Leveling the Playing Field: Default Policy and Its Effects on English Learner Reclassification Rates

Abstract

Reclassification, the process by which English learner (EL) students exit EL classification, often determines access to mainstream academic coursework. While existing research finds that many students who demonstrate English proficiency do not reclassify, few studies evaluate policies that effectively reclassify eligible students. We examine the impact of shifting reclassification responsibility from school districts to the state in Michigan. Using a difference-in-regression discontinuities (DiRDs) design, we find that state-level responsibility increases reclassification rates by 35 percentage points for students just above the threshold of reclassification eligibility. Effects are larger for Spanish speakers, suggesting automatic procedures may reduce linguistic bias in the reclassification process. Our findings contribute to the literature on default policies in K–12 education and policies that promote equitable EL education.

Equal access to educational opportunities and the equitable application of policy are fundamental principles in modern education policy, as exemplified by the Every Student Succeeds Act of 2015 (ESSA). Implemented in the 2016–17 school year, ESSA represents a shift toward standardizing student benchmarks and away from local discretion. Understanding the effects of policy decisions aimed at increasing standardization and uniformity in processes is critical for assessing whether this trend leads to more equitable decision-making. On the one hand, discretionary policies can allow for flexibility and nuanced decision-making regarding the education of individual students. On the other hand, they can also allow for differential treatment of similarly situated students, which may reflect biases against particular groups of students.

The trend toward education policy standardization is particularly true for multilingual students classified as English learners (ELs), who make up nearly 10% of the U.S. K–12 student population and attend schools in almost every district (National Center for Education Statistics [NCES], 2022). While classified as ELs, students are legally entitled to linguistic support services and additional resources to help them meet academic content standards as they develop English proficiency (ESSA, 2015). EL classification also often results in students being educated in different settings or receiving different educational services than non-ELs. Once ELs demonstrate English proficiency, ¹ they qualify to exit EL services through a process called “reclassification.” Reclassification is significant because it ends schools’ legal obligation to provide linguistic support services and routes students into mainstream academic coursework that is often unavailable to students classified as ELs. If timed appropriately, reclassification should result in a smooth transition to mainstream academic content and ensure students receive access to a developmentally appropriate educational setting (Robinson, 2011).

Once an EL student qualifies for reclassification, their state or school district must administratively formalize their exit from EL status using either a manual or an automatic reclassification process. Many states require districts to identify and manually reclassify students who have met reclassification criteria. Manual procedures often result in many eligible students remaining classified as ELs (Cimpian et al., 2017; Estrada & Wang, 2018; Mavrogordato & White, 2017). Further, reclassification under manual procedures is uneven, with eligible Spanish speakers being notably less likely to reclassify than students speaking other home languages (Mavrogordato & White, 2017; Umansky et al., 2020). This can lead to restricted opportunities to learn in academically rigorous settings and, in many cases, to students being denied access to mainstream core coursework (Estrada & Wang, 2018). A handful of states implement automatic reclassification procedures, in which students who meet reclassification criteria are automatically reclassified via state administrative data systems, thus removing the manual determination by districts. However, the ability of automatic procedures to (a) increase reclassification rates of eligible students and (b) alleviate linguistic biases in reclassification decision-making processes remains unexplored.

A state’s choice to implement a manual or automatic reclassification procedure has meaningful consequences because it sets the default approach to reclassification. In general, automatic processes assume that the qualifying student will reclassify unless an extenuating circumstance indicates the student is not ready. Manual procedures, on the other hand, require districts to decide to reclassify eligible students actively. In other words, automatic reclassification procedures generally require districts to “opt out” of reclassifying students, while manual procedures require districts to “opt in.” Default policies may increase uptake of a desired outcome (Jachimowicz et al., 2019), making them potentially powerful tools in states new to serving EL students, especially in contexts where educators have less experience identifying whether students will benefit from reclassification or in contexts where capacity is limited.

In light of ESSA’s (2015) call for standardization of reclassification, identifying processes that effectively reclassify eligible students is vital for policymakers to facilitate equitable reclassification rates for ELs. To better understand the efficacy of different reclassification procedures for reclassifying eligible students, we present estimates of the effect of shifting the statewide reclassification policy in Michigan. Michigan is a new immigrant diaspora state and, like most other states, serves a rapidly growing EL population. In the 2019–20 school year, Michigan shifted from a manual (school district responsibility) to an automatic (state responsibility) reclassification process. We use a difference-in-regression discontinuities (DiRDs) design to estimate (a) how shifting the default reclassification procedure in Michigan impacts eligible EL students’ likelihood of reclassifying and (b) how the magnitude of impact varies across subgroups of ELs.

We find that shifting from manual to automatic reclassification procedures in Michigan results in significant and meaningful default effects on reclassification rates for eligible students. In Michigan, eligible students are over 35 percentage points more likely to formally reclassify under automatic reclassification than manual reclassification. We also find preliminary evidence of larger effect sizes for ELs reporting Spanish as a home language, suggesting that automatic procedures have the potential to close gaps in reclassification rates for groups of eligible students who may experience bias under manual reclassification procedures. Given ESSA’s (2015) call to standardize reclassification policy, state education agencies may look toward procedures like automatic reclassification to facilitate standardization of the reclassification process and equitable reclassification rates for eligible students.

Our analysis contributes to the literature on default policies and EL education policy. First, we present what we believe to be the first evaluation of a default policy in K–12 education. The increasing prevalence of administrative data sets and the goal of closing education opportunity gaps have resulted in a rise in default policies in education, particularly around enrolling students in advanced coursework. For example, legislators in Texas recently adopted a policy that requires fifth graders who performed in the top 40% on a standardized math assessment to automatically be enrolled in advanced math for sixth grade (Richman, 2023). While researchers have examined the effectiveness of default policies in other fields (e.g., automatic voter registration [Garnett, 2022] and automatic enrollment into retirement savings [Madrian & Shea, 2001]), we have little information on how default policies function in K–12 education.

Additionally, we contribute to the literature on EL reclassification by evaluating the ability of two commonly used reclassification procedures to reclassify eligible students. Prior research identifies reclassification as a critical but elusive juncture in EL students’ educational trajectory, with a substantial population of reclassification-eligible students remaining classified as ELs and lacking access to mainstream coursework. However, no studies have investigated whether shifting to an automatic reclassification process leads to more students who meet reclassification criteria exiting EL status. Moreover, existing research on EL reclassification uses data from before ESSA’s (2015) implementation, which included a push for greater standardization of state reclassification procedures. This study uses post-ESSA data, which may apply to states’ current reclassification contexts.

Background

In what follows, we discuss the prior literature on the significance of timely EL reclassification and the factors influencing reclassification, as well as the role of default policies. This review will examine the adverse effects of prolonged EL status for students demonstrating English proficiency, the variability in state and district reclassification policy implementation, and the complexities of implementing manual reclassification procedures. We then discuss the impact of default policies across various contexts and highlight their potential as effective policy tools in education, particularly for improving EL reclassification outcomes.

EL Reclassification

Timely reclassification is vital as premature reclassification may lead students to struggle in mainstream coursework without linguistic scaffolds, and prolonged EL status may inadvertently lead to adverse social and academic consequences for students. While classified as ELs, students are legally entitled to linguistic supports that allow them to access academic content meaningfully (ESSA, 2015). Premature reclassification may negatively impact ELs’ outcomes in the mainstream classroom by providing fewer opportunities for English language development and limited linguistic scaffolding to access mainstream content (Cummins, 1980; de Jong, 2004). For those with relatively advanced English proficiency, the EL label itself and resulting barriers to core academic coursework can adversely affect academic achievement (Umansky, 2016). EL classification has also been linked to restricted access to core academic content and limited opportunities to interact with non-EL peers (Umansky, 2018), limited access to honors and college preparatory coursework at the secondary level, higher dropout rates, and decreased rates of college enrollment (Carlson & Knowles, 2016). Ultimately, EL services are intended to benefit students as they develop English proficiency, and reclassification upon demonstrating English proficiency ensures students have equitable access to challenging and appropriate coursework.

State-specific reclassification criteria and processes determine whether and when EL students reclassify (Morales & Lepper, 2024). ESSA (2015) requires states to determine reclassification criteria that are to be implemented evenly across districts in the state. Criteria must include an English language proficiency test score, but it can also include other objective or subjective criteria at the state’s discretion, such as standardized test scores or teacher recommendations. Determining appropriately rigorous reclassification criteria is paramount to ELs’ long-term success. For example, if policymakers set the reclassification criteria too low or too high, students may experience adverse effects of reclassification or prolonged tenure in EL status, respectively.

Although states establish standardized criteria for reclassification eligibility, meeting reclassification criteria does not guarantee that a student will reclassify. Existing studies have found that in many districts, a substantial number of eligible students are not reclassified due to variations in local policy interpretation and implementation (Cimpian et al., 2017; Estrada & Wang, 2018; Mavrogordato & White, 2020). For example, in a mixed-methods case study of reclassification procedures and outcomes in two California school districts, Estrada and Wang (2018) report that while one district reclassified nearly all students who met the criteria, another district reclassified only 67% of eligible students. Ultimately, the authors conclude that several factors drive differences in reclassification likelihood for eligible students, including excessive administrative burden on school district leaders to formally reclassify a student (e.g., requiring signature forms from parents, errors in applying criteria, lack of district monitoring of students and procedures) and staff perceptions of the benefits or drawbacks of reclassification compared to remaining EL-classified for specific students.

To better understand between-district variation in reclassification outcomes for students who meet reclassification criteria, Cimpian et al. (2017) compare two states using a regression discontinuity (RD) design. Across both states, the authors report substantial between-district and between-grade variation in how meeting test-based reclassification criteria predicts a student’s likelihood of reclassifying. For example, among districts with below-average reclassification rates in one state, meeting the reclassification criteria did not influence a student’s likelihood of reclassifying. In contrast, meeting the same criteria in districts with above-average reclassification rates significantly increased students’ likelihood of reclassifying. These studies suggest a complex interplay between reclassification criteria and outcomes, and they indicate that several factors, including district characteristics, district-level policies, procedural burden, and staff attitudes or knowledge, contribute to variation in reclassification likelihood among students who meet the established criteria.

Other studies identify heterogeneity in eligible students’ likelihood of reclassification based on their grade level. For example, Robinson (2011) finds that as EL students in California progress through school, they are less likely to reclassify upon meeting the criteria. Specifically, the reclassification rate for fourth graders meeting all criteria was 91%, compared to 64% for 10th graders, signifying greater teacher discretion in the reclassification process in high school. In contrast, using data from one large California school district, Umansky and Reardon (2014) report that 12% of ELs who meet test-based reclassification criteria do not reclassify in fifth grade. However, more ELs reclassify in 11th grade than qualify, suggesting educators are more likely to perceive an urgent need to reclassify students in later grades (Umansky & Reardon, 2014).

Finally, educators’ perceptions of EL students from different racial or ethnic backgrounds may also inform their reclassification decisions. For example, Umansky et al. (2020) report a higher likelihood of reclassification among Chinese-origin than Latinx ELs, even when Chinese-origin ELs do not meet reclassification criteria. Furthermore, Mavrogordato and White (2017) find that reclassification-eligible students who speak languages other than Spanish are five percentage points more likely to be reclassified than their Spanish-speaking peers. In summary, findings regarding heterogeneity in the reclassification rates of eligible students suggest that the reclassification processes can be influenced by various factors, including grade level, teacher discretion, and perceived urgency, highlighting the complexity of manual reclassification decisions. Given these collective findings, researchers suggest that reliance on more objective and standardized reclassification policies can improve discrepancies in reclassification for eligible students (Estrada & Wang, 2018; Okhremtchouk et al., 2018).

Default Policies

An automatic or default policy is a standard or predetermined choice automatically applied to an action if no alternative option is chosen (Herd et al., 2013). Policymakers establish defaults that reflect their preferred choice, and individuals must take deliberate action to opt for something different. Ultimately, policymakers implement these preselected choices because they are a subtle but powerful way to influence decisions and increase the uptake of a preferred option (Jachimowicz et al., 2019).

In practice, default policy options provide a straightforward means of implementation. For example, a well-known default policy is organ donation registration. This policy is often implemented by stating, “You are currently registered as an organ donor. Do you not want to be an organ donor?” By default, individuals are enrolled to be organ donors and must take deliberate action if they wish to opt out (Johnson & Goldstein, 2003). Despite their simplicity, default policies substantially influence individuals’ decision-making. Defaults have proven to be practical policy tools across a wide range of social issues, increasing organ donation rates (Johnson & Goldstein, 2003), voter registration (Garnett, 2022), and retirement savings (Madrian & Shea, 2001). Generally, the literature on default effects finds that decision-makers are likelier to choose the default option than an alternative (Jachimowicz et al., 2019).

Research studying the efficacy of default effects in education policy is small but growing and, thus far, has focused on higher education rather than K–12 education policy. Behlen and colleagues (2023) investigate the impact of defaults on universities’ final exam sign-up procedures in contexts that require students to register for final exams. They find that under default enrollment in final exams, students are more likely to participate in and succeed in final exams, underscoring the effectiveness of defaults in education. Additionally, Cox and colleagues (2020) provide insights into the factors that affect student loan borrowers’ decisions to opt for income-driven repayment plans instead of other loan repayment plans that may increase their chances of defaulting. Their findings highlight the importance of defaults and information provision in decision-making processes in this area and demonstrate that implementing income-driven repayment plans as the default can substantially decrease students’ likelihood of choosing riskier repayment plans. These studies indicate that defaults can be promising policy levers to improve students’ short- and long-term outcomes.

Specific to the EL reclassification policy, automatic procedures require districts to opt out students who meet reclassification criteria, while manual procedures require districts to actively opt in students to be reclassified. Many states currently employ a manual reclassification procedure in which school districts are responsible for identifying and reclassifying eligible students in district data systems. A handful of states have moved to an automatic reclassification procedure, which leverages state administrative data systems to automatically identify and reclassify eligible students upon receiving standardized test scores. Automatic procedures eliminate the burden on districts to identify and complete reclassification paperwork for eligible students, factors that contribute to disparities in the reclassification of eligible students (Estrada & Wang, 2018). In addition, they implicitly convey to districts that reclassification is the status quo for students who meet the criteria. It is also important to note that automatic reclassification policies do not imply that districts lose agency in determining which students reclassify. Exiting an EL student who has met eligibility criteria is the default, but automatic policies can be written such that districts can intervene and opt out of reclassification for individual students if they have reasons to believe continuing to receive English language development services would benefit the student.

Whether a manual or an automatic reclassification procedure leads to more equitable decision-making remains an open question. On one hand, manual policies allow for flexibility and nuance at the local level. These policies implicitly encourage local discretion by requiring district personnel to reclassify eligible students individually. For example, educators may identify students who meet test-based reclassification criteria but would benefit from continued EL services and retain that student in services. On the other hand, manual policies allow for differential treatment of similarly situated students. For example, Mavrogordato and White (2017) report that under manual reclassification procedures, educators at times rely on data unrelated to English proficiency, such as students’ personality traits and behavior. In these cases, a student may remain classified and receive EL services when they are in need of other, unrelated supports. In this case, automatic policies may result in more equitable decision-making because they remove the manual determination by districts.

Consistent with existing literature examining default policy effects, we hypothesize that an automatic reclassification policy will increase reclassification rates among eligible students compared to a manual reclassification policy. Additionally, because automatic reclassification is based solely on students’ test scores and completed via state data systems, we anticipate that the automatic policy will close gaps in reclassification rates across subgroups of eligible EL students whose prior research has identified as less likely to reclassify when eligible, particularly Spanish speakers (Mavrogordato & White, 2017; Umansky et al., 2020).

Michigan Policy Context

Michigan serves a linguistically diverse and growing number of EL students. In the past 10 years, the Michigan EL population has nearly doubled, and in 2023, ELs comprised 98,771 students, roughly 6.9% of Michigan’s K–12 student population.

Following federal requirements that states annually assess ELs’ English proficiency growth, Michigan and 40 other states use the WIDA ACCESS 2.0 (hereafter, WIDA) English proficiency assessment to evaluate ELs. All Michigan ELs take the WIDA assessment and other statewide standardized tests each spring. WIDA consists of four domains (listening, speaking, reading, and writing). Students receive a scale score for each domain and an overall scale score ranging from 100 (lowest score) to 600 (highest score; WIDA, 2024). Ultimately, scale scores correspond to an interpretive “proficiency level” score ranging from 1.0 (low) to 6.0 (high; WIDA, 2024). Scale score interpretations vary across grades, while proficiency levels can be compared across grades. We detail the history of Michigan’s reclassification criteria and procedures below.

Michigan districts serving ELs are legally obligated to provide students with language support services that enable them to access academic content in English. To meet this goal, Michigan provides supplemental per-pupil funding to districts serving ELs. This tiered funding system provides more funding to students with beginning English proficiency levels and less to those with more advanced English proficiency. Students with WIDA scores above 4.0 do not generate additional funding for their districts (Michigan Department of Education [MDE], 2023), so we do not anticipate financial incentives to keep students classified as ELs after they have attained English proficiency.

Manual Reclassification, 2016–17 through 2018–19

Before the 2019–20 school year, Michigan required students to meet multiple test-based criteria to qualify for reclassification. In addition, districts were responsible for manually reclassifying students in the state data reporting system. To qualify for reclassification, students needed to attain (a) a WIDA overall score of 4.5, (b) a WIDA reading domain score of 4.0, (c) a WIDA writing domain score of 4.0, and (d) score “proficient” in a locally chosen reading assessment. Regarding the locally chosen reading assessment criterion, districts were permitted to choose from several pre-approved options (e.g., NWEA, AIMSWeb, DIBELS Next, iReady Diagnostic, Star Early Literacy, and the state standardized reading assessment, M-STEP) and define the minimum score required to be considered “proficient,” creating variation across districts in the score needed to demonstrate proficiency and the overall difficulty of the chosen assessment (personal communication with MDE, 2023). For example, some districts set their proficiency threshold at the 25th percentile of their locally chosen assessment, while others set it at the 75th percentile (personal communication with MDE, 2023). Districts were not required to submit proficiency thresholds or student performance data on locally chosen reading assessments to the state. However, the proficiency threshold was generally lower on locally chosen reading assessments than on statewide standardized English Language Arts (ELA) assessments (personal communication with MDE, 2023). During the manual reclassification process, school district administrators identified and reclassified EL students who met state reclassification criteria using a multi-step process depicted in Figure 1.

Figure 1.

Manual reclassification procedure in Michigan, 2016–17 through 2018–19.

Automatic Reclassification, 2019–20 Through 2021–22

Acknowledging disparities between the number of students meeting reclassification criteria and those actually reclassifying, Michigan shifted from a manual to an automatic reclassification policy in the fall of 2019. Changes to the reclassification protocol in the fall of 2019 applied to the 2019–20 school year ² and beyond.

In shifting to an automatic process, MDE assumes responsibility for reclassifying students through state administrative data systems when the state receives annual WIDA scores from the WIDA consortium. While districts have the opportunity to review students identified for reclassification and can override reclassification decisions if they feel a student is unprepared to reclassify despite meeting criteria, the procedure automatically reclassifies eligible students. Importantly, this process removes all responsibility from districts to reclassify students in district and state data systems.

Beyond the shift to automatic reclassification, MDE made two significant changes to the reclassification criteria. First, MDE simplified the criteria to require only one assessment score as evidence of English proficiency. The revised reclassification criteria eliminated the WIDA reading, WIDA writing, and local ELA assessment proficiency thresholds. Second, the state raised the overall WIDA performance level required to qualify for reclassification from 4.5 to 4.8 out of 6.0. Table 1 outlines the differences in each period’s reclassification criteria and procedures.

Table 1

Reclassification Criteria and Procedures by Policy Period

Reclassificiation Policy Component	Manual reclassification (2016–17 through 2018–19)	Automatic reclassification (2019–20 through 2021–22)
Criteria	• Overall WIDA performance level of 4.5 or greater • WIDA reading performance level of 4.0 or greater • WIDA writing performance level of 4.0 or greater • “Proficient” in a locally chosen reading assessment	• Overall WIDA performance level of 4.8 or greater
Procedure	• District convenes a team and completes a nine-step reclassification process for qualified students	• State data system automatically reclassifies qualified students • Districts have the option to override automatic reclassification

Data

The data for our analyses come from the Michigan Education Data Center (MEDC) and include observations for all third through eighth grade EL students with valid WIDA scores between academic years 2016–17 through 2022–23 (N = 345,044). We restrict our sample to include only EL students who met at least a 4.0 reading and 4.0 writing performance level on the WIDA assessment and have a valid state standardized ELA test score (N = 59,045). We make this restriction to facilitate comparisons across policy periods, as these students would have been relatively close to the reclassification cutoff in both the manual and automatic reclassification periods. In addition, this restriction accounts for students needing to meet a 4.0 reading and 4.0 writing WIDA performance level and score proficient on a reading assessment to qualify for reclassification in the manual reclassification period.

Table 2 presents summary statistics for the primary analytic sample by policy period (manual reclassification = pre-period; automatic reclassification = post-period). About 35% and 26% of students reported speaking Spanish or Arabic as their home language, respectively. Approximately 4% of students were also classified as students with disabilities, and roughly 72% were identified as low-income. About 77% of students in the sample were enrolled in elementary grades (third through fifth), and about 23% were in middle grades (sixth through eighth). Students’ overall, reading, and writing performance level scores were similar across policy periods. This provides some evidence that students in our sample were similar academically before and after the policy change. WIDA overall and subdomain performance levels are similar across policy periods.

Table 2

Descriptive Statistics by Policy Period

Variable	Reclassification period
	Manual period	Automatic period	Full sample
	(2016–17 through 2018–19)	(2019–20 through 2022–23)
Covariate means (% of sample)
Female	54	55	54
Primary language: Spanish	36	31	35
Primary language: Arabic	25	27	26
Primary language: Other	37	42	39
SWD	4	4	4
Low-income	72	71	72
Grade (% of sample)
3	19	18	19
4	28	37	31
5	25	30	27
6	9	5	7
7	9	5	8
8	9	6	8
Test scores
WIDA overall scale score recentered around post-period reclassification threshold	2.19 (15.49)	2.17 (15.68)	2.18 (15.56)
WIDA overall performance level	4.87 (0.52)	4.87 (0.52)	4.87 (0.52)
WIDA reading performance level	5.58 (0.56)	5.51 (0.59)	5.56 (0.57)
WIDA writing performance level	4.35 (0.33)	4.36 (0.39)	4.35 (0.35)
Percent proficient on state standardized ELA assessment	50.56	49.72	50.26
Reclassification (%)
Qualified for reclassification	80	54	70
Reclassified	51	53	52
N	32,124	18,097	50,221

Note. Data for this analysis come from the Michigan Education Data Center. WIDA scale scores are recentered around their respective grade-level reclassification threshold for interpretation across grades (e.g., a recentered scale score of −1 can be interpreted as meaning the student missed the reclassification threshold in their grade and policy period by 1 point). The WIDA assessment reports scale scores between 100–600, and performance level scores are reported on a scale of 1.0–6.0. State ELA assessment = M-STEP ELA (Michigan’s state standardized ELA assessment for grades 3–7) or Preliminary-SAT reading score (grade 8). A student is considered “qualified for reclassification” in the pre-period if they met the grade-level WIDA overall scale score threshold for reclassification in the pre-period, and in the post-period if they met the grade-level WIDA overall scale score threshold for reclassification in the post-period. SWD = student with disabilities; ELA = English Language Arts.

To compare scale scores across grades in our analysis, we recenter the scale scores around 0, and 0 represents the minimum scale score required to qualify for reclassification in a given grade. As an example, a third-grade student who attained an overall scale score of 357 in the manual reclassification period would have a value of 1 (the reclassification threshold is 356 for third graders), and a fourth-grade student who attained an overall scale score of 367 in the manual reclassification period would also have a value of 1 (the reclassification threshold is 366 for fourth graders). Recentering the standardized scale scores allows us to estimate the effects of changing policy procedures for the total sample of students.

Recentering within grades also facilitates comparisons between the manual and automatic reclassification policy periods. MDE increased the overall WIDA score needed to qualify for reclassification between periods. To facilitate comparisons across policy periods, we recenter students’ scores around the post-period reclassification threshold. On average, ELs in our sample scored 2.18 points above the post-period reclassification threshold. To address changes made to the overall scale score reclassification threshold, we simulate raising the reclassification threshold in the pre-period to match the post-period threshold in our primary analysis. We discuss methods used to simulate raising the threshold in detail in the section titled “Endogeneity Issues.”

Of note, the state did not collect data on locally chosen reading assessments, which comprised one additional component of reclassification criteria during the manual reclassification period. To account for this, we use standardized, statewide ELA assessments (M-STEP ELA for third- through seventh-graders and PSAT Reading for eighth graders) as a proxy for students’ ELA proficiency on locally chosen assessments. Statewide ELA assessments were likely to be as difficult or more difficult than locally chosen reading assessments, so they should serve as a strong indicator for students’ performance on local reading assessments (personal communication with MDE, 2023). Due to the COVID–19 pandemic, statewide standardized ELA assessments were not administered during the 2019–20 school year. As a result, we do not include observations from 2019–20 in our main sample. ³

We consider students “qualified for reclassification” if they meet the WIDA overall performance thresholds. The sample in Table 2 reflects the set of students who met WIDA reading and writing performance thresholds of 4.0 (criteria for reclassification eligibility in the manual reclassification period). The manual reclassification period shows large gaps between the number of students eligible for reclassification and those who were reclassified (80% vs. 51%). In contrast, we see similar percentages of students qualifying and reclassifying in the automatic reclassification period (54% vs. 53%). The overall percent of students qualifying for reclassification likely shrunk between policy periods due to raising the overall score needed to qualify for reclassification and an overall decrease in student performance in Michigan after the COVID–19 pandemic (Kilbride et al., 2024). We discuss how we account for raising the reclassification threshold in the “Endogeneity Issues” section of this article. Although students’ performance declined in Michigan following the pandemic, student performance showed signs of progress beginning in the 2022–23 school year, which we incorporate into our post-period analytic sample. Our primary outcome of interest is whether an EL student reclassifies upon meeting reclassification criteria. Table 2 displays the percentage of students in our main sample who met each reclassification threshold under manual and automatic reclassification policies.

Endogeneity Issues

Our estimation strategy faces two primary endogeneity threats. First, when MDE shifted from manual to automatic reclassification, they also eliminated three components of the reclassification criteria (see Table 1). Second, MDE increased the overall WIDA score needed to qualify for reclassification upon shifting to an automatic reclassification process. This section discusses how our analysis accounts for these endogeneity threats.

Accounting for Eliminating Reclassification Criteria

To address the removal of WIDA reading, writing, and local reading assessment scores from state reclassification criteria, we apply a “frontier RD” approach to both the pre- and post-period samples (Reardon & Robinson, 2012). Frontier RD models subset the sample of students used in the analysis to those with scores above or below the cutoff score on all dimensions but one, then model the RD along only one cutoff score (Reardon & Robinson, 2012). Here, we subset the sample to students with at least a 4.0 performance level on the WIDA reading and writing subdomains required to qualify for reclassification during the pre-period. We also include a covariate for students’ recentered and standardized M-STEP ELA or PSAT reading scores in Equations (1a) and (1b) as a proxy for proficiency on local reading assessments. The resulting frontier RD sample produces results that are easily interpretable, reduces a multidimensional problem (shifting multiple reclassification criteria simultaneously) to a single dimension (only estimating the effect of shifting manual to automatic reclassification policies by accounting for other factors through sample selection), and isolates the effect of shifting from manual to automatic reclassification (Reardon & Robinson, 2012).

Accounting for Raising the Reclassification Eligibility Threshold

The second endogeneity threat concerns the change to the reclassification eligibility threshold for students’ overall WIDA scale scores. MDE increased the overall scale score required to qualify for reclassification across policy periods. Students were eligible for reclassification if they achieved a WIDA overall performance level of 4.5 during the pre-period and 4.8 during the post-period. To ensure comparisons between similar students who would have met reclassification criteria in either period, we simulate raising the reclassification threshold in the pre-period to match the post-period threshold. To create a sharp RD cut point at the higher reclassification threshold in the pre-period models, we assume that all students who did not meet the post-period reclassification threshold were not reclassified. In reality, some of the students below the simulated threshold were reclassified (as demonstrated in Figure 2, Upper Bound Intent-to-Treat [ITT] Estimate). As a result, our simulated sample (Figure 2, Lower Bound ITT Estimate) will provide a lower-bound estimate of the effect of shifting reclassification procedures because the simulated ITT effect in the pre-period is larger than in reality. Formally, this implies that we inflate the pre-period ITT effect estimate. This inflated estimate is subtracted from the post-period estimate such that $\mathrm{DiRD}={\beta }_1^{\mathrm{post}}-{\beta }_1^{\mathrm{pre}}$ . We report results for estimates using the simulated 4.8 reclassification threshold for comparison across policy periods. Figure 2 also includes an analogous post-period figure for reference. Table 3 provides information on analytic sample selection and reclassification eligibility and rates by year based on the post-period reclassification cutoff score.

Figure 2.

Upper- and lower-bound pre-period ITT estimates.

Table 3

Analytic Sample Selection and Classification Rates by Year

Year	Total students (N)	Eligible to reclassify based on post-period cutoff (N)	Eligible to reclassify based on post-period cutoff (%)	Qualified to reclassify and reclassified (N)	Reclassification rate of eligible students based on post-period threshold
2016–17	9,376	5,130	55	3,154	61
2017–18	10,801	6,192	57	4,460	72
2018–19	11,947	6,017	50	4,369	73
2019–20	Omitted due to COVID–19 pandemic
2020–21	4,370	2,174	50	2,144	99
2021–22	7,087	3,979	56	3,904	98
2022–23	6,640	3,637	55	3,596	99

Research Methods

We use two approaches to estimate the effect of qualifying for reclassification on reclassifying during the manual versus automatic reclassification periods. First, we use a sharp RD analysis to estimate the ITT effect of qualifying for reclassification on reclassifying during each of the two policy periods. The ITT effect estimates the impact of meeting or exceeding the reclassification threshold (recentered WIDA scale score) on the outcome (reclassification). Then, we use a DiRD approach to compare the two ITT effect estimates (Robinson-Cimpian & Thompson, 2016). The difference obtained from the DiRD framework provides a plausibly causal estimate of the effect of shifting from a manual to an automatic reclassification process.

Substantial research indicates that meeting reclassification criteria differentially impacts students’ likelihood of reclassifying based on their grade level (Robinson, 2011; Umansky & Reardon, 2014). To address this, we estimate all RD models separately for grade-level subsets. We present results for grade-level subsamples of students as well as a weighted average effect of the policy for all students in the sample.

Sharp RD Estimates

We first estimate the ITT effect of meeting the overall WIDA reclassification threshold in the “pre” period (manual reclassification) for student i in grade g:

$$\begin{array}{c}{Y}_{ig}^{\mathrm{pre}}={\beta }_0^{\mathrm{pre}}+{\beta }_1^{\mathrm{pre}}{C}_{ig}+f{\left(M_{ig}\right)}^{\mathrm{pre}}\\ \left[+X_{ig}{\beta }^{pre}\right]+v_{ig}^{\mathrm{pre}}.\end{array}$$

(1a)

This RD model predicts student i’s likelihood of reclassifying Y as a function f ⁴ of their recentered and standardized overall WIDA scale score M, an indicator for whether or not that score is above the recentered reclassification threshold C, and in some specifications, a vector X of additional covariates (recent immigrant status, special education status, low-income status, gender, home language, prior year overall WIDA score).

We restrict the bandwidth of WIDA scale scores used in the analysis to limit the influence of outlier students with very high or very low WIDA scores using the rdrobust command, which implements a data-driven process to determine an optimal bandwidth and estimates bias-corrected coefficients and robust standard errors (Calonico et al., 2014). We report results from the optimal bandwidths chosen by the rdrobust command for each grade, but we also report results for ½ and twice the size of the optimal bandwidths as robustness checks. Our preferred model uses a triangular kernel, but we also report results using a uniform kernel as a robustness check. For all RD models, we cluster standard errors at the school district level because districts were responsible for manually reclassifying students in the pre-period and overriding automatic reclassification in the post-period. The rdrobust command also adjusts for mass points in determining the bandwidth, meaning that it accounts for the running variable being less than fully continuous, as is the case with most education studies.

In Equation (1a), ${\beta }_1^{\mathrm{pre}}$ represents the ITT effect of just barely qualifying for reclassification on reclassifying in the pre-period (manual reclassification). Equation (1b) is a corresponding analysis for the “post,” automatic reclassification period:

$$\begin{array}{c}{Y}_{ig}^{\mathrm{post}}={\beta }_0^{\mathrm{post}}+{\beta }_1^{\mathrm{post}}{C}_{ig}+f{(M_{ig})}^{\mathrm{post}}\\ [+X_{i}g{\beta }^{\mathrm{post}}]+v_{ig}^{\mathrm{post}}.\end{array}$$

(1b)

The ITT estimates for each policy period and grade-level subsample give the impact of qualifying for reclassification on reclassifying in each policy period. ⁵

Difference-in-Regression Discontinuities

Next, we use a DiRD approach in an attempt to estimate the impact of shifting from manual to automatic reclassification on eligible students’ likelihood of reclassifying. The DiRD approach will estimate the difference in ITT effects from Equations (1a) and (1b). This estimate can inform policy by indicating whether the change altered eligible students’ likelihoods of reclassifying. Equation (1c) estimates the DiRD separately for each grade level using the sample of students included in the optimal bandwidths. ⁶ Estimates can be interpreted as the causal effect of shifting from a manual to an automatic reclassification process if there are no other confounding factors and are obtained by subtracting the post- and pre-period ITT effects:

$$\mathrm{DiRD}={\beta }_1^{\mathrm{post}}-{\beta }_1^{\mathrm{pre}}$$

(1c)

where $\mathrm{DiRD}$ is the coefficient estimate of the difference between ITT estimates ( ${\beta }_1^{\mathrm{post}},{\beta }_1^{\mathrm{pre}}$ ) in the post- and pre-periods.

Although we estimate $\mathrm{DiRD}$ separately for each grade level, we also report estimated effects for the full sample of students. These estimates are precision-weighted by grade level. To do this, we calculate a weighted average estimate inversely proportional to the standard error of each grade-level estimate. Equation (2) calculates the weighted average estimate of the effect of qualifying for reclassification on reclassifying: ⁷

$$\begin{array}{l}\mathrm{Weighted}\mathrm{Average}\mathrm{DiRD}\\ =\frac{{\Sigma }_i(\frac{1}{var{(DiRD)}_g}\times Grade-LevelEstimat{e}_i)}{{\Sigma }_i\frac{1}{var{(DiRD)}_g}}\end{array}$$

(2)

The weighted average DiRD is the combined estimate for all grades with weights inversely proportional to the variance; $\mathrm{var}{(\mathrm{DiRD})}_i$ represents the variance of the estimate for grade level g; $\mathrm{Grade}-{\mathrm{Level}\mathrm{Estimate}}_i$ is the estimate for grade level g. The weighted average effect provides an overall estimate of the impact of the policy change on all third to eighth-grade students (e.g., a weighted average effect of 0.26 corresponds to a 26 percentage-point increase in an eligible student’s likelihood of reclassifying after the policy change).

Subgroup Analyses

We next use a difference in DiRD (DiDiRD) framework to evaluate differential changes in reclassification rates related to the policy change for subgroups of ELs reporting different home languages. The estimate obtained from the subgroup analyses provides preliminary evidence of the ability of the policy change to ameliorate differential reclassification outcomes unrelated to English proficiency level. We begin by subsetting our sample to students in three subgroups based on their reported home language (Spanish, Arabic, and Other Home Language) and re-estimating Equations (1a) through (1c) and Equation (2). Then, Equation (3) estimates the DiDiRD separately for each grade level subgroup of students. ⁸ Estimates can be interpreted as the effect of shifting from a manual to an automatic reclassification process for a given subgroup of students and are obtained by subtracting the DiRD estimates for subgroups of students:

$$\mathrm{DiDiRD}={\mathrm{DiRD}}_{\mathrm{subgroup}1,\mathrm{g}}-DiR{D}_{\mathrm{subgroup}2,\mathrm{g}},$$

(3)

where $\mathrm{DiDiRD}$ is the estimate of the difference in DiRD estimates for a subgroup of students in grade $g$ . For example, if testing for a differential effect of the policy across home language, we would subtract the DiRD estimate for Spanish speakers from that of Arabic speakers in grade $g$ .

Internal Validity of Estimates

Our DiRD design relies on several assumptions to produce a causal estimate of the effect of shifting from manual to automatic reclassification. First, we assume that the running variable is not manipulated at the cutoff. Because educators and EL students know the cutoff score required to qualify for reclassification, they may act to manipulate student scores to either retain or reclassify students from EL status, potentially threatening the validity of our estimates. We test the assumption that the running variable is not manipulated at the cutoff using a McCrary test. McCrary (2008) suggests there should be no discontinuity in observations at the cutoff for this assumption to hold. A spike in observations on either side of the cutoff may indicate score manipulation. We report results from McCrary tests in Supplemental Appendix Table A1 and Supplemental Appendix Figures A1 and A2 to demonstrate no discontinuities in recentered scale scores at the cutoff using triangular and uniform kernels, confirmed using the rddensity command in Stata.

Next, the RD assumes that only treatment and outcomes change discontinuously at the cutoff. In other words, although treatment status should change at the cutoff, students must be otherwise similar on either side of the cutoff. If this assumption holds, the RD design produces causal estimates of the effect of the policy change, as the groups of students on either side can be used as counterfactuals for one another. Although we cannot test this assumption for unobservable student characteristics, we conduct tests for observable factors (such as gender, special education status, low-income status, and home language). We tested these factors by running separate RDs by grade level and policy period for the analytic sample, each time substituting a different variable as the outcome of interest. Results indicate that no observable student characteristics vary discontinuously at the cutoff other than reclassification likelihood. Supplemental Appendix Table A2 displays the results of these tests.

Finally, for the DiRD to be interpreted as a causal effect, there must be no other co-occurring change—other than those already addressed above, such as the shift in the threshold—that changed and could account for the manual-to-automatic reclassification effect estimate. Ideally, we could address this assumption through the use of an unaffected comparison group via a DiDiRD approach, but there is no unaffected group (e.g., never-ELs) who take the WIDA assessment and are not affected by either one of these policies. As such, there is no comparison that can be used to remove any secular trend from the reclassification policy change. While noting this caveat, it is also worth noting that raw reclassification rates based on the WIDA overall cutoff score for each respective policy period are fairly stable from year to year in Michigan, as we demonstrate in Table 4. Additionally, there are no other policy changes that we are aware of that could produce a discontinuous change in reclassification rates at the threshold in either the pre- or post-period. Thus, a sizable change from the pre- to post-period in the DIRD is plausibly attributable to the change from manual to automatic reclassification.

Table 4

Raw Reclassification Rates for Eligible ELs by Year

Year	Students eligible to reclassify (N)	Students reclassified (N)	Percentage of eligible students reclassified
2016–17	7,560	4,235	56
2017–18	8,818	5,811	66
2018–19	9,165	6,120	67
2019–20	5,104	4,936	97
2020–21	2,174	2,144	99
2021–22	3,979	3,904	98
2022–23	3,637	3,596	98

Note. Data come from the Michigan Education Data Center. Automatic reclassification was implemented in the 2019–20 school year. Between 2016–17 and 2018–19, students in the sample were eligible to reclassify if they attained a 4.5 overall WIDA performance level. Between 2019–20 and 2022–23, students in the sample were eligible to reclassify if they attained a 4.8 overall WIDA performance level. EL = English learner.

Results

Effects of Automatic Policy on the Likelihood of Reclassification

In this section, we will focus on the results of our lower-bound estimates, which provide the strongest test of the policy change and most directly address credible threats to internal validity of the design. We note that the patterns of significant effect estimates reported here for the lower-bound estimates also—and unsurprisingly—hold for the upper-bound estimates of the effect. See Supplemental Appendix Table A3 for the upper-bound estimates.

We now focus on our preferred model, which we argued yields a lower-bound estimate of the effect. Table 5 presents the results from this preferred model for reclassification likelihood by grade and policy period for the main model specification and an overall effect of the policy change by grade. Supplemental Appendix Table A4 displays results by alternative bandwidth and kernel specifications. Overall, we estimate a significant discontinuity (p < .001) in students’ likelihood of reclassifying upon meeting the reclassification threshold in both the pre- and post-period. This finding holds for all grade levels and weighted average effects in each policy period. Of note, the magnitude of the jump in eligible students’ likelihood of reclassifying varies by grade level. For example, during manual reclassification, eligible third graders were the most likely to reclassify ( ${\beta }_1^{\mathrm{pre}}$ = .806, p < .001). In other words, a third grader just above the reclassification threshold experienced an 80.6 percentage-point increase in their likelihood of reclassifying compared to a peer just below the threshold. In contrast, eligible fifth graders were least likely to reclassify ( ${\beta }_1^{\mathrm{pre}}$ = .393 [39.3 percentage points], p < .001). Although a fifth grader just above the reclassification threshold was more likely to reclassify than a peer just below the threshold, the effect of meeting the reclassification threshold was substantially smaller than in other grades. During manual reclassification, we estimate a weighted average effect of 0.634 (p < .001), meaning, on average, meeting the reclassification threshold increased a student’s likelihood of reclassifying by 63.4 percentage points.

Table 5

Lower-Bound Estimated Effect of Qualifying for Reclassification on Reclassifying Across Grades and Policy Periods

Grade Level	Manual reclassification	Automatic reclassification	Policy change
	2016–17 through 2018–19	2019–20 through 2021–22	(DiRD)
3	0.806***	0.979***	0.172***
SE	(0.023)	(0.013)	(0.026)
BW	[−50, 18]	[−11, 8]
N	5,566	1,453
4	0.561***	0.981***	0.420***
SE	(0.036)	(0.007)	(0.036)
BW	[−48, 25]	[−14, 24]
N	7,978	5,154
5	0.393***	0.972***	0.579***
SE	(0.035)	(0.013)	(0.038)
BW	[−48, 26]	[−10, 24]
N	7,281	3,957
6–8	0.580***	0.985***	0.405***
SE	(0.026)	(0.011)	(0.028)
BW	[−57, 18]	[−10, 9]
N	8,276	1,327
Weighted average	0.634***	0.980***	0.356***
SE	(0.014)	(0.005)	(0.016)
Standardized ELA score control	X
Local polynomial	1
Bandwidth	Optimal
Kernel	Triangular

Note. Robust standard errors clustered at the school district level appear in parentheses below the point estimates. We restrict the bandwidth of WIDA scale scores used in the analysis to limit the influence of outlier students with very high or very low WIDA scores using the rdrobust command, which implements a data-driven process to determine an optimal bandwidth and estimates bias-corrected coefficients and robust standard errors (Calonico et al., 2014). SE = standard error; BW = bandwidth; DiRD = difference-in-regression discontinuity; ELA = English Language Arts.

***

p < .001.

During the automatic reclassification period, students just above the reclassification threshold experienced much greater reclassification likelihood than those just below. On average, students just above the cut point were 98.0 percentage points more likely to reclassify than those just below. Further, there is less grade-level variation in eligible students’ likelihood of reclassifying during the automatic period. For example, eligible middle schoolers are the most likely to reclassify of all grade levels ( ${\beta }_1^{\mathrm{post}}=$ .985 [98.5 percentage points], p < .001), and eligible fifth graders are the least likely to reclassify ( ${\beta }_1^{\mathrm{post}}=$ .972 [97.2 percentage points], p < .001).

Estimated ITT effects across policy periods imply that manual and automatic reclassification features differentially impacted eligible students’ likelihood of reclassifying. Across all grade levels, we find a statistically significant DiRD estimate. The DiRD estimate is largest in fifth grade ( $\mathrm{DiRD}$ = 0.579, p < .001), where shifting from manual to automatic reclassification increased eligible students’ likelihood of reclassifying by 57.9 percentage points. This finding suggests that, of all grade levels, automatic reclassification “leveled the playing field” the most for fifth graders, who were the least likely of all grade levels to reclassify in the pre-period. In contrast, the DiRD effect is smallest in the third grade ( $\mathrm{DiRD}=$ 0.172, p < .001), meaning shifting from manual to automatic reclassification increased eligible third graders’ likelihood of reclassifying by roughly 17.2 percentage points. Although the DiRD estimate is the smallest in third grade, the policy change still resulted in a statistically significant increase in eligible students’ likelihood of reclassifying. The weighted average effect of the policy change is 0.356 for all students. On average, the policy change across the sample resulted in a 35.6 percentage point increase in any third through eighth-grade student’s likelihood of reclassifying upon meeting state reclassification criteria.

Differences in Policy Change Effects Across Language Subgroups of ELs

Prior research has found that eligible students’ likelihood of reclassifying varies based on students’ racial and linguistic backgrounds (Mavrogordato & White, 2017; Umansky et al., 2020). In other areas of education policy, automatic procedures are being implemented to ensure greater racial equity in service provision (e.g., automatic advanced course enrollment for students who score at the top of a standardized test distribution, Berg & Plucker, 2023). Where sample sizes allow, we evaluate the effect of the automatic reclassification policy across subgroups of ELs to test its ability to increase standardization in reclassification rates using a DiDiRD framework. We find that the shift to automatic policy had a larger effect on students reporting Spanish as a primary language compared to students reporting other primary languages. This implies that under manual reclassification, eligible Spanish speakers were less likely to reclassify than students speaking other primary languages, and automatic reclassification ameliorated some of the difference in reclassification rates for students reporting different home languages.

Among students near the reclassification threshold, we estimate that shifting from a manual to an automatic reclassification policy affected Spanish speakers more than students reporting another home language. Table 6 presents the DiRD point estimates of shifting from manual to automatic reclassification for students reporting Spanish and Arabic as their home languages. We compare the effects of the policy change for Spanish and Arabic speakers because these are the two most commonly spoken languages among Michigan ELs. We report comparisons to students reporting another primary language in Supplemental Appendix Table A5. Results are consistent with those reported here.

Table 6

DiRD Estimates Across Language Subgroups of ELs

Grade level	Primary language: Spanish			Primary language: Arabic
	Manual reclassification	Automatic reclassification	Policy change	Manual reclassification	Automatic reclassification	Policy change
	2016–17 through 2018–19	2019–20 through 2021–22	(DiRD)	2016–17 through 2018–19	2019–20 through 2021–22	(DiRD)
3	0.663***	0.964***	0.300***	0.844***	1.000***	0.156***
SE	(0.064)	(0.330)	(0.072)	(0.029)	0.000	(0.029)
Bandwidth	[−50, 8]	[−11, 9]		[−44, 9]	[−41, 3]
N	1,283	342		1,274	637
4	0.527***	0.972***	0.445***	0.543***	0.992***	0.450***
SE	(0.053)	(0.020)	(0.057)	(0.048)	(0.007)	(0.049)
Bandwidth	[−40, 19]	[−42, 13]		[−42, 19]	[−6, 14]
N	2,765	1,485		1,978	955
5	0.358***	0.956***	0.598***	0.396***	0.975***	0.579***
SE	(0.045)	(0.032)	(0.055)	(0.064)	(0.020)	(0.067)
Bandwidth	[−48, 21]	[−55, 19]		[−44, 23]	[−7, 12]
N	2,886	1,683		1,864	790
6–8	0.575***	0.953***	0.377***	0.596***	1.001***	0.405***
SE	(0.040)	(0.029)	(0.050)	(0.039)	(0.015)	(0.042)
Bandwidth	[−52, 13]	[−8, 10]		[−51, 14]	[−6, 13]
N	3,299	443		1,933	248
Weighted average	0.515***	0.964***	0.440***	0.684***	1.000***	0.350***
SE	(0.024)	(0.014)	(0.028)	(0.020)	0.000	(0.020)
Standardized ELA score control	X			X
Local polynomial	1			1
Bandwidth	Optimal			Optimal
Kernel	Triangular			Triangular

Note. Robust standard errors clustered at the school district level appear in parentheses below the point estimates. We restrict the bandwidth of WIDA scale scores used in the analysis to limit the influence of outlier students with very high or very low WIDA scores using the rdrobust command, which implements a data-driven process to determine an optimal bandwidth and estimates bias-corrected coefficients and robust standard errors (Calonico et al., 2014). When point estimate and standard error equal 1 and 0, respectively, estimates indicate that every student who qualified for reclassification was reclassified. This could be interpreted as a sharp RD. Because the third-grade post period point estimate for Arabic speakers is interpreted as a sharp RD, the precision weighted average is exactly the value of the sharp RD. SE = standard error; RD = regression discontinuity; DiRD = difference-in-regression discontinuity; ELA = English Language Arts; EL = English learner.

***

p < .001.

Results from the weighted average effects indicate that during manual reclassification, Spanish speakers just above the reclassification threshold were roughly 17 percentage points less likely to reclassify than Arabic speakers (0.515 vs. 0.684). In the automatic reclassification period, the weighted average effects of qualifying for reclassification on reclassifying are more similar across subgroups (0.964 for Spanish speakers vs. 1.000 for Arabic speakers). Overall, leveling out across subgroups’ likelihood of reclassifying under automatic reclassification implies that the policy had a greater impact on Spanish speakers (0.440) than Arabic speakers (0.350). Table 7 presents DiDiRD estimates of the effect of the policy change for Spanish and Arabic speakers. The DiDiRD estimates of the policy change confirm this finding, with the shift to automatic policy having a 9.0 percentage-point greater impact on Spanish speakers than on Arabic speakers. We report DiDiRD comparisons to students reporting another primary language in Supplemental Appendix Table A6.

Table 7

DiDiRD Estimates Across Language Subgroups of ELs

Intentionally blank	DiRD: Spanish	DiRD: Arabic	DiDiRD
Weighted average estimate	0.440***	0.350***	0.090***
SE	(0.028)	(0.020)	(0.034)

Note. DiRD = difference-in-regression discontinuity; DiDiRD = difference in DiRD; EL = English learner; SE = standard error.

***

p < .001.

Notably, the DiDiRD estimate appears to be primarily driven by differences in third-grade Spanish reclassification rates compared to Arabic. During manual reclassification, third-grade Spanish speakers just above the reclassification threshold experienced a roughly 66.3 percentage-point increase in their likelihood of reclassifying. In contrast, similar Arabic speakers experienced a roughly 84.4 percentage-point increase in reclassification likelihood. Other grade-level ITT effect estimates are similar across these linguistic subgroups during manual reclassification. For example, eligible fifth graders reporting Spanish as a home language experience a nearly 35.8 percentage-point increase in likelihood of reclassification, compared to an increase of nearly 39.6 percentage points at the threshold for students reporting Arabic as a home language.

Weighted average effects during the automatic reclassification period suggest that eligible Spanish speakers (0.964) continue to experience a lower likelihood of reclassification than Arabic speakers (1.000). However, the difference in weighted average effects across groups is much smaller than under manual reclassification. While these findings suggest that automatic reclassification reduced disparities in reclassification rates between language groups, we cannot fully disentangle language-based differences in reclassification from unobserved district-level characteristics. Part of the unobserved differences in reclassification rates could reflect variation in district-specific policies or resources rather than home language alone.

Robustness Checks

We conduct several robustness checks to estimate the ITT effect of qualifying for reclassification on reclassifying across grade levels and policy periods. We test the sensitivity of our main analytic models to different models, including alternative kernels, bandwidths, and clustered standard errors. We also estimate each model with and without the inclusion of covariates. Additionally, our main models include a control for standardized ELA scores as a proxy for achievement on a local ELA assessment (one component of reclassification criteria during manual reclassification). This excludes data from 2019–20, as standardized tests were not administered during the COVID–19 pandemic. In some alternative models, we exclude the standardized ELA score control and include reclassification data from 2019–20. Our estimations are also robust to models excluding data from both 2019–20 and 2020–21, given that both years were directly impacted by the COVID–19 pandemic (Supplemental Appendix Table A7). Given that we achieve balance on all covariates, we estimate a singl Ordinary Least Squares model in Supplemental Appendix Table A8. Results are robust to alternative estimation strategies, including logistic regression. We report point estimates for logistic regression estimations in Supplemental Appendix Table A9. Results are also robust to the possibility of unobserved between-district average differences, assessed by including district fixed effects in the models and obtaining similar results (Supplemental Appendix Table A10).

We also explored whether the policy change led to reductions in the variance of reclassification effects within and between school districts by estimating multilevel models where level 1 is students and level 2 is school districts, including random effects for the level-2 intercept and all slopes in Supplemental Appendix Table A11. Importantly, we find that the variance in the between-district random effects in the slope coefficient associated with attaining the threshold decreases from the pre- to post-automatic reclassification period. For example, the variance on the district-level random slope for scoring above the cutoff (labeled “Above Cutoff (slope)” in Supplemental Appendix Table A11) for third graders in the pre-period was 0.095, and it was substantially smaller at 0.015 in the post-period. Results from the robustness checks are comparable to results presented in the main findings and indicate that shifting to automatic reclassification had a large, positive effect on reclassification rates among eligible students and reduced variation in reclassification outcomes within and between districts.

Conclusion and Policy Implications

Using administrative data from Michigan, this study finds that automatic or default procedures can (a) substantially increase adherence to statewide standardized EL reclassification policy and (b) reduce linguistic or other disparities in access to reclassification compared to manual procedures. We find statistically significant, substantial effects of shifting from a manual to an automatic reclassification policy on reclassification rates of eligible EL students. The effects of shifting to an automatic reclassification policy are larger for specific subgroups of ELs, namely Spanish speakers.

While these findings do not speak to students’ outcomes following reclassification, which depend upon both reclassification criteria and procedures, they have implications for EL policy in light of ESSA’s (2015) mandate that states establish standardized EL reclassification protocol. Findings also have implications for education policy more broadly as states and school districts look to increase equity in students’ access to specialized programs, such as advanced coursework. Future research should consider how the application of default policies and selection of enrollment and reclassification criteria influences students’ access to specialized coursework and outcomes following enrollment in specialized coursework.

Implications for EL Policy

From an EL policy perspective, these findings corroborate earlier pre-ESSA research that finds substantial discrepancies between the population of students qualifying for reclassification and those reclassifying (Cimpian et al., 2017; Estrada & Wang, 2018). We extend this research base by highlighting automatic reclassification procedures as a mechanism to reduce these discrepancies. Recognizing that many eligible ELs were not reclassifying on time under manual procedures, Michigan implemented an automatic reclassification policy. Under manual reclassification, we confirm significant disparities in reclassification rates of eligible students. This suggests that reclassification decisions may have been based on factors other than reclassification criteria. Prior literature highlights several features of manual reclassification procedures that may contribute to disparities in reclassification rates, including excessive administrative burden on school districts, educators’ beliefs about the merits of reclassification, and EL students themselves, differences in state reclassification policy interpretation (Estrada & Wang, 2018; Mavrogordato & White, 2017), and variation in policy implementation across districts (Cimpian et al., 2017). This study provides causal evidence that shifting to an automatic procedure can create much greater parity in reclassification rates of eligible students.

Our findings and discussion raise two important caveats worth noting. First, some eligible students still do not reclassify under automatic procedures. This is because districts can override or opt out of automatic reclassification, a key feature of default policies. This feature allows for local discretion, particularly if a student meets test-based reclassification criteria, but educators feel the student could benefit from further linguistic support for other reasons. Second, reclassification does not in itself imply positive outcomes for students. Whether a student benefits from reclassification depends on the state’s reclassification criteria (Robinson, 2011). For example, if the reclassification criteria are set too low, students may reclassify too soon and struggle without linguistic support. If the criteria are set too high, students may remain in EL status when they would benefit from mainstream academic coursework. Assuming a state implements appropriately rigorous reclassification criteria, automatic procedures may be an effective policy for ensuring students reclassify when they demonstrate English proficiency. The combination of effective reclassification criteria and procedures can provide students with access to developmentally appropriate coursework.

As the EL population continues to grow and diversify rapidly, it is vital to consider how reclassification policies impact students within the EL subgroup differently. We provide the first causal evidence of the effect of a state’s choice of reclassification procedures on reclassification likelihood for ELs as a whole and among subgroups. Overall, we find that manual procedures impact subgroups of ELs differently. First, we find that eligible ELs are more or less likely to reclassify based on their grade. For example, under manual reclassification, roughly 20% of third-grade students eligible for reclassification did not reclassify. This research parallels Umansky and Reardon’s (2014) conclusion that in early grades, more students meet reclassification criteria than reclassify. However, Umansky and Reardon (2014) find that this trend reverses in middle school, with more students reclassifying than meeting eligibility criteria. In contrast, we find that under manual procedures, a significant proportion of sixth through eighth grade students who met reclassification criteria were not reclassified. Under automatic reclassification, these gaps close, and nearly all eligible third through eighth grade students reclassify. Taken together, these findings suggest that between-grade variation in eligible students’ reclassification likelihood exists under manual procedures. Automatic procedures may be more effective at standardizing reclassification rates for eligible students, a key goal of ESSA (2015).

In addition, research has identified subgroups of ELs that are less likely to reclassify upon meeting reclassification criteria, particularly ELs reporting Spanish as a home language (Mavrogordato & White, 2017; Umansky et al., 2020). Our estimates align with this research. Under manual reclassification procedures in which districts are responsible for reclassifying ELs, we find that eligible ELs who report Spanish as a home language are substantially less likely to reclassify than ELs who report other home languages. However, this discrepancy largely dissipates upon shifting to automatic reclassification, in which state data systems reclassify eligible ELs. This finding suggests potential bias against Spanish speakers under manual reclassification procedures.

Implications for Education Policy

Nationwide, state education agencies are grappling with the most effective ways to increase representation and enrollment in specialized educational services such as advanced coursework and gifted education (Blad, 2020). Many state education agencies are moving toward automatic enrollment to ensure students are served in a developmentally appropriate environment (Plucker, 2021). In light of this movement, rigorous causal evidence is needed to evaluate the ability of automatic policy to increase equity and representation in educational service enrollment. The present study offers the first evaluation of the effects of automatic policy in K–12 educational settings, finding that it can increase students’ likelihood of being served in a developmentally appropriate environment (e.g., by reclassifying upon demonstrating English proficiency).

This study faces several notable limitations. First, reclassification often entails a significant change in students’ instructional environment and has important implications for their short- and long-term outcomes. The present study focuses on evaluating the efficacy of automatic procedures in increasing adherence to state policy and ESSA guidance rather than assessing the effects of reclassification on students’ outcomes. Future research may explore outcomes for “compliers,” or eligible students who would reclassify under automatic procedures and not under manual procedures, to determine whether shifting the policy had positive or negative effects on student outcomes. Moreover, this study does not identify the mechanisms that caused lower reclassification of eligible students under manual reclassification. Future qualitative research may explore why manual reclassification procedures resulted in a substantially lower likelihood of reclassification among eligible students than automatic procedures.

In addition, although this analysis presents the first examination of automatic policies in K–12 education and incorporates post-ESSA data to examine effective EL reclassification procedures, our analysis has several contextual and methodological limitations. First, the policy change occurred one year before the COVID–19 pandemic. We acknowledge that the lingering effects of the pandemic, including disruptions to instruction, assessment, and student English proficiency growth patterns, may also influence our findings in unobserved ways. While our robustness checks suggest that pandemic-related shifts are not driving the results presented in this manuscript, we cannot fully rule out selection on unobservables at the district or household level, particularly if pandemic-related disruptions differentially affected certain subsets of students. Future qualitative research should investigate the extent to which this policy change and the pandemic led to substantive changes in the population of ELs that these districts served.

Second, reclassification eligibility was based on multiple test scores under manual reclassification procedures, but only a single test score under automatic reclassification procedures. The change in reclassification criteria may have led to a shifting of effort in which teachers or students focus more on attaining the single cutoff score (overall WIDA English proficiency score), whereas under manual procedures, they would have been focused on developing proficiency along multiple dimensions (overall WIDA English proficiency score and WIDA reading and writing subscores). However, we believe this shifting of effort is unlikely because the WIDA overall score is a weighted composite of students’ WIDA reading, writing, listening, and speaking subscores, meaning students need to attain high scores on each subdomain to meet the WIDA overall reclassification cutoff score.

Third, the change in eligibility criteria across policy periods introduces methodological challenges for interpreting the effects of the policy. In the post-period, eligibility for reclassification is determined by a single, observable cutoff score. This allows for a sharp RD design in which the effect of eligibility on reclassification can be interpreted as both the ITT and the local average treatment effect. In contrast, in the pre-period, eligibility was based on a combination of criteria, one of which is unobserved in our dataset (local reading assessment scores). As a result, the pre-period ITT represents the effect of crossing the WIDA overall score threshold on reclassification, but not necessarily the effect of being eligible for reclassification. Consequently, the difference between pre- and post-policy estimates may partly reflect unobserved changes in the strength of the first-stage RD rather than the causal effects of eligibility on reclassification itself. While we attempt to mitigate this limitation by conditioning the sample on students who meet other known reclassification criteria, we cannot fully observe or account for all factors influencing reclassification eligibility in the pre-period. Therefore, differences in the estimated treatment effects across policy periods should be interpreted with caution, as they may partially reflect shifts in the underlying eligibility process rather than just the change in reclassification procedures.

There are also several limitations of our DiRD design worth noting. The generalizability of our estimates is restricted to students just above or below the reclassification threshold, and this limits the applicability of our conclusions. Finally, these findings will not be generalizable to all states. Many states include subjective measures in their reclassification criteria (e.g., teacher recommendation, student grades), and subjective criteria are not collected by state data systems. As such, results and implications should be considered in a state with reclassification criteria captured by state administrative data systems.

Rigorous research is needed to examine the ways policy can expand or constrain educational opportunities for the growing and diversifying EL population in U.S. schools. Reclassification is one mechanism through which ELs gain access to the full range of academic coursework, and thus, policymakers should prioritize reclassifying students who demonstrate eligibility by meeting state reclassification criteria. This study offers one potential mechanism, automatic policy, that policymakers may consider to ensure greater equity in reclassification decisions among eligible students.

Supplemental Material

sj-pdf-1-epa-10.3102_01623737251400379 – Supplemental material for Leveling the Playing Field: Default Policy and Its Effects on English Learner Reclassification Rates

Supplemental material, sj-pdf-1-epa-10.3102_01623737251400379 for Leveling the Playing Field: Default Policy and Its Effects on English Learner Reclassification Rates by Caroline Bartlett, Joseph R. Cimpian and Madeline Mavrogordato in Educational Evaluation and Policy Analysis