Uniform Differential Item Functioning Across Gender, Grade Level and Racial Groups: A MIMIC Investigation of the Non-clinical Parent Ratings of the Pediatric Symptom Checklist-17

It is estimated that approximately 15% of children under the age of six have clinically significant mental health problems (Kann, 2016). This is concerning as early social, emotional and behavioral (SEB) problems are associated with later SEB development (Treyvaud et al., 2012; Waller et al., 2017; Wang et al., 2022), academic performance (Hammer et al., 2017; Polderman et al., 2010), school readiness (Gartstein et al., 2016), school engagement (Olivier et al., 2020), peer relationship (Kochel et al., 2012; Waller et al., 2017), school readiness (Denham & Brown, 2010; Shala, 2013), criminal activity and drug abuse (Halle & Darling-Churchill, 2016; Jones et al., 2015). Given the potential negative consequences of mental health problems, it is important to identify and treat these difficulties early in children’s lives.

Schools play an important role in identifying children with SEB problems (Bringewatt & Gershoff, 2010). Some schools have increased their early identification and screening procedures to identify at-risk students (Lane et al., 2012). Some school-based services, such as the multi-tiered system of support (MTSS), function well in preventing the occurrence of potential problems and reducing the risk level of targeted students (Fazel et al., 2014; Olubiyi et al., 2019; Sanchez et al., 2018). When early identification is combined with early prevention and intervention services, it can reduce the possibilities of students’ academic failures and future life difficulties (e.g., Lane & Menzies, 2003; Walker & Shinn, 2002) and minimize the impact of risk factors, as well as prevent further development of behavioral and social-emotional difficulties (Glover & Albers, 2007).

School-based universal mental health screening assessments have emerged as an efficient method to identify children at risk for SEB problems and may benefit from relevant interventions (Hunter et al., 2005; Kettler et al., 2014; Severson & Walker, 2002). However, roughly only 12% of schools use universal screening instruments to assess students’ SEB problems (Bruhn et al., 2014). The limited number of schools using universal screening instruments might be due to the short history of universal screening for young children’s SEB problems (Greenwood et al., 2011; Steed & Banerjee, 2016).

While screening young children for SEB problems is not yet widespread, it has been advocated due to the high rate of mental problems among youth (Essex et al., 2009) and the benefits provided to both schools and students. First, all students, regardless of whether they have risky behaviors or not, have the equal opportunity to be screened and receive possible support, thus reducing the likelihood of missing students who need intervention (Lane et al., 2012). Second, screening is recommended as an initial step for identifying risky behaviors, prevention, and intervention services in schools to reduce the negative effects of risky behaviors on students’ long-term outcomes (Walker et al., 2009; Webster-Stratton et al., 2011). Third, screening can help schools make better use of limited school resources such as counseling, prosocial skills training, and academic support (Walker et al., 2005). As many decisions may be made using a screening instrument, identifying a valid and reliable tool for effective universal screening in school is essential.

The utility and precision of screening assessments are based on their appropriateness for intended use, technical adequacy, and usability (Glover et al., 2007). Dowdy et al. (2010) recommended the following instruments for universal school-based mental health screening: Strengths and Difficulties Questionnaire (SDQ; Goodman, 1997); Pediatric Symptom Checklist (PSC; Gardner et al., 1999); Systematic Screening for Behavior Disorders (SSBD; Walker & Severson, 1992),  Behavior Assessment System for Children-Second Edition (BASC-2, Kamphaus & Reynolds, 2007), and Social, Academic, and Emotional Behavioral Risk Screener (SAEBRS, Kilgus et a., 2013). Among these forms, our focus is the Pediatric Symptom Checklist-17 (PSC-17). The screener is the shortened version of the full (35-item) PSC form and was originally a caregiver-completed questionnaire for children aged 4–16 years (Gardner et al., 1999). Compared to other screening tools, which are lengthy or expensive, PSC-17 has the advantages of free access for the public and a brief design with just 17 items, saving schools both time and money. The online availability (http://​www.​massgeneral.​org/​psychiatry/​assets/​PSC-17_​English.​pdf) makes it easier for school personnel to administer the assessment, including collecting, managing, and interpreting the assessment data. Further, the PSC-17 scores have demonstrated a strong correlation with other screening instruments such as Attention, Behavior, Language, and Emotion (ABLE; Barbarin, 2007), the BESS, and the SDQ (DiStefano et al., 2017), indicating sound validity evidence of the PSC-17. Therefore, the PSC-17 should be an acceptable option for schools to conduct universal screening of behavioral and emotional problems of students.

The PSC-17 has been validated and successfully used in clinical environments (Blucker et al., 2014; Borowsky et al., 2003; Chaffin et al., 2017; Gardner et al., 2007; Murphy et al., 2016; Stoppelbein et al., 2012). Three subscales underlying the PSC-17, including Externalizing Problems, Internalizing Problems, and Attention Problems, were identified with exploratory factor analysis (EFA) by Gardner et al. (1999) and confirmed with confirmatory factor analysis (CFA) (Chaffin et al., 2017; Murphy et al., 2016; Stoppelbein et al., 2012). Partial metric invariance was established across majority and minority groups (Stoppelbein et al., 2012). Bergmann et al. (2018) identified scalar invariance across gender. One study on the Externalizing Problems subscale of the PSC-17 identified one item (i.e., Item 4" refuses to share") with uniform differential item functioning (DIF) across genders and two items (i.e., Item10 " blames others" and Item14" teases others") demonstrating DIF across races (Studts et al., 2017). However, all these studies were conducted with parent ratings on the PSC-17 in primary care settings.

Recently, the PSC-17 has been used in school settings. DiStefano et al. (2017) identified a second-order factor structure of PSC-17 with two cross-loading items when used by teachers to rate preschool children. The first-order factors include Externalizing Problems, Internalizing Problems, and Attention Problems, and the second-order factor is Maladaptive Behavior. With a sample of teacher ratings of lower-grade elementary school children, Liu et al. (2020b) identified a 3-factor structure with one item loaded on a different latent factor. The study by Liu et al. (2020a) identified the high internal consistency and strong criterion-related validity of the PSC-17, indicating that this instrument is a high-quality universal screening tool for use in school settings. Gender invariance of teacher ratings of PSC-17 was established among preschool children. Boys exhibited more maladaptive behaviors than girls (Liu et al. (2018). Invariance of the PSC-17 was established across gender, race, and grade levels of children from the first grade to the second grade. Boys demonstrated higher levels of externalizing problems and attention problems than girls. Children from different racial groups or grade levels did not differ in their SEB functioning (Liu et al., 2020b). Measurement invariance of the PSC-17 was also established across teacher and parent respondents. Teachers perceived children as demonstrating a lower risk level of externalizing problems relative to parents. Parents and teachers did not differ in their perceptions of children's internalizing and attention problems (Gao et al., 2022). However, these studies, except the study by the (Gao et al., 2022), used only teacher-rated samples. To our knowledge, no study has examined the measurement equivalence of the PSC-17 rated by parents in non-clinical settings. To contribute to the research in measurement invariance of the non-clinical parent ratings of the PSC-17, we focused on testing measurement invariance in the PSC-17 across gender, grade level, and race/ethnicity.

Previous investigations of the PSC-17 in school settings have used teacher ratings of children’s SEB problems. However, as children behave differently in different settings, such as in school and at home (American Psychiatric Association, 2013), a multi-informant approach is important for assessing children’s behavioral problems (Wong et al., 2020). Besides, different informants may have unique perspectives on a child’s behavior (Achenbach, 2006; De Los Reyes & Kazdin, 2005). Teachers are more likely to take a normative approach in judging a child’s behavior, while parents may be a better judge of a child’s behavior from a more critical perspective (Konold et al., 2004). As only low to moderate correlations between parent and teacher ratings of children’s mental health have been identified, collecting multiple informants’ ratings will provide rich information about children’s behavioral problems (De Los Reyes et al., 2015). Therefore, the examination of non-clinical parent ratings on the PSC-17 is warranted.

The current study the aimed to investigate the factor structure of the PSC-17 and the uniform differential item functioning (DIF) of the PSC-17 item scores as a function of school children’s gender, grade level, and racial/ethnic groups. Specifically, the present study was designed to address the following research questions:

The Pediatric Symptom Checklist-17 (PSC-17) is the shortened form of the full PSC instrument (Gardner et al., 1999) designed to assess a child’s overall psychosocial functioning. The PSC-17 consists of 17 items that assess three sub-dimensions of maladaptive behavior: Internalizing Problems (e.g., “Feels sad, unhappy”), Externalizing Problems (e.g., “Fights with other children”), and Attention Problems (e.g., “Has trouble concentrating”). Respondents rate each symptom on a scale of 0 (never), 1(sometimes), and 2 (often). The 17 items are summed to produce a total score ranging from 0 to 34, with a higher score reflecting greater risk. A total score above 15 indicates an overall psychosocial health risk. The Internalizing Problems subscale consists of five items assessing children’s internalizing issues such as depression and anxiety. A score above five indicates risk on this subscale. Externalizing Problems subscale measures children’s disruptive behavior, such as aggression and hyperactivity. The Attention Problems subscale assesses whether children have trouble concentrating. Both Externalizing Problems and Attention Problems subscales consist of seven items, with sum scores greater than seven indicating risks.

A cluster sampling method was used in this study. Data were collected from children in nine public schools and child development centers in two states (California and South Carolina) involved in a funded grant project investigating universal screening. Parents provided consent for participation in the project; however, the method of consent varied by site. In California, parents were asked to give informed consent, while South Carolina sites used passive consent procedures. Both sites sent hard copies of forms home with children with their classroom work.

The children were nested within 91 classrooms. Hard copies of the PSC-17 forms were sent home for parents to complete. Forms were distributed in Spanish or English, depending on the home language. Parents were not compensated for participation; however, small incentives (e.g., pencils and stickers) were given to children. Families were not given individual feedback but were encouraged to contact the project investigators with any questions or for individualized information. We had bilingual project staff to assist Hispanic families with questions and concerns.

PSC-17 parent ratings were combined during three academic years (2016–17, 2017–18, 2018–19). The sample consists of 1,305 ratings of children aged from 3 to 6. The sample size is adequate based on the recommended sample size range from 300 to 460 cases, considering the number of indicators and factors, the magnitude of factor loadings, path coefficients, and the amount of missing data (Wolf et al., 2013). Institutional Review Board approval and informed consent were obtained before the data collection, and ethical treatment of subjects was followed during data collection and analysis procedures.

Female children (n = 591, 47.1%) and male children (n = 664, 52.9%) were evenly distributed in the sample. The sample of children rated by parents was predominantly Hispanic (53.6%), White (30.6%), African American (12.0%), and other racial groups, including Asian American, Pacific Islander/Native Hawaiian, American Indian/ Alaska natives, and multi-racial backgrounds (3.8%). The sample included children from different grade levels: Pre-kindergarten (50.1%), 5-year-old Kindergarten (40.4%), and Grade 1(9.5%); Grade level was coded as dichotomous. Approximately 50.1% of sampled children (n = 618) were from Pre-kindergarten, and 49.9% of sampled children (n = 615) were from K to grade 1.

All analyses were conducted using Mplus 8.4 software (Muthén & Muthén, 1998-2015). The weighted least squares with mean and variance adjusted (WLSMV) estimation method was chosen to accommodate the categorical nature of the PSC-17 data and address non-normality in the data (Finney & DiStefano, 2013). In Mplus, the method serves as the default when dealing with categorical data (Muthén & Muthén, 1998-2015. Missing data ranged from 0.3% to 2.9% across the 17 items. Due to the low proportion of missing data, pairwise deletion was utilized, meaning cases are excluded only if they have missing data on variables involved in data analysis. This approach is recommended for categorical data analysis as it maximizes case inclusion (DiStefano et al., 2017). Additionally, the nesting structure of the data (i.e., students nested in classrooms) was considered using a design effect to provide more accurate standard errors for parameter estimates. (Stapleton, 2013).

As the 3-factor structure of the PSC-17 used in the school setting was previously identified, we did not perform exploratory factor analysis. Confirmatory Factor Analysis (CFA) was conducted to test whether the three-factor solution is appropriate in the current study compared to previously established CFA models for teacher ratings of the PSC-17 in school settings (DiStefano et al., 2017; Liu et al., 2020b; Gao et al., 2022) and the parent-rated PSC-17 in clinical settings (Chaffin et al., 2017; Murphy et al., 2016; Stoppelbein, et al., 2012). The proposed model comprises three intercorrelated sub-scales: Internalizing Problems, Attention Problems, and Externalizing Problems.

To identify uniform differential item functioning (DIF) items, a Multiple Indicators Multiple Causes (MIMIC) model, defined as CFA models with covariates (Brown, 2015), was used. This model comprises a measurement model defining the relationship between indicators and latent variables (established at the CFA stage) and a structural model specifying the direct effects of covariates on item responses and latent factors (Jöreskog & Sörbom, 1996). MIMIC modeling assesses measurement invariance by allowing direct paths from grouping variables to observed variables (Kim et al., 2012). Uniform DIF occurs when the focal group consistently performs differently reference group after controlling for the level of the scale score (Scott et al., 2009). Potential uniform DIF items in the parent-rated PSC-17 were identified using children’s gender (0 = male, 1 = female), grade level(3 K to 4 K = 0, 5 K-1st grade = 1), and African American race/ethnicity(0 = no, 1 = yes), Hispanic(0 = no, 1 = yes), other racial/ethnic groups (0 = no, 1 = yes) as covariates. Male children, children from 3 to 4 K, and children from the White group served as reference categories. Item responses are regressed onto grouping variables to determine whether members of different groups vary in the possibility of endorsing any item response option after controlling their level on latent variables (Finch, 2005). MIMIC models were utilized to detect latent trait group differences by regressing the latent trait onto covariates while assuming that the hypothesized structure is invariant across groups (Green & Thompson, 2012). If the invariance measurement test identified items exhibiting uniform DIF, MIMIC modeling was used to examine whether the latent mean comparison across groups might be biased.

Following the approach used by Kim et al. (2012), a baseline MIMIC model was first constructed. In this baseline model, all latent variables identified in the CFA analysis were simultaneously regressed on all the covariates without any PSC-17 items included in the model (i.e., the direct effects of covariates on all items’ difficulty were constrained to zero). Next, the baseline models were compared with the relaxed MIMIC models where one direct effect of a covariate on a PSC-17 item was added sequentially, with all the covariates' direct effect on each latent variable (i.e., the direct effect of one covariate on one item’s difficulty was freely estimated). To compare these nested models, the WLSMV model chi-square difference test was conducted using the Mplus Difftest feature. A significant change in the chi-square difference with one degree of freedom between the baseline model and the less constrained model would indicate uniform DIF for the given item.

As high Type I error rates were reported when MIMIC modeling was used to identify nonvariant variables (Kim et al., 2012), the Oort adjustment to the Chi-square difference test was used to control Type I error inflation (i.e., false identification of uniform DIF for invariant items in the model comparison; Kim et al., 2012; Oort, 1998). Oort’s formula stated as \({K}{\prime}=[{x}_{0}^{2}/(K+{df}_{0}-1)]*\text{K}\), adjusts the critical chi-square value to account for the potential model misspecification in the baseline MIMIC model when analyzing categorical items. In the formula, K´ refers to the adjusted critical value for the chi-square difference test, and K is the critical value for the chi-square difference test (e.g., the critical value is 3.84 for 1 df at the 0.05 level of significance). χ₀² is the chi-square value for the baseline model, and df₀ is the degree of freedom for the baseline model. This method not only helps to control Type I error rates at or below the nominal level but also assists with maintaining high power across different study conditions (Kim et al., 2012).

If uniform DIF items were identified, the effects of gender, grade level, and race/ethnicity on the latent factor means were examined using MIMIC models to determine if latent factor means exhibit bias. The first MIMIC model was the baseline MIMIC model which only includes direct paths from covariates on latent factors. The second MIMIC model was the model in which the direct effects of the covariates on the identified DIF items, as well as latent factors, were included.

The CFA and MIMIC models were evaluated using the following indices commonly used with analysis of categorical data (Finney & DiStefano, 2013): (a) chi-square statistic, (b) comparative fit index (CFI), (c) root mean squared error of approximation (RMSEA), and (e) standardized root mean square residual (SRMR). As chi-square statistics are sensitive to sample size, the values were reported for model comparison purposes. CFI ≥ 0.90, RMSEA ≤ 0.08, and SRMR ≤ 0.10 indicated an acceptable model fit. CFI ≥ 0.95, RMSEA ≤ 0.05, and SRMR ≤ 0.08 suggested good fit (Hu & Bentler, 1999). 90% confidence interval (CI) close to the RMSEA point estimate should contain 0.05 to show the possible close fit (Browne & Cudeck, 1993).

Besides global fit, local fit indices were examined as global model fit might be affected by the magnitude of the factor loadings and the number of items (Greiff & Heene, 2017; McNeish et al., 2018). Residual values, which show the difference between the observed and estimated covariances, were evaluated. Large standardized residuals (e.g., |> 3.0|) suggest a possible local model misfit (Raykov & Marcoulides, 2012). The interpretability of parameter estimates was also examined.

Chi-square tests were conducted to examine the association between the PSC-17 item responses and children’s gender, grade, and race/ethnicity conditions. Applying a Bonferroni correction to adjust the alpha level for group comparison (0.05/17 = 0.002), results revealed significant differences in parental ratings across genders for three out of the five items in Attention Problems and four out of the seven items in Externalizing Problems (p < 0.001). Preschool children were rated differently from 5 K to Grade 1 students, showing differences in two items in Internalizing Problems and one item in Attention Problems (p < 0.001). Furthermore, parent ratings varied among children from different racial backgrounds, manifesting in differences for one item within Internalizing Problems, three within Attention Problems, and two within Externalizing Problems (p < 0.001).

T-tests were conducted to examine subscale scores (i.e., sum scores of the items from each scale) differences across grade levels or genders. Simultaneously, an ANOVA test was performed to assess variation in subscale scores across racial groups. The adjusted alpha level for group comparison, using the Bonferroni correction, was set at 0.017. T-test results indicated that parents rated 5 K children and boys as exhibiting a higher risk level of Internalizing Problems. ANOVA test findings revealed differences in parental ratings among children from different racial groups. Specifically, post-hoc analysis using the Tukey method test suggested that Hispanic children were rated as having a higher risk level of Attention Problems and Externalizing Problems compared to White children. Taken together, these findings underscore the impact of gender, grade level, and race/ethnicity on parent ratings of children’s psychosocial functioning as assessed by the PSC-17 (Table 1).

The single-factor and three-factor models were tested to evaluate the adequacy of the hypothesized 3-factor solution for PSC-17 item scores. A single-factor model allowed all items to load on a single latent factor. In the 3-factor confirmatory Factor Analysis (CFA) model, each item was freely estimated to load on one of the three factors; all factors were intercorrelated. Model fit indices are presented in Table 2.

Based on the criteria of model fit, small residuals, and interpretability of parameter estimates, the single-factor model exhibited poor model fit as all model fit indices were outside of the recommended boundaries (χ²(119) = 1302.676, RMSEA = 0.089, 90% CI [0.085, 0.094]; CFI = 0.854; SRMR = 0.12); this model was not considered further. The three-factor solution yielded an acceptable fit with all the model fit indices within the recommended bounds (χ²(116) = 540.308, RMSEA = 0 0.054, 90% CI [0.050, 0.059]; CFI = 0.948; SRMR = 0.077). Figure 1 shows a path diagram of the 3-factor model.

Standardized factor loadings indicated positive and significant associations of all items with the stated factors. The loading values ranged from 0.45 to 0.89, indicating that each subscale sufficiently captures the variance of the related variables. All the latent factors were significantly correlated. The correlation between latent factors was 0.41 (Attention Problems with Internalizing Problems), 0.57 (Externalizing Problems with Internalizing Problems), and 0.66 (Externalizing Problems and Attention Problems), suggesting that children at a higher risk level in one of the three SEB dimensions were more likely to develop problems in the other two dimensions. Standardized parameter estimates are provided in Table 3.

The coefficient omega for Internalizing Problems, Attention Problems, and Externalizing Problems are 0.87, 0.82, and 0.86, respectively, suggesting high internal consistency for the parent responses.

The baseline model was illustrated in Fig. 2 without a bolded path from gender to Item 9. This baseline model showed a significant model chi-square value (χ²(186) = 682.006, p < 0.001). Other model fit indices yielded adequate model fit with all the indices within the recommended bounds (RMSEA = 0.047, 90% CI [0.044, 0.051]; CFI = 0.930; SRMR = 0.075). Subsequently, the baseline MIMIC model was compared to multiple, less constrained MIMIC models (see Fig. 2 with the bolded path from grade to Item 9 as an example).

The WLSMV model chi-difference test was conducted for each model comparison, using an adjusted critical chi-square value of 13.86 for a nominal alpha of 0.05. As a result, four PSC-17 items with uniform DIF related to children’s racial background were identified. After identifying these items, the baseline MIMIC model was revised by adding the direct effects of race/ethnicity simultaneously on these four DIF items to the baseline MIMIC model. The final MIMIC model with DIF items yielded a good model fit (χ²(182) = 599.635, RMSEA = 0.044, 90%, CI [0.040, 0.048]; CFI = 0.941; SRMR = 0.070). Fit statistics for both the baseline MIMIC model and the final MIMIC model with DIF Items are shown in Table 2.

The final MIMIC model indicated uniform DIF for four PSC-17 items across racial /ethnic groups. The unstandardized parameter estimates for the direct effects of race/ethnicity on the items showing DIF are shown in Table 4. Unstandardized path coefficient values were reported, as they are preferred when covariates are categorical (Kline, 2023). Parents were more likely to rate African American children higher than White children on Item 14 (“Tease others”) in the Externalizing Problems. Compared to White children, Hispanic children were more likely to be rated lower by parents on Item 1(“Fidgety, unable to sit still) and Item 7 (“Has trouble concentrating”) but higher on Item 13 (“Acts as if driven by a motor”). No significant relationship was found between the PSC-17 items and children’s gender and grade level, indicating that the parent-rated PSC-17 functioned equally across children's gender and grade level.

The association between covariates (i.e., gender, grade level, and race/ethnicity) and the means of the three latent factors (i.e., Internalizing Problems, Attention Problems, and Externalizing Problems) was examined to determine the extent to which group comparisons at the latent factor level might be biased by items with DIF. The estimated direct effects of the covariates on the three latent factors from the baseline MIMIC model (i.e., model without the items included) and those from the final MIMIC model (i.e., model with the four items with DIF included) were compared as shown in Table 5. The comparison showed little difference between the two models, indicating that the identified uniform DIF of the four PSC-17 items does not significantly impact the relationship between covariates and the latent factor means. Hence, the latent mean differences should be the true group difference.

Boys exhibited more Attention and Externalizing Problems compared to girls (p < 0.001). However, boys and girls did not significantly differ in the mean level of Internalizing Problems. Preschool children had significantly lower means on Internalizing Problems(p < 0.001) and Attention Problems (p < 0.05) but higher means on Externalizing Problems (p < 0.05) than children from 5 K to the 1st Grade. Relative to White children, African American children had significantly lower means of Internalizing Problems (p < 0.01) and Externalizing Problems (p < 0.01). However, no differences in Attention Problems were identified between the two groups. Compared with White children, Hispanic children had lower mean scores in Internalizing Problems (p < 0.05), Attention problems(p < 0.001), and Externalizing Problems(p < 0.001). Children from other racial groups did not show significant differences from the White children regarding their behavioral problems.

The CFA results confirmed the underlying 3-factor structure of the PSC-17 (i.e., Internalizing Problems, Externalizing Problems, Attention Problems) identified in previous studies using the PSC-17 in school settings (DiStefano et al., 2017; Liu et al., 2020b; Gao et al., 2022) and clinical settings (Chaffin et al., 2017; Murphy et al., 2016; Stoppelbein, et al., 2012; Wagner et al., 2015). These findings support the multidimensionality of the parent-rated PSC-17. Therefore, school practitioners should move beyond the traditional classification of children as simply normal or abnormal regarding their psychosocial functioning. Instead, they can use the PSC-17 to assess specific areas of children’s psychosocial functioning, such as internalizing, externalizing, and attention problems, and target interventions that appropriately address these specific areas.

All parent-rated PSC-17 items performed consistently across child gender and grade level, indicating that school practitioners can use the parent-rated PSC-17 to compare children’s psychosocial functioning across these demographics. However, 4 out of the 17 items exhibited DIF across the child race. Specifically, parents of African American children were more likely to choose higher response options for “teases others”, a finding consistent with the study by Studs et al. (2017). Compared to parents of White children, parents of Hispanic children were more inclined to select higher responses for Item 13 (“Acts as if driven by a motor”) but lower responses for Item 1(“Fidgety, unable to sit still”) and Item 7(“Has trouble concentrating”). These results suggest that the scores on these four items do not have the same meaning for parents of White children and parents of minority children. According to Wainer (1995), the presence of DIF items in a scale can affect the scale-level measurement differently. Therefore, when school practitioners implement universal screening with parent-rated PSC-17, they should consider revising or deleting these DIF Items to enhance the sensitivity and specificity of identifying behavioral problems among diverse early-aged children. In particular, school practitioners should be cautious when making comparisons across children from different racial groups at the item level.

The DIF items did not affect the latent mean differences across child gender, grade level, or race. This might be because the uniform DIF of individual items did not consistently favor one group over others (Wiesner et al., 2015). The findings indicated that DIF items did not bias the scale-level measurement of the psychosocial functioning of children with the parent-rated PSC-17. Thus, school practitioners can use parent-rated PSC-17 to compare differences in children’s internalizing, externalizing, and attention problems across gender, grade level, and racial groups.

Latent mean comparison across genders showed that boys exhibited a higher risk of attention problems and externalizing problems than girls, which was consistent with previous findings (Liu at al., 2020b; Chi & Cui, 2020; Nigg & Nikolas, 2008; Reid et al., 2000). The finding indicates that schools and families should provide extra support to male students in enhancing their attention and externalizing behavior skills. In terms of racial group variation, compared to White children, African American children demonstrated a lower risk of internalizing and ixternalizing problems. Hispanic children showed lower risk levels of internalizing, attention, and externalizing problems compared to White children. The finding is supported by previous studies indicating that minority parents are less likely to identify their children’s maladjustment (Robert et al., 2005), and White parents are more sensitive to their children’s maladjustment than minority parents (Lau et al., 2004). As parents play a crucial role in helping children enhance behavioral skills at home, schools could organize training sessions for parents of diverse racial groups on how to effectively identify and support their children’s psychosocial functioning. Children in 5 K and the first grade exhibited a higher risk for internalizing and attention problems but a lower risk level for externalizing problems relative to preschoolers. This finding aligns with the previous studies showing that internalizing problems tend to increase as children age (Achenbach et al., 1991; Gilliom & Shaw, 2004) and externalizing problems decrease throughout childhood for most children (Fanti & Henrich, 2007; Shaw et al., 2003). Regarding attention problems, the finding is consistent with previous study results that ADHD symptoms developed during preschools become increasingly impairing during elementary school (APA, 2013). These findings imply that schools should provide universal screening for children at the preschool stage for early identification and intervention to reduce the possibility of further development of psychosocial difficulties at an older age.

The present study has several limitations. The current study was conducted with samples from pre-K to the first grade from public schools in two US states. The findings may not generalize to clinical samples, older children, and children from other different geographical regions. Hispanic children were overrepresented in the current sample (53.6% of participants). Large national studies with children from diverse backgrounds in different parts of the country should be conducted to determine whether the present results can be replicated. Additionally, the current study examined uniform DIF items from PSC-17 as a function of student gender, race/ethnicity, and grade level, but future research should consider factors such as socioeconomic status and language background for a more comprehensive view.