Individual Differences in the Effects of Neighborhood Socioeconomic Deprivation on Intertemporal Decision-Making and Psychotic-Like Experiences in Children

doi:10.21203/rs.3.rs-4618474/v1

Download PDF

Article

Individual Differences in the Effects of Neighborhood Socioeconomic Deprivation on Intertemporal Decision-Making and Psychotic-Like Experiences in Children

https://doi.org/10.21203/rs.3.rs-4618474/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

This study elucidates the influence of socioeconomic environments on neurodevelopment and psychiatric vulnerability in children. Employing advanced machine learning-based causal inference (IV Forest), we analyzed the impact of neighborhood socioeconomic deprivation on delay discounting and psychotic-like experiences (PLEs) among 2,135 children. Our findings reveal that greater neighborhood deprivation correlates with increased future reward discounting and elevated PLEs, particularly hallucinational symptoms, over 1-year and 2-year follow-ups. Vulnerable children in these settings exhibited notable neuroanatomical changes, including reduced limbic volume, surface area, and white matter, and heightened BOLD reactivity in the prefrontal-limbic system during reward tasks. These findings highlight the complex interplay between environmental factors and brain reward mechanisms in shaping PLE risk, advocating for early, targeted interventions in socioeconomically disadvantaged communities. This research not only extends our understanding of environmental influences on child psychology but also guides personalized intervention strategies and prompts reflection on broader societal impacts.

Biological sciences/Neuroscience/Social neuroscience

Biological sciences/Neuroscience/Cognitive neuroscience/Decision

Health sciences/Diseases/Psychiatric disorders/Psychosis

Intertemporal reward valuation

Psychotic-like experiences

Causal machine learning

Childhood socioeconomic environment

Heterogeneous treatment effects

In Critique of Practical Reason, Immanuel Kant champions the inherent power of human reason, suggesting that it is an a priori capacity independent of external factors, enabling individuals to engage in responsible actions¹. Nevertheless, a wealth of scientific studies in recent decades stands in opposition to the Enlightenment philosopher's claims, highlighting the significant impact of environmental factors on the development of personal identity and behavior.

Adverse childhood environments, such as low family income, malnutrition, physical or sexual abuse, and unsafe neighborhoods, are linked to an heightened risk of various mental or physical health issues, including psychosis^2–4, impoverished cognitive ability^5–7, anxiety, bipolar disorder, self-harm, depression^3,4,8, substance abuse, and obesity^9,10. Furthermore, these environments are associated with negative social outcomes, such as poor academic performance^11,12, low income, unemployment^13–18, higher rates of imprisonment, and increased likelihood of teen pregnancy¹⁹. Additionally, exposure to these adverse conditions in childhood is associated with a propensity for engaging in risky behaviors, including criminal activity²⁰, excessive consumption of calorie-dense foods²¹, substance use^22,23, deficient self-control²⁴, and disrupted reward processing²⁵.

The intricate relationship between challenging childhood environments, irresponsible behavior, and adverse social and health outcomes raises important questions. We hypothesized that childhood adversity causes impairment in one’s valuation system, leading to negative life outcomes. Children who experienced social adversities such as poverty show steeper discounting of future rewards in adulthood and have higher psychotic-like experiences (PLEs)^2,3,26–28. Lower socioeconomic status positively correlates with functional brain activity concordance and grey matter volume within reward-related areas (i.e., ventral striatum, putamen, caudate nucleus, orbital frontal cortex) and negatively with executive-related areas (i.e., frontal, medial frontal cortex)²⁹. A recent study reported that neuroanatomical features including total cortical volume, surface area, and thickness mediates the association of environmental risk factors and PLEs in children³.

In addition, individuals with steeper discounting of future rewards (i.e., value present rewards much higher than future rewards) tend to display a range of suboptimal behavior and outcomes. They are more likely to save less, invest less in their education, engage in criminal activities, exhibit lower academic performance, and accumulate less economic wealth^30–33. Such impairments in intertemporal valuation are not only associated with financial and social disadvantages but are also linked to psychiatric disorders, including psychosis, attention deficit/hyperactivity disorder (ADHD), and addiction^34,35.

Psychosis, in particular, is associated with distinctive alterations in intertemporal decision-making, characterized by steeper discounting of future rewards^36–38. This cognitive pattern in individuals with psychosis may be reflected in abnormal neural responses to non-relevant rewards, potentially driven by elevated levels of tonic dopamine^34,39–41. Comparative studies have demonstrated that individuals with psychosis discount future rewards more steeply than healthy controls^36,42,43, a pattern that is uniquely pronounced in psychosis compared to other psychopathologies such as primary mood disorders⁴², major depressive disorder⁴², and bipolar disorder⁴³. The disruption in dopaminergic projections from the ventral tegmental area to the mesocorticolimbic regions is hypothesized to impair reward anticipation and perception processes^39,40, potentially contributing to the phenomenology of delusions or hallucinations.

In the present study, our primary objective was to investigate the impact of neighborhood socioeconomic deprivation on adolescents' delay discounting and PLEs. Delay discounting, which is evidenced by the extent to which individuals’ discount future rewards, pertains to their intertemporal decision-making and impulsive behavior. Exposure to adversities at the neighborhood level during childhood has been shown to negatively influence neurocognitive development^7,44,45, subsequently resulting in psychiatric disorders^2,3,28 and unfavorable social outcomes, such as decreased income, reduced probability of college attendance, and limited employment opportunities¹⁷. This phenomenon is particularly pronounced in societies where discrimination based on family income or race/ethnicity restricts underprivileged families from selecting neighborhoods that present greater opportunities for upward social mobility, as observed in the United States¹⁷.

It is crucial to note that PLEs, frequently reported in children, are considered as a clinically significant risk indicator for psychosis and general psychopathology^46,47. Around 17% of 9–12 years old children report PLEs⁴⁸, and individuals with PLEs at age 11 had greater risk of developing psychotic disorders in adulthood^49,50. Prior studies revealed that PLEs are correlated to heightened vulnerability to other psychopathologies including suicidal behavior², mood, anxiety, and substance disorders^46,48, and exhibit the strongest association with environmental risk factors in comparison to other internalizing/externalizing symptoms during early adolescence³. The present study endeavors to explore the potential causal mechanisms underlying these associations.

Our second aim was to test whether the potential causal effects of neighborhood deprivation on children’s PLEs are heterogeneous based on individual’s delay discounting and its genetic, neural correlates. The heterogeneous nature of psychopathology has long posed significant challenges for clinical diagnosis and treatment^51,52. Given that the genetic and neural correlates of delay discounting substantially overlap with those of psychosis^40,41,53,54, the shared biological foundations between reward valuation and psychosis may result in heterogeneous effects of environmental exposure on an individual's PLEs. By investigating these potential variations, this study seeks to enhance the understanding of the complex interplay between environmental factors and individual predispositions in the development of psychopathology.

Identifying individual differences of treatment/exposure is crucial for the development of personalized health care. Delivering optimal health care for each patient necessitates the recognition of genetic markers, neurodevelopmental characteristics, and sociodemographic features associated with individual variations in treatment effects^55,56. However, previous studies employing traditional methods of testing the individual differences in treatment effects have often been unsuccessful in discerning the intricate interplay between genetic and environmental factors^57,58. Linear models with interaction terms of features selected a priori by the researcher may not fully reflect the complex and elusive gene-environment interplay, particularly in genetic and neuroscience research where the input features are usually high dimensional.

Using instrumental variable (IV) random forests (henceforth IV Forest), an up-to-date causal machine learning approach^59,60, we assessed the effects of neighborhood socioeconomic adversity on delay discounting and PLEs, and the potential individual differences within those effects. We leveraged multimodal magnetic resonance imaging (MRI) data from 11,876 preadolescent children aged 9 to 12 years old (the Adolescent Brain Cognitive Development (ABCD) Study). Integration of innovative analytical techniques and a large sample with diverse genetic and environmental backgrounds permits us to test the complex interactions between genetic and environmental factors, ultimately contributing to the development of more effective personalized health care strategies.

The demographic characteristics of the final sample (N = 2,135) are presented in Table 1. Within the sample, 46.14% were female, 76.63% of participants had married parents, the mean family income was $70,245, and 65.57% identified their race/ethnicity as white. To ensure the representativeness of the final sample, a supplementary table comparing the sample's demographic characteristics with those of the general United States population is provided in the Supplementary Information (Supplementary Table 1). This comparison serves to reinforce the validity and generalizability of the study's findings.

Table 1

**Socioeconomic/demographic characteristics of the participants.** *Age* is rounded to chronological month. *Race/Ethnicity* denote child’s self-reported racial / ethnic identity. *Household Income* is assessed as the total combined family income for the past 12 months. *Parental Education* is measured as the highest grade or level of school completed or highest degree received. *Family History of Psychiatric Disorders* represents the proportion of first-degree relatives who experienced mental illness.
Demographic Characteristics		N	Ratio (%)	Mean (SD)
Age		2,135		120.1541 (7.4658)
Sex	Male	1,517	53.86%
Sex	Female	985	46.14%
Marital Status of the first caregiver	Married	1,636	76.63%
	Widowed	12	0.56%
	Divorced	193	9.04%
	Separated	62	2.9%
	Never Married	142	6.65%
	Living with Partner	90	4.22%
Race/Ethnicity	White	1,400	65.57%
	Black	136	6.37%
	Hispanic	373	17.47%
	Asian	7	0.33%
	Other	219	10.26%
Parent’s Identity	Biological Mother	1,848	86.56%
	Biological Father	215	10.07%
	Adoptive Parent	39	1.83%
	Custodial Parent	12	0.56%
	Other	21	0.98%
Household Income		2,135		$70,245 (1.937)
Parental Education		2,135		17.2838 (2.3046)
BMI		2,135		18.4298 (3.8572)
Parental Age		2,135		40.8775 (6.3825)
Family History of Psychiatric Disorders		2,135		0.0958 (0.1125)

In our initial exploratory analysis, partial correlations were used to examine the relationship between psychopathological symptoms and delay discounting. Among the symptoms assessed (e.g., depression, anxiety, ADHD), only PLEs showed significant correlation with delay discounting (Spearman ρ= -0.067, p-FDR = 0.024 ~ ρ= -0.057, p-FDR = 0.035) (Supplementary Table 2). This finding underscores the unique association between PLEs and delay discounting, laying the groundwork for subsequent investigations into how delay discounting—along with its genetic and neural correlates—may be associated with the heterogeneous effects of neighborhood socioeconomic adversity on PLEs.

Given the non-randomized, observational nature of the ABCD Study, potential confounding factors, such as genetic, environmental variables, and their unobserved common causes, can lead to biased estimations⁶¹. A powerful and effective way to adjust such bias is the IV regression. Designed to address bias from unobserved confounders, IV regression is an effective method to conduct causal inference using non-randomized, observational data for research in various domains, including economics^62,63, psychology⁶⁴, neuroscience⁶⁵, and psychiatry⁶⁶.

In this study, the instrumental variable used was the presence of state-level source of income (SOI) laws at baseline year assessment, which prevent income discrimination in housing. These laws ensure that landlords accept housing vouchers, aiding low-income families in securing quality housing. As such, SOI laws are critical in enabling better residential environments by mitigating neighborhood socioeconomic adversity, thereby serving as an effective instrument for assessing the potential causal effects of neighborhood socioeconomic adversity (measured with Area Deprivation Index, henceforth ADI), on delay discounting and PLEs (see Fig. 1A).

We used IV Forest^59,60—a random forest-based IV regression⁶²—to adjust for unobserved confounding bias in identifying the potential causal effects of ADI on delay discounting and PLEs. The IV Forest method enabled us to derive nonparametric, doubly robust estimates of the average (group-level) and heterogeneous (individual-level) treatment effects of ADI on these outcomes (see Fig. 1B). This method is noted for delivering estimates with significantly lower mean-squared error compared to conventional k-nearest neighbor methods^59,60. Furthermore, its use of independent subsamples for model construction and validation ensures honest, overfitting-resistant estimates of average and heterogeneous treatment effects^59,60,67. Notably, this method is particularly useful for analyzing the complex, nonlinear interactions between genetic and environmental factors and their effects on neurocognitive development and PLEs, even within the confines of observational data⁵⁹.

Figure 2 presents the analytical framework of our study, examining the effects of neighborhood socioeconomic adversity on children's decision-making and mental health. ADI, recorded in the baseline year, serves as an indicator of this socioeconomic adversity. We assessed the impact of ADI on children’s intertemporal decision-making through delay discounting at a 1-year follow-up. PLEs, encompassing distress, delusional, and hallucinational symptoms, were evaluated at both 1-year and 2-year follow-ups. Our analysis spans multiple follow-up periods and PLE indicators to investigate the sustained influence of ADI over time and to explore differential effects on various PLE symptoms, particularly delusional versus hallucinational.

Average Treatment Effects of Neighborhood Socioeconomic Adversity on Delay Discounting and PLEs

IV Forest analyses revealed that a higher ADI has significant associations with a lower delay discounting (β= -1.73, p-FDR = 0.048) and a higher PLEs (distress score 1-year follow-up: β = 1.872, p-FDR = 0.048; distress score 2-year follow-up: β = 1.504, p-FDR = 0.039; delusional score 1-year follow-up: β = 5.97, p-FDR = 0.048; delusional score 2-year follow-up: β = 4.022, p-FDR = 0.048; hallucinational score 1-year follow-up: β = 3.761, p-FDR = 0.048; hallucinational score 2-year follow-up: β = 4.786, p-FDR = 0.039) (Table 2).

Table 2

**Potential causal effects of neighborhood socioeconomic adversity on intertemporal valuation and PLEs.** Average treatment effects of ADI on delay discounting and PLEs in the IV Forest models are shown. All p-values were corrected for multiple comparison using false discovery rate.
	IV Forests: Average Treatment Effects
	Estimates	Std. Error	95% Lower CI	95% Upper CI	P-FDR
Delay Discounting (1-year follow-up)	-1.730	0.748	-3.195	-0.265	0.048
Distress Score PLEs (1-year follow-up)	1.872	0.612	0.673	3.071	0.048
Distress Score PLEs (2-year follow-up)	1.504	0.592	0.345	2.664	0.039
Delusional Score PLEs (1-year follow-up)	5.970	2.911	0.264	11.676	0.048
Delusional Score PLEs (2-year follow-up)	4.022	1.987	0.127	7.917	0.048
Hallucinational Score PLEs (1-year follow-up)	3.761	1.902	0.033	7.489	0.048
Hallucinational Score PLEs (2-year follow-up)	4.786	1.861	1.139	8.433	0.039

To evaluate the robustness of our findings from the IV Forest, we calculated the E-values for the average treatment effects from the IV Forest. The E-values quantify the minimum strength of association that unobserved confounders would need to possess with both ADI and the outcomes, conditional on the observed covariates in the IV Forest model, to nullify the observed relationships⁶⁸. E-values indicated that unobserved confounders would need to have a relative risk greater than 9.13 for delay discounting and between 7.32 and 457.03 for PLEs to account entirely for the observed effects (Supplementary Table 3). This suggests a high threshold for unobserved confounding effects, thereby strengthening the validity of the causal inferences drawn from our analyses.

Supplementary analyses employing a conventional linear IV regression⁶² and an alternative causal machine learning method, i.e., Double ML^69,70, corroborated these findings. The conventional IV regression also showed that ADI has negative influence on childhood delay discounting (β= -0.468, p-FDR = 0.03) and positive PLEs (distress score 1-year follow-up: β = 0.609, p-FDR = 0.011; distress score 2-year follow-up: β = 0.78, p-FDR = 0.003; delusional score 1-year follow-up: β = 0.486, p-FDR = 0.028; delusional score 2-year follow-up: β = 0.578, p-FDR = 0.013; hallucinational score 1-year follow-up: β = 0.604, p-FDR = 0.011; hallucinational score 2-year follow-up: β = 0.827, p-FDR = 0.003). The partial-linear IV model of the Double ML algorithm showed significant effects of ADI on children’s delay discounting (β= -0.429, p-FDR = 0.044), distress score PLEs (1-year follow-up: β = 0.495, p-FDR = 0.023; 2-year follow-up: β = 0.609, p-FDR = 0.005), hallucinational score PLEs (1-year follow-up: β = 0.498, p-FDR = 0.018; 2-year follow-up: β = 0.683, p-FDR = 0.002), and 2-year follow-up delusional score PLEs (β = 0.417, p-FDR = 0.044). The negative effects of ADI on 1-year follow-up delusional score PLEs were marginally significant (β = 0.393, p-FDR = 0.051). These results of the conventional linear IV regression (Supplementary Table 4) and Double ML partial-linear IV regression (Supplementary Table 5) confirm the findings obtained from the IV Forest, further supporting the primary analyses and conclusions drawn from the study.

Heterogeneous Treatment Effects of Neighborhood Socioeconomic Adversity on PLEs, conditioned on the Genetic and Neural Correlates of Delay Discounting

Next, we tested whether the impact of ADI was heterogeneous across children, and, if so, whether the heterogeneity is linked to individual’s neurodevelopmental characteristics and the relevant genetic factors—assessed with genome-wide polygenic scores (GPS) and structural MRI and monetary incentive delay (MID) task fMRI data—correlated to intertemporal valuation. To identify the best subset of genetic and neural correlates of delay discounting, we first selected GPS and MRI brain regions of interest (ROIs) specifically related to delay discounting. To analyze the nonparametric correlations of multiple input variables, we used a random forest-based feature selection Boruta algorithm⁷¹. Its robustness and effectiveness in selecting relevant features in high dimensional, intercorrelated biomedical data (e.g., MRI) has been validated⁷¹ and consistently applied in genetics and neuroscience research^72–74. The variables significantly correlated with delay discounting (p-Bonferroni < 0.05) were GPS of cognitive performance, IQ, and education attainment; morphometric features (e.g., surface area, volume) in the limbic system (temporal pole, parahippocampal gyrus, caudate nucleus, rostral anterior cingulate, isthmus cingulate), inferior frontal gyrus (pars opercularis), and fusiform gyrus; mean beta activations of rewards/losses versus neutral feedback in the subcortical areas (thalamus proper, ventral diencephalon), precentral gyrus, supramarginal gyrus, temporal lobe (transverse temporal gyrus, superior temporal gyrus), and insula (Supplementary Table 6).

We then assessed the heterogeneous treatment effects of ADI on PLEs using three distinct IV Forest models: (1) the Delay Discounting model, incorporating sociodemographic features and delay discounting; (2) the Gene-Brain model, which included sociodemographic features and genetic and neural correlates of delay discounting (i.e., GPS and brain ROIs identified using the Boruta algorithm); and (3) the Integrated model which combined all the variables from the previous two models. All three models satisfied the overlap assumption (i.e., the estimated propensity scores are not close to one or zero), which is crucial for the validity of the estimated heterogeneous treatment effects (Supplementary Fig. 1). In line with prior studies^75,76, we obtained conditional average treatment effects, divided subjects into deciles (Q1: most vulnerable; Q10: most resilient) based on the conditional average treatment effects, and conducted three hypothesis tests⁷⁷ on each model to determine the most effective model for capturing the individual differences (heterogeneity) in the effects of ADI on PLEs: monotonicity, alternative hypothesis, and ANOVA.

Among the three models, only the Integrated model successfully demonstrated significant individual differences in the ADI effects on PLEs. This was evident in the impact of ADI on 1-year follow-up distress score PLEs (monotonicity test: p-FDR = 0.011; alternative hypothesis test: p = 0.002; ANOVA test: p < 0.001) and 1-year follow-up hallucinational score PLEs (monotonicity test: p-FDR = 0.038; alternative hypothesis test: p = 0.004; ANOVA test: p < 0.001), as presented in Fig. 3 and Table 3. In contrast, the Delay Discounting model and Gene-Brain model failed to pass the heterogeneity tests (monotonicity test: p-FDR ≥ 0.05; alternative hypothesis and ANOVA test: p ≥ 0.05).

Table 3

**Evaluation of individual differences in the potential causal effects of neighborhood socioeconomic adversity on children’s PLEs.** We employed three distinct tests to evaluate the heterogeneous treatment effects of ADI on PLEs: monotonicity test, alternative hypothesis test, and ANOVA test. These tests were applied to three developed IV Forest models: Delay Discounting model, Gene-Brain model, and Integrated model. All p-values were corrected for multiple comparison using false discovery rate. A star (*) denotes overall significance, indicating that a model passed all three heterogeneity tests.
		Monotonicity	Alternative Hypothesis				ANOVA		Overall Significance
		p-FDR	Estimate	Std. Error	t-value	p-FDR	F-value	p-FDR	Overall Significance
Delay Discounting Model	Distress Score PLEs (1-year follow-up)	0.037	-22.261	40.164	-0.554	0.799	1.114	0.48
	Distress Score PLEs (2-year follow-up)	0.094	-54.769	142.528	-0.384	0.799	1.081	0.48
	Delusional Score PLEs (1-year follow-up)	0.019	55.047	188.400	0.292	0.799	0.965	0.495
	Delusional Score PLEs (2-year follow-up)	0.303	23.289	91.550	0.254	0.799	0.852	0.568
	Hallucinational Score PLEs (1-year follow-up)	0.362	-86.756	241.597	-0.359	0.799	0.995	0.495
	Hallucinational Score PLEs (2-year follow-up)	0.098	-119.595	341.060	-0.351	0.799	1.010	0.495
Gene-Brain Model	Distress Score PLEs (1-year follow-up)	0.058	-9.522	4.244	-2.244	0.05625	2.289	0.027
	Distress Score PLEs (2-year follow-up)	0.099	-9.202	7.016	-1.312	0.285	1.608	0.161
	Delusional Score PLEs (1-year follow-up)	0.086	-7.463	3.262	-2.288	0.05625	2.631	0.01
	Delusional Score PLEs (2-year follow-up)	0.043	-9.512	4.886	-1.947	0.0936	3.493	0.001
	Hallucinational Score PLEs (1-year follow-up)	0.022	-8.917	4.123	-2.163	0.062	1.878	0.083
	Hallucinational Score PLEs (2-year follow-up)	0.146	-9.382	5.681	-1.651	0.162	2.695	0.009
Integrated Model	Distress Score PLEs (1-year follow-up)	0.011	-8.922	2.922	-3.053	0.0252	8.388	< 0.001	*
	Distress Score PLEs (2-year follow-up)	0.074	-9.343	3.469	-2.693	0.0252	7.361	< 0.001
	Delusional Score PLEs (1-year follow-up)	0.172	-6.946	2.460	-2.824	0.0252	6.816	< 0.001
	Delusional Score PLEs (2-year follow-up)	0.136	-6.854	2.617	-2.619	0.027	7.525	< 0.001
	Hallucinational Score PLEs (1-year follow-up)	0.038	-8.246	2.834	-2.910	0.0252	7.182	< 0.001	*
	Hallucinational Score PLEs (2-year follow-up)	0.089	-8.355	3.070	-2.721	0.0252	9.484	< 0.001

To elucidate the role of specific genetic and neural correlates within the heterogeneous effects of ADI on PLEs, we obtained Shapley additive explanation (SHAP) scores⁷⁸. SHAP scores provide insights into how each variable contributes positively or negatively to the differential effects of ADI on 1-year follow-up observations of distress score and hallucinational score PLEs. These scores help differentiate the roles of these factors across subgroups, ranging from low to high conditional average treatment effects, thereby providing a nuanced understanding of how ADI influences PLEs through various genetic and neural pathways.

In both distress score and hallucinational score PLEs, children who showed higher levels of ADI’s adverse effects on PLEs exhibited distinct neuroanatomical and functional brain patterns, particularly in the limbic system. These patterns included reduced neuroanatomical features such as smaller white matter and surface area in the right temporal pole, reduced area and volume in the right parahippocampal region, decreased left white surface area, smaller area in the right isthmus cingulate, reduced intracranial volume, smaller caudate nucleus volume, and lower total grey matter volume. Functionally, greater activation during MID tasks was observed in several areas including the posterior cingulate, right ventral diencephalon, right insula, left thalamus proper, and left precentral gyrus. Additionally, children more adversely affected by ADI, as indicated by higher conditional average treatment effects on PLEs, exhibited larger right fusiform volume, decreased activation in the left superior temporal gyrus, younger parental age, and lower BMI (Fig. 4).

The analysis also revealed that higher conditional average treatment effects on distress score PLEs was associated with higher cognitive performance GPS and a lower likelihood of being Hispanic. In contrast, for hallucinational score PLEs, greater importance was attributed to increased activation in the left supramarginal gyrus during MID tasks and more pronounced discounting of future rewards. These nuanced associations are depicted in Fig. 4.

Lastly, we conducted a supplementary analysis to test whether the effects of delay discounting between the impact of ADI on PLEs are captured with a conventional linear mediation model⁷⁹. This linear IV mediation model showed no significant mediation effects of delay discounting (β= -6.929E-6 [95% CI, -0.012 ~ 0.026] ~ 4.582E-6 [95% CI, -0.009 ~ 0.03]) (Supplementary Table 7).

In this study, we examined how neighborhood socioeconomic deprivation impacts children’s intertemporal choice behavior (delay discounting) and PLEs, considering the multifaceted effects of neighborhood adversity and its underlying biological, environmental, and behavioral drivers. Our findings can be distilled into two main points. Firstly, there was a notable link of living in socioeconomically disadvantaged neighborhoods to the propensity for children to prefer immediate rewards over larger, delayed ones—a behavior known as steep delay discounting (indicative of lower impulse control) and to a higher rate of PLEs. This association was significant even after adjusting for a range of confounding factors, both observed (e.g., familial socioeconomic status) and unobserved. Secondly, the influence of disadvantaged neighborhood environments on PLEs was found to be heterogeneous. This individual variability is influenced not just by delay discounting, but also by a confluence of factors including genetic predisposition for cognitive intelligence, and brain morphometry and functioning (task activation). Causal machine learning models utilized in our study have identified a spectrum of conditions that either exacerbate vulnerability or contribute to resilience, accounting for the diverse effects of neighborhood environments on children's PLEs.

Our findings hold implications for social science. Using causal machine learning models, such as IV Forest and Double ML, we provide consistent and clear results that residential adversity during childhood leads to steeper discounting of future rewards. We propose three possible interpretations to explain the effects of neighborhood socioeconomic adversity on individual’s intertemporal decision-making, focusing on individual’s discount rate, resource scarcity, and social trust.

The longstanding economic theory posits that an individual's rate of discounting future rewards (time preference) is an exogenous parameter of intertemporal choice, established a priori, and impervious to external influences³². Since the introduction of the discounted utility model⁸⁰ by Paul Samuelson in 1937, there has been limited exploration into whether environmental factors affect the development of an individual's parameter^32,81. Our study challenges this notion by offering concrete evidence that the development of an individual's time preference is subject to environmental influences, and thereby opening new avenues for understanding the dynamics of intertemporal decision-making.

Our second interpretation explores the cognitive impact of resource scarcity. Limited resources may overload cognitive capacity, diverting attention from long-term planning and precipitating poor financial decisions, such as impulsive purchasing and mismanagement of finances^82,83. The third interpretation emphasizes the role of social trust. Lack of trust and reliability in receiving promised future rewards may logically drive individuals to prefer immediate gratification^84,85.

These three interpretations, though seemingly distinct, converge in real-world contexts where socioeconomically disadvantaged families often face both resource scarcity and reduced social trust⁸⁶. This synthesis forms the basis of the ‘behavioral poverty trap’, wherein individuals raised in impoverished environments are prone to overvalue immediate rewards, leading to myopic behaviors such as overconsumption, inadequate savings^30,82, and heightened risk of psychiatric disorders including psychosis and addiction^34,54. These behaviors, in turn, perpetuate socioeconomic challenges and hinder escape from poverty^33,87.

We build on this behavioral poverty trap framework by identifying the potential causal influence of neighborhood environment on intertemporal choice, leveraging longitudinal observations of preadolescent children aged 9–12 years, a critical period for neurocognitive development. A plausible biological mechanism for this phenomenon is the effects of glucocorticoid on brain’s reward system. Prior studies indicate that adverse social environments induce chronic stress to children, elevating glucocorticoid hormones like cortisol^88–92. In particular, neighborhood socioeconomic deprivation has a more pronounced association with cortisol increases in children compared to any other social environmental factors⁹³. Long-term chronic stress from growing up in disadvantaged neighborhoods could result in epigenetic modifications affecting the mesocorticolimbic dopaminergic system, thereby altering the reward system^91,92. This alteration may lead to a heightened preference for immediate rewards and impulsive behaviors, such as unhealthy eating and substance abuse^{90–92,94−98}, further entrenching the cycle of socioeconomic disadvantage.

Our second findings extend this understanding by linking the heterogeneous effects of ADI on children’s PLEs with the intricate relationship between childhood social adversity and the reward system. Our findings suggest that these differential effects of neighborhood socioeconomic adversity are modulated by genetic predispositions and neurodevelopmental traits associated with delay discounting. Children who experience residential deprivation and are at a higher PLEs demonstrate several distinct characteristics, including lower BMI, younger parental age, and altered brain structures and functions associated with delay discounting. Notably, these children showed reduced volume or white matter in specific brain regions (right temporal pole, right parahippocampal gyrus, right caudate nucleus, right isthmus cingulate), along with a smaller intracranial and total grey matter volume. Functionally, these children showed greater activation during MID tasks in regions including the right posterior cingulate, right ventral diencephalon, right insula, left precentral gyrus, left thalamus proper, and left superior temporal. This is particularly pronounced in children with a greater propensity for hallucinatory symptoms, who also show increased activity in the left supramarginal gyrus.

It appears that variations in structural and functional aspects of the limbic system (the posterior cingulate, ventral diencephalon, insula, temporal pole, parahippocampal gyrus, and isthmus cingulate) play a crucial role in how socioeconomic hardship affects PLEs. This individual variability may be linked to individual differences in the glucocorticoid and reward system. The interaction between our genes and neurons, in response to chronic stress from poor socioeconomic conditions, may determine the differing impacts of such adversity on PLEs.

Although direct testing of this association within the ABCD Study samples was not feasible due to lack of relevant data, extensive animal and human corroborate our hypotheses. These studies suggest that maladaptive valuation of intertemporal rewards, namely the excessive discounting of future rewards, is linked to dysfunction of the prefrontal-limbic system, associated with psychopathologies such as psychosis in adolescents and adults^{34,35,38–41,99}. Animal models have demonstrated that adverse social environments trigger chronic dysregulation of glucocorticoid signaling in the hypothalamic-pituitary-adrenal axis and the dopaminergic mesocortical circuit⁹², through epigenetic control^91,92. This dysregulation disrupts the adolescent reward circuit. In humans, childhood exposure to social adversity leads to changes in the hypothalamic-pituitary-adrenal axis and contributes to psychosis through abnormal neurodevelopment of the limbic regions, the temporal pole, cingulate cortices, parahippocampal gyrus, and caudate nucleus^100–103. Young adults with a history of childhood social deprivation often show impaired reward processing, particularly in the cingulate and mesostriatal dopaminergic system^25,103–105.

The age of our study’s participants, 9–12 years old, is a critical period for development of the prefrontal-limbic system^103,106,107. Children with psychotic disorders often exhibit greater reductions in grey matter compared to their healthy peers^108,109. These neurodevelopmental alterations are associated with increased neuronal excitation, reduced inhibitory neural activities, and the resultant impulsive behaviors¹¹⁰. In line with our findings, previous research has shown a correlation between higher PLEs and neuroanatomical alterations in the right temporal fusiform, right temporal pole, and right parahippocampal gyrus¹⁰², as well as greater neural activations in limbic regions such as the insula and cingulate cortices during reward outcomes in MID task¹¹¹. Overall, our findings on the heterogeneous effects of neighborhood deprivation contribute to the growing body of literature showing the role of glucocorticoid and reward systems in modulating the adverse effects of environmental deprivation on psychosis^{101,103,105,112,113}.

In our study, we discovered that children, when exposed to deprived neighborhoods and already facing the challenges of residential disadvantage, were more likely to experience PLEs. Surprisingly, these children also showed a higher GPS for cognitive performance. At first glance, this finding seems to contradict prior research, which has consistently identified a negative relationship between PLEs and cognitive performance^28,114.

To understand this complex relationship, we turned to the bioecological model and the Scarr-Rowe hypothesis on gene-environment interactions^115–117. This theory proposes that the impact of genetic factors is lessened in unfavorable environments. An easy way to visualize this is by comparing it to plant growth: in poor soil, a plant can't get the nutrients it needs, which limits its growth despite its genetic potential to grow tall¹¹⁸. But, when these children face residential disadvantages, this protective gene-psychosis link weakens. Their genetic resilience decreases, making them more vulnerable to the negative impacts of such disadvantages on PLEs. Essentially, those with higher cognitive ability GPS lose more of their potential genetic protection, making them more susceptible to the adverse effects of their environment on PLEs.

Consistent with our findings, recent large-scale studies have demonstrated that the impact of genetics on brain structure, cognitive functions, and mental health disorders becomes less significant in harmful environments (e.g., abuse)^119,120. Conversely, in more supportive and enriched settings, like those associated with higher socioeconomic status, genetic influences are more noticeable (e.g., high socioeconomic status)^116,121,122. Together with these findings, our study contributes to a deeper understanding of how genetic and environmental factors interact to influence the development of psychopathology in children.

In this study, we utilized innovative causal machine learning techniques to test the negative impacts of neighborhood deprivation on childhood psychopathology. Specifically, we employed the IV Forest method that allows us to discern how residential deprivation influences children’s PLEs in a manner dependent on a variety of genetic risk factors (e.g., GPS of cognitive performance, educational attainment, and IQ^114,123,124) and environmental risk factors (e.g., family income^3,125), as identified in existing literature. Our findings were adjusted to account for potential biases from both observed and unobserved variables.

The machine learning algorithm we used was adept at modelling the complex interplay gene-environment interactions. Among the three IV Forest models we tested (i.e., Delay Discounting, Gene-Brain, Integrated), only the Integrated model—which included delay discounting, sociodemographic characteristics, and genetic and neural correlates of delay discounting—identified the significant heterogeneous effects of ADI on children’s PLEs. This suggests that the intricate interactions among environmental, genetic, neural factors, and delay discounting play a crucial role in how socioeconomic adversity impacts PLEs.

In contrast, traditional linear mediation analysis, which relies on predefined interaction terms in a deductive statistical framework, failed to identify any significant mediation effect of delay discounting between neighborhood deprivation and PLEs. This underscores the effectiveness of our advanced causal machine learning approach over conventional methods in detecting the subtle effects of various interacting factors on childhood psychopathology.

The IV Forest model represents a significant advancement over traditional analysis methods by enabling data-driven feature selection and the stratification of heterogeneous treatment effects^59,60. Unlike methods that rely on patterns predetermined by researchers, the IV Forest model inductively identifies complex and nonlinear interactions, providing a deeper and more nuanced understanding of the data. Traditional deductive approaches often suffer from low statistical power and bias^61,126, which inadequately capture the complexity of gene-environment interactions^57,58. For instance, employing conventional linear regression to model interactions among the 45 covariates in our Integrated model would necessitate the inclusion of over 35 trillion interaction terms. This is not only impractical due to its complexity but also prone to issues like reduced statistical power, poor interpretability, and collinearity.

Given these challenges, we believe that causal modeling approaches that assess heterogeneous treatment effects based on machine learning hold significant potential as powerful tools for advancing precision science in psychology and medicine. These approaches provide a more dynamic and accurate framework for understanding the multifaceted influences on psychopathology, demonstrating significant promise for future research in these fields.

Several limitations of this study warrant consideration. Firstly, we used ABCD Study, a non-randomized, observational cohort. Despite employing IV methods, including IV Forest and DoubleML, to adjust for both observed and potential unobserved confounders, the inherent limitation of the exclusion restriction assumption persists. This assumption, critical to the validity of the IV methods, cannot be directly verified with data. Albeit we substantiated this assumption with extensive prior research discussed in the Methods section, its validity may still be subject to scrutiny, as might the overall efficacy of the IV method in fully adjusting for residual confounding bias. To mitigate this, we calculated E-values for the average treatment effects of neighborhood disadvantage on delay discounting and PLEs. The large E-values calculated indicate that it would require unobserved confounders with a significantly strong association with both the exposure and the outcomes to negate our findings. Given the magnitude of these E-values and our comprehensive adjustments for confounding, it is unlikely that unobserved confounding could fully account for the observed relationships, thereby supporting the potential causal interpretations, despite not providing absolute proof of causality.

Secondly, since the majority of participants identified their race/ethnicity as white (63.76%, similar to the US population), the generalizability of our findings to other minor race/ethnicity might remain to be tested. Nonetheless, recent research suggests that temporal discounting measures are consistent across diverse populations worldwide (61 countries, N = 13,629)¹²⁷, which may mitigate concerns regarding the representativeness of our findings. Thirdly, the relatively short follow-up periods in our study (1-year and 2-year follow-up) may not adequately capture the long-term neurodevelopmental processes underlying intertemporal valuation and related psychopathology. Notably, additional follow-up data from the ABCD Study became available after we finalized this manuscript. As the ABCD Study continues to collect more longitudinal observations, longer follow-up periods in future studies could yield deeper insights. Fourthly, despite efforts to ensure representativeness by recruiting from diverse school systems across 21 research sites in the United States, our sample does not fully mirror the entire US population¹²⁸. To address this, we provide a supplementary table (Supplementary Table 1) comparing the demographic characteristics of our final sample with the general United States population enhancing the relevance and generalizability of our results. Lastly, future research should examine the heterogeneous effects of additional environmental risk factors—such as parenting behavior²⁸ and early life stress¹²⁰—as primary exposures to elucidate their potential causal effects on psychiatric disorders. Investigating how genetic and neural correlates interact with these risk factors will also advance our understanding of their unique contributions to individual differences in psychopathology.

This study highlights the differential effects of neighborhood disadvantage on intertemporal economic decisions and PLEs during early childhood. It underscores the importance of identifying diverse treatment effects by integrating genetic and environmental factors to guide personalized healthcare approaches. Furthermore, we propose that enhancing the childhood environment could contribute to the reduction of economic and health inequality gaps. Economic policies promoting positive intertemporal choice (e.g., increased savings, healthy diet) have predominantly focused on paternalistic welfare policies in adulthood. These policies often assume that an individual’s tendency to discount future rewards is fixed (“exogenous”)³². However, our findings suggest that policies or interventions aimed at enhancing the socioeconomic environment during childhood may foster improved intertemporal choice behavior, thereby reducing economic³³ and health inequality^23,129. By addressing the root of the problem, this indirect approach may assist individuals in developing the capacity to make more informed choices, ultimately promoting better outcomes.

The insights gleaned from our novel analytical methods revive longstanding philosophical inquiries: do humans possess reason or free will independent of their environment? If our ability to act responsibly is indeed shaped by external circumstances, this challenges the traditional rationale for penalizing criminal and morally objectionable behavior based on the assumption of free will. This inquiry underscores the need for further interdisciplinary research, bridging insights from psychology, sociology, neuroscience, ethics, and law, to explore the nuanced relationship between individual agency and environmental influences. Such research is crucial for understanding how external factors impact decision-making and behavior, thereby informing more nuanced approaches to ethical and legal accountability. It invites a reevaluation of responsibility and justice, suggesting that effective interventions and policies must consider the complex interplay of individual predispositions and environmental conditions in shaping behavior.

Study Participants

The ABCD Study recruited participants from 21 research sites across the nation, utilizing a stratified, probability sampling method to capture the sociodemographic variation of the US population¹³⁰. We used the baseline, first year, and second year follow-up datasets included in ABCD Release 4.0, downloaded on February 10, 2022.

Of the initial 11,876 ABCD samples, we removed participants without genotype data, MRI data, NIH Toolbox Cognitive Battery, delay discounting, residential address, ADI, and PLEs. Participants not meeting the ABCD Study’s MRI quality control standards were also excluded. As recommended by the ABCD team¹³¹, Johnson & Bickel’s two-part validity criterion¹³² was used to exclude subjects with inconsistent responses (i.e., indifferent point for a given delay larger than that of an indifference point for a longer delay). Missing values of covariates were imputed using k-nearest neighbors. The final samples included 2,135 children from a variety of race/ethnic groups.

Data

Neighborhood Disadvantage

Neighborhood disadvantage was measured with Residential History Derived Scores based on the Census tracts of each respondent’s primary addresses by the ABCD team. Consistent with prior research^3,44, we chose national percentile scores of the Area Deprivation Index (ADI) in baseline year, calculated from the 2011 ~ 2015 American Community Survey 5-year summary. It has 17 sub-scores regarding various socioeconomic factors such as median household income, income disparity, percentage of population aged more than 25 years or more with at least a high school diploma, and percentage of single-parent households with children aged less than 18 years, etc. Higher values of the ADI indicate greater neighborhood disadvantage.

Delay Discounting

Delay discounting was measured by the adjusting delay discounting task in the 1-year follow-up ABCD data^131,133. Each child was asked to make choices between a small immediate hypothetical reward or a larger hypothetical $100 delayed reward at multiple future time points (6h, one day, one week, one month, three months, one year, and five years). By increasing or decreasing the smaller immediate reward depending on the child’s response, the task records the indifference point (i.e., the small immediate amount deemed to have the same subjective value as the $100 delayed reward) at each of the seven delay intervals. Test-retest reliability of this delay discounting measure has been validated^134,135. Studies show that preadolescent children are capable of comprehending the delay discounting task and show similar patterns of discounting as adults¹³⁶.

To avoid methodological problems regarding mathematical discounting models (hyperbolic vs. exponential) and positively skewed parameters of discounting functions^135,137, we used the area under the curve, a model-free measure of delay discounting¹³⁷. The area under the curve measure of delay discounting rates (henceforth discount rates) ranges from 0 to 1, with lower values indicating steeper discounting and higher impulsivity.

Psychotic-Like Experiences

First and second-year follow-up observations of psychotic-like experiences (PLEs) were measured using the Prodromal Questionnaire-Brief Child Version (PQ-BC; child-reported). PQ-BC has a 21-item scale validated for use with a non-clinical population of children aged 9–10 years^138,139. In line with the previous research^{3,123,138,139}, we computed Total Score and Distress Score, each indicating the number of psychotic-like symptoms and levels of total distress. Total Score is the summary score of 21 questions ranging from 0 to 21, and Distress Score is the weighted sum of responses with the levels of distress, ranging from 0 to 126. Additionally, to test whether the heterogeneous treatment effects of neighborhood adversity differ among psychotic symptoms, Distress Score was divided into two separate scores: Delusional Score and Hallucinational Score^2,140. A higher value indicates greater severity of PLEs.

Genome-wide Polygenic Scores

Children’s genetic predispositions were assessed with genome-wide polygenic scores (GPS). Summary statistics from genome-wide association studies were used to generate GPS of cognitive intelligence (cognitive performance¹⁴¹, education attainment¹⁴¹, IQ¹⁴²), psychiatric disorders (major depressive disorder¹⁴³, post-traumatic stress disorder¹⁴⁴, attention-deficit/hyperactivity disorder¹⁴⁵, obsessive-compulsive disorder¹⁴⁶, anxiety¹⁴⁷, depression¹⁴⁸, bipolar disorder¹⁴⁹, autism spectrum disorder¹⁵⁰, schizophrenia¹⁵¹, cross disorder¹⁵²), and health and behavioral traits (BMI¹⁵³, neuroticism¹⁵⁴, worrying¹⁵⁴, risk tolerance¹⁵⁵, automobile speeding propensity¹⁵⁵, eating disorder¹⁵⁶, drinking¹⁵⁵, smoking¹⁵⁵, cannabis use¹⁵⁷, general happiness¹⁵⁸, snoring¹⁵⁹, insomnia¹⁵⁹, alcohol dependence¹⁶⁰). PRS-CSx, a high-dimensional Bayesian regression framework that places continuous shrinkage prior on single nucleotide polymorphisms effect sizes¹⁶¹, was applied to enhance cross-population prediction. This method has consistently shown superior performance compared to other methods across a wide range of genetic architectures in simulation and real data analyses¹⁶¹. Hyperparameter optimization for the GPSs was conducted using a held-out validation set of 1,579 unrelated participants. Adjustments for population stratification were performed based on the first ten ancestrally informative principal components to account for potential confounding effects.

Anatomical Brain Imaging: T1/T2, Freesurfer 6

Baseline year T1-weighted (T1w) 3D structural MRI acquired in the ABCD study were processed following established protocols^162,163: To maximize geometric accuracy and image intensity reproducibility, gradient nonlinearity distortion was corrected¹⁶⁴. After correcting intensity nonuniformity using tissue segmentation and spatial smoothing, images were resampled to 1 mm isotropic voxels. We used Freesurfer v6.0 (https://surfer.nmr.mgh.harvard.edu) for the following procedures: cortical surface followed by skull-stripping¹⁶⁵, white matter segmentation, and mesh creation¹⁶⁶, correction of topological defects, surface optimization¹⁶⁷, and nonlinear registration to a spherical surface-based atlas^168,169. Using Desikan–Killiany atlas¹⁷⁰, a standard atlas for Freesurfer and ABCD study, we extracted 399 brain ROI measures, including volumes, surface area, thickness, mean curvature, sulcal depth, and gyrification.

Functional MRI (fMRI): Monetary Incentive Delay (MID) task

The MID task was used measure the neural activation during anticipation and receipt of monetary gains and losses. In each trial, participants were shown a graphical cue of the 5 possible incentive types: large reward ($5), small reward ($0.20), large loss (-$5), small loss (-$0.20), or neutral ($0). The incentive cue is presented for 2,000 ms, followed by a jittered anticipatory delay (1,500–4,000 ms). Subsequently, a target to which participants respond to gain or avoid losing money was shown (150–500 ms), and feedback of their performance was provided (2,000 ms). A total of 40 reward, 40 loss, and 20 neutral trials were presented in pseudo-random order across the two task runs. Task parameters was dynamically manipulated for each subject to maintain 60% success rate¹⁶². We used baseline year observations of average beta weights of the MID task fMRI with Desikan-Killiany parcellations¹⁷⁰.

Covariates

To adjust for the potential confounding effects, sociodemographic covariates were included. Consistent with existing research on psychiatric disorders in ABCD samples^{3,123,138,171}, we controlled for the child’s sex, age, race/ethnicity, caregiver’s relationship to a child, BMI, parental education, marital status of the caregiver, household income, parent’s age, and family history of psychiatric disorders. The family history of psychiatric disorders, measured as the proportion of first-degree relatives who experienced psychosis, depression, mania, suicidality, previous hospitalization, or professional help for mental health issues³ was included as a covariate. Given that delay discounting and PLEs are associated with an individual's neurocognitive capabilities^172–174, NIH Toolbox total intelligence was used as a covariate. All covariates were from baseline year observations.

Statistical Analyses

Instrumental Variable Regression

The IV method controls unobserved confounding bias by utilizing an instrumental variable Z which affects the treatment/exposure variable of interest X but has no direct effect on the outcome variable Y⁶². We tested the endogeneity of ADI (i.e., whether ADI as a treatment/exposure variable correlates with the error term), and found significant bias from unobserved confounding (all Hausman test¹⁷⁵ for differences, p ≤ 0.0158). This underscores the necessity of employing IV regression approach to control for the significant confounding effects and to test the potential causal relationship of neighborhood disadvantage with delay discounting and PLEs.

The IV method relies on two main assumptions: the exclusion restriction and the strong instrument. The exclusion restriction asserts that the instrumental variable impacts the outcome Y exclusively through the treatment/exposure X, conditioned on observed covariates. Although this assumption cannot be directly tested from data, its plausibility is typically drawn from prior research and theoretical underpinnings⁶². In our study, the instrument variable Z for the exposure of ADI was the presence of state-level source of income (SOI) laws at baseline assessment, which prohibit income discrimination in the housing market. SOI laws are designed to ensure that landlords cannot refuse housing vouchers, which are provided to low-income families to assist in securing quality housing. Such legislation is critical because, despite the intention behind vouchers, many landlords prefer direct cash payments and might otherwise decline voucher-based payments. Thus, families residing in states with SOI laws are more likely to have better residential environments (i.e., lower neighborhood socioeconomic adversity).

Reports from the US Department of Housing and Urban Development indicate that SOI laws increase landlords’ acceptance of housing vouchers by 20.2%p to 59.3%p¹⁷⁶. Research links SOI laws with significant reductions in neighborhood poverty¹⁷⁷ and improved health outcomes in children, including lower hospitalization rates, less impulsive consumption¹⁷⁸, and substantially better mental health¹⁷⁹. Taken together, the presence of SOI laws may affect cognitive and psychiatric outcomes—particularly delay discounting and PLEs—solely by enhancing neighborhood environment, conditional on the individual sociodemographic characteristics such as family income, parental education, and race/ethnicity. This relationship supports the plausibility of the exclusion restriction, crucial for the validity of our IV method.

The second assumption requires the instrument variable to be strongly associated with the treatment/exposure X. F-statistic above ten is considered a strong instruments¹⁸⁰. The F-statistic for each model was F = 34.031 (p < 0.0001), suggesting that the instrument SOI laws is strongly associated with the treatment ADI. In other words, our IV model is not likely to suffer from weak instrument bias.

All continuous variables were standardized (z-scaled), and analyses were run using ivreg¹⁸¹ in R version 4.1.2. For all analyses in our study, threshold for statistical significance was set at two-tailed p < 0.05, with multiple comparison correction based on false discovery rate.

Causal Machine Learning for Treatment Effects

IV Forest (grf R package version 2.2.1)^59,60 is a novel causal machine learning approach extends from the conventional random forest framework¹⁸² with recursive partitioning, subsampling, and random splitting to identify the average treatment effects and its individual differences.

Initially, the IV Forest randomly splits the dataset into two independent subsets, S and T. Subset S is dedicated solely to the construction of individual trees within the forest, where each tree explores potential divisions—such as "Race/ethnicity = White"—to split S into groups ${S}_{1}$ and ${S}_{2}$, based on the fulfillment of the specified criteria. The selection of these splits is strategically chosen to maximize the differences in conditional average treatment effects estimates between the groups ${S}_{1}$ and ${S}_{2}$. Following the construction phase, subset T, independent of S, is employed for model validation. Fresh observations from T are introduced into groups with similar treatment responses by each tree. The aggregation of results from multiple trees is conducted through local weighting method, aimed at reducing overall estimate variance and improving accuracy. Using separate subsets for tree building and model validation ensures an honest estimation of conditional average treatment effects, systematically reducing the risk of overfitting⁶⁷.

Using IV Forests, we obtained augmented inverse propensity weighted estimates of average treatment effects, a doubly-robust estimator which can capture complex patterns of individual differences and do not rely on a priori model assumptions⁵⁹ such as linearity. This is particularly advantageous when the relationship between environmental variables and neurocognitive development is likely nonlinear^7,183,184. To measure the average outcome between treated versus untreated subjects, ADI was binarized (i.e., mean split).

In line with prior studies^75,76, we evaluated heterogeneous treatment effects by testing whether the average treatment effects are significantly different among subgroups defined by their relative resilience/vulnerability⁷⁷. These subgroups were defined across a decile spectrum, with Q1 representing the most vulnerable and Q10 the most resilient. We considered a model to have significant heterogeneous treatment effects only if it satisfied all three of the following criteria:

The monotonicity test evaluates the existence of at least one inequality in the average treatment effects across the deciles. This is achieved by whether to accept or reject the null hypothesis (${\mathcal{H}}_{0})$, which states that treatment effects are equal across all deciles. Essentially, the test determines whether there is a consistent, ordered relationship in the treatment effects from one decile to the next, indicating a monotonic trend.

The alternative hypothesis test evaluates whether the average treatment effect in the highest decile exceeds the combined average treatment effects in the remaining deciles Q2 through Q10$.$

The ANOVA test determines whether the average treatment effects are statistically different across deciles. In this context, the group mean in the ANOVA corresponds to the average treatment effect of each decile.

To ensure that the IV Forest estimations are robust across different random seeds, we developed 100-seed ensemble IV Forest model. Specifically, we used the following procedures:

For each iteration, randomly split the data in half (i.e., train vs test sets) to build a forest model with the first half and perform estimation with the other half. We repeated this process 100 times using different seeds in each iteration to build 100 forest models.

Combine the 100 forest models into one big IV Forest model and then rank the observations into deciles according to their estimated conditional average treatment effects.

Obtain augmented inverse propensity weighted average treatment effects for each decile and perform monotonicity, alternative hypothesis, and ANOVA tests.

Data availability

The ABCD Study dataset is openly available to all eligible researchers upon the submission of an access request via the National Institutes of Mental Health Data Archive (https://nda.nih.gov/abcd). Comprehensive written informed consent was obtained from the parents of participants, with children providing assent. The study protocols were approved by the University of California, San Diego's Institutional Review Board (IRB), under approval number 160091, in addition to receiving approval from the IRBs of the 21 participating data collection sites¹⁸⁵.

Code availability

All codes needed to replicate the results can be found at https://github.com/Transconnectome/DD-HTE.

Author Contributions:

J.P. and J.C. designed research; J.P. and M.C. performed research; J.P., M.C., E.L., B.-G.K., G.K., Y.Y.J. analyzed data; and J.P., M.C., and J.C. wrote the paper.

Competing Interest Statement:

Authors have no competing interests.

Kant, I. Critique of Practical Reason. (Hackett Publishing, 2002).
Karcher, N. R., Niendam, T. A. & Barch, D. M. Adverse childhood experiences and psychotic-like experiences are associated above and beyond shared correlates: Findings from the adolescent brain cognitive development study. Schizophrenia Research 222, 235–242 (2020). DOI: https://doi.org/10.1016/j.schres.2020.05.045
Karcher, N. R., Schiffman, J. & Barch, D. M. Environmental Risk Factors and Psychotic-like Experiences in Children Aged 9–10. Journal of the American Academy of Child & Adolescent Psychiatry 60, 490–500 (2021). DOI: 10.1016/j.jaac.2020.07.003
Gur, R. E. et al. Burden of Environmental Adversity Associated With Psychopathology, Maturation, and Brain Behavior Parameters in Youths. JAMA Psychiatry 76, 966 (2019). DOI: 10.1001/jamapsychiatry.2019.0943
Tomasi, D. & Volkow, N. D. Associations of family income with cognition and brain structure in USA children: prevention implications. Molecular Psychiatry (2021). DOI: 10.1038/s41380-021-01130-0
Vargas, T., Damme, K. S. F. & Mittal, V. A. Neighborhood deprivation, prefrontal morphology and neurocognition in late childhood to early adolescence. NeuroImage 220, 117086 (2020). DOI: 10.1016/j.neuroimage.2020.117086
Hackman, D. A. et al. Association of Local Variation in Neighborhood Disadvantage in Metropolitan Areas With Youth Neurocognition and Brain Structure. JAMA Pediatrics 175, e210426 (2021). DOI: 10.1001/jamapediatrics.2021.0426
Ye, J. et al. Socioeconomic Deprivation Index Is Associated With Psychiatric Disorders: An Observational and Genome-wide Gene-by-Environment Interaction Analysis in the UK Biobank Cohort. Biol Psychiatry 89, 888–895 (2021). DOI: https://doi.org/10.1016/j.biopsych.2020.11.019
Bor, J., Cohen, G. H. & Galea, S. Population health in an era of rising income inequality: USA, 1980–2015. The Lancet 389, 1475–1490 (2017). DOI: 10.1016/s0140-6736(17)30571-8
Dennis, E., Manza, P. & Volkow, N. D. Socioeconomic status, BMI, and brain development in children. Transl Psychiatry 12 (2022). DOI: 10.1038/s41398-022-01779-3
Leventhal, T. & Brooks-Gunn, J. The neighborhoods they live in: The effects of neighborhood residence on child and adolescent outcomes. Psychological Bulletin 126, 309–337 (2000). DOI: 10.1037/0033-2909.126.2.309
Leventhal, T. & Brooks-Gunn, J. A Randomized Study of Neighborhood Effects on Low Income Children's Educational Outcomes. Developmental Psychology 40, 488–507 (2004). DOI: 10.1037/0012-1649.40.4.488
Oreopoulos, P. The Long-Run Consequences of Living in a Poor Neighborhood*. The Quarterly Journal of Economics 118, 1533–1575 (2003). DOI: 10.1162/003355303322552865
Campbell, F. et al. Early Childhood Investments Substantially Boost Adult Health. Science 343, 1478–1485 (2014). DOI: 10.1126/science.1248429
Heckman, J. J., Stixrud, J. & Urzua, S. The effects of cognitive and noncognitive abilities on labor market outcomes and social behavior. Journal of Labor Economics 24, 411–482 (2006). DOI: https://doi.org/10.1086/504455
Knudsen, E. I., Heckman, J. J., Cameron, J. L. & Shonkoff, J. P. Economic, neurobiological, and behavioral perspectives on building America’s future workforce. Proceedings of the National Academy of Sciences 103, 10155–10162 (2006). DOI: https://doi.org/10.1073/pnas.0600888103
Chetty, R. & Hendren, N. The Impacts of Neighborhoods on Intergenerational Mobility I: Childhood Exposure Effects*. The Quarterly Journal of Economics 133, 1107–1162 (2018). DOI: 10.1093/qje/qjy007
Chetty, R., Hendren, N. & Katz, L. F. The Effects of Exposure to Better Neighborhoods on Children: New Evidence from the Moving to Opportunity Experiment. Am Econ Rev 106, 855–902 (2016). DOI: 10.1257/aer.20150572
Fryer, R. G. & Katz, L. F. Achieving Escape Velocity: Neighborhood and School Interventions to Reduce Persistent Inequality. Am Econ Rev 103, 232–237 (2013). DOI: 10.1257/aer.103.3.232
Heckman, J. J. Skill Formation and the Economics of Investing in Disadvantaged Children. Science 312, 1900–1902 (2006). DOI: 10.1126/science.1128898
Lowe, C. J., Morton, J. B. & Reichelt, A. C. Adolescent Obesity and Dietary Decision Making—a Brain-Health Perspective. The Lancet Child & Adolescent Health 4, 388–396 (2020). DOI: 10.1016/S2352-4642(19)30404-3
Shonkoff, J. P. Leveraging the biology of adversity to address the roots of disparities in health and development. Proceedings of the National Academy of Sciences 109, 17302–17307 (2012). DOI: 10.1073/pnas.1121259109
Shonkoff, J. P. Capitalizing on Advances in Science to Reduce the Health Consequences of Early Childhood Adversity. JAMA Pediatrics 170, 1003 (2016). DOI: 10.1001/jamapediatrics.2016.1559
Zhang, H., Lee, Z. X., White, T. & Qiu, A. Parental and social factors in relation to child psychopathology, behavior, and cognitive function. Transl Psychiatry 10 (2020). DOI: 10.1038/s41398-020-0761-6
Pechtel, P. & Pizzagalli, D. A. Disrupted Reinforcement Learning and Maladaptive Behavior in Women With a History of Childhood Sexual Abuse. JAMA Psychiatry 70, 499 (2013). DOI: 10.1001/jamapsychiatry.2013.728
Oshri, A. et al. Socioeconomic hardship and delayed reward discounting: Associations with working memory and emotional reactivity. Developmental Cognitive Neuroscience 37, 100642 (2019). DOI: 10.1016/j.dcn.2019.100642
Kim-Spoon, J. et al. Longitudinal pathways linking family risk, neural risk processing, delay discounting, and adolescent substance use. Journal of Child Psychology and Psychiatry 60, 655–664 (2019). DOI: 10.1111/jcpp.13015
Park, J. et al. Gene-Environment Pathways to Cognitive Intelligence and Psychotic-Like Experiences in Children. eLife (2023). DOI: 10.7554/elife.88117.3
Yaple, Z. A. & Yu, R. Functional and Structural Brain Correlates of Socioeconomic Status. Cerebral Cortex 30, 181–196 (2020). DOI: 10.1093/cercor/bhz080
Epper, T. et al. Time Discounting and Wealth Inequality. Am Econ Rev 110, 1177–1205 (2020). DOI: 10.1257/aer.20181096
Epper, T. et al. Preferences predict who commits crime among young men. Proceedings of the National Academy of Sciences 119, e2112645119 (2022). DOI: 10.1073/pnas.2112645119
Ericson, K. M. & Laibson, D. in Handbook of Behavioral Economics: Applications and Foundations 1 Vol. 2 (eds B. Douglas Bernheim, Stefano DellaVigna, & David Laibson) Ch. Chapter 1, 1–67 (North-Holland, 2019).
Laajaj, R. Endogenous time horizon and behavioral poverty trap: Theory and evidence from Mozambique. Journal of Development Economics 127, 187–208 (2017). DOI: https://doi.org/10.1016/j.jdeveco.2017.01.006
Maia, T. V. & Frank, M. J. From reinforcement learning models to psychiatric and neurological disorders. Nature Neuroscience 14, 154–162 (2011). DOI: 10.1038/nn.2723
Juechems, K. & Summerfield, C. Where Does Value Come From? Trends in Cognitive Sciences 23, 836–850 (2019). DOI: https://doi.org/10.1016/j.tics.2019.07.012
Ahn, W.-Y. et al. Temporal discounting of rewards in patients with bipolar disorder and schizophrenia. Journal of Abnormal Psychology 120, 911–921 (2011). DOI: 10.1037/a0023333
Ermakova, A. O. et al. Abnormal reward prediction-error signalling in antipsychotic naive individuals with first-episode psychosis or clinical risk for psychosis. Neuropsychopharmacology 43, 1691–1699 (2018). DOI: 10.1038/s41386-018-0056-2
Amlung, M. et al. Delay Discounting as a Transdiagnostic Process in Psychiatric Disorders. JAMA Psychiatry 76, 1176 (2019). DOI: 10.1001/jamapsychiatry.2019.2102
Maia, T. V. & Frank, M. J. An Integrative Perspective on the Role of Dopamine in Schizophrenia. Biol Psychiatry 81, 52–66 (2017). DOI: https://doi.org/10.1016/j.biopsych.2016.05.021
Millard, S. J., Bearden, C. E., Karlsgodt, K. H. & Sharpe, M. J. The prediction-error hypothesis of schizophrenia: new data point to circuit-specific changes in dopamine activity. Neuropsychopharmacology 47, 628–640 (2022). DOI: 10.1038/s41386-021-01188-y
Zald, D. H. & Treadway, M. T. Reward Processing, Neuroeconomics, and Psychopathology. Annu Rev Clin Psychol 13, 471–495 (2017). DOI: 10.1146/annurev-clinpsy-032816-044957
Brown, H. E., Hart, K. L., Snapper, L. A., Roffman, J. L. & Perlis, R. H. Impairment in delay discounting in schizophrenia and schizoaffective disorder but not primary mood disorders. npj Schizophrenia 4, 9 (2018). DOI: 10.1038/s41537-018-0050-z
Wang, H., Lesh, T. A., Maddock, R. J., Fassbender, C. & Carter, C. S. Delay discounting abnormalities are seen in first-episode schizophrenia but not in bipolar disorder. Schizophrenia Research 216, 200–206 (2020). DOI: https://doi.org/10.1016/j.schres.2019.11.063
Rakesh, D., Seguin, C., Zalesky, A., Cropley, V. & Whittle, S. Associations between neighborhood disadvantage, resting-state functional connectivity, and behavior in the Adolescent Brain Cognitive Development (ABCD) Study®: Moderating role of positive family and school environments. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging (2021). DOI: https://doi.org/10.1016/j.bpsc.2021.03.008
Taylor, R. L., Cooper, S. R., Jackson, J. J. & Barch, D. M. Assessment of Neighborhood Poverty, Cognitive Function, and Prefrontal and Hippocampal Volumes in Children. JAMA Network Open 3, e2023774 (2020). DOI: 10.1001/jamanetworkopen.2020.23774
van der Steen, Y. et al. ‘False-positive’ self-reported psychotic experiences in the general population: an investigation of outcome, predictive factors and clinical relevance. Epidemiology and Psychiatric Sciences 28, 532–543 (2019). DOI: 10.1017/s2045796018000197
Van Os, J. & Reininghaus, U. Psychosis as a transdiagnostic and extended phenotype in the general population. World Psychiatry 15, 118–124 (2016). DOI: 10.1002/wps.20310
Cannon, M. et al. Prevalence of psychotic symptoms in childhood and adolescence: a systematic review and meta-analysis of population-based studies. Psychological Medicine 42, 1857–1863 (2012). DOI: 10.1017/S0033291711002960
Kelleher, I. & Cannon, M. Psychotic-like experiences in the general population: characterizing a high-risk group for psychosis. Psychological Medicine 41, 1–6 (2011). DOI: 10.1017/s0033291710001005
Poulton, R. et al. Children's Self-Reported Psychotic Symptoms and Adult Schizophreniform Disorder. Archives of General Psychiatry 57, 1053 (2000). DOI: 10.1001/archpsyc.57.11.1053
Derks, E. M., Thorp, J. G. & Gerring, Z. F. Ten challenges for clinical translation in psychiatric genetics. Nature Genetics 54, 1457–1465 (2022). DOI: 10.1038/s41588-022-01174-0
Kapur, S., Phillips, A. G. & Insel, T. R. Why has it taken so long for biological psychiatry to develop clinical tests and what to do about it? Molecular Psychiatry 17, 1174–1179 (2012). DOI: 10.1038/mp.2012.105
Lancaster, T. M. et al. Polygenic Risk of Psychosis and Ventral Striatal Activation During Reward Processing in Healthy Adolescents. JAMA Psychiatry 73, 852–861 (2016). DOI: 10.1001/jamapsychiatry.2016.1135
Owens, M. M. et al. Functional and structural neuroimaging studies of delayed reward discounting in addiction: A systematic review. Psychological Bulletin 145, 141–164 (2019). DOI: 10.1037/bul0000181
Hamburg, M. A. & Collins, F. S. The Path to Personalized Medicine. New England Journal of Medicine 363, 301–304 (2010). DOI: 10.1056/nejmp1006304
Kent, D. M. et al. The Predictive Approaches to Treatment effect Heterogeneity (PATH) Statement. Annals of Internal Medicine 172, 35–45 (2019). DOI: 10.7326/M18-3667
Border, R. et al. No Support for Historical Candidate Gene or Candidate Gene-by-Interaction Hypotheses for Major Depression Across Multiple Large Samples. American Journal of Psychiatry 176, 376–387 (2019). DOI: 10.1176/appi.ajp.2018.18070881
Figlio, D. N., Freese, J., Karbownik, K. & Roth, J. Socioeconomic status and genetic influences on cognitive development. Proceedings of the National Academy of Sciences 114, 13441–13446 (2017). DOI: 10.1073/pnas.1708491114
Athey, S., Tibshirani, J. & Wager, S. Generalized random forests. The Annals of Statistics 47, 1148–1178 (2019). DOI: 10.1214/18-AOS1709
Wager, S. & Athey, S. Estimation and Inference of Heterogeneous Treatment Effects using Random Forests. Journal of the American Statistical Association 113, 1228–1242 (2018). DOI: 10.1080/01621459.2017.1319839
Akimova, E. T., Breen, R., Brazel, D. M. & Mills, M. C. Gene-environment dependencies lead to collider bias in models with polygenic scores. Scientific Reports 11 (2021). DOI: 10.1038/s41598-021-89020-x
Angrist, J. D., Imbens, G. W. & Rubin, D. B. Identification of Causal Effects Using Instrumental Variables. Journal of the American Statistical Association 91, 444–455 (1996). DOI: 10.1080/01621459.1996.10476902
Angrist, J. D. & Krueger, A. B. Instrumental Variables and the Search for Identification: From Supply and Demand to Natural Experiments. Journal of Economic Perspectives 15, 69–85 (2001). DOI: 10.1257/jep.15.4.69
Grosz, M. P., Rohrer, J. M. & Thoemmes, F. The Taboo Against Explicit Causal Inference in Nonexperimental Psychology. Perspectives on Psychological Science 15, 1243–1255 (2020). DOI: 10.1177/1745691620921521
Marinescu, I. E., Lawlor, P. N. & Kording, K. P. Quasi-experimental causality in neuroscience and behavioural research. Nature Human Behaviour 2, 891–898 (2018). DOI: 10.1038/s41562-018-0466-5
Ohlsson, H. & Kendler, K. S. Applying Causal Inference Methods in Psychiatric Epidemiology. JAMA Psychiatry 77, 637 (2020). DOI: 10.1001/jamapsychiatry.2019.3758
Athey, S. & Imbens, G. Recursive partitioning for heterogeneous causal effects. Proceedings of the National Academy of Sciences 113, 7353–7360 (2016). DOI: 10.1073/pnas.1510489113
VanderWeele, T. J. & Ding, P. Sensitivity Analysis in Observational Research: Introducing the E-Value. Annals of Internal Medicine 167, 268–274 (2017). DOI: 10.7326/M16-2607
Bach, P., Chernozhukov, V., Kurz, M. S. & Spindler, M. DoubleML-An Object-Oriented Implementation of Double Machine Learning in Python. Journal of Machine Learning Research 23, 1–6 (2022). DOI: https://www.jmlr.org/papers/v23/21-0862.html
Chernozhukov, V. et al. Double/debiased machine learning for treatment and structural parameters. The Econometrics Journal 21, C1-C68 (2018). DOI: 10.1111/ectj.12097
Kursa, M. B. & Rudnicki, W. R. Feature Selection with the Boruta Package. Journal of Statistical Software 36, 1–13 (2010). DOI: 10.18637/jss.v036.i11
Sveinbjornsson, G. et al. Multiomics study of nonalcoholic fatty liver disease. Nature Genetics 54, 1652–1663 (2022). DOI: 10.1038/s41588-022-01199-5
Wang, Z., Zhou, X., Gui, Y., Liu, M. & Lu, H. Multiple measurement analysis of resting-state fMRI for ADHD classification in adolescent brain from the ABCD study. Transl Psychiatry 13, 45 (2023). DOI: 10.1038/s41398-023-02309-5
Ball, G. et al. Machine-learning to characterise neonatal functional connectivity in the preterm brain. NeuroImage 124, 267–275 (2016). DOI: https://doi.org/10.1016/j.neuroimage.2015.08.055
Goligher, E. C. et al. Heterogeneous Treatment Effects of Therapeutic-Dose Heparin in Patients Hospitalized for COVID-19. JAMA (2023). DOI: 10.1001/jama.2023.3651
Athey, S. & Wager, S. Estimating Treatment Effects with Causal Forests: An Application. Observational Studies 5, 37–51 (2019). DOI: 10.1353/obs.2019.0001
Chernozhukov, V., Demirer, M., Duflo, E. & Fernández-Val, I. Generic Machine Learning Inference on Heterogeneous Treatment Effects in Randomized Experiments, with an Application to Immunization in India. National Bureau of Economic Research Working Paper Series No. 24678 (2018). DOI: 10.3386/w24678
Lundberg, S. M. et al. From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence 2, 56–67 (2020). DOI: 10.1038/s42256-019-0138-9
Tingley, D., Yamamoto, T., Hirose, K., Keele, L. & Imai, K. mediation: R package for causal mediation analysis. Journal of Statistical Software 59 (2014). DOI: https://doi.org/10.18637/jss.v059.i05
Samuelson, P. A. A Note on Measurement of Utility. The Review of Economic Studies 4, 155–161 (1937). DOI: 10.2307/2967612
Becker, G. S. & Mulligan, C. B. The Endogenous Determination of Time Preference. The Quarterly Journal of Economics 112, 729–758 (1997). DOI: 10.1162/003355397555334
Shah, A. K., Shafir, E. & Mullainathan, S. Scarcity Frames Value. Psychol Sci 26, 402–412 (2015). DOI: 10.1177/0956797614563958
Mani, A., Mullainathan, S., Shafir, E. & Zhao, J. Poverty Impedes Cognitive Function. Science 341, 976–980 (2013). DOI: 10.1126/science.1238041
Kidd, C., Palmeri, H. & Aslin, R. N. Rational snacking: Young children’s decision-making on the marshmallow task is moderated by beliefs about environmental reliability. Cognition 126, 109–114 (2013). DOI: https://doi.org/10.1016/j.cognition.2012.08.004
Jachimowicz, J. M., Chafik, S., Munrat, S., Prabhu, J. C. & Weber, E. U. Community trust reduces myopic decisions of low-income individuals. Proceedings of the National Academy of Sciences 114, 5401–5406 (2017). DOI: 10.1073/pnas.1617395114
Putnam, R. D., Leonardi, R. & Nanetti, R. Y. Making Democracy Work: Civic Traditions in Modern Italy. (Princeton University Press, 1994).
Adamkovič, M. & Martončik, M. A Review of Consequences of Poverty on Economic Decision-Making: A Hypothesized Model of a Cognitive Mechanism. Frontiers in Psychology 8 (2017). DOI: 10.3389/fpsyg.2017.01784
Johnson, L. E. et al. Patterns of poverty across adolescence predict salivary cortisol stress responses in Mexican-origin youths. Psychoneuroendocrinology 132, 105340 (2021). DOI: https://doi.org/10.1016/j.psyneuen.2021.105340
Fernald, L. C. H. & Gunnar, M. R. Poverty-alleviation program participation and salivary cortisol in very low-income children. Social Science & Medicine 68, 2180–2189 (2009). DOI: https://doi.org/10.1016/j.socscimed.2009.03.032
Anselme, P. & Güntürkün, O. How foraging works: Uncertainty magnifies food-seeking motivation. Behavioral and Brain Sciences 42, e35 (2019). DOI: 10.1017/S0140525X18000948
Barik, J. et al. Chronic Stress Triggers Social Aversion via Glucocorticoid Receptor in Dopaminoceptive Neurons. Science 339, 332–335 (2013). DOI: 10.1126/science.1226767
Niwa, M. et al. Adolescent Stress–Induced Epigenetic Control of Dopaminergic Neurons via Glucocorticoids. Science 339, 335–339 (2013). DOI: 10.1126/science.1226931
Malanchini, M. et al. Weak and uneven associations of home, neighborhood, and school environments with stress hormone output across multiple timescales. Molecular Psychiatry 26, 4823–4838 (2021). DOI: 10.1038/s41380-020-0747-z
Cabib, S. & Puglisi-Allegra, S. The mesoaccumbens dopamine in coping with stress. Neuroscience & Biobehavioral Reviews 36, 79–89 (2012). DOI: https://doi.org/10.1016/j.neubiorev.2011.04.012
Oswald, L. M. et al. Relationships Among Ventral Striatal Dopamine Release, Cortisol Secretion, and Subjective Responses to Amphetamine. Neuropsychopharmacology 30, 821–832 (2005). DOI: 10.1038/sj.npp.1300667
Douma, E. H. & de Kloet, E. R. Stress-induced plasticity and functioning of ventral tegmental dopamine neurons. Neuroscience & Biobehavioral Reviews 108, 48–77 (2020). DOI: https://doi.org/10.1016/j.neubiorev.2019.10.015
Mizoguchi, A. et al. Glucocorticoid receptor signaling in ventral tegmental area neurons increases the rewarding value of a high-fat diet in mice. Scientific Reports 11, 12873 (2021). DOI: 10.1038/s41598-021-92386-7
Barrot, M. et al. The dopaminergic hyper-responsiveness of the shell of the nucleus accumbens is hormone-dependent. European Journal of Neuroscience 12, 973–979 (2000). DOI: https://doi.org/10.1046/j.1460-9568.2000.00996.x
Vosberg, D. E., Leyton, M. & Flores, C. The Netrin-1/DCC guidance system: dopamine pathway maturation and psychiatric disorders emerging in adolescence. Molecular Psychiatry 25, 297–307 (2020). DOI: 10.1038/s41380-019-0561-7
DeRosse, P. & Barber, A. D. Overlapping Neurobiological Substrates for Early-Life Stress and Resilience to Psychosis. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 6, 144–153 (2021). DOI: https://doi.org/10.1016/j.bpsc.2020.09.003
Howes, O. D., McCutcheon, R., Owen, M. J. & Murray, R. M. The Role of Genes, Stress, and Dopamine in the Development of Schizophrenia. Biol Psychiatry 81, 9–20 (2017). DOI: https://doi.org/10.1016/j.biopsych.2016.07.014
Tognin, S. et al. Using Structural Neuroimaging to Make Quantitative Predictions of Symptom Progression in Individuals at Ultra-High Risk for Psychosis. Frontiers in Psychiatry 4 (2014). DOI: 10.3389/fpsyt.2013.00187
Howes, O. D. & Shatalina, E. Integrating the Neurodevelopmental and Dopamine Hypotheses of Schizophrenia and the Role of Cortical Excitation-Inhibition Balance. Biol Psychiatry 92, 501–513 (2022). DOI: https://doi.org/10.1016/j.biopsych.2022.06.017
Birn, R. M., Roeber, B. J. & Pollak, S. D. Early childhood stress exposure, reward pathways, and adult decision making. Proceedings of the National Academy of Sciences 114, 13549–13554 (2017). DOI: 10.1073/pnas.1708791114
Murray, R. M., Bhavsar, V., Tripoli, G. & Howes, O. 30 Years on: How the Neurodevelopmental Hypothesis of Schizophrenia Morphed Into the Developmental Risk Factor Model of Psychosis. Schizophrenia Bulletin 43, 1190–1196 (2017). DOI: 10.1093/schbul/sbx121
Shaw, P. et al. Neurodevelopmental Trajectories of the Human Cerebral Cortex. The Journal of Neuroscience 28, 3586 (2008). DOI: 10.1523/JNEUROSCI.5309-07.2008
Reynolds, L. M. & Flores, C. Mesocorticolimbic Dopamine Pathways Across Adolescence: Diversity in Development. Frontiers in Neural Circuits 15 (2021). DOI: 10.3389/fncir.2021.735625
Cannon, T. D. et al. Progressive Reduction in Cortical Thickness as Psychosis Develops: A Multisite Longitudinal Neuroimaging Study of Youth at Elevated Clinical Risk. Biol Psychiatry 77, 147–157 (2015). DOI: https://doi.org/10.1016/j.biopsych.2014.05.023
Douaud, G. et al. Schizophrenia delays and alters maturation of the brain in adolescence. Brain 132, 2437–2448 (2009). DOI: 10.1093/brain/awp126
Hensch, T. K. & Fagiolini, M. in Progress in Brain Research Vol. 147 115–124 (Elsevier, 2005).
Zeng, J. et al. Neural substrates of reward anticipation and outcome in schizophrenia: a meta-analysis of fMRI findings in the monetary incentive delay task. Transl Psychiatry 12, 448 (2022). DOI: 10.1038/s41398-022-02201-8
Howes, O. D. & Kapur, S. The Dopamine Hypothesis of Schizophrenia: Version III—The Final Common Pathway. Schizophrenia Bulletin 35, 549–562 (2009). DOI: 10.1093/schbul/sbp006
Howes, O. D. & Murray, R. M. Schizophrenia: an integrated sociodevelopmental-cognitive model. The Lancet 383, 1677–1687 (2014). DOI: https://doi.org/10.1016/S0140-6736(13)62036-X
Hubbard, L. et al. Evidence of Common Genetic Overlap Between Schizophrenia and Cognition. Schizophrenia Bulletin 42, 832–842 (2016). DOI: 10.1093/schbul/sbv168
Bronfenbrenner, U. & Ceci, S. J. Nature-nuture reconceptualized in developmental perspective: A bioecological model. Psychological Review 101, 568–586 (1994). DOI: 10.1037/0033-295X.101.4.568
Gottschling, J. et al. Socioeconomic status amplifies genetic effects in middle childhood in a large German twin sample. Intelligence 72, 20–27 (2019). DOI: https://doi.org/10.1016/j.intell.2018.11.006
Turkheimer, E., Haley, A., Waldron, M., D'Onofrio, B. & Gottesman, I. I. Socioeconomic Status Modifies Heritability of IQ in Young Children. Psychol Sci 14, 623–628 (2003). DOI: 10.1046/j.0956-7976.2003.psci_1475.x
Lewontin, R. Human Diversity. (Scientific American Library, 1995).
Taylor, M. J., Freeman, D., Lundström, S., Larsson, H. & Ronald, A. Heritability of Psychotic Experiences in Adolescents and Interaction With Environmental Risk. JAMA Psychiatry 79, 889 (2022). DOI: 10.1001/jamapsychiatry.2022.1947
Wang, H.-H. et al. Early life stress modulates the genetic influence on brain structure and cognitive function in children. Heliyon 10, e23345 (2024). DOI: https://doi.org/10.1016/j.heliyon.2023.e23345
South, S. C. & Krueger, R. F. Genetic and environmental influences on internalizing psychopathology vary as a function of economic status. Psychological Medicine 41, 107–117 (2011). DOI: 10.1017/s0033291710000279
Bates, T. C., Lewis, G. J. & Weiss, A. Childhood Socioeconomic Status Amplifies Genetic Effects on Adult Intelligence. Psychol Sci 24, 2111–2116 (2013). DOI: 10.1177/0956797613488394
Karcher, N. R. et al. Psychotic-like Experiences and Polygenic Liability in the Adolescent Brain Cognitive Development Study. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging (2021). DOI: https://doi.org/10.1016/j.bpsc.2021.06.012
Sanchez-Roige, S. et al. Genome-wide association study of delay discounting in 23,217 adult research participants of European ancestry. Nature Neuroscience 21, 16–18 (2018). DOI: 10.1038/s41593-017-0032-x
Newbury, J. et al. Cumulative Effects of Neighborhood Social Adversity and Personal Crime Victimization on Adolescent Psychotic Experiences. Schizophrenia Bulletin 44, 348–358 (2018). DOI: 10.1093/schbul/sbx060
Kent, D. M., Steyerberg, E. & van Klaveren, D. Personalized evidence based medicine: predictive approaches to heterogeneous treatment effects. BMJ 363, k4245 (2018). DOI: 10.1136/bmj.k4245
Ruggeri, K. et al. The globalizability of temporal discounting. Nature Human Behaviour (2022). DOI: 10.1038/s41562-022-01392-w
Compton, W. M., Dowling, G. J. & Garavan, H. Ensuring the Best Use of Data: The Adolescent Brain Cognitive Development Study. JAMA Pediatrics 173, 809–810 (2019). DOI: 10.1001/jamapediatrics.2019.2081
Cunha, F. & Heckman, J. J. The Economics and Psychology of Inequality and Human Development. Journal of the European Economic Association 7, 320–364 (2009). DOI: 10.1162/jeea.2009.7.2-3.320
Karcher, N. R. & Barch, D. M. The ABCD study: understanding the development of risk for mental and physical health outcomes. Neuropsychopharmacology 46, 131–142 (2021). DOI: 10.1038/s41386-020-0736-6
Luciana, M. et al. Adolescent neurocognitive development and impacts of substance use: Overview of the adolescent brain cognitive development (ABCD) baseline neurocognition battery. Developmental Cognitive Neuroscience 32, 67–79 (2018). DOI: https://doi.org/10.1016/j.dcn.2018.02.006
Johnson, M. W. & Bickel, W. K. An algorithm for identifying nonsystematic delay-discounting data. Experimental and Clinical Psychopharmacology 16, 264–274 (2008). DOI: 10.1037/1064-1297.16.3.264
Koffarnus, M. N. & Bickel, W. K. A 5-trial adjusting delay discounting task: Accurate discount rates in less than one minute. Experimental and Clinical Psychopharmacology 22, 222–228 (2014). DOI: 10.1037/a0035973
Matusiewicz, A. K., Carter, A. E., Landes, R. D. & Yi, R. Statistical equivalence and test–retest reliability of delay and probability discounting using real and hypothetical rewards. Behavioural Processes 100, 116–122 (2013). DOI: 10.1016/j.beproc.2013.07.019
Owens, M. M. et al. One-year predictions of delayed reward discounting in the adolescent brain cognitive development study. Experimental and Clinical Psychopharmacology (2021). DOI: 10.1037/pha0000532
Burns, P. et al. Examining children's ability to delay reward: Is the delay discounting task a suitable measure? Journal of Behavioral Decision Making 33, 208–219 (2020). DOI: 10.1002/bdm.2154
Myerson, J., Green, L. & Warusawitharana, M. AREA UNDER THE CURVE AS A MEASURE OF DISCOUNTING. Journal of the Experimental Analysis of Behavior 76, 235–243 (2001). DOI: 10.1901/jeab.2001.76-235
Karcher, N. R. et al. Assessment of the Prodromal Questionnaire–Brief Child Version for Measurement of Self-reported Psychotic-like Experiences in Childhood. JAMA Psychiatry 75, 853 (2018). DOI: 10.1001/jamapsychiatry.2018.1334
Karcher, N. R., Perino, M. T. & Barch, D. M. An item response theory analysis of the Prodromal Questionnaire-Brief Child Version: Developing a screening form that informs understanding of self-reported psychotic-like experiences in childhood. Journal of Abnormal Psychology 129, 293–304 (2020). DOI: 10.1037/abn0000502
Karcher, N. R., O’Brien, K. J., Kandala, S. & Barch, D. M. Resting-State Functional Connectivity and Psychotic-like Experiences in Childhood: Results From the Adolescent Brain Cognitive Development Study. Biol Psychiatry 86, 7–15 (2019). DOI: https://doi.org/10.1016/j.biopsych.2019.01.013
Lee, J. J. et al. Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals. Nature Genetics 50, 1112–1121 (2018). DOI: 10.1038/s41588-018-0147-3
Savage, J. E. et al. Genome-wide association meta-analysis in 269,867 individuals identifies new genetic and functional links to intelligence. Nature Genetics 50, 912–919 (2018). DOI: 10.1038/s41588-018-0152-6
Wray, N. R. et al. Genome-wide association analyses identify 44 risk variants and refine the genetic architecture of major depression. Nature Genetics 50, 668–681 (2018). DOI: 10.1038/s41588-018-0090-3
Nievergelt, C. M. et al. International meta-analysis of PTSD genome-wide association studies identifies sex- and ancestry-specific genetic risk loci. Nature Communications 10, 4558 (2019). DOI: 10.1038/s41467-019-12576-w
Demontis, D. et al. Discovery of the first genome-wide significant risk loci for attention deficit/hyperactivity disorder. Nature Genetics 51, 63–75 (2019). DOI: 10.1038/s41588-018-0269-7
Arnold, P. D. et al. Revealing the complex genetic architecture of obsessive–compulsive disorder using meta-analysis. Molecular Psychiatry 23, 1181–1188 (2018). DOI: 10.1038/mp.2017.154
Otowa, T. et al. Meta-analysis of genome-wide association studies of anxiety disorders. Molecular Psychiatry 21, 1391–1399 (2016). DOI: 10.1038/mp.2015.197
Howard, D. M. et al. Genome-wide meta-analysis of depression identifies 102 independent variants and highlights the importance of the prefrontal brain regions. Nature Neuroscience 22, 343–352 (2019). DOI: 10.1038/s41593-018-0326-7
Stahl, E. A. et al. Genome-wide association study identifies 30 loci associated with bipolar disorder. Nature Genetics 51, 793–803 (2019). DOI: 10.1038/s41588-019-0397-8
Grove, J. et al. Identification of common genetic risk variants for autism spectrum disorder. Nature Genetics 51, 431–444 (2019). DOI: 10.1038/s41588-019-0344-8
Ruderfer, D. M. et al. Genomic Dissection of Bipolar Disorder and Schizophrenia, Including 28 Subphenotypes. Cell 173, 1705–1715.e1716 (2018). DOI: https://doi.org/10.1016/j.cell.2018.05.046
Cross-Disorder Group of the Psychiatric Genomics Consortium. Identification of risk loci with shared effects on five major psychiatric disorders: a genome-wide analysis. The Lancet 381, 1371–1379 (2013). DOI: 10.1016/s0140-6736(12)62129-1
Locke, A. E. et al. Genetic studies of body mass index yield new insights for obesity biology. Nature 518, 197–206 (2015). DOI: 10.1038/nature14177
Nagel, M. et al. Meta-analysis of genome-wide association studies for neuroticism in 449,484 individuals identifies novel genetic loci and pathways. Nature Genetics 50, 920–927 (2018). DOI: 10.1038/s41588-018-0151-7
Karlsson Linnér, R. et al. Genome-wide association analyses of risk tolerance and risky behaviors in over 1 million individuals identify hundreds of loci and shared genetic influences. Nature Genetics 51, 245–257 (2019). DOI: 10.1038/s41588-018-0309-3
Watson, H. J. et al. Genome-wide association study identifies eight risk loci and implicates metabo-psychiatric origins for anorexia nervosa. Nature Genetics 51, 1207–1214 (2019). DOI: 10.1038/s41588-019-0439-2
Pasman, J. A. et al. GWAS of lifetime cannabis use reveals new risk loci, genetic overlap with psychiatric traits, and a causal effect of schizophrenia liability. Nature Neuroscience 21, 1161–1170 (2018). DOI: 10.1038/s41593-018-0206-1
Okbay, A. et al. Genetic variants associated with subjective well-being, depressive symptoms, and neuroticism identified through genome-wide analyses. Nature Genetics 48, 624–633 (2016). DOI: 10.1038/ng.3552
Jansen, P. R. et al. Genome-wide analysis of insomnia in 1,331,010 individuals identifies new risk loci and functional pathways. Nature Genetics 51, 394–403 (2019). DOI: 10.1038/s41588-018-0333-3
Walters, R. K. et al. Transancestral GWAS of alcohol dependence reveals common genetic underpinnings with psychiatric disorders. Nature Neuroscience 21, 1656–1669 (2018). DOI: 10.1038/s41593-018-0275-1
Ge, T., Chen, C.-Y., Ni, Y., Feng, Y.-C. A. & Smoller, J. W. Polygenic prediction via Bayesian regression and continuous shrinkage priors. Nature Communications 10 (2019). DOI: 10.1038/s41467-019-09718-5
Casey, B. J. et al. The Adolescent Brain Cognitive Development (ABCD) study: Imaging acquisition across 21 sites. Developmental Cognitive Neuroscience 32, 43–54 (2018). DOI: https://doi.org/10.1016/j.dcn.2018.03.001
Garavan, H. et al. Recruiting the ABCD sample: Design considerations and procedures. Developmental Cognitive Neuroscience 32, 16–22 (2018). DOI: https://doi.org/10.1016/j.dcn.2018.04.004
Jovicich, J. et al. Reliability in multi-site structural MRI studies: Effects of gradient non-linearity correction on phantom and human data. NeuroImage 30, 436–443 (2006). DOI: https://doi.org/10.1016/j.neuroimage.2005.09.046
Ségonne, F. et al. A hybrid approach to the skull stripping problem in MRI. NeuroImage 22, 1060–1075 (2004). DOI: https://doi.org/10.1016/j.neuroimage.2004.03.032
Dale, A. M., Fischl, B. & Sereno, M. I. Cortical Surface-Based Analysis: I. Segmentation and Surface Reconstruction. NeuroImage 9, 179–194 (1999). DOI: https://doi.org/10.1006/nimg.1998.0395
Fischl, B., Liu, A. & Dale, A. M. Automated manifold surgery: constructing geometrically accurate and topologically correct models of the human cerebral cortex. IEEE Transactions on Medical Imaging 20, 70–80 (2001). DOI: 10.1109/42.906426
Fischl, B., Sereno, M. I., Tootell, R. B. H. & Dale, A. M. High-resolution intersubject averaging and a coordinate system for the cortical surface. Human Brain Mapping 8, 272–284 (1999). DOI: https://doi.org/10.1002/(SICI)1097-0193(1999)8:4<272::AID-HBM10>3.0.CO;2-4
Ségonne, F., Pacheco, J. & Fischl, B. Geometrically Accurate Topology-Correction of Cortical Surfaces Using Nonseparating Loops. IEEE Transactions on Medical Imaging 26, 518–529 (2007). DOI: 10.1109/TMI.2006.887364
Desikan, R. S. et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage 31, 968–980 (2006). DOI: https://doi.org/10.1016/j.neuroimage.2006.01.021
van Dijk, M. T., Murphy, E., Posner, J. E., Talati, A. & Weissman, M. M. Association of Multigenerational Family History of Depression With Lifetime Depressive and Other Psychiatric Disorders in Children. JAMA Psychiatry (2021). DOI: 10.1001/jamapsychiatry.2021.0350
Anokhin, A. P., Grant, J. D., Mulligan, R. C. & Heath, A. C. The Genetics of Impulsivity: Evidence for the Heritability of Delay Discounting. Biol Psychiatry 77, 887–894 (2015). DOI: 10.1016/j.biopsych.2014.10.022
Shamosh, N. A. & Gray, J. R. Delay discounting and intelligence: A meta-analysis. Intelligence 36, 289–305 (2008). DOI: https://doi.org/10.1016/j.intell.2007.09.004
Keidel, K., Rramani, Q., Weber, B., Murawski, C. & Ettinger, U. Individual Differences in Intertemporal Choice. Frontiers in Psychology 12 (2021). DOI: 10.3389/fpsyg.2021.643670
Hausman, J. A. Specification Tests in Econometrics. Econometrica 46, 1251 (1978). DOI: 10.2307/1913827
Cunningham, M. et al. A Pilot Study of Landlord Acceptance of Housing Choice Vouchers. (US Department of Housing and Urban Development, Office of Policy Development and Research, Washington, DC, 2018).
Freeman, L. & Li, Y. Do Source of Income Anti-discrimination Laws Facilitate Access to Less Disadvantaged Neighborhoods? Housing Studies 29, 88–107 (2014). DOI: 10.1080/02673037.2013.824559
Pollack, C. E. et al. Association of Receipt of a Housing Voucher With Subsequent Hospital Utilization and Spending. JAMA 322, 2115–2124 (2019). DOI: 10.1001/jama.2019.17432
Kling, J. R., Liebman, J. B. & Katz, L. F. Experimental Analysis of Neighborhood Effects. Econometrica 75, 83–119 (2007). DOI: 10.1111/j.1468-0262.2007.00733.x
Staiger, D. & Stock, J. H. Instrumental Variables Regression with Weak Instruments. Econometrica 65, 557 (1997). DOI: 10.2307/2171753
Fox, J., Kleiber, C. & Zeileis, A. ivreg: Instrumental-Variables Regression by '2SLS', '2SM', or '2SMM', with Diagnostics v. R package version 0.6-3 (2023). https://zeileis.github.io/ivreg
Breiman, L. Random Forests. Machine Learning 45, 5–32 (2001). DOI: 10.1023/a:1010933404324
Berman, M. G., Stier, A. J. & Akcelik, G. N. Environmental neuroscience. Am Psychol 74, 1039–1052 (2019). DOI: 10.1037/amp0000583
Tooley, U. A., Bassett, D. S. & Mackey, A. P. Environmental influences on the pace of brain development. Nature Reviews Neuroscience 22, 372–384 (2021). DOI: 10.1038/s41583-021-00457-5
Auchter, A. M. et al. A description of the ABCD organizational structure and communication framework. Developmental Cognitive Neuroscience 32, 8–15 (2018). DOI: https://doi.org/10.1016/j.dcn.2018.04.003

There is NO Competing Interest.

SupplementaryInformation.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Individual Differences in the Effects of Neighborhood Socioeconomic Deprivation on Intertemporal Decision-Making and Psychotic-Like Experiences in Children

Status:

Version 1

Abstract

Figures

Introduction

Results

Average Treatment Effects of Neighborhood Socioeconomic Adversity on Delay Discounting and PLEs

Discussion

Methods

Study Participants

Data

Neighborhood Disadvantage

Delay Discounting

Psychotic-Like Experiences

Genome-wide Polygenic Scores

Anatomical Brain Imaging: T1/T2, Freesurfer 6

Functional MRI (fMRI): Monetary Incentive Delay (MID) task

Covariates

Statistical Analyses

Instrumental Variable Regression

Causal Machine Learning for Treatment Effects

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1