Developing a More Accurate Home Language Survey for Initial English Learner Classification: California's Pilot Study

• In Event: Strengthening the Validity and Utility of Home Language Surveys to Identify Potential English Learners

Abstract

Purpose. This study reports results comparing the accuracy of a more detailed home language survey (HLS) to a currently-used simple one.
Theoretical Framework. California has a growing population of students whose families speak a language other than English (LOTE) (Quality Counts, 2009). Accurately determining which students are initially fluent in English is an important educational function (Linquanti & Bailey, 2014). Step 1 in this determination is the HLS. Parent responses determine if a student proceeds to step 2: taking an initial English language proficiency (ELP) test (California: CELDT). Students who pass or are not required to take CELDT participate in mainstream classes. Students who score below a given cut-point are designated “English learner” (EL) and receive English language development support.
ELP tests and HLS are imperfect measures. For HLS, this may be due, in part, to question and decision rule simplicity. The current California HLS has three questions and any answer of LOTE requires the child to take CELDT. Parents who misunderstand the HLS as intended create false positives or negatives, which is then compounded by ELP test limitations (Linquanti & Bailey, 2014).
Methods. An expert panel created a 14-question Proposed HLS. Beginning in April 2015, we compared students’ initial CELDT results to their parent’s/guardian’s responses on the Current and Proposed HLS. We will complete the study in October 2015.
Research Question: how does the accuracy of the Proposed HLS in predicting which students will be designated as ELs compare to that of the Current HLS?
Logistic regression is used to determine the probability of a student scoring not English-proficient on CELDT, based on the number of LOTE responses. For each HLS, the probability of “failing” CELDT is calculated for each possible decision rule, from one LOTE response to all LOTE responses. We then compare all decision-rule probabilities across both HLS. Decision rules with higher probabilities are considered more accurate.
Data Sources. The sample includes newly-registering students for 2015/16 in five California districts. Data collection is on-going. We expect the final sample of approximately 2000 students. Each student has a completed Current and Proposed HLSs, and those required to take CELDT (using the Current HLS) have a CELDT score. The HLSs and CELDT screener are administered April–September 2015, with final data analysis in October. Students of parents/guardians who answer only “English” on the Current HLS will not have a CELDT score, a study limitation.
Results. A preliminary analysis of HLS data shows different response patterns between the two HLSs. For example, parents/guardians who answered LOTE to only question 3 of the Current HLS (what language parents use mostly with the child), answered LOTE from 2 to 9 times on the Proposed HLS. Thus, we expect important accuracy differences across decision rules for the two HLSs.
Significance. This study provides initial evidence of the extent to which we can strengthen the accuracy of commonly-used, simple HLSs with more complex ones, and so decrease initial EL misclassifications.

Authors

• Eric Haas, WestEd

• Robert T. Linquanti, WestEd