Online Stratification: How Academic Performance Indicators Relate to Choices Between Cyber Charter Schools
by Bryan A. Mann & Stephen Kotok - 2019
Background/Context: A primary argument that supports charter school policy assumes students favor schools with high academic performance ratings, leading to systemic school improvement. Previous research challenges this assumption but has limited generalizability because geographic and enrollment constraints limit student choice sets.
Purpose/Objective: This study examines student enrollment patterns within cyber charter schools in Pennsylvania, a state where elected policymakers tend to view choice as a means for school improvement. Cyber charter schools are advantageous to study in this context because they have fewer enrollment barriers, helping researchers account for constraints found in previous studies.
Research Design: Using consecutive years of student-level enrollment data, we use descriptive statistics and multinomial logistic regression analyses to answer the following questions: Is a particular cyber charter school more popular if it displays relatively higher performance on academic indicators? To what extent do enrollments in the highest performing cyber charter school relate to the demographics of students and school environments that they left?
Findings/Results: The findings suggest that despite the more accessible choice sets inherent in the cyber charter school sector, academic performance indicators still are not linked to popularity within the sector. Enrollment clustering persists along student demographics and feeder district traits.
Conclusions/Recommendations: These findings suggest that even in the cyber charter school sector where key enrollment restrictions are removed, inequitable enrollment patterns persist. These findings continue to challenge basic assumptions used in school choice policy framing. Policymakers should consider this evidence when and if they design and implement charter school policy, creating policy that accounts for inequitable enrollments that occur under current policy logic.
School choice advocates justify the use of school choice policy with the logic that enrollment preferences put pressure on schools within a choice environment to improve (Betts, 2005; Berends, Cannata, & Goldring, 2011; Chubb & Moe, 1990). The expectation of school improvement through school choice assumes that families have accurate knowledge and capacity to choose the schools that spur systemic improvement (Orfield & Frankenberg, 2013, p. 4455). Previous research about school choice preference is mixed, as some scholars suggest that families are more likely to select schools that promote improved growth on academic achievement tests (e.g., Hanushek, Kain, Rivkin, & Branch, 2007), while others argue non-academic logics explain choices (e.g., Orfield & Frankenberg, 2013). One limitation of previous research is that previous studies examine brick-and-mortar school choice scenarios where enrollments are restricted based on geography. These studies also tend to focus only on urban places (for a comprehensive review of the last two decades of school choice research, see Berends, 2015). To further understand the extent to which school choices reflect school performance indicators, we account for these prior limitations by examining a context in which school choices are less bound by geographic and other non-academic constraints.
Geographic constraints create inherent limitations for prior research because district boundaries, distance, and travel times limit available choices for families (Orfield & Frankenberg, 2013). Meanwhile, several other non-academic factors guide choice decisions (Berends & Zottola, 2009; Holme, 2002; Marsh, Carr-Chellman, & Sockman, 2009; Billingham & Hunt, 2016). Examining enrollment patterns linked to choices within the cyber charter school sector provides a dual advantage of examining choice sets that contain fewer geographic limitations and a sample of students who seek the same general type of school structure (i.e., a full-time online school). This means a sample of cyber charter school students mitigates previous geographic limitations and non-academic enrollment constraints inherent in prior research.
To evaluate enrollment patterns within a cyber charter sector, we focus on Pennsylvania, which is a state with high levels of cyber charter schooling. We use individual-level data for the 201112 cohort of students transferring into cyber charter schools. Using this data, our analyses examine the extent to which enrollments within the cyber charter school sector follow the market rationale embedded in choice assumptions generally and in Pennsylvania policy specifically (Pennsylvania Department of Education, 2004). Foreshadowing the results, our findings suggest that within-sector stratification based on social demographics persists despite the exclusion of geographic and other constraints.
To depict these trends, we first describe the demographic composition of cyber charter schools as compared to the demographic characteristics of other students in the state. The purpose of including these descriptive data is to consider generally if socioeconomic differences exist in the cyber charter school sector. Then, in order to fulfill the purpose of our research as detailed, we explore how enrollments within the cyber charter school sector relate to cyber charter school performance metrics and other characteristics of students and schools. These analyses answer two primary research questions:
R1: Is a particular cyber charter school more popular if it displays relatively higher performance on academic indicators?
R2: To what extent do enrollments into the highest performing cyber charter school relate to the demographics of students and school environments that they left?
The answers to these questions contribute to academic literature on school choice by accounting for previous research limitations. These answers also have value to policymakers because they extend knowledge on cyber charter school enrollment patterns, necessary because the current federal administration supports and seeks to expand their operation (Turner, 2017).
FRAMEWORK AND LITERATURE: CHOICES, QUALITY, AND CYBER CHARTER SCHOOLS IN PENNSYLVANIA
The framework that guides this research relates to the market assumptions linked to school choice theory in general. This study considers the assumption that students select the highest academically performing cyber charter schools. If students select schools whose previous students score lower on academic achievement tests and have lower graduation rates, then it is suggested that even within the cyber charter school sector, choices may not lead to systemic improvement in schooling outcomes. Further, potential equity concerns emerge if choice decisions into higher performing schools differ by student demographics.
ENROLLMENT CHOICES AND QUALITY
One premise of school choice reform is that widespread innovation and improvement occur when policy promotes choice and competition in the educational system (Friedman, 1955; Chubb & Moe, 1990; Moe & Chubb, 2009). Advocates of school choice argue that choosers in the system make rational choices about the schooling option they see as the best fit for their children. Their choices then put pressure on schools to improve to meet the needs of parent consumers or face enrollment losses. A failure to enroll students hinders a schools ability to target other students, and a resultant decline in enrollments should eventually lead to closure. Over time, this process is assumed to promote system-wide improvement because all schools improve their practice to entice students to enroll (e.g. Chubb & Moe, 1990; Berends et al., 2011).
Research has challenged the market rationale of choice advocates, suggesting that practice does not match theory. For instance, Orfield and Frankenberg (2013) argued that the original operationalization of choice rests on flawed assumptions (pg. 4555). These flawed assumptions include failure to consider equal access to transportation, school availability within a reasonable distance, and fair enrollment processes. Another criticism is that market advocates inaccurately assume families have equal access to reliable information about possible options (Bell, 2009; Holme, 2002; Teske & Schneider, 2001; Hastings & Weinstein, 2008). Indeed, previous research on school choice indicates that test scores and achievement data have not served as the best predictors of choices (Butler, Carr, Toma, & Zimmer, 2013). In addition to geographic constraints, an apparent misalignment between choices and perceived quality occurs because parents make selections based on alternate logics. Other preferences range from differing constructions of quality, alternative expectations of schooling, and racially motivated rationales (Berends & Zottola, 2009; Billingham & Hunt, 2016; Holme, 2002; Marsh et al., 2009). Parents also have segregated social networks, which tend to cause demographic clustering of school choices (Bell, 2009; Holme, 2002).
One of the primary byproducts of differences in choice preferences is that schools of choice, particularly charter schools, tend to be more segregated than traditional public schools (Frankenberg, Siegel-Hawley, & Wang, 2010). Although geography and housing segregation often contribute to charter school segregation, students living near each other and who have integrative school choices still have been shown to select options that make the system more segregated (Kotok, Frankenberg, Schafft, Mann, & Fuller, 2017; Stein, 2015). Our study informs these conversations, seeking to explore school choices, academic indicators, and stratification within the cyber charter school sector.
CYBER CHARTER SCHOOLS
As of 20132014, the year following the data in our study, cyber charter schools operated in 26 states and served about 200,000 students (Evergreen Education Group, 2014). Arizona, California, Ohio, and Pennsylvania have had the highest enrollments of cyber charter schools in the United States, all with more than 30,000 cyber charter students yearly (between 2% and 4% of their statewide student populations). The number of states allowing programs and number of students within states has consistently risen during the past decade (Evergreen Education Group, 2014).
While cyber charter schools represent only about 2% of all charter school students in the United States (Huerta, dEntremont, & González, 2006), studying their enrollment patterns offers unique opportunities for research. A study on cyber charter school choice patterns allows researchers to examine which schools are popular across a variety of demographic and geographic settings. Cyber charter schools purport to offer a more pure market scenario since students are allowed to select any of the alternative choices across geographic boundaries with fewer restraints, especially since cyber charter schools in Pennsylvania provide Internet access and computers to their enrolled students.
In addition to eliminating geographic boundaries, another advantage of examining cyber charter school enrollments is it enables researchers to isolate a single group of students (e.g., online students) and thus eases selection-bias concerns found in previous studies. Research suggests that cyber charter schools serve a unique, niche population of students with similar goals that relate to learning remotely from a distance (Ahn, 2011). A study of cyber charter schools therefore does more to equalize niche preferences than general studies on a diverse range of schools, better capturing how metrics of academic quality relate to school selection. This strategy of comparing cyber charter school students only to other cyber charter school students is more accurate than considering choosers of schools of choice to non-choosers in traditional schools, as typically seen in school choice research (i.e., Cowen, 2010). Therefore, despite the previous research that indicates test scores are poor predictors of choices, we intentionally use academic metrics as predictors because previous studies argue that the inadequacies were caused by certain limitations that induce a misalignment between test scores and enrollment. The cyber charter school scenario hypothetically should mitigate these limitations.
In addition to providing advantages to studying school choice theory, research on cyber charter schools has implications for policy and practice. While the use of cyber charter schools is growing in states and has landed on the federal agenda, current research relating to cyber charter school funding and practice is critical of cyber charter schools. DeJarnatt (2013) and the Pennsylvania Department of the Auditor General (2012) suggested there is a lack of financial accountability in Pennsylvania cyber charter schools. Huerta, dEntremont, and González (2006) raised similar concerns about cyber charter schools nationally, detailing a lack of legal oversight. Hasler Waters, Barbour, and Menchaca (2014) restated these concerns and showed in their review of cyber charter school literature that there is still a lack of accountability, unfounded funding structures, and inadequate school performance as measured by test scores and dropout rates.
Other research on cyber charter school performance shows that cyber charter schools tend to perform lower than traditional schools on test scores and graduation rates. For example, Miron, Horvitz, and Gulosino examined high school graduation rates and Adequate Yearly Progress (AYP) scores and showed lower graduation rates (37.6% compared to 79.4%) of cyber charter schools compared to all U.S. schools in 201112 (NEPC, 2013). Student-level studies conducted by the Center for Research on Education Outcomes (CREDO, 2011, 2015), focusing on cyber charter schools specifically in Pennsylvania and later adding other programs nationwide, indicated student growth on achievement tests in cyber charter schools is lower compared to traditional public schools. A peer-reviewed study on Ohio reflects these findings (Ahn & McEachin, 2017). However, the academic research is unclear as to whether the inadequate performance relates to school quality or if it is explained by other reasons such as the academic history of enrolled students.
The growth of cyber programs, coupled with the low-academic performance metrics, has created an imperative to understand the composition of students likely to enroll in cyber charter schools. According to a report from the National Education Policy Center (NEPC, 2016), cyber charter schools, when compared to traditional public schools, enroll lower shares of FRL-eligible students (29.4% compared to 49.9% nationally) and a higher than average proportion of White students (69.6% to 49.8%) (NEPC, 2016). These demographics are aggregate and are compared to national numbers that may include states that do not have cyber charter schools. Still, these populations of students historically tend to have higher levels of academic achievement in brick and mortar schools, but have lower test scores in cyber charter schools.
Taken together, prior research on cyber charter schools presents many concerns and questions relating to academic performance and enrollment patterns. Our study builds on these understandings because we use individual-level data to illuminate the student and school characteristics of a single state. Prior studies explore aggregate, national enrollments, and tend to generalize findings. Our study helps to understand specific, individual enrollments and how they relate to school academic performance metrics.
Pennsylvania is an ideal context to study cyber charter school enrollment because it has among the highest number of cyber charter school students in the United States (Evergreen Education Group, 2014). Cyber charter schools are embedded within the institutional structure of Pennsylvania policy and have an established history and reputation. The enrollment of students into the programs is robust and widespread enough to offer a rich setting for analysis, and the state has codified policy that captures the theoretical logic of market-based school choice policy.
Charter school policy in Pennsylvania started with Act 22 of 1997, and the Pennsylvania Department of Education (PDE) has explained that its charter schools have goals to [i]mprove pupil learning, [e]ncourage the use of different and innovative teaching methods and [p]rovide parents and pupils with expanded choices in the types of educational opportunities that are available within the public school system (2004). Choice assumptions are clearly embedded in the law, which represents how school choice is conceived theoretically: choice leads to innovation and improvement.
To fund all charter schools (including cyber charter schools) in Pennsylvania, the law mandates that when a student leaves for a charter school, their home school district pays about 70 percent of the per pupil expenditure of the home school district (which we label as feeder district) for every student who leaves (Hardy, 2015). This means individual students are not restricted financially when enrolling in cyber charter schools because tuition is paid with local tax dollars. As of the Oct. 1, 2014 , there were 13 authorized cyber charter schools in operation in the state, and the enrollment was 37,289 (30,458 non-special education, 6,831 special education), or approximately 2% of the statewide student population (Pennsylvania Department of Education, n.d.). If the sector were its own school district, it would be second only to Philadelphia as the largest school district in the state. There are some minor geographic footprint stipulations included in Pennsylvania cyber charter schools policy, but enrollments into cyber charter schools remain statistically not related to the location of the cyber charter school headquarters (Center on Reinventing Public Education, 2015, p.5).
Our analysis uses a dataset comprised of student-level data across two years in Pennsylvania. Using this data provided by the Pennsylvania Department of Education (PDE), we created a subsample of 4,088 students who exclusively attended a non-cyber public school (traditional public school or brick-and-mortar charter school) in 201011 and chose to then enroll in a cyber charter school in 201112. We then linked the individual data with 201011 school-level data from the Common Core of Data (CCD) and PDE. The rationale for selecting students who made a direct move from a brick-and-mortar school into a cyber charter school entailed the advantage of knowing the enrollment demographics of each students previous school. We would not have been able to capture these demographics if we had included students only enrolled in cyber charter schools, because our dataset does not include addresses or residential locations of individuals (and thus we had to extrapolate this based on their previous traditional school placement). We used 201112 because it was the most recent cohort available to us at the time of the research.
We primarily focused on the five largest cyber charter schools in the state at the time of this analysis, as they accounted for 90% of all cyber enrollees. These schools had varying levels of student achievement (see Table 1 in the findings section for list of these schools and their enrollment; the names of the schools are pseudonyms). We also included two smaller categories of cyber charter providers labeled as Intermediate Units (IUs) and small independent providers. IUs are organizational units between school districts and PDE, and thus IU cyber charter schools are those that individual IUs created but are still open to all students across the state. The independent cyber charter schools were three extremely small schools that we combined into a single category because of their small share in the sample. We felt comfortable collapsing IUs and independent cyber charter schools, as the individual enrollments were so small and the demographic and achievement data did not have great within-variation.
The outcome variable at focus in the study was a categorical variable that included each cyber charter school. In the tables, each school is also labeled with its enrollment size, percentage proficient rates on the prior years (201011) Math and Language Arts statewide tests, and the prior years graduation rate percentage. The purpose for including these indicators was to understand and report the metrics to which families would have had access while selecting the cyber charter school. As mentioned in the conceptual framework, since we only examine choices between cyber charter schools of only those students who select cyber charter schools, we mitigate concerns based on geography and selection bias based on niche. Therefore, this outcome variable allows us to explore if this specific sample of students indeed makes choices within the sector based on the popularly available academic achievement metrics. To ease interpretation, we provide the average percent of the three academic indicators and rely on this average in discussing our findings.
Individual student indicators. As mentioned, a primary concern in previous literature is that charter school enrollments exacerbate school segregation (Frankenberg et al., 2010). With this consideration, the first indicator we use is the race of a particular student, which was provided in the original dataset. Further, Pennsylvanias cyber charter school funding policy includes provisions that allow for greater funding to cyber charter schools for special education status (Hardy, 2015). Therefore, we include whether a student has an Individual Education Plan (IEP) as a proxy for special education status.1
Previous school traits. The NCES data allowed us to identify the students former school and characteristics of that school. In order to capture location, we included a variable for census-defined community type (urban, suburban, town, and rural). Geographic and selection bias concerns were assumed to apply less (or perhaps not at all)2 with these students choices because cyber charter schools are fully online and not place-based. The academic performance of the previous school likely weighs into a decision to leave a brick-and-mortar school, so we account for whether the students previous schools had met Adequate Yearly Progress (AYP) the year before the student transferred.3 We considered schools to be meeting performance expectations if they made AYP, were in only warning phase (one year of not meeting goals in one subject area), or making progress (passed AYP for first of two probationary years). We also used the percent of free-and-reduced-lunch (FRL) students at the previous school as a proxy for socioeconomic status. Although we lacked individual FRL data, school socioeconomic status has been found to be a critical factor on learning and social outcomes (Rothwell, 2012). Next, given that Philadelphia is the largest metropolitan area in the state, we created a variable to capture whether a student lived there or not. Other previous school traits include a dummy variable to indicate if the student came from a traditional public school (TPS) or brick-and-mortar charter school and a variable for dangerous schools based on a PDE persistently dangerous school measure for 201011 to identify if prior school danger related to choices among particular cyber charter schools.4
We use descriptive statistics to understand the net popularity of given schools. This allowed us to examine general trends and identify if there were differences between schools based on the variables we chose. These descriptive statistics allow us to show a clear picture of our sample5 and how particular traits, individual demographic as well as those of feeder schools, are linked to the popularity of given cyber charter schools.
The analytical portion of the study used multinomial logit analysis in order to infer which factors related to choices into each school. Multinomial logits are well suited for our research questions, as they are frequently used to analyze characteristics of choosers in scenarios with multiple choices such as college majors (Dickson, 2010), transportation types to school (Wilson, Marshall, Wilson, & Krizek, 2010), and school choice (Ford, 2011).
The following model depicts our basic multinomial logit model. Again, the outcome variable was what school the student attended in 201112. The charter school with the highest incoming enrollment served as the reference category, and all other coefficients are presented as relative likelihoods compared to the reference. It is often inferred in a multinomial logit that individuals are placing certain utility in a choice even though it is technically not observed or measured. Thus, we model our analysis as:
Uji = Xiαj + eji
In this particular model, j represents the choice, i denotes the student, and X represents the individual characteristics such as race/ethnicity, IEP status, urbanicity, danger, failure to make AYP, and percent FRL at the previous school.
DESCRIPTIVE ENROLLMENT CHARACTERISTICS OF CYBER CHARTER SCHOOLS
The first step of the analysis was to explore the descriptive data relating to cyber charter schools and consider how they compare to state averages. Table 1 depicts the distributions of the variables used in the study for Pennsylvania in general and then of each cyber charter school. The first column depicts the universe of students in Pennsylvania as a point of comparison. This comparison shows that cyber charter school transfers in general (second column) had slightly higher proportions of Black students, had slightly higher proportions of students from urban and rural locations (but far fewer from suburban locations), and had students from schools with higher proportions of FRL compared to state averages. These data show that descriptive enrollment patterns vary between cyber charter schools.
Next Generation, which was the most popular choice among transfers, had the highest number of Black new enrollees and had much higher rates of new enrollees with an IEP. Next Generation and Independent Charters were the only cyber charter schools in the sample to have above average shares of students with IEPs compared to the state. The group of Independent Charters stood out as serving a higher proportion of Hispanic students compared to other cyber charter schools. The Next Generation and Independent Charters students were also unique in the composition of the students feeder schools. Almost half the students were from urban areas, almost half were from the Philadelphia metropolitan area, and feeder schools for these cyber charter schools tended to have higher rates of FRL students than the other cyber charter schools. Additionally, about 40% of Next Generation and Independent Charters students left feeder schools that failed to make AYP the previous year, and slightly higher shares of these students came from persistently dangerous schools.
Meanwhile, High Tech, Horizons, Innovation, Network, and the IU cyber charter schools attracted a disproportionately lower share of minority students, urban students, and students with IEPs, although the popularity of these schools varied overall.
The findings also show differences among the types of communities that sent students to particular cyber charter schools. For instance, almost half of Next Generation students came from an urban school district, while High Tech and Network drew almost half of their students from suburban districts. Horizons attracted almost half of its students from rural areas (as well as only 15% of its transfers from urban areas).
THE POPULARITY OF HIGHER PERFORMING CYBER CHARTER SCHOOLS
The first research question asks if a cyber charter schools popularity reflects indicators of prior academic performance. Table 1 indicates wide variance among school enrollments for 201112 transfers, including a noticeable absence of clustering into the highest achieving schools. The most popular cyber charter school in the sample, Next Generation, had the lowest prior-year performance rating.
Despite the most popular cyber charter school having the lowest scores, enrollments were not consistent across social demographics and previous school traits. This disparity suggests that preference into higher or lower performing cyber charter schools is not universal. Next Generation, the school with the lowest performance indicators, enrolled the highest number of minority, urban, and IEP students and students from feeder schools with the highest proportions of FRL-eligible students. Meanwhile, High Tech, Horizons, Innovation, Network and the IU cyber charter schools tended to have higher prior-year percent proficient ratings and enrolled higher shares of White students without IEPs. Finally, Horizons Cyber School, which actually had the highest Math and ELA scores, was the second-least popular school. This school had the largest proportion of rural/town enrollees, the fewest urban students from feeder schools with high FRL rates, and among the highest non-minority populations of enrollees.
Together, these descriptive findings begin to raise questions about equity within the cyber charter school enrollment process. The achievement indicators presented in the tables are those that came in the year prior to enrollment, so they do not reflect the performance rating of the newly enrolling students. Part of the differences in prior-year performance ratings may occur as a result of receiving so many transfers from low-performing schools. However, this surfaces a question as to why new transfers do not then seek out the higher performing cyber school in accordance with the rationale that higher achieving options eventually dominate the sector.
ENROLLMENTS, POPULARITY, AND STRATIFICATION
To answer the second research question we used an analytic statistical approach to examine how enrollments into particular cyber charter schools related to the demographics of students and the school environments that they left. For this portion of the analysis, we used a multinomial logit (mlogit) to compare student choices to the reference category of Next Generation since it had the highest enrollment and among the lowest academic performance in the previous years (see Table 2; note the table is sorted left to right, from highest average performance rate to lowest). Coefficients in our mlogit are presented as relative likelihoods (RL) compared to the reference category6, which are similar to odds ratios in magnitude. If the coefficient is 1 or close to 1, it means there is little difference. Numbers closer to 0 indicate less likely predictors. Numbers greater than one indicate an exceedingly more likely scenario.
As mentioned previously, the fact that Next Generation attracted a plurality of students complicates the notion that students would cluster in the highest achieving cyber school. Further, the mlogit suggests a clear pattern where, on average, Next Generation was more likely to attract minority students, students with IEPs, urban students, and students from lower income districts. For instance, when controlling for all model variables, Horizon, the school with the highest performance indicators on the Math and ELA tests, was significantly less likely to attract Hispanic students (RL=0.34; p<.01), students from lower income districts (RL=0.32; p<.05), and students from failing schools compared to Next Generation (RL=0.44; p<.01). The fact that these coefficients are so large in magnitude (close to 0) demonstrates stark differences in student enrollments among the highest achieving and lowest achieving cyber charter schools. Similarly, Innovations, which also performed relatively well on performance metrics, was statistically less likely to enroll Black and Hispanic students, students with IEPs, students from lower income districts, and students from previous schools with poor AYP standings. The size of these predictors was also quite large in magnitude for Innovations.
Note. Standard Errors in Parenthesis; * p<.05; **p<.01; ***p<.001; the number of observations is 4,017. The coefficients are relative likelihoods compared to the reference category, meaning that greater than 1.0 is interpreted as more likely and less than 1.0 is interpreted as less likely. ^AP means average percent on performance outcomes (see Table 1).
There appear to be other contrasting enrollment patterns relating to students from the popular Next Generation school to students enrolling in the other schools. For instance, five of the six schools were significantly less likely to enroll students with IEPs than Next Generation, while the Independent Schools were more likely. In terms of race, five of six schools were statistically less likely to enroll Hispanic students than Next Generation (the independent schools were the only exception). Four out of the six cyber schools were statistically less likely to enroll Black students than Next Generation.
There were other key differences in the previous schools attended by the cyber school enrollees. In addition to the aforementioned Horizons and Innovation, the IUs and Network were statistically less likely to enroll students from lower income districts (with the IUs having the lowest coefficient). Students enrolling in Next Generation were statistically more likely to have been at a school with failing AYP status compared to three of the six schools (the other three were not significantly different). Even though Next Generation was more likely to enroll urban students where many brick-and-mortar charters are located, students transferring to Network were statistically more likely to have attended a brick-and-mortar charter school. Yet, students attending Independent Schools, which were also more likely urban, were the most likely to have formerly attended a brick and mortar charter school.
In sum, the mlogit model reveals that when controlling for all model variables, the lowest achieving cyber charter school was more likely to enroll minority, IEP, and students who came from schools with high levels of FRL that performed poorly on AYP metrics. The cyber charter sector is stratified, and there are equity concerns when considering these enrollments and performance metrics together. It does not appear the pure market is functioning strictly according to achievement criteria as expected by policy.
DISCUSSION AND CONCLUSION
Previous scholarship suggests a misalignment between academic performance indicators and enrollment, citing reasons such as geography and other non-academic factors (e.g., Orfield & Frankenberg, 2013). Our study explores a sample purportedly less influenced by these limitations, yet enrollments still do not consistently align with academic performance indicators and instead cluster based on social and school demographic traits. In a theoretical sense, our study further critiques assumptions that suggest choices will lead to systemic improvements based on academic performance indicators. In our findings, a few of the higher performing schools were among the most popular, but this popularity was not universal. Indeed, the most popular school was the lowest performing. Further, students with non-minority backgrounds from feeder schools with higher achievement were more likely to enroll in the cyber charter school that boasted higher performance, while lower performing cyber charter schools tended to enroll minority students who were from feeder schools with lower achievement scores.
One explanation as to why enrollment patterns may not to reflect quality may be due to a selection bias. Perhaps enrollments cause differences in school performance metrics rather than schooling metrics causing differences in enrollment patterns. This means one concern with our study may be that the academic scores are a function of the students enrolled. We understand and concede that there is no way to show whether this argument is true with our data, but we reiterate that our study used enrollment data from the year after the test scores that we report. This means students would have had the opportunity to respond to achievement metrics and distribute themselves into school settings accordingly. If quality were a pure predictor of enrollment, prior quality metrics would reallocate enrollment distributions differently than what we observed.
Additionally, despite selection bias possibly explaining performance metric trends, when we re-examine these schools based on more recent Pennsylvania Value Added Assessment measures (PVAAS, which are growth-metric data) we still see evidence of similar trends (with some variations in the distribution of growth metrics to the percent proficient scores). These new assessment measures, not used in our technical analysis because they are from different years than the individual-student dataset available to us, provide assurance that regardless of academic indicator used, school-level quality concerns do not drive enrollment redistribution. An additional pattern to note is that even though PDE regularly reports various metrics of performance to citizens across the state for many different years (causing one to expect changes in enrollment), the enrollment demographics of cyber charter schools have remained consistent throughout their history.
Finally, while conversations about what creates poor performance metrics are important, we feel that enrollment clustering based on social demographics is in and of itself a notable finding. This finding alone challenges the idea that choices relate to quality, as it is clear that other forces help drive choices due to the simple fact that stratification persists along demographics and traits of previous schools attended. This finding is not new within conversations related to equity in education, but it does put a new spin on this topic in a new setting. The cyber charter school marketplace is in many ways as pure of a neoliberal, market-based system as possible, devoid of many of the restrictions that scholars say cause stratification in brick-and-mortar settings. Despite the removal of these enrollment restrictions, the cyber charter school sector remains stratified in Pennsylvania.
CYBER CHARTER KNOWLEDGE
As seen with other studies (e.g., CREDO, 2015; Ahn & McEachin, 2017), the metrics in our study show concerning performance patterns of cyber charter schools. However, one element we find that other studies do not capture is that academic performance within the cyber charter sector is not consistent. The largest cyber charter school in our study had the lowest ratings on academic performance measures; meanwhile other cyber charter schools had higher scores. This suggests that discussions about the whole cyber charter school sector may overly generalize trends, meaning that the merits of a given cyber charter school should be considered while discussing these programs together. Additionally, poor performance measures in general should not preclude students from enrolling in schools if they are in situations in which online learning is a necessary solution (medical issues, pregnancy, expulsion, etc.). In addition to these considerations about cyber charter school research, our findings run counter to the aggregate understandings of previous reports (e.g., NEPC, 2016) that suggest cyber enrollments are of disproportionally high-income and non-minority students.
These findings together suggest that claims about cyber charter schools likely depend on the contextboth at the state and school levelat focus. In Pennsylvania, on aggregate, the enrollments tend to reflect the averages of the state, but at the school level they vary considerably. This reflects the same finding about inconsistent distributions of academic indicators across cyber charter schools. Considering these points, as well as considering the demographic patterns within the sector, leaders and policymakers should pay careful attention to choice distribution and equity.
RESEARCH, POLICY, LEADERSHIP, CONSIDERATIONS
Based on the understandings previously outlined in this discussion, a major question that stems from our research and that should guide future research is straightforward: If only some families make choices that are linked to academic performance indicators, then what drives the choices of the families who are choosing the schools that have lower performance? Or even more simply, why are there differences in enrollment across cyber charter schools? Finding the forces that explain enrollments were beyond the focus of our analysis, but a recent special issue in the Peabody Journal of Education may help shed light onto certain patterns driving enrollment differences: Marketing (Olson Beal, Stewart, & Lubienski, 2016). Articles in the special issue show that charter schoolsand the marketing strategies they employshape the populations of these schools (Jabbar, 2016; Wilson & Carlsen, 2016). In addition to marketing, previous literature suggests that families have segregated social networks, which inform and segregate school choices (Bell, 2009; Holme, 2002). Perhaps parents use segregated social networks in cyber charter school enrollment decisions, leading to the stratification we observed. While our study had no way of capturing whether enrollment trends occur thanks to a social network or marketing effect, we suspect that a force unmeasured in our analysis has driven enrollment differences between cyber charter schools and indicators of academic quality.
Although our thoughts about marketing and social networks are conjecture and outside the scope of our study, the patterns we did see provide the impetus for a future research possibility. If future studies do find that issues such as directed marketing campaigns are part of what is at fault, then perhaps there should be policy conversations related to how policymakers regulate the data released to the public as information that is more rationaltelling only specifically of program details and how these are associated directly with student learning. Policymakers also need to create plans about how to diffuse this information across different social networks.
However, there certainly may be other enrollment forces at play, including the learning model of the program and the expectations the programs set forth for their students. These understandings suggest that a direction for research is to begin to create profiles of cyber charter school choosers and understand who selects which school and the rationale for doing so. These studies would seek to understand how choosers obtain information, how they use it, and ultimately why they make their final choice among cyber charter school options.
Beyond adding to academic research knowledge, this study of cyber charter transfers has implications for school leaders in both the traditional public school (TPS) sector and charter sector in Pennsylvania and even nationally. One lesson that TPS school leaders can take from this knowledge is that if they want to retain their students, they may be getting hints on their districts deficiencies based on the cyber charter school enrollments. Our study begins to examine such trends and shows that quality based on outcome metrics is not the only influencer of choice. Traditional public schools may need to consider more than academic achievement when enticing students to stay. This may include embarking on programmatic strategies such as growing their own online programs. Additional research would explore these strategies and identify best practices that school districts can take to ensure students do not leave, especially if their leaving is a choice that does not promote academic growth.
Generally, leaders should pay particular attention to the schools students select. They may find themselves identifying that a student made an academically inadequate choice and seek to influence these students by suggesting they enroll in schools that have higher performance. Cyber charter leaders can also help in this process by intentionally coaching out students whom they know will not succeed in online school settings.
Pennsylvania currently incentivizes continual enrollment growth in the cyber charter school sector because cyber charter schools receive more funding for each student enrolled. While this structure is logical when assuming that popularity leads to improved systemic academic improvement, we show that the most popular cyber charter school had the lowest scores on performance metrics. Therefore, the state should take two steps: First, support research that clarifies enrollment choices, including how they align with academic fit, and second, incentivize choices that encourage academic growth.
The first of these recommendations, support research that clarifies enrollment choices, is essentially a recommendation to continue to improve the knowledge base. This is important because our study only depicts resultant enrollment trends and the inequity of these distributions rather than the process behind enrollment distributions. Greater funding for research would help researchers build profiles of enrolling students to understand their needs, choices, and academic goals. Right now, the knowledge of understanding choices in the cyber charter school sector is limited, but shows inequitable clustering based on enrollment placements.
Additional research should also help add confidence to the findings presented in our study. Our study comes with the limitations that the data are from 2011 and are missing some student-level indicators. With the support of PDE and the legislature, researchers could collect and use more refined student-level data to add confidence to the findings presented here.
With these considerations, the legislature and PDE, in tandem with families and educational leaders, can aim to incentivize choices and help design programs that meet the original goals of the policy: systematic and equitable educational improvement. These incentives and designs may come in many forms and avail themselves to important future conversations. However, the first step is for policymakers to acknowledge the deficiencies of the assumptions made with original policy: while families make choices for a number of reasons that may or may not be entirely rational, evidence suggests that enrollments do not necessarily favor higher achieving schools. The legislature should consider enrollment patterns presented here and support research to help mitigate concerns and move policy forward in a way that acknowledges the flaw in original assumptions.
1. There are, of course, other indicators, especially at the student level, that would have helped advance knowledge on choice decisions such as the level of academic attainment of a students parents. However, we could not include these due to the unavailability of data.
2. In addition to these arguments, we also explored enrollments using ArcGIS software. While some of the smaller cyber charter schools have slightly higher enrollments of students coming from locations nearer to them, the majority of cyber charter schools enroll students from across the state.
3. AYP was used at the time to share knowledge publically about performance, but it is no longer in use.
4. Only 19 schools were deemed persistently dangerous in 201011.
5. It is a sample of all cyber students generally, but it is the full population of all new enrollees.
6. Traditionally these coefficients are labeled risk ratios but we feel the term likelihood is more comprehensible in our reporting. The coefficients depict the likelihood a student with the particular demographic trait is to enroll in one of the six outcome schools compared to the reference school, Next Generation.
Ahn, J. (2011). Policy, technology, and practice in cyber charter schools: Framing the issues. Teachers College Record, 113(1), 126.
Ahn, J., & McEachin, A. (2017). Student enrollment patterns and achievement in Ohios online charter schools. Educational Researcher, 46(1), 44-57.
Bell, C. A. (2009). All choices created equal? The role of choice sets in the selection of schools. Peabody Journal of Education, 84(2), 191208.
Berends, M. (2015). Sociology and school choice: What we know after two decades of charter schools. Annual Review of Sociology, 41, 159180.
Berends, M., Cannata, M., & Goldring, E. B. (2011). School choice and school improvement. Cambridge, MA: Harvard Education Press.
Berends, M., & Zottola, G. (2009). Social perspectives on school choice. In M. Berends, M. G. Springer, D. Ballou, & H. J. Walberg (Eds.). Handbook of research on school choice (pp. 3554). New York, NY: Routledge.
Betts, J. R. (2005). The economic theory of school choice. In Betts J. R. & T. Loveless. (Eds.), Getting choice right (pp. 1439). Washington, DC: Brookings Institution Press.
Billingham, C. M., & Hunt, M. O. (2016). School racial composition and parental choice: New evidence on the preferences of white parents in the United States. Sociology of Education, 89(2), 99117.
Butler, J. S., Carr, D. A., Toma, E. F., & Zimmer, R. (2013). Choice in a world of new school types. Journal of Policy Analysis and Management, 32(4), 785806.
Center on Reinventing Public Education. (2015). The policy framework for online charter schools. Seattle, WA: R. Pazhouh, R. Lake, & L. Miller.
Center for Research on Educational Outcomes (CREDO). (2011). Charter school performance in Pennsylvania. Stanford, CA: CREDO.
Center for Research on Educational Outcomes (CREDO). (2015). Online charter school study. Stanford, CA: CREDO.
Chubb, J. E., & Moe, T. M. (1990). Americas public schools: Choice is a panacea. The Brookings Review, 412.
Cowen, J. (2010). Who chooses, who refuses? Learning more from students who decline private school vouchers. American Journal of Education, 117(1), 124.
DeJarnatt, S. L. (2013). Keep following the money: Financial accountability and governance of cyber charter schools. The Urban Lawyer, 45(4), 915951.
Dickson, L. (2010). Race and gender differences in college major choice. The Annals of the American Academy of Political and Social Science, 627(1), 108124.
Evergreen Education Group. (2014). Keeping pace with K12 digital learning: An annual review of policy and practice, 11th edition. Durango, CO: J. Watson, L. Pape, A. Murin, B. Gemin. & L. Vashaw.
Ford, M. (2011). School exits in the Milwaukee parental choice program: Evidence of a marketplace? Journal of School Choice, 5(2), 182204.
Frankenberg, E., Siegel-Hawley, G., & Wang, J. (2010). Choice without equity: Charter school segregation and the need for civil rights standards. Los Angeles, CA: The Civil Rights Project/Proyecto Derechos Civiles at UCLA; www.civilrightsproject.ucla.edu.
Friedman, M. (1955). The role of government in education. In R. A. Solo (Ed.), Economics and the public interest (pp. 127134). New Brunswick, NJ: Rutgers University Press.
Hanushek, E. A., Kain, J. F., Rivkin, S. G., & Branch, G. F. (2007). Charter school quality and parental decision making with school choice. Journal of Public Economics, 91(5), 823848.
Hardy, D. (2015, April 24). How are charter schools funded in Pennsylvania? Keystone Crossroads [Audio podcast]. Retrieved from http://crossroads.newsworks.org/index.php/local/keystone-crossroads/81092-multiple-choices-how-are-charter-schools-funded
Hasler Waters, L., Barbour, M. K., & Menchaca, M. P. (2014). The nature of online charter schools: Evolution and emerging concerns. Journal of Educational Technology & Society, 17(4): 379389.
Hastings, J. S., & Weinstein, J. M. (2008). Information, school choice, and academic achievement. The Quarterly Journal of Economics, 123(4), 13731414.
Holme, J. J. (2002). Buying homes, buying schools: School choice and the social construction of school quality. Harvard Educational Review, 72(2), 177206.
Huerta, L. A., dEntremont, C., & González, M. (2006). Cyber charter schools: Can accountability keep pace with innovation? Phi Delta Kappan, 88(1), 2330.
Jabbar, H. (2016). Selling schools: Marketing and recruitment strategies in New Orleans. Peabody Journal of Education, 91(1), 423.
Kotok, S., Frankenberg, E., Schafft, K. A., Mann, B. A., & Fuller, E. J. (2015). School choice, racial segregation, and poverty concentration: Evidence from Pennsylvania charter school transfers. Educational Policy, 31(4), 415447.
Marsh, R. M., Carr-Chellman, A. A., & Sockman, B. R. (2009). Why parents choose cybercharter schools. TechTrends, 53(4), 3236.
Moe, T. M., & Chubb, J. E. (2009). Liberating learning: Technology, politics, and the future of American education. New York, NY: John Wiley and Sons.
National Education Policy Center (NEPC). (2013). Virtual schools in the US 2013: Full-time virtual schools: Enrollment, student characteristics, and performance. Bolder, CO: G. Miron, B. Horvitz, & C. Gulosino.
National Education Policy Center (NEPC). (2016). Virtual schools report 2016: Directory and performance review. Boulder, CO: G. Miron & C. Gulosino.
Olson Beal, H., Stewart, M., & Lubienski, C. (2016). Marketing and public education evidence, emerging trends, and implications. Peabody Journal of Education, 91(1), 13.
Orfield, G., & Frankenberg, E. (2013). Educational delusions?: Why choice can deepen inequality and how to make schools fair. Berkeley, CA: University of California Press.
Pennsylvania Department of Education. (2004). Charter schools. Retrieved from http://www.education.pa.gov/Documents/Codes%20and%20Regulations/Basic%20Education%20Circulars/Purdons%20Statutes/Charter%20Schools.pdf
Pennsylvania Department of Education. (n.d.). PA.gov: Charter Schools. Retrieved from http://www.education.pa.gov/K-12/Charter Schools/Pages/default.aspx#.VwUinuIrKUk
Pennsylvania Department of the Auditor General. (2012). Charter and cyber charter education funding reform should save taxpayers $365 million annually. Harrisburg, PA: J. Wagner.
Rothwell, J. (2012). Housing costs, zoning, and access to high-scoring schools. Washington, DC: The Brookings Institution.
Stein, M. L. (2015). Public school choice and racial sorting: An examination of charter schools in Indianapolis. American Journal of Education, 121(4), 597627.
Teske, P., & Schneider, M. (2001). What research can tell policymakers about school choice. Journal of Policy Analysis and Management, 20(4), 609631.
Turner, C. (2017, February 4). Betsy DeVos' graduation rate mistake. National Public Radio. Retrieved March 20, 2017. Retrieved from https://www.npr.org/sections/ed/2017/02/04/513220220/betsy-devos-graduation-rate-mistake
Wilson, E. J., Marshall, J., Wilson, R., & Krizek, K. J. (2010). By foot, bus or car: Children's school travel and school choice policy. Environment and Planning A, 42(9), 21682185.
Wilson, T. S., & Carlsen, R. L. (2016). School marketing as a sorting mechanism: A critical discourse analysis of charter school websites. Peabody Journal of Education, 91(1), 2446.