BergmanMcFarlin school choice


1 ∗ Education for All? A Nationwide Audit Study of Schools of Choice Peter Jr. Bergman Isaac McFarlin y Universit Teachers College, Columbia University of Florida and NBER December , 2018 Abstract: School choice may allow schools to “cream skim” students perceived as easier to educat e. To test this, w e sent emails from fictitious parents to 6,452 schools in 29 states . The fictitious parent asked whether any student is eligible to apply and Washington, D.C to the school . Each email signaled a randomly assigned attribute of the and how to apply We that schools are less likely to respond to inquiries from students with poor child. find low achievement, or a behavior, eed. Lower response rates to students with a special n potentially significant special need are driven by charter schools. Otherwise, t hese results hold for traditional public schools in areas of school choice and high -value added schools . JEL Codes: I20, I21, I24, I28. ∗ The authors are co- leaders in the production of research presented herein, and their names are listed alphabetically. This research is supported by the University of Michigan Office of Research, Center for the Education of Women, Center for Public Policy in Diverse Societies; the Russell Sage Foundation and the thank Joe Altonji, David Card, Sarah Walton Family Foundation. We , Julie Cullen, John DiNardo, Cohodes Will Dobbie, Maria Ferreya, Caroline Hoxby, Larry Kotlikoff, Dick Murnane, Jonah Rockoff , Rich Romano, David Sappington, and Chris Walters for their detailed feedback. We also thank seminar participants at the Columbia University, Harvard University, University of California -Davis, University of Princeton University, Chicago, University of Florida, University of Michigan and the Tinbergen Institute as well as conference partici pants at APPAM, CESifo, IZA, NBER Labor Studies , Stanford GSE , and University of Wisconsin IRP . This study was approved by the University of Michigan Institutional Review Board (HUM00080890), Teachers College, Columbia University Institutional Review Board (15- 118), and by the University of Florida Institutional Review Board (IRB201702513).

2 1. Introduction school choice can lead to improvements in school access and productivity (cf. In theory, Friedman, 1962). , however, argue it enables schools to “cream skim” the easiest - Critics students perceived as harde r to educate to lower - to-educate students, which constrains . This segregation is concerning qua lity schools because school quality affects college Chetty (Deming, 2011; Angrist et al., 2013; enrollment, earnings, health, and criminality et al., 2011 Dobbie and Fryer, 2015) . To minimize ; Chetty et al., 2014; Wong et al., 2014; control admissions and , regulators use lotteries and common applications to this practice provide financial offsets to educate students with special needs ven if schools cannot . But e admit in the choice process may still let them influence who control whom they , frictions . Families often lack information about schools’ eligibility requirements applies , quality, and es (DeArmond et al., 2014; Hastings and Weinstein, 2008; Kapor admission process These frictions et al., 2017). the possibility that schools of choice manipulat e the raise applicant pool by provid ing less application information to the parents of children as more difficult to educate. perceived This is difficult to test. Differences in how families choose schools may reflect preferences rather than schools steering away applicants heterogeneous . Observational have yield ed mixed evidence about how have focused on specific contexts and studies (Lacireno- Paquet et al. , 2002; choice affects the distribution of students across schools 2009; Zimmer et al., 2009; Bifulco et al., Hoxby and Murarka, 2009; Zimmer and Guarino, 1 whether The difficulty in determining ; Walters, 2018). Nichols et al., 2015 -Barrer 2013; 1 Several studies also look at voucher systems (Epple et al., 2017) . Altonji et al. (2015) find little evidence that vouchers negatively that impact students remain in traditional public schools. Outside the United States, Hsieh and Urquiola (2006) show who the voucher (2015) use system in Chile caused high income students to move to private schools. Muralidharan and Sundararaman -achieving and high- a two- stage experiment in India to estimate whether th e introduction of private -school vouchers negatively impacts non -voucher outcomes. on student recipients; they find no evidence of spillovers 1

3 school controversy regarding the potential s’ dissuade certain students from applying fuels 2 (Cohodes and Dynarski, 2016). expansion of school choice ent to test whether schools give To overcome this difficulty, we conducted an experim 3 We sent to lower -performing or costlier -to-educate students . less application information emails from fictitious parents to 6,452 charter and traditional public schools of choice across 29 states and the District of Columb ia. The parent asked whether any student is 4 Each email signaled one of the following eligible to apply to and how to apply . the school randomly -assigned attribute s about the student : their disability status, poor behavior, high prior academic achievement, or no indicati on of these characteristics. We also or low randomly varied students’ implied race , household structure, and gender. students We find that schools respond less often to messages regarding whom schools may perceive as challenging to educate. The baseline response rate is 53 percent . more a s has a potentially restrictive special need are 5 Messages signaling that tudent less likely to receive a response than the percentage points baseline message. Messages signaling messages indicating low prior achievement are a behavior problem and 2 7 and percentage points less likely to receive a response, respectively . A message indicating good grades and attendance, however, is neither more nor less likely to receive a response than 2 -Picking Students” in the New There is also controversy in popular press on access to schools of choice. See “Are Charter Schools Cherry (December. 10, 2014), which features a debate by policymakers on charter school access. In an article in the Washington York Times , “The Masquerade of School Choice: A Parent’s Story” (April 1, 2017), a parent describes her exp erience with racial discrimination Post and school choice. 3 Our research design is correspondence or audit stud y. Prior research using this design includes Bertrand and Mullainathan (2004) to study racial discrimination in labor markets. Ayres and Siegelman (1995) investigate racial and gender discrimination in bargaining for the value of a credential from a for -profit postsecondary a car. In education settings, Darolia et al. (2015) and Deming et al. (2016) examine institution while Baker et al. (2018) study bias in online learning environments. Giulietti et al. (2017) examine racial disc rimination in the provision of public services in the United States. Inv estigating the sharing economy, Edelman et al. (2017) find evidence of racial bias in the online market for housing rentals. 4 Understanding eligibility requirements is the most commonly -cited barrier to selecting a school of choice (DeArmond et al, 2014). 2

4 the baseline message. these traditional public schools in A key question is whether results differ between choice and charter schools . Charter schools represent the fastest growing areas with school 5 ntry. To explore this question, our sample includes charter form of school choice in the cou schools matched to nearby traditional public schools of choice with the same entry grade level. We find that , overall, traditional public schools ’ response rates are similar to the response rates from school s across treatment messages. However, t here is a charter different response rate to messages that signal a child has a significant special need. Traditional public schools exhibit no differential response rate to these messag es, but charter schools are 7 percentage points less likely to respond to the m than to the baseline message . This result is important because students with disabilities are twice as expensive to educate than the typical student without a disability ( Moore et al. ; Chambers, , 1988 Collins and Zirkel, , and students with the severe disabilities can cost 8 -to-14 1998; 1992) -disabled student (Griffith, 2008). times to educate compared to the typical non These results hold for high -value -added schools, including urban, high- value added charter schools . Prior research has also shown that “no -excuses” charter schools also tend to have high value added ( Abdulkadiroglu et al., 2011; Angrist et al., 2013, 2016; Chabrier et al., 2016; Clark et al., 2015; Dobbie and Fryer, 2013 ; Dobbie and Fr yer, 2015; Hoxby and Murarka, 2009). We identified 272 such schools in our data; these schools have a value -added —one- half standard deviation above other charter schools. No -excuses charter schools are significantly less likely (10 percentage points) to respond to inquiries that signal a child’s potentially significant disability than to the baseline message. We present evidence that the monetary cost of serving students matters. States fund 5 See the National Alliance for Public Charter Schools’ report . 3

5 students with special needs in several different ways, including designated block grants , cost -reimbursements , and formulae that for special education for services rendered 6 to schools . funds (Griffith, 2008; Millard and Aragon, 2015) provide additional general , such as Wisconsin and Michigan , that reimburse We fin d that charter schools in states large share of the realized cost of serving special -needs students exhibit no for a districts rate to messages differential response a potentially high -cost special need. signaling Allowing charter schools to integrate with another Local Education Agency (LEA) 7 legal The -cost students across multiple schools . spread risk of serving higher could the provid e services to students with special needs falls on the LEA. Integration obligation to between a charter school and a traditional -public school LEA could enable the schools to 8 pool resources ’ LEA polic ies . . We coded each charter school’s LEA status based on states We find that s LEA status does not moderate the differential response rate to message disability. signaling a also find differences in response rates by the implied rac e of the family We but not by household structure . Schools may interpret these attributes as signal s of families’ socio- economic status . Overall, schools are 2 percentage points less likely to respond to emails signed by Hispanic -sounding names than to other messages. There is weaker evidence of a differential response rate for messages signed by Black -sounding names. These differences are largest in schools serving predominantly white students. implications for interpreting Our results have important use evidence studies that 6 Per -student education costs have steadily risen over time. Average expenditure per pupil for the 1990- 1991 school year was $9,936. In 2016 dollars, this increased to $13,119 in the 2014- 2015 school year (NCES Table 236.69). About 20 percent of the growth i n new education spending is directed to special education services (Hanushek and Rivkin, 1997). 7 An LEA is equivalent to a school district in most instances. 8 ool (e.g. the state or a district). LEA status varies within states because it can depend on what entity authorized the charter sch 4

6 from s’ admission lotteries to examine school effectiveness. Our findings do not school -based studies, but they underscore that lottery - undermine the internal validity of lottery 9 Hastings et al. (2006), . based studies are conditional on the set of students who apply lters (2014) and Kline and Walters (2016) find that students who may benefit the Wa most in terms of academic achievement are also the least likely to apply to high - performing schools or education programs. However, certain students –even those who may benefit most— may also impose high costs or negative behavioral spillovers (cf. Carrell and Hoekstra, 2010; Carrell et al., 2018) that reduce schools’ demand to serve to-educate students were evenly distributed across schools, these students. If the hardest- the im pacts of highly -effective schools might decrease due to negative behavioral or fiscal spillovers. This article makes three primary contributions to the literature on school choice. We provide the first experimental evidence testing whether schools of choice provide less application information to students whom schools may perceive as harder to educate. Second, we incorporate signals of several student attributes— beyond race and gender — across a wide the external variety and a large number of schools . These features raise validity of our study and allow us to investigate important dimensions of heterogeneity, traditional public schools versus charter schools or high - comparisons between such as low- value added schools. value added versus ighlights the importance of providing transparent information Lastly, our research h 10 DeArmond et to families to ensure all students have equal access to schools of choice. 9 Fryer (2014) finds that the practices of high -performing charter schools, when installed in traditional public schools, produce comparable gains in achievement. Abdulkadiroğlu et al. (2016) find those with lower applic ation costs to charter schools experience similar achievement gains to other students. 10 condary education For examples in th e U.S. context, see Hoxby and Turner ( 2013) and Bergman (2015 ) in post -secondary and se contexts. 5

7 al. cities and find that the most common barriers (2014) survey 4,000 parents across eight a school is understanding eligibility requirements, followed by transportation in choosing issues and obtaining accurate information about school characteristics. Hastings and (2008) show that providing school s’ test score in forma tion to families increases Weinstein their chances of selecting a higher -scoring school, which increases achievement. Corcoran et al. (2018) also show that providing information about nearby high schools’ selectiveness the quality of middle -school students’ high -school choices , and graduation rates improved particularly for non -English speaking households. Kapor et al. (2017) find that low -income families can misunderstand the school selection process and are less likely to be placed into their preferred school. I mpeding access to information about how to apply could opportunities for disadvantaged reduce students even when there is, ostensibly, open enrollment. normative question about what optimal policy require of choice There is a should schools with respect to whom they enroll. Our paper does not aim to identify an optimal policy, which depends on a particular social welfare function. Instead, our results inform how a socially optimal policy might be achieved. For instance, if a particular social welfare that all schools should have the capacity to serve any student, bolstering function implies existing efforts to reimburse schools for realized costs may ensure this opportunity for a variety of students. Other polic ies could coordin ate and simplify application processes to help everal education agencies have undertake n audits families make informed decisions. S like ours to monitor whether schools are providing information equitably across families . The rest of our paper proceeds as f ollows. Section 2 provides background on school choice for our sample. Section s 3 and 4 describe our intervention, experimental design presents our results and Section 6 concludes. and empirical analysis. Section 5 6

8 2. Background Information on School Choice and traditional public schools intra Our sample covers school choice in areas with -district charter schools. C harter schools are public, open -enrollment schools that have greater autonomy over their finances, staffing decisions, and curricula than traditional public schools , but they must admit students by lottery if more students apply to the school 11 Charter schools are the fastest growing form of school choice than can be accommodated. in the United States. Since 2010, more than 2,000 new charter schools have opened (NCES , 2015). They enroll nearly 3 million students at nearly 7,000 schools in 43 states and the District of Columbia. While their performance overall tends to be no better or worse than traditional public schools, charter schools in urban areas, which are often, no - excuses charter schools , have been shown to have large, positive impacts on student achievement ( Abdulkadiroglu et al., 2011; Angrist et al., 2013, 2016; Cohodes, 2018; Chabrier et al., 2016; Clark et al., 2015; Dobbie and Fryer, 2013; Dobbie and Fryer, 2015; Hoxby and Murarka, 2009). intra -district school choice Among the traditional public schools in our sample, y both state and local laws. operates in several ways. The rules are governed b For example, some states like Arizona make intra- district school choice mandatory for all school districts in the state. In states where it is voluntary, it is often practiced mostly by la rge urban school districts. At the local level, school districts commonly establish attendance areas, and students default to neighborhood -based assign ed schools but allow applications to other schools within the district subject to capacity constraints. Other to hood assignment and require that families apply school districts have no neighbor schools as a requirement for enrollment . Schools may also offer priority enrollment to 11 Specifically, lottery admission is required if they receive federal funding. 7

9 applicants based on residence, having a sibling within a school, or a safety concern. Among students based on a schools with more applicants than slots, schools first- may assign come, first -serve basis or a lottery . Both s and charter schools must comply with a number of traditional public school -discrimin ation laws, several of which have overlapping protections. The Civil Rights anti 12 Act of 1964 prohibits discrimination on the basis of race, religion, and country of origin. of 1973, Individuals with Disabilities Education Act and (IDEA) The Rehabilitation Act Americans with Disabilities Act of 1990 require LEAs to provide all necessary services to an students with physical or mental disabilities. IDEA mandates the creation of d Individualized Education Plan (IEP) for students with disabilities. The IEP, formulate with parents, school and other professional representatives, dictates what services the 13 to provide a student to address their special needs. LEA is legally obligated This last requirement may have greater are for charter schools that consequences authorized as their own LEA than for those authorized as part of an existing LEA or district (Heubert , 1997). The latter arrangement implies that charter schools may be able broader school district to help serve special -education to draw on resources from the while the former implies charter schools may have to address this requirement students, 14 entirely on their own. For this reason, charter schools that serve as their own LEA may respond to those that are not their different incentives during the application process than 12 The Civil Rights Act of 1964 further prohibits discrimination based on gender and sex, except for same -sex schools. 13 We summarize the above requirements as background information, but whether or not our findings constitute legal or illegal behavior on which is to determine experimentally whether schools practice any research question, of a school the is not germane to our first -order part form of differential treatment during the application process with respect to specific student characteristics. 14 rvices is multi -tiered. First, it Akin to other social insurance programs such as Medicare and Food Stamps, the economic justification for special education se provides a form of insurance to protect families who have children that are expensive to educate due to a disability; second, federal and state funding works as a form of insurance to protect local schools from the high cost of absorbing a disproportionate number of disabled students (Cullen and Rivkin, 2003). IDEA also permits the allocation of funds for a statewide “risk pool” to help LEAs serve students with high- cost disabilities ( Rhim et al., 2015). States may also designate charter schools to be part of a larger LEA specifically for IDEA purposes. There is some evidence that individual schools existing as their own LEA may form consortia to pool resources, making it easier to establish economies of scale and provide appropriate services for all students ( . NCESCS, 2017) 8

10 own LEA. Traditional public and charter school funding comes from federal, state, and local governments. tween charter school s and traditional public The degree of funding parity be 15 varies across states. Supplementary funds for students schools within the same state with disabilities can also vary based on state and local policies. Special education funds overwhelmingly (90%) come from state and local sources ( Cullen and Rivkin , 2003; Rhim et al. , 2015 ). As a point of reference, the average cost of educating a child wi th special needs is roughly 2.3 to 2.5 times that of without special needs ( Moore et al. , 1988; a child Chambers, 1998; Collins and Zirkel , 1992). A point of controversy is whether charter schools serve the most disadvantaged or costl iest -to-educate student at similar rates to traditional public schools . Nationally, the Government Accountability Office found that charter schools enroll a smaller proportion of students with severe disabilities than traditional public schools ( US Accountability Government , 2012 ce from specific locations is more nuanced. Setren (2015) finds ). But eviden Office 16 Hoxby and -area charter schools classify fewer students as special needs. that Boston Murarka (2009) find that New York City charter schools enroll more low -income and minority students than traditional public schools as a percentage of total enrollment . Using data from California and Texas, Booker et al. (2005 ) show that students who enroll in charter schools tend to have achievement than the students in the traditional lower public schools they left. 15 For example, the typical Oregon charter school receives only about 60 percent of the level of funding that a typical traditional public school receives while both charter and traditional public schools in Tennessee receive similar levels of funding ( Batdorff et al., 2014; Epple et al., 2016 ). 16 In part, this could be because the traditional public schools students were enrolled in prior to enrolling the charter school did not readily transfer their IEPs. 9

11 3. Experimental Design and Data Messages ed of email messages sent t o charter schools and traditional The field experiment consist school s of choice. We framed each message as coming from a parent looking for a public . The parent contacts the school to ask school about their child’s eligibility and how to apply. We developed our messages in consultation with chart er school and traditional - public school administrators who have received application inquiries via email. Our conversations with administrators at charter schools and traditional public schools found 17 The baseline message that p arents do make eligibility inquiries and provided examples. d that the parent is looking for a school for their son or daughter and they would indicate Each treatme nt like to know whether anyone can apply to that school and how to apply. message added a sentence to this baseline message to signal a child’s potential cost to educate, disadvantage or prior academic performance. This sentence indicated the child has one of the following: an IEP requiring they be taught in a classroom separate from mainstream students; poor behavior; bad grades; or good attendance and good grades. We show e xamples of the exact wording of these treatments in Figure 1 and Figure A1 . We chose these messages based on existing concerns about how schools may screen with potential applicants. Students certain disabilities may require additional support services . These students may an IEP that requires small- group instruction by a certified Special have teacher in a separate or “self -contained ” classroom. The poor -behavior message Education ties to a contention that some schools push out or screen children with behavior issues 17 For instance, parents frequently asked whether the school admits students with special needs or low grades. Administrators re ported other questions that parents ask as well, including whether parents have to volunteer, pay any fees, purchase school books or provide particular documentation about their child, whether there is a lottery or preferences for neighborhood families , and lastly, whether a or university admission. charter school diploma is valid for college 10

12 (Zimmer and Guarino, 2013) -grades message and the good grades and good . The poor reflects concerns schools may seek out students or screen attendance message that academic performance (Winters, Clayton, and Carpenter, 2017) . students based on their Lastly, characteristics of the parent and student were varied at random demographic across all messages. We randomized a signal of household structure by indicating that the parent and their spouse were making the inquiry (e.g. “My husband and I...”) or that just Bertrand and Mullainathan ( the name Following 2004), one parent making the inquiry. was 18 Hispanic , Black , or white. of the parent signaled the gender of the parent and whether they are The choice to randomly vary demographic signals also reflects concerns about how school may screen potential applicants. Minority background may signal socio- economic disadvantage and a child’s gender may signal behavior, disruptive as male students tend to be more disruptive in class than female students and Pan , 2013). (Bertrand and Sample Frames Experimental Design In the first experiment, conducted in late We conducted the experiment twice. we 2014, sent messages only to charter schools . In the second experiment, conducted in early 2018, we aimed to compare the response rates of traditional public schools in areas with int ra- district choice to the response of nearby charter schools. The three- year gap between rates the two experiments provide s a check on the consistency of the results across time. We constructed the sample frame for each experiment using the Common Core Data from the National Center for Education Statistics , which has information on the universe of both charter and traditional public schools . In the first experiment, we chose the 17 states with the largest number of charter schools. These 3,131 schools were roughly half 18 We chose names based on New York state data that associated names with race and gender. We then chose names that were overwhelmingly associated with one particular race and gender combination. 11

13 of the charter schools in the country at the time. The for the second sample frame was from the la (and the District of districts in 29 states experiment rgest 40 school 19 We matched charter schools Columbia) -district choice and charter schools. with intra to the nearest traditional public school with the same entry -grade level and within the same district boundari es. This sample consisted of 4, 338 schools, 1,016 of which were charter schools that were also in the first experiment, matched to 2,169 traditional public schools. Figure A 2 shows states with charter schools in light blue, states in our sample in dark blue, and states with no charter schools in gray. The sample has broad geographic coverage across the United States. App endix B discusses our sample construction in further deta il. In each experiment, w e sent two emails to each school three -to-four weeks apart . The treatment messages were randomly assigned at the school level in the first experiment . at the pair level ; identical For the second experiment, treatment messages were clustered - messages from the same fictitious parent were sent to each charter school and its match paired traditional public school. Within each experiment, n o school received the same message treatment twice and a school was assigned a treatment without replacement. We also the order in which schools were contacted. randomized Table 1 shows the results of regressing school characteristics on an indicator for each treatment. Each column restricts the sample to the baseline messages and the treatment message indicated in the column header. The results and joint test of covariates within each column show that rand omization generated assignments uncorrelated with school 19 choice policies. school ng a sample and investigating We selected the largest districts because of the fixed costs of creati 12

14 characteristics. shows the number of emails sent per treatment. Figure 1 More baseline and IEP messages were sent than behavior - and grades -related messages so that we would have 20 hose two treatments. Across the two experiments, additional precision with respect to t 21 we sent the same baseline, IEP, poor behavior and low -grades messages . The second experiment added the good -grades and good -attendance message to this list of treatments as well as the -parent household signal . two Data Data come from information on school websites, national databases of school demographic information , the census, a school -rating non -profit organization , and the responses to our emails . We hired research assistants to find and visit the website of every school in our sample frame. We then coded several variables including whether the school has a website, and, if so, whether the website includes a link to the school’s contact information on its landing page, a webform to contact the school a require ment to add a phone number , or We also used information on the school websites (and schools’ handbooks via the webform. the “no -excuses” charter school s in the sample . We based on these websites) to identify a template of characteristics common to such schools. this determination on We supplement ed these data with information on schools from two national databases: the Common Core of Data (CCD) and Civil Rights Data Collection (CRDC) . We use data to compute enrollment size, the share of students who are Black, d the CCD 20 Sending one baseline message to every school would have drastically reduced our power to detect treatment effects for each message type relative to the baseline message. In our first experiment, we wanted added precision for the IEP message, so we sent out relatively more IEP messages. Our primary specification, however, compares response rates to each treatment message to the response rate of the baseline message. A power analysis shows that power in the second experiment is maximized by sending roughly twice as many baseline messages as treatment messages. See Figure 4 for the exact message count per treatment in each experiment. 21 For instance, we randomly varied the subject line, the sign off Each message also had randomly assigned wording variations as well. (e.g. “thanks” or “thank you” or just signing the name) and the greeting. 13

15 Hispanic or white -price lunch at , and the share of students who receive free or reduced ed the latitude and longitude of a school and whether each school. These data also record it was We used the location data to link each located in an urban, suburban or rural area. the share of individuals by race, education, income school to census tract information on in a tract. d the number of students with a and disability From the CRDC data, we use disability for each school, which the CRDC breaks down by the portion of the day re not in a restricted environment (less than 40%, between 40% and 79% , and students a d these numbers into shares of enrollment. d the We also use 80% or more). We translate share of students who are chronically absent (missing 10% of days CRDC to calculate the , suspended, and have limited English proficiency. or more) GreatSchools d data provided by a nonprofit organization, , Third, we use which collects proficiency rates based on test scores for traditional public schools and charter . We average the proficiency across subjects (e.g. math schools across the country rates and grade levels for each school and reading) each . We use this measure to estimate school’ s value added by measuring the di fference between its observed average proficiency rate and the rate predicted based on the covariates specified above, state fixed effects, and an indicator for charter school or not . We then standardize this measure of value added according to the mean an d standard deviation within the sample. Fourth , we coded the responses of schools to our emails . We created an indicator for whether or not a school responds to a message, which is our outcome variable. Some schools provide automated responses (3% of emai ls receive an automated response) to our . Since each treatment is as likely to receive an automated response as another, messages this practice only raises overall response rates and our results are robust to excluding these messages (available upon request). 14

16 In the appendix, the characteristics of our sample schools (Table A1 ). On we show students from -income families , as shown by average, sample schools serve primarily low 52% who receive free or reduced -price lunch. Students are 29 % white, 32% the , Hispanic and 25% Black. Over half the sam ple is located in an urban area. We also show that school -level correlates of disadvantage predict lower response rates, and that high -value added schools are less likely to respond overall ( . ). Table A2 4. Empirical Strategy binary variable . W e Our outcome is a for whether a school responds to our inquiry or not estimate a linear -probability model as follows : β + Demographi = α + Treatment θ + ε cs δ + Wave γ + X R i i i i i i is our binary outcome. Treatment In this regression, R indicators is a vector of treatment i i for the mutually the IEP treatment, the behavior treatmen t, the -exclusive treatments: grades treatment, and the good -grades and good -attendance treatment. The low- Demographics randomized implied race , gender of the is a vector of indicators for the i child and parent , and whether the message indicates a two- parent household. X is a i vector of school covariates , including school demographics and whether or not the school is a traditional public school . Any missing covariates are imputed and an indicator for missing data is included in the regression. All regressions include indicators for the wave 22 23 of em ails. Standard errors are clustered at the school or school -by-pair level. To test whether traditional public schools respond at different rates to each type of message than charter schools , we interact each treatment with an indicator for whether 22 There are four waves of emails: two in the first experiment and two in the second experiment. 23 Specifically, clustering is at the school level for schools that are only in the first experiment, clustering is at the pair level for schools in both the first and second only in the second experiment, and clustering is at the school and pair level for schools that are are available experiment. In practice, the level of clustering does not alter the significance of the results; other versions of clustering upon request. 15

17 or no t the school is a traditional public school. The significance of this interaction effect responded more or less frequently than the tests whether the traditional public school other heterogeneity analyses, we either restrict the nearby charter school. To conduct sample according to a certain variable or fully interact it with our treatment messages as . specified in the text Our results replicate similarly across experiments, which may be unsurprising g iven messages were the same (except for the addition of new treatments) that the treatment and the policy environment was , we combine the two experiments in our similar. Thus of the results. The appendix results from . show presentation each experiment Table A3 separately . , however Lastly, we show our results are robust to different specifications and multiple -testing corrections. We show whether controls for school characteristics from the CCD and and pair fixed effects CRDC, census tract characteristics, affect our results. Given random assignment, n one of these additional covariates or fixed effects is required for In the adjust our main results— the effects of each treatment identification. appendix, we message and the differential effect between charter schools and traditional public schools -hypothesis testi ng (Table A5) . We use Holm’s step -down procedure —for multiple for groups of treatment effects specified in the text (Holm, 1979) . This procedure controls for the family (the probability of at least one false rejection of the null -wise error rate hypothesis ). 5. Results Table 2 presents the effects of our primary treatments— messages indicating poor behavior, the IEP, poor grades, and good grades and attendance— on response rates . The baseline message received at the a control mean at the bottom of Col umn (1) shows th 16

18 response 5 3% of the time. messages that the potential reply to signaling We find that schools are less likely to IEP (Column (1)) applicant has had bad grades, poor behavior, and an than the baseline . The bad -grades treatment reduces response rates by 2.4 percentage points. The message and poor behavior message reduce response rates by 5.1 percentage points IEP message 7.0 percentage points, respectively. T and he signal of good grad es and attendance has no discernable impact on response rates (the coefficient is 0.0 percentage points). In interpreting these results, note the IEP message signals a student who requires that tive or less costly services , the a restrictive environment. If the signal were for less restric hile schools do not actively provide more information to results may have differed. And w higher -performing students, this signal could be viewed by schools as “cheap talk.” Messages signaling these positive traits may be perceived as less truthful than the messages signaling other student traits. Table 2 treatments signaling race, Column (1) of also shows the results of our other gender and household structure. Only the message that indicates a Hispanic- soun ding name results in significantly lower 2.0 percentage points —response rates. The —by coefficient on B lack -sounding names is not significant (only with the addition of pair fixed effects is the result significant, as shown in Column (5) of Table A4 ). A n F-test for whether we can reject the null hypothesis that of demographic variables these treatments are jointly equal to zero cannot be rejected (p =0.14 for the specification in column 1). In the appendix ( Table A4 ), we show that the results above are extremely similar across different specifications . The coefficients do not vary significantly with the addition fixed effects. of school -level covariates , census tract -level covariates, or pair Heterogeneity across Traditional Public and Charter Schools 17

19 Given the growth rate of charter schools and concerns over whether they encourage all our differ between charter schools and students to apply findings , we test whether show results separately for traditional public schools . Columns (2) and (3) of Table 2 . Column (4) shows the traditional public schools and charter schools, respectively difference in response rates between traditional public schools and charter scho ols and its statistical significance. in areas of school choice generally respond at similar rates Traditional public schools as charter schools for each treatment, with one exception . Traditional public schools are 5.8 percentage points IEP message than significantly— —more likely to respond to the charter schools. This result remains statistically significant even after adjusting for (p=0.05 , Table A5 ). A test of whether multiple testing of these interaction terms the interaction effects for the primary treatments are jointly different from zero is also statistically significant (p =0.03 ). Treatment effects for signals of demographic characteristics do not differ significantly between traditional public schools and charter schools. Heterogeneity Across School Performance Levels and “No-Excuses” Charter Schools Access to high schools is important if school choice is to reduce gaps in -performing achievement across groups of students with different histories and attributes . Columns (1) and (2) of Table 3 shows treatment effects for all high and all low -value -added schools , which we define as above or below average value added in the sample . The IEP, behavior, and grades treatment effects are similar for h igh and low -value -added schools . Differences arise with the race implied by the messages, however: high value -added schools are less likely to respond to messages from families wi th Black or Hispanic -sounding names than messages from parents with Hispanic s -sounding name low value -added schools . Results for 18

20 (p =0.01 and p remain significant at the 1% level after adjusting for multiple testing ck-sounding name, respectively) result is sounding name and Bla . This =0.15 for Hispanic- -value added charter schools in the sample and by schools serving low driven by the high shares of minority students (results available upon request). Prior research has shown that urban , high -value added char ter schools can ). , 2011 (Abdulkadiroglu et al. significantly close racial gaps in student achievement shows the treatment effects on the primary treatment messages do not differ Column (3) significantly for this subgroup. Many “No -excuses” charter schools have demonstrated -achievement gaps in school districts. We identified a particular success in closing racial no-excuses schools from our sample . These schools tend to have much higher list of 272 test scores , value added, and serve larger shares of minority and low -income students relative to other charter schools. These schools also, however, serve fewer students with disabilities and have (see Table higher rates of suspensions than other charter schools excuses” charter schools is small, the coefficient on the IEP A6). While the sample of “no- 0 percentage points), negative, and significant . treatment remains large (1 Factors Moderating Funding Strategies States typically provide funding for special education students in three ways . Most common is formula -based funding . Th e formula may be a funding multiplier based on stu . Schools receive additional funds , but these are not necessarily ear - dent or staff counts marked for special -education services. Categorical funds or block grants similarly provide funds for schools , but these are ear -marked for special -education services. Finally, states to districts may provide partial or full reimbursement for their realized special education We generate a variable categorizing states by how they provide funding expenditures. 19

21 and interact this variable with our IEP treatment indicator. We restrict the sample we to charter schools, since traditional public schools are no less likely to respond to students 24 with an IEP to the baseline message. than Table 4 shows heterogeneous effects by funding strategy. Charter schools in states for (at least a portion) of their realized expenditures are 7 that reimburse schools percentage points more likely to respond to the IEP treatment message than charter schools in ot her states. This result is statistically significant at conventional levels (p=0.05). Charter schools in s tates with categorical funding have slightly higher response rates as well difference is not statistically significant . In states with for mula - , though the based funding, there are several weighting schemes ; we do not find any significant heterogeneity with respect to these different schemes (results available upon request). This pattern of results is consistent with an explanation that the costs of edu cating in most certain students are imperfectly compensated contexts . This pattern could create an incentive for schools to provide less encouragement to students with special application needs. Diversifying Risk: Local Education Agency (LEA) Status also shows heterogeneity by LEA status. LEAs, rath Table 4 an schools, bear the er th legal obligation to provide the services specified in students’ IEPs . Whether or not a charter school is its own LEA varies across and within states. Charter schools that a re their own LEA may have difficulty pooling resources and risk across schools . We multiple assess whether the opportunity to pool resources in this way is associated with differential response rates to the IEP message. We restrict the sample to charter sc hools and interact 24 Traditional public schools exhibit a similar pattern of heterogeneit y with respect to funding as charter schools, but the estimates are less precise much . 20

22 an indicator for LEA status with our IEP treatment. While own -LEA charter schools are slightly less likely to respond to the IEP treatment, this result is not statistically significant. Household Structure and Demographics explore whether signals of household structure exacerbate or attenuate Lastly, we a message indicated differential response rates across treatments. We randomized whether that a husband and wife are or just one parent is looking for a school and. We also randomized the race and gender of the child . Results restricting the sample to each of these groups are shown in Table A7 . Messages signaling a two -parent household , which could signal socio- tend to increase the likelihood of r esponse to economic advantage, children with poor behaviors and decrease the likelihood of response regarding children with good grades. Messages signaling a two -parent household also significantly increase Messages the likelihood of response for Black families. Hispanic with a -sounding name tend to have higher response rates to the good -grades treatment. Finally, treatment effects do not differ significantly for messages signaling a for the primary treatment messages compared to female students . male student 6. Conclusion While school choice improve access and quality , competitive pressures may can, in theory, also induce schools to keep costs low and discourage students perceived as costly to educate from enroll ing . These incentives may be present outside of school choice as well, but the opportunit y for schools to treat costlier or application process presents a distinct harder -to-educate students differently than with neighborhood -based school assignment. Using an audit study approach, we provide the first experime ntal assessment o f whether children who have with schools of choice fail to provide application information to families 21

23 a particular special need, behavior problem, or level , as well as of academic achievement race and gender much broader in scope than previous a perceived . Our study is also states and the District of Columbia research; we sample 29 and approximately half of the charter schools in the United States. We find that, on average, schools of choice are significantly less likely to provide informa tion to families with students who have low grades , behavior problems , or an IEP requiring they be taught in a separate classroom than to families of students without these attributes. Charter schools are significantly less likely to reply to students to t he IEP message than to the baseline message, while traditional public schools are not. T here is also evidence that schools are less likely to respond to families with Hispanic -sounding names . limitation of our analysis is that we chose one particular s One , and ignal of disability one that may be perceived as particularly costly for schools to provide services. Our findings may not generalize to other disabilities or students requiring other services, such as students with limited English proficiency. We did not incorporate a wider variety of treatments because of statistical power concerns. The implications of our results for o ptimal policy requires specifying a social welfare function. Some might favor maximizing a weighted average of student achievement , but pro many families see inclusion as essential for improving -social outcomes , especially as 25 Our results suggest that funding they pertain to students with special needs or diversity. is a key constraint on schools’ willingness to serv e students with disabilities . We show suggestive evidence that cost -reimbursement funding mitigates charter schools’ 25 am Rao (2018) on how the integration of castes in India impacted social preferences around For instance, see work by Gaut inequality, inclusion and altruism. 22

24 differential response rates for students with IEPs. All schools —including traditional publi c and charter —are less likely to respond to students with poor prior behavior and low . Conducting systematic audits could help deter this behavior. S grades everal education rter agencies undertake n audits via phone calls asking eligibility questions to cha have schools . Going forward, future research could use our methodology to determine schools’ responses to students with other characteristics, such as Limited English Proficiency and specific -learning disabilities. 23

25 References dul adiro ̆ glu, A tila, Jos h ua D Ab Angrist, k Susan M Dynarski, Thomas J Kane, and Parag A Pathak , “Accountability and flexibility in public schools: Evidence from Boston’s charters and pilots,” The Quarterly Journal of Economics , 2011, 126 (2), 699 –748. Ab k adiro ̆ glu, A tila, Jos h ua dul D Angrist, Peter Hull , and Parag A Pathak , “Charters without Lotteries: Testing Takeovers in New Orleans and Boston ,” American Economic Review , 2016, 106 (7), 1878–1920. Ab dul adiro ̆ glu, A tila, Parag Pathag, Jonathan Schellenberg and Christopher k “Do Parents Value School Effectiveness?” 2017 23912 , Walters, NBER Working Paper Joseph Altonji, -I Huang, and Christopher R Taber , “Estimating the G, Ching cream skimming effect of school choice,” Journal of Political Economy , 2015, 123 (2), 266–324. Angrist, Joshua D, Parag A Pathak, and Christopher R Walters , Charter School Effectiveness,” , American Economic Journal: Applied Economics “Explaining 5 (4), 1 2013, –27. Angrist, Joshua D., Sarah R. Cohodes, Susan M. Dynarski, Parag A. Pathak, and Christopher R. Walters , “Stand and Deliver: Effects of Boston’s Charter High Schools on College Preparation, Entry, and Choice,” Journal of Labor Economics , April 2016, 34 (2). Ayres, Ian and Peter Siegelman , “Race and Gender Discrimination in Bargaining , 1995, 85(3): 304- 321. for a New Car,” The American Economic Review 24

26 Baker, Rachel, Thomas Dee, Brent Evans, and Juhn John , “Bias in Online rom a Field Experiment,” Classes: Evidence f Center for Education Policy Analysis working paper, 2018. Batdorff, Meagan, Larry Maloney, Jay F May, Sheree T Speakman, , “Charter School Funding: Inequity Expands” Patrick J Wolf, Albert Cheng School Demon stration Project , University of Arkansas. April 2014. wp- content/uploads/charter -funding -inequity- expands.pdf Bertrand, Marianne and Jessica Pan , “The trouble with boys: Social influences and the gender gap in disruptive behavior,” Amer ican Economic Journal: Applied Economics , (1), 2013, 5 32–64. , “Are Emily and Greg More Employable Than and Sendhil Mullainathan Jamal? Experiment on Labor Market Discrimination,” Lakisha and American A Field Economic Review , 2004, 94 (4), 991–1013. Bergman, Peter, “Parent -Child Information Frictions and Human Capital Investment: Evidence from a Field Experiment CESifo Working Paper 5391 , 2015 . ,” Bergman, Peter, “The Risks and Benefits of School Integration for Participating Students: Evidence from a Randomized Desegregation Program,” No. 11602. Institute for the Study of Labor (IZA) , 2018. Bifulco, Robert, Helen F Ladd, and Stephen L Ross , “The effects of public school choice on those left behind: Evidence from Durham, North Carolina,” Peabody Journal (2), 130–149. 84 , 2009, Education of 25

27 Booker, Ron Zimmer, and Richard Buddin , Kevin, effects of charter “The schools on school peer composition,” 2005. Chabrier, Julia, Sarah Cohodes, and Philip Oreopoulos , “What Can We Learn from The Journal of Economic Perspectives , 2016, 30 (3), 57 –84. Charter School Lotteries?,” Chambers, JG , “Geographic Variations in Public Schools Costs (NCES 98- 04),” US National De partment of Education. Washington, DC: Center for Education Statistics Work- ing Paper , 1998. Chetty, Raj, John N. Friedman, Nathaniel Hilger, Emmanuel Saez, Diane Whitmore Schanzenbach, and Danny Yagan, “How does your kindergarten classroom affect your earnings? Evidence from Project STAR. ” The Quarterly Journal of , 2011, 126(4) Economics 1660. , 1593- Chetty, Raj, John N. Friedman, and Jonah E. Rockoff , “Measuring the impacts of teachers II: Teacher value- added and student outcomes in adulthood,” American Economic Review , 2014, 104(9) , 2633- 79. Clark, Melissa A, Philip M Gleason, Christina C Tuttle, and Marsha K Silver berg , “Do Charter Schools Improve Student Achievement,” Educational Evaluation and Policy Analysis , 2015, 37 (4), 419–436. Cohodes, Sarah , “Charter schools and the achievement gap ,” The Future of Children , winter, 2018. Cohodes, Sarah and Susan Dynarski, “Massachusetts charter cap holds back Evidence Speaks Reports, Economics Studies at Brookings disadvantaged,” , 2016, Vol 2 26

28 (1) pp. 1 -6. , “To What Extent, If Any, May Cost Be a Factor in Collins, LA and PA Zirkel Education Law Reporter , 1992, pp. 11–25. Special Education Cases?,” , “The Gender Gap in Charter School Corcoran, Sean P and Jennifer L Jennings Education Policy , forthcoming. Enrollment,” Cullen, Julie Berry and Steven G Rivkin , “The role of special education in school choice,” in “The economics of school choice,” University of Chicago Press, 2003, pp. 67– 106. Darolia, Rajeev, Cory Koedel, Paco Martorell, Katie Wilson, and Francisco Perez-Arce , “D o Employers Prefer Workers Who Attend For- Profit Colleges? Evidence from a Field Experiment,” Journal of Policy Analysis and 903. Management , 2015, 34(4): 881- DeArmond, Michael, Ashley Jochim, and Robin Lake , “Making School Choice Center on Reinventing Public Education Work,” , July 2014. sites/default/files/CRPE_MakingSchoolChoiceWork_Report.pdf David Deming, S. Hastings, Thomas J. Kane, and Douglas O. J., Justine Staiger , “School Choice, School Quality, and Postsecondary Attainment,” American Economic , 2014, 104 (3), 991–1013. Re- view Deming, David J . “Better schools, less crime?,” Quarterly Journal of Economics , 2011, 126(4), 2063- 2115. 27

29 Deming, David J ., Noam Yuchtman, Amira Abulafi, Claudia Goldin, and , “The Value of Postsecondary Credentials in the Labor Market: An Lawrence F. Katz Experimental Study,” , 106(3): 778- 806. American Economic Review Dobbie, Will and Jr. Fryer Roland G. , “Getting beneath the Veil of Effective Applied Schools: Evidence New York City,” American Economic Journal: from Economics , 2013, 5 (4), 28–60. G. Fryer and Roland , “The Medium -Term Impacts of High -Achieving Charter Schools,” Journal of Political Economy 123 (5), 985 –1037. , 2015, , “Racial Discrimination Edelman, Benjamin, Michael Luca, and Dan Svirsky American Economic in the Sharing Economy: Evidence from a Field Experiment,” Journal: Applied Economics , 2017, 9(2), 1- 22. Epple, Dennis, Richard Romano, and Miguel Urquiola , “School Vouchers: A Survey of the Economics Literature,” Journal of Economic Literature , 2017, 55(2), 441- 492. Epple, Dennis, Richard Romano, and Ronald Zimmer , “Charter Schools: A Survey of Research on their Characteristics and Effectiveness,” Handbook of Economics of Education , Editors, Eric Hanushek, Stephen Machin, and Ludger Woessmann, 2016, Vol. 5, 139 -208. Friedman, Milton, Capitalism and Freedom , Cambridge University Press, 1962. Fryer, Roland, tional Public “Injecting Charter Schools Best Practices into Tradi 28

30 Schools: Evidence from Field Experiments,” , 2014, The Quarterly Journal of Economics -1407. 29 (3), 1355 -Cost Special Education Griffith, Michael . “State Funding Programs for High , May 2008. Students,” Education Commission of the States: State Notes , “Charter Schools: Additional Accountability Office (GAO) Government tention Needed to Help Protect Access for Students with Disabilities. Report Federal At to Congres Requesters. GAO -12-543.,” US Government Accountability Office , 2012. sional . “Understanding the Twentieth- Hanushek, Eric and Steven Rivkin Century Journal of Human Resources , 1997, 32(1), 35 Growth in U.S. School Spending,” -68. Hastings, Justine, Thomas Kane, and Douglas O. Staiger , “Parental Preferences and School Competition: Evidence from a Publi c School Choice Program,” 2006. NBER Working Paper No. 11805, Hastings, Justine Jeffrey Weinstein , “Information, School Choice, and and Academic from Two Experiments,” Evidence The Quarterly Journal of Achievement: Economics , 2008, 123 (4), 1373–1414. Heubert, Jay P -Charter Schools? Federal Disability Law, , “Schools without Rules and the Paradoxes of Deregulation,” Harv. CR -CLL Rev. , 1997, 32 , 301. Holm, Sture , “A Simple Sequentially Rejective Multiple Test Procedure”, Scandinavian Journal of Statistics, 1979, 6(2) , pp. 65- 70. City: Who York Hoxby, Caroline and Sonali Murarka , “Charter schools in New 29

31 enrolls how they affect their students’ achievement,” Technical Report, National and Bureau of Economic Research 2009. Hoxby, Caroline M and Sarah Turner , “Expanding College Opportunities for High - Achieving, Low Income Students,” 2013. SIEPR Discussion Paper 12- 014. Hsieh, Chang -Tai and Miguel Urquiola , “The effects of generalized school choice on Journal of public achievement and stratification: Evidence from Chile’s voucher program,” Economics , 2006, 90 (8), 1477–1503. Kapor, Adam, Christopher Neilson, and Seth D Zimmerman , “Heterogeneous 201 Beliefs and School Choice,” Working Papers 612, Princeton University, 7. Kline, Patrick and Christopher R. Walters , “Evaluating Public Programs with Quarterly Journal of Economics, - Close Substitutes: The Case of Head Start,” 2016, pp. 1795 1848. Lacireno -Paquet, Natalie, Thomas T Holyoke, Michele Moser, and Jeffrey , “Creaming R Henig cropping: Charter school enrollment practices in response to versus market incentives,” Educational Evaluation and Policy Analysis , 2002, 24 (2), 145–158. Lazear, Edward P The Quarterly Journal of Economics , , “Educational Production,” 116 (3), 777–803. 2001, Loeb, Susanna, Jon Valant, and Matt Kasman , “Increasing Choice in the Market for Schools: Recent Reforms and their Effects on Student Achievement,” National Tax Journal , 2011, 64 (1), 141- 164. and Competitive Lubienski, Christopher , “Marketing Schools: Consumer Goods 30

32 Incen for Consumer Information,” Education and Urban Society , 2007, 40 (1), 118- tives 141. Lubienski, Christopher, Charisse Gulosino, and Peter Weitzel “School Choice , and Competitive Incentives: Mapping the Distribution of Educational Opportunities across Local Educational Markets,” 2009, 115. American Journal of Education Millard, Maria and Stephanie Aragon . “State funding for students with disabilities,” Education Commission of the States: 50 -State Review , June 2015. Moore, William Strang, Myron Mary, and Mark Braddock , Schwartz, “Patterns in Special Education Service Delivery and Cost.” 1988. Muralidharan, Karthik and Venkatesh Sundararaman , “The Aggregate Effect Experiment of School Choice: Evidence from a Two -Stage in India,” The Quarterly Journal of Economics , 2015, (3), 1011 –1066. 130 CES , “Table 216.90: Public National Center for Education Statistics (N ) state: Selected years, 1999- elementary and secondary charter schools and enrollment, by through 2013- 14,” in “Digest of Education Statistics,” National Center for Education 2000 2015. Statistics, National Center for Special Education in Charter Schools (NCSECS), “What is an Infrastructure? Special Education Support Systems in Charter Schools,” 2017, Issue Brief—Infrastructures , Nichols -Ba rrer, Ira, Philip Gleason, Brian Gill, and Christina Clark Tuttle , in KIPP Evaluation selection, attrition, and replacement dent middle schools,” Educational “Stu -73. , 2015, p. 162 and Policy Analysis 31

33 Rao, Gautam , “Familiarity Does Not Breed Contempt: Diversity, Discrimination ,” forthcoming, . and Generosity in Delhi Schools American Economic Review Lauren M., Paul O’Neill, Rhim, Ruck, Kathryn Huber, and Sivan Amy Tuchman , “Getting lost while trying to find the money: Special education finance in charter schools,” 2015. Scott, George A , “Charter Schools: Additional Federal Attention Needed to Help Protect Access for Students with Disabilities. Report to Congressional Requesters. GAO -12- 543.,” 2012. US Government Accountability Office , Setren, Elizabeth Education and English Language Learner Students in , “Special Boston Charter Schools: Impact and Classification,” 2015. “The Demand for Effective Charter Schools,” Journal of Walters, Christopher, , 2018, Forthcoming . Political Economy Winters, Marcus A., Grant Clayton, and Dick M. Carpenter II , “Are low - performing students more likely to exit charter schools? Evidence from New York City Economics of Education Review 2017, 56 pp. 110- 117. and Denver, Colorado,” Wong, Mitchell D., Karen M. Coller, Rebecca N. Dudovitz, David P. Kennedy, Richard Buddin, Martin F. Shapiro, Sheryl H. Kataoka Arleen F. Brown, Chi -Hong Tseng, Peter Bergman , and Paul J. Chung , “Successful schools and risky behavior -income adolescents ,” Pediatrics, s among low 2014, 134(2) . Zimmer, Ron, Brian Gill, Kevin Booker, Stephane Lavertu, Tim R Sass, and on achievement, attainment, John Witte , Charter schools in eight states: Effects 32

34 gration, and competition , Vol. 869, Rand Corporation, 2009. int e Zimmer, Ron and Cassandra Guarino , “Is there empirical evidence that charter -performing 2013, schools push out low students?,” , Educational Evaluation and Policy Analysis (4), 461 –80. 35 33

35 Figures (a) First Experiment (b) Second Experiment Figure 1 - Experimental Design for (a) First experiment sample and (b) Second experiment sample (Note: This figure shows the study design and examples of the treatment phrasings). 34

36 Tables alance across treatment messages Table 1 – Testing for baseline b Bad Good Bad Baseline IEP Behavior Grades Grades TPS 0.004 0.003 0.005 - 0.004 0.293 (0.455) (0.005) (0.005) (0.005) (0.005) % LEP students 0.047 0.033 - 0.038 - 0.010 0.123 (0.036) (0.18) (0.039) (0.034) (0.035) % Chronically absent students 0.159 - 0.026 0.030 0.010 0.008 (0.027) (0.026) (0.032) (0.187) (0.03) % Students with IEP 0.075 0.034 - 0.039 0.015 0.023 - (0.084) (0.062) (0.056) (0.061) (0.054) % Proficiency 0.463 0.021 0.060 0.057 0.035 (0.042) (0.042) (0.038) (0.225) (0.041) % Black students 0.254 - 0.010 - 0.021 0.029 0.008 (0.321) (0.02) (0.017) (0.02) (0.018) % Hispanic students 0.011 - 0.006 0.041 - 0.023 0.321 (0.024) (0.021) (0.31) (0.023) (0.022) % FRPL students 0.59 - 0.020 0.029 - 0.001 0.034* (0.026) (0.329) (0.026) (0.023) (0.024) Value added measure 0.017 - 0.006 - 0.011 - 0.013 - 0.005 (0.022) (0.992) (0.019) (0.02) (0.021) value , joint test - 0.385 0.21 0.854 0.734 P , 5 , 031 2 , 441 3 , 522 2 1 462 Observations , 434 Notes: Column (1) shows means for the baseline message and the standard deviations in Each other column is as follows: For each treatment, the sample is restricted parenthesis. indicated in the column header, and the treatment to baseline and treatment observation indicator is regressed on the covariates shown in each row . School -level demographic data including absenteeism, Free- Reduc ed Priced Lunch (FRPL), Limited English Proficiency (LEP) - , and disability data is retrieved from the Civil Rights Data Collection (CRDC) 2013 14 public dataset. % Proficiency measures the rate of students (0 -1) at or above proficiency, as reported by Great Schools in their most recent dataset from 2016. Value -added measure (VAM) is constructed by calculating the normalized difference between the observed proficiency and the predicted proficiency from a regression including school -level and tract - tandard errors (by pair and school) in parentheses. -robust s Two -way c luster ols. level contr The joint test is a test of whether the covariates are jointly different from zero. *** p<0.01, ** p<0.05, * p<0.1 35

37 Table 2 – Effects of Message Treatments on School Response Rates TPS Full Sample TPS - Charter diff. Sample Charter - 0.024** - 0.044** - 0.015 - 0.030 Bad Grades (0.011) (0.021) (0.013) (0.025) 0.052*** - 0.002 IEP - 0.065*** 0.058** - (0.010) (0.021) (0.011) (0.023) Bad Behavior - 0.070*** - 0.053*** - 0.076*** 0.021 (0.011) (0.021) (0.013) (0.024) - 0.013 Good Grades 0.005 0.015 0.001 (0.014) (0.021) (0.020) (0.029) 0.010 - 0.012 - 0.004 - 0.015 Black (0.010) (0.018) (0.011) (0.022) 0.023** Hispanic - 0.020** - 0.015 - 0.009 (0.010) (0.018) (0.011) (0.021) 0.003 - 0.009 0.008 - 0.017 Father (0.015) (0.009) (0.017) (0.008) - 0.011 - 0.007 Son - 0.011 0.004 (0.011) (0.008) (0.014) (0.007) 0.014 Two Parents 0.007 0.010 - 0.009 (0.011) (0.015) (0.015) (0.021) TPS 0.008 (0.013) Observations 14,806 4,296 10,510 14,806 Control Group Mean 0.533 0.496 0.548 0.533 Notes: Table shows school the results of a multivariate regression of an indicator for whether or not a -treatment indicators. Columns (1) -(3) show the results for different responded to the message on message samples: (1) full sample, (2) only traditional public schools and (3) only charter schools. Column (4) interaction between primary and secondary treatments and TPS . TPS is a n indicator variable (not a treatment) for whether a school is a traditional public school . All other variables included in the table are randomly assigned characteristics of the emails. Regressions include fixed effects by wav e and state. Two -way c luster -robust s tandard errors (by pair and school) in parentheses. *** p<0.01, ** p<0.05, * p<0.1 36

38 Table 3 - Differential Effect of Messages by VAM Performance and Type of Charter High VAM Low VAM High - VAM , Sample No Excuses Charter Urban Charter - 0.022 - 0.025 0.010 0.016 Bad Grades (0.018) (0.028) (0.052) (0.018) 0.104** 0.049*** - 0.056*** - 0.093*** IEP - - (0.016) (0.016) (0.024) (0.044) Bad Behavior - 0.067*** - 0.066*** 0.070** 0.009 - (0.018) (0.028) (0.054) (0.018) 0.035 - 0.034 0.054 0.033 Good Grades (0.022) (0.036) (0.063) (0.022) - 0.032** 0.007 Black 0.049** 0.036 - (0.016) (0.024) (0.050) (0.015) - 0.049*** 0.019 Hispanic 0.062** 0.029 - (0.015) (0.044) (0.016) (0.024) Father - 0.006 0.017 0.085** 0.012 (0.013) (0.013) (0.020) (0.037) - 0.006 - 0.012 - 0.007 - 0.069** Son (0.011) (0.010) (0.017) (0.031) 0.012 0.007 Two 0.002 0.046 Parents (0.017) (0.018) (0.029) (0.047) TPS 0.016 0.003 (0.020) (0.020) 8,835 Observations 2,516 3,457 689 Control Group Mean 0.549 0.538 0.538 0.559 Notes: Table shows the effect of different treatments on for different samples. TPS is a covariate for response rate . All other variables included in the table are randomly assigned characteristics of the emails. traditional public school Sample row shows the population for each regression: (1) Schools with high value -added measure (>0), (2) Schools with low high value -added measures (<0), (3) Urban charter schools with high VAM, and (4) “No Excuses” charter schools . Regressions include fixed effects by wave and state. Two luster -robust s tandard errors (by p air and school) in -way c parentheses. *** p<0.01, ** p<0.05, * p<0.1 37

39 Table 4 - Effect of IEP Message by State Funding Category and LEA status for Charter Schools (2) (1) -0.064*** IEP -0.080*** (0. 015 ) (0.016 ) IEP x Funding = Categorical 0.022 (0. 025) 0.068 IEP x Funding = Reimbursement ** (0. 035) IEP x Own LEA -0.008 (0. 023 ) 10,510 10,087 Observations Control Group Mean 0.548 0.548 Notes: Table shows the effect of different treatments on response rate for charter schools in the sample depending on state funding and whether they are their own LEA (Local Education Agency). The categories for forms of funding are “Formula” (base), “Categorical,” and “Reimbursement.” All other treatment variables are randomly assigned characteristics of the emails. Regressions include all treatment indicato rs (only IEP is shown), wave and state fixed effects. robust standard errors (by pair and school) in - Nested cluster parentheses. *** p<0.01, ** p<0.05, * p<0.1 38

40 Appendix A. Figures (a) (b) Figure A 1 - Message sent to parents based for (a) baseline and (b) IEP intervention (Note: This figure shows an example of one phrasing for the Baseline and IEP message) 39

41 Figure A 2 - Map of Sample Frame (Note: The map shows the sample frame for the experiment) 40

42 Appendix Tables 1 – Total Population of Traditional Public Schools and Charter Schools vs Sample Table A Sample CRDC Population Charter Charter Charter TPS TPS Charter (Exp 2) (Exp 1) (All) % Limited English Proficiency 0.094 0.186 0.096 0.091 0.12 0.081 (0.148) (0.162) (0.2) (0.154) (0.182) (0.167) Absenteeism Rate 0.188 0.112 0.162 0.183 0.15 0.17 (0.292) (0.384) (0.226) (0.333) (0.319) (0.353) % IDEA Students 0.093 0.117 0.088 0.081 0.108 0.086 (0.134) (0.166) (0.099) (0.146) (0.168) (0.1) % of IEP students in regular class <40% of time 0.008 0.004 0.017 0.002 0.003 0.002 (0.024) (0.023) (0.03) (0.034) (0.03) (0.009) % of IEP students in regular class 40 -79% of time 0.005 0.015 0.005 0.016 0.005 0.005 (0.019) (0.059) (0.02) (0.028) (0.021) (0.024) % of IEP students in regular class >80% of time 0.067 0.056 0.069 0.077 0.056 0.064 (0.134) (0.045) (0.096) (0.116) (0.088) (0.047) % Students Receiving In -School Suspension 0.016 0.028 0.022 0.015 0.022 0.018 (0.296) (0.304) (0.109) (0.208) (0.254) (0.227) % Students Receiving 1 Out -of-school Suspension 0.031 0.01 0.024 0.026 0.028 0.027 (0.263) (0.299) (0.077) (0.223) (0.204) (0.25) % Proficiency 48.146 42.177 48.374 48.703 40.329 44.475 (15.725) (22.273) (22.515) (20.445) (22.044) (23.182) Observations 89 , 898 5 , 813 2 , 170 4 , 283 2 , 582 1 , 701 Notes - 14 dataset. The CRDC does not have data on all : Means are reported for all schools that are available in CRDC 2013 schools in the study sample, which was from 2016 -17. TPS schools are all schools that are not classified by CRDC as alternative, special education, magnet, or charter schools. Charter schools are all schools that are classified by CRDC as charter schools. IDEA students refer to students in the Individuals with Disabilities Education Act. IEP students refers to students that have an Individualized Education Program covered by IDEA. Charter schools may be alternative, magnet, or special education schools. School proficiency scores show the percentage of students scoring at or above proficiency on state assessments across grades and subject as reported by Great Schools in their most recent dataset from 2016. Charter (All) column r epresents all the charter schools in the experimental sample for both experiments. Charter (Exp 1) and Charter (Exp 2) columns show the characteristics of the charter schools included in the first and second experiment, respectively. 41

43 Table A 2 – The Relationship between Response and School Locations and Demographics Responded 0.008 City (0.027) - Suburb 0.021 (0.027) Rural 0.015 (0.031) % FRL -0.070*** (0.019) 0.180*** % Female (0.067) - 0.321*** % Black (0.047) 0.006 % White (0.051) % Hispanic -0.252*** (0.047) Enrollment 0.000** (0.000) % Proficiency -0.001** (0.000) Proficiency (missing) - 0.055*** (0.018) Constant 0.662*** (0.059) 14,064 Observations R - 0.049 squared Table shows regression coefficients and standard errors Notes: a regression of school covariates on an indicator for for response. School -level demographic data including enrollment , FRPL population, race and ethnicity breakdowns is retrieved from National Center for Education Statistics (NCES) 16 school year. School Common Core of Data for the 2015- proficiency shows the % of proficient (or higher) students as reported by GreatSchool s in their most recent dataset from luster ed standard errors in parentheses. 2016. Nested c *** p<0.01, ** p<0.05, * p<0.1 42

44 Table A 3 - Effect of Message Treatments on Response Rates from Schools by Experiment nd nd st nd , 2 2 Experiment TPS , 2 Sample - Charter Experiment TPS Diff . 1 nd Experiment Charter, 2 Experiment Experiment 0.004 - Bad Grades - 0.044** - 0.043** - 0.005 0.044*** (0.017) (0.014) (0.021) (0.020) (0.029) 0.002 IEP - 0.042*** - - - 0.082*** 0.073** 0.055*** (0.013) (0.015) (0.021) (0.020) (0.029) Bad Behavior - 0.064*** - 0.072*** - 0.053*** - 0.091*** 0.037 (0.015) (0.021) (0.017) (0.020) (0.029) Good Grades - 0.003 0.013 - 0.019 0.027 (0.014) (0.021) (0.020) (0.029) Black - 0.024 - 0.003 - 0.004 - 0.002 - 0.003 (0.013) (0.018) (0.015) (0.018) (0.026) 0.013 Hispanic - 0.031** - 0.003 - 0.015 - 0.011 - (0.015) (0.013) (0.018) (0.018) (0.025) 0.012 - 0.001 - 0.009 0.007 - 0.016 Father (0.015) (0.015) (0.011) (0.021) (0.012) - 0.020* - 0.004 Son 0.007 - 0.000 - 0.007 - (0.011) (0.008) (0.011) (0.010) (0.015) 0.014 Two Parents 0.011 0.007 - 0.008 (0.021) (0.011) (0.015) (0.015) TPS 0.008 (0.013) Observations 6,210 8,596 4,300 8,596 4,296 R - squared 0.039 0.043 0.064 0.038 0.044 Control Group Mean 0.503 0.496 0.575 0.510 0.503 Notes: Table shows the effect of different treatments on response rate for the first (old) and second (new) experimental sample. TPS is a covariate for traditional public school. All other variables included in the table are randomly assigned characteristics of the emails. Regressions include fixed effects by wave and states. Sample row shows the population for each regression: (1) Old sample (full), which only included charter schools), (2) New sample (full), (3) only traditional public schools for new sample, (4) only charter schools for new sample, and (5) interaction betw een primary and secondary treatments and TPS for new sample . Nested cluster -robust s tandard errors (by pair and school for new sample (full and diff TPS- Charter) and only school for others) in parentheses. *** p<0.01, ** p<0.05, * p<0.1 43

45 Table A 4 - Effect of Message Treatments on Response Rates from Schools for Different Specifications (1) (2) (3) (4) (5) - - 0.027** - - 0.022* - 0.022* 0.024** Bad Grades 0.027** (0.011) (0.011) (0.011) (0.012) (0.012) IEP 0.052*** - 0.051*** - 0.051*** - 0.056*** - 0.056*** - (0.010) (0.010) (0.010) (0.010) (0.010) 0.069*** Bad Behavior - 0.069*** - - - 0.070*** - 0.069*** 0.070*** (0.011) (0.011) (0.011) (0.012) (0.012) 0.001 0.002 0.002 - 0.007 - 0.007 Good Grades (0.014) (0.014) (0.015) (0.015) (0.014) - 0.014 - 0.014 0.012 - 0.019* - 0.019* Black - (0.010) (0.010) (0.010) (0.011) (0.011) Hispanic - 0.020** - 0.021** - 0.021** - 0.018* - 0.017* (0.010) (0.009) (0.009) (0.010) (0.010) 0.003 Father 0.000 - 0.000 - 0.001 - 0.000 (0.008) (0.009) (0.008) (0.008) (0.009) Son - - 0.010 - 0.010 - 0.009 - 0.009 0.011 (0.007) (0.007) (0.007) (0.007) (0.007) 0.009 0.010 Two Parents 0.013 0.013 0.010 (0.011) (0.011) (0.011) (0.012) (0.012) 0.027** 0.030** 0.008 0.033** 0.031* TPS (0.013) (0.013) (0.014) (0.016) (0.016) 14,806 14,806 14,806 14,806 14,806 Observations R - squared 0.039 0.068 0.070 0.267 0.269 Control Group Mean 0.533 - level Controls 1 X X X School X - X School level Controls 2 X X Pair Fixed Effects X X - Tract level Controls X Notes: Table shows the effect of different treatments on response rate. TPS is a covariate for traditional public school. All other variables included in the table are randomly assigned characteristics of the emails. -level controls 1 include fraction of sc hool population that is black, Hispanic, free -and -reduced price School lunch eligible, female, the school's proficiency rating, and city, suburb, and rural status of the school's location. School -level controls 2 include fraction of the school population that has a disability, proportion of students with disabilities that are taken out of regular class <39%, 40 -79%, and 80%+ of the time, fraction of school population that is chronically absent, and fraction of the population that has received an out -of- school susp ension. Tract -level controls include disability shares, race/ethnicity shares, share of residents with a Bachelor’s Degree, median earnings, poverty rates, and food stamp recipients for the school’s (by pair and school) associated census tract. Nested c luster -robust standard e rrors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 44

46 Table A 5 – Effects of Message Treatments on School Response Rates Adjusting by Multiple Hypothesis Testing Full Sample TPS Charter TPS - Charter diff. Sample 0.015 - - 0.044* - 0.024* - 0.030 Bad Grades [0.064] [0.097] [0.512] [0.648] IEP - 0.052*** - 0.002 - 0.065*** 0.058* [<0.001] [0.917] [<0.001] [0.053] Bad Behavior - 0.070*** - 0.053** - 0.076*** 0.021 [<0.001] [0.038] [<0.001] [0.772] - 0.013 Good Grades 0.005 0.015 0.001 [0.941] [1.000] [0.817] [0.595] 0.010 - 0.012 - 0.004 - 0.015 Black [0.631] [0.818] [0.578] [1.000] - Hispanic - 0.02 0.015 0.009 - 0.023 [0.166] [1.000] [0.201] [1.000] - 0.009 0.008 - 0.017 0.003 Father [0.724] [1.000] [0.396] [1.000] 0.004 - 0.011 - 0.007 - 0.011 Son [0.441] [1.000] [0.599] [0.76] Two Parents 0.010 0.007 0.014 - 0.009 [0.719] [1.000] [0.708] [1.000] 14,806 Observations 4,296 10,510 14,806 Control Group Mean 0.533 0.496 0.548 0.533 Notes: Table shows - treatment indicators on an indicator variable for the results of regressing message whether or not a school responded to the message. Columns (1) -(3) show the results for different samples: (1) full sample, (2) only traditional public schools and (3) only charter schools. Column (4) interaction between primary and secondary t reatments and TPS . TPS is a n indicator variable (not a treatment) for whether a school is a traditional public school. All other variables included in the table are randomly assigned characteristics of the emails. Regressions include fixed effects by wave and state. Adjusted p- values by multiple hypothesis testing in squared parentheses. Multiple hypothesis adjustment was done using Holm’s method separately for primary and secondary treatments. *** p<0.01, ** p<0.05, * p<0.1 45

47 Table A 6 - Total Population of Charter Schools vs Sample Charter Schools and No Excuses Charter Schools CRDC Sample No Excuses Charter Charter Charter 0.094 0.096 0.157 % Limited English Proficiency (0.167) (0.19) (0.162) 0.17 Absenteeism Rate 0.112 0.162 (0.384) (0.333) (0.157) % IDEA Students 0.086 0.088 0.069 (0.166) (0.146) (0.054) 0.002 % of IEP students in regular class <40% of time 0.001 0.004 (0.03) (0.024) (0.006) % of IEP students in regular class 40 - 79% of time 0.005 0.005 0.004 (0.021) (0.012) (0.02) % of IEP students in regular class >80% of time 0.067 0.069 0.047 (0.096) (0.051) (0.088) % Students Receiving In School Suspension 0.016 0.018 0.041 - (0.304) (0.227) (0.064) % Students Receiving 1 Out - of - school Suspension 0.024 0.027 0.045 (0.223) (0.042) (0.299) 44.475 48.374 62.733 % Proficiency 21.945 (20.445) (22.515) ( ) Value Added Measure - 0.000 0.775 (0.976) (1.092) Observations 4283 272 5813 Notes : Means are reported for all schools that are available in CRDC 2013 - 14 dataset. The CRDC does not have data on all schools in the study 17. Charter schools are all schools that are sample, which was from 2016- classified by CRDC as charter schools. IDEA students refer to students in the Individuals with Disabilities Education Act. IEP students refers to students that have an Individualized Education Program covered by IDEA. Charter schools may be alternative, magnet, or special education schools. School proficiency scores show the percentage of students scoring at or above proficiency on state assessments across grades and subject as reported by Great Schools in their mo st recent dataset from 2016. Value -added measure (VAM) is constructed by calculating the normalized difference between the observed proficiency and the predicted proficiency from a regression including school- level and tract -level controls. VAM measures ca n only be estimated for the sample. 46

48 Table A 7 - Effect of Message Treatments on Response Rates by Different Randomized Characteristics One parent Son Daughter Black Hispanic Sample Two parent 0.031 - 0.023* - 0.023 - 0.023 - - 0.011 - 0.035* Bad Grades (0.012) (0.017) (0.028) (0.017) (0.020) (0.021) - 0.049* - 0.052*** - 0.051*** - 0.053*** - 0.053*** 0.038** IEP - (0.026) (0.011) (0.016) (0.015) (0.018) (0.018) - 0.025 - 0.080*** - 0.058*** - 0.084*** - 0.091*** - 0.061*** Bad Behavior (0.020) (0.027) (0.017) (0.017) (0.021) (0.013) - 0.015 0.029 - 0.012 - 0.027 0.058** 0.047* Good Grades (0.022) (0.021) (0.018) (0.027) (0.025) (0.026) 0.022 - 0.021* - 0.023* - 0.001 Black 0.000 (0.014) (0.011) (0.014) (0.022) (0.000) 0.002 - 0.025** - 0.024* - 0.018 0.000 Hispanic - (0.022) (0.011) (0.014) (0.014) (0.000) 0.011 0.001 0.007 - 0.001 0.017 0.010 Father (0.011) (0.011) (0.009) (0.013) (0.014) (0.018) 0.014 - 0.017** 0.010 - 0.022* Son - (0.013) (0.017) (0.014) (0.008) 0.005 0.025 - Two Parents 0.033* 0.004 (0.016) (0.016) (0.019) (0.020) TPS 0.003 0.010 0.003 0.012 0.006 0.007 (0.019) (0.015) (0.015) (0.015) (0.020) (0.019) 7,431 Observations 7,375 2,916 4,999 4,916 11,890 R - squared 0.052 0.041 0.039 0.042 0.044 0.044 0.551 Control Group Mean 0.535 0.532 0.522 0.546 0.523 Notes: Table shows the effect of different treatments on response rate for different samples. TPS is a covariate for traditional . All other variables included in the table are randomly assigned characteristics of the public school emails. Regressions include fixed effects by wave and states. Sample row shows the population for each regression identified in the e -mails: (1) two paren ts, (2) one parent, (3) student is a son, (4) students is a daughter, (5) black -sounding name, and (6) Hispanic -sounding name . Two -way c luster -robust s tandard errors (by pair and school) in parentheses. *** p<0.01, ** p<0.05, * p<0.1 47

49 Appendix B. Collection and Implementation Data First field experiment This subsection describes data collection for the first field experiment. Data collection of school contact information took place between August and October 2014. Email and web -form contact informa tion was collected for charter schools from the 17 states with the most charter schools. Once these states were identified we used the U.S. Department of Education’s Common Core of Data to determine which public schools identify as charter schools and limited our sample to these schools. The next step involved collecting contact information. To the extent possible, we visited charter school websites to obtain contact information. Sometimes websites were not easily located, so we used databases maintained by state Departments of Education. We emphasized the collection of contact information from school websites, because this is the likely contact information a parent would use if emailing a school. For both first and second field experiments, we prioritized the type of contact information collected. If there is an email address or webform that is used to field general inquiries, then we used this. Otherwise, we would identify a front office receptionist or office manager. If this was not available, then we wou ld look for collect the contact information for one of the school principals. Second field experiment This describes how we collected data on schools for the second field experiment for the school choice audit study. Data collection for the field experiment took place between Nov. 2017 and Jan 2018. We worked with research assistants to identify school districts across states that practice some -district school choice and that also have charter schools within their respective type of intra catchment areas. In identifying these school districts, we first focused on whether intra- district choice is practiced. We used geographic information systems (GIS) data from U.S. Census Bureau to identify local school district catchment areas. We also used latitude and longitude data on each school from the U.S. Department of Education’s Common Core of Data to identify wh ether charter schools are geographically situated in school districts that practice intra -district choice. One challenge is that there is a varia in how intra -district choice is practiced across school tion districts within states. These practices helped gu ide the selection of catchment areas to include in our sample. Open enrollment represents one end of the spectrum. This type of intra -district school choice is well- studied. See, for example, the research conducted by David Deming on Charlotte - district choice involving schools. Bergman (2016) investigates another type of inter- Mecklenburg transfers in California schools. Typically, open enrollment and transfer policies are well known and 48

50 formal application processes have been established. Less known are intra -district choice policies that require some form of administrative approval. For example, a superintendent may need to or, within a specific grant approval before parents can send a child to another school in a district, and receiving schools may have to agree upon a child moving from district, principals from sending one school to another. These types of intra -district choice might be viewed as more restrictive than -district choice, enrollment policies. For school districts that place strong restrictions on intra open- it may not be viewed as unreasonable that teachers, receptionists, front desk and office managers may be neither aware of the policy nor its eligibility requirements. restricted our attention to states with the most charter schools s California, Texas, We such a Florida, Arizona, Ohio, Michigan, New York, Nevada, and the District of Columbia. For each of school districts in terms of student enrollments. Some these states, we then identified the top 40 such as Arizona , have mandatory un iform intra -district school choice policies. Uniformity states, district school choice laws simplified the process of selecting school in implementation of intra- such as Tennessee , allow intra - district geographic areas to include in our sample. Other states, district choice but it is only practiced among its largest school district located in Nashville. This NCES table identifies more than 25 different types of policies across states with open enro llment 26 policies. For each state, our research assistants visited school districts’ websites to learn about the - ir intra Once we determined that a school district within a state practices intra district enrollment policy. - district choice, we made the decision to include that school district’s geographic area in our sampling frame (provided there are charter schools within the catchment area). 2015 school year. Although these are not the latest files from the The NCES data used are for the 2014- the most recent files that Common Core of Data, at the time we defined the sampling frame, they were are complete. For each charter school, research assistants found the nearest traditional public school. We matched the charter school to the nearest traditional public based on the type of school (i.e., regular, special education, etc.) the entry grade level. and We kept charter schools in our sample (and their respective matched pair) if the charter school enrollment was greater than 200 students . We also limited the number of schools we attempted to contact in a single school district 80. 26 e 4.2: “Numbers and types of open enrollment policies, by state: 2017.” See Tabl 49

Related documents