Draft: Please Do Not Cite Without Permission Schooling Decisions and Student Achievement Outcomes under Interdistrict Open Enrollment Deven Carlson University of Oklahoma Lesley Lavery Macalaster College Tyler Hughes University of Oklahoma September 2013 Abstract Motivated by a simple model of cross-district choice, this paper presents empirical analyses of three dimensions of Colorado’s statewide mandatory interdistrict open enrollment policy—the characteristics of participants, the schooling decisions that participants make through the program, and the relationship between interdistrict open enrollment participation and student achievement outcomes. Our analyses indicate that open enrollment participants disproportionately come from disadvantaged districts, although the students who participate are among the more advantaged within those districts. The analyses further suggest that students consider both quality and demographic composition of districts when making open enrollment decisions, although there is heterogeneity along racial/ethnic lines in some of the measures of these dimensions. Finally, we find open enrollment participation to have little effect on achievement outcomes for the full sample of participants, although there is some evidence that it has a negative effect on the achievement of Black students. We discuss the implications of these results for research and policy. 1 1. Introduction Virtually nonexistent only 25 years ago, today interdistrict open enrollment programs operate in a large majority of states and millions of students use these programs annually to attend schools located outside their district of residence. Despite the broad reach of these policies, we know little about their operations and effects, especially relative to more visible school choice policies such as private school vouchers, charter schools, or even magnet schools. There has been a small amount of student-level research on the characteristics of open enrollment participants (e.g, Phillips, Hausman, and Larson 2012; Lavery and Carlson 2013) and the effects of open enrollment policies on student achievement outcomes (e.g., Bifulco, Cobb, and Bell 2009), but the only analyses of schooling decisions made by participants in these programs rely on district-level data and can thus provide only limited insight into the topic (Reback 2008; Welsh, Skidmore, and Statz 2010; Carlson, Lavery, and Witte 2011). Motivated by a simple model of cross-district choice, this paper draws on individual-level data from the universe of students attending Colorado public schools between 2005-06 and 200910 to present empirical analyses of three dimensions of Colorado’s statewide mandatory interdistrict open enrollment policy—the characteristics of participants, the schooling decisions that participants make through the program, and the relationship between interdistrict open enrollment participation and student achievement outcomes. The results of our analysis reveal that a broad range of factors—district achievement, socioeconomic composition, distance, and others—play a role in the schooling decisions that families make under interdistrict open enrollment policies. Some of these factors relate to schooling decisions in ways that are largely predictable, but other factors enter into the decision calculus in less expected ways. Our results also reveal substantial heterogeneity in the determinants of schooling decisions across demographic and socioeconomic subgroups. Finally, we find open enrollment participation to 2 have little effect on achievement outcomes for the full sample of participants, although there is some evidence that it has a negative effect on the achievement of Black students. Taken as a whole, our analysis provides a broad perspective on the operation of interdistrict open enrollment policies. We proceed by first providing an overview of interdistrict open enrollment policies—devoting particular attention to the Colorado context—before presenting the conceptual framework that underlies our analysis. We then describe the data we use in our analyses, detail the specific analytical approach we take, and present the results of our analysis. We close with a discussion of the implications of our results for both research and policy. 2. Interdistrict Open Enrollment: Background and Context Interdistrict open enrollment programs take two primary forms—voluntary and mandatory. Voluntary policies permit school districts to decide whether or not to accept student transfers from other districts. Mandatory policies, in contrast, require school districts to accept transfers from other districts, although the policies typically specify a set of conditions under which districts can legally refuse to accept transfers; we discuss the most common of these in greater detail below. Both voluntary and mandatory policies generally prohibit districts from restricting student transfers out of the district. Although interdistrict open enrollment programs were among the first school choice policies to appear on the educational landscape, they are still relatively recent additions. The earliest voluntary programs only began to emerge in the early 1980s and the first mandatory statewide program did not exist until the implementation of Minnesota’s policy in 1991 (Boyd, Hare, and Nathan 2002). Enactment of that program, however, commenced a wave of state adoptions and by 2011 only 15 states and the District of Columbia were without some form of 3 interdistrict open enrollment (National Center for Education Statistics 2012). Table 1 presents the number of states with voluntary and mandatory interdistrict open enrollment policies.1 [Insert Table 1 about here] Our empirical analyses below rely on data from Colorado, which operates a mandatory statewide program. Consequently, we discuss this class of policy—both generally and in the specific context of Colorado—in further detail. Although the specifics of mandatory interdistrict open enrollment policies vary across states, there are three foundational features of these programs. First, the policies specify a process that allows students to attend public schools located outside their district of residence. Historically, public school students have been required—almost without exception—to attend the school specified by the district in which they reside. Second, although by definition mandatory policies require districts to accept transfers from other districts, they also typically specify a set of conditions under which school districts can legally refuse to accept interdistrict transfers. The list of allowable conditions for transfer refusal varies across states, but a lack of capacity and an applicant’s history of behavioral problems—such as a record of suspension, expulsion, or substance abuse—are the two most common conditions specified by state policies. Third, interdistrict open enrollment policies generally specify that per-pupil state aid follows program participants; funds are disbursed to the district of attendance rather than the district of residence. The precise amount of funding accompanying a student transferring across district lines varies by state, but it is almost certainly greater than the marginal cost of serving an additional student (Reback 2008). 1 Table 1 makes clear that two states have both voluntary and mandatory interdistrict open enrollment policies. In most of these cases, the mandatory policies require districts to accept transfers with a specific characteristic (e.g. low test scores, a learning disability, etc.) while acceptance of students without the specified characteristic(s) is voluntary. 4 In addition to these three foundational aspects of mandatory interdistrict choice policies, two other features of these transfer programs—transportation and desegregation policies— warrant brief discussion. Transportation is a major challenge faced by states implementing an interdistrict choice policy; there is no obvious solution for transporting program participants from their district of residence to the district they choose to attend. Over time, states have addressed issues of transportation in three primary ways. One set of states places all transportation responsibilities upon the parents of transferring students while a second group of states mandate that the district of residence provide all necessary transportation. Finally, a third collection of states do not address the issue of transportation at all in their open enrollment policies, thus leaving the issue to be sorted out by parents, the district of residence, and the district of attendance. In addition to variance in the responsibility for providing transportation, state policies also differ in the amount of funding provided to support the transportation of interdistrict transfers. Policies range from providing no transportation funding at all to fully reimbursing districts for the costs associated with busing interdistrict transfers.2 Finally, many states’ interdistrict transfer policies have historically permitted districts to refuse transfers—both into and out of the district—if the transfer would either violate the provisions of an established desegregation policy or upset the racial or socioeconomic balance of the district. The legality of such provisions, however, is in doubt after the recent U.S. Supreme Court decisions in Parents Involved in Community Schools Inc. v. Seattle School District and Meredith v. Jefferson County (Ky.) Board of Education, which prohibited schools and districts from considering race in school admissions processes. The Education Commission of the States maintains a database that describes several features of each state’s interdistrict open enrollment policy. The database can be found at: http://ecs.force.com/mbdata/mbtab4ne?sid=a0i70000000Xk5v&rep=OET 2 5 2.1. Colorado’s Statewide Mandatory Interdistrict Open Enrollment Program Colorado’s mandatory interdistrict transfer program is authorized by The Public Schools of Choice Act of 1990 (C.R.S. 22-36-101). Beginning with the 1994-95 school year, this legislation allowed students to attend any public school in the state without paying any tuition to the district in which the school was located. Like most mandatory interdistrict transfer policies, however, there are five conditions under which Colorado districts can legally refuse to accept a transfer application: A lack of space or teaching staff required to serve the student; The district or school is not equipped—either physically or with respect to curriculum— to serve the student; The student does not meet established eligibility criteria for participation in a requested program; Admission of the student would violate the terms of an established desegregation plan; The student has been expelled from another district. There is one more notable provision of Colorado’s interdistrict choice policy concerning student admission. Specifically, the policy states that if the number of transfer applications received by a district exceeds the number of available seats, the district is urged—but not required—to give enrollment priority to applicants who attend a low-performing school and achieve a proficiency level of unsatisfactory in one or more academic subjects according to the Colorado Student Assessment Program. With respect to funding, Colorado’s policy mirrors most programs nationally by disbursing state aid associated with a transferring student to the district of attendance. Finally, issues of transportation are not addressed in the relevant statutes. Colorado’s interdistrict open enrollment program quickly grew to serve a significant number of students. By the 2000-01 school year—only six years after the inception of the program—over 20,000 students were using the policy to attend a school located outside their 6 district of residence.3 Over the following decade the program tripled in size and in recent years it has served in excess of 68,000 students. Table 2 presents the number of students attending a school located outside their district of residence. For purposes of comparison it also presents the total K-12 enrollment in public schools in Colorado as well as the number of students enrolled in the state’s charter schools. The table indicates that about 3.2 percent of students attended a school located outside their district of residence during 2000-01 school year while approximately 8.1 percent of students did so in the 2011-12 school year. The corresponding numbers for charter school enrollment are 2.9 and 9.1 percent, respectively. [Insert Table 2 about here] 2.2. Previous Research Relative to the amount of scholarship on other school choice policies such as charter schools and private school vouchers, the research literature on interdistrict open enrollment is fairly limited. In the years after the initial wave of states adopted interdistrict choice policies— Minnesota, Massachusetts, Wisconsin, Ohio, and others—a small number of studies provided descriptive information about these programs. In the context of Massachusetts, both Fossey (1994) and Armor and Peiser (1998) used simple mean comparisons to examine differences between districts that were net senders versus net receivers of transferring students. Both studies found net receiving districts to be more advantaged than net senders on several measures; median family income, adult education, achievement scores, dropout rates, and per-pupil expenditures. In Ohio, Fowler (1996) surveyed district superintendents and asked about the factors that influenced their decision to participate (or not) in the state’s voluntary interdistrict transfer program. Participating superintendents cited the desire to increase enrollments and funding as the Data on the number of students utilizing Colorado’s interdistrict open enrollment program are not available prior to the 2000-01 school year. 3 7 primary driver of their decision. Superintendents who opted out of the program generally reported a lack of space as motivation for denying transfers. There was little scholarly analysis of interdistrict transfer policies in the decade or so following these early studies, but there has been renewed interest in these programs in recent years. This recent research can usefully be grouped into two main categories. The first set of studies analyzes the effects of interdistrict transfer programs on outcomes of interest, including student achievement, racial and socioeconomic segregation, and home values. Only two studies examine the effect of interdistrict open enrollment programs on student achievement, but both report positive results. Bifulco, Cobb, and Bell (2009) examine the effect of Connecticut’s interdistrict magnet transfer programs and find it to have a positive effect on the achievement scores of participants. In the context of Wisconsin, Welsch and Zimmer (2012) find positive systemic effects of the state’s cross-district transfer program; districts that experience large student losses exhibited improved test scores in the subsequent year. Studies of the stratifying effects of interdistrict choice programs reach more varied conclusions. The Institute of Metropolitan Opportunity (2013) found Minnesota’s cross-district transfer program to significantly increase racial stratification across districts while Powers, Topper, and Silver (2012) found Arizona’s program to have no such effects. Carlson’s (2013) analysis concludes that Colorado’s program produces a decrease in racial segregation across districts, but a slight increase in socioeconomic stratification. Considered together, these studies suggest that the segregating effects of interdistrict transfer programs are likely context-specific.4 The second group of studies focuses on the operations of interdistrict transfer programs, including characteristics of program participants and the correlates of cross-district transfer flows 4 Brunner, Cho, and Reback (2012) examine—both theoretically and empirically—the effect of interdistrict open enrollment on residential location decisions and housing values. They find that school districts located near highquality schooling options exhibit significant increases in both housing values and population density. 8 or schooling decisions made by program participants. Two analyses have used student-level data from Colorado to examine the characteristics of program participants. These studies provide evidence that participants are disproportionately non-white and come from socioeconomically advantaged contexts (Lavery and Carlson 2013; Holme and Richards 2009). Furthermore, students with special designations—such as gifted and talented, English language learner, or IEP—are much less likely to open enroll than their peers without those designations. Powers, Topper, and Silver (2012) report similar findings from an analysis of 27 metropolitan Phoenix school districts. Several studies have used district-level data to examine the characteristics associated with interdistrict transfer flows. These studies have been conducted in the contexts of Minnesota (Reback 2008), Wisconsin (Welsch, Statz, and Skidmore 2010), and Colorado (Holme and Richards 2009; Carlson, Lavery and Witte 2011). The analyses vary in their details, but they all provide evidence that—at least to some degree—transfer flows are correlated not only with district-level achievement scores, but also the demographic makeup, financial characteristics, and structural considerations, such as distance. Although these studies provide valuable evidence into the schooling decisions made by interdistrict open enrollment participants, the inferences that can be drawn from the studies are limited by their reliance on district-level data. Below we detail our approach to assessing whether similar conclusions emerge from analysis of five years of student-level data on the universe of students attending Colorado public schools. 3. A Simple Model of Interdistrict Choice Drawing upon previous theoretical work (e.g., Lankford and Wyckoff 1992; Lankford, Lee, and Wyckoff 1995; Deming, Hastings, Kane, and Staiger 2011) we present a simple model of interdistrict choice as motivation for our empirical analysis. The model begins with families 9 distributed across school districts and all children assigned to attend a school located in—and operated by—the district in which they reside. The presence of an interdistrict choice program, however, allows families the option of enrolling their child in a district other than the one in which they reside. Under this scenario, each household assesses the utility it would receive from each district schooling option, where Uik, k = 1, 2,…K, is the utility that household i would receive from enrolling its child in school district k. The set of K schooling options includes households’ district of residence, r. After evaluating the utility expected by each schooling option, the household elects to enroll its child in the district that maximizes utility. The household will participate in the interdistrict open enrollment program if Uir < Uik for any k ≠ r (1) We specify household utility in this model as a function of district quality, the demographic composition of the district, travel and search costs, and—following Deming et al. (2011)—an unobserved “fit” between households and each district alternative. More formally, Uik = U(Qk, Dik, Cik, λik) (2) where Q represents observable measures of quality—such as average standardized test scores— that are broadly valued by households, D represents the socioeconomic and demographic makeup of a district that each household may value differently, C represents the travel and search costs associated with attending a given school district, and λ is the unobserved fit noted earlier. Following Deming et al. (2011) we assume λ is distributed normally with mean zero and variance σ2. We further assume λ is independent of Q and also independent across district options for a given household. We specify the household utility function as linear and assume that each component enters it additively: Uik = Qk + Dik - Cik + λik (3) 10 Because we assume λ ~ N(0, σ2), it is the case that E(Uik) = Qk + Dik - Cik and we can rewrite (1) as Qr + Dir - Cir + λir < Qk + Dik - Cik for any k ≠ r (4) Normalizing C such that the cost of attending the district of residence is zero produces Qr + Dir + λir < Qk + Dik - Cik for any k ≠ r (5) which can be rearranged to produce λir < (Qk - Qr) + (Dik - Dir) - Cik for any k ≠ r (6) Under this model a family will open enroll out of their district of residence if there is any district where the combination of quality and demographic gains—less travel and search costs— provide more utility than the unobserved “fit” with their district of residence. The inclusion of λ in the model provides a justification, other than travel and search costs, for families that do not open enroll out of an observably poor district. However, the model still predicts that families are increasingly likely to open enroll as Qr and Dir decline, and as Qk and Dik increase. Finally, if there are multiple nonresident districts that satisfy (6), the utility maximizing nature of households leads them to select the district with the largest net gain. 4. Data Our empirical analysis draws on a dataset constructed from student-level records maintained by the Colorado Department of Education (CDE) supplemented with information from the U.S. Census Bureau. Beginning with the 2005-06 school year and extending through 2009-10, the CDE records contain annual information—enrollment status, demographic and socioeconomic characteristics, achievement outcomes, and school and district measures—on the universe of students attending public schools in Colorado. 11 The enrollment information included in the CDE records include a unique student identifier as well as annual records of the school each student attends and the district in which it is located.5 In addition to recording the district of attendance, the data also indicate whether a student attended a school operated by a district other than the one in which they reside—a measure of open enrollment. For open enrollers, the data also record an identifier for the students’ district of residence. Along with the individual-level information, the dataset also contains relevant contextual data on the schools and districts that students attend, such as dropout rates, mobility statistics, disciplinary information, staff data, available postsecondary options, fiscal information, and socioeconomic composition. For students who open enroll, this information is also available for students’ district of residence.6 Our data contain standard demographic characteristics—age, grade, gender, race/ethnicity—as well as measures of several other characteristics such as gifted and talented status, free or reduced lunch status, disability status, English language learner status, a measure of language proficiency, and students’ primary language. The CDE records also contain student achievement results. Specifically, the data record students’ scale scores on the reading and math portions of the Colorado Student Assessment Program (CSAP), which was administered to all students in grades 3-8 and 10 to meet the accountability provisions of the No Child Left Behind Act. We standardized the CSAP scale scores using the statewide mean and standard deviation for the proper year, grade, and subject. Taken as a whole, the dataset we created from the CDE 5 More specifically, the data contain a record for each school attended by a student during a given school year. The fact that the data contain multiple observations for students who attended more than one school in a given year represents a potential complication for student-level analyses. To address this issue, we implemented the following decision rule. First, for students with test scores, we kept the record containing the school in which the student was tested. This eliminated approximately half of the duplicate records. For the remaining students with multiple records—those without test scores—we kept the record in which the disposition code listing the reason that a student left a school was not applicable; in effect, we kept the student record for the school in which a student finished the year. 6 The data do not contain a record of the “school of residence” for open enrolling students. Consequently, we do not possess information about the schools out of which students are open enrolling. 12 records contains nearly 4.3 million observations from approximately 1.25 million unique students. We have extensive information on each student’s demographic and achievement profile, their open enrollment status, and data on the schools that students attend and the districts in which they reside. As the final step in creating our dataset, we supplemented the student-level CDE records with geographic data on school districts maintained by the U.S. Census Bureau. Specifically, we used ARCGIS software to calculate the geographic centroid of each school district and then, for each student-year observation, merged the longitude and latitude coordinates of the district of attendance into the dataset. For students who open enroll, we also merged the latitude and longitude coordinates for the district of residence. These geographic coordinates play an important role in our analyses, as we describe in greater detail below. 5. Empirical Analysis Our model of interdistrict choice presents testable propositions for several dimensions of open enrollment policies, including the characteristics of participants, the schooling decisions participants will make through the program, and—more indirectly—the educational outcomes that participants might exhibit. Here we explicitly identify the testable propositions within each of these dimensions and describe our approach to testing each of them empirically. 5.1. Characteristics of Participants The model presented above provides two main insights into the characteristics of interdistrict open enrollment participants. First, it holds that individuals with a good “match” with their district of residence will be less likely to open enroll, all else equal. Empirically, we might expect that families who have children with certain characteristics and educational designations—such as gifted and talented or special educational needs—select their district of 13 residence with careful consideration of how it “matches” their child on such dimensions. If accurate, students with these characteristics would be less likely to open enroll. Similarly, it is reasonable to expect that—all else equal—students with low achievement levels might perceive a poor “match” with their district of residence and thus be more likely to open enroll. Second, the model makes clear that open enrollment becomes increasingly likely as Qr and Dir decline. That is, a student will be more likely to open enroll as the quality of his district of residence declines or as its demographic profile becomes less desirable, a perception that may vary by student. We begin to assess whether these theoretical predictions are corroborated empirically by first presenting descriptive characteristics on the characteristics of interdistrict open enrollment participants. The top panel of Table 3 presents individual-level characteristics of participants for each of the five years in our dataset while the bottom panel presents average characteristics of participants’ district of residence for each year over this time period. In both cases the characteristics of nonparticipants are presented to provide a point of comparison. Overall, the statistics presented in Table 3 are largely consistent with the expectations laid out above. Specifically, Table 3 demonstrates that students with the designations noted previously—special education, gifted and talented, and ELL—all open enroll at disproportionately low rates. Moreover, the district-level characteristics are also generally in line with expectations. Open enrollers reside in districts that fare worse on several common measures of school quality— dropout rates, truancy rates, Advanced Placement (AP) course offerings, and average district achievement, at least in some years—relative to their non-open enrolling peers. Demographically, open enrollers reside in districts with higher levels of students eligible for free or reduced-price lunch, relative to non-open enrollers. The evidence regarding the relationship 14 between student achievement and open enrolling is more mixed; students who open enroll have slightly lower math scores, but somewhat higher reading scores. [Insert Table 3 about here] To further explore this issue, we estimate a simple logistic regression model that predicts interdistrict open enrollment participation during the 2006-07 school year as a function of student characteristics as well as the characteristics of a student’s district of residence and can be written as: Pr(𝑂𝑖𝑑2006 = 1) = 𝑙𝑜𝑔𝑖𝑡 −1 (𝜹𝑺𝒊𝟐𝟎𝟎𝟓 + 𝜽𝑫𝒅𝟐𝟎𝟎𝟓 ) (7) where the probability that student i residing in district d open enrolls in 2006-07 is a function of a vector of student and family background characteristics, S, and a vector of characteristics of the student’s district of residence, D; logit-1 (x) = ex/(1+ex). The variables contained in S and D mirror those presented in Table 3. Results from this model, which we estimate separately for kindergarteners, 6th graders, and 9th graders, are presented in Table 4 and are mostly consistent with the bivariate results presented in Table 3. Specifically, Table 4 again demonstrates that students with special designations are less likely to open enroll and students who reside in districts with lower levels of quality on the measures noted above are more likely to open enroll, even conditional on other factors. Finally, the mixed findings regarding the relationship between achievement and open enrollment participation that were observed in Table 3 are also present in the multivariate results. [Insert Table 4 about here] 5.2. Schooling Decisions In addition to insights about the characteristics of participants, the theoretical model also presents testable propositions regarding the schooling choices made by open enrollment participants. Specifically, the model holds that the probability of open enrolling into a given 15 district is increasing in both the quality of that district and the desirability of the demographic profile—represented by Qk and Dik, respectively—while decreasing in travel and search costs, Cik. As the starting point for assessing these predictions we used our data to construct, uniquely for each kindergartener observed open enrolling in our data, a choice set containing all districts located within 100 miles of the student’s district of residence.7 Then, using this dataset containing the district choice sets of all 16,747 open enrolling kindergarteners, we estimated the following conditional logistic regression model of the district each student chooses to attend: Pr(𝑆𝑖𝑘 = 1) = 𝑙𝑜𝑔𝑖𝑡 −1 (𝜹𝑸𝒌 + 𝜽𝑫𝒌 + 𝜷𝑪𝒊𝒌 + 𝜸𝒊 ) (8) where, S represents school districts in an individual’s choice set, i and k index individuals and school districts, respectively; Q represents a vector of observable school quality measures, such as average district achievement, the truancy and dropout rates, and the number of AP courses offered; D represents a vector of observable district demographic characteristics, including the racial/ethnic composition of the district and the percentage of students eligible for free or reduced-price lunch; C represents the costs of open enrolling, operationalized as the distance from an individual’s district of residence, and 𝜸 is a student fixed effect. To construct each student’s choice set we first created a dataset containing every pairwise combination of open enrolling kindergarteners and Colorado school districts. Then, using the school district centroids in our data, we used the Great Circle Distance Formula to calculate the distance between each student’s district of residence and every other school district in Colorado. The Great Circle Distance Formula is calculated as follows: distance = 3963.0 * arccos[sin(lat1/57.2958) * sin(lat2/57.2958) + cos(lat1/57.2958) * cos(lat2/57.2958) * cos(lon2/57.2958 lon1/57.2958)]. Upon calculation of the distances between a student’s district of residence and all other Colorado school districts we deleted those observations containing districts that were more than 100 miles from the student’s district of residence. We also deleted observations containing students’ districts of residence. The 100 mile cutoff may seem excessive upon first glance, but two main reasons underlie our belief that it is not. First, the distances are measured from the geographic center of one district to the geographic center of another. It is clearly not the case that all students live at the geographic centers of school districts, and we wanted our cutoff to account for this reality. Second, the cutoff of 100 miles was located at approximately the 98th percentile of observed distances. Specifications using cutoffs of 120 miles and 75 miles were also tested and the results were substantively similar and are available from the authors. Finally, we exclude any observed flows that exceeded 100 miles from the dataset. These represented a very small proportion of cases and are most likely attributable to enrollment in virtual schools. Virtual schools are undoubtedly an important topic of study, but they are beyond the scope of this analysis; we are only interested in interdistrict transfers where students physically attend a district other than the one in which they reside. 7 16 Table 5 presents the results from this model and although there are some unanticipated outcomes, the results are largely consistent with the theoretical expectations developed above. Most notably, the coefficient on average district achievement—the most common, if imperfect, measure of educational quality—is positive and significant, as is the coefficient on the number of AP courses offered by the district. Additionally, the coefficient on the district truancy rate is negative, but not statistically significant. Taken together, these results support the proposition that the probability of enrolling into a given district increases as its quality rises. The results in Table 5 also lend empirical support to the proposition that individuals become less likely to open enroll into a district as the cost of doing so increases, as evidenced by the negative and highly significant coefficient on the distance measure. Finally, the results indicate that individuals became less likely to open enroll into a given district as the percentage of students eligible for free or reduced-price lunch increased, but more likely to do so as the percentage of students who were Black or Hispanic increased. [Insert Table 5 about here] Although the results in Table 5 clarify the direction and statistical significance of the relationships between district characteristics and schooling decisions, the substantive magnitudes of the relationships are somewhat more difficult to discern. Consequently, we present Figure 1, which plots the predicted probability of selecting a given district across the observed range of each of the nine variables in the model when all other variables are held at their mean. Figure 1 makes clear that travel cost—as measured by distance—exerts the strongest substantive effect on district choices, with a predicted probability of nearly 0.4 for districts that are very close to a student’s district of residence all the way down to approximately 0.01 for districts that are 100 17 miles away.8 The plots also indicate that enrollment is a substantively meaningful determinant of schooling choices. All else equal, students are much more likely to open enroll into a large district than a small one. Although the data do not permit any firm inferences regarding the mechanisms responsible for this result, we speculate that enrollment into large districts may be at least partially driven by the availability—or expected availability—of particular educational programs or opportunities that the family perceives will be a good match for their child. This speculation could be examined empirically in future research. [Insert Figure 1 about here] Moving on to the measures of educational quality and district demographics, the plots in Figure 1 reveal that, while statistically significant, the relationships between schooling decisions and these measures are weaker, substantively speaking, than the distance and enrollment variables. For most of these measures, including the percentage of students eligible for free or reduced-price lunch and the variables measuring the racial or ethnic composition of the district, the change in the predicted probability of a student enrolling into a district at the observed minimum versus enrolling into a district at the observed maximum is in the range of approximately 0.05 to 0.10, depending upon the specific quality or demographic measure. The single unexpected result in Figure 1 is the positive relationship between the district dropout rate and open enrollment decisions, a result potentially attributable to the large inflows into Denver. 5.2.1. Heterogeneity by Socioeconomic Status and Race/Ethnicity Recent research on families’ schooling preferences suggests that, all else equal, families prefer schools where the student body matches their demographic and socioeconomic profile (e.g., Hastings, Kane, Staiger 2008). Our model of interdistrict choice recognizes this potential for preference heterogeneity by allowing the definition of a desirable district demographic profile 8 This result lends further support to our choice of the 100 mile cutoff. 18 to vary by student. Empirically, we assess the extent to which such heterogeneity is present by estimating equation (8) separately for White students, Black students, and Hispanic students as well as for students who are and are not eligible for free or reduced-price lunch. [Insert Table 6 about here] The results disaggregated by race/ethnicity are presented in the first three columns of Table 6 and the plots of the accompanying predicted probabilities—analogous to those in Figure 1—are presented in Figure 2. The results indicate that students in all three racial/ethnic groups we analyze become less likely to open enroll in a given district as the proportion of students eligible for free or reduced-price lunch in that district increases; the preference for attending socioeconomically advantaged districts appears to be common across racial/ethnic groups. However, the relationships between schooling decisions and the racial/ethnic composition of school districts differ across the three racial/ethnic groups we analyze, a finding consistent with previous research. For example, Figure 2 demonstrates that both White and Hispanic students become increasingly likely to open enroll into a district as the proportion of Hispanic students in the district increases. For Black students, however, the relationship is negative and significant. There is a positive relationship between the district into which a student open enrolls and the percentage of Black students for all three groups, but the relationship is substantively stronger for Black and White students relative to Hispanic students. There is also heterogeneity across these three groups along the quality and cost dimensions. Perhaps most notably, the relationships between schooling decisions and the average standardized test scores in a district vary across the three racial/ethnic groups; the relationship is strong and positive for White students but negative and significant for Black students; there is no statistically discernible relationship for Hispanic students. With respect to 19 distance, Figure 2 indicates that the relationship between distance and open enrollment decisions is stronger for White and Hispanic students than for Black students. Similarly, the relationship between district enrollment and the likelihood of open enrolling into a given district is weaker for Black students than it is for their White or Hispanic peers. [Insert Figure 2 about here] The fourth and fifth columns of Table 6 present the respective results from separate estimation of equation (8) over students who are not eligible for free or reduced-price lunch and those who are eligible to received subsidized meals. The accompanying predicted probabilities for these groups are presented in Figure 3. These results identify some measures of district demographics for which the two groups exhibit substantively similar results. For example, both groups become less likely to open enroll into a given district as the proportion of students eligible for free or reduced-price lunch increases. On other demographic measures, however, the two groups diverge in their results. Specifically, relatively to their eligible peers, students who are not eligible for subsidized lunches become increasingly likely to enroll into a given district as the proportion of minority students—both Black and Hispanic—increases. Figure 3 also demonstrates differential results for the two groups on some of the district quality measures, such as average district standardized test score or the truancy rate. For both measures, socioeconomically advantaged students become more likely to enroll into a given district as its quality increases; the corresponding relationships for students who are less socioeconomically advantaged are either negative or null. [Insert Figure 3 about here] 20 5.3. Student Achievement Outcomes Our model of interdistrict choice produces clear predictions about the characteristics of participants and the schooling decisions they will make, but it only indirectly addresses the issue of how interdistrict open enrollment participation might affect student educational outcomes. Specifically, the model addresses this issue to the extent that it incorporates district quality into the schooling decision process. The empirical analyses presented provide evidence that families do incorporate district quality as a consideration in their schooling decisions under interdistrict open enrollment. Furthermore, the analyses disaggregated by demographic and socioeconomic subgroups suggest that quality considerations play a larger role in schooling decisions for some subgroups than for others. Considered together, these findings provide a basis for expecting that interdistrict open enrollment could improve the achievement outcomes of participants. At the same time, however, both our model and empirical analyses make clear that school quality is not the sole determinant of families’ schooling choices—demographics and convenience play a large role. Furthermore, there are features of open enrollment participation that could have a negative effect on student achievement and other educational outcomes. Notable among these is the fact that open enrollment participation typically requires students to switch schools, and there is a large body of scholarship demonstrating that switching schools often has a disruptive effect on academic achievement, at least initially (Alexander, Entwisle, and Dauber 1996; Hanushek, Kain, and Rivkin 2004; Ingersoll, Scamman, and Eckerling 1989; Kerbow, Azcoitia, and Buell 2003; Lash and Kirkpatrick 1990, 1994; Rumberger et al. 1999; South, Haynie, and Bose 2007; Temple and Reynolds 1999; Xu, Hannaway, and D’Souza 2009; Zimmer et al. 2009; Engberg et al. 2012). Thus, the overall effect of open enrollment participation on student educational outcomes is a question of whether the potential positive aspects of participation—such as increased school quality or an improved educational “match” 21 with a student’s interests or needs—are greater than the relevant negative features, such as the penalty for switching schools. Given the theoretical ambiguity, we estimate a series of models to gain empirical insight into the matter. We analyze the relationship between student achievement and interdistrict open enrollment participation using two techniques that are well-suited to the panel structure of our data. We first specify a series of models containing a lagged outcome measure on the right-hand side—we take a value-added approach. Such an approach has been used in the analysis of other school choice programs, including charter schools (e.g., Hanushek et al. 2007; Sass 2006), private school vouchers (e.g., Witte et al. 2011; Carlson, Cowen, and Fleming 2013), and even open enrollment programs (Bifulco, Cobb, and Bell 2009). The primary appeal of the valueadded approach is that it represents a straightforward way to control for the accumulation of factors and experiences—both observed and unobserved—that may affect current achievement; it possesses the implicit ability to control for factors that evolve over time. The model we estimate can be written as: 𝑌𝑖𝑟𝑡 = 𝑌𝑖𝑟𝑡−1 + 𝜃𝑋𝑖𝑟𝑡 + 𝛿𝑂𝑖𝑟𝑡 + 𝜓 + 𝜀𝑖𝑟𝑡 (9) where student achievement, Y, for student i in district of residence r at time t is a function of lagged achievement, a vector of observed demographic characteristics X, an indicator of interdistrict open enrollment participation O, a grade-by-year-by-district fixed effect 𝜓, and an error term capturing additional unmeasured influences 𝜀.9 We estimate the model via OLS separately for reading and math achievement. The results from estimation of this model are presented in the bottom panel of Table 7. For purposes of comparison, the top two panels of Table 7 present results from estimation of 9 The district component of the grade-by-year-district fixed effect is the district of residence, not the district of attendance. 22 slight variants of the model presented in equation 9; the top panel presents results from a model containing just a measure of lagged achievement while the middle panel presents results from a model containing lagged achievement and the vector of demographic characteristics. Overall, the results are substantively similar across the three specifications and they suggest that open enrollment participation has little substantive effect on student achievement in either subject. Although all estimates of the relationship between open enrollment participation and math achievement are negative and statistically significant, the magnitudes of these estimates are only approximately 0.01 standard deviations. In reading, the estimates are positive and significant, but similarly small from a substantive standpoint. [Insert Table 7 about here] To assess whether these results are robust to an alternative analytical approach we estimate a second model containing a student fixed effect. This approach removes threats of bias from all time-invariant characteristics—observed or unobserved—and the resulting estimates from this model should be interpreted as the effect of interdistrict open enrollment participation on a student’s level of achievement. This contrasts with the estimates produced from the valueadded model depicted in equation (9), which represent the effect of interdistrict open enrollment participation on student achievement growth. The student fixed effects model we estimate can be written as: 𝑌𝑖𝑟𝑡 = 𝛼𝑖 + 𝜃𝑋𝑖𝑟𝑡 + 𝛿𝑂𝑖𝑟𝑡 + 𝜀𝑖𝑟𝑡 (10) where 𝛼 represents the student fixed effect and the remaining contents of the model were described previously. Results from estimation of equation (10) are presented in the middle panel of Table 8. For comparison purposes, the top panel of the table presents the results of the model containing just the student fixed effects and the open enrollment indicator—there are no time- 23 varying demographics. The results from these two model variants are virtually identical and indicate that open enrollment participation decreases students’ math and reading achievement by approximately 0.06 and 0.03 standard deviations, respectively. [Insert Table 8 about here] In each subject, the student fixed effects estimates of the relationship between student achievement and open enrollment participation are 0.03 to 0.04 standard deviations lower than the corresponding value-added estimates. To gain insight into whether these discrepancies are attributable to different analytic samples—the value-added model includes a lagged outcome measure and thus excludes a year of observations—or the different techniques utilized, we estimated equation (10) over the value-added analytic sample. These results are presented in the bottom panel of Table 8. In math, the results are substantively similar to the full-sample results, but in reading they are noticeably more negative. Together, these results suggest that the divergent results between the value-added and student fixed effects models are attributable to the different estimation techniques, rather than different analytic samples. The difference in the results across the two analytic approaches is consistent with open enrolling being an endogenous decision. Although the evidence is not definitive, the results are consistent with families choosing to open enroll after a particularly poor year of academic performance. Such a scenario could result in open enrollment participation being associated with lower achievement levels, but not lower growth in achievement, which is exactly the pattern we observe in Tables 7 and 8. 5.3.1. Heterogeneity in Achievement Outcomes by Race/Ethnicity and Socioeconomic Status Our earlier analysis revealed significant heterogeneity in the determinants of schooling choices across demographic and socioeconomic subgroups. These differences in the determinants of schooling choices suggests the potential for heterogeneity in the outcomes that 24 can be influenced by those choices. Consequently, we estimate equations (9) and (10) separately for the racial/ethnic and socioeconomic subgroups we examine previously. The results separated by race/ethnicity are presented in Table 9 and they demonstrate heterogeneity in outcomes. Whether we examine the value-added results of (9) or the student fixed effects results of (10), we see that the White and Hispanic results are similar to both the full-sample results and to each other. Specifically, the value-added results demonstrate interdistrict open enrollment participation has little substantive effect on student achievement growth while the student fixed effects results suggest that participation has a negative effect on the level of student achievement, particularly in math. Results for Black students, however, do not follow the same pattern. Both the value-added and fixed effects results indicate a negative relationship between open enrollment participation and student achievement. The magnitude of these results are in the range of 0.04 to 0.08 standard deviations. As we discuss in greater detail in the following section, these results generally mirror the patterns observed in our analysis of the determinants of schooling decisions. In that analysis, White and Hispanic students exhibited substantively similar results, but the results for Black students differed on measures of both demographics and quality. The results in Table 9 demonstrate that these differences carry over into the analysis of the relationship between open enrollment participation and student achievement outcomes. [Insert Table 9 about here] Table 10 presents the results of estimating equations (9) and (10) separately for students who are and are not eligible for free or reduced-price lunch. In contrast to the results disaggregated by race/ethnicity, the results broken down by subsidized lunch eligibility reveal little heterogeneity. For both groups the value-added results indicate little substantive effect of interdistrict open enrollment participation on student achievement outcomes while the fixed 25 effect results demonstrate a slightly more negative relationship. Detailed results can be found in Table 10 below. [Insert Table 10 about here] 6. Discussion and Conclusion Motivated by a simple model of cross-district choice, this paper presents empirical analyses of three dimensions of Colorado’s statewide mandatory interdistrict open enrollment policy—the characteristics of participants, the schooling decisions that participants make through the program, and the relationship between interdistrict open enrollment participation and individual student achievement outcomes. Considered as a whole, the analyses comprise a multi-part narrative that provides significant insight into the operations and effects of this largescale—if oft-overlooked—school choice policy, and the results have several implications for both research and policy. Our first line of inquiry concerned the characteristics of individuals who elected to participate in interdistrict open enrollment. Our model generated two insights on this topic. First, it predicted that individuals who were a good “match” with their district of residence would be less likely to open enroll. We conjectured that parents with children who have specific educational designations—such as gifted and talented or special educational needs—would place a disproportionate amount of attention on ensuring that their district of residence provides a good match on these dimensions and such students would thus be less likely to open enroll. Second, our model holds that the likelihood of open enrollment increases as the quality of the district of residence decreases. Both of these insights are generally corroborated empirically. Students with special designations are significantly less likely to open enroll than their non-designated peers. Additionally, Tables 3 and 4 demonstrate that, on average, open enrollers reside in 26 districts that perform worse on several measures of quality, including dropout rates, truancy rates, and number of AP course offerings, relative to non-open enrolling students. Interestingly, the average standardized test score of open enrollers’ district of residence are not appreciably different—and perhaps even somewhat better—than the average district scores for students who remain in their district of residence. Examination of the individual-level characteristics indicates that—conditional on the characteristics of their district of residence—students who open enroll are disproportionately advantaged from a socioeconomic standpoint. Upon gaining an understanding of the characteristics of individuals who participate in Colorado’s interdistrict open enrollment program, we turned to analyzing the schooling decisions participants make under the program. Our model holds that district quality and demographic composition will both enter into the decision process and the empirical analyses largely bear these predictions out, although the results indicate heterogeneity along racial/ethnic and socioeconomic lines for some of the quality and demographic measures. Specifically, our results provide suggestive evidence that, relative to White and Hispanic students, Black students may place a larger emphasis on demographic composition than traditional measures of quality— particularly average student achievement—when making schooling decisions through Colorado’s interdistrict open enrollment program. The insights drawn from our analysis of the determinants of schooling decisions inform our analyses of the relationship between interdistrict open enrollment participation and student achievement. Specifically, we recognize that the heterogeneity in the determinants of schooling decisions made under interdistrict open enrollment may also manifest itself in achievement outcomes. And indeed, the results indicate that interdistrict open enrollment participation has little substantive effect on the achievement of the full sample, but there is evidence of a modest 27 negative relationship between program participation and achievement outcomes for Black students. Such a finding is consistent with the results of our analysis of the determinants of schooling decisions described above. We employ both value-added and student fixed effects approaches in our analysis of the relationship between open enrollment participation and the two techniques return somewhat inconsistent results. Almost without exception, the student fixed effects results are more negative than the value-added results. Although such a discrepancy initially seems problematic, it can actually provide substantive insight into the operations of the program. Specifically, this pattern of results is consistent with a scenario where families decide to open enroll after a particularly poor year, academically speaking, but do not exhibit significant improvements in achievement following enrollment in a nonresident district. Of course, the results are only suggestive of the scenario described above and do not provide definitive evidence on the score. Consequently, a more in-depth, and perhaps qualitative, analysis of the conditions that influence interdistrict open enrollment decisions is a natural topic for future research. In addition, there are several other areas where the results presented above are suggestive of a specific scenario, but do not provide sufficient evidence to be fully confident in drawing specific conclusions and would thus benefit from additional inquiry. Specific examples include the sources of heterogeneity in schooling decisions and a more detailed analysis of how heterogeneity in schooling decisions relates to heterogeneity in schooling outcomes. Finally, future research would do well to assess whether different policy contexts produce substantively similar results. Along with the research implications that emerged from our analysis, the results also have several implications for policy. At a basic level, our analysis provides evidence that open 28 enrollment participation is unlikely to meaningfully increase student achievement, at least in the short term. At the same time, it also seems likely that the policy increases at least some participants’ satisfaction with their schooling option, an outcome with inherent value. Finally, the results of the analysis of the determinants of schooling decisions suggests that policymakers may wish to assess whether all families have easy access to information relevant to their schooling decisions. Similarly, administrators of Colorado’s interdistrict open enrollment program may consider examining the extent to which there are supply-side constraints on the decisions that families can make under the policy, and assess whether specific groups (e.g., students with disabilities, English language learners, etc) experience more constraints than other groups. Overall, this analysis provides us with a better understanding of the operations and effects of the nation’s largest school choice program—interdistrict open enrollment—but it is clear that there is still much to be learned. 29 7. References Alexander, K. L., Entwisle, D. R., & Dauber, S. L. (1996). Children in Motion: School Transfers and Elementary School Performance. Journal of Educational Research, 90 (1), 3-12. Armor, D. J., & Peiser, B. M. (1998). Interdistrict Choice in Massachusetts. In P. E. Peterson, & B. C. Hassel (Eds.), Learning from School Choice. (pp. 157-186). Washington D.C: Brookings Institution Press. Bifulco, R., Cobb, C. D., & Bell, C. (2009). Can interdistrict choice boost student achievement? The case of Connecticut’s interdistrict magnet school program. Educational Evaluation and Policy Analysis, 31(4), 323-345. Boyd, W.L., Hare, D., and Nathan, J. 2002. “What Really Happened?” In Hubert H (Ed.) Minnesota’s experience with statewide public school choice programs: Center for School Change. Humphrey Institute of Public Affairs, University of Minnesota; Minneapolis, MN: 2002. Carlson, D. (2013). School Choice and Educational Stratification. Working Paper. Carlson, D., Lavery, L., and Witte, J. (2011). The Determinants of Interdistrict Open Enrollment Flows. Educational Evaluation and Policy Analysis, 33(1), 76-94. Colorado Revised Statute. 22-36-101 Deming, D., Hastings, J., Kane, T., & Staiger, D. (2011). School choice, school quality and academic achievement. NBER Working Paper, 17438. Engberg, J., Gill, B., Zamarro, G., and Zimmer, R. (2012). Closing Schools in a Shrinking District: Do Student Outcomes Depend on Which Schools are Closed? Journal of Urban Economics, 71 (2): 189-203. Fossey, R. (1994). Open Enrollment In Massachusetts: Why Families Choose. Educational Evaluation And Policy Analysis, 16(3), 320. Fowler, F. C. (1996). Participation In Ohio's Interditrict Open Enrollment Option: Exploring The Supply-Side Of Choice. Educational Policy, 10(4), 518-536. Hanushek, E. A., Kain, J. F., & Rivkin, S. G. (2004). Disruption versus Tiebout Improvement: The Costs and Benefits of Switching Schools. Journal of Public Economics, 88 (9-10), 17211746. Hastings, J. S., Kane, T., & Staiger, D. (2008). Heterogeneous preferences and the efficacy of public school choice. NBER Working Paper, 2145. Holme, J. J., & Richards, M. P., (2009). School Choice and Stratification In Regional Context: Examining The Role Of Inter-District Choice. Peabody Journal Of Education, 84, 150-171. 30 Ingersoll, G. M., Scamman, J. P., & Eckerling, W. D. (1989). Geographic Mobility and Student Achievement in an Urban Setting. Educational Evaluation and Policy Analysis, 11 (2), 143-149. Institute of Metropolitan Opportunity. (2013). Open Enrollment and Racial Segregation in the Twin Cities: 2000-2010. University of Minnesota Law School: Minneapolis, MN. Retrieved September 3, 2013 from http://www.law.umn.edu/uploads/30/c7/30c7d1fd89a6b132c81b36b37a79e9e1/OpenEnrollment-and-Racial-Segregation-Final.pdf Kerbow, D., Azcoitia, C., & Buell, B. (2003). Student Mobility and Local School Improvement in Chicago. Journal of Negro Education, 72 (1), 158-164. Lankford, H., & Wyckoff, J. (1992). Primary and secondary school choice among public and religious alternatives. Economics of Education review, 11(4), 317-337. Lankford, R. H., Lee, E. S., & Wyckoff, J. H. (1995). An analysis of elementary and secondary school choice. Journal of Urban Economics, 38(2), 236-251. Lash, A. A., & Kirkpatrick, S. L. (1990). A Classroom Perspective on Student Mobility. The Elementary School Journal, 91 (2), 176-191. Lash, A. A., & Kirkpatrick, S. L. (1994). Interrupted Lessons: Teacher Views of Transfer Student Education. American Educational Research Journal, 31 (4), 813-843. Lavery, L., & Carlson, D. (2013). Dynamic Participation in Interdistrict Open Enrollment. Working Paper. National Center for Education Statistics (2012). State Education Reforms-Table 4.2. Numbers and types of state open enrollment policies, by state: 2011. Retrieved August 30, 2013 from http://nces.ed.gov/programs/statereform/tab4_2.asp Phillips, K. J., Hausman, C., & Larsen, E. S. (2012). Students Who Choose And The Schools They Leave: Examining Participation in Intradistrict Transfers.The Sociological Quarterly, 53(2), 264-294. Powers, J. M., Topper, A. M., & Silver, M. (2012). Public School Choice and Student Mobility in Metropolitan Phoenix. Journal of School Choice, 6(2), 209-234. Reback, R. (2008). Teaching To The Rating: School Accountability And The Distribution Of Student Achievement. Journal Of Public Economics 92(5-6): 1394–1415. Rumberger, R. W., Larson, K. A., Ream, R. K., & Palardy, G. J. (1999). The Educational Consequences of Mobility for California Students and Schools. Berkeley, CA: Policy Analysis for California Education. South, S. J., Haynie, D. L., & Bose, S. (2007). Student Mobility and School Dropout. Social Science Research, 36 (1), 68-94. 31 Temple, J. A., & Reynolds, A. J. (1999). School Mobility and Achievement: Longitudinal Findings from an Urban Cohort. Journal of School Psychology, 37 (4), 355-377. Xu, Z., Hannaway, J., & D'Souza, S. (2009). Student Transience in North Carolina: The Effect of School Mobility on Student Outcomes Using Longitudinal Data. Washington, DC: Urban Institute. Welsch, D.M., Statz, B., & Skidmore, M. (2010). An Examination Of Inter-District Public School Transfers In Wisconsin. Economics of Education Review, 29(1): 126-137. Welsch, D. M., & Zimmer, D. M. (2012). Do student migrations affect school performance? Evidence from Wisconsin's inter-district public school program. Economics of Education Review, 31(1), 195-207. Zimmer, R., Gill, B., Booker, K., Lavertu, S., Sass, T. R., & Witte, J. (2009). Charter Schools in Eight States: Effects on Achievement, Attainment, Integration, and Competition. Santa Monica, CA: RAND. 32 Tables and Figures Table 1. Number of states with interdistrict open enrollment programs, by program type Program type Number of states Any interdistrict open enrollment program 35 Voluntary only 21 Mandatory only 12 Both voluntary and mandatory 2 33 Table 2. Total enrollment in Colorado, interdistrict open enrollment, and charter schools, by year Year Total (K-12) OE Charter 2000-01 724,508 22,993 21,064 2001-02 742,145 23,979 24,658 2002-03 751,862 30,846 28,782 2003-04 757,668 35,752 31,529 2004-05 766,657 38,780 36,658 2005-06 780,708 42,278 44,254 2006-07 794,026 48,543 52,242 2007-08 802,639 51,430 56,772 2008-09 818,443 57,274 57,843 2009-10 832,368 60,916 66,556 2010-11 843,316 66,296 72,989 2011-12 854,265 68,829 77,853 Source: Colorado Department of Education 34 Table 3. Average characteristics, by open enrollment status and year 2005-06 Characteristic OE Not OE 2006-07 OE 2007-08 Not OE OE 2008-09 Not OE 2009-10 OE Not OE OE Not OE Individual Characteristics Male 49.5 51.5 49.5 51.5 49.2 51.6 48.8 51.5 49.0 51.6 White 64.5 62.1 63.2 61.2 63.4 60.7 63.4 60.2 63.0 60.0 Hispanic 23.9 27.4 24.4 28.1 24.1 28.5 24.6 28.9 24.8 29.0 Black 7.1 6.0 8.0 6.1 8.0 6.1 7.3 6.1 7.4 6.0 Asian 3.2 3.3 3.3 3.4 3.3 3.5 3.5 3.6 3.5 3.8 Native American 1.2 1.2 1.1 1.2 1.1 1.2 1.1 1.2 1.3 1.3 Free Lunch 19.0 25.9 19.0 26.4 19.7 27.6 21.9 28.4 23.3 30.6 Reduced Lunch 6.0 5.8 6.4 5.9 7.2 6.3 7.6 6.8 7.4 6.8 No Lunch 74.7 68.2 73.8 67.3 72.8 65.9 66.8 63.1 69.3 62.6 Gifted and Talented 7.3 9.9 4.8 7.0 4.7 7.2 4.8 7.3 4.8 7.5 Special education 6.7 12.3 6.4 12.2 6.4 12.5 7.1 12.7 6.8 13.0 English language learner 8.1 11.1 6.2 9.7 6.5 9.5 6.2 9.4 6.2 9.4 Standardized math score 0.01 0.00 -0.03 0.00 -0.01 0.00 -0.04 0.00 -0.02 0.00 Standardized reading score 0.08 0.00 0.04 0.00 0.06 0.00 0.05 0.00 0.06 0.00 ACT Composite 18.6 18.9 18.7 19.0 19.0 19.4 19.5 19.6 18.8 19.4 Characteristics of District of Residence Percent Free or Reduced Lunch 41.0 33.9 42.4 34.8 43.6 35.8 47 38.7 48.2 40.3 Percent White 54.8 61.8 53.3 61.2 53.3 60.7 53.1 60.5 50.1 56.7 Average District Achievement Reading -0.04 -0.01 -0.08 -0.01 -0.08 -0.01 0.01 -0.01 0.01 -0.01 Number of AP Courses Offered 10.6 11.7 14.6 17 15.6 18.1 15.0 18.1 15.6 18.4 District Enrollment 27,941 33,462 28,918 33,676 28,386 34,099 28,796 34,893 29,915 35,712 District Dropout Rate 5.2 4.3 5.2 4.3 4.2 3.6 4.1 3.5 3.6 3.0 Pupil/Teacher Ratio 17.1 17.4 17 17.3 16.9 17.3 17.0 17.3 17.6 17.9 District Truancy Rate 2.4 2.1 2.4 2.0 2.5 2.2 2.3 2.0 2.3 2.1 35 Table 4. Log Odds from logit model predicting open enrollment in 200607, by grade Characteristic KG 6th Grade 9th Grade Student Characteristics Female 1.0312 1.0485 1.0879** (0.0363) (0.0433) (0.0385) Hispanic 0.6291*** 0.8391*** 0.8793** (0.0368) (0.0514) (0.0445) Black 1.2173*** 1.1640* 0.9639 (0.0831) (0.1006) (0.0714) 1.4146*** 0.9098 0.7925** (0.1251) (0.1081) (0.0936) 0.7084** 0.7298 0.9495 Asian Native American Reduced-price lunch Not eligible for free/reduce lunch Lunch eligibility missing Gifted and talented Limited English proficiency Disability Math achievement (0.1235) (0.1651) (0.1453) 1.9178*** 1.4304*** 1.1064 (0.1670) (0.1255) (0.0911) 2.3288*** 1.8528*** 1.7081*** (0.1338) (0.1156) (0.0866) 6.7580*** 2.9959*** 1.0000 (0.9964) (1.0165) Omitted 0.6692 0.8571** 0.7761*** (0.1879) (0.0665) (0.0487) 0.3101*** 0.6277*** 0.6934*** (0.0267) (0.0686) (0.0629) 0.4479*** 0.8749* 0.8165*** (0.0428) NA (0.0657) 0.8700*** (0.0508) NA NA 1.0985** (0.0310) Reading achievement NA (0.0418) District Characteristics (Lagged) Percent eligible free/reduced lunch Percent White Average district achievementReading District made AYP 1.0012 1.0201*** 1.0209*** (0.0023) (0.0028) (0.0025) 0.9805*** 0.9788*** 0.9828*** (0.0028) (0.0027) (0.0024) 1.4655 26.3220*** 6.3312*** (0.3895) (5.3888) (1.1728) 1.4182*** 1.5148*** 2.1245*** (0.1280) (0.1418) (0.1786) 36 Table 4. Log Odds from logit model predicting open enrollment in 200607, by grade Characteristic Number of AP courses offered Enrollment (thousands) Dropout rate Student-teacher ratio Truancy rate Constant Observations KG 1.0014 6th Grade 9th Grade 0.9660*** 0.9749*** (0.0049) (0.0027) (0.0025) 0.8946*** 0.9115*** 0.9820** (0.0143) (0.0089) (0.0078) 1.0655*** 1.0998*** 1.0634*** (0.0109) (0.0125) (0.0101) 0.9448*** 0.9985 1.0525*** (0.0162) (0.0198) (0.0192) 1.0018 1.1845*** 1.1042*** (0.0233) (0.0331) (0.0284) 0.2996*** 0.0515*** 0.0182*** (0.1059) (0.0227) (0.0073) 64792 54875 57389 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 37 Table 5. Results of conditional logistic regression model predicting district into which student open enrolls Variable Coefficient Dropout Rate 0.050* (0.003) Truancy Rate -0.012 (0.007) Percent Free or Reduced Lunch -0.020* (0.001) Percent Black 0.040* (0.002) Percent Hispanic 0.018* (0.001) Number AP Courses Offered 0.026* (0.002) Enrollment (10,000s) 0.0224* (0.001) Average Reading Achievement 0.450* (0.086) Distance (10s) N Number of groups Log-Likelihood -0.470* (0.004) 1,210,022 16,747 -49,431 38 39 Table 6. Results of conditional logistic regression model predicting district into which student open enrolls, by subgroup Free or Variable White Black Hispanic No Lunch Reduced Lunch Dropout Rate 0.068* 0.003 0.041* 0.046* 0.051* (0.004) (0.013) (0.007) (0.004) (0.006) Truancy Rate -0.051* (0.010) 0.100* (0.016) 0.063* (0.013) -0.065* (0.010) 0.077* (0.009) Percent Free or Reduced Lunch -0.015* (0.001) -0.043* (0.004) -0.015* (0.002) -0.023* (0.001) -0.007* (0.002) Percent Black 0.051* (0.002) 0.064* (0.006) 0.017* (0.004) 0.056* (0.002) 0.009* (0.003) Percent Hispanic 0.025* (0.001) -0.022* (0.003) 0.018* (0.002) 0.023* (0.001) 0.004* (0.002) Number AP Courses Offered 0.014* (0.002) 0.049* (0.007) 0.034* (0.004) 0.025* (0.002) 0.038* (0.003) District Enrollment (10,000s) 0.250* (0.006) 0.199* (0.023) 0.207* (0.012) 0.223* (0.006) 0.208* (0.010) Reading Achievement 1.726* (0.096) -3.404* (0.143) 0.109 (0.207) 0.823* (0.106) -0.206 (0.157) -0.492* (0.005) 764,310 10,496 -31709.08 -0.401* (0.014) 119,865 1,478 -3868.85 -0.438* (0.009) 261,333 3,437 -10397.8 -0.474* (0.005) 863,733 11,654 -35141.28 -0.478* (0.008) 323,848 4,352 -12932.54 Distance (10s) N Number of groups Log-Likelihood 40 41 42 Table 7. Results of OLS regression of standardized reading and math achievement on open enrollment participation: Models contain lagged outcome measure in addition to variables noted in headers Variable Math Scores Reading Scores No demographics Open Enrollment -0.007* 0.021* (0.003) (0.003) Demographic characteristics Open Enrollment -0.011* 0.007* (0.003) (0.003) Demographic characteristics & grade-by-districtby year fixed effect Open Enrollment -0.011* 0.009* (0.003) (0.003) 43 Table 8. Results of OLS regression of standardized reading and math achievement on open enrollment participation: Model contains student fixed effects Variable Math Scores Reading Scores No demographics—Full sample Open Enrollment -0.063* -0.033* (0.003) (0.003) Open Enrollment Time-varying demographics—Full sample -0.057* -0.030* (0.003) (0.003) Open Enrollment Time-varying demographics—Value-added sample -0.061* -0.082* (0.004) (0.004) 44 Table 9. Results of OLS regressions of standardized reading and math achievement on open enrollment participation, by racial/ethnic subgroup and model type Variable Math Scores Reading Scores White Value-added Student fixed effects -0.008* 0.006 (0.002) (0.004) -0.061* -0.043* (0.005) (0.005) Black Value-added Student fixed effects -0.084* -0.041* (0.011) (0.014) -0.084* -0.056* (0.013) (0.017) Hispanic Value-added Student fixed effects -0.011 0.013* (0.006) (0.007) -0.057* -0.034* (0.007) (0.008) Note: Robust standard errors in parentheses below coefficients. Value-added models contain lagged outcome measure, demographic characteristics, and gradeby-district-by-year fixed effects. Student fixed effects models contain timevarying demographics and are estimated over value-added analytic sample. 45 Table 10. Results of OLS regressions of standardized reading and math achievement on open enrollment participation, by socioeconomic subgroup and model type Variable Math Scores Reading Scores Eligible for Free or Reduced-price Lunch Value-added Student fixed effects -0.022* 0.007 (0.005) (0.006) -0.058* -0.030* (0.007) (0.007) Ineligible for Free or Reduced-price Lunch Value-added Student fixed effects -0.004 0.007 (0.004) (0.004) -0.046* -0.027* (0.005) (0.005) Note: Robust standard errors in parentheses below coefficients. Value-added models contain lagged outcome measure, demographic characteristics, and grade-bydistrict-by-year fixed effects. Student fixed effects models contain time-varying demographics and are estimated over value-added analytic sample. 46