Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Jan 1.
Published in final edited form as: Child Abuse Negl. 2021 Dec 6;123:105369. doi: 10.1016/j.chiabu.2021.105369

Executive functions as a mediator of childhood maltreatment on adult psychopathology: A longitudinal mediation analysis comparing maltreatment factor models

Kyle R Kent a,*,1, Christian J Hopfer b, Robin P Corley c, Michael C Stallings a,c
PMCID: PMC8859768  NIHMSID: NIHMS1750691  PMID: 34883422

Abstract

Background:

Childhood maltreatment (CM), executive functions (EFs), and psychiatric disorders all correlate highly. Changes in EFs during adolescence related to CM present a possible mediating mechanism for the development of psychiatric disorders, yet no study has analyzed this longitudinally while comparing predictive capacity of different CM factor structures. We hypothesized that changes in EFs from adolescence to adulthood would mediate, in part, associations between CM, internalizing disorders (INT), and anti-social personality disorder (ASPD) while different subtypes of CM would differentially predict INT and ASPD.

Objective:

This study longitudinally examined the mediating effects of EFs on associations between CM, INT, and ASPD while comparing prediction of two CM factor structures.

Participants:

High-risk subjects selected for drug use in adolescence (N = 658) from mean ages 16 to 23.

Methods:

A Bayesian structural equation model was deployed to analyze change in EFs as a mediator of the relationship between CM and adult INT and ASPD. CM was measured using two factor structures: a single overall factor and four correlated factors representing CM subtypes.

Results:

CM significantly predicted INT and ASPD but there was no evidence that the relationship was substantially mediated through EFs. High correlations among subtypes of CM limited the unique predictions of each subtype on INT and ASPD.

Conclusion:

In this high-risk sample, the collinearity of CM subtypes obscured their predictions of outcome measures supporting the use of one CM factor. EFs did not significantly mediate associations between CM and psychiatric disorders, but further research on these relationships is warranted.

Keywords: Childhood maltreatment, Executive functions, Internalizing disorders, Bayesian structural equation model, Mediation

1. Introduction

During important developmental periods a substantial portion of children are exposed to different forms of childhood maltreatment (CM). A meta-analytic review of CM found that the median prevalence of neglect and sexual abuse in North America is 29% and 18% respectively (Moody et al., 2018), exposing almost a quarter of the population to events with potentially life-long deleterious effects (Rouse & Fantuzzo, 2009). CM is a difficult concept to operationalize and was, up until recently, only studied in its more physical manifestations (e.g., sexual or physical abuse; Dinwiddie et al., 2000; Stoltenborgh et al., 2015) despite other CM subtypes often co-occurring (Assink et al., 2019; Beal et al., 2019). When multiple subtypes of CM are included in a study (e.g., neglect, psychological, physical, and sexual abuse), they are often aggregated as a single sum of events or an overall factor (Felitti et al., 1998; Lau et al., 2005). In doing so, researchers lose the ability to analyze the potential unique effects of subtypes on life-outcomes. To adequately measure CM, we must first decide which approach, separating subtypes or combining them as a single factor, is most appropriate. Studies comparing measurement models show that either can be useful (Brumley et al., 2019; Lau et al., 2005; Warmingham et al., 2019). More analyses comparing CM models to predict life-outcomes are needed.

During brain maturation the development of neurological structures are vulnerable to the deleterious effects of stress (McCrory et al., 2017) with the greatest risk arising during developmentally sensitive periods (Teicher et al., 2003). Some of the last brain regions to develop are the frontal cortices (Gogtay et al., 2004) which underlie higher-order cognitive functioning (Pechtel & Pizzagalli, 2011). Those cognitive functions, specifically executive functions (EFs; e.g., inhibitory control, mental flexibility, attention), are thought to develop during growth spurts, like in late adolescence (Anderson et al., 2014), posing a risk for stressful events to impair EF development. In fact, the key brain regions that underlie EF, like the prefrontal cortex (Marsh et al., 2008), are disrupted in patients with psychiatric disorders (Langenecker et al., 2007) outlining a potential mechanism through which CM affects psychopathological development. This mechanism has been hypothesized as a latent vulnerability to CM (a stressful event) that causes maladaptive shifts in cognitive functioning that, over time, increase the risk of psychopathology (McCrory & Viding, 2015). The latent vulnerability is not manifested as a psychiatric disorder immediately but can be detected in changes in cognitive abilities that then mediate the relationship of CM to psychopathology. It is clear that these variables (CM, EFs, psychopathology) correlate cross-sectionally (Barzilay et al., 2019), but our understanding of their relationships must go beyond correlations. Analyzing them longitudinally is key to understanding the potential mechanistic process through which CM affects adult psychiatric disorders.

In this study, we analyzed a longitudinal high-risk sample selected for substance abuse and antisocial behavior in adolescence. We utilized measures assessed at two time-points (mean age 16 and 23) and analyzed changes in EF (assessed via performance IQ, mental flexibility, and task-shifting) as mediators, in part, of the relationship between CM, internalizing disorders (INT; depression, anxiety), and antisocial personality disorder (ASPD). Measurement models of CM were assessed using two models: as a single CM factor or four correlated factors of neglect, psychological, physical, and sexual abuse. We hypothesized: 1) that there would be a significant total and direct effect of CM on INT, 2) that changes in EFs would mediate, in part, these relationships (indirect effect), and 3) that CM factor subtypes would differentially predict INT. ASPD was added as an outcome to study psychopathology more hollistically.

2. Methods

2.1. Participants

The data in this study came from the Genetics of Antisocial Drug Dependence (GADD), a longitudinal study in its third assessment wave, that began ascertaining subjects between 2001 and 2007. The first two waves of GADD were used in this study since EF assessment ended after wave-2; data collection for wave-2 ended in 2014. The high-risk probands are a selected sample (selected for drug abuse and antisocial behavior at wave-1 during adolescence) identified through outpatient treatment programs in the Denver metropolitan area and the Colorado juvenile justice system (Kamens et al., 2013). All probands were age 13 to 19 at wave-1. Siblings of the probands who were age ≥ 12 were also recruited into the study. Subjects were excluded from participation if they exhibited current psychotic symptoms, were an imminent danger to self or others, or had an IQ < 70. Siblings older than 21 at wave-1 were also excluded. All subjects were paid $100 for participation at each timepoint, and all research protocols were approved by the Institutional Review Boards of the University of Colorado, Boulder and the University of Colorado, Denver. Note: some GADD subjects were recruited in San Diego, but they were not included in the current study because the abuse and neglect assessment was not administered to them.

In this study, subjects (N = 658) participated in two waves of data collection. At wave-1, there were 438 males (ages = 12–20, M = 15.95, SD = 1.72) and 220 females (ages = 12–20, M = 16.20, SD = 1.82). At wave-2, subjects were approximately 7 years older on average (males = 17–31, M = 22.56, SD = 2.45; females = 17–30, M = 22.54, SD = 2.40). Demographically, the sample is majority white (60%), Hispanic (16%), African American (10%), and 14% other.

2.2. Measures

2.2.1. Psychiatric disorders

The Computerized Diagnostic Interview Schedule IV (C-DIS-IV; Robins et al., 2000; Major Depressive Disorder and Generalized Anxiety) and the Diagnostic Interview Schedule for Children (age < 16; Shaffer et al., 2000) were used to measure Major Depressive Disorder, Generalized Anxiety Disorder, and Anti-social Personality Disorder based on DSM-IV symptoms. Diagnostic criteria were used to convert scores into a 0, 1, 2 scale (0 = no symptoms, 1 = some symptoms but no diagnosis, 2 = diagnosis). The Center for Epidemiological Studies Depression scale (CES–D; Radloff, 1977) was used as another measure of depression. The CES-D measures depressive symptomatology through a 20 question self-report format. Each question pertains to a depressive symptom; however, some items are reverse coded for scoring consistency. Subjects rated their symptoms on a scale assessing the frequency of occurrence of each symptom from 0 to 3 (0 = rarely or none of the time, 3 = most or all of the time). The responses for each subject were summed and converted to a 0 (little to no symptoms) or 1 (some symptoms, possibly warranting diagnosis) based on a conservative threshold of 18 defined by Hedayati et al. (2006; 69% sensitivity and 83% specificity).

2.2.2. Childhood maltreatment

CM was measured at wave-2 using a retrospective self-report measure called the Colorado Adolescent Rearing Inventory questionnaire (CARI-Q; Crowley et al., 2003). The questionnaire asks subjects about their upbringing prior to the age of 18 with questions relating to four main domains of childhood abuse and neglect (neglect, psychological, physical, and sexual abuse). Each domain was assessed using five items (20 total) with each question scored as a binary yes/no endorsement. Items begin with “During childhood (before age 18)…” and ask about a specific event, for example: neglect, “… you often went hungry or didn’t have necessary clothing”; psychological, “… adults you lived with yelled, cursed, or said threatening things to you”; physical, “… did an adult ever hit you with something that injured you or left a mark such as a bruise or a cut”; and sexual abuse, “… did an adult or someone older than you rub their body on you through clothing in a sexual way”.

2.2.3. Executive functions

EF Stroop and Trails measures were assessed at both wave-1 and wave-2 for all subjects recruited via substance abuse treatment facilities in Denver. However, subjects recruited through the Colorado Juvenile Justice were administered Stroop and Trails at wave-2 only. Further, both the Verbal and Performance IQ subscales (WISC/WASI) were administered at wave-1 but only the Performance subscales at wave-2. Thus, there is a degree of missingness with the EF measures. Further explanation of this structural missingness is discussed in the supplementary materials and the percentage of missingness for each variable is provided in Table 1. Since this missingness is structural rather than due to sample attrition, we consider the missing data missing at random and our approach using Bayesian estimation efficiently handles this missingness.

Table 1.

Descriptive statistics of continuous variables and frequencies of categorical variables by sex.

Executive functions
Male Female Missing
M SD M SD
Wave-1
 Stroop 35.68 11.03 38.13 12.24 47%
 Trails 41.49 17.79 39.63 17.05 47%
 P-IQ*** 50.95 8.99 47.71 9.75 7%
Wave-2
 Stroop 37.27 12.28 38.27 12.15 5%
 Trails*** 42.07 24.75 34.8 16.99 5%
 P-IQ*** 55.61 8.14 51.98 8.65 41%
Means of CM
Male Female Missing
M SD M SD
Neglect*** 0.62 0.99 0.88 1.27 1%
Psychological** 0.48 0.88 0.77 1.25 2%
Physical* 0.24 0.67 0.42 0.94 1%
Sexual*** 0.08 0.48 0.41 1.04 2%
Overall CM*** 1.42 2.08 2.48 3.62 1%
Prevalence of psychiatric disorder categories
Male Female Missing
0 1 2 0 1 2
Wave-2
 ASPD*** 26% 47% 27% 42% 46% 11% 0%
 GAD 85% 11% 4% 82% 10% 8% 0%
 MDD** 71% 14% 14% 59% 15% 25% 0%
 CES-D 65% 35% 60% 40% 0%
Open in a new tab

Presented here are the means and variances of the continuous variables and the prevalence of the categorical variables by sex. In the study, Child Maltreatment (CM) measures are considered binary, but for the presentation of their descriptive statistics the sums of each CM subtype and the overall CM are used. For GAD (generalized anxiety disorder), MDD (major depressive disorder), and ASPD (anti-social personality disorder), 0 = no symptoms, 1 = some symptoms, 2 = diagnosis. For CES-D (Center for Epidemiological Studies Depression scale), 0 = none to a few symptoms, 1 = enough symptoms to warrant a diagnosis. The “Missing” column is the percentage of participants who did not complete at least one question of the given measure. Sex differences were tested using Student’s t-tests and χ2 contingency table tests; results of these tests are indicated by the asterisk next to each variable’s name.

*

p < .05.

**

p < .01.

***

p < .001.

The Wechsler Abbreviated Scale of Intelligence III (WASI-III; Wechsler, 1999) and the Wechsler Intelligence Scale for Children III (WISC-III; Wechsler, 1991) were administered to operationalize performance IQ (P-IQ) at wave-1 and wave-2. The P-IQ subscale is a measure of perceptual organization and processing speed (Strauss et al., 2006) which was scored as age normed T-scores with a higher score indicating better functioning.

The Stroop Color and Word test (Stroop) assesses one’s ability to task shift and avoid or minimize interference in the cognitive processing of stimuli that contradict each other (Golden & Freshwater, 2002; Stroop, 1935). It was also shown to be a measure of attention, processing speed, cognitive flexibility (Jensen & Rohwer, 1966), and working memory (Kane & Engle, 2003). Further detail on Stroop administration and assessment is discussed in Friedman et al. (2008). In brief, it was scored by subtracting the averaged time spent on the initial two congruent/control word tasks from the incongruent/test word task. For ease of interpretation of the EF variables, we made higher scores indicate better performance. So, for a timed measure like Stroop (where a good score typically means lower values) the signs of the Stroop scores were flipped so that a higher score indicates better functioning.

The Trail Making Test A and B (Trails; Reitan & Wolfson, 2004) is considered to be a measure of mental flexibility. The task has two parts. In part A, the participants are shown 15 circles labeled with numbers and must draw a line between them in order (1–2-3-etc.). Part B is complicated by adding letters so participants must connect each circle in order, switching between numbers and letters (1-A-2-B-etc.). Each trial is timed, and the variable used here was the difference between completion time for test B and A, which we refer to as Trails. The same logic of reversing the signs of a timed measure to make higher scores indicate better performance was applied to Trails as it was to Stroop. Therefore, a higher Trails score indicates better functioning.

2.2.4. Parental educational attainment

Parental educational attainment was obtained via self-report by subjects at wave-2 and scored on a 0 to 9 scale (0 = none 9 = doctorate) similarly to Hollingshead (1975). We scored parental education as the mean of paternal and maternal educational attainment when data for both were present.

2.3. Procedures and design

All subjects were administered a battery of structured psychiatric interviews, cognitive tests, and self-report questionnaires pertaining to a range of information from demographics to drug use. In the first wave, probands were assessed in treatment facilities while their siblings were tested in their homes. At the second wave, both probands and their siblings were tested in their homes. The average time to complete the assessment battery was 2 to 3 h. Data were cleaned and prepared using R studio with all primary analyses being conducted in Mplus (version 8.4; Muthén & Muthén, 2017).

2.4. Statistical analysis

Due to concerns over missing EF data, many estimation techniques were considered. A Bayesian structural equation model (BSEM; Muthén & Asparouhov, 2012) approach was selected. Further discourse concerning our choice of estimation technique and validation of the Bayesian model (using the WAMBS checklist; Depaoli & de Schoot, 2017) is discussed in the supplementary materials. In brief, BSEM can handle complex models with a high degree of integration points, non-normal distributions, and missing data similarly to frequentist Maximum-Likelihood approaches (Muthén & Asparouhov, 2012; Yuan & MacKinnon, 2009). This makes BSEM an ideal candidate for a large mediation model with missing and categorical data. For all analyses, Bayesian estimation was used with a Markov Chain Monte Carlo algorithm and the Gibbs sampler. To estimate the latent factor model, Exploratory and Confirmatory Factor Analyses (EFAs; CFAs) were conducted. Model fit was assessed using the Posterior Predictive p-value (PP p; Asparouhov & Muthén, 2010a) which is akin to a χ2 fit using likelihood ratio tests. Bayesian estimation in Mplus does not produce estimates of model fit like CFI/TLI or RMSEA when categorical variables are present. To compensate, Wald χ2 tests were utilized to compare model constraints on nested models (Asparouhov & Muthén, 2020) along with comparing correlations of equal variance factors through a technique recommended by Asparouhov (2020).

Once measurement models were confirmed for latent CM constructs (an overall CM factor or four CM subtypes) then the same was done for latent dependent variables (EFw1, EFw2, INT). ASPD was the only externalizing variable available, so it was treated as an observed dependent variable. All variables were included in two Model structures along with the covariates age at each wave, sex, and parental educational attainment. Model 1 (Fig. 1A) consists of all aforementioned variables but used the single CM factor. Model 2 was the same but used four correlated factors representing each of the CM subtypes (Fig. 1B). PP p was used to estimate significance of each path, but credibility intervals (CI, the Bayesian equivalent of confidence intervals) were used as the primary measure of significance.

Fig. 1.

Fig. 1.

Open in a new tab

Shows the underlying structural equation model for the mediation analysis of Model 1 and 2 along with path coefficients. Boxes around variable names indicate observed variables, circles indicate latent variables. P-IQ = Performance Intelligence Quotient, CARI-Q = the Colorado Adolescent Rearing Inventory Questionnaire, which is our maltreatment measure, CM = Child Maltreatment, NEG = neglect, PSYC = psychological abuse, PHYS = physical abuse, INT = internalizing disorders, w1 = wave-1, w2 = wave-2. Covariates of sex and age at each wave were included in the models but are not shown so the figures are not cluttered. The dots between CARI-Q 1 and CARI-Q 20 represent all the CARI-Q questions which weren’t able to be shown. The loadings of the four neglect and abuse factors are replaced by their correlations for visualization purposes. NEG takes CARI-Q 1–3, PSYC takes CARI-Q 4–10, PHYS takes CARI-Q 11–15, and SEX abuse takes CARI-Q 16–20. Solid lines and bold font mean p < .001. Dashed lines and regular font mean the path was not significant.

Both Model 1 and Model 2 analyzed the mediation effects of changes of EFs from wave-1 to wave-2 on the relationships between CM and adult psychiatric disorders. A longitudinal model was selected in hopes of capturing change in EFs influenced by a latent vulnerability developed after CM as hypothesized here and by McCrory and Viding (2015). Although the age period of wave-1 EF assessment (mean age 16) cannot control for changes in EFs experienced in early childhood, the brain is still developing and would be susceptible to this latent vulnerability. Furthermore, CM is often not a discrete one-time event which allows for clearly defined periods of pre- and post-abuse, rather it is often chronic and experienced heterogeneously. Identifying subjects who experienced maltreatment before or after EF wave-1 assessment using age-of-onset would go against our understanding of CM as a phenomenon.

3. Results

3.1. Descriptive statistics

Table 1 provides the means, standard deviations, tests of sex differences, and degree of missingness for all observed variables at both waves. Although there appear to be some modest mean differences between self-identified males and females for the EF measures, significant sex differences were found only for P-IQ at wave-1 and wave-2 and Trails at wave-2. For maltreatment (middle portion of Table 1), the mean endorsements and standard deviations for the abuse subtypes and overall total number of the CM endorsed are shown. Although not shown, 52% of the subjects endorsed experiencing at least one instance of CM while 25% said they experienced three or more. Student’s t-tests tested sex differences for the CM subtype sums and total sum. Self-identifying women scored significantly higher on measures of neglect, sexual abuse, and total CM (with trending mean differences for psychological and physical abuse), indicating that, on average, women experienced more abuse and neglect than men. Females appeared to report significantly higher levels of MDD while males reported significantly higher ASPD. There were no differences in GAD and CES-D scores. Given the number of tests conducted, we used a conservative Bonferroni correction (0.05/15 ≈ 0.0033) to determine significance.

3.2. Measurement model fit

3.2.1. Childhood maltreatment

Measurement models (i.e., factor structures and model fits) were explored using Exploratory and Confirmatory Factor Analyses (EFA; CFAs). We considered two models for CM: 1) a one-factor model based on the common simple aggregation of maltreatment experiences, and 2) a four factor model based on the purported scoring of the CARI-Q (with one exception, see below) which posits four subdimensions of maltreatment: neglect, psychological, physical, and sexual abuse. EFA of the 20 maltreatment items empirically supported both the 1-factor and 4-factor models. Bayesian CFAs indicated that both models provided good fit to the data (1-factor model, PP p = .21; 4-factor model, PP p = .52 – CFI and RMSEA are not available for categorical variables in Bayesian estimation). Although a 4-factor model was efficacious, the final 4-factor model did not exactly follow the purported scoring of the CARI-Q. EFAs consistently showed that two of the neglect items (adults who cared for you were not warm and caring; or withdrew from you) loaded more strongly with the psychological abuse items than with the other neglect items (were there times when you went hungry, no family/household rules, left alone when younger than 12 with no supervision); all other items loaded on their expected subdimensions. Thus, the final 4-factor model allowed these two items to load on the psychological abuse latent factor, not the neglect latent factor. Factor loadings for the physical and sexual abuse subdimensions were consistent with the scoring of the CARI-Q.

3.2.2. Executive functions (EF)

Two CFAs were conducted to compare the factor loadings of Stroop, Trails B, and P-IQ. A single factor at wave-1 (EFw1) and a single factor at wave-2 (EFw2) both showed good fit to the data (EFw1, PP p = .37, CFI = 1.00, RMSEA <0.01; EFw2, PP p =.54, CFI = 1.00, RMSEA <0.01).

3.2.3. Internalizing disorder

A CFA confirmed the measurement model for INT at wave-2. MDD, GAD, and CES-D loaded on a single factor of INT providing a good fit to the data (PP p = .49 – CFI and RMSEA are not available for categorical variables in Bayesian estimation).

3.3. Mediation model results

Model 1 (Fig. 1A). The first model was designed to predict the effects of one overall maltreatment factor on INT and ASPD directly and when mediated by EFs. The goal of the analysis was to determine the extent to which the relationship between CM and psychopathology could be explained, in part, through changes in EF over time. Fig. 1A presents the standardized path coefficients between factors after controlling for covariates (age, sex, parental education). Only three paths (noted in bold print and solid lines) were significant: the direct effect of CM on INT, CM on ASPD, and the stability coefficient between EF at wave-1 and EF at wave-2. The final standardized results and the credibility intervals are shown in Table 2 and the mediation results in Table 3. CM significantly predicted INT and ASPD while EFw1 predicted EFw2. EFw1 had a small relationship (Coeff. = 0.180) with ASPD which had a PP p < .05 but the CIs include 0 indicating it is non-significant. This is likely explained by the high standard error (0.105) due to missing EF data. The total and direct effect of CM on INT and CM on ASPD was significant, but the total indirect effect and the specific indirect effects were not.

Table 2.

Standardized path coefficients for Model 1 and Model 2.

Model 1 regression paths
Coeff. S.E. PP p Lower CI Upper CI
INT
 CM 0.508 0.054 < 0.001 0.394 0.604
 EFw1 0.095 0.111 0.200 − 0.124 0.306
 EFw2 − 0.140 0.093 0.069 − 0.319 0.048
 Agew2 0.009 0.017 0.301 − 0.024 0.040
 Sex − 0.302 0.112 0.004 − 0.518 − 0.080
 Parent Ed. 0.009 0.022 0.338 − 0.033 0.052
ASPD
 CM 0.351 0.061 < 0.001 0.232 0.475
 EFw1 0.180 0.105 0.042 − 0.027 0.390
 EFw2 − 0.023 0.087 0.399 − 0.195 0.147
 Agew2 − 0.039 0.018 0.013 − 0.074 − 0.005
 Sex 0.662 0.105 < 0.01 0.457 0.866
 Parent Ed. − 0.022 0.020 0.137 − 0.061 0.016
EFw1
 CM 0.072 0.062 0.120 − 0.047 0.193
 Agew1 0.098 0.029 < 0.001 0.039 0.149
 Sex 0.172 0.120 0.077 − 0.067 0.402
 Parent Ed. 0.029 0.022 0.096 − 0.014 0.071
EFw2
 CM − 0.012 0.051 0.402 − 0.111 0.088
 EFw1 0.292 0.058 < 0.001 0.173 0.401
 Agew2 0.021 0.017 0.127 − 0.014 0.054
 Sex − 0.274 0.093 0.002 − 0.456 − 0.091
 Parent Ed. 0.003 0.019 0.430 − 0.034 0.041
Model 2 regression paths
Coeff. S.E. PP p Lower CI Upper CI
INT
 Neglect 0.729 0.829 0.178 −1.350 2.075
 Psychological 0.585 0.785 0.224 −1.223 1.984
 Physical − 0.541 0.650 0.162 −1.669 1.144
 Sexual − 0.320 0.404 0.171 −1.031 0.688
 EFw1 0.074 0.111 0.259 − 0.145 0.287
 EFw2 − 0.120 0.094 0.106 − 0.301 0.070
 Agew2 0.009 0.020 0.313 − 0.030 0.051
 Sex − 0.316 0.111 0.003 − 0.527 − 0.094
 Parent Ed. 0.010 0.022 0.323 − 0.033 0.052
ASPD
 Neglect 1.595 1.742 0.152 − 2.092 4.965
 Psychological 1.123 1.347 0.177 −1.519 3.968
 Physical −1.660 1.336 0.084 − 4.504 1.008
 Sexual − 0.899 0.887 0.108 − 2.814 0.840
 EFw1 0.213 0.151 0.068 − 0.068 0.535
 EFw2 − 0.001 0.122 0.497 − 0.239 0.248
 Agew2 − 0.048 0.023 0.008 − 0.098 − 0.007
 Sex 0.778 0.174 0.003 0.505 1.199
 Parent Ed. − 0.026 0.026 0.151 − 0.080 0.023
EFw1
 Neglect 0.022 0.308 0.471 − 0.620 0.641
 Psychological − 0.020 0.304 0.474 − 0.639 0.585
 Physical 0.054 0.277 0.413 − 0.478 0.622
 Sexual 0.029 0.166 0.422 − 0.286 0.366
 Agew1 0.091 0.034 < 0.001 0.039 0.176
 Sex 0.168 0.119 0.080 − 0.072 0.398
 Parent Ed. 0.029 0.022 0.099 − 0.015 0.072
EFw2
 Neglect 0.013 0.241 0.476 − 0.473 0.498
 Psychological − 0.049 0.240 0.411 − 0.542 0.434
 Physical 0.029 0.217 0.442 − 0.405 0.470
 Sexual − 0.004 0.131 0.486 − 0.261 0.263
 EFw1 0.288 0.059 < 0.001 0.167 0.400
 Agew2 0.014 0.022 0.267 − 0.027 0.057
 Sex − 0.279 0.094 0.002 − 0.459 − 0.091
 Parent Ed. 0.002 0.019 0.453 − 0.035 0.040
Open in a new tab

Presented are the path coefficients of the independent variables (indented names) on the dependent variables (header above the indented names) for Model 1 and Model 2. Coeff = the path coefficient or the slope of that variable in the model. S.E. = the standard error of the posterior distribution which is a likelihood distribution of the point estimate for the given parameter (or coefficient). PP p = the Posterior Predictive p−value, which is akin to frequentist p-value for any test statistic, is obtained through draws from the posterior distribution. Lower and Upper CI = the credibility intervals of the parameter estimates which is based on 2.5% to 97.5% intervals of the posterior distribution. w1 = wave-1, w2 = wave-2, EF = executive functions, INT = internalizing disorders, ASPD = anti-social personality disorder, CM = child maltreatment which is a single factor composed of its four abuse and neglect subtypes.

The far left column presents both the independent and dependent variables. The initial variable which is underlined is the dependent variable being regressed onto by the preceding independent variables (indented).

Table 3.

Standardized mediation results for Model 1 and 2.

Model 1
Predictor Outcome Effect Coeff. S.E. PP p Lower CI Upper CI
CM INT Total 0.511 0.052 < 0.001 0.402 0.606
CM INT Direct 0.508 0.054 < 0.001 0.394 0.604
CM INT Indirect 0.003 0.014 0.373 − 0.020 0.035
CM ASPD Total 0.362 0.061 < 0.001 0.246 0.486
CM ASPD Direct 0.351 0.061 < 0.001 0.232 0.475
CM ASPD Indirect 0.010 0.015 0.179 − 0.013 0.046
Model 2
Predictor Outcome Effect Coeff. S.E. PP p Lower CI Upper CI
Neglect INT Total 0.721 0.820 0.179 −1.337 2.049
Neglect INT Direct 0.729 0.829 0.178 −1.350 2.075
Neglect INT Indirect − 0.003 0.051 0.452 − 0.123 0.093
Psychological abuse INT Total 0.585 0.776 0.220 −1.190 1.970
Psychological abuse INT Direct 0.585 0.785 0.224 −1.223 1.984
Psychological abuse INT Indirect 0.002 0.050 0.464 − 0.102 0.104
Physical abuse INT Total − 0.537 0.644 0.163 −1.661 1.108
Physical abuse INT Direct − 0.541 0.650 0.162 −1.669 1.144
Physical abuse INT Indirect 0.001 0.046 0.478 − 0.083 0.108
Sexual abuse INT Total − 0.314 0.399 0.172 −1.025 0.687
Sexual abuse INT Direct − 0.320 0.404 0.171 −1.031 0.688
Sexual abuse INT Indirect 0.002 0.028 0.445 − 0.049 0.069
Neglect ASPD Total 1.575 1.720 0.151 − 2.035 4.895
Neglect ASPD Direct 1.595 1.742 0.152 − 2.092 4.965
Neglect ASPD Indirect − 0.004 0.089 0.473 − 0.227 0.148
Psychological abuse ASPD Total 1.101 1.329 0.177 −1.480 3.877
Psychological abuse ASPD Direct 1.123 1.347 0.177 −1.519 3.968
Psychological abuse ASPD Indirect − 0.009 0.089 0.432 − 0.213 0.155
Physical abuse ASPD Total −1.625 1.321 0.084 − 4.458 1.003
Physical abuse ASPD Direct −1.660 1.336 0.084 − 4.504 1.008
Physical abuse ASPD Indirect 0.017 0.082 0.352 − 0.099 0.235
Sexual abuse ASPD Total − 0.882 0.875 0.108 − 2.796 0.823
Sexual abuse ASPD Direct − 0.899 0.887 0.108 − 2.814 0.840
Sexual abuse ASPD Indirect 0.008 0.050 0.376 − 0.065 0.141
Open in a new tab

Presented are the mediation analyses from Model 1 and Model 2. The mediators of these predictions are EFw1 and EFw2. The total effect accounts for the relationship of all independent variables and mediations predicting the outcome variable. The direct effect is the specific prediction of the outcome when partialled for the mediators. The indirect effect is the relationship of the predictor through the mediators to the outcome. The specific indirect effects (the relationship of the predictor through the specific mediators to the outcome) are not shown but are in supplementary materials. A graphical representation of these relationships can be found in Fig. 1. Coeff = the path coefficient or the slope of that variable in the model. Posterior S.E. = the standard error of the posterior distribution which is a likelihood distribution of the point estimate for the given parameter (or coefficient). PP p = the Posterior Predictive p-value, which is akin to frequentist p-value for any test statistic, is obtained through draws from the posterior distribution. Lower and Upper CI = the credibility intervals of the parameter estimates which is based on 2.5% to 97.5% intervals of the posterior distribution. w1 = wave-1, w2 = wave-2, EF = executive functions, INT = internalizing disorders, ASPD = anti-social personality disorder, CM = Child Maltreatment which is a single factor composed of its four abuse and neglect subtypes.

Model 2 (Fig. 1B) aimed to delineate the effects of the CM subtypes by analyzing their influences on INT and ASPD directly and when mediated through EFs. As shown in Fig. 1, Model 2 includes four correlated factors representing the subdimensions of CM: neglect, psychological abuse, physical abuse and sexual abuse. The path coefficients between factors are presented in Fig. 1B and Table 2. The standardized results show that the EFw1 to EFw2 stability coefficient and covariates of age on EFw1 and sex on EFw2 were significant. None of the direct effects of the CM subscales on INT or ASPD nor the mediation effects were significant. However, note that the inter-correlations among CM subtypes were substantial and highly significant.

Post Hoc Analyses.

The finding that the different CM types had no significant unique effects on INT and ASPD were surprising since the overall CM (single factor) direct effect was highly significant. The null effects are likely explained by the substantial correlations and collinearity among each subtype. To address this, the CM subtypes were individually analyzed in the framework of Model 2, but only results between the subtypes and INT are reported. They are similar to when ASPD is the outcome. All of the subtypes showed highly significant direct effects on INT when individually analyzed, but there was no significant mediation via EFs (Table 4). In a complimentary analysis, each subtype was individually excluded from Model 2 and the resulting explained variance of INT was compared to the full model. A Wald χ2 test was conducted to see if the reduced model (removal of one subtype) worsened model fit. Results show that each subtype explained a large portion of the variance of INT (7–26%) but removing any one subtype from Model 2 did not significantly worsen model fit. This indicates that despite the CM subtypes individually explaining approximately 17% to 32% of the variance in INT, the substantial collinearity among them obscured their unique predictions of INT when modelled together.

Table 4.

Post Hoc: CM subtypes’ paths to INT when individually analyzed and individually excluded.

Individually analyzed INT
Coeff. S.E. PP p Lower CI Upper CI R2
Neglect 0.401 0.072 < 0.001 0.260 0.541 0.239
Psychological 0.507 0.055 < 0.001 0.392 0.606 0.315
Physical 0.444 0.075 < 0.001 0.288 0.584 0.268
Sexual 0.312 0.086 < 0.001 0.141 0.471 0.169
Post Hoc: Explained variance of INT by CM subtypes
INT
Model R2 ΔR2 Wald χ2 PP p
Baseline Model 0.400
Individually excluded
 Neglect 0.138 0.262 6.759 0.149
 Psychological 0.328 0.072 7.687 0.104
 Physical 0.27 0.130 7.226 0.124
 Sexual 0.299 0.101 4.894 0.298
Open in a new tab

This table shows post hoc analyses that sought to understand why the Child Maltreatment (CM) subtypes were not predictive of INT. In the analyses the overall structural equation model of Model 2 was maintained but the inclusion of the CM subtypes was modified. The first part of the table is the post hoc analysis where one CM subtype was analyzed at a time as the sole CM measure. This analysis intended to see if the CM subtypes predicted outcome variables when not correlated to the other subtypes. Only the path to and the explained variance of INT is included; the rest of the model results are not shown. The second part of the table is another post hoc analysis where three CM subtypes were included in the model and one was excluded. Presented for each subtype are the outcomes when that subtype was not in the model. This sought to see if excluding any CM types substantially changed the explained variance of INT and if the exclusion significantly worsened model fit. The baseline model is Model 2, so the change in R2 is the difference in explained variance when the given CM subtype is excluded; this difference is considered to be the explained variance of that subtype. A Wald chi-square test was conducted with df = 4. A significant Posterior Predictive p-value (PP p) indicates worse model fit. INT = internalizing disorders, CM = Child Maltreatment which is a single factor composed of its four abuse and neglect subtypes.

4. Discussion

In this study, we hypothesized that EFs (specifically changes in EF from adolescence to adulthood) would mediate, in part, the relationship between CM on INT with significant total, direct, and indirect effects. The same relationships were expected for ASPD which was included as an outcome (a measure of externalizing behavior) in our analyses to complement INT and more fully capture psychopathology. It was also hypothesized that different CM subtypes would show differential influences on EF and INT. For both INT and ASPD we found no evidence that EFs significantly mediated their relationships to CM. Associations between EFs and ASPD show a modest but trending relationship; large standard errors due to missing data meant it was not significant. The direct effects of the single CM factor (Model 1) on INT and ASPD were significant and accounted for most of the total effect of the model on both (99% for INT and 95% for ASPD). When CM subtypes were investigated, none of their unique predictions of EFs, INT, or ASPD were significant. The inability for the CM subtypes to predict outcomes is likely explained by the high level of co-occurrence (polyabuse) in this high-risk sample creating substantial inter-correlations (collinearity) obscuring their individual predictions. This supports measuring CM as a single common factor in high abuse samples.

The finding that overall CM predicts worse INT outcomes later in life is well founded (Carliner et al., 2017; McGuire et al., 2018; Miller & Brock, 2017) and this study adds to that body of evidence. What isn’t well understood, and is still hotly debated (Gabrielli & Jackson, 2019; Jackson et al., 2019), is how categorizing and classifying CM subtypes changes predictive power of CM measures. Our findings show that an overall measure of CM is sufficient to detect differences in INT. However, substantial collinearity among the CM subtypes in our high-risk sample resulted in non-significant unique predictions. This likely stems from the fact that abuse and neglect often co-occur (Kim et al., 2017) and 25% of our high-risk sample experienced three or more CM events (polyabuse). Our post-hoc analyses support this showing that when analyzed separately, the subtypes explained between 17% to 32% of the variance in INT by themselves. Yet when analyzed together, their collinearity obscured predictions of outcomes and exclusion of one subtype at a time did not worsen model fit. Lau et al. (2005) found similar results were in a polyabuse sample CM subtypes were not predictive of outcomes due to their inter-correlations. This may support classifying CM as a single factor since it captures the influence of co-occurring and correlated subtypes. However, the inability of factor analytic models to delineate the effects of subtypes of abuse may warrant a shift towards person-centered approaches such as Latent Class Analysis (Jackson et al., 2019). Further systematic comparisons of findings from high-risk (selected) and population-based samples are warranted.

The relationship between INT and CM was not significantly mediated by the change in EFs from wave-1 to wave-2 in either model. Although prediction of EFw1 to ASPD was trending, the lack of mediation may be due to the structural missingness on some of our EF measures. 47% of subjects were not administered the Stroop or Trails at wave-1 while 41% of subjects did not receive the P-IQ measure at wave-2. Bayesian estimation was used to address these limitations in the data; it is an efficient means of handling missing data (Asparouhov & Muthén, 2010a) and is asymptotically equivalent to Maximum Likelihood estimation (Asparouhov & Muthén, 2010b). It allows for parameter estimates to be derived using all available information similarly to Full Information Maximum Likelihood. Even with the use of all available information, point estimates of the EF mediation effect sizes were small implying that power was not the ultimate limiting factor. Notably, EF data were available from an average of 361 subjects, a sizeable sample. Even if larger samples altered significance, the interpretation of the modest effect sizes would be unchanged.

Further, some evidence suggests that EFs may not be strongly influenced by environmental pathways related to abuse and neglect. Instead, EFs may be more strongly affected by genetic factors. Fujisawa et al. (2017) found that EFs can be mostly explained through genetic differences. In a population-based twin sample, Friedman et al. (2018) showed that genetic differences are the strongest factors relating EFs to INT which remain relatively stable in early adulthood (Morrison et al., 2020). Only a small portion of change in EFs is explained by environmental factors (Friedman et al., 2016). It is possible that EFs are primarily established before age 16 (average wave-1 age), but our results show that EFs at wave-1 only partially explained EFs at wave-2 indicating substantial change. There is also some evidence that the EFs are strongly affected by other social determinants of health, like SES (Danese et al., 2016; Gur et al., 2019; Moore et al., 2016). That is why parental education was controlled for as a covariate. Finally, there are other cognitive functions like changes in threat appraisal, emotion regulation, and reward processing which may mediate the hypothesized relationships (for review see McCrory et al., 2017). Unfortunately, measures of these cognitive variables were unavailable. Our findings suggest that in this high-risk sample EFs are not strong mediators of the effects of CM on internalizing and externalizing disorders.

4.1. Limitations and future directions

Several limitations in this study should influence its interpretation. Our measure of CM is a retrospective report which are prone to underreporting, recall bias, and may measure different groups of CM victims than prospective reports (Baldwin et al., 2019). Furthermore, concerns that lower EFs may put individuals at a greater risk for experiencing traumatic events, including CM, creates confounds we cannot control for (Assink et al., 2019; Danese et al., 2016). Our subjects come from a high-risk sample (probands selected for substance abuse and antisocial behavior, and their siblings) rendering this study primarily generalizable to substance abusers and their families. Selection for a high-risk sample like ours may also include individuals with lower EFs due to substance use and other related stress exposure. It is possible that the measurement of EFs at mean age 16 and mean age 23 may be too late to capture changes in EFs after CM. However, EFs at wave-1 only partially explained wave-2 indicating substantial change.

Future studies should continue the use of longitudinal mediation analyses to determine if changes in EFs and INT are indeed influenced by maltreatment. Including polygenic risk-scores (Dudbridge, 2013) and SES variables in analyses may help elucidate whether the experiential and genetic factors related to INT interact. It may very well be that CM and EFs affect psychiatric disorders through separate pathways, one being genetic and the other environmental. Furthermore, when measuring CM, researchers should include severity, frequency, chronicity, and age of onset. Our study only examined yes/no endorsements of abuse and neglect events, but these other factors may hold key information about the effects of experiencing maltreatment.

4.2. Conclusion

The key findings from this study are that higher levels of internalizing disorders and anti-social personality disorder in young adulthood are associated with experiencing more CM. While a single CM factor predicted INT and ASPD, the high inter-correlations between the CM subtypes meant they could not uniquely predict these psychiatric outcomes. This lends support to using a single CM factor, but this finding may be influenced by characteristics of our high-risk, polyabuse sample. Despite some trending paths, the relationship of CM and adult psychopathology was explained primarily by a direct effect and was not significantly mediated through EFs assessed in adolescence. This may be due to missing EF data, although point estimates of the model paths and mediation effects were generally quite small. We hope this study will catalyze more research into mediating factors of the relationship between CM and psychiatric disorders.

Supplementary Material

Supp.Materials

Acknowledgements

Thanks to the Institute for Behavioral Genetics at the University of Colorado Boulder for their continued support on this study. We appreciate the subjects of the genetics for antisocial drug dependence study for participating and aiding in the advancement of our scientific knowledge. Personal thanks from the first author are extended to Christine, Richard, Jerry, and Judy Kent for their unwavering encouragement and to Dr. Michael “Flux” Caruso and Dr. Christopher Lowry for incredible mentorship.

Financial support

This work was supported by NIDA grants DA032555, DA042755, and DA035804.

Footnotes

Declaration of competing interest

The authors report no conflicts of interest.

Ethical standards

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.chiabu.2021.105369.

References

  1. Anderson V, Northam E, & Wrennall J. (2014). Developmental neuropsychology: A clinical approach. Psychology Press. [Google Scholar]
  2. Asparouhov T. (2020, July 6). Comparing models in bayesian framework. Mplus discussion. http://www.statmodel.com/cgi-bin/discus/show.cgi?9/8217. [Google Scholar]
  3. Asparouhov T, & Muthén B. (2010a). Bayesian analysis of latent variable models using Mplus (p. 60). [Google Scholar]
  4. Asparouhov T, & Muthén B. (2010b). Bayesian analysis using Mplus: Technical implementation (p. 38). [Google Scholar]
  5. Asparouhov T, & Muthén B. (2020). Advances in Bayesian model fit evaluation for structural equation models. Structural Equation Modeling: A Multidisciplinary Journal, 1–14. 10.1080/10705511.2020.1764360 [DOI] [Google Scholar]
  6. Assink M, van der Put CE, Meeuwsen MWCM, de Jong NM, Oort FJ, Stams GJJM, & Hoeve M. (2019). Risk factors for child sexual abuse victimization: A meta-analytic review. Psychological Bulletin, 145(5), 459–489. 10.1037/bul0000188 [DOI] [PubMed] [Google Scholar]
  7. Baldwin JR, Reuben A, Newbury JB, & Danese A. (2019). Agreement between prospective and retrospective measures of childhood maltreatment: A systematic review and Meta-analysis. JAMA Psychiatry, 76(6), 584–593. 10.1001/jamapsychiatry.2019.0097 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Barzilay R, Calkins ME, Moore TM, Wolf DH, Satterthwaite TD, Scott JC, … Gur RE (2019). Association between traumatic stress load, psychopathology, and cognition in the Philadelphia neurodevelopmental cohort. Psychological Medicine, 49(2), 325–334. 10.1017/S0033291718000880 [DOI] [PubMed] [Google Scholar]
  9. Beal SJ, Wingrove T, Mara CA, Lutz N, Noll JG, & Greiner MV (2019). Childhood adversity and associated psychosocial function in adolescents with complex trauma. Child & Youth Care Forum, 48(3), 305–322. 10.1007/s10566-018-9479-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Brumley LD, Brumley BP, & Jaffee SR (2019). Comparing cumulative index and factor analytic approaches to measuring maltreatment in the National Longitudinal Study of Adolescent to Adult Health. Child Abuse & Neglect, 87, 65–76. 10.1016/j.chiabu.2018.08.014 [DOI] [PubMed] [Google Scholar]
  11. Carliner H, Gary D, McLaughlin KA, & Keyes KM (2017). Trauma exposure and externalizing disorders in adolescents: Results from the National Comorbidity Survey Adolescent Supplement. Journal of the American Academy of Child & Adolescent Psychiatry, 56(9), 755–764. e3 10.1016/j.jaac.2017.06.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Crowley TJ, Mikulich SK, Ehlers KM, Hall SK, & Whitmore EA (2003). Discriminative validity and clinical utility of an abuse-neglect interview for adolescents with conduct and substance use problems. American Journal of Psychiatry, 160(8), 1461–1469. 10.1176/appi.ajp.160.8.1461 [DOI] [PubMed] [Google Scholar]
  13. Danese A, Moffitt TE, Arseneault L, Bleiberg BA, Dinardo PB, Gandelman SB, … Caspi A. (2016). The origins of cognitive deficits in victimized children: Implications for neuroscientists and clinicians. American Journal of Psychiatry, 174(4), 349–361. 10.1176/appi.ajp.2016.16030333 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Depaoli S, & van de Schoot R. (2017). Improving transparency and replication in Bayesian statistics: The WAMBS-checklist. Psychological Methods, 22(2), 240–261. 10.1037/met0000065 [DOI] [PubMed] [Google Scholar]
  15. Dinwiddie S, Heath AC, Dunne MP, Bucholz KK, Madden P. a. F., Slutske WS, … Martin NG (2000). Early sexual abuse and lifetime psychopathology: A co-twin–control study. Psychological Medicine, 30(1), 41–52. 10.1017/S0033291799001373 [DOI] [PubMed] [Google Scholar]
  16. Dudbridge F. (2013). Power and predictive accuracy of polygenic risk scores. PLoS Genetics, 9(3), Article e1003348. 10.1371/journal.pgen.1003348 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Felitti VJ, Anda RF, Nordenberg D, Williamson DF, Spitz AM, Edwards V, Koss MP, & Marks JS (1998). Relationship of childhood abuse and household dysfunction to many of the leading causes of death in adults: The adverse childhood experiences (ACE) study. American Journal of Preventive Medicine, 14(4), 245–258. 10.1016/S0749-3797(98)00017-8. [DOI] [PubMed] [Google Scholar]
  18. Friedman NP, Altamirano LJ, Corley RP, Young SE, Rhea SA, & Hewitt JK (2016). Stability and change in executive function abilities from late adolescence to early adulthood: A longitudinal twin study. Developmental Psychology, 52(2), 326–340. 10.1037/dev0000075 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Friedman NP, du Pont A, Corley RP, & Hewitt JK (2018). Longitudinal relations between depressive symptoms and executive functions from adolescence to early adulthood: A twin study. Clinical Psychological Science, 6(4), 543–560. 10.1177/2167702618766360 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Friedman NP, Miyake A, Young SE, DeFries JC, Corley RP, & Hewitt JK (2008). Individual differences in executive functions are almost entirely genetic in origin. Journal of Experimental Psychology. General, 137(2), 201–225. 10.1037/0096-3445.137.2.201 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Fujisawa KK, Todo N, & Ando J. (2017). Genetic and environmental influences on the development and stability of executive functions in children of preschool age: A longitudinal study of Japanese twins. Infant and Child Development, 26(3), Article e1994. 10.1002/icd.1994 [DOI] [Google Scholar]
  22. Gabrielli J, & Jackson Y. (2019). Innovative methodological and statistical approaches to the study of child maltreatment: Introduction. Child Abuse & Neglect, 87, 1–4. 10.1016/j.chiabu.2018.12.001 [DOI] [PubMed] [Google Scholar]
  23. Gogtay N, Giedd JN, Lusk L, Hayashi KM, Greenstein D, Vaituzis AC, … Thompson PM (2004). Dynamic mapping of human cortical development during childhood through early adulthood. Proceedings of the National Academy of Sciences, 101(21), 8174–8179. 10.1073/pnas.0402680101 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Golden CJ, & Freshwater SM (2002). Stroop color and word test: Revised 2002 adult manaual for clinical and experimental uses. Stoelting. [Google Scholar]
  25. Gur RE, Moore TM, Rosen AFG, Barzilay R, Roalf DR, Calkins ME, … Gur RC (2019). Burden of environmental adversity associated with psychopathology, maturation, and brain behavior parameters in youths. JAMA Psychiatry, 76(9), 966–975. 10.1001/jamapsychiatry.2019.0943 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Hedayati SS, Bosworth HB, Kuchibhatla M, Kimmel PL, & Szczech LA (2006). The predictive value of self-report scales compared with physician diagnosis of depression in hemodialysis patients. Kidney International, 69(9), 1662–1668. 10.1038/sj.ki.5000308 [DOI] [PubMed] [Google Scholar]
  27. Hollingshea AB. (1975). Four factor index of social status [unpublished]. https://sociology.yale.edu/sites/default/files/files/yjs_fall_2011.pdf#page=21.
  28. Jackson Y, McGuire A, Tunno AM, & Makanui PK (2019). A reasonably large review of operationalization in child maltreatment research: Assessment approaches and sources of information in youth samples. Child Abuse & Neglect, 87, 5–17. 10.1016/j.chiabu.2018.09.016 [DOI] [PubMed] [Google Scholar]
  29. Jensen AR, & Rohwer WD (1966). The stroop color-word test: A review. Acta Psychologica, 25, 36–93. 10.1016/0001-6918(66)90004-7 [DOI] [PubMed] [Google Scholar]
  30. Kamens HM, Corley RP, McQueen MB, Stallings MC, Hopfer CJ, Crowley TJ, … Ehringer MA (2013). Nominal association with CHRNA4 variants and nicotine dependence. Genes, Brain and Behavior, 12(3), 297–304. 10.1111/gbb.12021 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Kane MJ, & Engle RW (2003). Working-memory capacity and the control of attention: The contributions of goal neglect, response competition, and task set to Stroop interference. Journal of Experimental Psychology. General, 132(1), 47–70. 10.1037/0096-3445.132.1.47 [DOI] [PubMed] [Google Scholar]
  32. Kim K, Mennen FE, & Trickett PK (2017). Patterns and correlates of co-occurrence among multiple types of child maltreatment. Child & Family Social Work, 22 (1), 492–502. 10.1111/cfs.12268 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Langenecker SA, Kennedy SE, Guidotti LM, Briceno EM, Own LS, Hooven T, … Zubieta J-K (2007). Frontal and limbic activation during inhibitory control predicts treatment response in major depressive disorder. Biological Psychiatry, 62(11), 1272–1280. 10.1016/j.biopsych.2007.02.019 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Lau AS, Leeb RT, English D, Graham JC, Briggs EC, Brody KE, & Marshall JM (2005). What’s in a name? A comparison of methods for classifying predominant type of maltreatment. Child Abuse & Neglect, 29(5), 533–551. 10.1016/j.chiabu.2003.05.005 [DOI] [PubMed] [Google Scholar]
  35. Marsh R, Gerber AJ, & Peterson BS (2008). Neuroimaging studies of normal brain development and their relevance for understanding childhood neuropsychiatric disorders. Journal of the American Academy of Child & Adolescent Psychiatry, 47(11), 1233–1251. 10.1097/CHI.0b013e318185e703 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. McCrory EJ, Gerin MI, & Viding E. (2017). Annual research review: Childhood maltreatment, latent vulnerability and the shift to preventative psychiatry – The contribution of functional brain imaging. Journal of Child Psychology and Psychiatry, 58(4), 338–357. 10.1111/jcpp.12713 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. McCrory EJ, & Viding E. (2015). The theory of latent vulnerability: Reconceptualizing the link between childhood maltreatment and psychiatric disorder. Development and Psychopathology, 27(2), 493–505. 10.1017/S0954579415000115 [DOI] [PubMed] [Google Scholar]
  38. McGuire A, Cho B, Huffhines L, Gusler S, Brown S, & Jackson Y. (2018). The relation between dimensions of maltreatment, placement instability, and mental health among youth in foster care. Child Abuse & Neglect, 86, 10–21. 10.1016/j.chiabu.2018.08.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Miller ML, & Brock RL (2017). The effect of trauma on the severity of obsessive-compulsive spectrum symptoms: A meta-analysis. Journal of Anxiety Disorders, 47, 29–44. 10.1016/j.janxdis.2017.02.005 [DOI] [PubMed] [Google Scholar]
  40. Moody G, Cannings-John R, Hood K, Kemp A, & Robling M. (2018). Establishing the international prevalence of self-reported child maltreatment: A systematic review by maltreatment type and gender. BMC Public Health, 18(1), 1164. 10.1186/s12889-018-6044-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Moore TM, Martin IK, Gur OM, Jackson CT, Scott JC, Calkins ME, … Gur RC (2016). Characterizing social environment’s association with neurocognition using census and crime data linked to the Philadelphia neurodevelopmental cohort. Psychological Medicine, 46(3), 599–610. 10.1017/S0033291715002111 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Morrison CL, Rhee SH, Smolker HR, Corley RP, Hewitt JK, & Friedman NP (2020). Genetic and environmental influences on stressful life events and their associations with executive functions in young adulthood: A longitudinal twin analysis. Behavior Genetics. 10.1007/s10519-020-10017-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Muthén B, & Asparouhov T. (2012). Bayesian structural equation modeling: A more flexible representation of substantive theory. Psychological Methods, 17(3), 313–335. 10.1037/a0026802 [DOI] [PubMed] [Google Scholar]
  44. Muthén B, & Muthén L. (2017). Mplus (Version 8) [Computer software]. Muthén & Muthén. [Google Scholar]
  45. Pechtel P, & Pizzagalli DA (2011). Effects of early life stress on cognitive and affective function: An integrated review of human literature. Psychopharmacology, 214(1), 55–70. 10.1007/s00213-010-2009-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Radloff LS (1977). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1(3), 385–401. 10.1177/014662167700100306 [DOI] [Google Scholar]
  47. Reitan RM, & Wolfson D. (2004). The trail making test as an initial screening procedure for neuropsychological impairment in older children. Archives of Clinical Neuropsychology, 19(2), 281–288. 10.1016/S0887-6177(03)00042-8 [DOI] [PubMed] [Google Scholar]
  48. Robins LN, Cottler LB, Bucholz KK, Compton WM, North CS, & Rourke KM (2000). Diagnostic interview schedule for the DSM-IV (DIS-IV). [Google Scholar]
  49. Rouse HL, & Fantuzzo JW (2009). Multiple risks and educational well being: A population-based investigation of threats to early school success. Early Childhood Research Quarterly, 24(1), 1–14. 10.1016/j.ecresq.2008.12.001 [DOI] [Google Scholar]
  50. Shaffer D, Fisher P, Lucas CP, Dulcan MK, & Schwab-Stone ME (2000). NIMH diagnostic interview schedule for children version IV (NIMH DISC-IV): Description, differences from previous versions, and reliability of some common diagnoses. Journal of the American Academy of Child and Adolescent Psychiatry, 39 (1), 28–38. 10.1097/00004583-200001000-00014 [DOI] [PubMed] [Google Scholar]
  51. Stoltenborgh M, Bakermans-Kranenburg MJ, Alink LRA, & IJzendoorn M. H.v. (2015). The prevalence of child maltreatment across the globe: Review of a series of Meta-analyses. Child Abuse Review, 24(1), 37–50. 10.1002/car.2353 [DOI] [Google Scholar]
  52. Strauss E, Sherman EMS, & Spreen O. (2006). A compendium of neuropsychological tests: Administration, norms, and commentary. Oxford University Press. [Google Scholar]
  53. Stroop JR (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 18(6), 643–662. 10.1037/h0054651 [DOI] [Google Scholar]
  54. Teicher MH, Andersen SL, Polcari A, Anderson CM, Navalta CP, & Kim DM (2003). The neurobiological consequences of early stress and childhood maltreatment. Neuroscience & Biobehavioral Reviews, 27(1), 33–44. 10.1016/S0149-7634(03)00007-1 [DOI] [PubMed] [Google Scholar]
  55. Warmingham JM, Handley ED, Rogosch FA, Manly JT, & Cicchetti D. (2019). Identifying maltreatment subgroups with patterns of maltreatment subtype and chronicity: A latent class analysis approach. Child Abuse & Neglect, 87, 28–39. 10.1016/j.chiabu.2018.08.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Wechsler D. (1991). Manual for the Wechsler intelligence scale for children (3rd ed.). The Psychological Corporation. [Google Scholar]
  57. Wechsler D. (1999). Wechsler abbreviated scale of intelligence (3rd ed.). The Psychological Corporation: Harcourt Brace & Company. [Google Scholar]
  58. Yuan Y, & MacKinnon DP (2009). Bayesian mediation analysis. Psychological Methods, 14(4), 301–322. 10.1037/a0016972 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supp.Materials
NIHMS1750691-supplement-Supp_Materials.docx (32.6MB, docx)

RESOURCES