Differential item functioning of the revised Multigroup Ethnic Identity Measure (MEIM-R) in racially and income diverse youth with type 1 diabetes

Abstract

Objective

Racially minoritized youth with T1D are made vulnerable to disproportionately adverse health outcomes compared to White peers due to enduring systems of oppression. Thus, understanding modifiable psychosocial factors associated with diabetes-related outcomes in racially minoritized youth may help to buffer deleterious effects of racism. One factor meriting exploration is racial-ethnic identity. There is currently limited research on measures fit to assess ethnic identity in youth with chronic illnesses. This study’s purpose is to examine the factor structure, reliability, and validity of the revised Multigroup Ethnic Identity Measure (MEIM-R) in a racially- and income-diverse sample of youth with T1D across sociodemographic and illness-related proxies for one’s positionality in oppressive systems.

Method

As part of a larger study examining resilience, 142 youth with T1D ages 12–18 (M_age = 14.66, SD_age = 1.62, 55.6% Black/African-American, 44.4% White) completed the MEIM-R and various psychosocial measures. HbA1c levels and illness duration were extracted from medical records and caregivers reported income information. Confirmatory factor analyses compared the structural validity of competing MEIM-R models, and uniform and non-uniform differential item functioning (DIF) was explored across sociodemographic and illness-related factors.

Results

While a bifactor structure was supported, the MEIM-R was found to exhibit DIF by race and gender on multiple MEIM-R items and did not demonstrate linear bivariate relations with other psychosocial factors.

Conclusions

Since different MEIM-R item response patterns were observed across racial/ethnic and gender groups, caution is warranted in using this measure in racially and gender diverse youth with T1D.

Type 1 diabetes (T1D) is a demanding chronic illness requiring a stringent self-management routine impacting over 190,000 individuals under the age of 19 in the United States. Estimates of incidence have shown an alarming increase among Black youth ages 0–4 and White youth ages 10–14 (Lipman et al., 2013; Mayer-Davis et al., 2017; Mobasseri et al., 2020). T1D demands yield significant emotional and financial consequences for families. Maintaining a consistent treatment routine requires constant access to many expensive medications and diabetes supplies. Medical costs of T1D have been estimated to be approximately $16,000 per person annually, on average 2.3 times more for individuals with diabetes than those without (American Diabetes Association, 2018).

 

Diabetes care and complications

There is a complex interplay of factors contributing to thriving and/or languishing in families of youth living with T1D. Treatment focuses on maintaining blood glucose within a specific range for as long as possible to avoid compounding risk for future health complications (e.g., neuropathy, retinopathy, cardiovascular disease) (DiMeglio et al., 2018). It is recommended to have hemoglobin A1c (HbA1c, or average blood glucose levels over the past two to three months) values less than 7.0%–7.5% (which is approximately 154–169 mg/dl on average) and for blood glucose levels to be between 70–180 mg/dl 70% of the time (American Diabetes Association Professional Practice Committee, 2022). Youth work with their healthcare team to coordinate the administration of exogeneous insulin doses via multiple daily injections or insulin pump based on varying dietary intake, activity, blood glucose levels, and other factors.

Disparities in diabetes: intersecting systems of inequality

Complicating the care of diabetes further are various interconnected systems of oppression/privilege impacting individuals and communities (Velez & Spencer, 2018). The Phenomenological Variance of Ecological Systems Theory (PVEST) posits that individuals from systematically marginalized communities, particularly Black communities, are exposed to high-risk contexts yielding greater challenges with normative developmental tasks and increased likelihood of adversity (Spencer et al., 2006). Interactions among other systems of identity-based marginalization (e.g., gender, educational attainment, economic, health/ability) further restrict access to protective resources (Velez & Spencer, 2018). Race exists as a socially-defined construct within our society and is based on physical and cultural differences from the oppressors, resulting in systematic oppression, abuse, and health disparities (Velez & Spencer, 2018).

Within the context of diabetes, disparities in our healthcare system are the intended result of existing intersectional systems of inequality (Ogunwole & Golden, 2020). Individuals holding intersecting minoritized identities, such as those with minoritized racial identities and chronic illness, are especially burdened by these systems (Velez & Spencer, 2018). This is evident in youth with T1D identifying as Black, who are made vulnerable to significantly higher HbA1c levels (Keenan et al., 2021; Lipman et al., 2021) and diabetes-related stress (Fegan-Bohm et al., 2020) compared to their peers privileged as white. Youth living with economic marginalization (including “limited financial resources and marginalization related to social class”; Juntunen et al., 2022). And youth who are uninsured have evidenced higher HbA1c levels even when controlling for factors such as race and caregiver education (Petitti et al., 2009).

Racial-ethnic identity

Consequently, it is imperative to explore and foster protective factors that can buffer such effects. One construct meriting exploration is racial-ethnic identity (REI), broadly defined as a categorization of ancestry or self-identification with one or more cultural groups (Clarke et al., 2008). Alternatively, some may understand REI as a form of social identity encompassing one’s awareness, knowledge, and assessment of group membership (Tafjel, 1981). This subjective approach can make it difficult to conceptualize uniformly. This paper understands REI as described by Umaña-Taylor et al. (2014): a “multidimensional, psychological construct [reflecting] the beliefs and attitudes that individuals have about their ethnic-racial group memberships, as well as the processes by which these beliefs and attitudes change over time.”

Present literature highlights the importance of conceptualizing REI as a protective factor reducing the risk of adverse psychosocial and health-related outcomes. One study found higher levels of ethnic identity buffered the adverse effects of stress in a large sample of Filipino-Americans (Mossakowski, 2003). In a sample of Latino youth, ethnic identity moderated self-esteem with subsequent moderation on depressive levels (Umaña-Taylor & Updegraff, 2007). Additionally, higher levels of reported racial identity were linearly associated with lower pain and higher health-related quality of life in youth with sickle cell disease (Lim et al., 2012). Higher levels of ethnic identity were also strongly associated with greater improvement in HbA1c in Black/African-American and Latino adults with type 2 diabetes over time (Murayama et al., 2017).

Unfortunately, no known studies have measured REI nor its measurement accuracy across health-related outcomes and proxies for larger systems of oppression in youth with T1D. Given promising evidence that supports consideration of REI as a protective factor in other clinical populations, it is important to ensure REI is being accurately measured in youth with T1D, especially among diverse youth facing compounding systemic burdens associated with intersectional identities (Velez & Spencer, 2018). Integration of REI into health equity work may be helpful in determining whether REI plays an important role in the delivery and outcomes of patient-centered diabetes care considerate of youths’ cultural differences.

Measuring ethnic identity

Many published measures assess REI factors for specific ethnic groups rather than broadly assessing relevant factors for a variety of ethnic groups. However, the Multigroup Ethnic Identity Measure (MEIM) fills this gap (Phinney, 1992). Phinney (1992) employed a variety of shared factors to understand REI across groups, including self-identified ethnicity, behaviors and practices, affirmation and belonging, exploration and commitment, and attitudes toward other ethnic groups. The original MEIM consists of 20 items across three factors: positive ethnic attitudes/sense of belonging, ethnic identity achievement, and ethnic behaviors or practices (Phinney, 1992). Internal validation of this original measure found one single ethnic identity factor. Psychometric properties of the MEIM and its variations continue to be evaluated across a variety of populations with mixed factor structure outcomes, including a bifactor model, indicating a need for further examination of the MEIM-R factor structure across populations (Yap et al., 2014).

A six-item revision of the MEIM (MEIM-R) was created by Phinney and Ong (2007) consisting of two factors: exploration and commitment. The MEIM-R was administered to 241 racially and ethnically diverse university students. Confirmatory factor analyses (CFAs) of the new two-factor model indicated high a correlation between factors (r = 0.74); however, the extent to which these items tapped into the same construct (e.g., measurement invariance (MI)) across racial and ethnic groups was not assessed at that time. Adequate measurement invariance has been found across racial, ethnic, and age groups (Brown et al., 2014). It is important to continue to assess potential bias in a variety of populations, such as youth with chronic health conditions, in order to more aptly explore the role of REI in psychosocial and health-related outcomes.

Sample size limitations are common in measure validation studies in pediatric health samples. Dichotomization of non-categorical variables (e.g. income, HbA1c) in MI can reduce power, introduce bias, and result in uneven or inadequate sample sizes, making it difficult to draw conclusions (Bauer, 2017). Differential item functioning (DIF) mitigates this issue via item response theory to examine whether combinations of continuous variables and distinct groups (e.g., sociodemographic, illness-related) influence different probabilities of response patterns while controlling for latent scores. Because the variables are not unnecessarily further split into binary categories (as in MI), power is increased. Further, moderate-to-large main effect (i.e., intercept) DIF has been reliably found in sample sizes as small as 50–100 when using logistic regression with likelihood ratio testing (LR-LRT; Belzak, 2020).

Present DIF studies with the MEIM-R are limited; however, one study indicated that African-American adults report higher scores on the MEIM-R than European-American counterparts, with presence of DIF on item 1 (Chakawa et al., 2015). While the affirmation/belonging subscale of the original MEIM was associated with lower depressive symptomatology and higher health-related quality of life in a study of youth with obesity (Lim et al., 2016), there has not been a peer-reviewed assessment of bias with the original, nor revised, MEIM in a pediatric health context (Choquette, 2023). This further highlights the importance of exploring potential DIF of this measure among a diverse sample of youth with T1D.

The present study seeks to (Objective: 1) evaluate the psychometric properties of the MEIM-R in youth with T1D, (Objective: 2) assess potential DIF across sociodemographic factors which may be used as proxies for oppressive systems based on age, gender, race, and income; illness-related factors including HbA1c and illness duration will also be examined as potential indicators of differentially-allocated financial and health-related resources. Lastly, this study seeks to (Objective: 3) examine how the MEIM-R may relate to other psychosocial factors associated with diabetes-related outcomes, including familism, diabetes-specific family conflict, quality of life, stress, and psychological flexibility. It is hypothesized that the 2-factor MEIM-R will demonstrate adequate psychometric fit in youth with T1D; however, recommendations by Rhemtulla et al. (2012) highlight that measures with <4 categories should be treated as ordinal and those with >7 categories be treated as continuous. Given the MEIM-R hosts 5 category responses as well as recent recommendations highlighting the importance of exploring comparison models (Kline, 2023), it is important to compare various models as both ordinal and continuous. It is also expected that the presence of DIF will be evident by racial groups. Since past research has found that REI is predictive of positive psychosocial outcomes, it is expected that REI will demonstrate small positive associated with familism, diabetes-specific psychological flexibility, and health-related quality of life, while negatively associated with diabetes-specific family conflict and stress.

Methods

Participants and procedures

Participants were 142 youth ages 12–18 and their caregivers who were recruited from a pediatric endocrinology outpatient clinic in Memphis, Tennessee as part of the Predicting Resiliency in Youth with Type 1 Diabetes study (PRYDE), a longitudinal study examining potential psychosocial predictors of treatment adherence, glycemic control, and quality-of-life in youth with T1D. Data were collected at three separate timepoints (baseline, 6-month, and 12-month follow-ups). Study procedures were approved by Institutional Review Boards at Le Bonheur Children’s Hospital and the University of Memphis prior to data collection. Study staff recruited eligible youth during their standard diabetes care visits. Caregivers provided written consent on behalf of themselves and their children, and youth ages 14 and above provided additional written assent.

Participating youth met the following eligibility criteria: (1) diagnosed with T1D for 6+ months, (2) receiving (or intending to receive) care at the Le Bonheur Children’s Hospital endocrinology outpatient clinic for 1+ year, and (3) English-speaking. Exclusion criteria included (1) diagnosis of a severe developmental disability, (2) guardian unable to provide consent, or (3) if the youth respondent was pregnant. Questionnaires were either completed in-person or sent and returned via mail.

Of 220 families that were approached to participate in PRYDE, 195 provided informed consent. PRYDE included 183 youth and their caregivers who consented and completed baseline measures. The MEIM-R was added to the PRYDE study approximately 13 months after data collection began. Of the original sample, the present study examines 142 youth who first completed the MEIM-R at either baseline, 6-month, or 12-month follow-up (all three timepoints: N = 11; baseline: N = 39; 6-month: N = 17; 12-month: N = 45; baseline and 6-month: N = 14; baseline and 12-month: N = 0; 6-month and 12-month: N = 17). Descriptive information on the sample can be found in Table 1. The present study builds upon a previous project from the PRYDE data set (Choquette, 2023).

Table 1.

Descriptives and correlations.

Factor	1	2	3	4	5	6	7	8	9	10	11
1. HbA1c	–
2. Income	−0.33*	–
3. Race (Black/AA)	0.51*	−0.37*	–
4. Gender (male)	−0.06	0.18*	−0.10	–
5. Illness duration	0.25*	−0.17*	0.41*	−0.28*	–
6. Age (years)	−0.11	−0.03	0.07	−0.05	0.13	–
7. DSQ	0.30*	−0.10	0.20*	−0.18*	0.16*	−0.09	–
8. FS	−0.14	0.10	0.07	0.13	0.01	−0.05	−0.12	–
9. DFCS	0.20*	−0.27*	0.27*	−0.06	0.12	−0.03	0.44*	−0.12	–
10. DAAS	0.22*	−0.08	0.19	0.01	0.16*	−0.05	0.57*	0.59	−0.56*	–
11. MEIM-R (latent)	−0.25	0.16	0.24*	0.04	−0.07	0.12	0.09	0.08	−0.08	−0.10	–
Mean	10.51	$20K–25K	55.6	52.5	4.96	14.66	1.16	4.07	1.76	4.76	19.58
SD	2.14				0.32	1.62	0.63	0.98	0.64	1.98	5.39

Note. HbA1c = % hemoglobin A1c at respective timepoint, AA = African-American, DSQ = Diabetes Stress Questionnaire, FS = Familism Scale, DFCS = Diabetes Family Conflict Scale, DAAS = Diabetes Acceptance and Action Scale, MEIM-R = Multigroup Ethnic Identity Measure—Revised, SD = standard deviation. For gender, female = 1 and male = 0; For race, 0 = White and 1 = Black/African-American; % Frequency presented for binary variables. Median income range presented. N = 1 identified as Hispanic/Latino and White.

^* p < .05 or better.

Measures

Demographics

Youth reported demographic information including race, ethnicity, gender, age, and illness duration. Caregivers reported household income. Illness duration was calculated by obtaining date of diagnosis from participant medical records. HbA1c levels were also obtained from participant medical records at baseline, 6-, and 12-month timepoints. This study reports HbA1c levels at point of MEIM-R completion.

Racial-ethnic identity

Youth completed the revised Multigroup Ethnic Identity Measure (MEIM-R; Phinney & Ong, 2007). Participants responded to items on a 5-point Likert-style scale ranging from 1 (“strongly disagree”) to 5 (“strongly agree”). Cronbach’s alpha for the MEIM-R in this sample was .87.

Quality of life

Youth completed The Pediatric Quality of Life—Diabetes Module (PedsQL-DM) 3.0 to assess diabetes-related quality of life. The PedsQL-DM is a self-report measure prompting how problematic or difficult various domains of diabetes care have been for the respondent in the past month (Varni et al., 2003). Participants rated the difficulty on a scale of 0 (“never”) to 4 (“always”). Items are summed and transformed to a scale of 0–100, with higher scores indicating higher diabetes-related quality of life. Cronbach’s alpha for the PedsQL-DM3.0 in this sample was .88.

Familism

Youth completed a 5-item measure assessing familism utilizing items from two different scales consistent with previous literature (Villarreal et al., 2005) assessing perceptions of family importance and reflecting ideological beliefs about family. Youth rated how much they agree with statements about their family on a scale of 1 (“strongly disagree”) to 5 (“strongly agree”), with higher scores reflecting higher degree of perceived familism. Cronbach’s alpha for the familism scale in this sample was .93.

Diabetes-specific family conflict

Youth completed the Diabetes Family Conflict Scale (DFCS) to assess the impact of diabetes management on the caregiver–child relationship over the past month (Hood et al., 2007). The DFCS is a 19-item scale yielding subscales representing direct diabetes-related conflict (“I have argued with my parents about remembering to give shots”) and indirect diabetes-related conflict (“I have argued with my parents about what to eat when away from home”). Higher scores indicate higher reported conflict within the family. Cronbach’s alpha in this sample was .96.

Diabetes-specific psychological flexibility

The Diabetes Acceptance and Action Scale (DAAS-22) is a 22-item scale adapted to measure diabetes-specific psychological flexibility, a coping strategy involving engaging in values-consistent action associated with more optimal glycemic control (Berlin et al., 2020; Greco & Hayes, 2008). Youth utilized a 5-point Likert scale (0=”never true”, 4=”always true”) to reflect their beliefs about diabetes impairing their ability to engage with their values, engaging in diabetes-related avoidance behaviors, and diabetes-related cognitive fusion. To obtain the total score, items were reverse-scored and averaged. Cronbach’s alpha in this sample was .92.

Diabetes-specific stress

The Diabetes Stress Questionnaire (DSQ) is a 65-item self-report measure with eight subscales (Distress-Worry, Peer Stress, Averse Interpersonal Effects, Parental Stress, Hyperglycemia, Self-Care Regimen, Diet, and Hypoglycemia) yielding a total diabetes-related stress score (Delamater et al., 2013). Items are rated on a 4-point Likert scale (not at all, a little, pretty much, very much). Cronbach’s alpha in this sample was .97.

Analytic plan

Model fit

Given the range of potential factor structures of the MEIM and the present revised version as well as current recommendations to assess model fit, various CFAs assessed which factor structure best fit the MEIM-R among this unique sample of diverse youth with T1D (Objective: 1; Kline, 2023; Rhemtulla et al., 2012). The one-factor structure consisted of all MEIM-R items, whereas the two-factor structure featured the Exploration and Commitment factors. Two bifactor models were also examined (Reise et al., 2010; Figure 1). The first bifactor model included a “general” REI factor plus two “specific” factors (uncorrelated with the general factor). The second bifactor model included one general and one (uncorrelated) specific factor. Analyses were conducted utilizing MPlus Version 8.4. Model structures were compared as both categorical using weighted least mean squares (WLSMV) and continuous utilizing maximum likelihood with robust standard error (MLR). The data were examined for missingness, outliers, normality, and multicollinearity. The data evidenced 27 patterns of missingness across HbA1c, income, race, illness duration, and age, with covariance coverage ranging from 0.727 to 0.972, that were addressed via multiple imputation (100 datasets).

Figure 1.

One-factor, original 2-factor, one general/two specific and one general/one specific bifactor MEIM-R models.

Model fit was determined four recommended indices: chi square, root-mean-square estimates of approximation (RMSEA), confirmatory fit index (CFI), and the standard root mean square residual (SRMR; West et al., 2012). Models with non-significant chi square values are preferred as they indicate well fitting models where no significant differences between the observed and model-implied covariance matrices are found. RMSEA values and 90% confidence interval (CI) reflect how poorly a model fits, with values of 0.05 or below indicating good fit, and values between 0.08 and 0.10 indicating mediocre fit. RMSEA values should be interpreted cautiously with smaller sample sizes. The CFI compares the fit of the model to the fit of a null model, with a good fit being greater than 0.95. The SRMR represents the square root of the difference between residuals in the sample covariance matrix. The hypothesized model should ideally be below 0.08 to be acceptable, though sample sizes smaller than 200 can cause the SRMR to perform poorly.

In potential consideration of a bifactor model structure, further indices warrant consideration for appropriateness of factor structure. Explained common variance (ECV) represents the proportion of shared variance explained by that factor, with ECVS representing the strength of the specific factors in comparison to the explained variance of items loading onto the factor (Rodriguez et al., 2016). Omega is a measure of internal reliability, with all items considered for the general factor. OmegaS represents only the items loading onto the specific factors (Rodriguez et al., 2016). Omega hierarchical (OmegaH) indicates the percent of systematic variance in raw scores that can be attributed to individual differences on the general factor, with higher OmegaH scores (>0.8) reflecting unidimensionality. OmegaHS reflects the proportion of systematic variance of the specific factors after accounting for variability attributed to the general factor (Rodriguez et al., 2016). Relative omega is calculated by dividing OmegaH (or OmegaHS) by Omega (or OmegaS). For the general factor, relative omega indicates the proportion of variance in the multidimensional composite driven by the general factor. For specific factors, this is the proportion of variance not explained by the general factor (Rodriguez et al., 2016). “H” represents construct replicability and is a correlation between the factor and an optimally-weighted composite, with higher values (>0.8) indicating a well-defined latent variable (Rodriguez et al., 2016). Lastly, factor determinacy (FD) represents the correlation between factor scores and the factors, themselves, with values of >0.9 being recommended for use (Gorsuch, 1983).

Differential item functioning

DIF analyses examine patterns of item-level responses among different participant characteristics via regression of the item on the categorical grouping variable and/or continuous “covariates” (Objective: 2). DIF within this sampled explored whether construct-irrelevant differences in item-level responses are a function of respondent race, age, grade, gender, income, HbA1c levels, and illness duration. If bias is present, significant direct effects of these proxies on each item while controlling for latent total scores may be observed at all levels along the latent trait (uniform DIF), or significant latent interactions which reflect whether factor loadings vary as a function of respondent characteristics in which DIF does not occur equally at all points (non-uniform DIF).

Validity

In the absence of a second REI measure, validity evidence can be challenging. As such, the present study examined bivariate correlations of MEIM-R total scores with psychosocial variables associated with diabetes outcomes to explore convergent validity with REI (Objective: 3). Psychosocial factors of interest include familism and diabetes-specific family conflict, stress, psychological flexibility, and health-related quality of life.

Results

Factor analysis

Of the models, a two-factor continuous model evidenced decent fit; however, model fit improved substantially through examination of bifactor models (Table 2). In the bifactor models, the correlation of the general ethnic identity factor to the specific factor(s) was set to 0. This allowed the interpretation of the specific factor(s) to be understood as latent variables representing the shared residual variances among the MEIM-R items unrelated to the general factor. The first continuous bifactor model (one general ethnic identity factor with two specific factors) demonstrated the strongest model fit. The second continuous bifactor model (one general and one specific factor) evidenced similar yet slightly poorer fit criteria; however, it boasted stronger factor loadings on the general factor (Table 3) as well as stronger bifactor indices (see Supplementary Table S1; Objective: 1).

Table 2.

Model fit comparisons.

Model	BIC	χ 2	p	df	Parameters Est. (#)	RMSEA [90% CI]	CFI	SRMR
1-Factor Continuous	2431.528	54.73	<.001	9	18	0.19 [0.14, 0.24]	0.89	0.05
1-Factor Categorical	–	68.85	<.001	9	30	0.22 [0.17, 0.27]	0.94	0.04
2-Factor Continuous	2400.181	2.9	.574	8	17	0.10 [0.04, 0.15]	0.96	0.03
2-Factor Categorical	–	33.76	<.001	8	29	0.15 [0.10, 0.20]	0.98	0.03
Bifactor Continuous (1 general and 2 specific factors)	2412.456	0.913	.633	2	25	0.00 [0.00, 0.132]	1.00	0.007
Bifactor Categorical (1 general and 2 specific factors)	–	996.57	<.001	15	37	0.00 [0.00, 0.16]	1.00	0.01
Bifactor Continuous (1 general and 1 specific factor)	2412.685	6.013	.198	4	23	0.00 [0.00, 0.15]	0.992	0.025

Note. BIC values only used to compare continuous models.

Table 3.

Standardized factor loadings of the general and specific factors.

Item	Item text	General factor	Specific factor
1	I have spent time trying to find out more about my ethnic group, such as its history, traditions, and customs.	0.575**	0.220*
2	I have a strong sense of belonging to my own ethnic group.	0.740**	−0.141
3	I understand pretty well what my ethnic group membership means to me.	0.850**	−0.526**
4	I have often done things that will help me understand my ethnic background better.	0.772**	0.108
5	I have often talked to other people in order to learn more about my ethnic group.	0.850**	0.389**
6	I feel a strong attachment towards my own ethnic group.	0.689**	0.016

Bifactor indices and evaluation

ECV values indicated the general REI factor explained most of the shared variance (0.870). Omega values for the general REI factor (0.905) and the specific factor (0.905) indicated high levels of reliability. The OmegaH value for the general factor was above 0.8 (0.905), reflecting ethnic identity as a unidimensional construct. The OmegaHS value for the specific factor (0.000) indicated that it does not account for systematic variance. The general REI factor achieved an H value of 0.903, meaning the latent construct was well-defined. However, the specific factor was not above the 0.8 threshold. The FD values for the general and specific factors were all ≥0.9, indicating that each factor score was appropriate for interpretation. As appropriate for this intended bifactor model, the standardized factor loadings and the bifactor indices of the specific factor were not strong, suggesting the presence of a “nuisance” factor (factors that arise because of content, method, etc.) that potentially interferes with measurement of the target construct (Reise et al., 2010).

Differential item functioning

DIF models examined the influence of each of these demographic and illness-related variables controlling for the general and specific factors (Objective: 2; Table 4 provides standardized results). Upon examination of the direct effects of sociodemographic variables on item-level responses, evidence of uniform DIF was present by race on items 4 and 5 of the MEIM-R following Holm’s procedure p-value correction (Holm, 1979). Participants identifying as Black/African-American were more likely to rate these items higher than White participants, even when controlling for the general and specific factors. In examining the presence of non-uniform DIF, the moderating effect of the general factor and gender on item 4 was present following Holm’s procedure.

Table 4.

Differential item functioning analysis of the general and specific factors with corrected p values via Holm’s procedure using a target family wise error rate of p<.05.

Parameter	Estimate	SE	p value	Corrected p value
MEIMQ1 regressed on
General (factor loading)	0.465	0.174	.007	.476
Specific (factor loading)	−0.084	0.162	.604	1.000
A1C	0.009	0.066	.886	1.000
Income	−0.01	0.043	.816	1.000
Race	0.799	0.334	.017	1.000
Gender	−0.153	0.272	.574	1.000
Illness duration	−0.812	0.436	.063	1.000
Age	0.05	0.088	.567	1.000
A1C × General	0.101	0.062	.105	1.000
Income × General	0.081	0.072	.261	1.000
Race × General	−0.22	0.289	.446	1.000
Gender × General	0.291	0.472	.537	1.000
Illness duration × General	1.053	0.6	.079	1.000
Age × General	−0.184	0.131	.160	1.000
MEIMQ2 regressed on
General (factor loading)	0.248	0.206	.228	1.000
Specific (factor loading)	−0.774	0.111	<.001	<.001
A1C	−0.024	0.073	.749	1.000
Income	0.039	0.047	.401	1.000
Race	1.061	0.351	.002	.148
Gender	0.2	0.249	.423	1.000
Illness duration	−1.197	0.427	.005	.355
Age	0.016	0.069	.819	1.000
A1C × General	0.133	0.086	.124	1.000
Income × General	0.05	0.066	.449	1.000
Race × General	−0.787	0.313	.012	.804
Gender × General	0.043	0.372	.907	1.000
Illness duration × General	1.496	0.885	.091	1.000
Age × General	0.003	0.129	.983	1.000
MEIMQ3 regressed on
General (factor loading)	0.438	0.202	.030	1.000
Specific (factor loading)	−0.523	0.116	<.001	<.001
A1C	−0.069	0.056	.218	1.000
Income	−0.014	0.031	.641	1.000
Race	0.826	0.266	.002	.148
Gender	0.494	0.235	.035	1.000
Illness duration	−0.421	0.413	.308	1.000
Age	−0.027	0.057	.640	1.000
A1C × General	0.109	0.045	.016	1.000
Income × General	0.067	0.038	.079	1.000
Race × General	−0.213	0.335	.524	1.000
Gender × General	−0.43	0.228	.059	1.000
Illness duration × General	1.137	0.586	.052	1.000
Age × General	0.006	0.099	.952	1.000
MEIMQ4 regressed on
General (factor loading)	0.746	0.148	<.001	<.001
Specific (factor loading)	−0.297	0.18	.099	1.000
A1C	−0.01	0.047	.834	1.000
Income	−0.037	0.035	.290	1.000
Race	1.117	0.25	<.001	<.001
Gender	0.228	0.188	.224	1.000
Illness duration	−0.378	0.413	.361	1.000
Age	−0.099	0.056	.080	1.000
A1C × General	0.102	0.044	.020	1.000
Income × General	0.127	0.045	.005	.355
Race × General	−0.646	0.308	.036	1.000
Gender × General	0.677	0.181	<.001	<.001
Illness duration × General	2.233	0.647	.001	.075
Age × General	−0.198	0.064	.002	.148
MEIMQ5 regressed on
General (factor loading)	0.502	0.138	<.001	<.001
Specific (factor loading)	−0.595	0.108	<.001	<.001
A1C	−0.075	0.056	.180	1.000
Income	−0.07	0.041	.083	1.000
Race	1.127	0.282	<.001	<.001
Gender	0.281	0.267	.294	1.000
Illness duration	−0.625	0.403	.121	1.000
Age	−0.198	0.071	.005	.355
A1C × General	0.113	0.075	.131	1.000
Income × General	0.089	0.062	.153	1.000
Race × General	−0.745	0.406	.066	1.000
Gender × General	0.599	0.582	.303	1.000
Illness duration × General	0.107	0.959	.911	1.000
Age × General	0.075	0.128	.559	1.000
MEIMQ6 regressed on
General (factor loading)	0.205	0.154	.181	1.000
Specific (factor loading)	−0.591	0.12	<.001	<.001
A1C	0.006	0.066	.928	1.000
Income	−0.048	0.051	.341	1.000
Race	0.601	0.394	.128	1.000
Gender	0.341	0.288	.236	1.000
Illness duration	−0.078	0.464	.866	1.000
Age	−0.085	0.083	.308	1.000
A1C × General	−0.06	0.074	.417	1.000
Income × General	0.117	0.065	.072	1.000
Race × General	0.259	0.499	.604	1.000
Gender × General	−0.564	0.481	.241	1.000
Illness duration × General	0.311	0.996	.755	1.000
Age × General	−0.057	0.139	.680	1.000

Note. MEIM = Multigroup Ethnic Identity Measure.

Validity

Using a model that controlled for uniform and non-uniform DIF, latent correlations were determined between the latent general ethnic identity factor and observed total scores for PedsQL-DM 3.0, a familism scale, the DSQ, the DFCS, and the DAAS. This determined whether ethnic identity relates to other psychosocial factors associated with diabetes-related outcomes (Objective: 3). As shown in Table 3, the general REI factor had small, non-significant associations with every variable except race, with youth identifying as Black scoring higher total MEIM-R scores.

Discussion

The present study explored the psychometric properties of the MEIM-R in a racially- and income-diverse sample of youth with T1D by assessing the factor structure (Objective: 1), examined potential bias related to sociodemographic proxies of oppression through DIF (Objective: 2), and subsequently utilized bivariate correlations with psychosocial and sociodemographic factors (Objective: 3). Contrary to previous MEIM-R factor structures examined among healthy youth (Brown et al., 2014) our results indicated that a bifactor model with one general REI and one specific factor for the MEIM-R provided the best model fit in this sample of diverse youth with T1D. The high OmegaH value for the general factor reflects the MEIM-R as a unidimensional construct, and the collection of bifactor indices and factor loadings for the specific factor suggests that the uncorrelated specific factor may be a useful strategy to disentangle construct-irrelevant variance in the MEIM-R items. A bifactor model has previously been found appropriate in a sample of college-aged individuals using the original MEIM (Yap et al., 2014), yet there is no literature examining a bifactor model of the MEIM-R. Further, no studies have examined the psychometric properties of the 6-item MEIM-R among youth with chronic illnesses.

While the MEIM-R previously evidenced DIF in White Americans scoring higher on item 1 of the MEIM-R relative to their Black/African-American counterparts (Chakawa et al., 2015), the present study found significant direct effects of the racialized category on items 4 and 5, indicating that youth with T1D identifying as Black/African-American scored higher on these items than their White peers. Indeed, previous MI evidence of the original MEIM encouraged caution in MEIM use among diverse samples due to challenges at the scalar level (Yap et al., 2014). Since White youth are the majority racial group, they may not have as much experience with racial socialization as their Black peers and are rewarded for silence on their own REI in promoting the status quo, yielding a lower overall sense of REI toward their own ethnic group (Moffitt & Rogers, 2022). This explanation could be further supported in that much of the educational content produced demonstrates bias and sympathy in favor of White voices and experiences in the course of history (Jimenez, 2020).

Moreover, previous research has indicated that youth with minoritized ethnic identities are more likely to mention ethnicity when describing themselves compared to White peers (Akiba et al., 2004). Because racially minoritized youth are exposed to higher levels of racism and discrimination at the hands of White peers, they are more likely to discuss topics related to race and meaning-making surrounding their experiences with oppression (Moffitt & Rogers, 2022). Critical consciousness (i.e., reflection, motivation, and action against oppression) is one potential area needing further exploration in its association with REI and how it may protect against adverse outcomes in minoritized youth (Castro et al., 2022). Critical motivation, known as one’s perceived ability and desire to advance equity and justice (Rapa et al., 2020), has been associated with improved well-being in youth facing racial/ethnic marginalization (Castro et al., 2022).

Contrary to prior evidence (Chakawa et al., 2015), a significant moderating effect was found on the path between gender with the general factor and item 4 (“I have often done things that will help me understand my ethnic background better”; i.e., non-uniform DIF), indicating it was a significantly stronger predictor for those who identified as “male”. Interestingly, previous research posits that women and girls may be expected to carry on cultural traditions and values, with adolescent Black girls reporting higher in ethnic identity search relative to their adolescent Black boy peers (Phinney, 1990). Umaña-Taylor and colleagues observed that Latina adolescents experience earlier developmental progression of REI relative to white peers (Umaña-Taylor et al., 2009). White youth have reported heightened focus on their gender identity relative to their ethnic identity, whereas youth holding minoritized ethnic identities equally considered both their gender and REI (Spears Brown, 2007). However, little research overall has examined gender differences in REI. Other areas of oppression due to intersecting identities may impact women and girls’ ability to engage in tasks related to REI; for example, men and boys may be more “privileged” with opportunities for REI exploration. Further research is needed to better understand potential gender differences in REI.

There are a few limitations to this study warranting consideration. While age was not found to be significant, the cross-sectional nature of this study design limited the ability to observe previously-established developmental changes in REI over time (Umaña-Taylor et al., 2014). Longitudinal research indicated that REI development increases over time in minoritized youth, with exposure to racism stimulating further REI development (Umaña-Taylor et al., 2009). Once a set of MEIM questions are established without the presence of DIF, it will be important to examine longitudinal invariance over developmental periods. Further, limited ethnic diversity in this sample prevented examination of DIF in Hispanic/Latino youth with T1D.

Another concern is the measurement of gender as a binary variable (i.e., boy/girl) in this study, which prohibited examination of potential differences for youth with T1D holding a minoritized gender identity. Our present understanding acknowledges cultural and societal expectations influencing gender/gender identity development, which can be fluid (American Psychological Association, 2015). As such, future studies should seek to provide responses more inclusive of gender identity selections.

Contrary to what was hypothesized, the MEIM-R did not significantly correlate with other psychosocial or demographic measures utilized in this study aside from race, limiting determination of convergent validity. Lim et al. (2016) found that the Affirmation/Belonging subscale of the 12-item MEIM correlated significantly with health-related quality-of-life outcomes, but no prior studies have examined T1D-specific psychosocial measures and outcomes. However, the role of ethnic identity on improved HbA1c has been supported in adults with T2D (Murayama et al., 2017). Our results suggest that REI may serve in a moderating role in the relation between constructs and items, which may suggest a broader moderating role with other psychosocial or sociodemographic variables (Mossakowski, 2003; Umana-Taylor & Updegraff, 2007). As such, complex and/or non-linear relations may be present and provide a better statistical and conceptual match to theory, seeking evidence of validity, and exploring the potential role of REI in the lives of youth with T1D and their families. (Liese et al., 2022; Velez & Spencer, 2018).

While the present study adds to previous literature by supporting a bifactor approach with the MEIM-R, DIF analyses indicated that use of the MEIM-R as a total “sum” score may not be appropriate without utilizing quantitative strategies to account for DIF. This warrants caution to the field of potential differences in response patterns on the MEIM-R by racialized/ethnic groups and gender which may yield inaccurate measurement of REI. Future studies examining the MEIM-R could reconsider returning to the full length MEIM, developing a new item pool, examining DIF among other racialized/ethnic groups not included in this study, as well as conducting a longitudinal study to explore how patterns of MEIM-R responses may change over development. Further, examination of convergent validity of the MEIM-R relative to other measures of REI could help inform the field in its appropriateness for use.

Supplementary material

Supplementary material is available online at Journal of Pediatric Psychology (https://academic.oup.com/jpepsy/).

Author contributions

Adora E. Choquette (Conceptualization [lead], Formal analysis [lead], Writing – original draft [lead], Writing – review & editing [lead]),Kristoffer S. Berlin (Data curation [equal], Funding acquisition [lead], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Supervision [equal], Writing – review & editing [equal]),Kishan R. Desai (Writing – review & editing [equal]), Rachel L. Ankney (Data curation [equal], Resources [equal], Writing – review & editing [equal]), Rachel Tillery-Webster (Writing – review & editing [equal]), Kasey R. Harry (Writing – review & editing [equal]), LaTasha Holden (Writing – review & editing [equal]), Jessica L. Cook (Writing – review & editing [equal]), Mary Keenan-Pfeiffer (Writing – review & editing [equal]), Katherine A. Semenkovich (Writing – review & editing [equal]), Kimberly L. Klages (Writing – review & editing [equal]), Tiffany J. Rybak (Writing – review & editing [equal]), Gabrielle G. Banks (Writing – review & editing [equal]), Kathryn Sumpter (Resources [equal], Writing – review & editing [equal]), and Angelica R. Eddington (Resources [equal], Supervision [equal], Writing – review & editing [equal]).

Funding

This work was supported in part by a grant from The University of Memphis Faculty Research Grant Fund to KSB. This support does not necessarily imply endorsement by the University of Memphis of research conclusions.

Conflicts of interest

The authors have no conflicts of interest to disclose.

Ethical approval

This study was approved by the University of Tennessee Health Science Center and The University of Memphis Institutional Review Board.