Support for this project comes from the National Institute of Mental Health (T32 MH18834) and Maternal and Child Health Bureau, LEAH grant (5T71MC08054). The author would like to thank Wendy Kliewer and Andres De Los Reyes for their feedback on earlier versions of this manuscript and Catherine Bradshaw for consultation on statistical analyses. The author would also like to thank the Project on Human Development in Chicago Neighborhoods for providing access to the data.
Correspondence concerning this article should be addressed to Kimberly Goodman, Johns Hopkins Bloomberg School of Public Health-Mental Health, 624 N. Broadway St., Suite 806, Baltimore, MD 21205. Electronic mail may be sent to email@example.com.
This study extends research examining the implications of parent–youth informant discrepancies on youth victimization. Latent class analysis (LCA) identified dyads distinguished by patterns of parent and youth report of victimization. Analyses examined how latent classes were related to adjustment (i.e., anxiety/depression, aggression, and delinquency) concurrently and at follow-up assessment (~2.5 years) in a socioeconomically and ethnically diverse sample. Participants were 485 youths (58.1% male; M age = 12.83 years, SD = 1.60) and their primary caregivers from the Project on Human Development in Chicago Neighborhoods. This study compared three classes of youths: (a) Parent > Youth (24.0%), (b) Youth > Parent (21.5%), and (c) Relative Agreement (54.5%). Findings did not support the hypothesis that groups reflecting parental underreporting of youth victimization experiences would show poor adjustment relative to all other classes longitudinally. Surprisingly, youths who self-reported lower levels of victimization than parents reported were at risk for maladjustment over time. This type of discrepant dyad may deserve more careful attention than previously considered in the literature.
Epidemiological literature highlights poor agreement between parents and youths on ratings of youth exposure to violence (ETV), as prevalence rates are consistently lower for parent report than for child self-report (e.g., Ceballo, Dahl, Aretakis, & Ramirez, 2001; Howard, Cross, Li, & Huang, 1999; Kuo, Mohler, Raudenbush, & Earls, 2000; Richters & Martinez, 1993). Risk and protective factors for youth ETV may be quite different depending on the informant used (Kuo et al., 2000; Offord et al., 1996). Furthermore, informant discrepancies can have implications for screening, treatment referral, and planning of mental health interventions for victimized youth (Guterman & Cameron, 1999; Guterman, Hahm, & Cameron, 2002). As a result, informant discrepancies regarding youth victimization require careful consideration for both epidemiological research and clinical interventions.
Although poor cross-informant agreement can pose interpretive problems for research, discrepant perspectives between parents and youths may also provide useful information and predictive utility for research on youth adjustment (e.g., Ferdinand, Van der Ende, & Verhulst, 2004; Guion, Mrug, & Windle, 2009; Pelton, Steele, Chance, & Forehand, 2001). Findings that prevalence rates are lower for parent report than for child self-report of violence exposure suggest that discrepancies may reflect a lack of parental knowledge or parental unawareness of youths' violence exposure (Ceballo et al., 2001; Howard et al., 1999; Richters & Martinez, 1993). Based on the supposition that parents who are unaware of youths' ETV are limited in their ability to help youths cope adaptively with violence (Richters & Martinez, 1993), some research has examined whether parent–youth agreement on violence exposure is related to youth adjustment (Ceballo et al., 2001; Howard et al., 1999; Zimmerman & Pogarsky, 2011). Consistent with this literature, researchers have recently proposed a theoretical framework to conceptually link informant discrepancies on reports of victimization to youth adjustment. The Discrepancies in Victimization Implicate Developmental Effects (DiVIDE) model suggests that when discrepancies between parents and youths on ratings of youth victimization reflect a lack of youth disclosure or parental unawareness of youth victimization, such discrepancies lead to impaired social support and coping resources that, in turn, negatively affect youth adjustment (Goodman, De Los Reyes, & Bradshaw, 2010). The current study provided a preliminary test of the DiVIDE model by investigating heterogeneity in discrepant reports of youth victimization and examining the ways in which discrepant reports of victimization were related to youth adjustment over time.
Although it is not possible to definitively tease apart underestimation on the part of one informant (e.g., parent) from overestimation on the part of another (e.g., child), research on adolescent social development provides theoretical support for the idea that discrepancies reflect parental unawareness of youth experiences (Goodman et al., 2010). Adolescence is a time marked by decreases in parental monitoring, as parents and youths spend less time together, and youths disclose less information about their whereabouts and behaviors (Collins & Laursen, 2004). Recent work highlights the importance of youth disclosure for parents' knowledge of youths' experiences (Kerr & Stattin, 2000; Smetana, Metzger, Gettman, & Campione-Barr, 2006). Further, informants themselves have noted nondisclosure of information as one key reason for discrepancies between adolescents and their parents on ratings of behavior (Bidaut-Russell et al., 1995).
Importantly, many adolescents report social constraints in talking with caregivers about violent events (Ozer & Weinstein, 2004). Constraints on disclosure may cause individuals to inhibit discussion of the event or suppress thoughts and thereby impair adaptive coping (Kliewer, Lepore, Oskin, & Johnson, 1998; Lepore, Silver, Wortman, & Wayment, 1996). In fact, some findings suggest that violence-exposed youths who feel constrained in talking about their experiences are more likely to experience internalizing symptoms (Kliewer et al., 1998; Ozer & Weinstein, 2004). Although this literature has considered both direct and indirect forms of violence exposure, disclosure of victimization experiences may be particularly meaningful because victimization is an especially strong personal affront and potentially isolating experience (Goodman et al., 2010; O'Donnell, Schwab-Stone, & Ruchkin, 2006).
Our understanding of parent–youth disagreement on victimization and its implications is limited for several reasons. First, several studies documenting parental underreporting (relative to youth self-report) of youth violence exposure focused exclusively on indirect exposure or witnessed violence (i.e., Hill & Jones, 1997; Kuo et al., 2000; Thomson, Roberts, Curran, Ryan, & Wright, 2002; Zimmerman & Pogarsky, 2011). Relatively few investigations also included direct exposure (victimization) in the analyses of parent–youth agreement (e.g., Ceballo et al., 2001; Howard et al., 1999; Richters & Martinez, 1993). Second, studies linking informant disagreement with youth maladjustment have relied on indices of agreement on the presence or absence of events without taking into account which informant was reporting higher levels of violence exposure (Ceballo et al., 2001; Howard et al., 1999). In one noteworthy exception, recent work indicates that some caregivers report higher levels of witnessed violence than children self-report, highlighting that not all instances of informant disagreement reflect parental underestimation (Zimmerman & Pogarsky, 2011). However, this work computed discrepancies based on summed indices of witnessed violence and did not provide information on patterns of discrepant reports for specific forms of violence exposure. Further, it is not clear whether parents' relative overreporting or underreporting tends to be consistent across items.
Goodman et al. (2010) suggested that future work examine whether there are underlying subgroups in the population with different patterns of reporting agreements. With a focus on victimization, the present study employed latent class analysis (LCA) to examine heterogeneity and patterns of discrepant reports. This technique considers that different subgroups of individuals may underlie the population such that variables are related to one another in different ways for different groups of people (Laursen & Hoff, 2006). Specifically, the first goal of the current study was to identify groups of dyads based on parent and youth ratings of victimization such that dyads were most similar within groups and dissimilar between groups.
Prior work suggests that both internalizing and externalizing outcomes are associated with parent–youth concordance or absolute agreement in ratings violence exposure. Howard et al. (1999) found that parent–youth disagreement on ratings of violence exposure was related to poor parent–child communication, low parental monitoring, symptoms of distress, low self-esteem, low problem-solving, and perpetration of violence. In another study, parent–youth agreement on victimization significantly added to the prediction of internalizing (but not externalizing) symptoms after controlling for demographic variables and parent report of youths' ETV (Ceballo et al., 2001). Because these studies were cross-sectional, however, it is not clear whether maladjustment is an outcome or merely an associative characteristic of informant agreement. Importantly, recent longitudinal work by Zimmerman and Pogarsky (2011) found that parental underestimation of witnessed violence was associated with future child psychological symptoms (both internalizing and externalizing problems), whereas parental overestimation was not. Building on this body of literature, the second goal of the present study was to examine the ways in which discrepant reports of victimization are related to adjustment, both concurrently and longitudinally. The present study is unique in that it examined changes in adjustment as an outcome of discrepancies.
The overall purpose of the present study was to extend the literature on parent–youth discrepancies on victimization and their links to child adjustment outcomes by employing a person-centered approach. This study focused on pre- and early adolescence, a developmental period marked by decreases in parental monitoring and youth disclosure. Specifically, LCA identified latent groups of dyads distinguished by patterns of parent and youth report of youth victimization. This study conceptualized age and gender as covariates of the latent classes, as these demographic factors are associative characteristics of parent–youth discrepancies on reports of violence exposure (Goodman et al., 2010). Study hypotheses were twofold. The working hypothesis was that latent classes would reflect patterns of parent–youth (dis)agreement on ratings of victimization. Specifically, at least two “disagreement” classes were expected, with one class in which parents report less victimization than youths' self-report and another class in which parents report higher levels of victimization than youths' self-report. The second hypothesis was that groups reflecting parental underreporting of youth victimization experiences (i.e., classes in which parents report less youth victimization than youths' self-report) would show increased anxiety/depression, increased aggression, and increased delinquency relative to all other classes.
This study examined data from two cohorts of youth from the Project on Human Development in Chicago Neighborhoods (PHDCN). A stratified probability sample of 80 Chicago neighborhoods was used in the PHDCN, sampled from 21 strata (seven racial or ethnic groups by three socioeconomic levels). Within PHDCN, a series of coordinated longitudinal studies followed youths and their primary caregivers to examine individual, family, and peer influences on adjustment (for additional description, see Earls & Buka, 1997; Sampson, Raudenbush, & Earls, 1997). Specifically, this study used victimization data from Wave 2, and adjustment data from Waves 2 and 3 of the PHDCN. Whereas Wave 1 assessed only witnessed violence, the second wave of data collection included both primary caregivers' and child informants' reports of victimization frequency in the past year.
Of interest was studying the extent to which disagreement on the presence of victimization related to indices of psychosocial maladjustment. A prerequisite for studying these relations was examining instances in which the construct of interest was present or expressed. This required at least one informant to perceive that victimization had occurred. Thus, the analyses that follow focus on those participant dyads for which at least one informant reported victimization. This decision to focus on dyads in which disagreement could occur is consistent with recent multi-informant work using LCA to reflect patterns of disagreement for other constructs such as children's social-emotional adjustment (e.g., De Los Reyes, Alfano, and Beidel, 2011; De Los Reyes, Youngstrom, et al., 2011).
This study included two cohorts of youths (cohorts ages 9 and 12 at Wave 1) who were approximately 10–15 years of age at the second wave of data collection. As 90% of primary caregivers were biological parents, primary caregivers are hereafter referred to as parents. Of the total number of cases for which both parent and youth provided data on youths' ETV (total N =1,339), over half (n =854) indicated that no victimization occurred in the past year according to parent and child report. These cases (64%) were excluded from the study analyses, resulting in a study sample of cases for whom at least one informant reported victimization in the past year (total N =485). Frequencies for demographic characteristics (i.e., youth ethnicity, neighborhood socioeconomic status [SES], caregiver education level, and youth sex) for the sample are reported in Table 1. The youth sample was 41.9% female and 58.1% male, with an average age of 12.83 years (SD = 1.60). The sample was socioeconomically and ethnically diverse. Neighborhood SES data were available for 79% of informants (n =383). As indicated in Table 1, 144 (29.7%) lived in neighborhoods with low SES, 141 (29.1%) lived in medium SES neighborhoods, and 98 (20.2%) lived in high SES neighborhoods.
Table 1. Frequencies for Demographic Information of Final Sample (N = 485)
Study sample restricted to cases for whom at least one informant reported one or more forms of victimization in the past year.
Ethnicity of youth
Caregiver education level
Below high school
Some high school
Finished high school
Some education beyond high school
Neighborhood socioeconomic status
Project on Human Development in Chicago Neighborhoods data were collected in three waves over a period of 7 years (1995–2001), with each wave of data collection separated by approximately 2.5 years. The primary method of data collection was face-to-face interviewing in participants' homes. Interviewers provided all respondents with a description of the study purposes and procedures, and participants were given the opportunity to discontinue the interview at any time. A Certificate of Confidentiality was obtained for this study, and issues of confidentiality were discussed as part of the consent and assent process.
Exposure to violence
My Exposure to Violence (My ETV; Selner-O'Hagan, Kindlon, Buka, Raudenbush, & Earls, 1998) examines a subject's lifetime and past-year exposure to 18 different violent events that have either been witnessed or personally experienced. Prior work investigating the psychometric properties of My ETV provides strong support for the validity of this measure in the current study sample (Brennan, Molnar, & Earls, 2007). The present study included six items that reflect interpersonal violence: being chased (“chased, but not caught, when you thought that you could really get hurt?”), hit (“hit, slapped, punched, or beaten up?”), attacked with a weapon (“attacked with a weapon?”), being shot at (“shot at”?), sexual assault (“sexually assaulted, molested, or raped?”), and threatened (“someone threatened to seriously hurt you?”). Frequency of exposure during the past year was measured on a 6-point scale (never, once, two or three times, 4–10 times, 11–50 times, more than 50 times). All victimization items were recoded from their original 6-point scale to dichotomous indicators (never, one or more times) to reduce sparseness in the data for LCA. A parallel parent report measure, My Child's Exposure to Violence, was administered to parents.
At Waves 2 and 3, youths and parents completed parallel measures of psychological symptoms. This study included anxiety/depression, aggression, and delinquency subscales of the Child Behavior Checklist (CBCL; Achenbach, 1991a) and Youth Self Report (YSR; Achenbach, 1991b). Based on a 3-point scale, 0 (not true), 1 (somewhat true), and 2 (very true), respondents reported how true each item (behavior) was during the past 6 months.
This study used child report of child demographics (i.e., ethnic status and gender) and parent report of parental education. Neighborhood SES was previously computed in the PHDCN by summing the following standardized neighborhood-level measures: median income, percentage college educated, percentage of households with income >$50,000, percentage of families living below the poverty line, percentage of families on public assistance, and percentage of households with income <$50,000 (Fauth, Roth, & Brooks-Gunn, 2007; Sampson et al., 1997).
Latent class analysis is a person-centered analytic approach used when the construct of interest (in this study, parent–youth reporting agreement) is made up of qualitatively different groups, but the group membership must be inferred from the data, because it is unobserved (Lanza, Flaherty, & Collins, 2003; McCutcheon, 1987). As described by Goodman et al. (2010), a person-centered approach such as LCA can elucidate patterns of agreement in the population such that parent–child dyads vary in whether parents over- or underestimate youths' violence exposure in some domains and not others. Whereas variable-centered analytic approaches assume that the population is homogeneous with respect to how predictors operate on the outcomes, person-centered approaches assume that variables are related to one another in different ways for different groups of people (Laursen & Hoff, 2006). LCA separates persons (here parent–youth dyads) into mutually exclusive groupings such that, across the indicators, groups are maximally similar to each other and thus maximally dissimilar to members assigned to other groups.
In the present study, parent and youth reports of the past-year occurrence of six victimization events (12 victimization events, total) were used as the observed indicators in LCA. Analyses were conducted using the Mplus 6.0 statistical package (Muthén & Muthén, 1998), with model selection based on fit indices and substantive interpretation. There is no one single index for selecting the best latent class model, as several fit statistics are available and do not always agree as to which model is the one with the optimal number of classes (Nylund, Asparouhov, & Muthen, 2007). Tests that balance model–data misfit and parsimony (number of model parameters estimated), such as Akaike's information criterion (AIC) and Bayes' information criterion (BIC) differ according to the weight attributed to parsimony. BIC tends to underestimate the number of classes, while AIC tends to select a model with too many classes (Magidson & Vermunt, 2004). Models with the lowest AIC, BIC, and sample-size adjusted BIC (SSA BIC) values, or the point at which these indices begin to level off (as shown in a scree plot), suggest the best fit (Sclove, 1987). The Lo–Mendell–Rubin (LMR) likelihood ratio test compares the estimated model to a model with one less class (e.g., between a four-class and a three-class model), and a nonsignificant p value suggests that the additional class does not result in a significant improvement in fit. Finally, Nylund, Bellmore, et al. (2007) also emphasized the importance of substantive interpretation to guide selection of the LCA model. That is, a model is only useful to the extent that a researcher can interpret the results in terms of substantative theory.
Preliminary Descriptive Analyses
Overall, youths endorsed higher rates of victimization than parents, consistent with prior literature. Youths were most likely to report being hit (43.1%), followed by chased (26.8%), threatened (17.5%), attacked (7%), shot at (4.1%), and sexually assaulted (1.4%). Parents were most likely to report that their child was hit (36.9%), followed by chased (20%), threatened (8.9%), attacked (3.1%), sexually assaulted (0.8%), and shot at (0.6%).
Of the 417 cases with parent adjustment (CBCL) data at Wave 2, 80 cases (19.2%) were missing CBCL data at Wave 3. Of the 423 cases with youth adjustment (YSR) data at Wave 2, 128 cases (30.3%) were missing data at Wave 3. Participants who did and did not participate in both waves of the study were compared with all demographic variables using t tests and chi-square difference tests. Attrition was unrelated to neighborhood SES, caregiver education level, child age, sex, and ethnicity. Further, attrition was unrelated to all adjustment indices at baseline.
Selection of Latent Class Model
Latent class models were fit for one through five classes. The LCA model began with a model that included parent and youth report of victimization (12 indicators) and no covariates. Fit indices for this model are displayed in Table 2. Whereas the BIC values increased beyond three classes, the SSA BIC values continued to decrease with the inclusion of additional classes. However, the SSA BIC values began to level off (i.e., did not show appreciable decreases) after the three-class solution. As indicated in the table, the LMR likelihood ratio test indicated that the three-class solution was the best fit. Specifically, the p value indicated that a three-class solution fit the data better than a two-class solution, whereas a four-class solution did not fit the data better than a three-class solution, and a five-class solution did not fit the data better than a four-class solution. Thus, the three-class solution was selected as the best fitting model. Next, age and gender were added to the model as covariates. The LMR likelihood ratio test once again indicated that the best fitting model was a three-class solution.
Table 2. Latent Class Analysis Fit Indices
Number of classes
LMR LRT p value
Adjusted LMR LRT p value
Fit indices for latent class model with no covariates. Bayesian information criterion (BIC; Schwartz, 1978);); Sample-size-adjusted Bayesian information criterion (SSA BIC; Sclove, 1987); Akaike information criterion (AIC), Lo–Mendell–Rubin adjusted likelihood ratio test (LMR LRT; Lo, Mendell, & Rubin, 2001); and the sample-size-adjusted LMR. Bolded model indicates best fitting model according to LMR LRT.
Class proportions were very similar for the LCA model with covariates and the model without covariates. The three-class model without covariates yielded the following three classes: (a) a class in which parents report higher levels of victimization than youths (Parent > Youth, 23.8%), (b) a class in which youths report higher levels of victimization than parents (Youth > Parent, 22.8%), and (c) a class in which parents and youths show relative agreement in victimization (Relative Agreement, 53.4%). Similarly, the three-class solution including youth age and gender as covariates yielded the following three classes of parent–youth dyads: (a) Parent > Youth, 24%; (b) Youth > Parent, 21.5%, and (c) Relative Agreement, 54.5%. Thus, inclusion of age and gender as covariates did not qualitatively change the interpretation of the groups.
Characteristics of Victimization Classes
Table 3 displays conditional probabilities, or the probability of a response given that the respondent is in a particular latent class. Figure 1 displays the conditional probabilities, based on the data in Table 3. Discrepant classes were distinguished by informants' discrepant reports about whether the child was hit. Within the Parent > Youth class, 100% of parents reported that their child was hit in the past year, whereas only 17% of youths reported this form of victimization. Probabilities for all other forms of victimization were relatively low within this class. Conversely, the Youth > Parent class was characterized by 100% of youths reporting that he or she was hit in the past year, whereas only 4% of parents reported this form of victimization. Probabilities for all other forms of victimization were low within this class. Finally, the Relative Agreement class, comprising 54.5% of the sample, was characterized by the absence of salient discrepant reports in probabilities of reporting victimization. Whereas the discrepant classes were distinguished by reports that the child was hit, both informants reported more diverse forms of victimization within the Relative Agreement class. However, it is important to note that youths were more likely than parents to report victimization within this class. Youths were most likely to endorse being chased (45%), followed by hit (32%), threatened (29%), attacked (12%), shot at (7%), and sexually assaulted (2%). Parents in this class reported lower incidence of youth victimization, with probabilities of parent-reported victimization as follows: chased (30%), hit (22%), threatened (11%), attacked (4%), shot at (1%), and sexually assaulted (1%).
Table 3. Conditional Probabilities for Reports of Youth Victimization
Class 1: Parent > Youth (24%)
Class 2: Youth > Parent (21.5%)
Class 3: Relative Agreement (54.5%)
Conditional probabilities for victimization items.
Class membership was regressed on the covariates gender and age. Using the Relative Agreement class as the normative comparison group, two comparisons were made for each covariate: (a) the likelihood of being in Youth > Parent class compared with the Relative Agreement class, and (b) the likelihood of being in the Parent > Youth class compared with the Relative Agreement class. The gender logistic regression coefficients for the Youth > Parent class (.97, p =.01) and Parent > Youth class (.88, p <.01) indicated that females were more likely to be in the discrepant classes, in comparison with the Relative Agreement class. The age logistic regression coefficients for the Youth > Parent class (−.26, p <.05) and Parent > Youth class (−.27, p <.01) indicated that discrepant classes were comprised of youths who were younger, in comparison with the Relative Agreement class.
Analyses also explored whether there were class differences for other demographic variables (i.e., ethnic group status and neighborhood SES). These variables were not conceptualized as covariates of the latent classes, but rather explored as potential associative characteristics of the classes. The auxiliary function in Mplus (Muthén & Muthén, 1998) was used to investigate class differences for ethnic group status and neighborhood SES. It is important to note that an auxiliary variable is not included in the LCA model and does not influence the composition of latent classes (Marsh, Ludtke, Trautwein, & Morin, 2009; Muthén & Muthén, 1998). Thus, unlike covariates, auxiliary variables are not treated as antecedents of the latent classes. Some ethnic differences emerged. Youths in the Parent > Youth class were more likely to be Black than youths in the Youth > Parent class (χ2 = 4.24, p <.05). Youths in the Parent > Youth were also less likely to be Hispanic than youths in the Youth > Parent class (χ2= 15.28, p <.001) and Relative Agreement class (χ2= 6.69, p =.01). However, class status was unrelated to neighborhood SES (χ2= 0.36, p =.83).
Relationship Between Latent Class Membership and Adjustment
The auxiliary function in Mplus was used to relate distal outcomes (i.e., adjustment variables) to the latent classes. The auxiliary function provides a test of statistical significance for comparing class means on each outcome variable. The approach uses pseudo-class draws that take into account the probabilities that a case will fall into a particular group, thereby reducing error that occurs through modal assignment or assigning each case to only one class (Marsh et al., 2009; Muthén & Muthén, 1998). Table 4 displays means and standard deviations for concurrent adjustment scales (anxiety/depression, aggression, delinquency) for each reporter. As indicated in Table 4, the Youth > Parent class did show higher levels of maladjustment relative to the No/Low Victimization and Parent > Youth class on all three youth-reported (YSR) indices at Wave 2. However, these findings were not consistent for parent-reported adjustment indices.
Table 4. Relationship Between Concurrent Adjustment and Class Membership
(1) Parent > Youth
(2) Youth > Parent
(3) Relative Agreement
Values presented are means with standard errors in parentheses.
Parent report of adjustment at Wave 2 (Child Behavior Checklist)
1 > 2, 1 > 3
Youth report of adjustment at Wave 2 (Youth Self-Report)
2 > 1, 3 > 1
2 > 1, 3 > 1
2 > 1, 3 > 1
The results examining whether parent–youth discrepancies predicted youth adjustment longitudinally were not consistent with hypotheses. Table 5 displays class differences for changes in adjustment (adjustment at wave 3 − adjustment at wave 2). Positive values indicate increases in symptoms over 2.5 years, whereas negative values indicate decreases in symptoms. The Parent > Youth class showed increased symptoms for depression and anxiety over 2.5 years, relative to the Youth > Parent class (χ2= 5.10, p <.05) and Relative Agreement class (χ2= 4.40, p <.05). Further, class comparisons revealed similar trends for youth-reported delinquency and aggression. The Parent > Youth class showed increased symptoms for aggression over 2.5 years, relative to the Youth > Parent class (χ2= 3.30, p =.07) and Relative Agreement class (χ2= 3.70, p =.06). Similarly, the Parent > Youth class showed a trend for increased symptoms of delinquency over 2.5 years, in comparison with the Relative Agreement class (χ2= 3.69, p =.06).
Table 5. Relationship Between Changes in Adjustment and Class Membership
(1) Parent > Youth
(2) Youth > Parent
(3) Relative Agreement
Values presented are means with standard errors in parentheses. Change scores were computed by subtracting adjustment at Wave 2 from adjustment at Wave 3.
Marginally significant difference (p <.07).
Changes in parent report of adjustment (Child Behavior Checklist) over 2.5 years
Changes in youth report of adjustment (Youth Self-Report) over 2.5 years
1 > 3
1 > 3
1 > 2, 1 > 3
This study extends the literature investigating parent–youth informant discrepancies on reports of victimization and the implications of discrepant perspectives for youth adjustment. Prior literature highlights discrepant reports of violence exposure between parents and youths, with preliminary evidence suggesting that discrepancies are linked with psychological maladjustment in youth. However, as mentioned previously, much of this work has focused exclusively on witnessed violence, whereas relatively little work has focused on victimization. Further, this work has not examined patterns of discrepant perspectives or prevalence of these different patterns in the population. Finally, prior studies examining victimization were cross-sectional and did not shed light on whether poor adjustment is in fact an outcome or merely an associative characteristic of discrepancies on ratings of youth victimization.
Findings were consistent with prior work indicating that parents report lower levels of victimization than youths based on overall prevalence rates. As prior work has overlooked instances in which parents report more victimization than their children self-report, the present study addressed this gap by employing a person-centered analytic approach to classify dyads according to ratings of parent and youth report. The data supported the hypothesis that at least two “disagreement” classes would emerge in the population, with each class reflecting a different direction of discrepant reports. Specifically, over one fifth of dyads were typified by youths reporting higher levels of victimization than parents, and nearly one fourth of dyads were typified by parents reporting higher levels of victimization than youths. Both of these discrepancy groups were characterized by discrepancies in reports that the child was hit in the past year. Over one half of dyads, referred to as the Relative Agreement class, were characterized by less discrepancy in the likelihood of parent and youth report of victimization.
It is important to note that the Relative Agreement class was labeled as such because this class was not typified by large discrepancies in parent and youth report of victimization as were the other two classes. However, within this class, the probability of youth-reported victimization was still higher than the probability of parent-reported victimization. Moreover, the forms of victimization reported were more diverse than the two discrepant classes. For this reason, the Relative Agreement class was an especially stringent comparison group in this study for detecting class differences. Further, unlike prior work that included dyads agreeing on the absence of victimization, this study excluded cases in which both informants reported no victimization. This study was also unique in that the latent classes reflected type of victimization in addition to patterns of discrepancy.
As mentioned previously, an important theoretical foundation for the present study is the supposition that parents' underreporting of youth victimization (relative to youth report) reflects impaired communication or nondisclosure, parental unawareness of victimization, and a potential lack of coping resources for victimized youths (Goodman et al., 2010). Thus, a key hypothesis of the present study was that youths who self-report greater victimization than parents would show increased anxiety/depression, increased aggression, and increased delinquency relative to all other classes. Concurrent associations provided some support for this supposition, whereas longitudinal findings examining changes in adjustment after 2.5 years did not support this hypothesis.
Contrary to hypotheses, youths who reported less victimization than parents (i.e., Parent > Youth class) were at risk for increases in youth-reported anxious or depressed symptoms over 2.5 years relative to youths classified as Relative Agreement and Youth > Parent. Youths in the Parent > Youth class also showed a trend for increased symptoms of aggression and delinquency in comparison to the Relative Agreement group. One might surmise that youths' relative underreporting of victimization reflects coping efforts such as repressing or denying that victimization has occurred (Goodman et al., 2010). In fact, some literature suggests that disengagement coping—a construct that includes denial and avoidance—is related to youth psychopathology (Compas, Connor, Saltzman, Thomsen, & Wadsworth, 2001). Relatedly, youths' underreporting relative to parent report may also reflect shame, as this emotion is associated with attempts to deny, hide, or escape the shame-inducing situation. Interestingly, shame proneness is an affective disposition linked to a host of maladjustment outcomes (Tangney, Wagner, & Gramzow, 1992). Youths' distrust in the research process may also result in a reluctance to disclose information. Finally, it is important to note that any processes contributing to youths' relative underreporting of victimization also likely generalize to underreporting of other psychological domains (e.g., depression, delinquency, and aggression).
Implications for Research, Policy, and Practice
Intervening to prevent the psychological sequelae of victimization may be challenging if some youths underreport victimization experiences in the context of interviews or self-report screening measures. One important challenge for future work involves determining the best methods for screening youths for ETV as part of prevention efforts conducted in school or community settings. Based on group-level prevalence rates revealing that youths report higher levels of violence exposure than parents, several researchers have suggested that youth report is at least as valuable as—if not superior to—parent report (Buka, Stichick, Birdthistle, & Earls, 2001; Ceballo et al., 2001; Richters & Martinez, 1993; Thomson et al., 2002). The findings from this study suggest that parent report of victimization might add valuable predictive utility beyond youth report. Although one cannot make inferences regarding the validity of either informant, this study does indicate that sole reliance on any one informant may result in missed information. Future work might also examine the implications of parent–youth discrepancies on victimization for use of mental health services, as parents and children frequently fail to agree on problems that warrant mental health treatment in clinics (Yeh & Weisz, 2001). Studies might further explore formal service use and informal sources of support that exist for different types of discrepant dyads.
Limitations and Future Directions
One salient limitation of the current study is the lack of attention to context in which victimization occurs. Examining context is an exciting direction for future work to understand why parents and youths differ in their reports of child victimization. Several researchers posit that informant discrepancies reflect differences in the settings in which behavior is observed by different informants (e.g., De Los Reyes & Kazdin, 2005; Kraemer et al., 2003). Although recent empirical work supports the theory that informant discrepancies provide information about the contexts in which behaviors occur (De Los Reyes, Henry, Tolan, & Wakschlag, 2009), the connection between context and informant-specific reports has rarely been considered in the study of ETV. In one noteworthy exception, some literature suggests that parents are more likely to report youths' ETV that took place in the home, whereas youths are more likely to report exposure that took place in school (Thomson et al., 2002). However, this work was limited to witnessed violence, rather than victimization.
It is important to note that there may be considerable variability and some interpretive ambiguity associated with the item “hit, slapped, punched, or beaten up.” For the age group under investigation, getting “hit” or “slapped” may include corporal punishment by family members (Straus & Stewart, 1999). It is therefore possible that the Parent > Youth class may include parents who report this form of victimization because they believe the item content includes corporal punishment, whereas the youths do not have this same interpretation. This possibility also underscores the importance of attending to context. Because “community” is a heterogeneous term that can include a number of settings (home, school, neighborhood) as well as a number of perpetrators (family, friends, strangers), context is frequently overlooked or inconsistently defined in the literature on youth exposure to community violence (Guterman, Cameron, & Staller, 2000). Future work that examines discrepancies within specific contexts may ultimately shed light on why discrepancies occur, and under what circumstances discrepancies are risk factors for dysfunction.
Past-year incidence was considered to be an optimal timeframe for collecting informant reports in order to maximize the likelihood of accurate recall, especially because lifetime prevalence reports may be inaccurate when children are recalling stressful events that occurred at a very young age (Howe, Toth, & Cicchetti, 2006). However, future research might also consider discrepancies based on lifetime incidence. It is also possible that parent–youth discrepancies on victimization are most predictive of maladjustment when these discrepancies are stable over the course of early adolescence (Goodman et al., 2010). This study did not consider stability of discrepant perspectives, as classes reflected discrepancies in one limited time period (i.e., past year). Future research might apply latent transition analysis (LTA) to identify dyads that show stability or change in class status over time (Lanza et al., 2003).
Finally, this study was based on an assumption that discrepant perceptions are maladaptive, and the theoretical framework therefore emphasizes deficits (e.g., lack of coping resources, lack of parental knowledge, and lack of child disclosure) that may explain why disagreement is a “risk factor.” Future work based on a resilience framework—focusing on positive adaptation and development despite adversity—might also be fruitful (Luthar, Cicchetti, & Becker, 2000). What is the role of shared perspectives (parent–youth agreement) and caregiver support for victimized youths who demonstrate resilience? Further, some measurement of felt acceptance and social support from parents might help us to understand processes through which parent–youth agreement is adaptive (Goodman et al., 2010).
Despite its limitations, this study advances the literature on several fronts. Researchers must frequently reconcile or use conflicting information from different informants. This study applied a person-centered approach to integrate information from parent and youth informants on ratings of youth victimization. Latent class analysis revealed considerable heterogeneity in the population with regard to patterns of parent and youth report of youth victimization. This study added to a growing body of literature that conceptualizes informant discrepancies as useful information and as a risk factor for poor adjustment. The findings underscore the importance of attending to direction of discrepancy (i.e., which informant reports higher levels of victimization) when examining how informant disagreement is related to youth adjustment. Surprisingly, findings suggested that youths who self-reported lower levels of victimization than parents reported may be at risk for poor adjustment. This type of discrepant dyad may deserve more careful attention than previously considered in the literature. Findings suggest several important questions and directions for future research that seeks to understand informant discrepancies as a risk factor for youth maladjustment.