One Size Does Not Fit All: Conceptualizing Alternative Data Approaches to Improve Single-Language Developmental Language Disorder Screening for Multilingual Public School Students

Abstract

Purpose:

Universal screening for developmental language disorder (DLD) is a promising preventative approach for decreasing negative educational outcomes associated with DLD. Unfortunately, schools with multilingual student populations often face barriers in following best-practice guidelines for language testing and screening students in all their languages. We explored the outcome of a school district administering a single-language DLD screener to a multilingual student body, as well as alternative approaches for using single-language screening data to identify DLD risk in multilingual students.

Method:

A school district engaged in a research–practice partnership administered an English-only DLD screener to 703 kindergartners with English language learner (ELL) status. We analyzed screener outcomes and the relationship of identified DLD risk status and English proficiency level for the ELLs. We explored alternative data analysis approaches of local norms, adjusted cut-points, change in screener score from fall to spring, and a hybrid of adjusted cut-points and change to determine which approaches identified DLD risk status for ELLs at a rate more aligned with DLD prevalence rates and the intended rates of a screener.

Results:

The English-only language screener overidentified multilingual students as at risk for DLD. There was a significant negative correlation between risk identification and student English proficiency level. All alternative methods for data analysis lowered the rate of identification of ELLs as at risk for DLD.

Conclusions:

Single-language testing overidentifies multilingual students as at risk for DLD. We caution school districts against a monolingual screening approach to screen multilingual students and encourage them to consider alternative approaches to using English-only screening data for multilingual students.

Developmental language disorder (DLD) is a neurodevelopmental condition that affects a child's learning, understanding, and/or use of spoken language (Leonard, 2020). Given that it is a relatively common developmental disorder, with an estimated prevalence of 7.58%, it is likely that two children in every classroom of 30 will present with DLD (Norbury et al., 2016; Tomblin et al., 1997). Despite its prevalence, DLD is considered a “hidden disorder” due to its inconsistent, varied, and sometimes subtle features, resulting in 50%–70% of children with DLD going unidentified (Norbury et al., 2016; Tomblin et al., 1997). The underidentification of DLD is highly consequential, as, without the appropriate supports, the disorder can have profound negative impacts on a child's educational outcomes, including reading difficulties and poor performance across academic domains (Catts, 1993; Ziegenfusz et al., 2022).

Universal Screening for DLD

Currently, DLD is most commonly identified reactively, following a referral from a concerned teacher or parent, if it is identified at all. This “wait to fail” model is problematic as it precludes a child with less readily perceptible symptoms of DLD from benefiting from early intervention (Adlof & Hogan, 2019; Tomblin et al., 1997). Screening for DLD early in a child's school experience is an essential step that school districts can take to proactively mitigate DLD's negative impact on students' academic outcomes (Adlof & Hogan, 2019; Bao et al., 2024; Catts & Hogan, 2020). It is already common practice for public schools in the United States to screen for conditions that can impact academic performance. For example, in addition to screening for vision and hearing problems, new state laws are increasingly mandating universal screening for reading challenges (Catts & Hogan, 2020; Komesidou et al., 2022; Ward-Lonergan & Duthie, 2018). Positively, a reading screener is likely to capture many children with DLD as the overlap between DLD and reading difficulty is known to be as high as 50%–80% (Adlof & Hogan, 2018). However, a screener that only captures reading problems in terms of word reading accuracy and fails to screen for or independently capture and report underlying spoken language difficulties will miss the approximately 50% of children with DLD who develop adequate word reading skills but will ultimately struggle with reading comprehension due to their spoken language challenges (Adlof & Hogan, 2018; Bao et al., 2024).

The rationale for a universal screening of spoken language skills to increase the early identification of DLD is clear. The beginning of kindergarten has been identified as an ideal time to screen, as, by that age, the highly variable trajectories of child language development have primarily stabilized and school attendance is compulsory for most children (Hulme et al., 2024; Komesidou et al., 2022; McGregor, 2020). Thus, screening upon entry to school can be both possible and reliable. Unfortunately, school districts and school personnel face numerous barriers to implementing schoolwide language screenings, including the availability and accessibility of psychometrically sound DLD screeners that are appropriate for the culturally and linguistically diverse student body of any given district (Bao et al., 2024). In their 2024 review of commercially available screeners for DLD, Bao et al. (2024) found a dearth of available screeners for languages other than English and identified a clear need for DLD screeners that are appropriate for and sensitive to language variation and dialect use. Given the growing population of English language learners (ELLs) in U.S. schools, it is imperative to consider multilingualism and English language proficiency levels when conceptualizing and ultimately implementing universal language screenings.

Identification of DLD and DLD Risk in a Multilingual Population

Decades of research on multilingualism ¹ have highlighted the language differences between a multilingual student and a monolingual student (Grosjean, 1989). Multilingual children have increased heterogeneity and complexity in their linguistic environments and, consequently, their linguistic performance. Thus, careful consideration for evaluating language skills in multilingual populations is imperative (Freeman & Schroeder, 2022; Lugo-Neris et al., 2015; Nayeb et al., 2021). Empirical studies demonstrate that applying monolingual approaches for language evaluation to a multilingual population is likely to be biased against multilingual children (Peña et al., 2023; Sullivan & Bal, 2013). This is in part due to multilingual children's distributed language experiences, which can lead to mixed language dominance. Testing a multilingual child in only one of the languages that they speak is unlikely to capture their full repertoire of linguistic skill, thus impacting their score on a standardized language test (Lugo-Neris et al., 2015). Further, language assessment that happens only in the student's nondominant language is expected to underestimate a multilingual child's actual language ability due to reduced proficiency in a lesser used or newly acquired language; this increases the likelihood of overidentifying multilingual students as at risk for (in the case of a screener) or having (in the case of a diagnostic assessment) a language disability (Bialystok et al., 2008, 2010; Peña et al., 2011). Overidentification can lead to provision of unwarranted special education services and can be counterproductive to the child's academic trajectory (Artiles et al., 2005; Ford et al., 2017). Therefore, both related research and recommendations from the American Speech-Language-Hearing Association (ASHA) indicate that language testing for the purpose of identification of a language disorder in a multilingual population should be conducted in both languages of the child (ASHA, 2023). Additionally, the Individuals with Disabilities Education Act (IDEA, 2004) stipulates that materials used to assess a child should be provided and administered in the child's native language and should be implemented “in the form most likely to yield accurate information on what the child knows and can do academically, developmentally, and functionally.” These considerations for multilingual assessment apply regardless of whether the language testing is administered to diagnose or to screen for risk of a language disorder (Lugo-Neris et al., 2015).

Despite the availability of research indicating best practice for language assessment and screening in a culturally and linguistically diverse student body, legitimate barriers exist that prevent school systems and professionals from administering multilingual language evaluations appropriately. In their survey of 166 speech-language pathologists (SLPs), Arias and Friberg (2017) found that the largest barriers to multilingual language assessment included lack of interpreters, lack of time, lack of training, and lack of available standardized assessments. Limited access to interpreters and school personnel who speak the languages necessary to test a child in their native language presents an increasingly insurmountable barrier to multilingual assessment and screening as the heterogeneity of multilingualism within the student body increases (Arias & Friberg, 2017; Freeman & Schroeder, 2022). While some districts may have a multilingual student population that primarily speaks the same heritage language (i.e., Spanish), other districts may have highly diverse multilingual populations in which students speak many different heritage languages, complicating efforts for a school staff to test in both languages of the child. Taken together, the challenges that exist in the context of a school setting preclude well-meaning districts and professionals from implementing the best practices put forth in the research. It is disconcerting, but unsurprising, that it remains common for multilingual children to be tested in English only, oftentimes with testing instruments developed for monolinguals with monolingual norms (Altman et al., 2022; Arias & Friberg, 2017). Screening tools in languages other than English are even more limited than diagnostic assessment tools. Therefore, while it is clearly problematic, school districts implementing universal language screenings may administer English language screeners to a linguistically diverse student body as it is not currently feasible to administer multiple screeners in multiple languages.

Considerations for Single-Language Testing and Screening: Alternative Approaches

Local Norms

Prior clinical research has aimed to explore viable solutions to the problem of single-language testing for multilingual students. This research has established the possibility of using monolingual tasks in multilingual assessment if the norms are specifically developed for the target population (Altman et al., 2022; Gathercole, 2013; Peña et al., 2011). One method to adapt a monolingual language test for multilingual populations is to use local norms. Prior research suggests that students who speak a language other than the language of instruction should be compared to each other, rather than to the full sample of their peers; this should support a more equal distribution of monolingual versus multilingual students being identified as at risk (Bedore & Peña, 2008; Martini et al., 2021). Altman et al. (2022) found that when using a monolingual language assessment to identify DLD in multilingual children, empirically derived local multilingual standards had higher diagnostic accuracy for multilingual children as compared to the assessment's monolingual norms. The researchers empirically derived multilingual norms by generating different cut-points that both considered individual language background (chronological age and age of onset of multilingualism) and were based off the means and standard deviations of the multilingual group only; they found optimal sensitivity and specificity for cut-points of 1.25 and 1.5 SDs below the local mean.

Adjusted Cut-Points

Another method to adapt a monolingual language test for multilingual populations is to lower the cut-off criteria for the multilingual population (Altman et al., 2022; Martini et al., 2021; Thordardottir, 2015). For example, Barragan et al. (2018) found that the recommended norms for the Clinical Evaluation of Language Fundamentals–Fourth Edition, Spanish (Wiig, Semel, & Secord, 2006), identified 53.5% of their sample population of dual-language learners from low socioeconomic backgrounds receiving English-only education as language impaired based on a cut-point score of 85. Lowering this cut-point to 78 reduced the number of children identified as language impaired to 9.6% and yielded a sensitivity and specificity that could be considered acceptable (sensitivity = 86%, specificity = 80%; Barragan et al., 2018; Plante & Vance, 1994).

Change Scores

An additional factor to consider when using an English-only language test to identify multilingual students with risk for DLD is a floor effect. This is particularly important to consider at the beginning of the students' school experience, when English language proficiency levels may be minimal. The skills being measured on a language screener are a “moving target,” in that language can constantly be changing in response to interactions in the child's home and school environments, as well as through the process of typical cognitive development. Children entering kindergarten in an English-only educational context will continue to develop the skills needed to complete language screening tasks as they progress through their first year of school (Catts et al., 2008; Speece, 2005). Given that language experience and instruction at school supports language development, administering a language screener in English to an ELL too early in their school experience could lead to a floor effect because the students may not have the language proficiency to complete the screener tasks (Barragan et al., 2018; Catts et al., 2008; Restrepo, 1998).

Therefore, there is precedence to adjust the scoring schedule to account for floor effects, as well as to use measures of growth to circumvent the “moving target” problem; by screening at more than one timepoint, decisions can be made based on multiple data points gathered over an extended period of time (Catts et al., 2008; Speece, 2005). This concept is consistent with that of dynamic assessment, an evidence-based approach recommended for multilingual assessment that aims to capture a child's learning potential by measuring change over time (Castilla-Earls et al., 2020; Gutierrez-Clellen & Peña, 2001). Adding a dynamic assessment approach to screening can be beneficial in evaluating students' ability to acquire and retain new information over time (Goodrich et al., 2023).

Taken together, if a school district faces barriers to screening multilingual students in their first language (L1), empirical evidence suggests that alternative data analysis approaches including deriving local norms, adjusting cut-points, and measuring change over time may present as viable options for using a single-language screener while simultaneously keeping identification rates consistent with what would be expected for a screener given the prevalence of the disorder in the general population (see White & Schatschneider, 2024).

Implementation Science and Universal DLD Screening

Multilingual language assessment and screening is an area of practice with an observable research-to-practice gap. As a result, over- and underidentification of multilingual children with DLD persists. This gap is caused by barriers that prevent practitioners from applying research-backed methods, regardless of their level of awareness of evidence-based practices for multilingual language evaluation (Odom et al., 2019). Operating within the constraints of their professional practice settings, practitioners may not have access to additional strategies, such as use of local norms, tailored cut-points, or alternative assessment methods that researchers may use to improve the outcomes of assessments that are administered in less-than-ideal circumstances. With the increase in statewide mandates for universal literacy screenings and the strong rationale for concurrent or supplemental universal DLD screenings, schools across the nation may face a problem of how to put research into practice, that is, how to universally administer language screeners to a linguistically diverse population. Therefore, multilingual language screening presents an excellent opportunity for employing a dissemination and implementation science lens to facilitate the translation of research findings into routine school practice by leveraging research-to-practice partnerships (RPPs; Alonzo et al., 2022) to co-construct ecologically valid solutions that center the needs, strengths, and constraints of the school district (Douglas et al., 2015; Komesidou et al., 2022).

The Present Study

The first aim of the present study is to explore the problem that districts face when attempting to administer a universal language screener to a heterogeneous multilingual student body due to inability to administer the screener in each of the students' languages. There is little documented information about the identification rates of multilingual children when using an English-only language screener in a district with a large and highly variable population of English learners. Understanding the outcomes of single-language screening can help to elucidate the extent to which this screening method provides useful information about multilingual students' risk for DLD. The second aim of the present study is to leverage an RPP to collaboratively explore solutions to this problem through thoughtful and tailored data analysis. School districts with culturally and linguistically diverse student bodies could benefit from using varied statistical approaches to analyze data from single-language testing to efficiently and appropriately screen multilingual students for language challenges. This could improve students' overall academic experience and trajectory by accurately identifying risk before students begin to struggle academically, while avoiding identifying typically developing ELLs for special education

It is important to note that because the students in the study participated in the DLD screening per the discretion of their school district as a part of typical school practices, not all the screened students underwent follow-up diagnostic testing. Further, data from district-administered diagnostic testing were not made available to the research team as part of the partnership. Therefore, the absence of follow-up diagnostic testing for the specific student population included in this study limits our ability to understand the true sensitivity and specificity of the given screener, regardless of the data analysis approach applied. Nonetheless, studying the extant screening data for a multilingual population in the context of previously established literature on preventative screening processes and prevalence rates provides valuable insights that can support school districts in implementing language screening practices with linguistically diverse populations.

Our research questions and hypotheses were as follows:

1. What percentage of multilingual students are identified as at risk for DLD using their fall score on an English-only language screener, in each of the categories of student language proficiency?

We hypothesize that a high percentage of the multilingual kindergartners will be identified as at risk for DLD using their score on an English-only language screener administered in the fall and that the percentage of kindergartners identified as at risk will be negatively correlated with English proficiency level (EPL). Based on prior research demonstrating that language assessments administered in a child's nondominant language tend to underestimate their language skills, we expect that the English-only language screener will identify a higher number of multilingual children—enrolled in a variety of English instruction programs for ELLs—as at risk than would be expected given the prevalence of DLD in the general population.

•  What percentage of multilingual students are identified as at risk for DLD using their spring score on an English-only language screener, in each of the categories of language proficiency?

We hypothesize that overidentification of multilingual children as at risk for DLD will continue to be an issue when using spring screening scores but that the spring screener will identify fewer students overall than the fall screener. This hypothesis is supported by research that language proficiency levels in the second language (L2) affect performance on L2 language tasks (Barragan et al., 2018; Restrepo, 1998). We believe that, by the spring of their kindergarten year, most students—enrolled in various English instruction programs for ELLs—will have increased their English proficiency to some extent, supporting a higher score on the language screener in the spring. We hypothesize that we will continue to see a relationship between identified EPL and at-risk status, in that students with the lowest EPLs will be most likely to be identified as at risk for DLD in the spring.

•  Are there alternative approaches to using English-only screening data and routinely collected English language proficiency data to identify multilingual students as at risk for DLD?

Based on the prior empirical evidence discussed above, we hypothesize that using local norms, lowering cut-points, using a change score derived from the two screening timepoints, and/or a hybrid of these approaches will lower the rate of identification of multilingual students to more closely align with the expected prevalence rates in the general population and intended rates for a DLD screener.

Method

Research Practice Partnership

Students in the current study were enrolled in one of three school districts participating in an RPP as part of a large-scale longitudinal study, Orthography and Word Learning (OWL), which examined language and reading trajectories of children with and without DLD (Alonzo et al., 2022; Komesidou et al., 2022; Radville et al., 2024). Institutional review board approval was obtained for the larger longitudinal study, and consent forms were signed for participating students. Students were enrolled in a large metropolitan district (~9,200 students) in a small northeastern U.S. city (~70,000 population; see Alonzo et al., 2022). This district formed an RPP with a research university in 2018. In addition to OWL data collection, a key RPP goal was supporting the district's implementation of universal kindergarten dyslexia screening, mandated by a recently passed state law (Alonzo et al., 2022; Komesidou et al., 2022). Researchers advised on existing literacy screening tools as well as proposed their own screening instrument that would screen for both dyslexia and DLD. The district chose to implement this researcher-developed language and literacy screener (OWL Screener; see Komesidou et al., 2022) as part of their typical district-wide yearly assessment/screening practices. The research team agreed to provide training, professional development, and data reports and analysis to support the district through the screening process. The RPP developed and signed a memorandum of understanding stipulating the reciprocal sharing of data. Per state and district guidelines, the school informs parents/guardians of universal screening plans (Department of Elementary and Secondary Education, 2023); individual consent is not required for general screenings administered to all students (IDEA, 2004).

A notable benefit of the district's choice to use the researcher-developed OWL screener was the opportunity to simultaneously identify children who were at risk for both dyslexia and DLD. However, a downside to using this tool for this specific district was that the English-only language screener, like most publicly available language screeners, was not designed for their large and heterogeneous population of multilingual students. As of 2023, there were a total of 69 different languages spoken by students in the district (retrieved from the district's English Language Dashboard). In the fall of 2019, the first year that the district administered the OWL screener, 40% of the enrolled kindergarten students (n = 801 of 2,026) were considered limited English proficient and enrolled as ELLs. While best-practice guidelines are to assess a student's language skills in all their languages, the district faced many of the aforementioned barriers to following this guideline, such as limited access to multilingual screeners or personnel. This led them to proceed with the English-only language screening tool. The RPP created a plan to review the screener outcomes as they became available and adjust as needed.

The district administered the English-only OWL screener to their incoming kindergarten class in the fall of 2019–2020 and 2020–2021. By the 2020–2021 academic year, the percentage of students in the incoming kindergarten class with limited English proficiency had increased to 53% (n = 811 of 1,529). Recognizing the need to better capture multilingual students' language profiles, the district worked with the researchers to refine screening procedures. The RPP collaboratively decided to add a spring screening in 2021–2022 to allow ELL students additional time to gain English proficiency, as well as provide a second timepoint that could potentially capture language growth. Additionally, the researchers would provide support and data expertise to interpret the fall and spring 2021–2022 language screener data of the district's multilingual student population. As such, screening was completed as part of business-as-usual educational practices for all enrolled students, as per the practices established by the RPP.

Study Sample

Data were gathered as part of routine educational practices newly implemented as part of the RPP described above from students enrolled in kindergarten during the 2021–2022 academic year at two timepoints, October 2021 (fall) and April 2022 (spring). Of the 2,090 students enrolled in kindergarten in 2021, 1,882 (90%) completed the language portion of the OWL screener during at least one timepoint (fall or spring). Of those kindergartners who completed at least one language screening timepoint, 703 (37%) were ELLs, as documented by the district's report of their ELL education program and their EPL. This EPL is derived from student completion of a language proficiency measure (WIDA, 2020) and describes student English proficiency in terms of six EPLs: Entering (1), Emerging (2), Developing (3), Expanding (4), Bridging (5), and Reaching (6), with EPL 1 representing the lowest level of English proficiency and EPL 6 representing the highest. Upon entry to school, students whose parents/guardians report they speak a language other than English at home complete the WIDA. The WIDA then provides an EPL and an ELL designation, if applicable. Students who receive a minimum proficiency score of 5 on the WIDA are not considered ELL. Those that score below 5 are considered ELL. The study sample is composed of the 703 ELL kindergarteners with EPLs below 5 who completed language screener(s) in the 2021–2022 academic year. Per district report, 29.4% of these students (n = 207) were EPL 1, 17.5% (n = 123) were EPL 2, 51.8% (n = 364) were EPL 3, and 1.3% (n = 9) were EPL 4 (see Figure 1). The district did not report EPL scores greater than 4 for any of the 2,090 kindergarten students.

Figure 1.

The district's multilingual students, grouped by English proficiency level (EPL), included in the present analysis.

The school district offers three types of education programs for ELL students: Sheltered English Immersion, Dual Language Education, and Transitional Bilingual Education. Students in Sheltered English Immersion receive daily instruction in English 100% of time to support their developing English language proficiency. Content area instruction is delivered in English by Sheltered English Immersion–endorsed teachers. The Dual Language Program uses a model in which kindergarten classes receive instruction in Spanish 80% of the time and English 20% of the time and progressively receive 10% more English instruction each subsequent year. The Transitional Bilingual Education is a program for students with EPLs 1 and 2 in which instructional materials are in English and Spanish, beginning with most content in Spanish and progressing into using more English content. Due to our data sharing agreement with the district, only details regarding the students' English language exposure at school are known; no other language exposure information was provided. See Table 1 for more information about the distribution of student demographics.

Table 1.

Student demographics, including EPL, multilingual program, and FRPL status.

Variable	n	%
English proficiency level (EPL)
EPL 1	207	29.4
EPL 2	123	17.5
EPL 3	364	51.8
EPL 4	9	1.3
Multilingual program
Dual Language Education	29	4.1
Sheltered English Immersion	618	87.9
Transitional Bilingual Education	16	2.3
Provisional	1	0.1
Parent waived services	27	3.8
Missing	12	1.7
Free or Reduced-Price Lunch (FRPL) status
Free	10	1.4
Free transitional assistance	477	67.9
Reduced transitional assistance	42	6.0
Missing	174	24.8

Note. N = 703. In the United States, FRPL status is a commonly used proxy for socioeconomic status.

Procedure

Beginning in 2019, classroom teachers were trained by the research team in the administration and scoring of the OWL screener as part of routine educational practices. Due to student absences and additional factors resulting from typical school-day procedures, 673 students completed the screener in Fall 2021, 589 students completed the screener in Spring 2022, and 559 students completed the screener at both timepoints. The research team gathered the scores from teachers and shared screening results with them and the school district.

Measures

Students completed two measures: the WIDA Screener for kindergarten to assess their EPL and the OWL Screener, which consisted of both a language and literacy subtest. For the present analyses, only the scores from the WIDA Screener and language subtest of the OWL Screener were analyzed.

English Proficiency Screening

The WIDA Screener for kindergarten (WIDA, 2020) is the only tool state-approved for screening the English proficiency of kindergarten students. The WIDA is a story and activity-centered test experience that consists of a storybook section, followed by listening, speaking, reading, and writing components. Kindergartners who take the WIDA in their first semester of school complete the storybook and the listening and speaking sections only, which takes up to 20 min to complete. It is individually administered via paper and pencil and scored by a trained administrator. After completing the WIDA, students receive a proficiency level score that describes performance in terms of six WIDA EPLs as described above (see the Study Sample section).

DLD Screening

All students completed a digital version of the OWL Screener, a researcher-created screening measure that is not publicly available but was offered to the district to address their goal of screening all kindergarten students for dyslexia and DLD (see Komesidou et al., 2022). Classroom teachers supported students in completing the screener on their personal digital devices as part of routine classroom procedures. The language subtest of the OWL screener included 10 test items assessing receptive morphosyntax. Students listened to a spoken sentence (e.g., “the dog but not the cat is sleeping”) and identified the corresponding image from an array of four options. Students who scored 7/10 or below were identified as at risk for DLD.

While the sensitivity and specificity of this screener are not available for this specific student population, the area under the curve (AUC), which provides an estimate of screening accuracy, was reported for a similar type of screener, which served as an inspiration for the creation of the OWL screener, administered to 32 first graders (11 with DLD, 21 with typical language skills) and 65 second graders (11 with DLD, 54 with typical language skills; Hendricks et al., 2019). AUC values between .7 and .8 are considered acceptable, AUC values between .8 and .9 are considered excellent, and AUC values above .9 are outstanding (Hosmer & Lemeshow, 2000). The AUC for the identification of DLD in first and second graders was .837, 95% CI [0.756, 0.919], and therefore considered excellent.

Data Analysis Plan

Statistical analysis was completed using the IBM SPSS statistics software. For Research Questions 1 and 2, we determined what percentage of children fell at or below the original 7/10 cut-point and reported these data by EPL group. We used chi-square tests to determine if the rate of identification was equally distributed across EPL groups. For Research Question 3, we aimed to lower the rate of risk identification to be closer to what would be expected given DLD prevalence rates in the general population (7.58%; Norbury et al., 2016) and best practices for multi-tiered system of supports (MTSS), such as response to intervention. In the MTSS literature, it is reported that approximately 20% of students are identified as at risk and receive additional support (for review, see Bao et al., 2024). We also planned to consider the needs of each EPL group in the case that our analysis revealed screener score and rate of risk identification differed significantly by EPL. Thus, we aimed to see DLD risk identification rates of approximately 20%–25% of the multilingual student population and of each EPL group.

We chose to apply four additional data analysis approaches to identify risk: local norms, adjusted cut-points, change scores, and a hybrid of change scores and adjusted cut-points. For the local norms approach, we used descriptive statistics to determine the mean fall and spring screener scores and standard deviations for each EPL group. A one-way analysis of variance (ANOVA) test was used to determine whether there were significant differences in performance among EPL groups at either the fall or spring timepoint. If screener scores varied by EPL, then students who fell 1 or more SDs below the mean for their EPL group would be considered at risk for DLD using this approach. For the adjusted cut-points approach, we lowered the original cut-point for identifying risk from 7/10 to 5/10, 4/10, and 3/10 and determined the percentage of screened students at each EPL falling below each of these cut-points for both the fall and spring. We looked within each EPL group to identify the cut-point that would place 20%–25% or fewer of students in the “at-risk for DLD” category. For the change score approach, we aimed to identify students who demonstrated below-average growth in screener score from fall to spring compared to peers and who therefore may be at risk for DLD relative to students demonstrating increased learning potential (Castilla-Earls et al., 2020; Gutierrez-Clellen & Peña, 2001). We used descriptive statistics to determine the mean change in screener score from fall to spring at each EPL. Then, we determined the number of students who demonstrated below-average change for their group. Using this approach, students who demonstrated growth that was below average compared to their peers would be considered at risk for DLD. Lastly, we recognized that some students may obtain a high score on the screener and consequently demonstrate less growth in screener score from fall to spring. We therefore used a hybrid approach to identify students who neither scored above a given cut-point (7/10 or 5/10 with the intent to tailor to EPL) at either screener timepoint nor demonstrated average or above-average change in score from fall to spring.

For each alternative data analysis approach used to identify risk (local norms, adjusted cut-points, change scores, and a hybrid of these), we calculated descriptive statistics to determine the percentage of students identified as at risk using the given approach within each EPL group. Chi-square tests allowed us to determine if these approaches led to at-risk rates that were equivalent across EPL groups. To compare differences in identification rates between each of the alternate classification approaches and the original approach applied to the same group of children, we used McNemar–Bowker tests. The McNemar–Bowker test is a nonparametric statistical test used to assess whether there is a significant difference in the proportion of paired binary responses. It is appropriate for within-subject comparisons when the outcome is dichotomous (e.g., at risk vs. not at risk) and focuses on the discordant pairs—cases where classification outcomes differ across the two methods.

Results

Research Question 1: Percentage of Multilingual Students Identified as at Risk for DLD Using Fall Score Only

Using the original cut-point of 7/10 points on the screener to determine risk for DLD, 85% of the 673 students (n = 569) who completed the OWL screener in the fall of 2021 were identified as at risk. The percentage of students identified at each EPL ranged from 63% (EPL 4) to 92% (EPL 2) and can be found in Table 2. There was a significant relationship overall between EPL group and percent identified as at risk such that higher EPL groups had lower proportions of children identified at risk, χ²(6, 703) = 23.44, p < .001.

Table 2.

Percentage of multilingual kindergarten students identified as at risk for developmental language disorder by English proficiency level (EPL) group and data analysis approach.

Data analysis approach	EPL 1		EPL 2		EPL 3		EPL 4
	Fall (N = 200)	Spring (N = 164)	Fall (N = 118)	Spring (N = 96)	Fall (N = 347)	Spring (N = 322)	Fall (N = 8)	Spring (N = 7)
Original cut-point	91% (n = 182)	78% (n = 128)	92% (n = 108)	77% (n = 74)	79% (n = 274)	49% (n = 158)	63% (n = 5)	0% (n = 0)
1. Local norms	16% (n = 32)	12% (n = 20)	18% (n = 21)	16% (n = 15)	19% (n = 67)	15% (n = 48)	13% (n = 1)	0% (n = 0)
2a. Cut-point 5	75% (n = 150)	54% (n = 88)	64% (n = 75)	45% (n = 43)	47% (n = 162)	15% (n = 48)	13% (n = 1)	0% (n = 0)
2b. Cut-point 4	63% (n = 125)	31% (n = 64)	43% (n = 51)	23% (n = 28)	31% (n = 109)	9% (n = 31)	0% (n = 0)	0% (n = 0)
2c. Cut-point 3	44% (n = 87)	23% (n = 38)	27% (n = 32)	16% (n = 15)	19% (n = 67)	5% (n = 17)	0% (n = 0)	0% (n = 0)
3. Change score		35% (n = 55)		40% (n = 36)		33% (n = 101)		33% (n = 2)
4a. Hybrid 7		32% (n = 53)		34% (n = 34)		26% (n = 82)		0% (n = 0)
4b. Hybrid 5		27% (n = 47)		19% (n = 20)		12% (n = 38)		0% (n = 0)

Research Question 2: Percentage of Multilingual Students Identified as at Risk for DLD Using Spring Score Only

Using the original cut-point of 7/10 points on the screener to determine risk for DLD, 61% of the 589 students (n = 360) who completed the OWL screener in the spring were identified as at risk at the spring timepoint. Overall, the rate of identification of multilingual students as at risk for DLD decreased from the fall to the spring timepoint by 24%, and this difference was significant according to the McNemar–Bowker test, χ²(3, 703) = 155.5, p < .001. Identification rates also decreased at each EPL.

The percentage and number of students identified as at risk in the spring at each EPL ranged from 0% (EPL 4) to 78% (EPL 1) and can be found in Table 2. There was again a significant relationship overall between EPL group and percent identified as at risk in the spring, χ²(6, 703) = 71.53, p < .001. Students with the lowest EPLs were identified as at risk at the highest rate, with a decreasing rate of identification as English proficiency increased.

Research Question 3: Alternative Analytic Approaches to Using English-Only Screening and EPL Data to Identify Multilingual Students as at Risk for DLD

To answer Research Question 3, we explored a total of four additional novel approaches to analyze the screener data. Table 2 provides a comparison of DLD risk identification rates for each of the data analysis approaches employed across EPL groups.

Local Norms

To determine risk status by using a local norms approach, we calculated the mean and standard deviation of the screener scores for children within each EPL group in the fall and spring. The mean (M) fall screener score and standard deviation for each EPL ranged from M = 3.98, SD = 2.34 (EPL 1), to M = 7.25, SD = 1.16 (EPL 4), and are reported in Table 3. A one-way ANOVA indicated that there were significant differences in fall scores between EPL groups, F(3, 669) = 21.97, p < .001. Post hoc pairwise comparisons using Bonferroni adjustments revealed that EPL 1 differed significantly from EPL 3 (p < .001) and EPL 4 (p < .001) and that EPL 2 differed significantly from EPL 3 (p < .01) and EPL 4 (p < .05). EPL 1 did not differ significantly from EPL 2, and EPL 3 did not differ significantly from EPL 4.

Table 3.

Mean language screener scores and standard deviations for Fall and Spring screening timepoints out of 10 total test items.

English proficiency level (EPL)	Fall		Spring
	M	SD	M	SD
EPL 1	3.98	2.34	5.35	2.45
EPL 2	4.66	2.07	5.70	2.29
EPL 3	5.51	2.33	7.21	1.93
EPL 4	7.25	1.16	8.71	0.76

The mean spring screener scores for each EPL group ranged from 5.35 to 8.71 and are reported in Table 3. A one-way ANOVA indicated that there were significant differences in spring scores between EPL groups, F(3, 585) = 34.73, p < .001. Post hoc pairwise comparisons using Bonferroni adjustments revealed that EPL 1 differed significantly from EPL 3 (p < .001) and EPL 4 (p < .001) and that EPL 2 differed significantly from EPL 3 (p < .001) and EPL 4 (p < .005). EPL 1 did not differ significantly from EPL 2, and EPL 3 did not differ significantly from EPL 4. The mean change in language score from fall to spring was M = 1.47, SD = 2.45.

We determined the number of students at each EPL scoring 1 or more SDs below that within-EPL-group mean for both fall and spring timepoints. These students would be considered at risk for DLD using the local norms approach. There was not a significant relationship between EPL group and percent identified as at risk using local norms in the fall, χ²(6, 703) = 2.69, p = .847. The overall rate of identification using this approach was 17.2% (see Table 2), and this was a significantly lower rate of identification than the original 7/10 cut-point approach used in the fall, McNemar–Bowker, χ²(3, 703) = 448.0, p < .001. There was a significant relationship between EPL group and percent identified as at risk using local norms in the spring, χ²(6, 703) = 14.13, p < .05. This was driven by the fact that 0/7 students in EPL 4 scored below 1 SD from the local mean in the spring. The overall rate of identification using this approach was 11.8% (see Table 2), and this was a significantly lower rate of identification than the original 7/10 cut-point approach used in the spring, McNemar–Bowker, χ²(1, 703) = 277.0, p < .001.

Adjusted Cut-Point Approach

To determine risk by using an adjusted cut-point approach, we lowered the screener cut-point to 5/10, 4/10, and 3/10 across EPL groups. When the cut-point for identification of risk was lowered from 7/10 to 5/10, the rate of DLD risk identification significantly decreased from 85% (n = 569) to 58% (n = 388) using the fall timepoint score on the OWL screener, McNemar–Bowker, χ²(1, 703) = 181.0, p < .001, and from 61% (n = 360) to 30% (n = 179) using the spring timepoint score, McNemar–Bowker, χ²(1, 703) = 181.0, p < .001. The number and percentage of students that would be identified as at risk at each EPL using a cut-point of 5/10 on either the fall or spring screener ranged from 0% (n = 0, EPL 4, spring) to 75% (n = 150, EPL 1, fall) and can be found in Table 2.

There was a significant relationship between EPL group and percent identified as at risk using a cut-point of 5/10 in both the fall, χ²(6, 703) = 51.38 p < .001, and the spring, χ²(6, 703) = 102.07, p < .001. Notably, a fall 5/10 cut-point only identified 25% or fewer students as at risk for students in the highest English proficiency group (EPL 4, 13% identified, n = 1). When applied to spring screener scores, a cut-point of 5/10 identified 15% of students in EPL 3 (n = 48) as at risk for DLD. This cut-point identified greater than 25% of students in EPLs 1–2 as at risk in both fall and spring.

The rate of identification of students as at risk for DLD continued to reduce at each EPL when the cut-point was lowered further to 4/10 and 3/10. The number and percentage of students that would be identified as at risk at each EPL using a cut-point of 4/10 on either the fall or spring screener ranged from 0% (n = 0, EPL 4, fall and spring) to 63% (n = 125, EPL 1, fall) and can be found in Table 2. The percentage of students that would be identified as at risk at each EPL using a cut-point of 3/10 on either the fall or spring screener ranged from 0% (EPL 4, fall and spring) to 44% (EPL 1, fall) and can be found in Table 2.

There was a significant relationship between EPL group and percent identified as at risk using a cut-point of 4/10 in both the fall, χ²(6, 703) = 57.58 p < .001, and the spring, χ²(6, 703) = 74.63, p < .001. There was also a significant relationship between EPL group and percent identified as at risk using a cut-point of 3/10 in both the fall, χ²(6, 703) = 41.82, p < .001, and the spring, χ²(6, 703) = 47.34, p < .001. Notably, when applied to spring screener scores, a cut-point of 4/10 identified 23% of students in EPL 2 as at risk, and a cut-point of 3/10 identified 23% of students in EPL 1 as at risk. These percentages represent the closest identification rates to our aim of identifying 25% or fewer students as at risk for DLD.

Change Score Approach

To use the change score approach, we identified the students that demonstrated below-average growth in screener score compared to their peers. A total of 559 students completed the screener in both the fall and the spring (n = 157 at EPL 1, n = 91 at EPL 2, n = 305 at EPL 3, and n = 6 at EPL 4). The average change in screener score from fall to spring across EPL groups was +1. Of the 559 students, 65% (n = 365) demonstrated a positive change in screener score from fall to spring, and 35% (n = 194) did not. Identifying students that did not demonstrate a positive change in score as at risk led to a significant reduction in identification rate compared to the originally applied 7/10 cut-point approach, McNemar–Bowker, χ²(3, 703) = 307.3, p < .001. The number and percentages of students that did not demonstrate a positive change in their screener scores ranged from 33% (EPL 4, n = 2; EPL 3, n = 101) to 40% (EPL 2, n = 36) and can be found in Table 2. There was not a significant relationship between EPL group and percent identified as at risk using the positive change approach, χ²(6, 703) = 10.27, p = .114.

Hybrid Approach

Using the hybrid approach, students that did not demonstrate a positive change or earn a high score on at least one screening timepoint would be considered at risk for DLD. The number and percentage of students that did not demonstrate a positive change in screener score from fall to spring or earn a score of 7 or higher at either screening timepoint ranged from 0% (EPL 4, n = 0) to 34% (EPL 2, n = 34) and can be found in Table 2. This reduced identification rate was significantly lower than the original 7/10 cut-point approach, McNemar–Bowker, χ²(2, 703) = 400.0, p < .001. The number and percentage of students that did not demonstrate a positive change in screener score from fall to spring or earn a score of 5 or higher at either screening timepoint ranged from 0% (EPL 4, n = 0) to 27% (EPL 1, n = 47) and can be found in Table 2. There was a significant relationship between EPL group and percent identified as at risk using the hybrid approach cut-point 7, χ²(6, 703) = 14.92, p < .05, as well as the hybrid approach cut-point 5, χ²(6, 703) = 24.88, p < .001.

Discussion

The present study addresses a significant research-to-practice gap in multilingual language assessment by examining the use of an English-only universal language screener for DLD in a multilingual student population. This investigation first analyzes outcomes when a school district administers such screening to multilingual students, then identifies alternative analytical approaches for the collected data. These alternative methods aim to improve identification rates of multilingual children at risk for DLD, bringing these rates more closely aligned with established prevalence expectations. By focusing on practical solutions to this assessment challenge, this research contributes to narrowing the divide between theoretical knowledge and educational practice in multilingual contexts.

Screening in English-Only Overidentifies Risk for DLD in Multilingual Students

We analyzed the outcomes of giving an English-only language screener to multilingual kindergartners when they entered the district in the fall. Using the original cut-point of 7/10 for the OWL screener (i.e., scoring below 7 indicates risk for DLD), we found that 85% of multilingual kindergartners were identified as at risk for DLD according to their fall screener score. This finding is in line with our hypothesis that the rate of identification of DLD in a multilingual student population when using an English-only measure would be much higher than the prevalence rate of the disorder in the general population (i.e., 7.58%; Norbury et al., 2016; Tomblin et al., 1997) or the rate expected to be identified as at risk on a language screener (i.e., 20%–25%; Bao et al., 2024). This finding is also consistent with the literature that applying monolingual approaches for language evaluation to a multilingual population is likely to be biased against multilingual children (Peña et al., 2023; Sullivan & Bal, 2013) and that language assessment and screening that happens only in the student's nondominant language underestimates a multilingual child's actual language ability and increases the likelihood of overidentifying them as at risk for DLD (Bialystok et al., 2008, 2010; Peña et al., 2011).

In addition to understanding the total rate of identification in a multilingual student body when given an English-only language screener in the fall of their kindergarten year, we examined how risk status, as determined by the fall language screener, would differ across English proficiency levels. We found that students with the lowest EPLs were identified as at risk at the highest rate, with a decreasing rate of identification as English proficiency increased. Specifically, 91% and 92% of students at the lowest EPLs (1 and 2, respectively) were identified as at risk for DLD according to their fall screener score. Identification rates remained high for students with higher EPLs (79% and 63% of students for EPLs 3 and 4, respectively). These findings are indicative of possible floor effects occurring at the fall timepoint. This is likely due to testing students in their L2 before they have sufficient exposure to it and is expected based on prior research (Catts et al., 2008). The difference in identification rates across EPLs suggests that those students with higher English proficiency appeared more ready to be screened in English, despite still being overidentified as at risk for DLD using their fall screener score.

Given the possibility of finding floor effects due to screening students too early in their school experience, before they had developed sufficient English language proficiency (Catts et al., 2008), we examined outcomes of screening students for DLD in the spring using the same English-only screening measure. Adding a second screening timepoint was recommended by the research team under the premise that measuring growth as part of the screening process may be helpful when assessing a skill that is still developing at the time of assessment (Catts et al., 2008; Speece, 2005). Consistent with this literature and our hypothesis, we found that the rate of identification of students as at risk for DLD decreased in the spring, after students had completed several months of instruction. While 85% of students were identified as at risk for DLD according to their fall screener score, 61% of students were identified as at risk in the spring. Additionally, identification rates decreased from the fall to spring at each EPL. This demonstrated the benefits of screening language skills at two timepoints for a student population that is still developing proficiency in the language in which they are being tested. However, despite the observed reduction in identification rates by the spring of their kindergarten year, the screener continued to overidentify students as at risk for DLD. Our findings confirmed our hypothesis that English-only language screening would overidentify the multilingual children in this school district as at risk for DLD and substantiated the importance of identifying alternative analytical approaches for English-only screening and English proficiency data to identify multilingual students as at risk for DLD.

Alternative Data Analysis Approaches Yield More Appropriate Rates of DLD Risk Identification

Our data revealed significant differences in how different EPL groups perform on the language screener. We therefore considered the needs of each EPL group individually when applying alternative data analysis approaches, aiming to see identification rates of approximately 20%–25% of the multilingual student population and of each EPL group.

Local Norms

When we looked at the number of students who scored 1 SD or more below the mean for their EPL group, regardless of screening timepoint, we found that anywhere from 12% to 19% of students would be identified as at risk for DLD if applying a local norms approach. This significantly lower identification rate seen was consistent with our hypothesis and prior empirical evidence for tailoring norms to the target population (Altman et al., 2022; Gathercole, 2013; Peña et al., 2011). Since we created the local norms, we mathematically assured that less than 20% of students would be in the risk category in each group, hence the lack of significant relationship between EPL group and risk status when using this approach in the fall. By only placing children who performed significantly lower than other children of similar English proficiency as at risk, we aimed to avoid overidentifying less proficient English speakers as being at risk for DLD. However, we note that identification rates using this approach are below the expected range of 20%–25%, and this may lead to underidentification of students as at risk for DLD, potentially precluding some multilingual kindergartners from receiving the support that they need.

Adjusted Cut-Points

In aiming to identify 20%–25% of students at each EPL as at risk for DLD, we found that lowering cut-points reduces the rate of overidentification for multilingual students, consistent with prior empirical evidence (Barragan et al., 2018; Thordardottir, 2015). Additionally, we found that the most appropriate screening cut-point and screening schedule varied by EPL, which is consistent with the literature on floor effects showing that language proficiency can impact testing performance (Catts et al., 2008). We found that the nine students with the highest English proficiency (EPL 4) appeared ready to be screened in English with a cut-point lowered to 5/10 when they entered school in kindergarten. This approach identified 12.5% of students in EPL 4 as at risk for DLD in the fall and 0% in the spring. Screening identification rates in this group should be interpreted with caution given the small number of students in EPL 4 that were screened (n = 7). Students with EPLs 1–3 were still identified at high rates (> 25%) when various cut-points were used to identify risk based on their fall screening scores. Therefore, students with lower EPLs appeared to benefit most from an additional semester of schooling to support English language growth before being able to gain useful information from single-language screening results. When the preferred cut-points were applied to spring screener scores for each EPL group, a cut-point of 5/10 identified 15% of students in EPL 3 as at risk, a cut-point of 4/10 identified 23% of students in EPL 2 as at risk, and a cut-point of 3/10 identified 23% of students in EPL 1 as at risk. While we found adjusting cut-points useful in lowering the rate of at-risk identification, the variability in using this approach across EPLs may introduce additional challenges for a school district. Additionally, it encourages waiting until the spring to identify students, which can be problematic for students that could benefit from earlier intervention. Further, it appeared that students in EPL 1 benefited from a notably low cut-point, indicating that they only needed to score 3/10 to not be considered at risk, which we note is only marginally above the number of items we would expect children to answer correctly if performing at chance. Thus, the practical implications of such low screener scores supported our decision to explore other approaches of using the data to identify risk.

Change Scores

Our findings from the adjusted cut-point approach revealed that screening multilingual students for DLD in the spring could be beneficial in supporting their performance on the English-only screener. Since we attributed the higher screener scores seen in the spring to English language growth, we also wanted to capture the growth itself, as growth and learning potential has been implicated as an indicator of typical language development (Castilla-Earls et al., 2020; Gutierrez-Clellen & Peña, 2001). When we identified students who did not demonstrate a positive change in screener score from fall to spring (the average full point increase across groups) as at risk, we did not find a significant relationship between EPL and risk status, thus suggesting that children in each EPL group are equally likely to grow by 1 point across the school year. Therefore, it appears that this approach can reliably be used to identify risk across all EPL groups, making this a feasible option for school districts to identify DLD risk status without the unintended influence of EPL. This finding is consistent with the literature on dynamic assessment, which uses a test–teach–retest approach to monitor language learning progress (Castilla-Earls et al., 2020; Gutierrez-Clellen & Peña, 2001). Lack of language learning, regardless of the language being learned and the students' language background, is a hallmark of DLD (Tomblin et al., 1997) and therefore could indicate risk for children of any language proficiency level. However, when we looked at those students who did not demonstrate any positive change on the screener, we identified approximately 35% of students on average across EPLs, which is still higher than our goal identification rate of 20%–25%. Additionally, while we view a 1-point increase as a small change in score, identifying students who did not gain at least 2 points would identify approximately half of the multilingual kindergartners as at risk. Therefore, in our sample, using positive change alone as a method for risk identification may still overidentify students, despite the apparent reliability of the approach across EPLs.

Hybrid

We recognized that there may be students who do not demonstrate a positive change on their language screener but who already performed well enough on the screener that they scored above a given cut-point. Therefore, we expected that a hybrid approach that identifies risk if a student either fails to demonstrate a positive change in screener score or does not score above a pre-identified cut-point might bring DLD identification rates closer to the expected prevalence of DLD in the population. We found that using a hybrid approach of positive change and scoring greater than 6 (cut-point 7/10) on at least one screener yielded identification rates closer to our target range of 20%–25% for students at higher EPLs (EPLs 3 and 4, 26% and 0% identified, respectively) and a hybrid approach of positive change and scoring greater than 4 (cut-point 5/10) on at least one screener yielded identification rates closer to our target range for students at lower EPLs (EPLs 1 and 2, 27% and 19% identified, respectively). Using a different approach for higher versus lower EPLs is consistent with our findings that, at both the fall and spring timepoints, screener scores did not differ significantly within the lower EPL groups (1–2) or the higher EPL groups (3–4), but they did differ significantly across low EPL (1–2) and high EPL (3–4).

Overall, our hypothesis to our third research question was substantiated, in that all the additional approaches that we explored lowered the overall DLD risk identification rate. These approaches appear to represent a more appropriate and effective use of single-language screening data for a multilingual population, as compared to the original identification approach that was developed for monolingual students. This finding is consistent with prior empirical evidence that tailoring analytical approaches to account for multilingual differences can bring identification rates closer to expected screener rates and the prevalence rates in the general population (Altman et al., 2022; Gathercole, 2013; Peña et al., 2011).

Clinical Implications

Based on our findings, we provide preliminary clinical implications for practitioners. Despite the limitations of our study addressed below, we believe providing these implications and suggestions is important because the issue of using single-language screening data for multilingual children is likely to become even more pressing for districts with new state laws mandating reading screenings and advocacy for DLD screenings underway. We encourage all districts to screen for DLD when students enter school. DLD is a common developmental disorder (Norbury et al., 2016), and screening for it early in a child's school experience can proactively mitigate its negative impact on students' academic outcomes (Adlof & Hogan, 2019; Bao et al., 2024; Catts & Hogan, 2020). As a first step to implementing language screening, we suggest that districts refer to the list of commercially available DLD screeners reviewed by Bao et al. (2024) and review their findings and recommendations. However, we caution districts that the researchers found a dearth of available screeners for languages other than English. Therefore, they focused their reviews and recommendations on the commercially available screeners in English. Thus, use of their recommended screening measures would likely require adaptations such as those described in this study. Given that the population of ELLs enrolled in public schools across the nation continues to rise (National Center for Education Statistics, 2024), we expect to see school districts with a largely multilingual student body face similar challenges to those discussed in the present study when selecting a screening tool for DLD.

The results from our study clearly indicate that an English-only language screener will overidentify multilingual students with risk for DLD, particularly for students with the lowest English proficiency. Aligned with prior research on multilingual language assessment, we conclude that a monolingual screening approach should not be applied to multilingual students. It will overidentify students as at risk and yield numbers that are not manageable for the district to assess comprehensively, therefore counteracting the purpose of screening students. Our results add to the literature on the feasibility of single-language testing when using norms that are appropriate for the tested population (Altman et al., 2022; Gathercole, 2013; Peña et al., 2011). Further, they serve as preliminary guidance for school districts to more appropriately and effectively analyze data if administering an English-only language screener to multilingual students is unavoidable. Also, this provides impetus for SLP leaders in public school settings to engage in district-level advocacy to support thoughtful approaches to DLD screening. We outline recommended steps that districts can follow if using an English-only DLD screener to screen a multilingual population, and we encourage school SLPs to advocate for DLD screening in their districts and to share relevant research and evidence-based screening recommendations with their district leaders.

Recommended Steps for Districts Administering Single-Language DLD Screeners

Based on our findings, we recommend the following steps for districts with a large multilingual student population if English-only DLD screening is the only option:

Screen at two timepoints (fall and spring). Our results indicate that screening in the spring allows time for students to further develop their English proficiency, which will support them in achieving a higher and more useful score on an English-only language screener. It also avoids possible floor effects that could be expected from a fall-only screener (Catts et al., 2008). However, waiting until the spring to screen could prevent students who need additional support from receiving earlier intervention. Therefore, we also recommend screening in the fall, which can identify some students sooner than later and additionally provides the opportunity to look for growth in screener score when administered again the following semester.

Use local norms (fall). Identifying the number of students who score 1 SD or more below the mean score at their EPL will yield approximately 16% percent of students who may be at risk for DLD. While this method may not capture everyone who is at risk, it appears to be a reliable method to identify a manageable number of students who may benefit from additional support as early as the fall of their kindergarten year. We suggest determining and using local norms at each EPL to identify students as at risk in the fall. For example, we found significant differences in screener score between lower EPL Groups 1 and 2 and higher EPL Groups 3 and 4. For students identified through a local norms approach in the fall, we suggest immediately initiating the process for them to receive further assessment by referring directly to the SLP.

Use a hybrid approach (spring). Following the spring screening timepoint, we recommend that districts return to the screening data for further identification. We suggest an approach that takes change scores, which demonstrate learning potential and language growth, into consideration for determining risk. However, determining risk by looking for positive change in screener score alone is still likely to overidentify students (in our sample, it identified approximately 35% of all students as at risk for DLD). Therefore, we recommend using a hybrid approach that identifies risk if a student either fails to demonstrate an average or above change in screener score or does not score above a pre-identified cut-point.

First, districts can determine the mean change from fall to spring across EPL groups. Identify students whose change score falls below this mean. Then, trial cut-points to identify approximately 20%–25% of students in each EPL group whose change score fell below the mean or whose screener score fell below the trialed cut-point on at least one screener. In our sample, EPL Groups 1 and 2 were significantly different in mean screener score as compared to Groups 3 and 4. Therefore, we found a cut-point of 7/10 to be appropriate for students at EPLs 3 and 4 and 5/10 for students at EPLs 1 and 2. We recommend that students identified as at risk in the spring using a hybrid approach should be directly referred to the SLP for further assessment.

We propose our recommendations with the suggestion of aiming to identify 20%–25% of the multilingual population as at risk for DLD to provide these students with the opportunity for further diagnostic assessment. Given the discussed barriers to a comprehensive diagnostic assessment for a multilingual student, some schools or districts may continue to face limitations in providing appropriate further assessment for this number of students, especially given that monolingual students will be identified as well. We therefore suggest the possibility of using these data analysis approaches for an English language screener as an initial step and then adding an additional measure to help triage the comprehensive assessment process. For example, administering an additional task, such as nonword repetition, or sending out a parent or teacher language/development questionnaire, could add one more data point that would ultimately support triangulation of data for an official diagnosis (Castilla-Earls et al., 2020). In the meantime, this additional data point could be used to prioritize the initiation of next steps for any students who also demonstrate difficulties on this added measure.

The needs of any given school district will be highly variable, depending on factors such as student demographics, availability of school personnel, and screening needs. We recommend that districts make screening decisions that are in line with their specific student demographics and screening needs. Overall, our findings reiterate the importance of taking multilingualism into consideration when assessing students' language skills, whether for the purpose of diagnosing or identifying risk for DLD. Based on our results, we encourage districts to take a thoughtful and tailored approach to looking at their screening data. We also recommend that school districts collaborate with SLPs in their district and form partnerships with researchers in the field, as SLPs and SLP-researchers can provide valuable support in analyzing language data and identifying students who will benefit from language intervention.

Limitations

A limitation of this study is that the researcher-developed OWL screener has not yet been validated. Therefore, the screener does not have pre-established sensitivity and specificity rates or predictive accuracy data to compare our results against. Positively, the AUC results from administration of a similar type of screener—an inspiration for the OWL screener—to a different student population of first and second graders were considered excellent (Hendricks et al., 2019). Additionally, because the screener was administered as part of typical school procedures and screening data were shared as part of the RPP, we do not have the diagnostic outcomes of the students that were screened in this study; therefore, we cannot know with certainty the percentage of students in each EPL that should truly be identified as at risk for DLD. An additional limitation worth noting is that we do not have data on the students' language skills and development in their L1, which is an important factor in the differential identification of DLD. These factors limit our ability to determine effective data analysis approaches with certainty. Rather, we present possible alternative methods for analyzing outcomes on an English-only language screener with the aim of bringing the number of multilingual students identified closer to what would be expected given the prevalence of DLD in the general population.

While it is impossible to know for certain that the lowered identification rates accurately captured the students who are truly at risk for DLD, the alternative data analysis approaches demonstrate promise given that they were applied methodically. For example, the hybrid approach appears promising in capturing those students at risk for DLD, as lower-than-average language scores coupled with lack of language growth are hallmarks of the disorder. It is also important to note that every district will have a different set of student demographics, screening needs, and capacity to support the screening, diagnosis, and intervention process. Given that our data come from only one school district, the approaches that were most effective for the population of students in the present study may not generalize to all districts. Lastly, the approaches that we identify and recommend for screening a multilingual student body if using an English-only screener represent a preliminary framework for districts to consider, as they have not yet been trialed and validated in practice.

Future Directions

Given the limitations of the present study, more research is needed to study the effectiveness of the recommended approaches on appropriately identifying multilingual children as at risk or not at risk for DLD. We recommend that researchers continue to partner with school districts and support them in implementing universal language screenings. Research studies as part of these partnerships should study the effectiveness of applying the recommended approaches from this study to a screener that has already been validated. Validating the OWL screener administered in this study is a suggested next step to provide additional context for our results. We also recommended following the students who were identified as at risk for DLD through use of the suggested approaches to ascertain their later diagnostic outcomes following a comprehensive assessment to provide more robust psychometric data. Given the limited availability of screeners in other languages and the barriers to assessing children in all of the languages that they speak (Arias & Friberg, 2017; Bao et al., 2024; Freeman & Schroeder, 2022), further research is also needed to continue understanding the effectiveness of alternative methods for screening multilingual children for DLD, such as the use of language agnostic approaches and cognitive tasks (Antonijevic-Elliott et al., 2019; Ebert & Pham, 2019; Freeman & Schroeder, 2022). We suggest studying the outcomes of using these measures in conjunction with our recommended data analysis approaches for administering an English-only, language-based screener for DLD, as a combined approach may yield strong sensitivity and specificity while limiting the resources required of a given school district.

Data Availability Statement

Data are available upon reasonable request to the corresponding author.

Funding Statement

The first author was a trainee supported by Project INTERSECT (ED/OSERS-OSEP Project H325D230037: Project Director: Emily Lund). Research reported in this publication was supported by National Institute on Deafness and Other Communication Disorders Grant R01DC016895 (Principal Investigators: Tiffany P. Hogan and Julie A. Wolter). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the policy of the Department of Education, and you should not assume endorsement by the Federal Government.