3.1. Data Characteristics and Variables
3.1.2. Reclassification Criteria
For the purposes of this study, reclassification was coded as a dichotomous variable that took on a value of 1 for a student’s final year as an EL, and 0 for all prior years. This variable is the outcome in the survival analysis models, which are in essence predicting the probability that a student will meet the criteria necessary to be reclassified starting in the following school year.
According to technical documentation, reclassification decisions in this state are to be made based solely on students’ ELP scores (see, e.g, [State 2011 ELPA Technical Report], p.4). Up through the 2012-13 school year (the final year of data for the 2008 cohort, only), the state used an ELPA that produced two scale scores: oral language, and
speaking, and the two scale scores were created by pairing these raw scores – listening and speaking for oral language, and reading and writing for written language – and scaling the two combinations separately. In their individual score reports, students
received scaled scores and proficiency levels for oral and written language, as well as one overall proficiency level.
The four performance levels for all three reporting categories were Beginning, Intermediate, Advanced, and Proficient. The cut score for proficiency and reclassification was between Advanced and Proficient, meaning students had to score in the highest performance level (Proficient, Level 4) to be reclassified. Similarly, to be reclassified, or generally to advance from one overall proficiency level to the next, students needed to clear the cut score on both scales. In sum then, only students who scored in the highest performance levels on both subscores were eligible for reclassification, according to the state’s criteria.
A student’s reclassification is formally signaled in the dataset when the student ceases to be marked with the state’s special LEP code. According to the state’s data manual,
Students who test out of LEP by reaching proficiency on the [ELPA] are still entitled to accommodations and some types of services for two years; however, once the students have tested out of LEP, they must NOT be
recorded as LEP with [the state’s LEP] code. (p. 188-9)
Thus – assuming that districts adhere to the state’s policies about reclassification – beginning in the next school year after a student meets the reclassification cut score, the student is no longer coded as an EL in the state records.
For the purposes of this study, it was instructive to examine the extent to which the state’s use of this LEP flag code reflected the state’s documented reclassification
policies – in other words, did the test scores of students who were reclassified meet the performance standards specified for this transition? To explore how well schools and districts adhered to the state’s reclassification policies, I created contingency tables that crossed students’ performance levels with their EL status in the following school year. Overall, out of 20,457 instances where students’ ELP performance made them eligible for reclassification, 93 percent (19,029 cases) were reclassified in the following school year, as evidenced by the disappearance of the state’s limited English proficient code from the student’s file in the following school year. These 19,029 students also represented 96 percent of all students who were reclassified during the observation period. Thus, the vast majority of reclassification decisions followed the state’s prescriptions.
An analysis of the deviations from the rule suggested that it was more common for students who met the reclassification criteria to be retained as ELs than for students who fell short to be reclassified. Among all students who met the reclassification criteria, 1,428 (approximately 7 percent) were not reclassified. Meanwhile, among all students who were reclassified, 723 (approximately 4 percent) fell short of the prescribed criteria. In the latter case, one fifth (154 cases; 21 percent) were students who had met the
criterion on one test and were in the second highest performance level on the other, and roughly one tenth (10.6 percent; 77 cases) were students who scored in the second highest category on both assessments. The remaining 492 cases were mostly students who were missing data altogether. In general, although it is possible that coding errors may be at play in some of the observed instances of deviation from the reclassification rule, for the purposes of this study I assumed that any deviations from the rule were
genuine situations (i.e., not coding errors) where school- or district-level judgment was allowed to override state policy.
As a final point, it is important to note that the state did switch to a different ELP assessment with different reclassification rules starting in the 2012-13 school year. This new assessment not only had new cut scores, but also used an entirely new decision rule for reclassification – a combination model where students had to meet an overall
performance level and certain minimum scores on each of the four subtests to be reclassified. Because of the timing of this change, no reclassification decisions in my observation period were made on the basis of this new test (those decisions would be reflected in student records starting in the 2013-14 school year). Scores from this assessment are available in the final year of the 2008 cohort’s data, however, and are relevant for considerations about students who are still ELs at the end of the observation period (see section 4.1.3 for this discussion). This shift is also important, because it clearly limits the generalizability of my conclusions to years beyond those observed here. Since the entire reclassification shifted, it remains to be seen whether the relationships observed here changed in subsequent school years.