Assessing the Parenting Stress Index-Short Form (PSI-SF) for Latinx Parents of Children with Intellectual and Developmental Disabilities Within Exploratory Structural Equation Modeling and Structural Equation Modeling-Based Generalizability Theory Frameworks

Abstract

Although the Parenting Stress Index-Short Form (PSI-SF) is widely applicable, there has been a dearth of research on the psychometric properties and item wording effects associated with the cultural sensitivity of the instrument. This study investigated the psychometric properties, the factor structure, and the negative and positive item wording effects of PSI-SF for 96 Latinx parents of children with intellectual and developmental disabilities in the United States within exploratory structural equation modeling, and structural equation modeling-based generalizability theory frameworks. This study produces the following noteworthy findings. First, the ESEM bifactor model with negatively and positively phrased items best captured the theoretical frameworks underlying the structure of the PSI-SF. Second, adding method factors enhanced the model fits of all the factor models. Third, this study discovered several items with great proportions of method factors. This study recommended that the items with high proportions of method factors, which may likely be culturally insensitive to Latinx parents, be examined further and refined. With the use of accurately assessed PSI-SF scores, practitioners will be better able to support and empower Latinx parents of children with IDD and better meet their needs through the FIRME program, a parent advocacy initiative.

1. Introduction

The Parenting Stress Index (PSI) is one of the most commonly employed instruments to assess parental stress in a diverse range of families and children. It is especially helpful in determining the causes and levels of stress, which can result from a variety of effects like the child’s conduct, the psychological traits of the parent, and the dynamics of the parent–child relationship as a whole [1].

The Parenting Stress Index-Short Form (PSI-SF) is a condensed form of the original PSI that preserves the validity and reliability of the full form while offering a quicker evaluation. It assesses stress related to parenting in three key areas: Parental Distress (PD), which is a term used to describe stress that is specifically associated with being a parent and feelings of overburden; Parent–Child Dysfunctional Interaction (PCDI), which investigates stress resulting from the parent’s belief that their relationship with the child falls short of their expectations; and Difficult Child (DC), which assesses stress related to challenging child behaviors [1].

Previous research on the factor structure and psychometric properties of the Parenting Stress Index-Short Form (PSI-SF) endeavored to demonstrate its utility as a reliable and valid measure of parenting stress [2], but there has been a lack of agreement on the factor structure of the PSI-SF (e.g., [3,4]). Some studies have supported the three factor structure of the PSI-SF, which assesses parental distress (PD), parent–child dysfunctional interaction (PCDI), and difficult child (DC) through confirmatory factor analyses across diverse populations (e.g., [5,6,7,8,9]). On the other hand, in other studies, the three factor structure of the PSI-SF was not supported well (e.g., [3,4,10]).

Additionally, the PSI-SF has been suggested to be modified when used with parents of culturally and linguistically diverse (CLD) backgrounds, including Latinx [8,9]. Rios et al. (2021, 2024) [8,9] pointed out that cultural factors like familism (the emphasis on strong family connections) may influence how stress is perceived, especially in the PCDI domain. Latinx parents often report heightened stress levels related to both economic hardships and cultural expectations around family cohesion, which are not always fully captured by standard PSI-SF measures.

Despite the widespread use of the PSI-SF, there has been a dearth of research on the psychometric properties, factor structure, and item wording effects associated with the cultural sensitivity of the instruments for Latinx parents of children with intellectual and developmental disabilities (IDD). Therefore, first, this study evaluates the psychometric properties, the factor structure (structural validity), and negative and positive item wording effects on model fits, method factors, and measurement error based on 96 Latinx parents of children with IDD in the United States. Second, this study identifies culturally insensitive items with high proportions of negative and positive item wording effects and suggests that these items need to be examined further and refined to appropriately reflect the stress levels faced by Latinx parents of children with IDD. To achieve these goals, and for reasons described in the foregoing sections, the researchers applied one factor, three factor, and bifactor representations of the PSI-SF using confirmatory factor analysis, exploratory structural equation modeling techniques, and SEM-based GT procedures.

2. The Parenting Stress Index (PSI) and Its Short Form

The Parenting Stress Index (PSI) is a widely recognized tool used to evaluate the level of stress that parents experience in their caregiving roles. It is particularly valuable for identifying both the sources and extent of stress, which can arise from various factors such as the child’s behavior, the parent’s psychological characteristics, and the overall parent–child relationship dynamics. Designed for parents of children aged zero to twelve years, the PSI is utilized in both clinical and research settings to guide interventions aimed at alleviating parenting stress. The PSI comprises 120 items scored on a five-point Likert scale, measuring stress across two main domains: the Child Domain, which assesses stress related to specific child characteristics (e.g., demandingness, adaptability, mood), and the Parent Domain, which focuses on stressors linked to the parent’s psychological well-being, sense of competence, and social support. The PSI provides a total stress score, with higher scores indicating greater levels of parenting stress, and is frequently used to identify areas where intervention may be needed to support both the parent and child [1].

The Parenting Stress Index-Short Form (PSI-SF) is a streamlined version of the original PSI, developed to provide a faster assessment while maintaining the reliability and validity of the full form. Consisting of 36 items, the PSI-SF significantly reduces the time needed for completion. The PSI-SF generates a total stress score, with higher scores indicating greater stress levels, similar to the full version. Due to its brevity and accuracy, the PSI-SF is widely used in clinical contexts, especially when time is limited, and is also valuable in research settings for its ability to offer quick yet insightful evaluations for intervention planning. Both the PSI and PSI-SF are valuable for identifying stress that can negatively impact parenting and child outcomes. This tool is often used in conjunction with other assessments to guide interventions aimed at reducing stress and supporting families, especially those facing additional challenges, such as families or those raising children with IDD [1].

The PSI-SF is used to test the effects of the Families Included in Receiving Better Special Education Services (FIRME) program, a parent advocacy program that aims to educate Latinx parents of children with IDD to advocate for school services and empower them through non-adversarial advocacy skills by evaluating parenting stress for this population [9]. The use of the PSI-SF received important insights from the FIRME program. Contrary to expectations, findings from the study indicated that many Latinx parents experienced an increase in stress following participation in the program. This was reflected in higher PSI-SF scores, particularly in the parental distress and difficult child subscales. The increase in stress may be attributed to the additional responsibilities and emotional burdens that came with engaging in advocacy work, as well as the challenges of navigating complex systems to support their children’s needs. These results underscore the complexity of advocacy efforts, suggesting that while such programs build critical skills, they may also contribute to heightened stress for parents, necessitating further support mechanisms to manage the added pressures [9].

3. Previous Research on the Factor Structure and Psychometric Properties of the PSI-SF

The PSI-SF has been found to exhibit strong internal consistency, with Cronbach’s alpha values typically exceeding the acceptable threshold of 0.70 for each subscale, indicating high reliability [9,11]. Additionally, the total stress score derived from the PSI-SF has shown robust correlations with related constructs such as parental mental health, child behavior problems, and family functioning, providing evidence of its construct validity [11].

Even though the PSI-SF is generally accepted as a psychometrically sound tool for measuring parenting stress [2,12], there has been disagreement on the factor structure of the PSI-SF (e.g., [2,3]). Some studies have supported the three factor structure of the PSI-SF, which assesses parental distress (PD), parent–child dysfunctional interaction (PCDI), and difficult child (DC) through confirmatory factor analyses, which have shown that this structure provides a good fit across diverse populations, reinforcing its applicability in both clinical and research settings (e.g., [5,6,7,8,9,13]). However, in other studies, the three factor structure of the PSI-SF was not supported well (e.g., [2,3,10]). For example, using CFA, Haskett et al. (2016) [2] demonstrated that the three factor model did not fit well into a mixed sample of 185 parents, 68% of whom were African Americans, and suggested a new two factor model. There is conflicting evidence to support the three factor model.

Although cross-cultural validation studies have also supported the utility of the PSI-SF in culturally and linguistically diverse populations [14], it has been suggested that minor adjustments might be necessary to enhance its cultural applicability when used with parents of culturally and linguistically diverse (CLD) backgrounds, including Latinx [8,9]. For example, Rios et al. (2021, 2024) [8,9] pointed out that cultural factors like familism (i.e., the emphasis on strong family connections) may influence how stress is perceived, especially in the PCDI domain. Latinx parents often report heightened stress levels related to both economic hardships and cultural expectations around family cohesion, which are not always fully captured by standard PSI-SF measures. Studies like Rios et al. (2021, 2024) [8,9] call for ongoing research and adaptations to the tool, ensuring that it captures the cultural nuances and specific stressors experienced by these families.

4. An Overview of Factor Analytic Techniques, Generalizability Theory, and Method Effects

The literature reviews up to this point have focused on the PSI-SF and previous research on the psychometric properties of the PSI-SF. The following section presents the background information on confirmatory factor analysis, exploratory structural equation modeling, methods effects, and generalizability theory.

5. Confirmatory Factor Analysis and Exploratory Structural Equation Modeling

Confirmatory factor analysis (CFA) is a structural equation modeling (SEM) technique that evaluates the relationships between latent factors and observed indicators (e.g., items) [15]. CFA is a popular statistical method for examining the intricate multifaceted structures, such as correlated-factor, hierarchical, and bifactor models, that underlie personality inventories. Although CFA is theory-driven, its rigorous assumptions about perfect zero cross-loadings and residual covariances frequently result in unacceptable model fits and significant parameter biases in the estimation of factor loadings and correlations [15,16,17,18].

Asparouhov and Muthén (2009) [19] proposed exploratory structural equation modeling (ESEM) as a substitute approach to address the shortcomings of conventional factor analytic methods by combining the methodological benefits of CFA, SEM, and Exploratory Factor Analysis (EFA). Contrasting with traditional CFA, ESEM is more data driven. Items in ESEM load on all factors but are intended to load significantly higher on targeted than on non-targeted factors. According to prior research, ESEM outperforms CFA/SEM in terms of model fits and precision of parameter estimates (e.g., [19,20,21]). ESEM also enables goodness-of-fit evaluation, multiple-group invariance tests, longitudinal differential item functioning, higher-order factor structures, growth modeling, and other uses ([17,19,22,23]).

6. The Bifactor Model

In this study, one factor, three factor, and bifactor models are applied. Bifactor (also referred to as nested-factor or general-specific) models contain a latent structure consisting of a general factor that accounts for the commonality of all indicators and unrelated group factors that indicate systematic variance unique to nested and non-overlapping subsets of indicators [15,16,24,25]. The bifactor model postulates that covariation in observed scores is explained by both global and more specific factors [18].

The bifactor model has the following benefits ([24]). First, the bifactor model can serve as a baseline model to be compared to the second-order model using a likelihood ratio test because the second-order model is nested within the bifactor model [26,27]. Second, the bifactor model can be utilized to evaluate the significance of domain-specific factors that are impertinent to the general factor. For example, applying an original conception of general and domain-specific factors to PSI-SF [1], it is assumed that there was a factor representing general parental stress as well as domain-specific factors reflecting separate stress levels such as parental distress, parent–child dysfunctional interaction, and difficult child. With the effects of the general factor partialed out, bifactor models display the strength of correlation between the individual group factors and their respective indicators. Lastly, while simultaneously controlling for each other’s impacts, the bifactor model can be utilized to estimate results using both the general and group factors [15,24].

7. Method Effects

Method effects indicate a respondent’s systematic tendencies to respond to questions based on construct-irrelevant factors [15,18,28]. Such effects can result from many sources producing further covariation among item scores within self-report measures that involve a combination of positively and negatively phrased items. When not addressed properly, method effects can distort interpretations of scores, the nature of constructs being studied, and evidence of reliability and validity ([28,29]).

8. Generalizability Theory

Generalizability theory (GT; [30,31,32,33]) has been shown to be a useful paradigm for understanding the manner in which different sources of measurement error affect assessment scores when interpreting the reliability and validity of results from psychological assessments. GT can be used to quantify sources of measurement error, to assess how changing measurement methods can improve psychometric qualities, as well as to create score reliability and validity indices for particular applications. GT-based procedures can be undertaken employing structural equation models (SEMs; e.g., [34,35,36]). One significant, albeit understudied, application of GT-SEM analyses is to estimate the proportions of observed score variance that are attributed to item wording effects in psychological assessment. The effects of item phrasing are frequently handled within such models by introducing separate orthogonal factors for negatively and/or favorably phrased items, or by linking uniquenesses for negatively and/or positively phrased items (e.g., [37]). However, these strategies are often employed to improve model fit rather than to establish how much of the observed score variance is explained by item wording.

9. Purpose of the Present Study

Despite the wide applicability of the PSI-SF, there has been a lack of research on the psychometric properties, factor structure, and item wording effects related to the cultural sensitivity of the tool in Latinx parents of children with IDD. To the best of our knowledge, no studies examined the psychometric properties, the factor structure (structural validity), and negative and positive item wording effects of the PSI-SF based on Latinx parents of children with IDD in the United States using CFA and ESEM bifactor models and SEM-based Generalizability theory frameworks.

The primary goal of this study is twofold. First, this study evaluates the psychometric properties, the factor structure (structural validity) and item wording effects of the PSI-SF on model fits, method factors, and measurement error based on Latinx parents of children with autism and other disabilities in the United States. Second, this study identifies culturally insensitive items with high proportions of negative and positive item wording effects and suggests that these items need to be refined to appropriately reflect the stress levels faced by Latinx parents of children with IDD. To achieve the goals, this study applied one factor, three correlated-factor, and bifactor models using confirmatory factor analysis, exploratory structural equation modeling techniques, and SEM-based GT procedures.

The following main research questions and sub-questions will guide this study.

RQ1. What are the psychometric properties of the PSI-SF for the Latinx parents of children with IDD?

(1)

What factor models will best represent the framework of the Parenting Stress Index-Short Form (PSI-SF)?

(2)

To what extent does the inclusion of negatively and positively phrased items affect the model fit and the magnitude of factor loadings, method factors, and measurement error?

RQ2. How accurately does the PSI-SF represent the stress experienced by low-income Latinx parents of children with IDD?

(1)

Which items will result in the high(est) proportion of method factors in the SEM-based congeneric GT bifactor models?

10. Methods

10.1. Participants

The study involved 96 Latinx parents of children with IDD from two states in the United States. Participants were eligible if they were Latinx parents of children with IDD and enrolled in an advocacy training program. For this study, the term Latinx referred to individuals either born in or with heritage from Latin America [38,39]. Participants were excluded if they were not Latinx and did not enroll in an advocacy program. Children included in the study were between the ages of five and eighteen and were living at home, as this was essential for exploring parent–child dynamics related to the study’s objectives.

The sample size was determined based on previous research in this area [11]. We performed the power analysis using the semTools package in R [40]. The power analysis indicated that the sample sizes of 96 and 54 achieved the powers of 0.995 and 0.85, respectively, and this study yielded roughly 0.70 power with the sample size of parents of autistic children (N = 42). It was sufficient for the statistical analyses conducted, which met the objectives of the study.

While expanding the sample further could have been beneficial, practical constraints such as time, resources, and recruitment challenges limited our ability to do so. As such, data from only 96 families was available. Even though data were collected in two states over a prolonged period of time (namely five years), obtaining this sample size was already a challenge for two main reasons. Firstly, the number of Latinx parents of children with IDD is relatively low. Secondly, the acute period in which families were recruited to participate is extremely demanding. The majority of this data was collected during COVID-19. Participating in research in this demanding and emotional phase may be perceived as an additional burden by parents.

10.2. Data Collection Procedures

First, Institutional Review Board (IRB) approval was obtained. Purposeful sampling was employed to select participants for the study [41]. Recruitment took place through local and statewide agencies, community organizations (e.g., parent support groups, Latino/a-serving churches), and social media platforms like Facebook. The recruitment strategy incorporated personalismo, a culturally responsive approach involving the development of confianza (trust) between families and professionals [42]. A bilingual Latina researcher built relationships with Latinx parents of children with IDD by volunteering with Latino/a organizations and conducting community-based research. Eligible participants then completed a survey related to the advocacy training program. This survey, offered in both English and Spanish, was completed entirely in Spanish by all participants. The advocacy training program, discussed in detail in previous research [8,9,43,44], provided 12 h of instruction on special education policy, non-adversarial advocacy skills, and empowerment. The training, delivered in-person in Spanish, was facilitated by native Spanish-speaking Latina instructors. Each participant received a USD 20 stipend, and the second author entered survey data into SPSS software [45] for analysis.

10.3. Parenting Stress Index Scale-Short Form (PSI-SF, Abidin, 1990)

The Parenting Stress Index-Short Form (PSI-SF, [1]) is a 36-item questionnaire with five Likert-type response options ranging from one (strongly disagree) to five (strongly agree). High overall scores indicate that participants have greater parental stress. The PSI-SF is available in Spanish [46]. The PSI-SF has been translated into Spanish and successfully used with Latinx parents of children with IDD with high reliability (e.g., [11]). The first and second authors classified the items of the PSI-SF into 15 negatively phrased items and 21 positively phrased items.

10.4. Data Analysis

Descriptive statistics, reliability coefficients, and 21 models as described in

Table 1. The factor models analyzed.

Estimation Methods	Models
One factor	CFA WLSMV
	CFA WLSMV with negatively phrased items
	CFA WLSMV with both negatively and positively phrased items
Three factor	CFA WLSMV
	CFA WLSMV with negatively phrased items
	CFA WLSMV with both negatively and positively phrased items
Bifactor	CFA WLSMV
	CFA WLSMV with negatively phrased items
	CFA WLSMV with both negatively and positively phrased items
One factor	ESEM WLSMV
	ESEM WLSMV with negatively phrased items
	ESEM WLSMV with both negatively and positively phrased items
Three factor	ESEM WLSMV
	ESEM WLSMV with negatively phrased items
	ESEM WLSMV with both negatively and positively phrased items
Bifactor	ESEM WLSMV
	ESEM WLSMV with negatively phrased items
	ESEM WLSMV with both negatively and positively phrased items
Bifactor	CFA WLSMV congeneric GT
	CFA WLSMV congeneric GT with negatively phrased items
	CFA WLSMV congeneric GT with both negatively and positively phrased items

Note. CFA = confirmatory factor analysis; ESEM = exploratory structural equation modeling; WLSMV = weighted least squares with mean and variance adjusted.

were analyzed based on all parents, a group of parents of children with autism, and a group of parents of children with other disabilities using lavaan package in R [47] and Mplus 8.10 [48]. We used congeneric SEM models in which trait, method factor, and error score variance can differ across items by allowing both uniquenesses and unstandardized factor loadings to vary. Figure 1, Figure 2 and Figure 3 display a one factor model denoting PSI, a three factor model representing PD, PCDI, and DC, and a bifactor model of the PSI-SF comprising a general factor of PSI and group factors of PD, PCDI, and DC.

The model fit was evaluated using the comparative fit index (CFI; [49]), Tucker–Lewis index (TLI; [50]), and root-mean-square error of approximation (RMSEA; [51]). Values larger than 0.90 and 0.95 for the CFI and TLI indicate an acceptable and excellent fit to the data, respectively, while values less than 0.06 for the RMSEA support an outstanding model fit and values less than 0.08 suggest an adequate model fit.

10.5. Sample Characteristics

Table 2. Sample characteristics (N = 96).

Characteristics		Percent (Frequency) or Mean (SD)
Respondent gender: Female		96.9% (93)
Education background	Some high school	28.1% (27)
	High school degree	31.3% (30)
	Some college	22.9% (22)
	Four-year degree	8.3% (8)
	Graduate/Professional degree	9.4% (9)
Annual household income	Less than USD 15,000	30.2% (29)
	Between USD 15 and 29,000	32.3% (31)
	Between USD 30 and 49,000	22.9% (22)
	Between USD 50 and 69,000	7.3% (7)
	Between USD 70 and 99,000	6.3% (6)
	Over USD 100,000	1% (1)
Ethnicity	Mexican	85.4% (82)
	Puerto Rican	4.2% (4)
	Central American	4.2% (4)
	South American	5.2% (5)
	Other/mixed	1% (1)
Child disability	Autism	43.8% (42)
	Other	56.2% (54)
Child gender	Male	77.1% (74)
	Child age	9.56 (4.72)

indicates the demographics of the 96 parents. The sample was predominantly female, with 97% (n = 93) of participants being women, and their average age was 40.854 years (SD = 6.85). The children were mostly male (77.1%, n = 74), and 58.33% (n = 56) of the children were 9 years old or younger with an average age of 9.56 (SD = 4.72). Parents self-reported their child’s disability, with 43.8% (n = 42) identifying their child as having autism. Most families had an annual household income of less than USD 49,000 (85%, n = 82). The majority of parents reported that they received some high school education (28.1%, n = 27), graduated from high school (31%, n = 30), and had some college education (22.9%, n = 22). Only 17.7% (n = 17) claimed to hold a four-year or graduate degree.

11. Results

11.1. Descriptive Statistics

Table 3. Descriptive statistics.

Sample	Scale	Mean: Scale (Item)	SD: Scale (Item)	Alpha	Omega
All	PSI	116.385 (3.233)	25.133 (0.698)	0.931	0.934
	PD	37.167 (3.097)	10.390 (0.866)	0.897	0.904
	PCDI	40.083 (3.34)	10.032 (0.836)	0.857	0.866
	DC	39.135 (3.261)	9.367 (0.781)	0.85	0.856
Autism	PSI	120.905 (3.358)	19.699 (0.547)	0.907	0.914
	PD	39.810 (3.317)	9.163 (0.764)	0.882	0.888
	PCDI	41.952 (3.496)	8.166 (0.680)	0.834	0.846
	DC	39.143 (3.262)	6.411 (0.534)	0.699	0.721
Other	PSI	112.870 (3.135)	28.342 (0.787)	0.94	0.944
	PD	35.111 (0.975)	10.895 (0.303)	0.898	0.908
	PCDI	38.630 (1.073)	11.131 (0.309)	0.869	0.875
	DC	39.130 (1.087)	11.201 (0.311)	0.895	0.9

Note. PSI = parenting stress index; PD = parental distress; PCDI = parent–child dysfunctional interaction; DC = difficult child.

included descriptive statistics for all 96 Latinx parents, parents of children with autism, parents of children with other disabilities (means and standard deviations), and conventional reliability estimates (alpha and omega; see Table 3). The total average PSI, PD, PCDI, and DC scores of parents of children with autism (M = 120.905 for PSI; M = 39.810 for PD; M = 41.952 for PCDI; M = 39.143 for DC) were much higher than those of parents of children with other disabilities (M = 112.870 for PSI; M = 35.111 for PD; M = 28.63 for PCDI; M = 39.130 for DC). Parents of children with autism also reported greater stress on average than parents of children with other disabilities at the item level (e.g., 3.317 vs. 0.975 for PD; 3.496 vs. 1.073 for PCDI; 3.262 vs. 1.087 for DC). Alpha reliability estimates ranged from 0.851 to 0.930 for all; from 0.699 to 0.907 for parents of children with autism; and from 0.869 to 0.940 for parents of children with other disabilities. Omega coefficients ranged from 0.856 to 0.934 for all; from 0.721 to 0.914 for parents of children with autism; and from 0.875 to 0.944 for parents of children with other disabilities.

11.2. Examining the Factor Structure of the PSI-SF Within CFA and ESEM Frameworks

In the following section, we investigated the factor structure of the PSI-SF for all Latinx parents, and the parents who have children with autism and other disabilities and the negative and positive item wording effects on model fit and factor loadings; we also compared differences between these two groups using CFA and ESEM one factor, three factor, and bifactor models.

11.2.1. Model Fit for All Groups (N = 96)

The results in

Table 4. The model fit statistics for all parents (N = 96).

CFA WLSMV	Parameters	Chi-Square	df	CFI	TLI	RMSEA
One factor	182	1363.274	594	0.812	0.801	0.116
One factor with negative	195	1249.847	581	0.837	0.823	0.11
One factor with negative and positive	218	1073.902	558	0.874	0.858	0.098
Difference				0.062	0.057	−0.018
Three factor	185	1051.463	591	0.888	0.88	0.09
Three factor with negative	198	978.698	578	0.902	0.893	0.085
Three factor with negative and positive	221	926.133	555	0.909	0.897	0.083
Difference				0.021	0.017	−0.007
Bifactor	218	895.783	558	0.918	0.907	0.079
Bifactor with negative	231	850.639	545	0.925	0.914	0.076
Bifactor with negative and positive	254	780.503	522	0.937	0.924	0.072
Difference				0.019	0.017	−0.007
ESEM WLSMV	Parameters	Chi-square	df	CFI	TLI	RMSEA
One factor	182	1363.274	594	0.812	0.801	0.116
One factor with negative	195	1249.847	581	0.837	0.823	0.11
One factor with negative and positive	218	1073.902	558	0.874	0.858	0.098
Difference				0.062	0.057	−0.018
Three factor	251	881.28	525	0.913	0.896	0.084
Three factor with negative	264	832.998	512	0.922	0.904	0.081
Three factor with negative and positive	287	767.259	489	0.932	0.912	0.077
Difference				0.019	0.016	−0.007
Bifactor	284	803.32	492	0.924	0.903	0.081
Bifactor with negative	297	767.98	479	0.929	0.907	0.079
Bifactor with negative and positive	320	689.875	456	0.943	0.921	0.073
Difference				0.019	0.018	−0.008

Note. df = degree of freedom; CFA = confirmatory factor analysis; ESEM = exploratory structural equation modeling; WLSMV = weighted least squares with mean and variance adjusted; CFI = comparative fit Index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; Difference = difference in model fits between a model without negatively and positively phrased items and a model with both negatively and positively phrased items.

revealed that the CFA bifactor model with the inclusion of negatively phrased items produced more improved model fits than the corresponding CFA model without those items in relation to CFIs, TLIs, and RMSEAs (CFI = 0.925 vs. 0.918, TLI = 0.914 vs. 0.907, RMSEA = 0.076 vs. 0.079). Adding both negatively and positively phrased items also enhanced model fits (CFI = 0.937 vs. 0.918, TLI = 0.924 vs. 0.907, RMSEA = 0.072 vs. 0.079). These trends also held in ESEM models. ESEM bifactor models with the inclusion of negatively phrased items alone and both negatively and positively phrased items produced better model fits than the corresponding ESEM model without those items in terms of CFIs, TLIs, and RMSEAs (see Table 4).

CFA and ESEM one factor models provided the poorest fits. Overall, the best-fitting model was the ESEM bifactor model with negatively and positively phrased items in terms of CFI (0.943) and the CFA bifactor model with negatively phrased items in relation to TLI (0.924) and RMSEA (0.072). Based on the criteria of CFIs and TLIs ≥ 0.90 and RMSEAs ≤ 0.08, the CFA bifactor model and the ESEM three factor and bifactor models without negatively phrased items and the CFA and ESEM three factor and bifactor models with negatively phrased items alone and both negatively and positively phrased items yielded acceptable fits.

11.2.2. Model Fit for Parents of Children with Autism (N = 42)

The results in

Table 5. The model fit statistics for parents of children with autism (N = 42).

CFA WLSMV	Parameters	Chi-Square	df	CFI	TLI	RMSEA
One factor	177	856.877	594	0.781	0.768	0.103
One factor with negative	189	800.652	582	0.818	0.803	0.095
One factor with negative and positive	212	752.313	559	0.839	0.819	0.091
Difference				0.058	0.051	−0.012
Three factor	180	831.6	591	0.8	0.787	0.098
Three factor with negative	192	780.637	579	0.832	0.818	0.091
Three factor with negative and positive	215	729.961	556	0.855	0.836	0.086
Difference				0.055	0.049	−0.012
Bifactor	213	736.753	558	0.851	0.832	0.087
Bifactor with negative	225	715.893	546	0.859	0.837	0.086
Bifactor with negative and positive	249	667.548	522	0.879	0.854	0.081
Difference				0.028	0.022	−0.006
ESEM WLSMV	Parameters	Chi-square	df	CFI	TLI	RMSEA
One factor	177	856.877	594	0.781	0.768	0.103
One factor with negative	190	799.385	581	0.818	0.803	0.095
One factor with negative and positive	213	750.943	558	0.84	0.819	0.091
Difference				0.059	0.051	−0.012
Three factor	246	1832.739	525	0.92	0.904	0.066
Three factor with negative	259	594.546	512	0.931	0.916	0.062
Three factor with negative and positive	282	555.694	489	0.945	0.929	0.057
Difference				0.025	0.025	−0.009
Bifactor	279	571.633	492	0.934	0.915	0.062
Bifactor with negative	291	546.14	480	0.945	0.928	0.057
Bifactor with negative and positive	314	511.526	457	0.955	0.938	0.053
Difference				0.021	0.023	−0.009

Note. df = degree of freedom; CFA = confirmatory factor analysis; ESEM = exploratory structural equation modeling; WLSMV = weighted least squares with mean and variance adjusted; CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; Difference = difference in model fits between a model without negatively and positively phrased items and a model with both negatively and positively phrased items.

indicate that ESEM three factor and bifactor models yielded better model fits on average than the CFA models regarding CFIs, TLIs, and RMSEAs. Adding a method factor of negatively phrased items and method factors of both negatively and positively phrased items to all models yielded salient improvements in CFIs, TLIs, and RMSEAs. For the ESEM three factor model, adding method factors of both negatively and positively phrased items improved CFIs, TLIs, and RMSEAs on average by 0.025, 0.025, and −0.009, respectively. The model fit of the ESEM bifactor model also improved by 0.021 for CFI, 0.023 for TLI, and decreased by 0.009 when adding method factors of negatively and positively phrased items.

The ESEM bifactor model with method factors of negatively and positively phrased items provided the best fits (CFI = 0.955, TLI = 0.938, RMSEA = 0.053). Based on the criteria of CFIs and TLIs ≥ 0.90 and RMSEAs ≤ 0.08, all ESEM three factor and bifactor models with and without negatively and positively phrased items yielded acceptable fits (see Table 5).

11.2.3. Model Fit for Parents of Children with Other Disabilities (N = 54)

The results in

Table 6. The model fit statistics for parents of children with other disabilities (N = 54).

CFA WLSMV	Parameters	Chi-Square	df	CFI	TLI	RMSEA
One factor	184	1068.537	594	0.826	0.815	0.122
One factor with negative	196	970.575	582	0.857	0.846	0.111
One factor with negative and positive	219	855.418	559	0.891	0.877	0.099
Difference				0.065	0.062	−0.023
Three factor	187	930.883	591	0.875	0.867	0.103
Three factor with negative	200	858.56	578	0.897	0.888	0.095
Three factor with negative and positive	221	832.662	557	0.899	0.886	0.096
Difference				0.024	0.019	−0.007
Bifactor	220	823.879	558	0.902	0.89	0.094
Bifactor with negative	233	786.007	545	0.912	0.898	0.09
Bifactor with negative and positive	256	738.147	522	0.921	0.904	0.088
Difference				0.019	0.014	−0.006
ESEM WLSMV	Parameters	Chi-square	df	CFI	TLI	RMSEA
One factor	184	1068.46	594	0.826	0.815	0.122
One factor with negative	196	970.575	582	0.857	0.846	0.111
One factor with negative and positive	218	891.345	560	0.878	0.863	0.105
Difference				0.052	0.048	−0.017
Three factor	253	738.066	525	0.922	0.906	0.087
Three factor with negative	265	691.022	513	0.935	0.92	0.08
Three factor with negative and positive	288	651.333	490	0.941	0.924	0.078
Difference				0.019	0.018	−0.009
Bifactor	286	658.895	492	0.939	0.922	0.079
Bifactor with negative	298	629.15	480	0.945	0.928	0.076
Bifactor with negative and positive	321	577.039	457	0.956	0.939	0.070
Difference				0.017	0.017	−0.009

Note. df = degree of freedom; CFA = confirmatory factor analysis; ESEM = exploratory structural equation modeling; WLSMV = weighted least squares with mean and variance adjusted; CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; Difference = difference in model fits between a model without negatively and positively phrased items and a model with both negatively and positively phrased items.

reveal that the model fits for all models with method factors of both negatively and positively phrased items were better on average than the fits for corresponding models without method factors. Improvements in fit when adding method factors were most pronounced for the CFA one factor model (CFI = 0.891 vs. 0.826, TLI = 0.877 vs. 0.815, RMSEA = 0.099 vs. 0.122) and less pronounced in CFIs (0.956 vs. 0.939), TLIs (0.939 vs. 0.922), and RMSEAs (0.070 vs. 0.079) for the ESEM bifactor model.

The ESEM bifactor model with both negatively and positively phrased items produced the highest model fits in terms of CFI (0.956), TLI (0.939), and RMSEA (0.070). According to the cutoff rules of CFIs and TLIs ≥ 0.90 and RMSEAs ≤ 0.08, the ESEM three factor and bifactor models with and without negatively and positively phrased items yielded acceptable fits.

11.2.4. Summary

When considering model fit indices collectively, ESEM bifactor models with the inclusion of method factors for both negatively and positively phrased items produced enhanced model fits than the corresponding models without those method factors within and across all parents, parents of children with autism, and other disabilities. ESEM bifactor models with negatively and positively phrased items within each sample also provided the highest and most acceptable model fits.

In the next section, we evaluated primary factor loadings and explained variance due to trait and method factors of the three best-fitting models, ESEM bifactor models without method factors with the inclusion of a method factor of negatively phrased items only, and method factors of both negatively and positively phrased items.

11.3. Factor Loadings and Explained Variance

Table 7. The mean trait and method factor loadings.

Sample	Analysis	Stress	PD	PCDI	DC	Neg	Pos	EV Due to Trait	EV Due to Negative	EV Due to Positive
All	ESEM	0.512	0.532	0.232	0.370			0.648
	ESEM with Negative	0.509	0.534	0.214	0.420	0.114		0.588	0.056
	ESEM with both	0.513	0.184	0.056	0.082	0.097	0.225	0.655	0.048	0.125
Autism	ESEM	0.457	0.444	0.024	0.195			0.646
	ESEM with Negative	0.454	0.466	0.019	0.168	0.099		0.641	0.057
	ESEM with both	0.452	0.173	−0.001	0.058	0.099	0.067	0.636	0.060	0.091
Other	ESEM	0.525	0.409	0.253	0.474			0.657
	ESEM with Negative	0.536	0.289	0.228	0.504	0.039		0.663	0.054
	ESEM with both	0.534	0.135	0.073	0.134	0.025	0.167	0.723	0.051	0.105

Note. Stress = parenting stress; PD = parental distress; PCDI = parent–child dysfunctional interaction; DC = difficult child; Neg = negatively phrased items; Pos = positively phrased items; EV = explained variance.

presents the mean primary factor loadings on the general factor (stress), group factors (PD, PCDI, and DC), method factors of negatively and positively phrased items, and explained variance accounted for by a trait (stress) and method factors for negatively and positively phrased items. Parents of children with autism had a higher average method factor loading (0.099) than those of children with other disabilities (0.025), which coincides with its greater improvements in model fit when the method factor was added.

The general factor (PSI: parenting stress) and the group factor (PD, PCDI, and DC) are clearly represented by the loadings across all samples except for PCDI and DC in autism. PCDI was not well defined by high factor loadings (M = 0.024 for ESEM bifactor; M = 0.019 for ESEM bifactor with negatively phrased items; M = −0.001 for ESEM bifactor with both negatively and positively phrased items) in parents of children with autism. For example, when a method factor of negatively phrased items was added in the ESEM bifactor model, PCDI was not well defined because PSI13 and PSI19-22 had negative loadings. DC also had items with negative factor loadings such as PSI 26, 30, 31, 33, and 34. As shown in the table, explained variance due to method factors was not substantial in each sample, ranging from 0.056 to 0.060. Explained variance due to trait varied from 0.588 to 0.663.

11.4. Applications of SEM-Based Congeneric Bifactor GT Models

Employing generalizability theory structural equation modeling frameworks, the subsequent analyses focused on the partitioning of variances and method factors for CFA congeneric bifactor models with added method factors of negatively and positively phrased items based on all parents, parents of children with autism, and parents of children with other disabilities.

11.4.1. Model Fit

The results in

Table 8. CFA congeneric bifactor model fit statistics.

All	Parameters	Chi-Square	df	CFI	TLI	RMSEA
Bifactor	144.000	653.533	558.000	0.897	0.883	0.042
Bifactor with negative	157.000	628.576	545.000	0.910	0.896	0.040
Bifactor with negative and positive	180.000	602.544	522.000	0.913	0.895	0.040
Difference				0.016	0.012	−0.002
Autism	Parameters	Chi-square	df	CFI	TLI	RMSEA
Bifactor	144	595.278	558.000	0.839	0.818	0.040
Bifactor with negative	157	577.052	545.000	0.862	0.840	0.038
Bifactor with negative and positive	180	543.225	522.000	0.908	0.890	0.031
Difference				0.069	0.071	−0.009
Other	Parameters	Chi-square	df	CFI	TLI	RMSEA
Bifactor	144	621.011	558.000	0.905	0.892	0.046
Bifactor with negative	157	604.036	545.000	0.911	0.897	0.045
Bifactor with negative and positive	180	566.622	522.000	0.932	0.918	0.040
Difference				0.028	0.026	−0.006

Note. df = degree of freedom; CFA = confirmatory factor analysis; WLSMV = weighted least squares with mean and variance adjusted; CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; Difference = difference in model fits between a model without negatively and positively phrased items and a model with both negatively and positively phrased items.

reveal that control for item wording effects provided better fits. Adding method factors of both negatively and positively phrased items yielded salient improvements in CFIs by 0.069, TLIs by 0.071, and RMSEAs by −0.009 for parents of children with autism. The bifactor model with both negatively and positively phrased items for parents of children with other disabilities provided the best model fits for CFI (0.932) and TLI (0.918) while the corresponding model for parents of children with autism produced the greatest model fit in terms of RMSEA (0.031).

11.4.2. Partitioning of Construct, Item-Level Method Factors, and Measurement Error Variance for the CFA Congeneric Bifactor GT Model

The SEM-based GT models allow for individual item-level partitioning and thus determine proportions of observed score variance caused by item wording effects and multiple sources of measurement error at both composite and item levels.

Table 9. The partitioning of variance for the congeneric bifactor model with two method factors of both negatively and positively phrased items based on all parents (N = 96).

All	Stress as a Trait	General	Group	Neg	Pos	Error
Total	0.943	0.810	0.132	0.009	0.002	0.047
Item 1	0.143	0.005	0.138	0.233		0.625
Item 2	0.238	0.050	0.188	0.319		0.443
Item 3	0.394	0.022	0.372		0.001	0.605
Item 4	0.527	0.167	0.360	0.019		0.454
Item 5	0.594	0.174	0.420	0.008		0.398
Item 6	0.408	0.240	0.168	0.006		0.587
Item 7	0.562	0.334	0.228		0.008	0.430
Item 8	0.622	0.457	0.165		0.023	0.355
Item 9	0.630	0.458	0.172		0.006	0.364
Item 10	0.627	0.422	0.205	0.001		0.372
Item 11	0.610	0.376	0.234	0.003		0.387
Item 12	0.728	0.368	0.361	0.000		0.271
Item 13	0.740	0.738	0.002	0.000		0.260
Item 14	0.840	0.793	0.047	0.056		0.104
Item 15	0.511	0.510	0.001	0.077		0.411
Item 16	0.425	0.385	0.039	0.148		0.427
Item 17	0.378	0.367	0.011		0.045	0.577
Item 18	0.227	0.081	0.146	0.094		0.679
Item 19	0.451	0.242	0.209	0.032		0.517
Item 20	0.440	0.164	0.276	0.156		0.404
Item 21	0.537	0.098	0.438		0.141	0.322
Item 22	0.197	0.178	0.019		0.035	0.768
Item 23	0.582	0.563	0.019		0.052	0.366
Item 24	0.343	0.343	0.000		0.088	0.569
Item 25	0.349	0.177	0.172		0.106	0.545
Item 26	0.553	0.457	0.096		0.007	0.441
Item 27	0.481	0.218	0.263		0.002	0.517
Item 28	0.734	0.231	0.503		0.019	0.247
Item 29	0.341	0.068	0.273		0.163	0.495
Item 30	0.501	0.211	0.290		0.026	0.474
Item 31	0.162	0.064	0.098		0.089	0.749
Item 32	0.067	0.036	0.031		0.211	0.722
Item 33	0.513	0.437	0.076		0.010	0.477
Item 34	0.428	0.324	0.104		0.000	0.572
Item 35	0.289	0.287	0.002	0.030		0.681
Item 36	0.105	0.104	0.001		0.000	0.895
Mean	0.452	0.282	0.170	0.074	0.052	0.486

,

Table 10. The partitioning of variance for the congeneric bifactor model with two method factors of both negatively and positively phrased items based on parents of children with autism (N = 42).

Autism	Stress as a Trait	General	Group	Neg	Pos	Error
Total	0.888	0.759	0.129	0.025	0.024	0.064
Item 1	0.137	0.005	0.133	0.233		0.629
Item 2	0.471	0.113	0.358	0.189		0.341
Item 3	0.353	0.024	0.329		0.048	0.599
Item 4	0.271	0.057	0.214	0.154		0.574
Item 5	0.352	0.041	0.311	0.216		0.432
Item 6	0.609	0.550	0.059	0.001		0.389
Item 7	0.695	0.485	0.210		0.058	0.247
Item 8	0.431	0.366	0.065		0.091	0.479
Item 9	0.584	0.117	0.467		0.002	0.413
Item 10	0.618	0.380	0.239	0.005		0.377
Item 11	0.682	0.328	0.354	0.054		0.264
Item 12	0.794	0.375	0.419	0.020		0.186
Item 13	0.738	0.734	0.003		0.024	0.238
Item 14	0.786	0.645	0.141		0.074	0.140
Item 15	0.440	0.440	0.000		0.038	0.523
Item 16	0.416	0.311	0.105	0.101		0.483
Item 17	0.592	0.541	0.051		0.002	0.406
Item 18	0.174	0.016	0.158	0.155		0.671
Item 19	0.468	0.182	0.285	0.004		0.528
Item 20	0.177	0.130	0.047	0.276		0.547
Item 21	0.368	0.121	0.247		0.070	0.562
Item 22	0.144	0.143	0.001		0.001	0.855
Item 23	0.364	0.350	0.014		0.003	0.633
Item 24	0.592	0.562	0.030		0.116	0.292
Item 25	0.057	0.013	0.045		0.219	0.724
Item 26	0.556	0.556	0.000		0.016	0.428
Item 27	0.268	0.233	0.035		0.049	0.683
Item 28	0.325	0.325	0.001		0.043	0.632
Item 29	0.104	0.077	0.028		0.743	0.153
Item 30	0.330	0.166	0.163		0.186	0.484
Item 31	0.049	0.049	0.000		0.013	0.937
Item 32	0.303	0.007	0.296		0.184	0.513
Item 33	0.202	0.187	0.016		0.140	0.658
Item 34	0.077	0.046	0.031		0.075	0.848
Item 35	0.761	0.344	0.417	0.000		0.238
Item 36	0.114	0.010	0.104		0.004	0.883
Mean	0.400	0.251	0.149	0.108	0.096	0.500

and

Table 11. The partitioning of variance for the congeneric bifactor model with two method factors of both negatively and positively phrased items based on parents of children with other disabilities (N = 54).

Other	Stress as a Trait	General	Group	Neg	Pos	Error
Total	0.918	0.829	0.089	0.008	0.039	0.035
Item 1	0.077	0.001	0.076	0.381		0.542
Item 2	0.121	0.037	0.084	0.662		0.218
Item 3	0.383	0.009	0.374		0.006	0.611
Item 4	0.674	0.178	0.496	0.000		0.326
Item 5	0.771	0.206	0.565	0.008		0.221
Item 6	0.383	0.109	0.274	0.006		0.611
Item 7	0.517	0.244	0.273		0.003	0.480
Item 8	0.737	0.527	0.210		0.011	0.251
Item 9	0.746	0.648	0.098		0.004	0.250
Item 10	0.645	0.452	0.192	0.003		0.352
Item 11	0.606	0.397	0.210	0.011		0.383
Item 12	0.691	0.374	0.317	0.038		0.271
Item 13	0.720	0.720	0.000		0.001	0.279
Item 14	0.929	0.916	0.013		0.036	0.035
Item 15	0.538	0.537	0.001		0.029	0.433
Item 16	0.374	0.374	0.000	0.077		0.549
Item 17	0.393	0.351	0.042		0.035	0.571
Item 18	0.333	0.116	0.217	0.022		0.645
Item 19	0.403	0.276	0.127	0.009		0.589
Item 20	0.508	0.198	0.310	0.089		0.403
Item 21	0.904	0.064	0.839		0.075	0.021
Item 22	0.165	0.144	0.021		0.107	0.728
Item 23	0.672	0.651	0.020		0.010	0.318
Item 24	0.270	0.264	0.007		0.000	0.729
Item 25	0.461	0.397	0.065		0.182	0.356
Item 26	0.430	0.428	0.002		0.156	0.414
Item 27	0.260	0.260	0.000		0.261	0.479
Item 28	0.331	0.330	0.001		0.274	0.394
Item 29	0.231	0.174	0.057		0.278	0.490
Item 30	0.386	0.368	0.019		0.201	0.413
Item 31	0.081	0.078	0.003		0.457	0.463
Item 32	0.095	0.085	0.011		0.128	0.777
Item 33	0.717	0.524	0.193		0.108	0.176
Item 34	0.562	0.489	0.073		0.122	0.316
Item 35	0.620	0.219	0.401	0.045		0.335
Item 36	0.618	0.246	0.372		0.003	0.379
Mean	0.482	0.316	0.166	0.104	0.108	0.411

summarize the partitioning of stress as a trait, method factors, and total error for CFA bifactor models with negatively and positively phrased items at both composite and item levels. Indices in Table 9, Table 10 and Table 11 reveal that method effects of negative wording alone account for proportions of variance ranging from 0.000 to 0.662 at the item level and from 0.008 to 0.025 at the total score level. Results further indicate that the method factor for positively worded items accounts for large proportions of total score variance (0.002 to 0.039) and item-level score variance (0.000 to 0.743) beyond what was accounted for by the negative wording method factors within those models.

Overall, mean proportions of negative wording method effects are slightly greater in the parents of children with autism (M = 0.108) than in those of children with other disabilities (M = 0.104). The mean proportions of positive wording method effects are larger in the parents of children with other disabilities (M = 0.108) than in those of children with autism (M = 0.096).

For all parents, items 1 (0.233) and 2 (0.319) display considerable negative wording method effects while item 32 (0.211) shows the strongest positive wording effect. For parents of autistic children, item 29 exhibits the strongest positive item wording effect (0.743). Item 20 (0.276) shows the greatest negative wording method effects followed by items 1 (0.233) and 5 (0.216). For parents of children with other disabilities, item 2 displays the highest negative item wording effect (0.662), whereas item 31 shows the greatest positive wording method effects (0.457). To mitigate such effects, those items would be the ideal candidates for replacement.

12. Discussion and Implications for Research and Practice

Despite the widespread usage of the PSI-SF, there has been little research into its psychometric properties, factor structure, and item wording effects in Latinx parents of children with IDD. This study evaluated the psychometric properties, the factor structure (structural validity), and item wording effects of the PSI-SF on model fits, method factors, and measurement error based on Latinx parents of children with autism and other disabilities in the United States. Additionally, this study identified items with great proportions of negative and positive item wording effects, which may be culturally insensitive, and suggested that these items need to be refined to appropriately reflect the stress levels faced by Latinx parents of children with IDD. Important implications of the present study for research and practice and related suggestions for future research are summarized below.

12.1. Psychometric Properties of the PSI-SF

Representing the Structure of the Parental Stress and Item Wording Effects

The results of this study demonstrated that ESEM bifactor models with positively and negatively phrased items effectively captured the conceptual underpinnings of the PSI-SF structure. Hence, this study indicates that the bifactor structure is the best depiction of the PSI-SF theoretical framework, in contrast to the majority of earlier studies that supported the three factor structure of the PSI-SF (e.g., [5,6,7,8,9]). In the bifactor representation, parental stress (PSI) is a general factor, and parental distress (PD), parent–child dysfunctional interaction (PCDI), and difficult child (DC) are group factors (refer to Figure 3). In this regard, this work adds to the evidence of the PSI-SF’s psychometric qualities while also making unique proposals for the PSI-SF’s bifactor structure.

Furthermore, this study found that all models with method factors for negatively and positively phrased items yielded better model fits than corresponding models without such a factor. Method effects were strongest for the sample of parents of children with autism. These findings coincide well with those from previous studies [34,52,53,54,55] that highlight improvements in fit after controlling for negatively and positively phrased items. Method effects are systematic patterns in responding to questions unrelated to the assessed constructs. Research findings may be interpreted incorrectly if method effects are not taken into consideration [15,18].

Since no studies evaluated the item wording effects of the PSI-SF, the present analyses significantly extend these prior studies of the PSI-SF (e.g., [3,4,5,6,7,8,9]) by demonstrating that much greater improvements in model fit can be achieved by including negatively and positively phrased items for one factor, three factor, and bifactor models, and underscoring the need to control for such construct-irrelevant factors and evaluate their potential effects whenever a combination of positively and negatively phrased items are included in assessment measures [15,37,56].

12.2. Evaluation and Recommendations for Improving the PSI-SF Items

The results from the SEM-based GT analyses allow for the identification of the items with high proportions of method factors.

Table 12. Items with high proportions of method factors for all parents.

Items of the PSI-SF	Negative	Positive
Item 1: I often have the feeling that I cannot handle…	0.233
Item 2: I find myself giving up more of my life to meet my children’s needs…	0.319
Item 16: Sometimes I feel my child doesn’t like me…	0.148
Item 20: My child is not able to do…	0.156
Item 21: It takes a long time and it is very hard…		0.141
Item 29: My child reacts very strongly when something…		0.163
Item 32: I have found that getting my child to do…		0.211

,

Table 13. Items with high proportions of method factors for parents of children with autism.

Items of the PSI-SF	Negative	Positive
Item 1: I often have the feeling that I cannot handle…	0.233
Item 2: I find myself giving up more of my life to meet my children’s needs…	0.189
Item 4: Since having this child, I have been unable…	0.154
Item 5: Since having a child, I feel that I am almost never…	0.216
Item 18: My child doesn’t seem to learn…	0.155
Item 20: My child is not able to do…	0.276
Item 25: My child seems to cry or fuss…		0.219
Item 29: My child reacts very strongly when something…		0.743
Item 30: When playing, my child doesn’t often…		0.186
Item 33: Think carefully and count the number of things…		0.140

and

Table 14. Items with high proportions of method factors for parents of children with other disabilities.

Items of the PSI-SF	Negative	Positive
Item 1: I often have the feeling that I cannot handle…	0.381
Item 2: I find myself giving up more of my life to meet my children’s needs…	0.662
Item 25: My child seems to cry or fuss…		0.182
Item 26: My child generally wakes up…		0.156
Item 27: I feel that my child is very moody…		0.261
Item 28: Compared to the average child, my child has a great…		0.274
Item 29: My child reacts very strongly when something…		0.278
Item 30: When playing, my child doesn’t often…		0.201
Item 31: My child’s sleeping or eating schedule was much…		0.457

summarize the items with greater proportions of negative and positive item wording effects. For all parents, items 1 (0.233) and 2 (0.319) display significant negative wording method effects while item 32 (0.211) has the greatest positive wording effect. For parents of autistic children, item 29 exhibits the strongest positive item wording effect (0.743). Item 20 (0.276) shows the largest negative wording method effects followed by items 1 (0.233) and 5 (0.216). For parents of children with other disabilities, item 2 presents the highest negative item wording effect (0.662), whereas item 31 establishes the greatest positive wording method effects (0.457). Across all samples, the common items that exhibit the high proportions of negative item wording effects are items 1 and 2 while the item that consistently displays the large proportions of positive item wording effects is item 29. To alleviate such effects, those items would need to be revised.

Individual item-level partitioning can help choose items that enhance score consistency for the intended purpose of an assessment while eliminating those that do not. For example, if we viewed parental stress as a trait as a relevant construct while the wording effects of negatively and positively phrased items as an irrelevant construct are expected to be taken into account, and considered items with low proportions of method factor variance and low proportions of error to best serve that aim, then, items 1, 2, 29 and 31 would be least effective and would require more investigation and refinement in this regard.

The parents’ inability to comprehend the contents of the items can be attributed to the phrasing of the items, which may be culturally insensitive. These results imply that several items of the PSI-SF may not be culturally sensitive enough to accurately represent the stress levels experienced by these Latinx parents, given that the majority of participants are low-income and lack a college degree. When the second author gave the PSI-SF survey orally, the parents repeatedly questioned her, asking, “What does this question mean?” This study suggests that items with large proportions of the item wording effects that may not be culturally responsive need to be refined to appropriately reflect the stress levels faced by Latinx parents of children with IDD. Determining the effects of PSI-SF item phrasing on Spanish-speaking Latinx parents requires further investigation and testing by researchers.

Other areas for future inquiry include the applications of Bayesian SEM analyses and Bayesian SEM-based GT analyses with different informative priors, the multivariate GT analysis, the use of alternative rotation methods in ESEM analyses [19,57,58,59], and the use of the hierarchical model for theoretical inquiries and practical applications.

13. Limitations

In conducting this study, several limitations are identified, which may affect the interpretation of the results. Acknowledging these limitations is crucial for contextualizing our findings and informing future research. First, the small sample size used in this study cautions against drawing broad statistical conclusions. Unfortunately, there is no agreement in the literature about what constitutes an adequate sample size for SEM, despite the fact that it is a crucial decision in SEM. There is evidence to suggest that even with relatively small sample sizes, basic SEM models could be meaningfully tested [60,61,62]; however, typically, the sample size of 100–150 is thought to be the minimum sample size needed to perform SEM [63,64,65]. Although the power analysis revealed that the sample size used for this study was adequate for the statistical analyses carried out, a sample size of 42 may not be sufficient because we need at least 50 samples to obtain the power of 0.80. This may increase the possibility of Type II error. Nevertheless, the findings provide significant ramifications. Despite the small sample size, this study underscores the benefits of this investigation, which established a novel bifactor structure for the theoretical framework underlying the PSI-SF, offering vital insights into the scant literature on PSI-SF for Latinx families of children with IDD. These results have the potential to inform future research and intervention programs tailored to help measure parental stress more accurately and provide support and empowerment for Latinx parents of children with IDD.

Second, the overrepresentation of women in our sample is most likely due to the typical caregiving tasks that women take on, particularly in Latinx households with children with IDD. Women, as mothers, may be most familiar with their children’s IDD needs, so their care might be regarded as an asset. However, gender-specific sampling may misrepresent these parents’ stress levels. Thus, the findings should be interpreted with caution. Future studies may need to determine a strategy for actively recruiting male participants.

14. Conclusions

This study provided crucial insights into the paucity of research and contributed to advancing theory in developing a new bifactor structure of the theoretical framework on PSI-SF, identifying potentially culturally insensitive items, and suggesting the revision of the PSI-SF for 96 Latinx parents of children with IDD by systematically applying CFA and ESEM factor analyses and SEM-based GT procedures. This work highlights the necessity of controlling for these construct-irrelevant method factors and assessing their potential effects whenever a combination of positively and negatively phrased items are contained in assessment measures [15,37,56]. Applying the CFA and ESEM factor analytic techniques and SEM-based GT procedures significantly extended evidence of the psychometric quality of the PSI-SF scores and provided guidelines for future research into other multidimensional measures of psychological constructs. In addition, researchers in future studies might replicate the present modeling procedures with larger sample sizes to determine the extent to which they generalize to other measures.

This study recommended that the items with high proportions of method factors, which may likely be culturally insensitive to Latinx parents, be examined further and refined. The present analyses will assist in identifying items that are insensitive to the cultural norms of Latinx parents, modify the measure as needed to ensure sensitivity to the specific cultural values, and develop a culturally responsive PSI-SF assessment tool that lessens or eliminates the effects of item wording and appropriately and accurately addresses the stress experienced by Latinx parents of children with IDD. Through the FIRME program, a parent advocacy initiative, practitioners will be better able to support and empower Latinx parents of children with IDD and better meet their needs with the use of accurately assessed PSI-SF scores. This study will also guide researchers to create and alter other psychological tools that can adapt to different cultural contexts and meet the needs of diverse ethnic and cultural groups.