Offspring of Mothers with and without Major Depressive Disorder

Abstract

Attentional biases toward depression-relevant stimuli (e.g., sad faces) are thought to contribute to the development and maintenance of depression, but research in youth is limited and primarily cross-sectional. Therefore, it remains unclear whether such biases are risk factors for, versus consequences of, depression (or both), and whether these links may be stronger among youth already at risk for depression due to a family history of the disorder. To address these questions, the present multi-wave longitudinal study followed 242 mothers and their offspring (ages 8–14; 51.65% girls, 81.40% non-Hispanic White) over two years with assessments every six months (5 assessments total) to examine prospective relations between attentional biases (measured via a dot probe eye-tracking task) and depressive symptoms in youth. Of the mothers, 123 had a history of major depressive disorder (MDD) during their child’s life and the rest had no lifetime history of MDD. Using random intercept cross-lagged panel models, we found that depressive symptoms predicted prospective changes in attentional biases, but not the reverse, and the pattern of attentional bias change differed based on maternal history of MDD. Specifically, among offspring of mothers with MDD, higher depressive symptoms at a given assessment predicted future increases in attentional bias to sad faces, whereas among offspring of never depressed mothers, higher depressive symptoms predicted subsequent decreases in attentional bias to happy faces. These findings suggest that experiences of depression may contribute to different patterns of attention to emotional cues based on maternal history of MDD.

General Scientific Summary:

This multi-wave longitudinal study examines links between children’s attention to facial displays of emotion and their depressive symptoms, and whether these links differ by maternal history of major depressive disorder (MDD). Depressive symptoms predicted increases in attention to sad faces among offspring of mothers with MDD and decreases in attention to happy faces among offspring of never depressed mothers, suggesting the potential for unique pathways of risk.

Theorists have proposed that information processing biases, including biases in attention, interpretation, and memory, contribute to the onset and maintenance of depression (Clark et al., 1999; Gotlib & Joormann, 2010). Attentional biases refer to systematic tendencies to preferentially attend to certain types of emotional information (Mennen et al., 2019) and are thought to emerge during development and confer risk across the lifespan. According to cognitive models of depression (Clark et al., 1999; Gotlib & Joormann, 2010), depressed and at-risk individuals are hypothesized to exhibit increased attention toward, or difficulty disengaging attention from, depression-relevant stimuli (e.g., sad faces). Supporting this hypothesis, studies have shown that depressed adults exhibit higher levels of sustained attention to sad faces, with some evidence of reduced attention to positive stimuli, such as happy faces (for reviews, see Armstrong & Olatunji, 2012; Huang et al., 2023; Peckham et al., 2010; Suslow et al., 2020).

Far fewer studies have examined links between attentional biases and depression in youth, and existing findings have been mixed. Some studies report greater attention toward sad faces among clinically depressed or symptomatic youth (Buhl et al., 2023; Hankin et al., 2010) and in community samples (Platt et al., 2015), whereas others find attentional avoidance of sad faces, particularly among currently depressed children and offspring of mothers with a history of major depressive disorder (MDD) (e.g., Gibb et al., 2009; Harrison & Gibb, 2015). Notably, a recent longitudinal meta-analysis found attentional biases showed small and often inconsistent prospective associations with later internalizing symptoms (Vos et al., 2025), suggesting that their predictive utility may vary across developmental or contextual factors. Together, these findings suggest that attentional biases may not function as uniform vulnerability markers across youth, especially among youth already at risk for depression due to a maternal history of the disorder.

Currently, therefore, it remains unclear whether these attentional biases reflect a risk factor for depression, emerge as a consequence of prior depression, or a combination of both. Evidence in adults suggests that attentional biases increase risk for depression (e.g., Clasen et al., 2013; Beevers et al., 2011). One study has also found that attentional biases to sad faces predict future depressive symptom increases in adolescents (Feurer et al., 2020), raising the question of whether depression may also shape attentional processing over time in youth. According to the scar hypothesis, experiences with depression can leave lasting cognitive changes, which persist even after the depression remits (Lewinsohn et al., 1981). Alternative models propose dynamic, bidirectional relations between mood and cognition (e.g., Teasdale, 1988), raising the possibility that depressive symptoms may prospectively shape and be shaped by attentional processing. Consistent with this possibility, youth studies show transactional relations between other forms of cognitive vulnerabilities and depressive symptoms over time (e.g., Calvete et al., 2013; Gibb et al., 2006; Kopala-Sibley et al., 2015; Shankar & Gibb, 2024). These findings underscore the need for developmental research clarifying the temporal relations between attentional biases and depression in youth.

The primary goal of this study, therefore, was to examine prospective relations between youths’ attentional biases and depressive symptoms across a multi-wave longitudinal study with assessments every six months for two years (five assessments total). At each assessment, youth completed a modified dot probe task (cf. MacLeod et al., 1986) which assessed their attention to emotional, compared to neutral, stimuli. Although attentional biases in the dot probe task have traditionally been indexed via reaction time difference scores, these indices exhibit unacceptably low reliability and limited interpretability (see, e.g., Gibb, Beevers, & McGeary, 2016; Price et al., 2015; Rodebaugh et al., 2016). We therefore focused on eye tracking indices of attentional biases for a more direct and psychometrically robust measure of gaze allocation, allowing clearer interpretation of sustained attention patterns that are theoretically relevant to depression (Armstrong & Olatunji, 2012; Gibb et al., 2016a; Price et al., 2015). We predicted prospective relations between attentional biases for sad faces and depressive symptoms over time, such that each would predict prospective changes in the other across measurement waves. Although we focused on attentional biases for sad faces, we examined attentional biases for happy and angry faces to examine the specificity of the effects.

A secondary aim was to determine whether these relations may be stronger among youth already at risk for depression. There is clear evidence that offspring of mothers with a history of MDD are at increased risk for depression themselves (for a review, see Goodman, 2020). Research further suggests that these youth exhibit altered patterns of attention to emotional stimuli, including biases for sad faces, and differ from offspring of never depressed mothers in attentional responding (Gibb et al., 2016a; Gibb et al., 2023; Joormann et al., 2007; Kujawa et al., 2011). Offspring of mothers with a history of MDD have greater genetic vulnerability and may be exposed to more frequent or intense negative emotional cues, which in turn may shape how they attend to stimuli over time (Gotlib et al., 2020). In at-risk samples (i.e., children of depressed mothers) attentional avoidance of sad faces has been proposed as an emotion regulation strategy (e.g., Gibb et al., 2016b), with risk hypothesized to increase as these youth are no longer able to disengage attention from depression-relevant stimuli. Although maternal depression has been proposed to heighten the effects of cognitive vulnerabilities in youth, relatively few studies have tested whether maternal history of MDD moderates prospective relations between attentional biases and depressive symptoms. That said, there is evidence that other cognitive vulnerabilities confer risk more strongly among offspring of mothers with MDD (Abela et al., 2006; Abela et al., 2012; Burkhouse et al., 2015; Gibb et al., 2012), though few studies have formally tested a moderation hypothesis. Based on the existing research, therefore, we predicted that the link between attentional biases for sad faces and future changes in depressive symptoms would be stronger among offspring of mothers with a history of MDD than among offspring of never depressed mothers. We also examined whether maternal history of MDD moderated the reverse direction of influence, though no specific hypotheses were made.

In line with these aims, the present study tested bidirectional, prospective associations between attentional biases and depressive symptoms across five waves and examined maternal MDD history as a moderator of these prospective paths. We hypothesized that attentional biases for sad faces and depressive symptoms would predict changes in one another across time points, and that these associations would be stronger among youth already at risk for depression due to a maternal history of MDD. Analyses of attentional biases for happy and angry faces were examined for specificity. Finally, sensitivity analyses examined whether any observed effects were maintained after statistically controlling for the child age or sex, given evidence that both attentional processing and depression risk may vary across development and by sex (Chaplin et al., 2009; Chen et al., 2015; Jenness et al., 2021).

Method

Participants

Participants for this study were 255 mother-child dyads recruited from the local community through various advertisements (e.g., flyers and newspaper ads) as part of a larger project examining risk for the intergenerational transmission of depression. Data collection occurred between January 2010 and August 2014. Mothers were required to either have a history of major depressive disorder (MDD) during their child’s life according to the Diagnostic and Statistical Manual of Mental Disorders 4th ed. (DSM-IV; American Psychiatric Association, 2000) or have no lifetime history of any mood disorder. Only mothers were included because a history of MDD in mothers served as the defining risk variable for the broader longitudinal project, and diagnostic data were not available for fathers. Exclusion criteria for both groups included history of schizophrenia, alcohol or substance abuse within the last 6 months, or lifetime history of bipolar disorder.

Participating youth were required to be biological offspring of the mothers and between the ages of 8–14 at baseline. If more than one child in a family met the inclusion criteria, one child was randomly selected for participation. Of the original 255 dyads, four were excluded because a mother in the control group developed a first onset of MDD during the two-year follow-up. An additional nine participants were excluded for not having usable eye-tracking data for at least one of the five timepoints, resulting in a final sample of 242 mother-child dyads (123 in the MDD group, 119 in the never depressed group). At baseline, 17 youth met criteria for a lifetime history of MDD, with two meeting criteria for current MDD. Included dyads did not differ from those excluded on any demographic or clinical characteristics (all ps > .130). The average age of youth included in this study was 11.35 years (SD = 1.91) at baseline, with 51.65% girls. Of these youth, 81.40% were non-Hispanic White, 4.55% were Black, 11.16% were multiracial, and the remaining 2.89% were from other racial/ethnic groups. Among the mothers, most were non-Hispanic White (86.50%), and the remaining were Black (4.64%), multiracial (4.13%), or from other racial/ethnic groups (5.37%). The median annual family income was $50,001 to $55,000.

Measures

To determine current and lifetime diagnoses of DSM-IV Axis I disorders among mothers, the Structured Clinical Interview for DSM– IV Axis I Disorders (SCID-I; First et al., 1994) was used. As noted above, 123 mothers had a history of MDD during their child’s life and 119 had no lifetime history of any mood disorder. Of the mothers in the MDD group, 15.45% (n = 19) were in a current MDD episode at baseline.¹ To assess interrater reliability, 20 randomly selected SCID-I interviews were coded by a second interviewer and reliability for diagnoses of MDD were found to be excellent (κ = 1.00). To better characterize the sample, we assessed youth’s history of MDD using the Schedule for Affective Disorders and Schizophrenia for School-Aged Children–Present and Lifetime Version (K-SADS-PL; Kaufman et al., 1997), which was separately administered to youth and their mothers. As noted above, at the baseline assessment 17 youth met criteria for a lifetime history MDD, with two meeting criteria for current MDD. As with SCID-I interviews, we assessed the interrater reliability of K-SADS MDD diagnoses by having 20 interviews coded by a second interviewer. Reliability for diagnoses of MDD in the children was excellent (κ = 1.00).

Children’s attentional biases toward facial displays of emotion were measured using a Tobii T60XL infrared video-based eye-tracker (60 Hz data rate; 1,920 × 1,200 pixels; 24-inch monitor) while they completed a modified dot probe task (cf. MacLeod et al., 1986). Participants sat approximately 65 cm from the monitor with unrestricted head movement, and gaze from both eyes was recorded throughout the task. To ensure accurate gaze tracking, participants first completed a five-point calibration of the eye tracker by looking at specific points at the center and corners of the monitor. Calibration accuracy was confirmed through a visual inspection of fixations recorded during the calibration process. A fixation during the task was defined as a sustained gaze directed at a predefined area of interest for at least 100 ms. The dot probe task included 192 trials presented across two blocks, with a brief break in between. Each trial began with a central fixation cross, followed by a 1,000 ms presentation of a face pair (angry-neutral, happy-neutral, or sad-neutral), drawn from the NimStim standardized stimulus set (Tottenham et al., 2009) and created using 32 actors (16 male, 16 female). Faces measured 15.50 cm in height × 12.75 cm in width (13.60° × 11.20° visual angle), and stimulus order was randomized. After the face pair, a probe (one or two asterisks) replaced one of the images and participants indicated the number of asterisks using a response box.² Probes were equally likely to replace emotional or neutral faces, and intertrial intervals ranged from 750 to 1,250 ms. Attentional biases were indexed as the duration of gaze to emotional relative to neutral faces. To do this, we used regression analyses with gaze duration to neutral stimuli predicting gaze duration to emotional stimuli for each emotional-neutral pairing at each time point and saved the unstandardized residual, reflecting attention to emotional relative to neutral faces for each emotion type. These residual bias scores were used for all analyses. Spearman-Brown split-half reliabilities for the attention bias scores to angry, happy, and sad faces averaged across waves were .83, .68, .63, respectively. All three attention bias measures showed medium to large six-month retest stability across the waves of data (see Table 2).

Table 2.

Correlations among study variables

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20
1. Mom MDD	--
2. T1 Attention to Sad	−.11	--
3. T2 Attention to Sad	−.02	.35	--
4. T3 Attention to Sad	−.06	.35	.25	--
5. T4 Attention to Sad	.05	.35	.45	.36	--
6. T5 Attention to Sad	.09	.19	.31	.38	.47	--
7. T1 Attention to Happy	−.08	.53	.30	.16	.31	.18	--
8. T2 Attention to Happy	−.00	.30	.62	.22	.36	.26	.36	--
9. T3 Attention to Happy	−.11	.30	.41	.55	.24	.17	.25	.28	--
10. T4 Attention to Happy	−.01	.24	.48	.32	.63	.35	.36	.38	.44	--
11. T5 Attention to Happy	.10	.32	.30	.23	.50	.61	.33	0.39	.31	.58	--
12. T1 Attention to Angry	−.03	.65	.47	.47	.38	.37	.63	.40	.41	.30	.39	--
13. T2 Attention to Angry	.07	.35	.70	.33	.40	.40	.29	.65	.37	.34	.34	.54	--
14. T3 Attention to Angry	−.07	.39	.49	.66	.38	.39	.28	.45	.64	.41	.25	.57	.60	--
15. T4 Attention to Angry	−.03	.26	.47	.55	.72	.56	.17	.35	.42	.69	.47	.38	.43	.57	--
16. T5 Attention to Angry	.08	.23	.35	.41	.50	.65	.10	.30	.34	.38	.57	.36	.46	.46	.60	--
17. T1 CDI	.23	−.03	.12	.10	.13	.11	.05	.12	.11	.17	.08	−.01	.04	.03	−.04	−.10	--
18. T2 CDI	.26	−.01	−.02	.15	.11	−.04	.06	.05	.09	.17	−.17	−.01	−.03	−.03	.03	−.17	.62	--
19. T3 CDI	.25	−.03	−.07	.18	.13	.07	.06	.05	.16	.19	−.05	−.05	.04	.01	.01	−.13	.55	.73	--
20. T4 CDI	.24	−.05	−.07	.09	.00	−.05	.09	.04	.10	.00	−.06	−.03	.01	−.04	−.09	−.26	.37	.55	.62	--
21. T5 CDI	.16	−.15	−.18	−.04	−.05	.08	−.06	−.09	.01	−.06	−.12	−.15	−.16	−.12	−.06	−.13	.25	.47	.51	.60

Note. MDD = Major Depressive Disorder. CDI = Children’s Depression Inventory. Gray shading is used for ease of viewing and highlights correlations between variables from the same domain (e.g., all Sad, Happy, Angry, or CDI variables). Correlations > |.13| significant at p < .05. Correlations > |.16| significant at p < .01.

Youth depressive symptoms were measured at each of the five timepoints using the Children’s Depression Inventory (CDI; Kovacs, 1981), which is a 27-item self-report measure commonly used with youth ranging from 7 to 17 years old. The CDI has been shown to have strong reliability and validity in community samples of youth (Smucker et al., 1986) and studies have supported measurement invariance, including longitudinal invariance, across ages within the age range of this sample (e.g., Jelínek et al., 2021; Lei et al., 2016; Scott et al., 2016; Stumper et al., 2019). In the current study, the CDI demonstrated good reliability in the present study across each of the five assessments (αs = .85−.89).

Procedure

Upon arrival at the laboratory for the baseline assessment, mothers were asked to provide informed consent, and offspring were asked to provide assent to be in the study. Next, mothers were administered the SCID-I while children completed the dot probe task and CDI. After the mother completed the SCID-I, a different interviewer administered the K-SADS to mothers and children separately. Participants then completed follow-up assessments approximately every 6 months for 2 years (five assessments total). In practice, the average duration between follow-ups was 6.17 months (SD = 0.91). At each follow-up assessment, children completed the dot probe task and CDI again. Participating families were compensated monetarily for their time. All study procedures were approved by the university’s Institutional Review Board.

Analysis Plan

Random intercepts cross-lagged panel models (RI-CLPM; Hamaker et al., 2015) were used to evaluate transactional relations between children’s attentional biases to emotional faces and depressive symptoms over the two-year follow-up. This approach allowed us to differentiate within-person fluctuations from between-person differences in our variables of interest, across time. The hypothesized models included reciprocal cross-lagged paths predicting prospective changes in depressive symptoms from attentional biases, and vice versa, controlling for autoregressive effects. In RI-CLPMs, between-person associations between attentional biases and depressive symptoms are captured by the covariance between the random intercepts, reflecting stable, trait-like differences across individuals averaged over time. The general model is presented in Figure 1. To examine whether between-subject differences in child age or sex influenced the findings, exploratory analyses were conducted to re-estimate all models including age and sex as predictors of the random intercepts. Finally, although we were primarily in interested in the role of youth depressive symptoms, we also tested models with youth anxiety, assessed at each wave with the Multidimensional Anxiety Scale for Children (March et al., 1997) to examine the specificity of our effects.

Figure 1.

Random Intercepts Cross-Lagged Panel Model (RI-CLPM) Examining Relations between Attentional Biases to Emotional Stimuli and Depressive Symptoms Across Time

Note. AB = Attention Bias; CDI = Children’s Depression Inventory.

Model building followed a three-step approach with each emotion type examined separately. First, for each form of attentional bias (i.e., sad, happy, angry), we first fit an unconstrained model in which all paths were allowed to vary across time and across groups. Next, we tested an across-wave equivalence model in which autoregressive and cross-path weights were constrained to be equal across waves of data within each group. To compare the models, we conducted nested model comparisons using chi-square (χ²) difference tests to determine whether adding these constrained significantly reduced overall model fit. Finally, we tested whether maternal MDD group moderated any of the effects by constraining all paths, as well as the correlation between latent intercepts, to be equal across groups. Again, we used nested model comparisons to compare the relative fit of the models. A significant reduction in model fit when the paths were constrained indicated that the strength or direction of these cross-lagged effects varied by maternal MDD history, consistent with moderation. Conceptually, this approach is comparable to testing a statistical interaction, as it examines whether the relation between two variables (e.g., attentional bias and later depressive symptoms) depends on group membership. Model fit was evaluated using standard indices, including the comparative fit index (CFI) and root mean squared error of approximation (RMSEA). Good model fit was indicated by CFI > .90 and RMSEA < .08, with excellent fit indicated by CFI > .95, RMSEA < .06.

Transparency and Openness

This study was not preregistered. Although data from this project cannot be posted to an online repository because not all participants gave consent for their data to be shared, data are available from the first author upon reasonable request. All outputs for statistical analyses are available at https://osf.io/3jnsu/?view_only=37cd497dff104031a61f52b33b7a2793. Data were analyzed using AMOS, version 29.0.0 (IBM Corp., 2022). In this paper, we report how we determined our sample size, all data exclusions, manipulations, and measures in the study.

Results

Of the 242 mother-child dyads who participated in the baseline assessment, 205, 193, 166, and 171 completed follow-up assessments at 6, 12, 18, and 24 months, respectively, and 81.40% completed at least 3 of the 5 assessments. A preliminary evaluation of the data revealed that scores on the Children’s Depression Inventory (CDI) exhibited significant skew. To meet normality assumptions, square root transformations were applied prior to conducting further analyses. Given the presence of missing data across time points, we examined whether the data were missing at random to determine the suitability of data imputation. Little’s Missing Completely at Random (MCAR) test (Little & Rubin, 1987), for which the null hypothesis is that the data are missing completely at random, was nonsignificant, χ²(1039) = 1083.46, p = .160, supporting the estimation of missing values. Given this, we used full information maximum likelihood (FIML) estimation in AMOS to handle missing data. This approach uses all available observed data to estimate model parameters and yields less biased and more efficient estimates than traditional approaches such as pairwise or listwise deletion (see Little et al., 2014).

Descriptive statistics, including means, standard deviations, and correlations among study variables, are presented in Tables 1 and 2. Of the youth, 6.4%, 4.4%, 2.4%, 3.2%, and 5.6% scored above the recommended clinical cutoff of 16 on the CDI (Timbremont et al., 2004) at timepoints 1, 2, 3, 4, and 5, respectively. Therefore, although drawn from the community, there was a meaningful range of depressive symptoms displayed among children in this sample.

Table 1.

Descriptive statistics for study variables

	Mothers with MDD (n = 123)	Never Depressed Mothers (n = 119)
Child Gender (% Girls)	49.59%	53.78%
Child Age (T1)	11.37 (2.01)	11.34 (1.82)
Child Race (% Non-Hispanic White)	70.73%	90.76%
T1 Attention to Sad	−4.51 (38.48)	4.55 (46.49)
T2 Attention to Sad	−1.05 (47.17)	1.05 (55.17)
T3 Attention to Sad	−2.65 (39.61)	2.86 (52.49)
T4 Attention to Sad	2.73 (46.14)	−2.73 (54.72)
T5 Attention to Sad	4.45 (44.23)	−3.89 (46.83)
T1 Attention to Happy	−3.62 (39.21)	3.65 (47.95)
T2 Attention to Happy	−0.07 (54.66)	0.18 (48.95)
T3 Attention to Happy	−5.14 (45.30)	5.54 (56.70)
T4 Attention to Happy	−0.71 (49.26)	0.71 (55.88)
T5 Attention to Happy	5.36 (53.16)	−4.69 (46.62)
T1 Attention to Angry	−1.86 (50.60)	1.87 (62.90)
T2 Attention to Angry	4.12 (64.49)	−4.21 (61.23)
T3 Attention to Angry	−4.22 (58.15)	4.54 (72.26)
T4 Attention to Angry	−2.01 (64.68)	2.01 (66.18)
T5 Attention to Angry	4.64 (57.61)	−4.06 (53.94)
T1 CDI	7.54 (5.98)	4.84 (5.31)
T2 CDI	6.75 (7.18)	3.64 (4.35)
T3 CDI	6.08 (5.56)	3.56 (4.31)
T4 CDI	5.40 (5.85)	2.91 (4.16)
T5 CDI	5.44 (5.72)	3.69 (5.37)

Note. Unless otherwise noted, values represent means (and standard deviations). MDD = Major Depressive Disorder. CDI = Children’s Depression Inventory. Attention variables reflect residual difference scores for gaze duration to emotional faces relative to neutral faces. To facilitate comparisons with other studies, statistics for the CDI are based on untransformed variables.

Attentional Biases to Sad Faces

Random intercept cross-lagged panel models (RI-CLPMs) were then used to test transactional relations between attentional biases to sad faces and depressive symptoms across the multi-wave follow-up. For attentional biases to sad faces, the across-wave equivalence model did not fit worse than the unconstrained model, χ²(24) = 25.11, p = .400, and provided a good fit to the data, χ²(66) = 79.75, p = .119, CFI = .97, RMSEA = .03. However, when we tested a model with all paths constrained to be equal across groups, this fit significantly worse than the model in which the paths were allowed to vary across MDD groups, χ²(5) = 13.27, p = .021, indicating significant moderation by maternal MDD of one or more paths. Thus, we retained the model assuming across-wave equivalence within groups but allowing paths to vary between groups. The results of this model are presented in Table 3.

Table 3.

Random Intercept Cross-Lagged Panel Model (RI-CLPM) Results

	Mothers with MDD			Never Depressed Mothers
	beta	S.E.	p	beta	S.E.	p
Sad Attention Bias
Autocorrelation: L_ABT → L_ABT+1	0.111	0.092	0.226	−0.094	0.087	0.280
Autocorrelation: L_CDIT → L_CDIT+1	0.490	0.083	<.001	0.151	0.082	0.066
Cross path: L_ABT → L_CDIT+1	0.001	0.002	0.629	0.000	0.001	0.980
Cross path: L_CDIT → L_ABT+1	8.253	3.269	0.012	4.404	4.061	0.278
Correlation between Latent Intercepts	r =−.276	p =.242		r =.251	p =.048
Happy Attention Bias
Autocorrelation: L_ABT → L_ABT+1	0.170	0.092	0.065	0.071	0.086	0.405
Autocorrelation: L_CDIT → L_CDIT+1	0.477	0.083	<.001	0.158	0.080	0.049
Cross path: L_ABT → L_CDIT+1	0.003	0.001	0.055	0.000	0.001	0.958
Cross path: L_CDIT → L_ABT+1	4.170	3.438	0.225	−10.751	3.685	0.004
Correlation between Latent Intercepts	r =.000	p = 1.00		r =.111	p =.389
Angry Attention Bias
Autocorrelation: L_ABT → L_ABT+1	0.290	0.063	<.001	0.290	0.063	<.001
Autocorrelation: L_CDIT → L_CDIT+1	0.313	0.061	<.001	0.313	0.061	<.001
Cross path: L_ABT → L_CDIT+1	0.001	0.001	0.164	0.001	0.001	0.164
Cross path: L_CDIT → L_ABT+1	5.854	3.017	0.052	5.854	3.017	0.052
Correlation between Latent Intercepts	r =.079	p =.494		r =.072	p =.494

Note. MDD = major depressive disorder. AB = attention bias. CDI = Children’s Depression Inventory. Beta values are unstandardized. For all models, across-wave equivalence was supported. Therefore, autoregressive and cross-lagged paths were equal across waves. For attentional biases to sad and happy faces, estimates differ across MDD groups. For attentional biases to angry faces, there was no moderation by mother MDD history, so estimates are equal across groups.

Among children of mothers with MDD, depressive symptoms significantly predicted future increases in attentional biases to sad faces, beta = 8.25, SE = 3.27, p = .012, with standardized beta (β) values ranging from .18 to .25.³ However, attentional biases to sad faces did not significantly predict prospective changes in depressive symptoms, beta = .001, SE = .002, p = .629, β = .03 to .04. For children of never depressed mothers, neither attentional biases to sad faces, beta = .000, SE = .001, p = .980, β = −.002 to −.001, nor depressive symptoms, beta = 4.40, SE = 4.06, p = .278, β = .07 to .10, significantly predicted each other over time. At the between subjects level, the random intercepts for attentional bias to sad faces and depressive symptoms were not significantly correlated among children of mothers with MDD, r = −.28, p = .242, but were significantly positively associated among children of never depressed mothers, r = .25, p =.048.

Attentional Biases to Happy Faces

Turning next to attentional biases to happy faces, the unconstrained model failed to converge due to a non-positive definite covariance matrix caused by the covariance between the random intercepts in the maternal MDD group. When this covariance was fixed to zero, the model converged successfully and, therefore, this constraint was included in all additional model comparisons. The across-wave equivalence model did not fit worse than this less constrained model, χ²(24) = 32.89, p = .106, and provided a good fit to the data, χ²(67) = 104.68, p = .002, CFI = .93, RMSEA = .05. However, when we tested a model with all paths constrained to be equal across groups, this fit significantly worse than the model in which the paths were allowed to vary across MDD groups, χ²(5) = 16.00, p = .007, indicating significant moderation by maternal MDD of one or more paths. Thus, we retained the model assuming across-wave equivalence within groups but allowing paths to vary between groups (see Table 3).

Within this model, attentional biases for happy faces did not significantly predict prospective changes in depressive symptoms in either the maternal MDD group, beta = .003, SE = .001, p = .055, β = .09 to .12, or the never depressed group, beta = .000, SE = .001, p = .958, β = .003 to .004. Similarly, depressive symptoms did not significantly predict future changes in attentional biases to happy faces in the MDD group, beta = 4.17, SE = 3.44, p = .225, β = .08 to .10. However, among children of never depressed mothers, higher levels of depressive symptoms predicted future decreases in attentional biases to happy faces over time, beta = −10.75, SE = 3.69, p = .004, β = −.31 to −.15. At the between subjects level, the correlation between random intercepts was fixed to zero in the maternal MDD group as described above. In the never depressed group, the correlation was small and nonsignificant, r = .11, p = .389.

Attentional Biases to Angry Faces

Finally, for the models examining attentional biases to angry faces, the across-wave equivalence model again provided a good fit to the data, χ²(66) = 83.33, p = .018, CFI = .96, RMSEA = .04, and did not fit significantly worse than the unconstrained model, χ²(24) = 27.39, p = .286. We then tested the more constrained model requiring that autoregressive paths, cross-lagged paths, and the association between the random intercepts be equal across maternal depression groups. This model also demonstrated good fit to the data, χ²(71) = 101.69, p = .010, CFI = .95, RMSEA = .04, but did not fit significantly worse than the baseline model, χ²(5) = 9.36, p = .095, indicating no evidence of moderation by maternal MDD. The results of this final model are presented in Table 3. Examining the effects in this model across groups, there was a marginally non-significant effect of depressive symptoms on future increases in attentional biases to angry faces, beta = 5.85, SE = 3.02, p = .052, β = .10 to .12. In contrast, attentional biases to angry faces did not significantly predict prospective changes in depressive symptoms, beta = .001, SE = .001, p = .164, β = .05 to .08. At the between subjects level, the correlation between the random intercepts for attentional bias to angry faces and depressive symptoms was not significant, r = .08, p = .494.

Exploratory Analyses

Exploratory moderation analyses examined child age and sex as predictors of the random intercepts for both depressive symptoms and attentional biases (sad, happy, and angry). Including these predictors did not alter the overall pattern or significance of results. Sensitivity analyses examining anxiety symptoms similarly yielded no evidence of prospective associations between attentional biases and anxiety in either direction, across models. Details of these analyses are available at https://osf.io/3jnsu/?view_only=37cd497dff104031a61f52b33b7a2793.

Discussion

The current study examined transactional relations between attentional biases for emotional faces and depressive symptoms in youth over a two-year period, with a particular focus on whether relations between attentional biases for sad faces and depressive symptoms were stronger among youth at elevated risk for depression based on maternal history of MDD. Focusing first on attentional biases to sad faces, we found that, among offspring of mothers with a history of MDD, higher depressive symptoms at a given assessment wave predicted subsequent increases in attentional bias to sad faces. This effect was unique to high-risk youth and was not observed among offspring of never depressed mothers. Attentional biases to sad faces did not predict prospective change in depressive symptoms in either group. Depressive symptoms were also more stable among offspring of mothers with MDD, suggesting more persistent mood disturbances in the high-risk group. At the between-person level, average depressive symptoms were associated with greater attention to sad faces only among offspring of never depressed mothers, suggesting that changes in attentional biases to sad faces are more likely following experiences of depression in at-risk youth, but still appear to be a correlate of depression in offspring of never depressed mothers.

These results are consistent with theory and research suggesting that experiences of depression may contribute to the development or worsening of cognitive vulnerabilities, which may then further increase risk for future depression later in life (Gibb et al., 2006; Lewinsohn et al., 1981; Shankar & Gibb, 2024). The current findings suggest that this scarring effect may be more pronounced among offspring of mothers with a history of MDD. Although there are various mechanisms by which intergenerational risk may be transmitted, including inherited vulnerability, greater exposure to negative or dysregulated affect, and heightened stress within the caregiving environment (Goodman & Gotlib, 1999; Goodman, 2020; Hammen & Brennan, 2003), it is also clear that children of mothers with MDD are exposed to facial displays of sadness, which may increase the salience of these types of cues when they experience depressive symptom increases themselves. Although we focused on mothers with a history of at least one episode of MDD during their children’s lives compared to mothers with no lifetime history of depression, MDD is often recurrent and may be associated with depressive symptom elevations even when not meeting criteria for the disorder (Goodman, 2020; Solomon et al., 2000), thus increasing the potential salience of sad facial stimuli in at-risk youth.

Contrary to hypothesis, attentional biases to sad faces did not predict prospective increases in depression in either group. Although several factors could have contributed to this null finding, at least two are particularly salient and should be investigated in future research. First, these biases are thought to develop during childhood and early adolescence before becoming more stable vulnerability factors. Therefore, it is possible that the predictive power of attentional biases increases in later adolescence and into adulthood. This possibility is supported by research in adults showing that attentional biases for sad faces does predict prospective changes in depression (Beevers et al., 2011; Clasen et al., 2013). A second possibility is that these biases, like many cognitive vulnerabilities, operate within a vulnerability-stress framework, serving to increase risk for depression in the presence, but not absence, of stress. These possibilities are not mutually exclusive and there is some evidence that attentional biases moderate the impact of stress on depressive symptom increases and that this effect is stronger in older, compared to younger, adolescents (Feurer et al., 2020). Future research is needed to further explore these hypotheses.

With respect to attentional biases for happy faces, within-person associations with depressive symptoms differed by maternal depression status. Among offspring of mothers with a history of MDD, no significant prospective associations were observed. In contrast, among offspring of never depressed mothers, higher depressive symptoms at a given assessment prospectively predicted decreases in attention to happy faces, with no evidence for the reverse direction of influence. Combined with the findings observed for attentional biases to sad faces, this suggests that experiences with depression have differential effects on attentional processes among youth with and without a maternal history of MDD. Specifically, whereas increases in depression predict increases in attention to sad faces among youth with a maternal history of MDD, they predict decreases in attention to happy faces among youth of never depressed mothers. Therefore, even among lower-risk youth, experiences of depression may erode a putatively protective information-processing style by reducing attention to positive emotional stimuli.

Finally, with respect to attentional biases for angry faces, results were equivalent across mother MDD groups, with no significant prospective associations in either direction between attentional biases and depressive symptoms. Although there was some trending evidence that depressive symptoms may predict future increases in attentional biases to angry faces, this effect was not statistically significant and should be interpreted with caution. This said, prior research has reported links between depression and attention to angry faces (e.g., Woody et al., 2016; Platt et al., 2015), suggesting that depression may be linked to broader processing of negative emotional information. We should also note that attentional biases to angry faces have been most consistently associated with anxiety (for reviews, see Armstrong & Olatunji, 2012; Dudeney et al., 2015), though there was no evidence for this link in our exploratory analyses of youth anxiety. In studies of anxiety, however, attentional biases to angry faces are typically observed at shorter stimulus durations (e.g., ≤ 500 ms), reflecting early, automatic vigilance to threat (Abend et al., 2018) whereas we focused on sustained attention over 1000 ms. Future studies specifically focusing on initial orienting of attention may provide a more powerful examination of potential transactional relations between attentional biases for angry faces (or threat-relevant stimuli more generally) and symptoms of anxiety over time in youth.

The present study has several notable strengths, including its multi-wave longitudinal design, use of eye tracking to assess attentional biases, and focus on a high-risk sample. First, our multi-wave longitudinal design allowed us to examine prospective, within-person associations between attentional biases and depressive symptoms across a critical developmental period. Our analytic approach, using random intercept cross-lagged panel models, further strengthened our ability to distinguish within-person fluctuations from stable between-person differences over time. Second, the use of eye-tracking provided a direct and temporally precise measure of attentional allocation. Importantly, we assessed attentional biases for multiple emotional expressions (sad, happy, and angry faces) enabling a more detailed investigation of emotion-specific attentional patterns. Additionally, the use of a high-risk sample, clinically defined based on structured diagnostic interviews with mothers, allowed us to test how maternal risk moderates these processes during a key developmental period, when cognitive vulnerabilities are hypothesized to still be developing.

Constraints on Generality

There are also some aspects of the study that may limit generalizability of the findings and highlight important avenues for future research. First, our sample included 8–14-year-olds assessed for two years, and additional research with younger and older samples is needed to determine how these transactional relations may change across development. Although results were maintained after accounting for the effects of child age and sex at the intercept level, the sample was underpowered to formally evaluate age as a moderator of prospective paths. Second, despite the strengths of our analytic approach, there are limits to which causal conclusions can be drawn from RI-CLPMs and researchers have recently proposed alternative approaches to rule out potential confounding influences that could be incorporated in future research (Kenny & McCoach, 2025; Muthén & Asparouhov, 2024). Third, although maternal history of MDD is a well-established marker of elevated depression risk, it represents only one component of familial vulnerability. We did not assess paternal psychopathology or consider concurrent symptoms in mothers or broader family stressors, any of which may also shape attentional and emotional development in youth and could moderate transactional associations. Future studies should incorporate more comprehensive assessments of family mental health to clarify the specificity and breadth of these effects. Fourth, our sample was predominantly non-Hispanic White, which may limit the generalizability of findings. Replicating this work in more racially, ethnically, and culturally diverse populations is an important next step to determine whether similar patterns emerge across groups with different lived experiences and contextual factors. Fifth, we focused on depressive symptoms in youth and future research is needed to determine whether the results generalize to clinically significant depressive diagnoses. Finally, although previous research has supported measurement invariance, including longitudinal invariance, of the CDI in this age range (Jelínek et al., 2021; Lei et al., 2016; Scott et al., 2016; Stumper et al., 2019), our sample size was not sufficient to provide a formal test in this sample across fives waves of data. Future research is also needed to elucidate how best to examine measurement invariance of attention bias scores across in longitudinal studies.

Conclusion

Overall, the current findings underscore the idea that relations between attentional biases and depressive symptoms in youth are dynamic, developmentally sensitive, and are likely shaped by context. Importantly, they suggest that the influence of depression on youth attentional biases may differ for offspring of mothers with a history of MDD compared to offspring of never depressed mothers. Specifically, in offspring of mothers with MDD, experiences of depression appear to increase levels of a known vulnerability factor – attentional biases to sad faces – perhaps because stimuli reflecting sad facial expressions are particularly salient for these youth. In contrast, among offspring of never depressed mothers, experiences of depression appear to decrease levels of a putative protective factor, attention to happy facial stimuli. If replicated, this could suggest the presence of distinct pathways of risk for youth with and without a maternal history of depression. Across analyses, depressive symptoms emerged as a primary predictor of subsequent attentional bias to emotional information, particularly attention to sad faces among offspring of mothers with MDD. This pattern was not observed among youth without maternal risk, suggesting that maternal depression may amplify the extent to which emerging depressive symptoms shape downstream cognitive-affective processes. More broadly, the findings indicate that attentional biases in youth may be better conceptualized as symptom-linked processes rather than fixed vulnerabilities.

Clinically, these findings suggest that shifts in attention to emotional information may be more likely to emerge after increases in depressive symptoms, particularly among offspring of mothers with a history of MDD during their lives. For offspring of mothers with MDD, elevated symptoms prospectively predicted greater attention to sad faces over time, consistent with the idea that symptom escalation may “tune” attention toward depression-relevant cues. In contrast, among youth without a maternal history of MDD, increases in depressive symptoms predicted reduced attention to happy faces, suggesting that symptom increases may erode attention to positive social signals. Together, these patterns highlight depressive symptoms as an early, practical target for monitoring and intervention, and suggest that attention-based markers may be most informative when interpreted in the context of maternal risk and symptom change rather than as uniform premorbid vulnerabilities.