The longitudinal effects of neglect on aggression and delinquency: Testing competing hypotheses

Abstract

Childhood neglect is the most common form of maltreatment, yet it remains understudied, with questions remaining about how it imparts harm across development. This study reports on a theory-driven test of whether duration of exposure to neglect across childhood (cumulative hypothesis) or timing of exposure during specific developmental periods (critical period hypothesis) better explains the likelihood of adolescent aggression and delinquency. We examined the effects of longitudinal patterns of exposure to two types of neglect (failure to provide and lack of supervision) on aggression and delinquency at ages 12 and 16. Using LONGSCAN data (N = 1,354), we created dummy variables for failure to provide and lack of supervision across four-year developmental periods (0–4, 5–8, 9–12, and 13–16). Using a model-testing approach estimating a series of models that each represent a specific hypothesis, we identified the best fitting model using general linear tests and fit statistics. Our results indicate that there is an early adolescence critical period for lack of supervision for age 16. These findings add to a growing body of evidence demonstrating that neglect is harmful and requires attention from child welfare throughout childhood and into adolescence.

1. Introduction

Neglect is the most common form of maltreatment, comprising 76% of maltreatment cases (U.S. Department of Health & Human Services, 2023), yet is relatively understudied. Although neglect has similar effects to other forms of maltreatment, there are also unique effects of neglect (e.g., Hildyard & Wolfe, 2002), suggesting the importance of research that examines neglect distinct from abuse. It is clear there is an association between neglect and delinquency and aggression (Graham et al., 2010; Lee & Villagrana, 2015; Logan-Greene & Semanchin Jones, 2015; Watts, 2017), but most longitudinal studies examining exposure to child maltreatment as predictors of delinquency and aggression have not separated neglect from other forms of maltreatment (Stewart et al., 2008; Thornberry et al., 2010). Thus, we can further our understanding of this association by exploring the longitudinal effect of neglect, specifically, on delinquency and aggression. Further, although scholars have identified subtypes of neglect (Coohey, 2003; Jonson-Reid et al., 2013), and while studies of maltreatment often differentiate between types of abuse, neglect is typically treated as a singular phenomenon. Consequently, knowledge about the longitudinal associations between subtypes of neglect and delinquency and aggression can provide insight into how to intervene to counter the negative effects of neglect.

2. Background and literature review

2.1. Types of neglect

Although there are multiple conceptualizations of neglect in the literature, this study focuses on two main subtypes of neglect that are frequent subtypes in child welfare jurisdiction coding: physical neglect and supervisory neglect. These two types of neglect are critical to the observable development of the child. Physical neglect refers to the caregiver’s failure to provide for the child’s physical needs, including food and water, safe and sanitary living conditions, medical care, and education (Coohey, 2003; Jonson-Reid et al., 2013). Failure to provide neglect is about providing the material resources and supports necessary for the child to be healthy, engaged, and capable of growing and achieving developmental tasks. On the other hand, lack of supervision refers to the caretaker’s failure to protect the child from their environments (Coohey, 2003). This type of neglect can be thought of as boundary setting for the child: a caretaker establishing limits to the context within which the child can explore and experiment safely, as well as protecting the child from unsafe people and contexts. Thus, as the child ages, appropriate boundaries and supervision will change for each developmental stage (Ryan et al., 2013). For example, an infant must be watched constantly lest they roll off a bed or encounter a choking hazard. For a toddler, they must be protected from sharp objects, electrical outlets, or from jumping off chairs and sofas. On the other hand, an adolescent requires more freedom and autonomy to exercise their independence and explore their identities while ensuring that they are complying with reasonable rules and expectations for their age. Lack of supervision during adolescence is often conceptualized as inadequate parental monitoring (e.g., knowing who they are with during their free time) and setting curfews (Ryan et al., 2013). Thus, failure to provide and lack of supervision can be considered complimentary forms of neglect of the child’s development.

2.2. Longitudinal perspectives on neglect

Neglect is more likely to be chronic than other types of maltreatment (Semanchin Jones & Logan-Greene, 2016), highlighting the importance of understanding the longitudinal effects of neglect. Longitudinal measures of neglect often take into account two dimensions: duration and timing. Duration refers to the total amount of time the child is exposed to neglect while timing refers to when neglect is experienced by the child. It is possible that the effect of neglect depends on either duration, timing, or both. For example, if the effect of neglect depends on duration, then negative effects continue to accrue throughout the duration of exposure. This cumulative hypothesis is depicted in the top panel of Fig. 1 and is one approach to measuring “chronicity” in that more chronic neglect will result in longer exposure duration. This cumulative hypothesis suggests that neglect has a direct effect on the child’s behavior, wherein longer exposure will have stronger and more lasting impacts on the child’s behavior. Studies of chronicity have found that longer exposure or ongoing exposure to neglect has been found to be related to delinquency and aggression (Graham et al., 2010; Logan-Greene & Semanchin Jones, 2015; Ryan et al., 2013).

Fig. 1.

Longitudinal hypotheses.

Alternatively, the effect of neglect may depend on the timing of exposure. It is possible that there is a critical period in development during which experiences of neglect may be especially disruptive to development and most likely to contribute to delinquency and aggression outcomes in adolescence. From this perspective, the timing of neglect may be the key factor that determines the negative effects. In support of this hypothesis, Graham et al. (2010) showed that a longitudinal measure of neglect that takes into account developmental periods, rather than a measure of total duration of exposure, had a stronger association with peer aggression. The bottom panel of Fig. 1 shows two possible versions of this critical period hypothesis: middle childhood as the critical period and early adolescence as the critical period.

Studies have shown that there is an association between exposure in specific developmental periods and adolescent outcomes, although a clear critical period has not emerged. There is evidence that early or middle childhood experiences of neglect (Fig. 1, middle panel) are associated with increased adolescent impacts (Chapple et al., 2005; Evans & Burton, 2013; Watts, 2017).

On the other hand, other studies have found evidence to suggest that adolescence may be a critical period for adolescent aggression and delinquency (Dubowitz et al., 2019, 2022; Thornberry et al., 2010). Thornberry et al. (2010) tested whether there is a critical period in childhood or adolescence by creating “developmental classifications.” They defined two types of exposure to maltreatment: only in childhood or any exposure in adolescence. Although Thornberry et al. found that any adolescent maltreatment was associated with higher levels of general offending, involvement in violent crimes, and official arrests/incarceration, there are limitations to the study to warrant further examination. Thornberry et al.’s developmental classification did not distinguish between duration and timing, since 39% of the any adolescent group also experienced maltreatment in childhood. Thus, many of the children in the any adolescent group also experienced a longer duration of maltreatment. Furthermore, this study tested the effect of maltreatment, which includes both abuse and neglect, rather than identifying the specific effect of neglect. More recently, Dubowitz et al. (2019,2022) conducted a latent class analysis and identified three longitudinal patterns of neglect, which was primarily differentiated by self-reported neglect: limited neglect, chronic neglect, and late neglect (starting in adolescence). They found that those in the late neglect group were at higher risk for substance use and delinquency compared to those in the chronic neglect group. However, their study did not differentiate between types of neglect and did not provide a direct test of the competing hypotheses.

2.3. Research linking neglect with adolescent aggression and delinquency

Studies have established that there is an association between neglect (general) and both delinquency and aggression (Graham et al., 2010; Lee & Villagrana, 2015; Logan-Greene & Semanchin Jones, 2015; Watts, 2017). In terms of specific types of neglect, researchers have found an association between failure to provide and delinquency and aggression (Chapple et al., 2005; Evans & Burton, 2013; Logan-Greene & Semanchin Jones, 2015). However, results are mixed for associations between lack of supervision and delinquency and aggression (Kobulsky et al., 2022; Logan-Greene & Semanchin Jones, 2015; Maughan & Moore, 2010). More research is needed to understand the effects of specific types of neglect and how their duration and timing affects these outcomes.

2.4. The current study

This current study advances knowledge of the longitudinal effects of neglect by parsing out the effects of duration versus timing. Prior studies have operationalized duration and timing of neglect within a single model, which can yield measures that conflate the two dimensions of longitudinal neglect. We use an innovative model-building approach to test two competing hypotheses about the longitudinal association between neglect and delinquency and aggression in adolescence: the cumulative hypothesis, which asserts the importance of the duration of exposure to neglect, versus the critical period hypothesis, which asserts that the timing of exposure of neglect has the more negative effect. These two competing hypotheses are compared using an approach developed specifically to test competing hypotheses regarding exposures across the life course (Mishra et al., 2009). In contrast to standard approaches where results focus on the significance of coefficients within each model, in this novel approach, models which represent the two hypotheses are estimated and compared, and thus the results focus on overall model fit. This approach has been used to test exposure to a variety of phenomena, including poverty (Mazza et al., 2017), family structure (Pearce et al., 2014), and adversity (Dunn et al., 2019) on outcomes, including behavioral problems (Mazza et al., 2017), socio-emotional well-being (Pearce et al., 2014), and alcohol use (Britton et al., 2015). More recently, this approach has been used to examine the longitudinal effect of neglect on internalizing mental health symptoms (Logan-Greene et al., 2025).

Although there is evidence to support both hypotheses, prior studies do not conduct a direct test between the two hypotheses. The evidence for the negative effect of childhood experiences of neglect, and for failure to provide specifically, is stronger. Therefore, the first hypothesis this study will test is: There is an early childhood critical period for failure to provide, since disrupting key developmental tasks early on is more likely to have cascading effects throughout the child’s life. On the other hand, the evidence for lack of supervision is less clear, but in Thornberry’s study, the “any adolescent” group was likely to also have had longer exposure. Thus, the second hypothesis this study will test is: There is a cumulative effect for lack of supervision.

3. Method

Data from the Longitudinal Studies on Child Abuse and Neglect (LONGSCAN) were used to test these competing hypotheses (Runyan et al., 1998). LONGSCAN includes five pooled cohort samples from several regions across the U.S.: East (Baltimore), Midwest (Chicago), Northwest (Seattle), Southwest (San Diego), and the South (North Carolina). The sampling method varied by site, such as from pediatric clinics serving low-income, inner-city children; high risk at birth; moderate risk for Child Protective Services (CPS); and matched sample of families reported to CPS and neighborhood controls. Such an approach yields a sample at high-risk for maltreatment but with variability of experiences. Each of the sites used nearly identical methods and measurement tools, and children and parents were interviewed repeatedly from 0 to 18 (Runyan et al., 1998). Thus, this dataset is unique in the ability to test the longitudinal effects of neglect. Although the data collection for infants/baseline began in 1991, this dataset remains, to our knowledge, the only available longitudinal study of maltreatment that could be used to test these competing hypotheses. See Runyan et al. (1998) for more details about the LONGSCAN study.

Since the interviews occurred at uneven intervals (at ages 4, 6, 8, 12, 14, 16, and 18), some of the shorter time periods were collapsed to create four even time periods which also approximately mapped on to four different developmental periods: ages 0–4 (infancy and early childhood); ages 5–8 (middle childhood); ages 9–12 (late childhood); and ages 13–16 (early adolescence). Thus, measures during infancy and early childhood were based on data collected at age 4; middle childhood based on data collected at ages 6 and 8; late childhood based on data collected at age 12; and early adolescence based on data collected at ages 14 and 16.

Sample statistics are presented in Table 1. The initial sample consisted of 1,354 children. The sample was diverse with 53.2% Black, 26.1% Caucasian, 11.9% mixed race, 7.2% Hispanic/Latino, and 1.6% another race. The sample was approximately balanced by child’s sex, with 51.5% female.

Table 1.

Sample statistics.

	Mean	Standard Deviation
Delinquency
Age 12	1.29	0.95
Age 16	1.37	1.10
Aggression
Age 12	2.78	1.31
Age 16	2.44	1.35
Neglect	Percent	Frequency
Failure to Provide
Ages 0–4	39.40%	342
Ages 4–8	14.75%	128
Ages 8–12	11.64%	101
Ages 12–16	5.41%	47
Lack of Supervision
Ages 0–4	26.50%	230
Ages 4–8	17.51%	152
Ages 8–12	11.52%	100
Ages 12–16	7.37%	64
Race/Ethnicity
Caucasian	25.91%	200
African American	55.70%	430
Hispanic	7.12%	55
Mixed Race	10.49%	81
Other Race	0.26%	2
Gender	48.39%	420
	51.61%	448
Poverty	68.09%	510
Foster Care	41.75%	263
Sexual Abuse	18.32%	159
Physical Abuse	36.29%	315
Study Site:
East	18.43%	160
Midwest	21.43%	186
Northwest	21.31%	185
South	15.21%	132
Southwest	23.62%	205

3.1. Measures

3.1.1. Neglect

Neglect is measured as a dichotomous variable at all four time periods using screened-in allegations. These are official reports made to child protective services (CPS) during the study period. All screened-in allegations of neglect were included in this analysis due to evidence that substantiation is not a reliable marker of the existence or seriousness of maltreatment (Hussey et al., 2005; Kohl et al., 2009). LONGSCAN researchers utilized a maltreatment classification system (Barnett et al., 1993) to apply consistent maltreatment codes to differing local definitions. Two categories of neglect were established: failure to provide (FTP) and lack of supervision (LOS; Dubowitz et al., 2005). Thus, neglect is operationalized as any screened-in allegations of neglect (=1) in each of the four periods (infancy and early childhood, middle childhood, late childhood, and early adolescence) for both types of neglect: FTP and LOS. Across the four time periods, this results in 16 possible combinations, which are enumerated in the first column of Table 2 where the four digits represent the four study periods. For example, the first profile (0 0 0 0) reflects no allegations of neglect (i.e., no screened in allegations) in any of the four time periods, while the last profile (1 1 1 1) indicates allegations of neglect in all four time periods. The profile of (0 1 0 0) would indicate allegations of neglect between ages 5–8. The frequency of each neglect profile for failure to provide and lack of supervision are reported in the second and third columns of Table 2.

Table 2.

Frequency of each neglect profile and associated hypotheses.

Neglect profile	Failure to provide	Lack of supervision	Critical period hypothesis				Cumulative hypothesis
	Frequency	Frequency	Critical 1	Critical 2	Critical 3	Critical 4	Cumulative 2	Cumulative 3	Cumulative 4
0000	430	498	0	0	0	0	0	0	0
0001	12	16	0	0	0	1	0	0	1
0010	27	24	0	0	1	0	0	1	1
0011	7	8	0	0	1	1	0	1	2
0100	37	62	0	1	0	0	1	1	1
0101	3	7	0	1	0	1	1	1	2
0110	8	14	0	1	1	0	1	2	2
0111	2	9	0	1	1	1	1	2	3
1000	224	141	1	0	0	0	1	1	1
1001	11	5	1	0	0	1	1	1	2
1010	24	17	1	0	1	0	1	2	2
1011	5	7	1	0	1	1	1	2	3
1100	47	33	1	1	0	0	2	2	2
1101	3	6	1	1	0	1	2	2	3
1110	24	15	1	1	1	0	2	3	3
1111	4	6	1	1	1	1	2	3	4

Note: Each of our competing models has a single predictor which is binary for the critical value models, and numeric for the cumulative models. These columns show how the unique neglect profiles (i.e., rows) are collapsed to create the value of that predictor specific to each hypothesis indicated in the header of the column.

These 16 neglect profiles enumerated in Table 2 are combined into simplified models to represent the critical period and cumulative hypotheses. Several variations of the two main hypotheses were tested, as outlined in Table 3, yielding a total of 7 competing models. The two panels on the right-hand side of Table 2 show how the 16 profiles are combined into a model to represent each of the hypotheses listed. Table 3 explains how each hypothesis is translated into a model based on the constraints listed in the last column. For example, if there is a critical period in infancy and early childhood (Critical 1, ages 0–4), it is conjectured that only the model coefficient corresponding to the main effect of the first period is non-zero. Thus, only neglect profiles with allegations in the first period are identified with a 1 in the Critical 1 column in table 2, with all other profiles identified with a 0.

Table 3.

Critical Period and Cumulative Hypotheses.

Hypothesis	Hypothesis Description	Version Tested	Ages	Specific Hypothesis	Constraints
Critical Period	Only the presence or absence of a neglect allegation in identified developmental period is associated with negative effects, with neglect allegations at other developmental periods not being associated.	Critical 1	0–4	Exposure in infancy & early childhood (period 1)	Coefficient for all other periods = 0
		Critical 2	5–8	Exposure in middle childhood (period 2)	Coefficient for all other periods = 0
		Critical 3	9–12	Exposure in late childhood (period 3)	Coefficient for all other periods = 0
		Critical 4	13–16	Exposure in early adolescence (period 4)	Coefficient for all other periods = 0
Cumulative	The negative effects of neglect are proportional to the amount of exposure to neglect through an identified developmental period.	Cumulative 2	0–4; 5–8	Linear effect of neglect exposure through middle childhood (in periods 1 and 2)	Coefficient for periods 3 and 4 = 0
		Cumulative 3	0–4; 5–8; 9–12	Linear effect of neglect exposure through late childhood (in periods 1, 2, 3)	Coefficient for period 4 = 0
		Cumulative 4	0–4; 5–8; 9–12; 13–16	Linear effect of exposure through early adolescence (in periods 1, 2, 3, 4)

3.1.2. Delinquency and aggression

The two outcome variables were measured at ages 12 (at the end of late childhood) and 16 (at the end of early adolescence) using subscales from the parent reported Child Behavior Checklist (CBCL; Achenbach, 1991). The delinquency subscale includes items pertaining to behaviors such as a child that runs away, sets fires, and steals. The parent reported aggression subscale includes items pertaining to behaviors such as whether the child argues, brags, is mean, and has a temper. Raw scores were used since they have maximum variation and are more interpretable. Internal consistency for the delinquency subscale was good (α = 0.76) and for the aggression subscale was excellent (α = 0.91) (Knight et al., 2010).

3.1.3. Control variables

To test the robustness of the results, the models were also estimated with a set of related control variables. Race/ethnicity was coded into a set of dichotomous variables: White, Black, Hispanic, mixed race, and other. Gender was coded as male (=1) or female (=0). Family poverty status was coded to identify whether the family income was below the federal poverty line (=1) between ages 0–4. This variable was constructed from family income and household size compared to the federal poverty guideline for the household size in the year the data were collected. Foster care status (=1) indicated whether the child was ever placed in foster care. There were also two indicator variables of abuse: whether there was ever an allegation of physical abuse (=1) or an allegation of sexual abuse (=1) during the study period. Finally, a set of dichotomous variables were included which identified each of the study sites.

3.2. Analytic approach

First, bivariate analyses were conducted to explore the associations between neglect and both aggression and delinquency. T-tests were used to identify associations between delinquency and aggression at ages 12 and 16 with exposure to FTP and LOS at each of the four time periods of the study.

The linear model-based approach for comparing competing models as outlined in Mishra et al. (2009) was used. The saturated model, which was fit using SAS PROC REG, is defined based on the 16 profiles enumerated in Table 2 and consisting of all main effects for the four periods and all possible interaction terms between the four periods. Then, reduced models were calculated for each of the 7 competing models subject to constraints that are determined by the competing hypotheses as outlined in Table 3. This analysis approach was repeated separately for the two types of neglect (FTP and LOS) and for the two outcomes (aggression and delinquency) at ages 12 and 16.

To compare the competing models, the F-statistic for the general linear test was used to assess each candidate model against the saturated model, with a significant p-value indicating that the saturated model provides significantly better model fit. Thus, non-significant p-values indicated plausible candidate models. The subset of plausible models were subsequently compared to each other using the Akaike Information Criterion (AIC), where the lowest value indicates the best model fit. To assess the robustness of findings, the identical process was repeated with the control variables added to the model. Given the complicated nature of the models and lack of cell coverage for the calculations, each control variable was added one at a time; adding all control variables simultaneously is not possible for this statistical technique. For a more detailed explanation of these analyses, please see (Logan-Greene et al., 2025).

4. Results

Table 4 shows the bivariate associations between neglect in each time period with the two outcomes at both 12 and 16 years of age. The bivariate association between FTP is statistically significant for delinquency and aggression at age 12 and aggression at age 16. However, at age 16, only FTP during ages 0–4 and 13–16 are statistically significant with delinquency. For LOS, there is a similar but slightly different pattern of associations: LOS at all three periods is statistically significant for delinquency and aggression at age 12, but at age 16, LOS is only statistically significant with both delinquency and aggression in early adolescence. These descriptive results may indicate that there is a cumulative effect for age 12 outcomes and a critical effect of early adolescence for age 16 outcomes.

Table 4.

Bivariate associations between neglect and delinquency and aggression, by child age.

	Age 12 Outcomes N = 951			Age 16 Outcomes N = 867
Failure to Provide	No	Yes	p-value	No	Yes	p-value
	Delinquency			Delinquency
Ages 0–4	2.35(2.72)	2.90(3.03)	0.004	2.80(3.64)	3.57(4.29)	0.007
Ages 5–8	2.43(2.71)	3.36(3.53)	0.004	3.05(3.97)	3.41(3.67)	0.35
Ages 9–12	2.46(2.81)	3.34(3.07)	0.002	3.04(3.88)	3.62(4.21)	0.16
Ages 13–16				3.01(3.88)	4.70(4.44)	<0.01
	Aggression			Aggression
Ages 0–4	8.84(6.93)	10.40(7.86)	0.002	7.23(6.86)	8.61(7.56)	0.006
Ages 5–8	9.04(7.00)	11.97(8.73)	<0.001	7.55(6.99)	9.04(8.06)	0.05
Ages 9–12	9.17(7.22)	11.67(7.92)	<0.001	7.57(6.95)	9.36(8.53)	0.045
Ages 13–16				7.59(7.07)	10.96(8.16)	<0.01
Lack of Supervision	No	Yes	p-value	No	Yes	p-value
	Delinquency			Delinquency
Ages 0–4	2.42(2.82)	2.95(2.92)	0.01	2.95(3.71)	3.52(2.94)	0.08
Ages 5–8	2.48(2.88)	2.98(2.70)	0.04	3.00(3.89)	3.60(2.95)	0.09
Ages 9–12	2.47(2.80)	3.35(3.25)	0.01	3.05(3.91)	3.56(4.02)	0.22
Ages 13–16				2.91(3.67)	5.52(5.84)	<0.01
	Aggression			Aggression
Ages 0–4	9.03(7.21)	10.58(7.56)	0.03	7.64(7.12)	8.16(7.32)	0.35
Ages 5–8	9.08(7.23)	11.25(7.65)	<0.01	7.57(7.05)	8.74(7.65)	0.07
Ages 9–12	9.20(7.09)	11.71(9.00)	<0.01	7.63(7.02)	8.90(8.18)	0.14
Ages 13–16				7.38(6.82)	12.69(9.41)	<0.01

Table 5 presents the model fit statistics for the 7 models for FTP and the two outcomes and two outcome periods. For the effect of FTP on delinquency, at age 12 two models do not show significantly worse fit than the saturated model (cumulative 2 and cumulative 3), but the model with the lowest AIC (and thus the best fitting model) is the cumulative 3 model. At age 16, only the cumulative 4 model does not show significantly worse fit than the saturated model, which also has the lowest AIC. For the effect of FTP on aggression, there is a similar pattern. At age 12, the best fitting model based on both the p-value and AIC is the cumulative 3 model. At age 16, both model fit indices indicate that the cumulative 4 model is the best fitting model.

Table 5.

Model Fit Statistics for Failure to Provide.

	Delinquency at Age 12 N = 951				Delinquency at Age 16 N = 867
	df	F-statistic	P-value	AIC	df	F-statistic	P-value	AIC
Saturated Model				2592.04				2626.75
Critical 1	6, 943	3.08	<0.01	2598.50	14, 851	1.92	0.02	2625.67
Critical 2	6, 943	2.65	0.01	2595.94	14, 851	2.62	<0.01	2635.30
Critical 3	6, 943	2.99	<0.01	2597.99	14, 851	2.52	<0.01	2633.99
Critical 4					14, 851	1.89	0.02	2625.26
Cumulative 2	6, 943	1.73	0.11	2590.47	14, 851	1.95	0.02	2626.08
Cumulative 3	6, 943	0.78	0.59	2584.75	14, 851	1.85	0.03	2624.78
Cumulative 4					14, 851	1.47	0.12	2619.49
	Aggression at Age 12 N = 951				Aggression at Age 16 N = 867
	df	F-statistic	P-value	AIC	df	F-statistic	P-value	AIC
Saturated Model				3201.21				2673.27
Critical 1	6, 943	4.19	<0.01	3214.24	14, 851	1.97	0.02	2972.87
Critical 2	6, 943	3.02	<0.01	3207.33	14, 851	2.18	<0.01	2975.85
Critical 3	6, 943	3.89	<0.01	3212.44	14, 851	2.30	<0.01	2977.48
Critical 4					14, 851	1.84	0.03	2971.16
Cumulative 2	6, 943	2.46	0.023	3203.98	14, 851	1.74	0.04	2969.70
Cumulative 3	6, 943	1.45	0.19	3197.92	14, 851	1.64	0.06	2968.40
Cumulative 4					14, 851	1.32	0.19	2963.95

Table 6 present the model fit statistics for the 7 models for LOS and the two outcomes and two outcome periods. For the effect of LOS on delinquency, similar to FTP, at age 12 the p-value indicates both the cumulative 2 and cumulative 3 models while the AIC indicates cumulative 3 is the best fitting model. At age 16, both model fit indices indicate that the critical 4 model is the best fitting model. For aggression, the cumulative 3 model is the best fitting model at age 12, whereas the critical 4 model is the best fitting model at age 16. The parameter estimates for the best fitting models are reported in Table 7.

Table 6.

Model fit statistics for lack of supervision.

	Delinquency at Age 12 N = 951				Delinquency at Age 16 N = 867
	df	F-statistic	P-value	AIC	df	F-statistic	P-value	AIC
Saturated Model				2597.33				2624.97
Critical 1	6, 943	2.15	0.045	2598.28	14, 851	2.67	<0.01	2634.31
Critical 2	6, 943	2.55	<0.01	2600.66	14, 851	2.61	<0.01	2633.40
Critical 3	6, 943	2.49	0.02	2600.29	14, 851	2.63	<0.01	2633.75
Critical 4					14, 851	1.11	0.34	2612.67
Cumulative2	6, 943	1.27	0.27	2592.96	14, 851	2.44	<0.01	2631.15
Cumulative 3	6, 943	0.60	0.73	2588.96	14, 851	2.32	<0.01	2629.41
Cumulative 4					14, 851	1.75	0.04	2621.58
	Aggression at Age 12 N = 951				Aggression at Age 16 N = 867
	df	F-statistic	P-value	AIC	df	F-statistic	P-value	AIC
Saturated Model				3203.60				2664.89
Critical 1	6, 943	3.62	<0.01	3213.23	14, 851	3.05	<0.01	2979.37
Critical 2	6, 943	3.02	<0.01	3209.57	14, 851	2.86	<0.01	2976.70
Critical 3	6, 943	4.19	<0.01	3116.64	14, 851	3.05	<0.01	2979.33
Critical 4					14, 851	1.03	0.42	2951.49
Cumulative 2	6, 943	1.90	0.08	3203.03	14, 851	2.82	<0.01	2976.26
Cumulative 3	6, 943	1.46	0.19	3200.37	14, 851	2.79	<0.01	2975.74
Cumulative 4					14, 851	2.21	<0.01	2967.80

Table 7.

Coefficients for best fitting models.

Failure to provide	Model	Delinquency		Aggression
		Age 12	Age 16	Age 12	Age 16
N		951	867	951	867
Constant		1.165	1.241	2.603	2.285
Coefficient * (FP1 + FP2 + FP3)	Cumulative 3	0.190**		0.274***
Coefficient * (FP1 + FP2 + FP3 + FP4)	Cumulative 4		0.186***		0.22***
Model AIC		2584.75	2619.49	3197.92	2963.95
Saturated AIC		2592.04	2626.75	3201.21	2673.27
Lack of Supervision
N		951	867	951	867
Constant		1.191	1.321	2.632	2.373
Coefficient * (LS1 + LS2 + LS3)	Cumulative 3	0.179***		0.269***
Coefficient * LS4	Critical 4		0.710***		0.941***
Model AIC		2588.96	2612.67	3200.37	2951.49
Saturated AIC		2597.33	2624.97	3203.60	2664.89

Note: FP = Failure to Provide. LS = Lack of Supervision.

^***= p < 0.0001.

4.1. Model robustness – adding control variables

Table 8 reports on the best fitting models with the addition of control variables. Overall, the LOS results for delinquency and aggression at age 16 were robust to the addition of all control variables, with the best fitting model consistently the critical 4 (early adolescence) model. The FTP results for delinquency and aggression at age 16 were less robust to the addition of study site, physical abuse, and sexual abuse, although there was consistency in that results changed from cumulative 4 to critical 4 (early adolescence). For delinquency at age 16, the addition of foster care as a control variable also changed the results from cumulative 4 to critical 4 (early adolescence).

Table 8.

Best fitting model with the addition of control variables.

	Lack of Supervision				Failure to Provide
	Delinquency		Aggression		Delinquency		Aggression
	Age 12	Age 16	Age 12	Age 16	Age 12	Age 16	Age 12	Age 16
No Controls	Cumulative 3	Critical 4	Cumulative 3	Critical 4	Cumulative 3	Cumulative 4	Cumulative 3	Cumulative 4
Race/Ethnicity	Cumulative 3	Critical 4	Cumulative 3	Critical 4	Cumulative 3	Cumulative 4	Cumulative 3	Cumulative 4
Gender	Cumulative 3	Critical 4	Cumulative 3	Critical 4	Cumulative 3	Cumulative 4	Cumulative 3	Cumulative 4
Poverty	Cumulative 3	Critical 4	Cumulative 3	Critical 4	Cumulative 3		Cumulative 3	Cumulative 4
Foster care	Cumulative 3	Critical 4	Cumulative 3	Critical 4	Cumulative 3	Critical 4	Cumulative 3	Cumulative 4
Study Site	Cumulative 3	Critical 4	Critical 2	Critical 4	Critical 2	Critical 4	Critical 2	Critical 4
Physical Abuse	Cumulative 3	Critical 4	Cumulative 3	Critical 4	Cumulative 3	Critical 4	Cumulative 3	Critical 4
Sexual Abuse	Critical 1	Critical 4	Cumulative 3	Critical 4		Critical 4	Cumulative 3	Critical 4

Note: Bold text indicates that the addition of the control variable changed the best-fitting model. Shaded cells indicate that the model did not converge.

Results for LOS and FTP for delinquency and aggression at age 12 were also less robust overall, although they were robust to the addition of demographic control variables (gender, race/ethnicity, and poverty) as well as foster care status and physical abuse. With the addition of sexual abuse as a control variable, the results for LOS for delinquency at age 12 changed from cumulative 3 to critical 1 (infancy and early childhood). The addition of study site as a control changed the results for LOS for aggression at age 12 and FTP for delinquency and aggression at age 12 from cumulative 3 to critical 2 (middle childhood). Note that two models failed to converge: the addition of sexual abuse in the FTP for delinquency at age 12, and the addition of poverty to the FTP for delinquency at age 16.

5. Discussion

This study extends existing research that has shown that there is an association between neglect and delinquency and aggression by directly testing competing hypotheses about the longitudinal impact of neglect on these negative outcomes. Contrary to the first hypotheses, these analyses did not yield an early childhood critical period effect of failure to provide on aggression and delinquency in adolescence. Rather, results from this study suggest that failure to provide may have an accumulating effect on delinquency and aggression at both ages 12 and 16. If failure to provide has an accumulating association with the child’s behaviors, then regardless of the age of the child, providing for the basic needs of the child is important. However, these results were not robust to the addition of foster care status, study site, physical abuse, and sexual abuse. Given that the sampling method varied by study site, controlling by study site also controls for maltreatment experiences, similar to the other control variables that changed the results – foster care status, physical abuse, and sexual abuse. In contrast, the results were robust to the addition of demographic control variables (gender, race/ethnicity, poverty). Different types of maltreatment and neglect are highly correlated, and thus, controlling for the maltreatment variables may also control for the variation in neglect, yielding inconsistent results.

The results for lack of supervision also did not support our hypothesis that there is a cumulative effect of supervision. While the results for age 12 suggest that there is a cumulative effect of lack of supervision on the adolescent’s aggression and delinquency, these results were not robust to the addition of control variables. On the other hand, the most robust results of this study were for delinquency and aggression at age 16, which indicated that early adolescence (ages 13–16) is a critical period for lack of supervision. Thus, the results of this study are consistent with Dubowitz et al.’s (2019, 2022) findings that neglect in adolescence are associated with higher risk for negative adolescent and young adult outcomes compared to chronic neglect.

However, Dubowitz et al.’s (2019) study empirically identified longitudinal patterns of neglect, and although their measure of neglect was multidimensional, they did not parse out subtypes of neglect, and thus their study did not allow for differential associations between subtypes of neglect and outcomes. This study, on the other hand, found robust findings for a critical period of a lack of supervision in early adolescence for delinquency and aggression at age 16, but not for failure to provide. Thus, these results highlight the importance of attending to subtypes of neglect for multiple reasons. Many studies focus on child maltreatment and do not separate neglect from abuse (e.g., Lee & Villagrana, 2015). Yet, knowledge of the specific effect of neglect versus abuse has important implications for policies and interventions, since neglect is an act of omission whereas abuse is an act of commission. Second, there is no clear standard on subtypes of neglect, nor how to measure neglect (e.g., Coohey, 2003; Graham et al., 2010; Haworth et al., 2024). As scholarship on neglect continues to develop, it may be worth developing a clear standard for measuring specific types of neglect. For example, this study shows significant results for both failure to provide and lack of supervision, whereas other studies do not (e.g., Logan-Greene & Semanchin Jones, 2015). This suggests that future studies should continue to use measures of specific types of neglect, rather than a general measure of neglect.

It is noteworthy that although the findings for failure to provide were less robust, the changes for the age 16 outcomes for failure to provide were from the cumulative model to early adolescence as a critical period model. Although results from this study were not robust, suggesting both an accumulating and critical period effect of neglect, results nevertheless indicated that failure to provide during early adolescence may have a particular significance. The results from this study, alongside Dubowitz et al.’s studies, contribute to a growing body of evidence that highlights the importance of studying neglect in adolescence. Similar analyses should be conducted with different samples to establish the robustness of these results and to clarify how these control variables impact the results.

In addition to highlighting the possibility that subtypes of neglect may have differential effects on outcomes, another strength of this study is the estimation of models at ages 12 and 16. The results for lack of supervision for the models at ages 12 and 16 suggest an accumulating effect of neglect until age 12, followed by robust findings for a critical period of adolescence (ages 13–16). These results support the notion that the effects of neglect may change with age (Evans & Burton, 2013). Thus, although this study did not find support for a critical period in early childhood, these results may not necessarily be in conflict with prior studies that found an association between childhood experiences of neglect and adolescent outcomes (Chapple et al., 2005; Evans & Burton, 2013; Watts, 2017). These prior studies used measures of neglect that were limited to childhood, without consideration of chronicity nor exposure in adolescence.

Additionally, building on prior studies of chronicity (Graham et al., 2010; Logan-Greene & Semanchin Jones, 2015; Ryan et al., 2013), finding of a stronger effect of adolescent maltreatment conflates duration and timing, and thus may support the cumulative hypothesis as well. In fact, Thornberry et al.’s (2010) finding of a stronger effect of adolescent maltreatment conflates duration and timing may support the cumulative hypothesis as well. Additionally, this study adds to the burgeoning understanding of the most appropriate longitudinal measure of neglect. Graham at al. (2010) showed that longitudinal measures should take into account developmental periods. These results support this finding, especially given that the longitudinal effect of lack of supervision may change in adolescence. Taken as a whole, it is clear that point-in-time measures of neglect are limited, and future studies should take into account both the duration and timing of exposure to neglect.

At the same time, it is clear that the nature of supervision changes as the child ages. In adolescence, lack of supervision is often conceptualized by researchers as being closely related to parental monitoring (Ryan et al., 2013), referring to whether parents generally know where their adolescents are and what they are doing, especially as related to delinquent behaviors. Parental monitoring may not be the best way to measure supervisory neglect, since it may also capture the quality of the parent–child relationship related to communication, trust, and negotiation (Fletcher et al., 2004; Stattin & Kerr, 2000). Thus, as Ryan et al. (2013) pointed out, this measure may also conflate the parent–child relations, or parent–child conflict, with lack of supervision. Thus, it is possible that the change in the results between ages 12 and 16 might reflect such changes in the relationships participants had with their caregivers at the time.

5.1. Practice and policy implications

Increased evidence for the harm done by neglect shows the need for child welfare agencies to respond as urgently to neglect as they do other forms of maltreatment. Overall, these findings suggest that there may be an accumulating effect of neglect or a critical period in early adolescence. Either way, adolescence is an important period for intervention to disrupt the effects of neglect. In other words, disrupting neglect, regardless of the developmental period and including adolescence, may protect healthy child development. To disrupt failure to provide neglect in adolescence, providing concrete services to families may be a key intervention. For example, public assistance programs that provide food, housing, medical and dental care, and other basic needs for children and families may help families provide for their children. These programs should be protected for families with children of any age, including adolescents, as they may reduce the potential harm that results from failure to provide neglect.

To address instances of lack of supervision neglect, programs and policies can provide supervision for the child of all ages. Caregivers who work low-income jobs may struggle to find appropriate supervision for their children, as these types of jobs often yield unpredictable, irregular, and nonstandard (i.e., weekends and nights) schedules which may make childcare difficult to plan and afford (Lambert et al., 2014). Thus, one approach is to increase access to childcare for those with irregular schedules. Alternatively, investing in more after-school programming may also be an effective, safe, community-based intervention for this type of neglect (Heymann & Earle, 2011). Providing no-cost after-school activities, particularly in less affluent neighborhoods, will provide not only supervision to children, but also opportunities to develop and grow. For adolescence, after-school activities may take the form of drama club or sports teams or part-time work activities – this can be more flexible for young teens, but the point is to provide structured and supervised activities, especially for caregivers with fewer resources. Through these structured and supervised activities, adolescents’ exposure to risky situations, which Kobulsky et al. (2022) found to be associated with a higher likelihood of delinquency, can be minimized.

5.2. Limitations

Results from this study should be interpreted with caution in light of its limitations. First, there are a number of limitations with this dataset. The sample is not nationally representative, and the sites recruited participants differently, which means that maltreatment exposure varied by site. Additionally, these data were collected in the 1990s and 2000s and may be dated in some respects. However, we are unaware of more updated data – thus, the field needs more recent, high-quality longitudinal data to capture societal changes that affect child development. Additionally, there are limitations to these measures of neglect, which still do not capture all types of neglect and only reflect neglect that is reported to CPS. Yet, these measures of neglect are established measures and common subtypes within child welfare systems.

There are also some limitations with respect to the analyses. First, given the analytic approach, only one control variable could be tested at a time. It is possible that the results would change if all the control variables could be included in a single model. Second, this study only tests two hypotheses, thus providing a simplistic test comparing duration versus timing. In fact, the cumulative models assumed a linear association between duration and outcome, when it is possible that there is a nonlinear association. Furthermore, it is possible that the lack of robustness suggests that both duration and timing are important. For example, it is possible that chronicity that is initiated later in development, which reflects a combination of duration and timing, may be the more accurate hypothesis. Additionally, there are other possible hypotheses that are not tested in this study, such as the recency hypothesis: the most recent period is the period with an effect on the outcome. Future studies can build on this initial study by using this method to compare more sophisticated models that incorporate duration and timing, capture nonlinear associations, or test other hypotheses such as the recency hypothesis. Third, our four developmental periods which span 3–4 years each provide a limited measure of chronicity. Future studies should use datasets with more frequent measures of neglect that can provide a more precise measure of chronicity, which would also facilitate the testing of more sophisticated hypotheses, such as late onset chronicity of neglect. Fourth, this study did not provide an exhaustive examination of subtypes of neglect, such as emotional neglect; future studies should consider emotional neglect and other subtypes of neglect. Finally, the outcomes were measured contemporaneously with some of the time periods used, and thus this study does not provide evidence to support causal relationships.

6. Conclusion

This study advances knowledge of neglect by using a novel model-testing approach to test competing hypotheses about the longitudinal effects of two types of neglect on both delinquency and aggression. The effects of failure to provide may accumulate throughout childhood and appear to be associated with increased delinquency and aggression. Similarly, lack of supervision also may accumulate throughout childhood, although the findings indicate that lack of supervision during early adolescence is a critical period. In both cases, this disrupts conventional wisdom that neglect only matters early in child development. Rather, this makes a case for the importance, not of just preventing harm (i.e., acts of commission), but also of providing sufficient supports to ensure healthy development (i.e., acts of omission). More than that, ensuring the child and family have sufficient resources to facilitate healthy development has the potential to disrupt the development of delinquency and aggression, and thus such investments in child development matter, no matter the age of the child.

Data availability

The authors do not have permission to share data.