About this poster

---

Panel information
Panel 3. Schooling and Education

Abstract

---

Raciolinguistics is the study of how students who speak a minority language and are children of color experience discrimination based on their intersecting identities (Rosa & Flores, 2017). These children have long been marginalized and viewed through a deficit-based lens that frames them primarily as “non-English proficient” without much consideration of what they know in their home language (MacSwan, 2018). However, there is a growing awareness of the benefits of multilingualism, and the need to reframe these approaches to capitalize on linguistic strengths (Bialystok, 2018).

Methods
Purpose
The purpose of this research was to calibrate the new items on the Multitudes expressive vocabulary universal screening measure and create a unitary scale across English and Spanish using conceptual scoring.

Participants
A sample of 1,219 bilingual children in kindergarten and 1st grade across the state of California.

Measure
The Multitudes expressive vocabulary test including about 75 items that were the same in English and Spanish per grade.

Procedures
Children were administered the assessment in both English and Spanish in the winter of 2023 to calibrate test items and establish item difficulty in each language. The data were then analyzed to explore relationships in item difficulty between English and Spanish and the possibility of creating a unitary scale.

Analyses and findings
We first conducted preliminary basic descriptive item analysis within each language. For the Spanish items, the proportion of correct responses ranged from 0 to 0.85 with an average of 0.29, and the point-biserial correlations ranged from 0.09 to 0.73 with an average of 0.51. For the English items, the proportion of correct responses ranged from 0.02 to 0.81 with an average of 0.33, and the point-biserial correlations ranged from 0.07 to 0.77 with an average of 0.52.
We then compared proportion or correct responses and point-biserial correlations between the English and Spanish versions of the same item, we observed a moderate correlation. The correlation was 0.73 for proportion of correct responses and 0.58 for point-biserial correlations for the remaining 62 items in the analysis.

The moderate correlations between psychometric properties of English and Spanish versions provided a rationale for conceptual scoring. The basic item analysis was run again using conceptual scoring and the proportion of correct responses ranged from 0.05 to 0.98 with an average of 0.52, and the point-biserial correlations ranged from 0.19 to 0.65 with an average of 0.46.

The Rasch model was fitted by using the TAM package in R with the marginal maximum likelihood estimation to the conceptual scores (combined responses from both languages). Table 1 summarizes the item parameters and item fit statistics. The item location estimates ranged from -4.59 to 3.65 with an average of 0 and standard deviation of 1.59. The distribution of item location estimates compared to the person location estimates is shown using a Wright Map (See Figure 1).

We also examined the unidimensionality assumption for the conceptual scores. One way to evaluate the assumption of unidimensionality after fitting a Rasch model is to calculate the
proportion of variance in responses that can be attributed to item and person locations on the primary latent variable, as estimated using the Rasch model. One first can find the variance associated with the observed responses (VO) and the variance associated with residuals (VR) so that we can compare these values. The percent of variance explained in observed responses by the Rasch model can be found by, (VO – VR)/ VO. In this dataset, the percent of variance explained by the Rasch model was found to be 40.4%, indicating strong evidence of unidimensionality.

⇦ Back to session