Early-life adversities are highly prevalent and are associated with a wide range of negative consequences in behavioral and neural development of adolescents (Holland et al., Reference Holland, Khandaker, Dauvermann, Morris, Zammit and Donohoe2020; North, Fox, & Doom, Reference North, Fox and Doom2023; Rakesh et al., Reference Rakesh, Kelly, Vijayakumar, Zalesky, Allen and Whittle2021a; Ramos et al., Reference Ramos, Dion, Bőthe, Girouard, Hébert, Wong and Bergeron2022). To better understand the specific mechanisms linking disparate adversities and development, accounting for the notion that different types of adversities often co-occur and may share common features, recent studies have increasingly adopted dimensional models, which generally include three dimensions: threat, deprivation, and unpredictability (Ellis, Figueredo, Brumbach, & Schlomer, Reference Ellis, Figueredo, Brumbach and Schlomer2009; McLaughlin, Sheridan, & Lambert, Reference McLaughlin, Sheridan and Lambert2014; Sheridan & McLaughlin, Reference Sheridan and McLaughlin2014). Although all three dimensions are closely linked to children's negative developmental outcomes, the unique effects of unpredictability and its underlying mechanisms are still unclear compared with the other dimensions (Liu & Fisher, Reference Liu and Fisher2022; Wade, Wright, & Finegold, Reference Wade, Wright and Finegold2022), highlighting the need for further empirical investigation.

Unpredictability refers to spatial-temporal variation in threat or deprivation (Ellis, Sheridan, Belsky, & McLaughlin, Reference Ellis, Sheridan, Belsky and McLaughlin2022). Two perspectives, the ancestral cue and statistical learning perspectives, offer insights into the approaches of measuring unpredictability (Young, Frankenhuis, & Ellis, Reference Young, Frankenhuis and Ellis2020). The ancestral cue perspective proposes that humans evolved to detect unpredictability; therefore, unpredictability can be assessed with cues (e.g. parental transition) that reliably indicate high unpredictability. Statistical learning perspective suggests that humans evaluate the level of unpredictability by integrating variations in lived experiences throughout development; therefore, unpredictability can be measured by collecting series data indicating statistical properties, such as variance and autocorrelation (Young et al., Reference Young, Frankenhuis and Ellis2020). Moreover, based on topological approach and related studies, how children understand and interpret stressful experiences might shape their biological and psychosocial development, beyond the impact of exposure to stressful events (Baldwin & Degli Esposti, Reference Baldwin and Degli Esposti2021; Smith & Pollak, Reference Smith and Pollak2021, Reference Smith and Pollak2022; Ugarte & Hastings, Reference Ugarte and Hastings2022). In this study, we adopted an ancestral cue perspective, focusing on children's perception of unpredictability, to examine the associations between unpredictability and subsequent neural and behavioral outcomes in adolescence.

According to the life history theory (Roff, Reference Roff and Levin2002; Stearns, Reference Stearns1992), individuals tend to adopt faster life history strategies (e.g. early pubertal maturation, sexual behavior, and reproductive timing, as well as increased impulsivity and risk taking) to enhance evolutionary fitness in unpredictable environment (Belsky, Schlomer, & Ellis, Reference Belsky, Schlomer and Ellis2012; Ellis et al., Reference Ellis, Figueredo, Brumbach and Schlomer2009). Although being evolutionarily adaptive, faster life history strategies are associated with adverse developmental outcomes in the long-term, especially incurring more risky and aggressive behaviors, which are typical types of externalizing problems (Ellis et al., Reference Ellis, Figueredo, Brumbach and Schlomer2009, Reference Ellis, Sheridan, Belsky and McLaughlin2022; Ellis, Shakiba, Adkins, & Lester, Reference Ellis, Shakiba, Adkins and Lester2021; Martinez et al., Reference Martinez, Hasty, Morabito, Maranges, Schmidt and Maner2022). The associations between unpredictability and internalizing problems, however, are less consistent (Farkas, Baptista, Speranza, Wyart, & Jacquet, Reference Farkas, Baptista, Speranza, Wyart and Jacquet2024; Li & Belsky, Reference Li and Belsky2022; Li, Sturge-Apple, Jones-Gordils, & Davies, Reference Li, Sturge-Apple, Jones-Gordils and Davies2022; Lindert et al., Reference Lindert, Maxwell, Liu, Stern, Baram, Poggi Davis and Glynn2022; Spadoni et al., Reference Spadoni, Vinograd, Cuccurazzu, Torres, Glynn, Davis and Risbrough2022). Informed by the life history theory, recent studies started to highlight the unique role of unpredictability in children's development above and beyond the other dimensions (Li et al., Reference Li, Song, Xiang, Fu, Zhou and Chen2023; Liu & Fisher, Reference Liu and Fisher2022; Wang, Cao, Zheng, Chen, & Zhu, Reference Wang, Cao, Zheng, Chen and Zhu2023); however, it remains uncertain how unpredictability shapes socioemotional development. Elucidating the underlying neural mechanisms will improve our understanding of the deep reasons behind the associations between unpredictability and diverse types of behavioral problems.

Different dimensions of adversity may be interactively associated with neurodevelopment and psychopathology. For example, social deprivation exacerbated the effects of childhood violence exposure on the development of amygdala-orbitofrontal cortex white matter connections (Goetschius et al., Reference Goetschius, Hein, Mitchell, Lopez-Duran, McLoyd, Brooks-Gunn and Monk2020). To our best knowledge, however, limited studies have examined the interactive effects of unpredictability and other dimensions of adversity on psychopathology and the neural underpinnings, hindering our understanding of the conditions under which unpredictability may have an impact. In a recent review, Colich, Rosen, Williams, and McLaughlin (Reference Colich, Rosen, Williams and McLaughlin2020) proposed that whether unpredictability associated with deviated development depends upon various features of the environment including resource availability. Therefore, deprivation, involving limited or reduced social and cognitive inputs from the environment (McLaughlin et al., Reference McLaughlin, Sheridan and Lambert2014; Sheridan & McLaughlin, Reference Sheridan and McLaughlin2014), may moderate the associations between unpredictability and neural as well as socioemotional development. Moreover, the associations between adversity and brain development often differ by sex (Rakesh et al., Reference Rakesh, Kelly, Vijayakumar, Zalesky, Allen and Whittle2021a; Rakesh, Allen, & Whittle, Reference Rakesh, Allen and Whittle2023), although it is uncertain of sex differences in unpredictability–neurodevelopment associations. As such, this study examined the moderating effects of deprivation and sex on the associations between unpredictability and neurodevelopment. Considering environmental unpredictability mainly reflected variations in threat/deprivation at the level of family in this study, we particularly focused on deprivation in the neighborhood context, as neighborhood is also an important aspect of environment (Rakesh, Seguin, Zalesky, Cropley, & Whittle, Reference Rakesh, Seguin, Zalesky, Cropley and Whittle2021b).

Resting-state functional connectivity (rsFC) provides a powerful approach to examine the neurobiological pathways linking adversities and child development (Daliri & Behroozi, Reference Daliri and Behroozi2014). Previous work has mainly focused on rsFC of the frontolimbic circuitry (e.g. Brieant, Sisk, & Gee, Reference Brieant, Sisk and Gee2021; Kaiser et al., Reference Kaiser, Clegg, Goer, Pechtel, Beltzer, Vitaliano and Pizzagalli2018), though more widespread networks may be associated with adversities (e.g. Rakesh et al., Reference Rakesh, Kelly, Vijayakumar, Zalesky, Allen and Whittle2021a), such as the higher-order brain networks, including the cingulo-opercular network (CON), default mode network (DMN), and fronto-parietal network (FPN). The CON, encompassing the dorsal anterior cingulate cortex and bilateral anterior insula, consistently activates during tasks involving error detection and ongoing task management (Menon & Uddin, Reference Menon and Uddin2010). The DMN, comprising mainly midline cortical regions, such as the anterior medial prefrontal cortex and posterior cingulate cortex, activates when individuals are not focused on external tasks but instead are engaged in self-reflection or introspection (Raichle, Reference Raichle2015). The FPN, which consists of dorsolateral prefrontal and posterior parietal cortices, functions as a control network orchestrating behavior towards specific goals (Marek & Dosenbach, Reference Marek and Dosenbach2018). Moreover, connectivity between task-positive networks (typically activated during tasks, e.g. CON and FPN) and task-negative networks (typically deactivated during tasks, e.g. DMN; Yu et al., Reference Yu, Linn, Shinohara, Oathes, Cook, Duprat and Sheline2019) also has important functions. For example, connectivity between the FPN-B (subnetwork of FPN), dorsal attention network, CON and lateral DMN was associated with switching and inhibiting behaviors (Beaty, Cortes, Zeitlen, Weinberger, & Green, Reference Beaty, Cortes, Zeitlen, Weinberger and Green2021); greater negative connectivity between FPN and DMN was linked to less mind wandering during tasks necessitating external attention (Deck et al., Reference Deck, Kelkar, Erickson, Erani, McConathey, Sacchetti and Medaglia2023; Kelly, Uddin, Biswal, Castellanos, & Milham, Reference Kelly, Uddin, Biswal, Castellanos and Milham2008). Aberrant rsFC, both within and between CON, DMN, and FPN, were implicated in children's internalizing and externalizing problems (Chahal, Miller, Yuan, Buthmann, & Gotlib, Reference Chahal, Miller, Yuan, Buthmann and Gotlib2022; Rakesh et al., Reference Rakesh, Kelly, Vijayakumar, Zalesky, Allen and Whittle2021a, Reference Rakesh, Seguin, Zalesky, Cropley and Whittle2021b). As such, our study mapped unpredictability-related psychopathology onto the rsFC within and between large-scale networks (i.e. CON, DMN, and FPN) to offer a more comprehensive view of the brain mechanisms linking adversity and psychopathology.

Moreover, the brain networks undergo dynamic restructuring in late childhood and adolescence (Fair et al., Reference Fair, Dosenbach, Church, Cohen, Brahmbhatt, Miezin and Schlaggar2007; Grayson & Fair, Reference Grayson and Fair2017; Lin et al., Reference Lin, Zhu, Gao, Chen, Toh, Styner and Gilmore2008); however, the existing work of adversity has mostly relied on cross-sectional designs (e.g. Rakesh et al., Reference Rakesh, Seguin, Zalesky, Cropley and Whittle2021b), which cannot reveal how adversity impacts brain maturation over development. Normative developmental changes include positive associations between age and rsFC within networks (Truelove-Hill et al., Reference Truelove-Hill, Erus, Bashyam, Varol, Sako, Gur and Davatzikos2020; but see Chahal et al., Reference Chahal, Miller, Yuan, Buthmann and Gotlib2022), as well as negative associations between age and between-network connectivity (Stevens, Reference Stevens2016; but see Sanders et al., Reference Sanders, Harms, Kandala, Marek, Somerville, Bookheimer and Barch2023). Deviations from the typical development of brain networks could be detected under pathological conditions (Dennis & Thompson, Reference Dennis and Thompson2013; Rakesh et al., Reference Rakesh, Kelly, Vijayakumar, Zalesky, Allen and Whittle2021a). The stress acceleration theory suggests that early-life adversity accelerates neural development, marked by faster maturation of the cortico-limbic circuits (Callaghan & Tottenham, Reference Callaghan and Tottenham2016; Gee et al., Reference Gee, Gabard-Durnam, Flannery, Goff, Humphreys, Telzer and Tottenham2013a, Reference Gee, Humphreys, Flannery, Goff, Telzer, Shapiro and Tottenham2013b). Despite the widespread influences of adversity on brain, limited longitudinal work has examined the effects of adversity on changes in higher-order brain networks rsFC with results indicating both accelerating (Rakesh et al., Reference Rakesh, Allen and Whittle2023) and delaying effect (Rakesh et al., Reference Rakesh, Kelly, Vijayakumar, Zalesky, Allen and Whittle2021a). The inconsistency highlights the need for additional longitudinal work, especially those with larger sample sizes (Chahal et al., Reference Chahal, Miller, Yuan, Buthmann and Gotlib2022; e.g. Adolescent Brain Cognitive Development Study [ABCD]), to better elucidate how adversity, especially unpredictability, shapes the development of large-scale brain networks implicated in psychopathology.

In summary, this study had four major aims. First, we examined the effects of environmental unpredictability on the changes in rsFC within and between three major higher-order networks – CON, DMN, and FPN from baseline to 2-year follow-up. We hypothesized that rsFC within networks would increase and rsFC between networks would decrease from baseline to 2-year follow-up, with greater unpredictability accelerating these changes based on the stress acceleration theory; however, given prior limited and inconsistent findings, this hypothesis was exploratory. Second, we examined whether neighborhood deprivation moderated the associations between unpredictability and changes in rsFC of brain networks. We hypothesized that unpredictability would be associated with atypical development of network rsFC only when adolescents were also exposed to increased deprivation. As exploratory analyses, we also examined whether sex played a moderated role, as well as the three-way interaction of unpredictability, deprivation, and sex; we did not make specific hypotheses regarding sex differences and three-way interaction due to limited prior research. Third, we examined the mediating effects of changes in rsFC and hypothesized that unpredictability would be associated with atypical development of brain network rsFC, which in turn would increase adolescents' behavioral problems. Fourth, as exploratory analyses, we examined whether neighborhood deprivation moderated the indirect pathways from environmental unpredictability to adolescent behavioral problems through changes in rsFC.

Methods

Participants

Participants were from the ABCD study (https://abcdstudy.org/). The ABCD study is an ongoing longitudinal study that has recruited 11 868 children (9–10 years of age) from 21 study sites across the United States (Casey et al., Reference Casey, Cannonier, Conley, Cohen, Barch and Heitzeg2018). We used baseline, 1-year, 2-year, and 3-year follow-up data from the 5.0 release. After excluding participants who did not have resting-state functional magnetic resonance imaging (rsfMRI) data, or whose rsfMRI data were recommended for exclusion by the ABCD analytic core at baseline and 2-year follow-up, the final sample consisted of N = 5885 participants (see Table 1 for demographic information). All participants provided informed consent or assent, detailed ethics information can be found in Clark et al. (Reference Clark, Fisher, Bookheimer, Brown, Evans, Hopfer and Yurgelun-Todd2018).

Table 1.

Demographic information

Environmental and behavioral data

Environmental unpredictability and threat

We used a subset of items from the Life Events Scale (LES; Grant, Compas, Thurm, McMahon, & Gipson, Reference Grant, Compas, Thurm, McMahon and Gipson2004; Hoffman et al., Reference Hoffman, Clark, Orendain, Hudziak, Squeglia and Dowling2019; Tiet et al., Reference Tiet, Bird, Davies, Hoven, Cohen, Jensen and Goodman1998) to measure environmental unpredictability at 1-year follow-up. LES is a 33-item questionnaire measuring children's stressful life events. We operationalized and measured unpredictability from ancestral cue perspective (Young et al., Reference Young, Frankenhuis and Ellis2020) by using classic items similar to those employed in previous research (see online Supplementary information for all selected 16 items; Belsky et al., Reference Belsky, Schlomer and Ellis2012; Mittal, Griskevicius, Simpson, Sung, & Young, Reference Mittal, Griskevicius, Simpson, Sung and Young2015). If children reported having experienced the stressful event and perceiving it as bad, the score of the degree of disruption was used, where 0 = not at all, 1 = a little, 2 = some, 3 = a lot, otherwise the score was indicated as 0. Then we computed the sum score of the degree of disruption. Therefore, the higher the total score, the greater level of environmental unpredictability the child perceived.

Additionally, we selected four items pertaining to the experiences or witness of severe trauma events to measure perceived environmental threat, consistent with measures employed in previous research (Dennison et al., Reference Dennison, Rosen, Sambrook, Jenness, Sheridan and McLaughlin2019; see online Supplement information for selected items). The higher the total score, the greater level of environmental threat the child perceived.

Neighborhood deprivation

We used the Child Opportunity Index (COI) 2.0 to measure neighborhood deprivation (Noelke et al., Reference Noelke, McArdle, Baek, Huntington, Huber, Hardy and Acevedo-Garcia2020). The COI 2.0 was derived from residential geocodes based on primary address information provided at baseline (Acevedo-Garcia et al., Reference Acevedo-Garcia, McArdle, Hardy, Crisan, Romano, Norris and Reece2014). For detailed information on the computational procedure and ethical considerations, please refer to the online Supplementary information and Fan et al. (Reference Fan, Marshall, Smolker, Gonzalez, Tapert, Barch and Herting2021). The index scores captured neighborhood resources and conditions from three domains: education; health and environment; social and economic. Using the composite score may hide the distinct effect of different types of deprivation, which were revealed in previous studies (Dennison et al., Reference Dennison, Rosen, Sambrook, Jenness, Sheridan and McLaughlin2019). Therefore, we used the weighted average z-scores of each domain and the overall COI (weighted average of three domain averaged z-scores), nationally normed. The lower the z-scores, the greater levels of deprivation adolescents experienced.

Internalizing and externalizing problems

We used the Child Behavior Checklist (CBCL; Achenbach, Reference Achenbach, Kreutzer, Deluca and Caplan2018; Karcher & Barch, Reference Karcher and Barch2021), which measured children's psychopathology and behaviors over the past 6 months. Parents reported at baseline and at 3-year follow-up. Given we examined sex differences and longitudinal changes in this study, we used raw scores rather than sex- and age-corrected T-scores, from the internalizing and externalizing symptom subscales, which were rated on a 3-point Likert-type scale ranging from ‘0 = not true’ to ‘2 = very true’.

Imaging data

Participants underwent neuroimaging scans using standardized protocols across sites (see Casey et al. [Reference Casey, Cannonier, Conley, Cohen, Barch and Heitzeg2018] for detailed imaging procedures). In this study, we used rsfMRI data at baseline and at 2-year follow-up. The pre-processed time courses for each participant were mapped onto the cortical surface, using the standardized ABCD pipeline (Hagler et al., Reference Hagler, Hatton, Cornejo, Makowski, Fair, Dick and Dale2019). Using these time courses, within- and between-network connectivity (Pearson correlation) were computed based on the Gordon functional parcellation (Gordon et al., Reference Gordon, Laumann, Adeyemo, Huckins, Kelley and Petersen2016), and then Fischer z transformed. We focused on the changes in rsFC within and between the three major networks: CON, DMN, and FPN (Fig. 1).

Figure 1.

Three higher-order brain networks in Gordon cortical network parcellation.

Statistical methods

First, to test the associations between environmental predictability and rsFC, we conducted models using cluster robust standard errors (CR-SEs; the highest level – site ID was modeled as clustering variable, TYPE = COMPLEX; McNeish, Stapleton, & Silverman, Reference McNeish, Stapleton and Silverman2017) to account for the clustering structure (i.e. multiple children from the same family and site) in Mplus Version 7.4 (Muthén & Muthén, Reference Muthén and Muthén1998–2014), with network rsFC values as outcomes (in separate models) and environmental unpredictability as the predictor. Second, we examined the moderating roles of neighborhood deprivation and sex on the associations between environmental unpredictability and rsFC. If the moderating effects were significant, Johnson–Neyman method was used to assist further interpretation of moderating effects (Johnson & Fay, Reference Johnson and Fay1950). Furthermore, we examined the mediating roles of rsFC linking environmental unpredictability and adolescents' behavioral problems, using 95% bias-corrected confidence intervals (CIs) with 5000 bootstrap samples in the mediation analysis. We also examined the potential moderating effect of neighborhood deprivation on the indirect paths from environmental unpredictability to behavioral problems via rsFC.

We controlled for threat, sex, race, age, scanner type, and mean framewise displacement when predicting rsFC at 2-year follow-up. We also controlled for rsFC at baseline; therefore, the outcome was residualized to eliminate baseline connectivity effects, leaving only variances attributable to developmental change. In models predicting behavioral problems at 3-year follow-up, we also covaried threat, sex, race, age, and internalizing/externalizing problems at baseline. In all models, full information maximum likelihood estimation was utilized to address missing data in the study variables (please see online Supplementary Table S1 for missing variables, rates, and patterns). We controlled for multiple comparisons using the false discovery rate (p < 0.05).

Results

Correlations between study variables and changes in rsFC

Correlations between all study variables were shown in online Supplementary Fig. S1 (see online Supplementary information). Pair-sample t test indicated that rsFC within networks increased and rsFC between networks decreased from baseline to 2-year follow-up (online Supplementary Table S2).

Associations between environmental unpredictability and changes in rsFC

We found that greater environmental unpredictability was associated with a smaller increase in rsFC within DMN and a smaller decrease in rsFC between CON and DMN (B = −0.042, s.e. = 0.010, p < 0.001, 95% CI [−0.064 to −0.025]; B = 0.035, s.e. = 0.008, p < 0.001, 95% CI [0.021–0.052]; online Supplementary Table S3).

Moderating roles of neighborhood deprivation and sex

We found that neighborhood educational deprivation moderated the associations between environmental unpredictability and rsFC within DMN, FPN, as well as between CON and DMN (B = 0.034, s.e. = 0.012, p = 0.004, 95% CI [0.011–0.057]; B = 0.025, s.e. = 0.008, p = 0.003, 95% CI [0.009–0.041]; B = −0.028, s.e. = 0.010, p = 0.004, 95% CI [−0.048 to −0.009]; online Supplementary Table S4).

Johnson–Neyman plots (Fig. 2) showed that when the level of neighborhood educational deprivation was high, greater environmental unpredictability was associated with a smaller increase in rsFC within DMN and a smaller decrease in rsFC between CON and DMN; when the level of neighborhood educational deprivation was low, however, the association was not significant. We also found that when the level of neighborhood educational deprivation was extremely high, greater environmental unpredictability was associated with a smaller increase in rsFC within FPN; when the level of neighborhood educational deprivation was moderate, the association was not significant; when the level of neighborhood educational deprivation was low, greater environmental unpredictability was associated with a greater increase in rsFC within FPN.

Figure 2.

The moderating effect of neighborhood educational deprivation (ND_E; lower ND_E values indicate higher neighborhood educational deprivation) on the associations between environmental unpredictability (EU) and changes in rsFC within DMN (a), FPN (b), as well as between CON and DMN (c). CON, cingulo-opercular network; DMN, default mode network; FPN, fronto-parietal network.

The moderating effects of other domains of deprivation, overall deprivation, sex, or the three-way interactions among environmental unpredictability, neighborhood deprivation, and sex were not significant (online Supplementary Tables S5–S12).

Mediating roles of changes in rsFC between environmental unpredictability and adolescents' behavioral problems

Based on the significant results of the associations between environmental unpredictability and changes in rsFC, we conducted further analyses to test if rsFC within DMN and between CON and DMN mediated the associations between environmental unpredictability and internalizing as well as externalizing problems, respectively. The results showed that greater environmental unpredictability was associated with a smaller decrease in rsFC between CON and DMN, which predicted more externalizing problems (indirect effect = 0.001, s.e. = 0.000, p = 0.007, 95% CI [0.001–0.002]; Fig. 3 and online Supplementary Table S13).

Figure 3.

The mediating effect of changes in CON-DMN rsFC between environmental unpredictability and adolescents' externalizing problems. CON, cingulo-opercular network; DMN, default mode network; FPN, fronto-parietal network. The parameter estimates in the table are all standardized. *p < 0.05, **p < 0.01, ***p < 0.001.

Moderating role of neighborhood educational deprivation in the mediation model

Considering the moderating effect of neighborhood educational deprivation on the associations between environmental unpredictability and rsFC within DMN, FPN, as well as between CON and DMN, we conducted moderated mediation models to examine whether neighborhood educational deprivation moderated the paths from unpredictability to adolescents' behavioral problems through rsFC of these networks. Results showed that neighborhood educational deprivation moderated the indirect path from environmental unpredictability to internalizing/externalizing problems via rsFC between CON and DMN (moderated mediating effect = −0.001, s.e. = 0.001, p = 0.028, 95% CI [−0.003 to −0.001]; moderated mediating effect = −0.002, s.e. = 0.001, p = 0.023, 95% CI [−0.004 to −0.001]; Fig. 4 and online Supplementary Table S14).

Figure 4.

The moderating effect of neighborhood educational deprivation (ND_E; lower ND_E values indicate higher neighborhood educational deprivation) on the indirect path from environmental unpredictability (EU) to externalizing problems (EP) via changes in CON-DMN rsFC. CON, cingulo-opercular network; DMN, default mode network; FPN, fronto-parietal network.

Particularly, only when the level of neighborhood educational deprivation was high, greater environmental unpredictability was associated with a smaller decrease in rsFC between CON and DMN, which was linked to greater externalizing problems and internalizing problems; however, the result pertaining to internalizing problems became non-significant after handling outliers (see sensitivity analysis), thus, we did not provide further explanations for this result.

Sensitivity analysis

To assess the potential bias introduced by outliers in model estimates, we conducted a sensitivity analysis using the winsorization method. Specifically, observations with values exceeding 4 standard deviations from the mean on any model variables were replaced with the nearest maximum or minimum value (Dixon & Yuen, Reference Dixon and Yuen1974; Hair, Black, Babin, & Anderson, Reference Hair, Black, Babin and Anderson2013). The results were highly consistent with those obtained prior to outlier treatment, however, one moderated mediation model (i.e. environmental unpredictability and neighborhood deprivation interactively predicted internalizing problems via changes in rsFC between CON and DMN) became non-significant after handling outliers (moderated mediating effect = −0.001, s.e. = 0.001, p = 0.030, 95% CI [−0.003 to −0.000]; see online Supplementary Tables S15–S18).