Search

Grit, defined as perseverance and passion for long-term goals, has garnered attention in educational research for its potential to predict academic achievement and well-being. Despite this, its psychometric properties and applicability across diverse populations remain underexplored. This study aims to validate the Grit-O scale and explore its relations with various psychosocial and academic outcomes among underrepresented undergraduate STEM scholars. Analyses utilized a mixed-methods approach, including confirmatory and exploratory factor analyses, bivariate correlations, and moderated multiple regression analyses. Results indicate that the Grit-O scale did not demonstrate adequate fit in its original form. However, an amended 11-item scale comprising two correlated factors (perseverance and consistency) showed improved fit and was retained for further analysis. Exploratory analyses revealed consistent factor structures across binary gender groups, suggesting scale robustness across gender identities. Neither perseverance nor consistency significantly predicted entry into a STEM graduate program. The intricate relationship between grit and various academic and psychosocial outcomes offers valuable insights into the influence of perseverance and consistency. Correlations and moderated multiple regression analyses highlight nuanced associations between grit dimensions and academic and mental health outcomes. This study sheds light on Grit-O's psychometric properties, but its utility for measuring academic achievement remains inconclusive. The mixed findings underscore the need for continued psychometric scrutiny and a nuanced understanding of academic achievement dynamics, particularly for underrepresented groups. Future grit research should prioritize rigorous validation of the measure to ensure that the model appropriately fits the population and can be reliably used to draw meaningful conclusions. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

When measuring latent constructs, for example, language ability, we use statistical models to specify appropriate relationships between the latent construct and observe responses to test items. These models rely on theoretical assumptions to ensure accurate parameter estimates for valid inferences based on the test results. This dissertation specifically concerned the local independence (LI) assumption (or presence of local dependence (LD) equivalently) in two specific item response theory (IRT) models -- the partial credit model (PCM) and the generalized partial credit model (GPCM). With a primary focus on several commonly used global fit indices (GFIs), this study investigated how parameter estimates in the studied IRT models were affected by varying degrees of LD, whether the studied GFIs were responsive to model misspecifications due to LD, and any insights regarding making reasonable decisions in terms of good-fit models. The results revealed that, first, parameter estimates in both the PCM and GPCM were severely biased, especially the latent trait and item step parameter estimates. The degree of biases increased with higher levels of LD across different model and data conditions. Other general factors, including the test length and number of item response categories, could also impact biases in parameter estimates. Second, all studied GFIs were responsive to model misfit caused by the presence of LD. Specifically, studied GFIs exhibited moderate to strong associations with biases of the latent trait estimates in expected directions. Lastly, based on the performance of the studied GFIs under baseline and LD-induced conditions, the absolute fit indices (RMSEA and SRMSR) appeared to be more informative than the incremental fit indices (CFI and TLI) in assessing the global fit of the PCM and GPCM. This study provides insights into adopting specific decision criteria of good model fit under various model and data conditions, aiming to reduce potential false positive rates while maintaining reasonable true positive rates. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

This dissertation compares the performance of equivalence test (EQT) and null hypothesis test (NHT) procedures for identifying invariant and noninvariant factor loadings under a range of experimental manipulations. EQT is the statistically appropriate approach when the research goal is to find evidence of group similarity rather than group difference; despite this, the conventional approach to measurement invariance analysis relies upon NHT. EQT has proved effective for invariance detection using global model-data fit statistics in simulated and real-world data (Counsell et al., 2020) but its use in partial measurement invariance (PMI) analysis for evaluation of factor loading differences between groups has not been previously investigated. A Monte Carlo simulation is used to investigate test performance under experimental manipulations of power, model complexity, degree of noninvariance, and choice of EQT bound. Jung and Yoon (2017) use a similar approach to demonstrate that the direct comparison of factor loading differences across groups using NHT confidence intervals effectively detect population patterns of invariant and noninvariant loadings. I compare the performance of NHT to EQT using both conventional benchmarks of factor loading equivalence inferred from the work of Counsell and Cribbie (2015), Nye, Bradburn, Olenick, Bialko, and Drasgow (2019), and Shi et al. (2019) as well as EQT bounds selected post hoc to maximize recovery of the population pattern of invariance or partial invariance. I also investigate the likelihood of false invariance detections for truly noninvariant loadings and false noninvariance detections for truly invariant loadings over a range of EQT bounds, sample sizes, and magnitudes of noninvariance. The results of this simulation indicate that EQT can detect population patterns of invariance or partial invariance, but requires high power (i.e., large sample sizes and strong communalities) as well as optimal selection of the EQT bound. In experimental conditions where these requirements are met, EQT recovers the correct population pattern of invariant and noninvariant factor loadings in more simulation replicates than the NHT approach (Jung & Yoon, 2017). Inspection of individual loading detection rates reveals that false noninvariance errors occur most frequently when power is low, while false invariance errors are common whenever the true magnitude of noninvariance is small and the selected EQT bound is larger than the loading difference. In the EQT framing, false invariance errors in the presence of minor noninvariance might also be interpreted as accurate detections of practically equivalent loadings. These findings support the use of EQT for PMI analysis and reinforce previous findings that EQT is an effective but power-intensive procedure in the multivariate context (Alter & Counsell, 2023). The choice of EQT bound has implications for both the power of the test and the consequences of accepting loading differences falling within the selected EQT bound as invariant. Researchers using EQT to detect PMI should collect large samples in order to maximize the likelihood of detection, and should consider how much misspecification their selected EQT bounds might tolerate. Future development of PMI simulation tools could support the selection and justification of EQT bounds in applied settings. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

The landmark theoretical model of Community Cultural Wealth (CCW) was developed almost twenty years ago as a critique of American meritocracy and the deficit views of Students of Color (Yosso, 2005). Most research on CCW since that point has been qualitative in nature, specifically in-depth interviews to understand how Students of Color exact their aspirational, familial, linguistic, navigational, resistant, and social capital as they persist through education. The current study aimed to quantify and validate CCW as a six-factor model and examine the effects of CCW on academic self-efficacy, sense of belonging, grade point average, and mental health in a sample of Black and Latino/a college students. 567 college students from colleges across the United States participated in an online survey with CCW, student outcomes, and demographic items. Results indicated that all six cultural capital confirmatory factor analyses (CFA) exhibited good model fit and converged into a second-order Community Cultural Wealth latent construct, consistent with theory. CFA indicated good model fit for first-order factors for academic self-efficacy, sense of belonging, and mental health. Community Cultural Wealth significantly positively predicted all student outcomes, including explaining 62% of the variance in Academic Self-Efficacy, 25% of the variance in Sense of Belonging, and 14% of the variance in Mental Health. Grade point average was significant but was a weak outcome variable, and participation in college readiness programs did not mediate any relationships as predicted. The quantification of CCW as a six-factor model consistent with theory is a key finding, as it is the first time it has ever been done. Future research should involve buy-in from participants to solidify items used to measure CCW and how CCW affects student outcomes in different student populations. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

The purpose of this study was to investigate the effectiveness of four different models (bifactor, CTC(M-1), CTCU and unidimensional) as to optimal model selection when the wording effect associated with negatively worded items was present. A Monte Carlo simulation study was conducted to compare model-data fit and accuracy in parameter estimates of the four models. Design factors include: two true models (CTC(M-1), CTCU models) x three sample sizes (small N=250, moderate N=500, large N=1,000) x two item ratios (positive items:negative items = 2:1 or 1:1) x three distributions (distribution of negative items; symmetric, moderately asymmetric, and extremely asymmetric). The generated data were analyzed with two different estimators (MLMV, WLSMV) under four different models (bifactor, CTC(M-1), CTCU, and unidimensional). Outcomes examined included fit index values (chi-square, RMSEA, SRMR, CFI, TLI, AIC, BIC), power, Type I error, and relative bias in factor loadings on general trait factor estimated from the four tested models. As a second study, the four tested models were applied to empirical data set from the RSE scale to evaluate whether the simulation study results would be observed in practice. Results showed that three tested models (bifactor, CTC(M-1), and CTCU) fit the data equivalently well regardless of study conditions. Misspecified unidimensional models would erroneously be selected as an optimal model when certain study conditions (small sample, asymmetrically distributed negative items, unbalanced number of positive and negative items, WLSMV estimator) were present. Among tested models, the CTC(M-1) tended to produce most accurate factor loadings for both positive and negative items. According to the study's findings, there could be overlaps in the fit index values under several alternative solutions. Researchers should not select one model over the other solutions only because it produced the best fit index values. Instead, a model selection should be determined based on the study's objectives or on substantive considerations. Additionally, researchers should carefully examine the data's features when misspecified unidimensional models yield goof-fit. Unidimensional models ignoring negative wording effects could be recognized as a good-fitting model in some circumstances because of different performance of fit indices. Several practical advice and guidelines are provided for researchers using unidimensional scales with mixed item wording. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

This multiple-manuscript dissertation explored the measurement invariance (MI) testing with multiple-group confirmatory factor analysis (MG-CFA) approach from different perspectives. Study 1 explored MI from a theoretical perspective by conducting a systematic review study on MI practices in education. The findings of this study indicated inconsistency in MI practices and showcased the limitations of the MI practices conducted by researchers in the field of education. Study 2 examined MI from an empirical perspective by implementing a cultural MI test of Strengths and Difficulties Questionnaires for elementary school students in the United States and China. The study provided a step-by-step demonstration of how to conduct an MI test appropriately. Study 3 investigated MI from a methodological perspective with a simulation study. This study examined the impact of model size and group size ratio on the sensitivity of fit measures to detect MI. The study found that model size, in combination with group size ratio, affected the power of CFI, RMSEA, and SRMR for identifying the metric or scalar noninvariance. Study 2 and Study 3 originated from the issues with MI practice identified from Study 1 and served as extensions of Study 1. These three manuscripts contributed to the research on MI testing with the MG-CFA approach theoretically, empirically, and methodologically. Overall, this series of studies help researchers gain a better understanding of the application of MI from different perspectives. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

This dissertation explored whether unnecessary linguistic complexity (LC) in mathematics and biology assessment items changes the direction and significance of differential item functioning (DIF) between subgroups emergent bilinguals (EBs) and English proficient students (EPs). Due to inconsistencies in measuring LC in items, Study One adapted a rubric counting instances of specific grammatical features in items and introduced a method for evaluating lexical features in items. Four raters were asked to count the presence of five grammatical features in assessment items and determine whether each feature contained construct-relevant vocabulary. The items were drawn from four content assessments administered to Massachusetts high school students: two biology assessments and two mathematics assessments. These counts of grammatical and lexical features were modeled in factor analyses to evaluate the multidimensionality of LC and subsequent fit of multidimensional LC models. While there were problems with raters consistently counting construct-irrelevant grammatical features, multidimensional models of LC fit acceptably well. Factor scores obtained from the measurement models for lexical complexity, relative clauses, and complex noun phrases created in Study One were used for Study Two. In Study Two, Rasch hierarchical generalized linear models (HGLMs) were created to evaluate DIF between different subgroups of EBs and EPs on a biology assessment and a mathematics assessment, as including LC as an item covariate may predict item responses differently by comparison group. Seven comparison groups were evaluated across two assessments (mathematics and biology): EPs versus EBs, EPs versus short-term EBs, EPs versus long-term EBs, short-term EBs versus long-term EBs, EPs versus Spanish-speaking EBs, EPs versus non-Spanish-speaking EBs, and non-Spanish-speaking EBs versus Spanish-speaking EBs (reference group versus focal group, respectively). For each comparison group, at least five models were created: a comparison model with all participants in the comparison group with that only accounts for the main effect of focal group status, a "base model" that evaluated DIF for the comparison groups with no LC item covariates, a model including lexical complexity as an item covariate ("LEX predictor"), a model including complex noun phrases as an item covariate ("NP predictor"), and a model including relative clauses as an item covariate ("RC predictor"). If LC predictor models improved model fit, models with multiple LC predictors were created. For the EP versus EB comparison groups on the mathematics assessment, model fit only improved with the NP predictor model, while the LEX, NP, and RC predictor models improved model fit for the EB versus EB comparison groups; a model with all LC predictors improved model fit for the EB versus EB comparison groups. For the biology assessment, the LEX, NP, and RC predictor models improved model fit for all comparison groups; a model with all LC predictors improved model fit for all comparison groups. The main effects of the item covariates (LC factor scores) and their interactions with focal group status were evaluated, as were the number of items within a comparison group that had changes in DIF significance or direction when including a LC predictor. All LC predictors had consistent main effects across comparison groups. For the mathematics assessment, items with higher complex noun phrases factor scores were consistently more difficult for all comparison groups (NP predictor model), and items with higher lexical complexity (LEX predictor model, all predictors model) or relative clauses factor scores (RC predictor model, all predictors model) were consistently more difficult for all EB versus EB comparison groups. For the biology assessment and all comparison groups, items with higher lexical complexity (LEX predictor model, all predictors model) or complex noun phrases factor scores (NP predictor model, all predictors model) were consistently more difficult, and items with lower relative clauses factor scores (RC predictor model, all predictors model) were consistently more difficult, with one exception. In the all predictors models for the EB versus EB comparison groups, only relative clauses had a significant main effect. There were some changes in interactions with LC predictors and focal group status. For the mathematics assessment and EP versus EB comparison groups, complex noun phrases interactions favored EPs. For the mathematics assessment and EB versus EB comparison groups, generally the interactions in the single LC predictor models generally favored STEBs compared to LTEBs and non-Spanish-speaking EBs compared to Spanish-speaking EBs, but when all LC predictors were included, no interactions between LC predictor and focal group status were significant. For the biology assessment and EP versus EB comparison groups, lexical complexity and complex noun phrases factor scores interactions generally favored EPs, and relative clauses factor scores interactions favored EBs and EB subgroups. For the biology assessment and EB versus EB comparison groups, regardless of whether examining the single LC predictor or all predictors models, no interactions between focal group status and LC predictor were significant. Changes in DIF significance and direction were compared between the base model and LC predictor models for all comparison groups. For the mathematics assessment and EP versus EB comparison groups, after conditioning on complex noun phrases, items with complex noun phrases generally exhibited significant DIF favoring EBs, regardless of whether the complex noun phrases factor scores were high (one standard deviation above the mean) or low (due to floor effects, the lowest complex noun phrases factor score). For the biology assessment, all items exhibited significant DIF favoring EBs after accounting for lexical complexity, most items exhibited non-significant DIF after accounting for complex noun phrases or relative clauses, and items were mixed between exhibiting non-significant DIF or significant DIF favoring EBs after accounting for all LC predictors. While items with high relative clauses factor scores exhibited non-significant DIF, some items with low relative clauses factor scores exhibited significant DIF favoring EPs after accounting for relative clauses. Items with two or more high factor scores exhibited non-significant DIF, but items with two or more low factor scores exhibited significant DIF favoring EBs after accounting for all LC predictors. These results were fairly consistent across different EP versus EB comparison groups, although different items were flagged for DIF in initial models not accounting for LC predictors. Items were less difficult for EBs than EPs after accounting for LC features, which suggests the abilities of EBs are underestimated due to LC in items, even if the items have low LC. Considering subgroup differences in these EIRMs, the key takeaway is that while different items are flagged as exhibiting significant DIF for different EP versus EB comparison groups when examining DIF with no LC predictors, there are few subgroup differences in items changing DIF significance or direction after accounting for LC predictors. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]