Depressed Mood and Maternal Report of Child Behavior Problems: Another Look at the Depression–Distortion Hypothesis

Abstract

Caregiver depression has been described as leading to overreport of child behavior problems. This study examines this “depression–distortion” hypothesis in terms of high-risk families of young adolescents. Questionnaire and diagnostic interview data were collected from mothers, teachers, and fathers, and self-report information was obtained from youth between ages 10 and 14 years. First, convergent and discriminant validity were demonstrated for internalizing and externalizing multiagent constructs. Second, the depression–distortion hypothesis was examined, revealing a modest effect of maternal depression, leading to the inflation of reported son externalizing and daughter internalizing problems. The data suggest the need to consider multiple influences on parental perceptions of child behavior and psychopathology in research and clinical settings.

1. Introduction

Child behavior problems and maternal depression have been linked in a multitude of investigations involving both clinical and community samples (Downey & Coyne, 1990; Goodman & Gotlib, 1999). Children of mothers who experienced clinical levels of depression were found to be at increased risk for psychopathology in general and for behavior problems in particular (Fendrich, Warner, & Weissman, 1990; Weissman et al., 1984; Welsh-Allis & Ye, 1988). This association between mothers’ depressive symptoms and child behavioral and emotional difficulties has been demonstrated for families of preschoolers participating in the treatment of child disruptive disorders (Webster-Stratton & Hammond, 1988) and community samples that included children of various ages (Egeland, Kalkoske, Gottesman, & Erickson, 1990; Krain & Kendall, 2000). The strength of the association between maternal depressive symptoms and child behavior problems is borne out by the consistency of findings across developmental stages from preschool (Campbell, Pierce, Moore, Marakovitz, & Newby, 1996) through adolescence (Thomas, Forehand, & Neighbors, 1995). In addition, studies using diverse measurement approaches, including diagnostic interviews, self-report questionnaires, and observations, have yielded convergent findings (Egeland et al., 1990; Fendrich et al., 1990; Hops et al., 1987).

It should be noted that maternal depressive symptoms have been shown to predict increased risk for child externalizing and internalizing type problems differently as a function of child gender (Essex, Klein, Cho, & Kraemer, 2003). On the other hand, divergent results have also been obtained. Hammen (1991), for example, did not detect gender differences in the impact of maternal depression. Interestingly, Hops (1996) found that girls of depressed mothers experienced more pronounced effects, particularly when older, whereas other investigators reported more extensive effects for younger boys. More recently, maternal depression was found to be associated with both externalizing and internalizing behavior problems for boys, but not linked with behavior problems/symptoms for female offspring (Carter, Garrity-Rokous, Chazan-Cohen, Little, & Briggs-Gowan, 2001). Given these inconsistent findings, we explored the existence of gender differences in associations between maternal symptoms of depression and child behavior problems, but did not formulate specific a priori hypotheses about the exact nature of such differences.

A number of mechanisms have been proposed to explain the association between maternal depression and child behavior problems (Campbell et al., 1996; Cummings & Davies, 1994; Goodman & Gotlib, 1999; Patterson, 1980). First, mothers’ symptoms of depression may lead to an actual increase in child behavior problems, either directly via symptoms of depression (e.g., dysphoria) that manifest while parents interact with their children, or through the effects of depressive symptoms on parenting/parent–child interactions. A presentation characterized by emotional unavailability, negative affect, and cognitions associated with depression likely diminish the quality of parent–child interactions (Cummings & Davies, 1994; Goodman & Gotlib, 1999). In addition, maternal depression has been linked with the use of ineffective behavior management practices (e.g., harsh or inconsistent discipline and inadequate supervision) and marital/family discord, which in turn may contribute to the development and maintenance of child behavior problems (Forehand, Lautenschlager, Faust, & Graziano, 1986).

The depression–distortion hypothesis (Richters & Pellegrini, 1989) has also been proposed as an explanatory mechanism for the association between maternal depression and child behavior problems. Specifically, dysphoric emotions associated with depression were hypothesized to activate a negative perceptual bias in the mothers’ ratings of child behaviors that, in turn, leads to overreporting of child adjustment difficulties (Field, 1992; Geller & Johnston, 1995; Griest, Wells, & Forehand, 1979; Johnston & Short, 1993). A number of studies have called into question the accuracy of maternal report of child behaviors when mothers experience symptoms of depression (Breslau, Davis, & Prabucki, 1988; Friedlander, Weiss, & Traylor, 1986). Schaughency and Lahey (1985), for example, warned about the risk of pathologizing child behavior on the basis of maternal ratings if the level of maternal depression is not taken into account. Friedlander et al. (1986) concluded that maternal depressive symptoms had a significant impact on the report of child behavior, although they did not regard maternal ratings as completely invalid.

However, evidence contrary to the depression–distortion hypothesis has also been presented. For instance, Pellegrini (1989) and Richters (1992) examined correlations between ratings of child behavior problems provided by mothers and by teachers and reported that the associations were not affected by the level of maternal depressive symptomatology or mood disorder diagnosis, and they noted that this pattern of findings was not consistent with the depression–distortion hypothesis. When Richters and Pellegrini (1989) evaluated state–trait dimensions of maternal depression as mediators of the agreement between teachers and mothers, their findings failed to demonstrate any significant differences in agreement as a function of maternal depression status. However, children of mothers who experienced a current depressive episode or who reported a history of one or more depressive episodes were described by mothers and teachers as having exhibited more frequent behavior problems than had control children.

Investigations into discrepancies between different sources of information about child symptoms of psychopathology and behavior problems, and the impact of parents’ own symptoms on discrepancies between their ratings and those of other informants, have continued. Treutler and Epkins (2003), for example, examined whether parents’ symptoms made unique contributions to mothers’ and fathers’ reports of children’s behavior problems, as well as to discrepancies between mother–child, father–child, and father–mother reports. Maternal and paternal symptomatology contributed to father–mother discrepancies regarding internalizing and externalizing type difficulties, indicating that differences between caregivers’ reports were a function of both caregivers’ symptoms. Chi and Hinshaw (2002) investigated the depression–distortion hypothesis by examining the effects of maternal depressive symptoms on cross-informant discrepancies in reports of child behavior problems for a sample of children identified as presenting with attention deficit hyperactivity disorder (ADHD), combined type. Overall, maternal depressive symptoms predicted negative biases in their reports of their child’s ADHD symptoms, general behavior problems, and their own negative parenting style. Interestingly, increases in mothers’ symptoms of depression were associated with maternal reports of negative parenting, but not with indicators based on laboratory observations of parent–child interactions, suggesting possible depression–distortion effects. De Los Reyes and Prinstein (2004) applied the depression–distortion hypothesis to the assessment of peer victimization from the perspective of the victim, and examined whether adolescents’ depressive symptoms and aggressive behavior were associated with discrepancies between self- and peer reports of peer victimization experiences. Adolescents’ aggression was associated with underestimation of peer victimization directed toward the teen, on self-report instruments relative to peer reports, whereas depressive symptoms were associated with overestimations of own peer victimization on self-report compared with peer reports.

Structural equation modeling (SEM) techniques have been used to directly test the hypothesis that maternal depression contributes to over-report of behavior problems and thus leads to divergence from indicators derived from other informants, and an increase in the error associated with maternal ratings of child behavior problems. Fergusson, Lynskey, and Horwood (1993) evaluated three models: (a) maternal depression was unrelated to maternal report of child difficulties, (b) maternal depression causally influenced maternal report of child difficulties, and (c) maternal depression was correlated with error in maternal report of child difficulties. A sample of adolescent boys was included in this research, and information regarding conduct disorder and attention deficit disorder was gathered from teachers, mothers, and children. Fergusson and colleagues detected moderate associations between maternal depression and reporting errors, supporting the latter two models. However, questions remain regarding the applicability of these findings to girls as well as to other areas of child functioning (e.g., internalizing-type difficulties, including symptoms of depression and anxiety).

This study applied statistical advancements to understanding the extent to which maternal depression distorts reports of child and adolescent psychopathology (Fergusson et al., 1993; Richters, 1992). First, an array of agents (e.g., mothers, fathers, teachers, and children) provided information about child adjustment that served as the bases for the multi-agent constructs of internalizing and externalizing problems. Second, a series of analyses was undertaken to ascertain convergent and discriminant validity. These analyses were conducted to address the need to establish superior validity of multiagent constructs used as criteria for ascertaining maternal reporting errors (Fergusson et al., 1993; Richters, 1992). Finally, direct tests of the contribution of maternal depressive symptoms to over-reporting, and thus the error associated with mother report of externalizing and internalizing difficulties in the context of multi-agent constructs, were performed separately for sons and for daughters. The depression–distortion hypothesis was tested by using a series of SEM models, following methodological recommendations (Fergusson et al., 1993; Stoolmiller, 1998). These tests were conducted separately for females and for males based on past research suggesting gender differences in psychopathology and behavior problems among youth (e.g.; Hayward & Sanborn, 2002; McManus, Alessi, Grapentine, & Brickman, 1984; Rutter, Izard, & Read, 1986).

First, it was hypothesized that the validity evaluations conducted in the course of this study would support the validity of the multi-agent constructs addressing internalizing and externalizing behavior problems. Specifically, results of the evaluation of convergent and discriminant validity were expected to support the appropriateness of independent, but related, internalizing and externalizing problems constructs. In addition, it was hypothesized that models accounting for correlated residuals across different indicators provided by the same informant would be superior to those not accounting for the non-independence of these error terms (Garber, Quiggle, Panak, & Dodge, 1991; Nelson, Hammen, Brennan, & Ullman, 2003). Second, relations between maternal depressive symptoms and child behavior problems were hypothesized, with higher levels of maternal depression being associated with higher levels of child behavior problems. Third, we hypothesized that higher levels of maternal depressive symptoms would be linked with increases in the measurement error associated with mothers’ report of adolescent psychopathology.

2. Method

2.1. Participants

The data presented in this study were collected in the context of the Adolescent Transitions Program (ATP), a group intervention curriculum for teens and their parents. Participants in this study were self-referred and had learned about the program through a variety of sources, including community and newspaper advertisements and teacher and counselor referrals (see Dishion & Andrews, 1995; Dishion, Andrews, Kavanagh, & Soberman, 1996). The program was fully explained during the first telephone contact, which had been initiated by the parents inquiring about ATP. During this initial contact, interested parents participated in a telephone screening interview developed on the basis of risk factor research conducted by Bry, McKeon, and Pardina (1982). Parents were questioned regarding 10 dimensions of child risk (closeness to parents, emotional adjustment, academic engagement, involvement in positive activities, experience seeking, problem behaviors, the child’s substance use, peer substance use, family substance use history, and stressful life events), and the criterion for inclusion in the intervention was based on a cut-off of 4 factors out of 10 being endorsed by the parent (Bry et al., 1982; Dishion & Andrews, 1995). This inclusion/exclusion criterion led to the retention of 50% of the original sample, all of whom participated in the baseline assessment. Thus, data were collected from mothers (N = 219), teachers (N = 218), and fathers (N = 123) of 221 children (113 girls and 108 boys), who also provided self-report data. The youth were between ages 10 and 14 years, M (SD) = 12.31 (.80) for females and M (SD) = 12.43 (.87) for males. Slightly more than 47% of the participating families represented one-parent households, determined on the basis of the mother’s report of having been a single parent for 1 or more years. The average number of children per family was 2.2.

This sample was primarily low income, but moderately well educated. The majority of mothers (85.7%) and fathers (78.6%) graduated from high school; more than 50% of mothers and 45% of fathers had some college education. However, the median annual household income was between $15 000.00 and $20 000.00, and the level of unemployment (23.2%) and the percentage of families receiving financial assistance (58%) were relatively high. More than 90% of the families were European American, roughly representing the ethnic distribution in the community (Dishion & Andrews, 1995).

2.2. Measures

Child Behavior Checklist (CBCL)

The CBCL (Achenbach, 1991) is a widely used measure of children’s social and academic functioning and behavioral problems. The behavior problems portion includes 118 items that yield scores for eight narrowband clinical scales, two broadband (composite) scales, externalizing and internalizing behavior problems, and the total behavior problems score. The externalizing and internalizing composite scores, based on the reports of mothers and fathers, were used in this investigation. Adequate reliability and validity have been reported for the CBCL (Achenbach, 1991). With the ATP sample, youth internalizing items were found to be internally consistent on the basis of mother (α = .86) and father report (α = .88); externalizing items were also found to be reliable (α_mothers = .92; α_fathers = .93).

2.2.1. Teacher Report Form (TRF)

The TRF (Achenbach, 1991) represents a teacher report version of the CBCL. The TRF has been frequently used in a wide range of studies, providing evidence of its reliability and validity. This measure yields a number of narrowband and composite scores parallel to the parent-report measure and can be scored in a manner that produces content consistent with that of the CBCL (Achenbach, 1991). Internalizing and externalizing composite scores, based on teacher report, were used in this study (α_{externalizing} = .94; α_{internalizing} = .89).

2.2.2. Center for Epidemiological Studies Depression Scale (CES–D)

The CES–D (Radloff, 1977) is a 20-item self-report scale of depression, with questions addressing presence or absence of negative and positive thoughts, feelings, and behaviors, as well as the somatic manifestations of depression. Satisfactory reliability and validity have been demonstrated for this screening measure of depressive symptoms. A total score (sum) was computed for mothers participating in this study, and items were found to be internally consistent (α = .92). The CES–D represents one of the indicators of maternal depression evaluated in this study. The frequency of clinically significant reports of depression for mothers (34%) was somewhat higher in this sample than for the CES–D normative sample (21%; Radloff, 1977), based on the recommended cut-off score of 16 on the CES-D (Radloff, 1977).

2.2.3. Brief Symptom Inventory (BSI)

The BSI (Derogatis, 1993) is a 53-item self-report symptom inventory developed to capture psychological symptomatology for clinical and community samples. A brief form of the Symptom Checklist 90–Revised (SCL–90–R), it uses a 5-point Likert scale ranging from 0 (not at all distressful) to 4 (extremely distressful). The measure is scored to generate profiles spanning nine symptom dimensions and three global distress markers. Two of the nine symptom dimensions were used in this study: mother report of depression (six items; α = .86) and youth report of both depression (six items; α = .85) and hostility (five items; α = .73). Mother depression score was used in the context of the depression construct, along with the CES–D indicator, whereas youth depression and hostility scores were included in the internalizing and externalizing multi-agent constructs, respectively.

2.3. Procedure

Measures reported in this study were administered at baseline of a longitudinal intervention evaluation project. Mothers and fathers completed the CBCL, teachers responded to the TRF, and youth filled out the BSI to assess youth externalizing and internalizing problems at baseline. At the same time, mothers completed both indicators of maternal depression (CES–D and the BSI). Parents and children completed their questionnaires at home at their convenience, returned them to researcher staff during a scheduled visit, and were reimbursed $10 for their effort. Teachers completed their portion of the assessment in the school setting, returned completed forms to the researchers, generally by mail, and were reimbursed $8.00 for participation. Parents and children participated in the ATP intervention after the baseline assessment. The intervention protocols are described in detail in Dishion and Andrews (1995), Dishion et al. (1996), and Poulin, Dishion, and Burraston (2001).

2.3.1. SEM Analytic Strategy

EQS 6.1 (Bentler, 2004) was used for all SEM analyses. All SEM analyses were conducted using maximum likelihood estimation for parameter estimation (Myung, 2003), with missing data handled using maximum likelihood estimators (Arbuckle, 1996). The following fit indices, in addition to chi-square tests, were reported for all models tested: Adjusted Goodness of Fit Index (AGFI; Raykov & Marcoulides, 2000), Akaike Information Criterion (AIC; Akaike, 1987), Incremental Fit Index (IFI; Bollen, 1989), Comparative Fit Index (CFI; Bentler, 1990), and Root Mean-Square Error of Approximation (RMSEA; Loehlin, 1998). Collectively, these fit indices satisfied the following criteria: (a) minimally impacted by sample size (Bollen, 1986), (b) sensitivity to model misspecification, and (c) lack of upward or downward systematic bias (Fan, Thompson, & Wang, 1999). For the AGFI, CFI, and IFI, values larger than .90 are indicative of adequate model fit, with values ranging between .95 and 1.00 considered a well-fitting model. For both the RMSEA and the AIC, lower values reflect improved model fit relative to higher values, with the AIC not constrained by a lower or upper bound. However, for the RMSEA, values between 0 and .05 reflect good model fit, with values between .06 and .10 reflecting adequate model fit. RMSEA values greater than .10 reflect models with questionable fit. Finally, change in chi-square tests were used to examine differences between tested models.

To examine the convergent and the discriminant validity of child psychopathology, a confirmatory factor analytic (CFA) approach was used, with a single-factor model, a single factor with correlated errors model, and a two-factor (externalizing and internalizing) model with correlated errors compared. Specifically, mother and father CBCL, teacher TRF, and child self-reported BSI Time 1 assessments of internalizing and externalizing behaviors were used. Correlated errors were anticipated because of the potential for rater bias, which can occur when an individual reports on one or more constructs (e.g., an indication of one symptom may increase the likelihood of reporting other symptoms across different areas of functioning; Nelson et al., 2003). The best-fitting model resulting from the CFA approach described earlier was adopted for subsequent analyses.

After the convergent and discriminant validity of the multi-agent construct were established, the depression–distortion hypothesis was examined. For these analyses, the null model was one without a path from the latent construct of maternal depression to the error term associated with maternal report of child (sons’ or daughters’) internalizing/externalizing symptoms. The model compared against the null model included a path from the latent construct of maternal depression to maternal report of child internalizing/externalizing symptoms. This set of analyses resulted in four alternative models being tested against their respective null models.

3. Results

3.1. Preliminary Analyses

Prior to conducting SEM analyses, we examined the potential impact of missing data from fathers (n = 101, or 47.34% of potential father respondents had missing data) on maternal and child outcomes relevant to the multi-informant nature of the SEM analyses and the potential implications for maternal and child well-being in the presence of missing father data. Unfortunately, adequate data regarding divorce/marriage rates for the majority of participants were not reported (74.3% did no respond to this question). Nevertheless, of the 58 participants who responded to this particular question, 60.4% indicated being currently married, with 39.6% reporting either being divorced or never married. This information suggested that mothers and children participating in this study experienced rates of family disruption consistent with widely reported rates of divorce across the country, and that it would not be appropriate to conclude that missing data from fathers (44.7%) was solely a function of family disruptions such as divorce. A series of 2 (child gender: male vs. female) × 2 (paternal report present: missing paternal data vs. no missing paternal data) ANOVAs were conducted on the following variables: a combined maternal depression score consisting of maternal CES-D and maternal BSI depression, child reports of their own BSI depression and hostility symptoms, and maternal and teacher reports of CBCL child internalizing and externalizing difficulties. No significant effects were observed for child reports of their own BSI depression or hostility symptoms or maternal and teacher reports of CBCL child internalizing difficulties (all p > .05). However, significant main effects for gender were observed with maternal, F(1, 215) = 6.95, p < .05, and teacher, F(1, 211) = 20.89, p < .05, report of externalizing difficulties. Both mothers and teachers indicated significantly higher externalizing symptoms for males (M = 20.58 and 14.85, respectively) compared with females (M = 16.81 and 6.98, respectively). These findings, consistent with those of recent investigations, showed similar discrepancies between male and female children on externalizing difficulties (Rescorla et al., 2007a; Rescorla et al., 2007b). Significant main effects of paternal report present on maternal- and teacher-reported externalizing difficulties were not observed, F(1, 215) = 0.33, p > .05 and F(1, 211) = 0.049, p > .05, respectively, nor were significant gender × paternal report present effects on maternal- and teacher-reported child externalizing difficulties noted, F(1, 215) = 0.36, p < .05 and F(1, 211) = 0.074, p > .05, respectively.

For the purposes of this set of analyses, maternal CES-D and BSI depression scores were summed to create a composite maternal depression score because of the association between maternal CES-D and BSI depression scores, r(216) = .82, p < .001. Results of analyses relying on this depression composite as a dependent variable indicated that there were no main effects for child gender, F(1, 214) = 0.061, p > .05, or paternal report present, F(1, 214) = 1.76, p > .05, nor was there a child gender × paternal report present interaction, F(1, 214) = 0.51, p > .05. Thus, evidence suggests that the ability of paternal caregivers to participate or not participate (for any reason) in the current study will have a negligible, if any, systematic impact on subsequent analyses.

Finally, given the multi-informant nature of this study, associations between maternal, paternal, teacher, and child reports of internalizing and externalizing symptoms were examined. These findings are presented in Table 1, with most associations emerging as significant and occurring in the anticipated direction across informants.

Table 1.

Associations between Multiinformant Ratings of Child Internalizing and Externalizing Difficulties

Informant	Mother EXT	Father EXT	Teacher EXT	Child EXT	Mother INT	Father INT	Teacher INT	Child INT
Mother EXT	—	.72**	.42**	.13*	.46**	.21**	−.01	.07
Father EXT		—	.42**	.08	.21*	.55**	.17	−.06
Teacher EXT			—	.09	−.01	.13	.46**	.01
Child EXT				—	.15*	.09	.11	.59**
Mother INT					—	.35**	.17*	.22**
Father INT						—	.24**	.17
Teacher INT							—	.18**

Note. Sample sizes varied across associations (Ns = 219, 123, 218, and 221 for mothers, fathers, teachers, and children, respectively), which is reflected in the p-values associated with each correlation. EXT = externalizing; INT = internalizing.

^*p < .05;

^**p < .01

3.2. Validation of Externalizing and Internalizing Criterion Measures

Findings associated with the discriminant validity analyses supported a model with two separate but related constructs of externalizing and internalizing behavior problems. The initial single-factor model provided a poor fit to the data. The final two-factor model with error provided a significant improvement in fit, Δ χ² (df = 1) = 34.86, p < .001, relative to the single-factor, correlated error model. Examination of factor loadings from the final model (range .17 to .83, average loading .49) provided support for convergent validity of the different indicators (i.e., mother, father, teacher, and child report) for the internalizing and externalizing constructs.

3.3. Depression–Distortion Hypothesis Results

Given the evidence of higher levels of externalizing difficulties in male offspring relative to female offspring as noted by both mothers and teachers (see Preliminary Analyses above), the study proceeded with separate depression–distortion SEM analyses for sons and daughters, with separate analyses on internalizing and externalizing constructs for each. The null and alternative model estimates for all depression–distortion models are presented in Table 2, with the alternative models (i.e., daughters: Models 2 and 4; sons: Models 6 and 8), including paths of interest from maternal depressive symptoms to the error term associated with mothers’ report of child behavior problems (depicted schematically in Figures 1–4). It should be noted that in EQS software, residuals must be designated as independent variables (e.g., have no one-way arrow leading to them from another variable). In order to test the depression–distortion models by using EQS, the residuals associated with mother report of child internalizing and externalizing behaviors had to be redesignated as a latent/unobserved variable. This redesignation resulted in a change from an independent to a dependent variable status, but had no effect on the meaning of the variable in the model or on the associated paths (Bentler & Wu, 1995).

Table 2.

Fit Indices Associated with Analyses of Convergent/Discriminant and Predictive Validity, and Depression–Distortion Hypothesis

Analyses	Model	Chi-square	(df)	AGFI	AIC	IFI	CFI	RMSEA (90% confidence)
Discriminant validity
	Single factor	250.84***	(20)	.70	210.84	.60	.59	.23 (.20 to .25)
	Single factor w/error	57.12***	(16)	.89	25.11	.94	.94	.11 (.08 to .14)
	Two factor w/error	22.61	(15)	.96	−7.74	.99	.99	.05 (0 to .08)
Depression–distortion (daughters)
	Internalizing Model 1	14.06	(9)	.92	−3.95	.97	.97	.07 (0 to .14)
	Internalizing Model 2	4.63	(7)	.97	−9.37	1.02	1.00	0.00 (0 to .08)
	Externalizing Model 3	12.58	(9)	.93	−5.42	.99	.99	.06 (0 to .13)
	Externalizing Model 4	11.89	(8)	.93	−4.11	.99	.99	.07 (0 to .14)
Depression–distortion(sons)
	Internalizing Model 5	16.50*	(9)	.93	0.50	.99	.99	.10 (.02 to .16)
Internalizing Model 6		16.45*	(8)	.93	0.45	.99	.99	.10 (.02 to .16)
Externalizing Model 7		8.19	(8)	.96	−7.81	1.01	1.00	.02 (0 to .11)
Externalizing Model 8		3.39	(7)	.98	−10.61	1.03	1.00	0.00 (0 to .07)

^*p < .05;

^**p < .01;

^***p < .001

Figure 1.

Depression–distortion Model 2: Daughter internalizing problems.

Figure 4.

Depression–distortion Model 8: Son externalizing problems.

3.3.1. Daughters

Model 2 (Figure 1), reflecting the distortion hypothesis, produced a significantly better fit to the data than did the corresponding Model 1, Δχ² (df = 2) = 9.42, p < .01. In Model 2, the path from the latent variable of maternal depression to the error term associated with maternal report of daughter internalizing problems was significant, indicating that increased maternal depression was associated with higher error term values. However, Model 4, with a path from the latent variable of maternal depression to the error term associated with maternal report of daughter externalizing behavior (Figure 2), failed to improve overall model fit, relative to Model 3, Δχ² (df = 1) = .69, p > .05. Furthermore, the path from the latent variable of maternal depression to the error term of interest was not significant, indicating that maternal depression did not have a distorting effect on maternal report of daughters’ externalizing symptoms.

Figure 2.

Depression–distortion Model 4: Daughter externalizing problems.

3.3.2. Sons

Model 6 with a path from the latent variable of maternal depression to the error term associated with maternal report of son internalizing behavior (Figure 3) did not provide a better fit to the data than did Model 5, Δχ² (df = 1) = .05, p > .05. Likewise, the path from the latent variable of maternal depression to the error term associated with maternal report of son internalizing behavior was not significant. However, a significant association between the latent variable of maternal depression and the latent variable of son internalizing behavior problems was present, which suggests a direct link between maternal depression and increased internalizing symptoms in male offspring. Model 8 (Figure 4), including a path from the latent variable of maternal depression to the error term associated with maternal report of son externalizing behavior, demonstrated a significantly better fit for the data than did Model 7, Δχ² (df = 1) = 4.8, p < .05. It is important to note that consistent with the depression–distortion hypothesis, the path from the latent variable of maternal depression to the error term associated with maternal report of son externalizing disorder was significant, which indicated that increased maternal depressive symptoms were associated with increased error in maternal reporting of sons’ externalizing behaviors.

Figure 3.

Depression–distortion Model 6: Son internalizing problems.

4. Discussion

This study addressed the depression–distortion hypothesis for a sample of parents whose children can be described as “at risk” for the development of adjustment difficulties (Dishion & Andrews, 1995). The study was designed to be consistent with a number of methodological recommendations relevant to questions regarding the depression–distortion hypothesis (e.g., Fergusson et al., 1993; Richters, 1992). A number of informants were included in the evaluation of behavioral/emotional difficulties for children, enabling the construction of multi-agent latent variables (LVs) for internalizing and externalizing problems. The LVs were evaluated rigorously to ensure convergent and discriminant validity. It was hypothesized that the results of this investigation would support relations between maternal depression and child behavior problems, and that this association would be explained by the depression–distortion effect (i.e., relations between maternal symptoms of depression and the error associated with maternal report of child behavior problems). Thus, higher levels of maternal depressive symptoms were expected to lead to over-reporting of child behavior problems, resulting in an increased error term for the maternal report indicator.

The impact of maternal depression on the error associated with maternal report of child behavior problems (i.e., the depression–distortion hypothesis) was directly evaluated by using SEM procedures. They enabled us to discern the appropriateness of models that included paths from maternal depression LV to the error terms associated with maternal report of child behavior problems. Specifically, these paths of interest (i.e., paths from maternal depressive symptoms to the error component of maternal report child behavior scores), as well as the improvement in the overall model fit, were examined. In addition, we evaluated potential gender differences without specifying predicted outcomes on an a priori basis, because of conflicting findings linking maternal depression and child behavior problems for boys and for girls (Carter, et al., 2001; Hammen, 1991; Hops, 1996).

The first step of the analytic process involved the development of multi-agent constructs addressing internalizing and externalizing behaviors. In the second step, construct validity was evaluated, with a focus on convergent and discriminant validity. Convergent and discriminant validity were assessed through a series of CFA models. Findings associated with the discriminant validity analyses supported a model with two separate, but related, indicators of externalizing and internalizing behaviors, and included correlated error terms (i.e., significant paths between error terms associated with the internalizing and externalizing scores provided by the same informant). A moderate association between indicators of internalizing and externalizing problems was predicted and is consistent with the existing literature demonstrating the association among behavior problems in different domains (Capaldi & Stoolmiller, 1999; Lilienfeld, 2003; Nelson et al., 2003; Youngstrom, Findling, & Calabrese, 2003). Predictions about the appropriateness of correlated error terms were also supported by the model testing performed in the context of the convergent/discriminant validity analyses. The latter finding is not particularly surprising, and speaks to the fact that each informant responded in a similar manner to items comprising the externalizing and internalizing domains of child behavior problems. This result could be anticipated from previous discussions of potential respondent bias (Garber et al., 1991), but has generally not emerged in earlier research addressing the depression–distortion hypothesis, because previous studies often focused on a single domain of child functioning or failed to include multi-agent constructs (Fergusson, et al., 1993; Field, 1992; Geller & Johnston, 1995). A notable exception is a study conducted by Nelson et al. (2003), wherein correlated residuals for different informants of child behavior problems were hypothesized and demonstrated for mother and youth report of internalizing and externalizing problems. Our findings are consistent with the Nelson et al., results in confirming the superiority of models that include correlated residuals. Convergent validity was supported by an examination of the factor loading associated with the most appropriate solution. Thus, results of the analyses addressing convergent and discriminant validity provided support for the construct validity of these LVs, consistent with earlier reports advocating the use of such constructs (Capaldi & Patterson, 1989, 1991).

Results of this study can be described as partly consistent with the depression–distortion hypothesis and prior reports of a negative depression-related bias in parental ratings of child behaviors (Field, 1992; Geller & Johnston, 1995; Griest, Wells, & Forehand, 1979; Johnston & Short, 1993). Overall, the association between maternal symptoms of depression and child behavior problems (distorted or veridical) was supported in six of the eight examined models. This association did not emerge as statistically significant only in the analyses of maternal depressive symptoms and daughter externalizing behavior problems. A significant association between maternal depression and child behavior problems was supported by the initial model addressing boys’ externalizing symptoms. However, after the path representing the relations between maternal depression and the error component of externalizing problems ratings for boys was introduced, the coefficient associated with the path from maternal depression to the externalizing LV became nonsignificant. Conversely, a significant association between a maternal depression and boys’ internalizing problems LV was observed even after the introduction of the path representing the distorting effect of maternal depressive symptoms. Thus, there was no support for the depression-related distortion in the evaluation of the association between maternal depressive symptoms and son internalizing problems. Rather, SEM analyses supported veridical relations between maternal depressive and son internalizing symptoms, with son symptoms increasing at higher levels of maternal dysphoria.

Evaluations of the associations between maternal depressive symptoms and child behavior problems have focused primarily on disruptive or externalizing difficulties, especially for boys (Campbell et al., 1996; Egeland et al., 1990; Webster-Stratton & Hammond, 1988). However, results of our investigation indicate that the association between mothers’ depression and sons’ internalizing problems (e.g., symptoms of depression and anxiety) may be equally important. This study’s findings are consistent with a recent report of similar gender-related differences, wherein boys’, but not girls’, internalizing symptoms were linked with the effect of maternal depression (Essex et al., 2003). Specifically, boys exposed to maternal depression in infancy demonstrated a preponderance of internalizing behaviors, whereas for girls, internalizing symptoms increased particularly as a result of exposure to marital conflict occurring in the toddler/preschool period. Our findings suggest that for male children, mothers’ symptoms of depression may have a direct effect on son’s internalizing behavior problems later in childhood. A presentation that includes emotional unavailability, negative affect, and cognitions associated with depression likely contributes to this effect (Cummings & Davies, 1994; Goodman & Gotlib, 1999) and leads to a similar constellation of difficulties for male offspring. Alternatively, this link between maternal symptoms of depression and boys’ internalizing symptoms may be a function of genetic influences, more pronounced in the male offspring. Previous research, however, has not provided evidence of differences in the genetic transmission of risk for depression for male and for female offspring, suggesting an environmental basis for the gender differences observed in this study (Rice, Harold, & Thapar, 2005).

Perhaps most important, significant positive coefficients emerged as indicators of the association between maternal symptoms of depression and the error component of maternal report for child behavior problems, indicating that the amount of reporting error increased at higher levels of depressive symptomatology, consistent with the depression–distortion hypothesis. It should be noted that the significance of this contribution varied as a function of domain of adjustment difficulties (i.e., internalizing versus externalizing) and child gender. Our findings indicated that mothers with higher levels of depressive symptoms over-reported externalizing behavior problems for sons and internalizing difficulties for daughters. Specifically, maternal depression accounted for 18% of the variance for son externalizing problems and 14% of the variance for daughter internalizing problems, respectively.

Thus, results of this study did not demonstrate a consistent pattern of increased error in estimates of child behavior problems resulting from maternal depression. Rather, this biasing effect was noted only for “gender concordant” behavior problems. That is, boys have been generally described as exhibiting higher levels of externalizing type difficulties, whereas higher levels of internalizing problems have been reported for girls. For example, estimates for prevalence of conduct disorder among children have ranged from 4% to 10%, with antisocial behavior being three times more common for boys (American Psychiatric Association, 1994). Boys have been rated by observers as more aggressive when engaging in naturalistic activities (Condry & Ross, 1985). Internalizing behaviors can be expected to cause greater difficulties for girls, given that women are significantly more likely than men to experience a major depressive episode, with this gender differentiation becoming evident by adolescence (American Psychiatric Association, 1994; Hankin et al., 1998).

Perhaps the perceptual bias related to symptoms of depression is activated for parent report of child behaviors in a manner consistent with overall attitudes and expectations. Presumably, mothers would expect their sons to exhibit more acting-out difficulties and their daughters to demonstrate more problems with anxiety or depression. These expectations could be amplified, leading to over-reporting, as mothers experience higher levels of depression. Prior research has not raised the possibility of depression–distortion effects being linked with child gender; however, this possibility deserves further evaluation in light of this study’s findings. It should be noted that these domain-by-gender differences were not anticipated on an a priori basis, and thus should be replicated in the future.

This study has a number of implications for research and clinical applications related to parental symptomatology and child adjustment difficulties. Our findings indicate that research and clinical activities should be conducted in a manner sensitive to the possibility of parental over-reporting for male externalizing and female internalizing difficulties. Specifically, gender-related issues should be taken into account when collecting data and providing clinical services. Studies addressing issues relevant to child development and psychopathology often do not incorporate gender-related analyses (Fergusson et al., 1993) because both genders are not represented adequately in the samples. Under these circumstances, investigators have limited opportunity to detect gender-related effects, which are especially important in studies addressing externalizing and internalizing behaviors, given prior reports of gender differences for these areas of child adjustment and this study’s findings.

The single most significant limitation of this study involves our inability to comprehensively investigate the proposed causal mechanisms responsible for the link between maternal symptoms of depression and child behavioral and emotional difficulties. Maternal depression has been linked with the use of ineffective behavior management practices and marital/family conflict, which in turn may contribute to the development and maintenance of child behavior problems (Forehand et al., 1986). Although these parenting/family factors (i.e., behavior management practices and family conflict) were beyond the scope of the present investigation, the influence of these variables could in part explain the association between maternal depressive symptoms and the error portion of the estimates of child behavior problems. That is, family conflict, for example, could mediate the association between maternal depression and reporting error associated with child behavior problems indices. This possibility should be directly evaluated in future research.

Further limitations of this investigation should be noted, including that the sample represents a group of parents whose children have been identified to be at risk for the development or exacerbation of adjustment difficulties (Dishion & Andrews, 1995). Results should be generalized with care, especially to children under 10 years of age, but are likely applicable both to parents in the community whose children are experiencing some behavioral difficulties and to parents of clinically referred children. Although generally representative of the local population, the study sample was limited in terms of its cultural and ethnic representation. In addition, the possibility that children and parents may be inflating child symptoms in a gender-stereotyped manner is relevant to this study and should be considered and evaluated in future research. That is, children likely formulate ideas about their own symptoms on the basis of information obtained from adults (teachers, and parents in particular), which may lead them to inflate their estimates of their own symptomatology, consistent with adults’ notions of gender roles.

Finally, the BSI was used as an indicator of emotional and behavioral difficulties with youth younger than age 13, because of the longitudinal nature of the larger project (i.e., ATP) from which the data were drawn for the present investigation. This limitation should be remedied in future research, in so far as the use of this instrument with younger children has not been validated to the extent as has its use with those older than 13.

Future investigations should attempt to replicate the gender-related findings reported in this study and address the potential influence of family/parenting variables that may be in part responsible for the association between parental depression and higher levels of reporting error noted earlier. In addition, studies should include various samples such as “at-risk” and normative or community samples of families, as well as families with caregivers with a variety of psychiatric diagnoses/symptoms, not just depression. In this regard, researchers also should more closely evaluate the impact of paternal diagnoses/symptoms. Last but not least, future work should attempt to recruit samples representative of the various ethnic and cultural groups that make up the U.S. population.

Finally, assessments conducted in the context of clinical services provided to children and families rely largely on maternal reports of child behavior problems. Maternal perceptions of child behavior are clearly important and should be addressed in the course of clinical interventions. However, multiple sources of information would likely yield a more accurate picture of child adjustment, especially when mothers report higher levels of depressive symptoms. Results of this study suggest that clinical evaluations must be sensitive to potential overreporting of externalizing problems for boys and internalizing difficulties for girls.