Positive parenting moderates associations between childhood stress and corticolimbic structure

Abstract Childhood stress has a deleterious impact on youth behavior and brain development. Resilience factors such as positive parenting (e.g. expressions of warmth and support) may buffer youth against the negative impacts of stress. We sought to determine whether positive parenting buffers against the negative impact of childhood stress on youth behavior and brain structure and to investigate differences between youth-reported parenting and caregiver-reported parenting. Cross-sectional behavioral and neuroimaging data were analyzed from 482 youth (39% female and 61% male, ages 10–17) who participated in an ongoing research initiative, the Healthy Brain Network (HBN). Regression models found that youth-reported positive parenting buffered against the association between childhood stress and youth behavioral problems (β = −0.10, P = 0.04) such that increased childhood stress was associated with increased youth behavior problems only for youth who did not experience high levels of positive parenting. We also found that youth-reported positive parenting buffered against the association between childhood stress and decreased hippocampal volumes (β = 0.07, P = 0.02) such that youth who experienced high levels of childhood stress and who reported increased levels of positive parenting did not exhibit smaller hippocampal volumes. Our work identifies positive parenting as a resilience factor buffering youth against the deleterious impact of stressful childhood experiences on problem behaviors and brain development. These findings underscore the importance of centering youth perspectives of stress and parenting practices to better understand neurobiology, mechanisms of resilience, and psychological well-being.

values were within the "good" range, with Chronbach's alpha for full APQ (Youth) = 0.875 and Chronbach's alpha for Positive Parenting Subscale (Youth) = 0.815. Internal reliability estimates were also calculated both for the full APQ (Parent Report) and the Positive Parenting Subscale (Parent Report) using the total sample of 485 observations. Chronbach's alpha values were within the "good" range, with Chronbach's alpha for full APQ (Parent) = 0.855 and Chronbach's alpha for Positive Parenting Subscale (Parent) = 0.843.

Youth Self-Report
We operationalized youth problem behavior in two ways, the first being through the Youth Self-Report Total Problems Raw Score. Studies of the YSR's psychometric properties report discriminant and convergent validity among clinical, non-clinical, and diverse samples and acceptable reliability metrics (e.g., mean testretest reliability = 0.79, mean internal consistency = 0.83) (Achenbach et al., 1995). We used the Total Problems Raw Score over the T-score due to literature advising Raw YSR scores to be used in statistical analyses. There is evidence that T scores truncate variability, and that conclusions based on these truncated scores are less reliable (Thurber & Sheehan 2012). The creators of the YSR also advocate for using raw scores in data analyses (Achenbach & Rescorla, 2001).
Internal reliability estimates were calculated for the Youth Self Report Total Raw score using the total sample of 485 observations. Chronbach's alpha values were well within the "excellent" range with Chronbach's alpha for YSR Total score = 0.993.

Strengths & Difficulties Questionnaire
We also considered an alternative measure of youth behavioral problems, the Strengths and Difficulties Questionnaire (SDQ). Specifically, we used SDQ Ttal Difficulties Score as an alternative, caregiver-reported operationalization. The SDQ is a well-validated and widely used instrument with acceptable psychometric properties (i.e., average test-retest reliability = 0.72; average internal consistency = 0.71, average interrater reliability = 0.39) (Goodman, 2001). In the HBN Battery, the 30-item questionnaire was completed by caregivers. The SDQ the following subscales and total score: 1) Emotional Symptoms, 2) Conduct Problems, 3) Hyperactivity/Inattention, 4) Peer Problems, 5) Prosocial Behavior, and 6) a Total Difficulties Score. For additional information on the SDQ, including psychometric properties and language translations, please see: https://www.nctsn.org/measures/strengths-and-difficulties-questionnaire-parent-report.
Internal reliability estimates were calculated for the Strength and Difficulties Questionnaire (SDQ) Total Difficulties score using the total sample of 485 observations. Chronbach's alpha values were well within the "good" range with Chronbach's alpha for SDQ Total Difficulties score = 0.884.

Financial Support Questionnaire (FSQ)
The Financial Support Questionnaire (FSQ) is a measured developed by the Child Mind Institute/Healthy Brain network and measures a variety of socioeconomic-related parameters, including employment status of primary caregiver (i.e., the caregiver at the study visit; FSQ_02; 0 = unemployed, 1 = employed) and employment status of the child's second caregiver (FSQ_03; 0 = unemployed, 1 = employed, 2 = no second primary caregiver). We recoded the second-caregiver employment variable (FSQ_03) so that a response of "no second primary caregiver" was given the value "0". We then summed values from primary caregiver employment and secondary caregiver employment for an overall variable representing parental employment (ie.., 0 = no caregiver employment, 1 = one caregiver is employed, 2 = two caregivers are employed). From the FSQ, we also considered Income (i.e., "what is your annual household income?") and recoded the categorical Income variable by multiplying each value by the midpoint of each income range bracket and then taking the natural log.

Barratt Simplified Measure of Social Status (BSMSS)
The Barratt Simplified Measure of Social Status (BSMSS) is a measure of "Social Status" based on the work of Hollingshead (1957Hollingshead ( , 1975. This measure considers caregiver education level/attainment and occupational "prestige." In analyses considering socioeconomic status as a covariate, we used the Barratt Total score. We also constructed a composite variable that a) selects the higher of the two occupations and education levels between two caregivers (and considers the occupation and education level of the single caregiver, if applicable), b) mean-centered each variable (i.e., took the "z-score"), and c) derived the mean of the two.

Wechsler Intelligence Scale for Children-V (WISC-V)
The Wechsler Intelligence Scale for Children-V (WISC-V) is a widely used test that provides a global index of cognitive performance. The main variable we considered as a measure of intelligence the Full Scale Intelligence Quotient (FSIQ), or a numerical representation of a child's intellectual ability.

MRI Site Information, Data Acquisition, Data Processing, and Qualtity Metrics
The HBN project conducted MRI scans at 4 different sites: Citigroup Biomedical Imaging Center, the City University of New York, Rutgers University, and Staten Island. Eighty-two participants were scanned at the HBN Diagnostic Research Center in Staten Island (SI), 200 scanned at Rutgers University Brain Imaging Center (RU), 186 scanned at the CitiGroup Cornell Brain Imaging Center (CBIC), and 14 scanned at the City University of New York (CUNY). Table S1 depicts the scanning parameters for each scanner. These scans were collected on different MRI scanners, specifically a 1.5 T Siemens Avanto scanner at SI, a Siemens 3T Tim Trio scanner at RU, Siemens 3T Prisma scanner at RU and CBIC, and a Siemens 3T Prisma scanner at CUNY. All structural MRI scans were 3D T1-weighted sequences and specific scan sequence parameters are included in the table below. As discussed in the manuscript, we used linear mixed effects models to account for (MRI-related) site variations given differences in MRI acquisition. Of note, all neuroimaging data used in this study are openly available for download with proper data usage agreement via the International Neuroimaging Data-sharing Initiative (fcon_1000.projects.nitrc.org/indi/cmi_healthy_brain_network/). Table S1. Scan Parameters Table S1. Scanning parameters for each scanner used in the HBN Study. Data obtained from all four scanners are used in this project.

MRI Data Processing
Freesurfer is a widely documented and freely available morphometric processing tool suite (http://surfer.nmr. mgh.harvard.edu/) The technical details of these procedures are described in prior publications Fischl et al., n.d., 2002Fischl et al., n.d., , 2004Fischl, Sereno, Tootell, et al., 1999). Briefly, this processing includes motion correction and intensity normalization of T1-weighted images, removal of non-brain tissue using a hybrid watershed/surface deformation procedure (Fischl et al., 2004), automated Talairach transformation, segmentation of the subcortical white matter and deep gray matter volumetric structures (including hippocampus, amygdala, caudate, putamen, ventricles), tessellation of the gray matter white matter boundary, and derivation of cortical surface area and cortical thickness. Of note, the "recon-all" pipeline with the default set of parameters (no flag options) was used and no manual editing was conducted. In keeping with our past work (Gilmore et al., 2021), Freesurfer outputs were checked via research staff for major errors, and via automated methods (see next section). After successful processing and quality assurance, we extracted volumes for our subcortical structures of interest-the hippocampus and amygdala. We calculated the total volume (for the hippocampus or amygdala) by summing volumes from the left and right hemispheres of each structure. Freesurfer was implemented using Brainlife.io, (brainlife.app.0, https: //doi.org/10.25663/bl.app.0), which is a free, publicly funded, cloud-computing platform for reproducible neuroimaging pipelines and data sharing (Avesani et al., 2019), for additional information, visit http:// brainlife.io/).

Image Quality Metrics
We assessed image quality to exclude particularly high-motion scans and limit the impact of image quality on subcortical volume quantification. Past work from our group (e.g., Gilmore et al., 2021) has found that T1-weighted image quality is related to volumetric measures from commonly used morphometric tools suites (e.g., Freesurfer). We therefore assessed image quality to: 1) Exclude particularly high-motion scans; and 2) limit the impact of image quality on subcortical volume quantification. To assess MRI quality, we generated a quantitative metric ("CAT12 score") using the Computational Anatomy Toolbox 12 (CAT12) (Gaser et al., 2016). This metric considers four summary measures of image quality: noise-to-contrast ratio, coefficient of joint variation, inhomogeneity-to-contrast ratio, and root-mean-squared voxel resolution. CAT12 normalizes and combines these measures using a kappa statistic-based framework. The score is a value from 0 to 1, with higher values indicating better image quality. Additional information is available at: http://www.neuro.uni-jena.de/cat/index.html#QA.

Statistical Analyses
Here, we elaborate on the analytic plan from our manuscript in further detail. We first fit ordinary least squares regression models to examine the interactions of stress and parenting in predicting youth behavioral problems. This involved entering the total scores from the YSR as the dependent variable, and sex, age, site, youth-reported negative life events, youth-reported positive parenting, and an interaction between youthreported negative life events and youth-reported positive parenting as independent variables. All continuous variables were mean-centered. Next, to model relations between variables of interest and deal with potential variations in research sites, we fit linear mixed-effects models (LMEM) using the R package 'lmer' (Bates et al., 2015). For each brain region of interest (i.e., total hippocampal volume, total amygdala volume), we first fit a LMEM with a random effect of site, main effects of sex, age, estimated total intracranial volume, youth-reported negative life events, youth-reported positive parenting, and an interaction between negative life events and youthreported positive parenting. This was motivated by our interest in youth perceptions of parenting behaviors. Notably, we also fit a second set of LMEMs for each brain region of interest again including a random effect of site, main effects of sex, age, eTIV, and youth-reported negative life events, this time including caregiverreported positive parenting and an interaction between youth-reported negative life events and caregiverreported positive parenting. This was to examine whether caregiver perceptions of positive parenting had different associations with brain volumes (compared to youth perceptions). All continuous variables were mean-centered for these analyses. For all significant interaction terms, we then conducted follow-up analyses of the simple slopes of subgroups (the mean of all individuals, +1 SD of the mean, and -1 SD of the mean) using the "interactions" R library (Long, 2019). R code and output for all analyses are included in this document.
Across models, we also tested for differences between any significant (and non-significant) interaction terms. This was done when comparing the interaction results for youth-reported negative life events and youthreported positive parenting (in predicting structural volumes) to interaction results for youth-reported negative life events and caregiver-reported positive parenting. For such comparisons, we used Williams's Test, a statistical comparison used when two correlation coefficients are calculated from a single sample, may be collinear, and not statistically independent (Dunn & Clark, 1971;Williams, 1971). This was done using the "psych" package in R (Revelle, 2022).
Additional sensitivity analyses (presented in the supplemental materials) examine both caregiver and youth reports of negative life events, positive parenting, and youth behavior within our main models; main effect models without interaction terms; four different operationalization of socioeconomic status; the contribution of psychopathology and cognitive ability in smaller overall samples (n = 226 and n = 320, respectively); questions of interest using General Additive Mixed Models (GAMMs) to account for non-linear age effects; models of stress with a non-linear quadratic term in LMEMs and GAMMs; total gray matter volume as an alternative brain scaling variable; left and right hemispheres of brain regions separately; associations between hippocampal volumes and youth behavioral problems; and comparisons of key variables between the HBN imaging sample anad the total HBN sample.    Table of Key Variables   This correlation table can also be viewed on GitHub.   Table S4. M and SD are used to represent mean and standard deviation, respectively. Values in square brackets indicate the 95% confidence interval for each correlation. The confidence interval is a plausible range of population correlations that could have caused the sample correlation. * indicates p < .05. ** indicates p < .01. We computed a two-sample Z-test to analyze a potential difference among the two distributions (i.e., the distributions of age after splitting the sample by biological sex). The Z-test was non-significant (Z-statistic = -1.482, p = 0.138) indicating that the age distributions do not significantly differ by sex.

Regression Diagnostics for Main Manuscript Models
We examined a variety of regression diagnostics including outlier values, QQ plots for studentized residuals, and leverage plots for our main models below (i.e., childhood stress and youth-report parenting predicting youth behavioral problems; childhood stress and youth-report parenting predicting total hippocampal volumes).

Regression Diagnostics: Childhood stress and youth-report parenting on youth-report behavioral problems
Examination of outlier values, QQ plots for studentized residuals, and leverage plots revealed two points that may be influential in the model considering the interaction between (youth report) stress and (youth report) parenting on (youth report) behavioral problems.

Caregiver-Report of Youth Behavioral Problems
In our main manuscript, we detail results of a model considering associations between youth behavioral problems (from the Youth-Self Report [YSR] Total Score) in relation to childhood stress, parenting (youth-report), and the interaction of stress and parenting. This was due to our interest in centering youth perspectives in our research. The results of this model suggested a potential buffering effect of positive parenting within the association between stress and behavioral problems. Put another way. there is no relation between stress and behavioral problems, for children exposed to the highest levels of positive parenting.

Associations between Youth-Report Stress and Youth-Report Parenting on SDQ Total Score
The results of this model suggest a significant interaction of childhood stress and youth-reported positive parenting on caregiver-reported youth problems (i.e., SDQ total difficulties score). This replication of our above result (i.e., with the YSR as an outcome variable instead of the SDQ) underscores that there is an interaction between childhood stress and youth-reported positive parenting on youth behavioral problems regardless of whether the youth or caregiver is the informant on youth behavioral problems.

Associations between Youth-Report Stress and Caregiver-Report Parenting on SDQ Total Score
This next model investigated the interaction of childhood stress and caregiver-reported positive parenting on caregiver-reported youth problems (i.e., SDQ total difficulties score). Consistent with previous results, there is no significant interaction between childhood stress and caregiver-reported positive parenting on behavioral problems.

Socioeconomic Status (SES) Variables
Most studies to date exploring stress and parenting on youth outcomes have operationalized stress as socioeconomic disadvantage, while less research has focused on exposure to stressful and negative life events. In order to test the unique contribution of actual experience of stress on youth behavioral problems and corticolimbic structure and the moderating effect of positive parenting, we consider below various operationalizations of SES as covariates in our models and correlations with study variables.

Correlations between Stress and Multiple Operationalizations of SES
We considered correlations between Childhood Stress (youth-reported Negative Life Events) and four different operationalizations of socioeconomic status, including a) a categorical variable representing parent employment status (Parental.Employment), b) a log-transformed income variable (Income), c) a variable representing the highest parental education achievement and occupational status from the Barratt Simplified Measure of Social Status (BSMSS) (Barratt.Edu.Occ), and d) the Barratt measure total "social status" score (Barratt.Total).
As reflected in the results below, the correlations between stress and SES variables are quite low (ranging from r = -0.0085 to -0.0314) and non-significant at the p < 0.05 level.

Correlations between Parenting and Multiple Operationalizations of SES
We considered correlations between both a) youth-report Positive Parenting and b) caregiver-report Positive Parenting and each of the 4 SES variables described below.
All but one of these 8 correlations were very low and non-significant at the p < 0.05 level, with the exception of the correlation between caregiver-report Positive Parenting and Income (r = -0.1182, p = 0.02486).

Sensitivity Analysis with Socioeconomic Status (SES) as an Additional Variable in Main Models Childhood Stress, Youth-Report Parenting, and SES on Youth-Report Behavioral Problems
We ran models calculating the association between Childhood stress and youth-report parenting on youthreport behavioral problems using 4 different operationalizations of socioeconomic status: a) a categorical variable representing parent employment status (Parental.Employment), b) a log-transformed income variable (Income), c) a variable representing the highest parental education achievement and occupational status from the Barratt Simplified Measure of Social Status (BSMSS) (Barratt.Edu.Occ), and d) the Barratt measure total "social status" score (Barratt.Total).
Analyses using the parent employment status variable (Parental.Employment) led to a reduction in sample size from n = 410 to n = 358. Analyses using the income variable (income) led to a reduction in sample size from n = 410 to n = 313.
Analyses using the variables from the Barratt measure (ie., a variable representing the highest level of parental education attainment and occupation status [Barratt.Edu.Occ] and a variable representing overall social status [Barratt.Total]) preserved our sample size the most, with a reduction in sample from n = 410 to n = 404. In order to preserve as large a sample size as possible, we used the Barratt.Edu.Occ variable as a covariate in all our main analyses.

Childhood Stress, Youth-Report Parenting, and SES on Total Hippocampal Volume
Given the varying sample sizes for the different operationalizations, it is difficult to draw across-the-board conculsions regaring the role of socioeconomic status within our models exploring the interactive effect between childhood stress and positive parenting on youth behavioral problems and total hippocampal volumes. However, in the analyses that preserve the largest sample sizes (i.e., in considering SES variables from the Barratt measure) it appears that socioeconomic status does not explain the significant associations between chilhood stress, youth-reported positive parenting, and smaller hippocampal volumes.

Correlations between stress and IQ
We used the WISC Full Scale Intelligence Quotient (FSIQ) score as an index of cognitive ability. We considered the correlation between FSIQ and stress, finding that the two constructs were not correlated (r = 0.0117, 95% CI [-0.0890, 0.11217], t = 0.22753, df = 378, p-value = 0.8201).

Childhood Stress and Youth-Report Parenting on Youth-Report Behavioral Problems with IQ
The inclusion of WISC IQ score leads to an over 20% reduction in our sample size (from n = 410 to n = 320). We therefore chose not to include IQ as a covariate in our main analyses.

Sensitivity Analysis with Psychopathology Diagnosis as an Additional Variable in Main Models
Given past research noting connections between psychopathology and hippocampal neurobiology, we constructed additional models that included a measure of psychopathology in our models examining stress, parenting, and the interaction of stress and parenting. We used a binary indicator of the presence (or absence) of psychopathology derived from the Kiddie Schedule for Affective Disorders and Schizophrenia (KSADS) Diagnostic Interview. This variable was originally coded as 1= "No", 2= "Yes", 3= "dropped out of study before diagnosis was given". We recoded this variable to exclude datapoints with a value of 3 for the purposes of this analysis. We named this binary diagnosis variable Any.Psychopathology as presented in the following model.
Upon selecting datapoints for which the Any.Psychopathology variable was available, there is a large drop in sample size for these analyses (n=226 vs. n=400-480 in other models). We first replicated our main analysis of interest among this smaller sample (i.e., childhood stress and youth-reported positive parenting predicting total hippocampal volumes) and found that the main effect of childhood stress and the interaction between childhood stress and youth-reported positive parenting is no longer significant in this smaller sample.

Youth-Report Parenting, Stress, and Psychopathology on Total Hippocampal Volume (LMEM)
The binary diagnosis variable (i.e., Any.Psychopathology) was added to our main model predicting total hippocampal volume from an interaction between stress and youth-reported positive parenting, along with our main covariate set. The variable Any.Psychopathology was not significantly associated with total hippocampal volumes in this sample.

Youth-Report Parenting and Stress on Total Hippocampal Volume (GAMM)
In the main manuscript, we used linear mixed effects models to examine relations of interest. However, there may be non-linear effects of age on brain volume. To model these potential patterns of change, we fit Generalized Additive Mixed Models (GAMMs) with package 'gamm4', "Generalized Additive Mixed Models using 'mgcv' and 'lme4' " (v 0.2-6, Woods & Scheipl, 2020). GAMMs allow for modeling of non-linear data by incorporating smooth functions, or splines, and accounting for random effects (e.g., study site) using mixed effects models. This is particularly critical to flexibly capture linear or nonlinear age effects and is in keeping with recent cutting edge work in developmental cognitive neuroscience (Larsen et al., 2020).
For each brain region of interest (i.e., total hippocampal volume, total amygdala volume), we first fit a GAMM with a random effect of site, a smooth (non-linear) term for age, and main effects of sex, estimated total intracranial volume (Total.Intracranial.Vol), negative life events, youth-reported positive parenting, and an interaction between negative life events and youth-reported positive parenting. This was motivated by our interest in youth perceptions of parenting behaviors. Notably, we also fit a second set of GAMMs for each brain region of interest again including a random effect of site, a smooth term for age, and main effects of sex, total intracranial volume, and Youth-Reported negative life events, this time including caregiverreported positive parenting and an interaction between youth-reported negative life events and caregiverreported positive parenting. This was to examine the importance of youth (as opposed to caregiver) reports of parenting. The results of this GAMM are consistent with the LMEM reported in the manuscript; namely, that there is a significant interaction between stress and youth-reported positive parenting on total hippocampal volume.

Caregiver-Report Parenting and Stress on Total Hippocampal Volume (GAMM)
The results of this GAMM are again consistent with the LMEM reported in the manuscript; namely, that using caregiver-reported positive parenting (instead of youth-reported positive parenting) results in a nonsignificant interaction between stress and positive parenting on total hippocampal volume. This supports our claim that centering youth-reports of parenting behaviors are important when considering the effects of stress and hippocampal development.

Youth-Report Parenting and Stress on Total Amygdala Volume (GAMM)
The results of the next two GAMMs are consistent with the LMEs invesgiating total amygdala volume. Neither youth-reported positive parenting nor caregiver-reported positive parenting significantly interact with stress to predict total amygdala volume.

Probing Questions of Interest Using a Quadtratic Stress Term to Account for Potential Non-Linear Stress Effects
We ran a series of GAMMs modeling stress as a quadratic variable and a spline/smooth function applied to the interaction between stress and age. For total amygdala volume, as well as right and left amygdala volume, we fit a GAMM with a random effect of site, a quadratic stress (i.e, negative life events) term, main effects of sex, estimated total intracranial volume (Total.Intracranial.Vol), youth-reported positive parenting, and a a smooth (non-linear) function applied to the interaction between negative life events and age. We also ran 3 additional models with the same DVs (ie.., total, left, and right amygdala) but instead using total Gray Matter Volume as the brain scaling covariate. Results indicate non-significant effects of stress and the interaction between stress and age. Notably, cross-sectional data is not well-suited to interrogate effects of interest (i.e., effects between stress and age) given that complex non-linear relations between amygdala structure and stress experiences play out over time.

Investigating Total Brain Region Volumes with Total Grey Matter Volume as an Alternative Covariate to Estimated Total Intracranial Volume
Our main models in the manuscript included estimated total intracranial volume (eTIV /Total.Intracranial.Vol) as our brain covariate to control for individual differences in brain size. We ran sensitivity analyses with an alternative measure of brain size, Total Gray Matter Volume (Total.Gray.Vol) for each of our main LMEMs. Overall, we found that the inclusion of Total.Gray.Vol versus Total.Intracranial.Vol does not significantly impact model results.
In parallel to the main models presented in the manuscript, there remained a significant interaction between youth-reported parenting and stress on changes in total hippocampal volume when Total.Gray.Vol was included as the brain scaling variable instead of Total.Intracranial.Vol. These sensitivity analyses underscore that the significant interaction between stress and youth-reported positive parenting on hippocampal volumes is consistent even when accounting for individual differences in brain size in various ways. As in our main models, there was not a significant association between the interaction of caregiver-reported parenting and stress on hippocampal volume when Total.Gray.Vol was included as the brain scaling variable instead of Total.Intracranial.Vol. These findings continue to highlight importance of centering youth reports to understand associations between stress, parenting, and the hippocampus.
Aligned with the main models, the inclusion of Total.Gray.Vol instead of Total.Intracranial.Vol in models with the amygdala did not change results (i.e., associations remained non-significant).

Investigating Left and Right Hemispheres of Brain Regions Separately
Our main models in the manuscript investigated the interactions between childhoos stress and both youthand caregiver-reported positive parenting on total hippocampal and amygdala volume. The total volume variables were derived by adding the volumes for the left and right hemispheres of the hippocampus and amygdala, respectively. We ran sensitivity analyses investigating the interactions between childhood stress and both youth-and caregiver-reported positive parenting on each the left and right hemisphere of the hippocampus and amygdala. Briefly, we found that the interaction between stress and youth-reported positive parenting was significant for the left and right hippocampus, and was not significant for either the left or right amygdala. Consistent with the pattern of results reported in our manuscript, the interaction between childhood stress and caregiver-reported positive parenting was not significant for the left hippocampus, right hippocampus, left amygdala, or right amydala.
These analyses underscore that, when centering the youth's perspective of positive parenting, there is a consistent buffering effect of positive parenting on the association between stress and hippocampal volumes, whether that be total volume, left hemisphere, or right hemisphere. Our models do not provide evidence for an analogous effect for the amgydala.

Left Hippocampus
Observations 482

Examining Associations Between Hippocampal Volumes and Youth Behavioral Problems
Our previous results found significant interactions between stress and positive parenting (from the youth perspective) on left, right, and total hippocampal volume. We lastly investigated the association between total hippocampal volume and youth behavioral problems. We considered both YSR Total Score (Youthreported behavioral problems) and SDQ (caregiver-reported behavioral problems) as dependent variables in separate models.
We ran analyses considering YSR Total Score as linear mixed-effects models (LMEM) and as linear models (LM) due to issues with singularity for YSR Total Score in a LMEM. Across both types of models, results were consistent, with a non-significant relationship between total hippocampal volume and the YSR. We also examined associations with left and right hippocampal volumes, finding a significant association youth behavioral problems (YSR) and left hippocampal volumes.  -values (p-val). Random effects include the random effect of Imaging Site and the number of grouping variables (4 imaging sites). Model output also depicts overall model fit statistics and number of observations.

Total Hippocampal Volumes on Youth Behavioral Problems (Caregiver-Report/SDQ) as a linear mixed-effects model
We did not find a significant relationship between SDQ total score and total hippocampal volume.

Left Hippocampus on Youth Behavioral Problems (Caregiver-Report/SDQ) as a linear mixedeffects model
The association between the left hippocampus and total SDQ score was significant, indicating that increased youth behavioral problems are significantly associated with decreased volume in the left hippocampus. These results are visualized after the model output below. We did not find parallel results for the right hippocampus.

Comparing Key Variables between HBN-Imaging Sample to HBN-Total Sample
This data is drawn from the Healthy Brain Network (HBN) study, a large-scale, publicly-available dataset with measures from over 4,000 children across New York City. Importantly, not all participants have completed all measures. The total amount of datapoints for relevant study variables ranges from 482 to 3,437 depending on a) whether participants were part of the imaging protocol and b) missingness on study variables.
We ran missing data analyses comparing the subsample with imaging data (N = 482) to the larger HBN sample on the following key study variables: Youth-Reported Negative Life Events, Youth-Reported Positive Parenting, Caregiver-Reported Positive Parenting, and Youth Self-Report scores. We created a binary indicator variable imaging that we used as the independent variable in all below analyses to test for any differences between the imaging subsample and larger HBN sample (0 = not part of imaging sample, 1 = part of imaging sample).
For more information on the Healthy Brain Network or to get access to data, visit this site: http://fcon_ 1000.projects.nitrc.org/indi/cmi_healthy_brain_network/