Abstract

Educational accountability policies are a popular tool to close the achievement gaps between advantaged and disadvantaged students. However, these policies may exacerbate inequality if families from advantaged backgrounds are better able to advocate for their children and thus circumvent policy. We investigate this possibility in the context of the early grade retention policy in Florida, which requires all students with reading skills below grade level to be retained in the third grade, yet grants exemptions under special circumstances. We find that Florida's third-grade retention policy is in fact enforced differentially depending on children's socioeconomic background, especially maternal education. Holding exemption eligibility constant, scoring right below the promotion cutoff results in an increase in the probability of retention that is 14 percent greater for children whose mothers have less than a high school degree compared with children whose mothers have a bachelor's degree or more. We also find that the discrepancies in retention rates are mainly driven by the fact that students with well-educated mothers are more likely to be promoted based on subjective exemptions, such as teacher portfolios.

1.  Introduction

Research consistently demonstrates a strong, positive relationship between parents’ socioeconomic status (SES) and children's educational achievement. Whereas the last half-century has seen a slow narrowing of the achievement gap between black and white students, the gap between high- and low-SES students has remained persistent, actually increasing by approximately 40 percent over the last thirty years (Reardon 2011). This achievement gap is already present when children enter school in kindergarten and, despite the numerous policies aimed at leveling the educational playing field for disadvantaged students, it does not dissipate as children progress in their schooling (Duncan and Magnuson 2011).

Universal educational policies are a popular tool to address inequalities, with the underlying belief being that disparities can be overcome by holding all students to the same high standards and ensuring that all families have access to the same opportunities. However, these policies may be ineffective—and may actually exacerbate inequality—if high-SES families are better able to advocate for their children, make informed decisions, circumvent policy, or take advantage of opportunities in their children's schooling.1 We investigate this possibility in the context of a statewide grade retention policy aimed at ensuring that all students enter the fourth grade proficient in reading.

There is mounting evidence across social science disciplines that parents’ behavior regarding their child's schooling does in fact differ depending on SES. Lower-SES parents have been found to be less likely to request a specific teacher (Jacob and Lefgren 2005), challenge their child's placement into a lower curriculum track (Barg 2012), and question the pedagogical authority of their child's teacher during parent–teacher conferences (Weininger and Lareau 2003). Ethnographic work by Lareau and Calarco (2012) found that compared with lower-class parents, middle-class parents had greater knowledge of their child's school environment and experiences and were aware of a much wider variety of opportunities for intervention in their child's schooling. Where middle-class parents “approached interactions with the school as an ongoing negotiation” (Lareau and Calarco 2012, p. 74), lower-class parents rarely asked for any educational modifications, even when they felt their child might benefit from one. Furthermore, on the rare occasions that lower-class parents did try to engage the school, they often used less successful strategies, approaching school staff in an angry, confrontational manner, whereas middle-class parents were more apt to calmly but firmly try to engage school personnel in a partnership.

These differences in parents’ behavior can amount to real impacts on the effectiveness of educational policy and have important educational consequences for children. One prominent example is school choice. Heralded as a mechanism to level the playing field between children in different neighborhoods, school choice policies allow children living in neighborhoods with poorer performing schools to have options beyond their neighborhood school. The evidence suggests, however, that less-educated and lower-income parents respond differently than middle-class parents when presented with a choice among schools, resulting in increased segregation by SES without improved academic performance (Fiske and Ladd 2000; Cullen et al. 2005; Hsieh and Urquiola 2006). Hastings, Kane, and Staiger (2006a) specifically find that the preference for school's mean test score increases with parent's income while preference for proximity decreases, resulting in two distinct types of parents—those with a preference for test scores regardless of proximity (who are more likely to be higher income), and those with a preference for proximity regardless of test scores (who are more likely to be lower income). These differences in preferences among parents translate into differences in academic achievement by SES. Children of parents who place high weights on academics experience academic gains when randomly assigned to their first-choice school, whereas children of parents who place a low weight on academics experienced academic losses (Hastings, Kane, and Staiger 2006b).

Despite the potential for socioeconomic differences in parental knowledge, preferences, and behaviors to exacerbate inequality, there is little large-scale empirical evidence on the impact of parents’ SES in the face of a broad policy that is intended to be enforced universally. The current examples from the literature have consisted of situations that either require an active choice by parents, in the case of school choice, or are cases where parents can choose to intervene regarding discretionary school decisions not based on formal policy. This paper builds upon the prior research into SES, parental behavior, and academic achievement by exploring the idea that through these socioeconomic differences in behavior, seemingly universal educational policies may be differentially enforced for students of different backgrounds.

We examine this question in the context of a statewide grade retention policy enacted by the Florida state legislature in 2002. The Florida policy mandated that, in the absence of a specific exemption, promotion from the third grade to the fourth grade would be conditional upon meeting a minimum standard of reading. For a number of reasons, this policy presents an ideal opportunity to study the importance of family SES in the face of a broad universal policy. In particular, although the policy allowed for exemptions in order to provide schools with flexibility in cases where there were extenuating circumstances that rendered retention inappropriate, a large proportion of students—over 40 percent—were granted an exemption. By allowing exemptions to the rule, this program provides a natural measure of differential enforcement: whether children were more likely to be granted an exemption based upon their socioeconomic background.

The Florida setting is particularly advantageous because we have had the unusual opportunity to link educational records to birth record data.2 Doing so is essential in order to deeply investigate the role of SES in the implementation of a universal educational policy. Administrative data in education are limited to measures of race and free or reduced-price lunch (FRPL) status and do not include other types of background or parental characteristics. National longitudinal data, although they contain a broad range of background information, do not have sufficient numbers of observations, even if they happen to be timed in such a way as to observe children when a particular policy is implemented. Linking educational records to birth records, however, allows us to examine the impact of maternal characteristics including education, marital status, place of birth, and age, on the implementation of the Florida retention policy.

Furthermore, because the Florida policy relies on a strict score cutoff for determining retention, we are also able to leverage a regression discontinuity design to conduct a difference-in-difference estimation of socioeconomic differences in policy implementation for the marginal student. This approach allows us to difference out any retention rate differences across students from different socioeconomic backgrounds who just make the cutoff and are therefore not affected by the retention policy. These differences might arise due to inherent biases teachers or school personnel might have, which would cause them to differentially retain students based on SES in the absence of the policy, or due to differences in reading skills that are not captured by test scores between high- and low-SES students.

We focus our analyses on differences in implementation using maternal education as the defining indicator of SES while exploring implementation differences by other indicators as well. We focus on maternal education because it has been found to be the strongest predictor of children's academic achievement (Haveman and Wolfe 1995) and we believe that the mechanisms that would likely lead to differential exemption—namely, having the knowledge, agency, and desire to intervene in the policy's implementation—are most likely to be impacted by maternal educational attainment.

We find that Florida's third-grade retention policy is in fact enforced differentially depending on children's socioeconomic background, especially maternal education. Holding exemption eligibility constant, scoring right below the promotion cutoff results in an increase in retention probability that is 7 percentage points larger for children whose mothers have less than a high school degree as compared to children whose mothers have a bachelor's degree or more, leading to a retention rate that is 14 percent higher for the former group right below the cutoff. Smaller increases in retention probability are associated with having a foreign-born mother and qualifying for FRPL. We also find the discrepancies in retention rates are mainly driven by the fact that students with well-educated mothers are more likely to receive subjective exemptions, such as teacher recommendations for advancement (officially known as portfolios). These socioeconomic disparities are robust to a variety of different functional forms and bandwidths and remain stable even when comparing children only to other children within the same school.

2.  Policy Background

In 2002, as part of an increasingly popular nationwide movement toward early grade retention to ensure reading proficiency that has only gained additional traction in the years since Rose and Schmike (2012), the Florida legislature mandated that third-grade students meet the Level 2 benchmark or higher (the second lowest of five levels) on the Florida Comprehensive Assessment Test (FCAT) reading exam in order to be promoted to the fourth grade.3 The focus on third-grade reading scores highlights the belief among educators that it is at this time when reading proficiency becomes crucial for success across subjects and children transition from “learning to read” to “reading to learn.” Students who do not score at a Level 2 or higher and do not obtain an exemption are subject to retention in conjunction with a number of other interventions intended to ensure that they are able to be promoted the following year. Retained students must also be assigned to a high-performing teacher, receive intensive reading instruction during their retained year, and be given the opportunity to attend a summer reading program prior to the next school year.4

There are a number of “good cause exemptions” that allow students to be promoted to the fourth grade despite failing to score at the Level 2 benchmark or above. Students are eligible for an exemption if they have limited English proficiency and have received fewer than two years of instruction in an English for Speakers of Other Languages program, have certain disabilities, or have received intensive reading remediation for two years and have already been retained twice between kindergarten and third grade. Additionally, students are able to obtain an exemption by demonstrating they are reading at a level equal to or above a Level 2 on the FCAT by performing at an acceptable level on an alternative standardized reading assessment approved by the State Board of Education (e.g., 51st percentile or above on the Stanford 10 reading exam,5 a level of proficiency much higher than that needed to meet a Level 2 on the FCAT), or by demonstrating proficiency through a teacher-developed portfolio.6 A key feature of the good cause exemption policy that might lead to differential retention rates by SES is that parents need to contact their child's teacher or principal if they believe their child is eligible for a good cause exemption.7 Therefore, holding exemption eligibility constant, students from upper-class families might be less likely to be retained (compared with students from disadvantaged backgrounds) if their parents are more likely to advocate for their children's eligibility (as they are better informed about the retention policy) or are more likely to succeed in their efforts (as they have closer relationships with the teachers and/or the principals or use more successful negotiation strategies).8

In spite of the intended benefits of the program and the extra services that retained children receive, there are numerous reasons why parents may prefer not to have their child retained. Perhaps most importantly, there is mixed evidence that retaining a child improves her academic outcomes. The overwhelming majority of the earlier literature on grade retention found that students perform significantly worse than their promoted peers in the years following their retention. Holmes (1989) and Jimerson (1999) provide high-quality meta-analysis of earlier research on grade retention. However, these studies were unable to account for omitted variables bias, making it likely that some of the perceived negative impact of grade retention was due not to retention itself but instead to the fact that students who were retained were more likely to perform poorly in the future through some other factors about themselves that also influenced the decision to retain them in the first place. More recent work using quasi-experimental designs to better address these identification challenges have found that, especially in early grades, retention may have a positive impact on test scores in the short term (Jacob and Lefgren 2005, 2009; Roderick and Nagaoka 2005; Greene and Winters 2007, 2012; Schwerdt, West, and Winters 2015). Exploiting a nonlinearity in the relationship between test scores and the probability of retention for third-, sixth-, and eighth-grade students in Chicago, Jacob and Lefgren (2005, 2009) find that retention and mandatory summer school had a small positive effect on achievement in the short term for third graders but not for sixth graders. They also find that retention had a negative impact on high school graduation rates for eighth graders but not sixth graders. Recent research specifically on Florida's retention policy found that although retained students outperformed their same-age peers in the short term, after six years these achievement gains faded out entirely. Retained students were, however, less likely to be retained in a later grade and were no more or less likely to graduate from high school (Schwerdt, West, and Winters 2015).

Parents may also be concerned about the effects that grade retention will have on their child's social and emotional well-being. They may worry about their child having to adjust to a new peer group or being stigmatized by teachers and peers. Parents may also be concerned that being retained will make their children view themselves and their abilities more negatively, harming their self-esteem and decreasing their school engagement. These objections are likely to be particularly pronounced for middle-class parents who have greater personal resources at their disposal to assist their child at home and who are more likely to uncover and better able to process research on the impacts of grade retention. Furthermore, they may view grade retention as particularly stigmatizing due to its extreme uncommonness among the children of parents in their social class.9

3.  Data and Empirical Strategy

Data

The data for our analyses are drawn from two sources. The first are natality data provided by the Florida Department of Health. These data cover the universe of births in the state of Florida between the years 1992 and 2002. At the time of each birth the mother and her health care provider complete a survey that covers maternal demographic information, pregnancy behaviors, and infant health at birth. The data report the mother's age, years of education, race, place of birth, place of residence, and marital status; behaviors during pregnancy such as tobacco and alcohol usage and prenatal care; information on prior births (if any); and information on birth outcomes.10

These data are then matched (79 percent matched)11 to educational data containing information on all Florida students attending public schools from the 2000–01 through 2007–08 school years.12 The educational data include information on the school the child attended, his teachers, student characteristics such as ethnicity, gender, FRPL eligibility, special education classification, English proficiency, and FCAT reading and math scores. Unlike other studies relying solely on educational records, the matching of these data to birth records gives us a unique opportunity to explore whether maternal education and other socioeconomic characteristics, including mother's marital status, age, and country of origin, impact children's likelihood of receiving a retention exemption.13 In our main analysis, we utilize six cohorts of students entering third grade for the first time between 2002 and 2007 (712,231 unique students), all of whom were subjected to the retention policy. In additional analyses, we compare students entering third grade for the first time during the two years prior to the retention policy (2000 and 2001) to those after. The total number of third graders in our population over the eight cohorts is 879,328.14

Table 1 shows the proportion of students who failed to reach Level 2 proficiency on the third-grade FCAT reading exam each year, the proportion retained, and the proportion retained separately for those who scored below versus above the cut point. During the first year of the policy, the proportion of students retained increased from 3.36 percent to 15.04 percent. Among students who scored below the promotion cutoff, the retention percentage grew from 11.16 percent to 67.13 percent. Over the six years of policy implementation for which we have data, the proportion of students retained dropped from a high of 15.04 percent the first year to a low of 6.75 percent during 2005, and has increased slightly since. Some of the decrease in retention stems from a decrease in the proportion of students scoring below the cutoff (a decrease from over 21 percent to between 12 percent and 16 percent for many of the later cohorts). The remainder is due to a decrease in the proportion of failing students who are retained, primarily driven by the increase in the percentage of failing students receiving an exemption. In the first year of the policy, more than 67 percent of students scoring below the cutoff were retained. That proportion has remained below 60 percent for all subsequent cohorts and below 50 percent for two cohorts. During the same time frame, the percentage of students receiving an exemption increased for all exemption categories, more than doubling from 26 percent to 54 percent.

Table 1.
Summary Statistics by Cohort-Year
Before PolicyAfter Policy
20002001200220032004200520062007
% Below 23.72 23.80 21.17 18.59 16.90 11.73 16.21 13.79 
% Retained 3.38 3.36 15.04 11.56 9.87 6.75 7.89 7.33 
% of Below retained 10.39 11.16 67.13 58.27 54.77 51.83 45.24 48.64 
% Above retained 1.20 0.92 1.05 0.90 0.74 0.76 0.67 0.73 
% of Below received         
Limited English Proficiency exemption NA NA 1.61 0.94 0.92 1.31 0.83 0.81 
Special education exemption NA NA 8.54 15.03 17.55 22.12 19.09 20.50 
Alternative test exemption NA NA 7.01 9.01 9.37 6.93 11.91 12.77 
Teacher portfolio exemption NA NA 3.50 8.20 10.00 8.87 14.57 12.90 
Twice retained exemption NA NA 0.57 1.20 1.36 1.53 1.78 1.76 
N 59,849 107,248 117,695 113,558 114,867 119,366 121,623 125,122 
Before PolicyAfter Policy
20002001200220032004200520062007
% Below 23.72 23.80 21.17 18.59 16.90 11.73 16.21 13.79 
% Retained 3.38 3.36 15.04 11.56 9.87 6.75 7.89 7.33 
% of Below retained 10.39 11.16 67.13 58.27 54.77 51.83 45.24 48.64 
% Above retained 1.20 0.92 1.05 0.90 0.74 0.76 0.67 0.73 
% of Below received         
Limited English Proficiency exemption NA NA 1.61 0.94 0.92 1.31 0.83 0.81 
Special education exemption NA NA 8.54 15.03 17.55 22.12 19.09 20.50 
Alternative test exemption NA NA 7.01 9.01 9.37 6.93 11.91 12.77 
Teacher portfolio exemption NA NA 3.50 8.20 10.00 8.87 14.57 12.90 
Twice retained exemption NA NA 0.57 1.20 1.36 1.53 1.78 1.76 
N 59,849 107,248 117,695 113,558 114,867 119,366 121,623 125,122 

The proportion of students scoring a Level 1 on the FCAT reading exam differs dramatically by SES (table 2). For instance, during the two years prior to the implementation of the policy, nearly 39 percent of children whose mothers had less than a high school degree scored at this level, less than 8 percent of children whose mothers had a bachelor's degree or more did. These percentages have dropped to 29 percent and 5 percent, respectively, over the eight years since the retention policy was introduced. Similar patterns emerge when we compare students based on FRPL eligibility, mothers’ marital status, and place of birth, even though the differences are substantially smaller along the last dimension. There are significant discrepancies along racial/ethnic lines as well, with 30 percent of black students scoring below the retention cutoff in the post-policy years, compared with 10 percent of white students.

Table 2.
Summary Statistics by Socioeconomic Status: Before Policy vs. After
Before PolicyAfter Policy
% Below% Retained% of Below Retained% of Above Retained% Below% Retained% of Below Retained% of Above Retained
Maternal education less than high school 38.65 6.23 12.98 1.98 28.84 15.13 49.30 1.28 
Maternal education high school diploma 24.56 3.35 10.37 1.07 15.13 8.97 47.47 0.76 
Maternal education some college 16.05 1.91 8.37 0.68 10.40 4.93 43.45 0.46 
Maternal education bachelors+ 7.72 0.74 6.11 0.29 4.53 1.92 38.36 0.18 
FRPL eligible 35.15 5.14 11.68 1.59 24.78 12.88 48.97 0.99 
FRPL ineligible 12.95 1.68 8.84 0.62 7.86 3.66 41.53 0.43 
Mother U.S.-born 23.03 3.50 11.50 1.11 16.23 8.10 45.85 0.78 
Mother foreign-born 26.44 2.88 8.95 0.69 17.71 9.40 51.26 0.39 
Mother married 17.61 2.35 9.75 0.77 11.58 5.60 44.89 0.45 
Mother not married 35.18 5.25 11.95 1.61 24.75 12.84 49.12 0.90 
White 14.80 2.45 11.03 0.96 9.94 4.67 41.66 0.58 
Black 40.59 5.70 11.91 1.45 29.23 15.24 49.93 0.91 
Hispanic 27.10 2.87 8.44 0.80 19.40 10.07 50.34 0.66 
Before PolicyAfter Policy
% Below% Retained% of Below Retained% of Above Retained% Below% Retained% of Below Retained% of Above Retained
Maternal education less than high school 38.65 6.23 12.98 1.98 28.84 15.13 49.30 1.28 
Maternal education high school diploma 24.56 3.35 10.37 1.07 15.13 8.97 47.47 0.76 
Maternal education some college 16.05 1.91 8.37 0.68 10.40 4.93 43.45 0.46 
Maternal education bachelors+ 7.72 0.74 6.11 0.29 4.53 1.92 38.36 0.18 
FRPL eligible 35.15 5.14 11.68 1.59 24.78 12.88 48.97 0.99 
FRPL ineligible 12.95 1.68 8.84 0.62 7.86 3.66 41.53 0.43 
Mother U.S.-born 23.03 3.50 11.50 1.11 16.23 8.10 45.85 0.78 
Mother foreign-born 26.44 2.88 8.95 0.69 17.71 9.40 51.26 0.39 
Mother married 17.61 2.35 9.75 0.77 11.58 5.60 44.89 0.45 
Mother not married 35.18 5.25 11.95 1.61 24.75 12.84 49.12 0.90 
White 14.80 2.45 11.03 0.96 9.94 4.67 41.66 0.58 
Black 40.59 5.70 11.91 1.45 29.23 15.24 49.93 0.91 
Hispanic 27.10 2.87 8.44 0.80 19.40 10.07 50.34 0.66 

Note: FRPL: free or reduced-price lunch.

More interestingly, retention probabilities differ markedly along SES conditional on scoring below the retention cutoff. For instance, after the policy was enacted, failing students with the most educated mothers are 11 percentage points (roughly 25 percent) less likely to be retained than failing students with the least educated mothers (38.4 percent for the most educated versus 49.3 for the least educated). These retention gaps are also observed between FRPL-eligible and ineligible students (49 versus 41.5 percent), between students with foreign-born mothers and students with U.S.-born mothers (51.3 versus 45.9 percent), between students with married mothers and unmarried mothers (45 versus 49.1 percent), and across racial/ethnic groups (41.7 percent for white students versus 50 percent for black and Hispanic students). However, these observed differences do not necessarily imply that high-SES parents are circumventing the retention policy; instead, they may reflect the actual differences in exemption eligibility.

Table 3 presents the descriptive statistics for our entire regression discontinuity sample (2002–07 cohorts) in the first column, retained students who scored below the cutoff during the policy period (second column), and promoted students who scored below the cutoff during the policy period (third column). Conditional on scoring below the promotion cutoff, promoted students are more advantaged than students who were retained. Compared with retained students, promoted students were older, less likely to be FRPL eligible, more likely to be white, more likely to have been born to a married mother, less likely to have a foreign-born mother, and more likely to have a mother who had at least some college education at the time of their birth.

Table 3.
Descriptive Statistics
AllBelow Cutoff: RetainedBelow Cutoff: Promoted
Maternal education less than high school 0.236 0.428 0.394 
 (0.425) (0.495) (0.489) 
Maternal education high school 0.382 0.406 0.403 
 (0.486) (0.491) (0.490) 
Maternal education some college 0.230 0.132 0.154 
 (0.421) (0.339) (0.361) 
Maternal education bachelor's degree or more 0.153 0.0338 0.0487 
 (0.360) (0.181) (0.215) 
Limited English proficient 0.0439 0.115 0.0985 
 (0.205) (0.320) (0.298) 
Special education student 0.155 0.305 0.485 
 (0.362) (0.460) (0.500) 
FRPL eligible 0.516 0.799 0.746 
 (0.500) (0.401) (0.435) 
Mother foreign-born 0.244 0.283 0.241 
 (0.430) (0.450) (0.428) 
Mother married 0.617 0.408 0.450 
 (0.486) (0.492) (0.498) 
Mother's age at birth 27.087 25.337 25.486 
 (6.222) (6.154) (6.118) 
Black 0.249 0.463 0.416 
 (0.433) (0.499) (0.493) 
White 0.508 0.267 0.336 
 (0.500) (0.443) (0.472) 
Hispanic 0.191 0.238 0.211 
 (0.393) (0.426) (0.408) 
Other race/ethnicity 0.052 0.032 0.037 
 (0.221) (0.176) (0.188) 
Male 0.504 0.579 0.586 
 (0.500) (0.494) (0.492) 
Relative age (in months) −0.101 0.292 4.001 
 (5.677) (6.043) (4.798) 
3rd-grade FCAT reading scale score 311.411 212.378 219.041 
 (61.02) (39.91) (39.75) 
3rd-grade FCAT math scale score 323.422 234.043 252.142 
 (66.45) (57.69) (60.60) 
3rd-grade SAT10 reading percentile rank 59.04 17.73 22.27 
 (27.49) (10.80) (15.21) 
N 712,231 55,864 62,333 
AllBelow Cutoff: RetainedBelow Cutoff: Promoted
Maternal education less than high school 0.236 0.428 0.394 
 (0.425) (0.495) (0.489) 
Maternal education high school 0.382 0.406 0.403 
 (0.486) (0.491) (0.490) 
Maternal education some college 0.230 0.132 0.154 
 (0.421) (0.339) (0.361) 
Maternal education bachelor's degree or more 0.153 0.0338 0.0487 
 (0.360) (0.181) (0.215) 
Limited English proficient 0.0439 0.115 0.0985 
 (0.205) (0.320) (0.298) 
Special education student 0.155 0.305 0.485 
 (0.362) (0.460) (0.500) 
FRPL eligible 0.516 0.799 0.746 
 (0.500) (0.401) (0.435) 
Mother foreign-born 0.244 0.283 0.241 
 (0.430) (0.450) (0.428) 
Mother married 0.617 0.408 0.450 
 (0.486) (0.492) (0.498) 
Mother's age at birth 27.087 25.337 25.486 
 (6.222) (6.154) (6.118) 
Black 0.249 0.463 0.416 
 (0.433) (0.499) (0.493) 
White 0.508 0.267 0.336 
 (0.500) (0.443) (0.472) 
Hispanic 0.191 0.238 0.211 
 (0.393) (0.426) (0.408) 
Other race/ethnicity 0.052 0.032 0.037 
 (0.221) (0.176) (0.188) 
Male 0.504 0.579 0.586 
 (0.500) (0.494) (0.492) 
Relative age (in months) −0.101 0.292 4.001 
 (5.677) (6.043) (4.798) 
3rd-grade FCAT reading scale score 311.411 212.378 219.041 
 (61.02) (39.91) (39.75) 
3rd-grade FCAT math scale score 323.422 234.043 252.142 
 (66.45) (57.69) (60.60) 
3rd-grade SAT10 reading percentile rank 59.04 17.73 22.27 
 (27.49) (10.80) (15.21) 
N 712,231 55,864 62,333 

Notes: Standard deviations are in parentheses. FRPL: free or reduced-price lunch; FCAT: Florida Comprehensive Assessment Test; SAT10: Stanford Aptitude Test.

These patterns all hold when all demographics are included simultaneously in an ordinary least squares regression (table 4, column 1). Once we condition on students’ standardized test performance, many of these associations decrease substantially (table 4, column 2). This is because the further a student's score is below the cutoff, the more likely she is to be retained, and because less advantaged students have lower scores on average, not only are they more likely to score below the promotion cutoff, they are also more likely to score below the cutoff by a greater margin. Yet even accounting for test scores, the association between maternal education and retention remains substantial; children of mothers with a bachelor's degree or more who score below the promotion cutoff remain 6 percentage points less likely to be retained than children of mothers with less than a high school degree, even after controlling for differences in achievement.

Table 4.
Association Between Student Background Characteristics and the Likelihood of Being Retained in the Face of Failing to Meet Promotion Cutoff
(1)(2)
Maternal education high school −0.034*** −0.024*** 
 (0.003) (0.003) 
Maternal education some college −0.076*** −0.056*** 
 (0.005) (0.005) 
Maternal education bachelor's degree or more −0.122*** −0.086*** 
 (0.008) (0.008) 
Free or reduced-price lunch eligible 0.047*** 0.025*** 
 (0.004) (0.004) 
Mother foreign-born 0.022*** 0.028*** 
 (0.005) (0.005) 
English not native −0.018*** −0.010*** 
 (0.003) (0.003) 
Mother married 0.001*** 0.002*** 
 (0.0003) (0.0003) 
Mother's age at birth 0.020*** −0.012*** 
 (0.004) (0.004) 
Black 0.020*** 0.010* 
 (0.006) (0.005) 
Hispanic −0.020*** −0.041*** 
 (0.006) (0.006) 
Limited English proficient −0.120*** −0.196*** 
 (0.004) (0.004) 
Special education student −0.017*** −0.018*** 
 (0.0002) (0.0002) 
Relative age (in months) 0.036*** 0.028*** 
 (0.003) (0.003) 
Male  −0.029*** 
  (0.002) 
3rd-grade FCAT math scale score  −0.007*** 
  (0.0001) 
3rd-grade SAT10 reading percentile rank  −0.002*** 
  (0.0001) 
3rd-grade SAT10 math percentile rank  −0.024*** 
  (0.003) 
N 117,148 114,457 
(1)(2)
Maternal education high school −0.034*** −0.024*** 
 (0.003) (0.003) 
Maternal education some college −0.076*** −0.056*** 
 (0.005) (0.005) 
Maternal education bachelor's degree or more −0.122*** −0.086*** 
 (0.008) (0.008) 
Free or reduced-price lunch eligible 0.047*** 0.025*** 
 (0.004) (0.004) 
Mother foreign-born 0.022*** 0.028*** 
 (0.005) (0.005) 
English not native −0.018*** −0.010*** 
 (0.003) (0.003) 
Mother married 0.001*** 0.002*** 
 (0.0003) (0.0003) 
Mother's age at birth 0.020*** −0.012*** 
 (0.004) (0.004) 
Black 0.020*** 0.010* 
 (0.006) (0.005) 
Hispanic −0.020*** −0.041*** 
 (0.006) (0.006) 
Limited English proficient −0.120*** −0.196*** 
 (0.004) (0.004) 
Special education student −0.017*** −0.018*** 
 (0.0002) (0.0002) 
Relative age (in months) 0.036*** 0.028*** 
 (0.003) (0.003) 
Male  −0.029*** 
  (0.002) 
3rd-grade FCAT math scale score  −0.007*** 
  (0.0001) 
3rd-grade SAT10 reading percentile rank  −0.002*** 
  (0.0001) 
3rd-grade SAT10 math percentile rank  −0.024*** 
  (0.003) 
N 117,148 114,457 

Notes: Robust standard errors, clustered at the relative reading score level, are given in parentheses. Estimates obtained from ordinary least squares regressions on all students who scored below the promotion cutoff for 2002—07 cohorts. FCAT: Florida Comprehensive Assessment Test; SAT10: Stanford Aptitude Test.

***p < 0.01; *p < 0.1.

Empirical Framework

In order to examine whether Florida's retention policy is differentially enforced depending on student SES, we present both graphical evidence and difference-in-difference estimates of the impact of scoring below the promotion cutoff for students from different socioeconomic backgrounds, subtracting out differences in retention probabilities between socioeconomics groups above the promotion cutoff. As in traditional regression discontinuity designs, whether a student scores just below or just above the cutoff is close to random, and thus students just below and just above the cutoff should not systematically differ from one another. This allows us to look at differences in the impact of scoring just below the promotion cutoff on retention, with differences between groups just above the promotion cutoff serving as a counterfactual for what we would expect to see in the absence of the policy. We estimate these impacts with the following equation:
Ri=φ+δFi+k(Si)+k(Si)*Fi+γSESi+λSESi*Fi+k(Si)*SESi+k(Si)*Fi*SESi+ξELIGi+ψELIGi*Fi+k(Si)*ELIGi+k(Si)*Fi*ELIGi+αXi+βXi*Fi+k(Si)*Xi+k(Si)*Fi*Xi+Year*School+νi,
(1)
where Ri is the probability of retention for student i; Fi is an indicator for failing to meet the promotion cutoff, k(Si) is a polynomial function of the relative reading score; SESi is the vector of SES measures including maternal education, place of birth, age and marital status at birth, student's FRPL eligibility, and race/ethnicity; ELIGi is a vector of student exemption eligibility measures (i.e., limited English proficiency and special education status, student test scores in alternative tests, student age at the time she first enters the third grade compared with her peers), Xi is a vector of other student characteristics including gender, and νi is an error term.15 By including student exemption eligibility measures, which are then interacted with scoring below the promotion cutoff, we focus on differences in retention probability taking into account any differences in the proportion of students who fall into one of these formal exemption categories. Using this framework, our main parameter of interest, λ, presents the percentage point difference in the jump in probability of retention at the discontinuity for students from high-SES families, as compared to the rate for students from low-SES families—the difference-in-difference estimate for each of the SES measures in which we are interested. This specification also includes school-by-year fixed effects to take into account differences in retention trends within schools over time.

There are several strategies for specifying the underlying functional form when relying on a regression discontinuity design, the first is to estimate the equation nonparametrically using kernel-weighted local polynomial smoothing as proposed by Hahn, Todd, and Van der Klaauw (2001) and later developed by Porter (2003) to include higher-order polynomial estimators. Nonparametric strategies reduce the possibility of misspecification bias inherent in parametric models. However, when the selection variable is discrete, as in this case, a nonparametric estimator may lead to more biased estimates because it is not possible to compare averages within an arbitrarily small neighborhood around the cutoff (Lee and Card 2008). We therefore follow Lee and Card (2008) and estimate equation 1 parametrically using a linear specification, allowing for differences in slope on either side of the discontinuity, and limiting the analysis to students within a bandwidth of 20 points. We present graphical evidence that this specification appears to fit the data well, and check the robustness of our findings using different bandwidths (5, 10, 50, and all) and polynomial orders (1, 2, and 3). Also following Lee and Card, we cluster standard errors at the relative reading score level. We also present subgroup analysis to explore whether SES has a differential impact on rates of retention by student, teacher, or school characteristics.

4.  Results

Figure 1 presents the local linear smoothing of the probability of retention on the relative reading score for cohorts subjected to the retention policy, calculated separately for each side of the cutoff using the triangle kernel, with the solid circles representing the retention rate for each test score. In the upper panel, we use all possible scores and a bandwidth of 20 points, whereas the lower panel uses only the scores within the 20 points on either side of the cutoff with a bandwidth of 5 points. These figures show that students just below the promotion cutoff are approximately 35 percentage points more likely to be retained than students who score just at the promotion cutoff. Within the 20-point bandwidth, the relationship between relative reading score and retention probability appears linear, with the local linear fit very closely aligned with the linear model, falling within the 95 percent confidence interval for nearly the whole 20-point range. Figure 2 presents a falsification test, replicating the same graphical analysis as figure 1 but for the cohorts entering third grade during the two years before the policy was implemented. It is clear from these figures that the retention rate is approximately the same before and after the policy for students scoring just at or above the promotion cutoff (approximately 5 percent), but there is no discontinuity at the cutoff before the policy. This makes us confident that the jump in retention at the cutoff that is seen after the policy is introduced is in fact due to the retention policy.

Figure 1.

The Relationship Between Third-Grade Reading Scores and Grade Retention—–After Policy

Notes: Based on 2002—07 cohorts. Solid lines represent predicted values from local linear regression, estimated separately for the two sides of the cutoff using a bandwidth of 20 points (bandwidth of 5 points in the lower panel) and the triangle kernel function; the shaded area represents 95% confidence interval.

Figure 1.

The Relationship Between Third-Grade Reading Scores and Grade Retention—–After Policy

Notes: Based on 2002—07 cohorts. Solid lines represent predicted values from local linear regression, estimated separately for the two sides of the cutoff using a bandwidth of 20 points (bandwidth of 5 points in the lower panel) and the triangle kernel function; the shaded area represents 95% confidence interval.

Figure 2.

The Relationship Between Third-Grade Reading Scores and Grade Retention—–Before Policy

Notes: Based on 2000—01 cohorts. Solid lines represent predicted values from local linear regression, estimated separately for the two sides of the cutoff using a bandwidth of 20 points and the triangle kernel function; the shaded area represents 95% confidence interval.

Figure 2.

The Relationship Between Third-Grade Reading Scores and Grade Retention—–Before Policy

Notes: Based on 2000—01 cohorts. Solid lines represent predicted values from local linear regression, estimated separately for the two sides of the cutoff using a bandwidth of 20 points and the triangle kernel function; the shaded area represents 95% confidence interval.

Figures 3 and 4 use local linear regressions estimated separately on each side of the promotion cutoff to depict the relationship between relative reading score and retention around the promotion cutoff for students from different socioeconomic backgrounds after and before the policy implementation, respectively. Figure 3 shows that whereas retention rates above the promotion cutoff appear to be approximately the same regardless of family SES, for students who fail to meet the promotion cutoff there are stark differences in retention rates by SES, especially by maternal education. For instance, across reading scores, students with more highly educated mothers are less likely to be retained, and this relationship appears to be monotonic, with each additional level of maternal education translating to a smaller fraction of students actually being retained in the event that they fail to meet the cutoff for promotion to fourth grade. Figure 4 provides a comparison by looking at the same relationship during the two years before the policy was implemented. Although students from high-SES families are less likely to be retained across the reading score spectrum, there is no significant jump at the retention cutoff, providing evidence that high-SES parents responded differently to the retention policy.

Figure 3.

The Relationship Between Third-Grade Reading Scores and Grade Retention Around the Cutoff by Socioeconomic Status—–After Policy

Notes: Based on 2002—07 cohorts. Lines represent predicted values from local linear regression, estimated separately for the two sides of the cutoff using a bandwidth of 5 points and the triangle kernel function. FRPL: free or reduced-price lunch; HS: high school.

Figure 3.

The Relationship Between Third-Grade Reading Scores and Grade Retention Around the Cutoff by Socioeconomic Status—–After Policy

Notes: Based on 2002—07 cohorts. Lines represent predicted values from local linear regression, estimated separately for the two sides of the cutoff using a bandwidth of 5 points and the triangle kernel function. FRPL: free or reduced-price lunch; HS: high school.

Figure 4.

The Relationship Between Third-Grade Reading Scores and Grade Retention Around the Cutoff by Socioeconomic Status—–Before Policy

Notes: Based on 2000—01 cohorts. Lines represent predicted values from local linear regression, estimated separately for the two sides of the cutoff using a bandwidth of 5 points and the triangle kernel function. FRPL: free or reduced-price lunch; HS: high school.

Figure 4.

The Relationship Between Third-Grade Reading Scores and Grade Retention Around the Cutoff by Socioeconomic Status—–Before Policy

Notes: Based on 2000—01 cohorts. Lines represent predicted values from local linear regression, estimated separately for the two sides of the cutoff using a bandwidth of 5 points and the triangle kernel function. FRPL: free or reduced-price lunch; HS: high school.

Table 5 presents the effect of failing to meet the promotion cutoff for fourth grade on the likelihood of being retained. Note that all estimations are based on our preferred discontinuity sample within a 20 test-score-point bandwidth around the cutoff, using the post-policy years. The first column confirms the graphical analysis presented in figure 1, revealing that the students who scored right below the cutoff were about 30 percentage points more likely to be retained than students who were right above. The second column introduces the SES indicators and their interactions given in equation 1. Similar to the findings presented in figure 3, the difference-in-difference estimates suggest that students from high-SES families (especially those with well-educated mothers) are significantly less likely to be retained in response to just failing to meet the promotion cutoff. The impact of scoring below the cutoff is 1.6 percentage points smaller for students with mothers who have a high school degree, 3.9 percentage points smaller for those whose mothers have some college, and 9.1 percentage points smaller for those whose mothers have a bachelor's degree or more when compared with students whose mothers have less than a high school degree. The impact of falling right below the promotion cutoff also varies significantly by FRPL eligibility (higher by 3.7 percentage points for eligible students), by mothers’ place of birth (3.5 percentage points for students with foreign-born mothers), and by race/ethnicity (2.9 percentage points higher for black students compared with white and Asian students).

Table 5.
Effect of Scoring Below the Promotion Cutoff on Retention in Third Grade
(1)(2)(3)(4)(5)
Below 0.291*** 0.264*** 0.530*** 0.517*** 0.526*** 
 (0.006) (0.014) (0.018) (0.029) (0.033) 
Maternal education high school × Below  −0.016** −0.018** −0.018* −0.023** 
  (0.007) (0.009) (0.010) (0.011) 
Maternal education some college × Below  −0.039*** −0.044*** −0.044*** −0.044*** 
  (0.007) (0.006) (0.014) (0.015) 
Maternal education bachelor's+ × Below  −0.091*** −0.072*** −0.073*** −0.081*** 
  (0.019) (0.015) (0.020) (0.024) 
Mother foreign × Below  0.035*** 0.038*** 0.036*** 0.033** 
  (0.010) (0.008) (0.012) (0.014) 
Mother married × Below  −0.022 −0.009 −0.011 −0.013 
  (0.014) (0.011) (0.010) (0.012) 
Mother's age at birth × Below  −0.001 −0.0003 −0.0001 −0.0001 
  (0.001) (0.0010) (0.001) (0.001) 
Black × Below  0.029** −0.007 −0.008 −0.006 
  (0.011) (0.009) (0.011) (0.012) 
Hispanic × Below  0.010 −0.010 −0.009 −0.004 
  (0.010) (0.010) (0.014) (0.016) 
Free or reduced-price lunch eligible × Below  0.037*** 0.031*** 0.032*** 0.027** 
  (0.011) (0.009) (0.011) (0.012) 
Limited English Proficient × Below   −0.013 −0.020 −0.017 
   (0.012) (0.016) (0.019) 
Special education × Below   −0.095*** −0.093*** −0.094*** 
   (0.008) (0.011) (0.012) 
SAT10 reading > 50th percentile × Below   −0.125*** −0.132*** −0.147*** 
   (0.012) (0.015) (0.018) 
Standardized FCAT math × Below   −0.017** −0.020** −0.016* 
   (0.007) (0.008) (0.010) 
SAT10 reading percentile × Below   −0.005*** −0.005*** −0.005*** 
   (0.0003) (0.0004) (0.001) 
SAT10 math percentile × Below   −0.002*** −0.001*** −0.001*** 
   (0.0002) (0.0002) (0.0003) 
Relative age × Below   −0.007*** −0.007*** −0.006*** 
   (0.001) (0.001) (0.001) 
School fixed effects No No No Yes No 
Teacher fixed effects No No No No Yes 
N 107,874 107,860 105,896 105,896 105,896 
(1)(2)(3)(4)(5)
Below 0.291*** 0.264*** 0.530*** 0.517*** 0.526*** 
 (0.006) (0.014) (0.018) (0.029) (0.033) 
Maternal education high school × Below  −0.016** −0.018** −0.018* −0.023** 
  (0.007) (0.009) (0.010) (0.011) 
Maternal education some college × Below  −0.039*** −0.044*** −0.044*** −0.044*** 
  (0.007) (0.006) (0.014) (0.015) 
Maternal education bachelor's+ × Below  −0.091*** −0.072*** −0.073*** −0.081*** 
  (0.019) (0.015) (0.020) (0.024) 
Mother foreign × Below  0.035*** 0.038*** 0.036*** 0.033** 
  (0.010) (0.008) (0.012) (0.014) 
Mother married × Below  −0.022 −0.009 −0.011 −0.013 
  (0.014) (0.011) (0.010) (0.012) 
Mother's age at birth × Below  −0.001 −0.0003 −0.0001 −0.0001 
  (0.001) (0.0010) (0.001) (0.001) 
Black × Below  0.029** −0.007 −0.008 −0.006 
  (0.011) (0.009) (0.011) (0.012) 
Hispanic × Below  0.010 −0.010 −0.009 −0.004 
  (0.010) (0.010) (0.014) (0.016) 
Free or reduced-price lunch eligible × Below  0.037*** 0.031*** 0.032*** 0.027** 
  (0.011) (0.009) (0.011) (0.012) 
Limited English Proficient × Below   −0.013 −0.020 −0.017 
   (0.012) (0.016) (0.019) 
Special education × Below   −0.095*** −0.093*** −0.094*** 
   (0.008) (0.011) (0.012) 
SAT10 reading > 50th percentile × Below   −0.125*** −0.132*** −0.147*** 
   (0.012) (0.015) (0.018) 
Standardized FCAT math × Below   −0.017** −0.020** −0.016* 
   (0.007) (0.008) (0.010) 
SAT10 reading percentile × Below   −0.005*** −0.005*** −0.005*** 
   (0.0003) (0.0004) (0.001) 
SAT10 math percentile × Below   −0.002*** −0.001*** −0.001*** 
   (0.0002) (0.0002) (0.0003) 
Relative age × Below   −0.007*** −0.007*** −0.006*** 
   (0.001) (0.001) (0.001) 
School fixed effects No No No Yes No 
Teacher fixed effects No No No No Yes 
N 107,874 107,860 105,896 105,896 105,896 

Notes: Robust standard errors, clustered at the relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using a degree of 1 and within 20 points of the promotion cutoff, using 2002—07 cohorts. The coefficients on the interaction terms represent the difference in retention discontinuity at the cutoff by the listed student attribute. All regressions include year fixed effects and gender. FCAT: Florida Comprehensive Assessment Test; SAT10: Stanford Aptitude Test.

***p < 0.01; **p < 0.05; *p < 0.1.

There are several ways to interpret these findings. First, these observed differences in the impact of failing the third-grade reading test might indeed reflect the differences in parental behavior, with high-SES parents more actively advocating for their children's promotion or having more success in their efforts. Alternatively, these differences might reflect the actual differences in formal exemption eligibility if, for instance, students from high-SES families are more likely to demonstrate reading proficiency on alternative tests. In the third column of table 5, we consider the latter possibility and control for the aforementioned measures of exemption eligibility. The results remain almost unchanged, with the exception that the differential impact of failing the reading test between black and white/Asian students is completely explained by the differences in exemption eligibility.

Finally, holding exemption eligibility constant, the impact of failing might vary between students from different socioeconomic backgrounds if advantaged students are enrolled in schools or assigned to teachers that are more likely to grant exemption requests for failing students. Column 4 introduces third-grade school fixed effects, and column 5 introduces teacher fixed effects; both columns show that high-SES students remain less likely to repeat the third grade compared with their disadvantaged peers in the same school.16

In table 6, we use the fully interacted school fixed effects model (table 5, column 4) and compare the estimates in the face of the policy to those during the two years before the policy's implementation. If the observed differences in the impact of scoring in the lowest reading achievement level are indeed driven by differential parental responses to the implementation of the retention policy, we would expect no such differences in the years leading to the policy. The pre-policy estimates presented in the first column are in accord with the graphical evidence presented in figure 4 and reinforce the hypothesis that high-SES parents are responding differently to the retention policy.

Table 6.
Effect of Scoring Below the Promotion Cutoff on Retention in Third Grade: Before vs. After Policy
BeforeAfter
Below 0.000 0.517*** 
 (0.019) (0.029) 
Maternal education high school × Below 0.006 −0.018* 
 (0.013) (0.010) 
Maternal education some college × Below −0.017 −0.044*** 
 (0.015) (0.014) 
Maternal education bachelor's+ × Below −0.020 −0.073*** 
 (0.021) (0.020) 
Mother foreign × Below 0.012 0.036*** 
 (0.013) (0.012) 
Mother married × Below 0.009 −0.011 
 (0.012) (0.010) 
Mother's age at birth × Below 0.001 −0.0001 
 (0.001) (0.001) 
Black × Below 0.005 −0.008 
 (0.014) (0.011) 
Hispanic × Below −0.019 −0.009 
 (0.016) (0.014) 
Free or reduced-price lunch eligible × Below 0.006 0.032*** 
 (0.013) (0.011) 
N 32,722 106,758 
BeforeAfter
Below 0.000 0.517*** 
 (0.019) (0.029) 
Maternal education high school × Below 0.006 −0.018* 
 (0.013) (0.010) 
Maternal education some college × Below −0.017 −0.044*** 
 (0.015) (0.014) 
Maternal education bachelor's+ × Below −0.020 −0.073*** 
 (0.021) (0.020) 
Mother foreign × Below 0.012 0.036*** 
 (0.013) (0.012) 
Mother married × Below 0.009 −0.011 
 (0.012) (0.010) 
Mother's age at birth × Below 0.001 −0.0001 
 (0.001) (0.001) 
Black × Below 0.005 −0.008 
 (0.014) (0.011) 
Hispanic × Below −0.019 −0.009 
 (0.016) (0.014) 
Free or reduced-price lunch eligible × Below 0.006 0.032*** 
 (0.013) (0.011) 
N 32,722 106,758 

Notes: Robust standard errors, clustered at the relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using a degree of 1, within 20 points of the promotion cutoff, and including all interacted controls, gender, and school and year fixed effects found in table 5, column 4. The coefficients on the interaction terms represent the difference in retention discontinuity at the cutoff by the listed student attribute, before and after the policy, respectively.

***p < 0.01; *p < 0.1.

The main mechanism that might drive the aforementioned differential impact of failing on retention is the differences in receiving an exemption among students from different socioeconomic backgrounds. To investigate this hypothesis, table 7 replicates the analysis given in table 5, replacing the retention indicator with exemption receipt as the outcome. The findings reveal a strong relationship between SES and the impact of scoring below the promotion cutoff on receiving an exemption, even after controlling for formal exemption eligibility. In particular, just-failing students whose mothers have some college education are 5.1 percentage points, and those whose mothers have a bachelor's degree or more are 6.7 percentage points, more likely to receive an exemption compared with similar students whose mothers have less than a high school degree. The impact of scoring below the cutoff on receiving an exemption is also lower for students with foreign-born mothers and black students. These findings are robust to the inclusion of school fixed effects and indicate that a significant portion of the differences in the impact of failing on retention across socioeconomic groups can be explained by the differences in exemption receipt between these groups.

Table 7.
Effect of Scoring Below the Promotion Cutoff on Receiving Exemption in Third Grade
(1)(2)(3)(4)(5)
Below 0.469*** 0.466*** 0.227*** 0.240*** 0.240*** 
 (0.006) (0.026) (0.018) (0.028) (0.030) 
Maternal education high school × Below  0.008 0.014 0.014 0.010 
  (0.007) (0.009) (0.010) (0.011) 
Maternal education some college × Below  0.044*** 0.051*** 0.048*** 0.048*** 
  (0.010) (0.011) (0.014) (0.014) 
Maternal education bachelor's+ × Below  0.073*** 0.067*** 0.066*** 0.043* 
  (0.012) (0.015) (0.021) (0.024) 
Mother foreign × Below  −0.041*** −0.043*** −0.045*** −0.039*** 
  (0.010) (0.008) (0.012) (0.013) 
Mother married × Below  0.004 −0.007 −0.004 −0.003 
  (0.008) (0.007) (0.010) (0.011) 
Mother's age at birth × Below  0.001 0.001 0.0002 0.0003 
  (0.001) (0.001) (0.001) (0.001) 
Black × Below  −0.067*** −0.027** −0.025** −0.024** 
  (0.011) (0.010) (0.011) (0.012) 
Hispanic × Below  0.001 0.019 0.020 0.019 
  (0.012) (0.015) (0.014) (0.016) 
Free or reduced-price lunch eligible × Below  0.017 0.023** 0.019* 0.022* 
  (0.012) (0.010) (0.011) (0.012) 
Limited English Proficient × Below   0.025 0.031* 0.022 
   (0.019) (0.016) (0.019) 
Special education × Below   0.105*** 0.108*** 0.111*** 
   (0.012) (0.011) (0.013) 
SAT10 reading > 50th percentile × Below   0.114*** 0.118*** 0.122*** 
   (0.010) (0.018) (0.020) 
Standardized FCAT math × Below   0.031*** 0.034*** 0.0341*** 
   (0.006) (0.008) (0.008) 
SAT10 reading percentile × Below   0.005*** 0.005*** 0.005*** 
   (0.0002) (0.0004) (0.001) 
SAT10 math percentile × Below   0.002*** 0.002*** 0.001*** 
   (0.0002) (0.0002) (0.0003) 
Relative age × Below   0.009*** 0.008*** 0.007*** 
   (0.001) (0.001) (0.001) 
School fixed effects No No No Yes No 
Teacher fixed effects No No No No Yes 
N 107,874 107,860 105,896 105,896 105,896 
(1)(2)(3)(4)(5)
Below 0.469*** 0.466*** 0.227*** 0.240*** 0.240*** 
 (0.006) (0.026) (0.018) (0.028) (0.030) 
Maternal education high school × Below  0.008 0.014 0.014 0.010 
  (0.007) (0.009) (0.010) (0.011) 
Maternal education some college × Below  0.044*** 0.051*** 0.048*** 0.048*** 
  (0.010) (0.011) (0.014) (0.014) 
Maternal education bachelor's+ × Below  0.073*** 0.067*** 0.066*** 0.043* 
  (0.012) (0.015) (0.021) (0.024) 
Mother foreign × Below  −0.041*** −0.043*** −0.045*** −0.039*** 
  (0.010) (0.008) (0.012) (0.013) 
Mother married × Below  0.004 −0.007 −0.004 −0.003 
  (0.008) (0.007) (0.010) (0.011) 
Mother's age at birth × Below  0.001 0.001 0.0002 0.0003 
  (0.001) (0.001) (0.001) (0.001) 
Black × Below  −0.067*** −0.027** −0.025** −0.024** 
  (0.011) (0.010) (0.011) (0.012) 
Hispanic × Below  0.001 0.019 0.020 0.019 
  (0.012) (0.015) (0.014) (0.016) 
Free or reduced-price lunch eligible × Below  0.017 0.023** 0.019* 0.022* 
  (0.012) (0.010) (0.011) (0.012) 
Limited English Proficient × Below   0.025 0.031* 0.022 
   (0.019) (0.016) (0.019) 
Special education × Below   0.105*** 0.108*** 0.111*** 
   (0.012) (0.011) (0.013) 
SAT10 reading > 50th percentile × Below   0.114*** 0.118*** 0.122*** 
   (0.010) (0.018) (0.020) 
Standardized FCAT math × Below   0.031*** 0.034*** 0.0341*** 
   (0.006) (0.008) (0.008) 
SAT10 reading percentile × Below   0.005*** 0.005*** 0.005*** 
   (0.0002) (0.0004) (0.001) 
SAT10 math percentile × Below   0.002*** 0.002*** 0.001*** 
   (0.0002) (0.0002) (0.0003) 
Relative age × Below   0.009*** 0.008*** 0.007*** 
   (0.001) (0.001) (0.001) 
School fixed effects No No No Yes No 
Teacher fixed effects No No No No Yes 
N 107,874 107,860 105,896 105,896 105,896 

Notes: Robust standard errors, clustered at the relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using a degree of 1 and within 20 points of the promotion cutoff, using 2002—07 cohorts. The coefficients on the interaction terms represent the difference in the discontinuity in exemption receipt at the cutoff by the listed student attribute. All regressions include year fixed effects and gender. FCAT: Florida Comprehensive Assessment Test; SAT10: Stanford Aptitude Test.

***p < 0.01; **p < 0.05; *p < 0.1.

Table 8 breaks down the effect of failing the third-grade reading test on exemption receipt by the type of exemption received, focusing on the heterogeneity by maternal education. The results suggest that the relationship between maternal education and the effect on exemption receipt is mainly driven by discrepancies in alternative test and teacher portfolio exemptions. For instance, just-failing students whose mothers have a bachelor's degree or more are 4.4 percentage points more likely to receive a teacher portfolio exemption compared with failing students whose mothers have less than a high school degree. This finding provides further evidence that high-SES parents are more likely to advocate for the promotion of their children and/or are more likely to succeed in their efforts.

Table 8.
Effect of Scoring Below the Promotion Cutoff on the Type of Exemption Received in Third Grade, by Maternal Education
Maternal Education Level
High School Diploma × BelowSome College × BelowBachelors+ × Below
Received an exemption 0.014 0.048*** 0.066*** 
 (0.010) (0.014) (0.021) 
Limited English proficient exemption 0.003 0.003 0.007* 
 (0.002) (0.002) (0.004) 
Special education exemption −0.002 −0.006 −0.008 
 (0.005) (0.007) (0.010) 
Alternative test exemption 0.016** 0.037*** 0.025 
 (0.007) (0.012) (0.017) 
Teacher portfolio exemption 0.001 0.016 0.044*** 
 (0.007) (0.010) (0.017) 
Twice retained exemption −0.004* −0.002 −0.003 
 (0.002) (0.003) (0.004) 
N 105,896 105,896 105,896 
Maternal Education Level
High School Diploma × BelowSome College × BelowBachelors+ × Below
Received an exemption 0.014 0.048*** 0.066*** 
 (0.010) (0.014) (0.021) 
Limited English proficient exemption 0.003 0.003 0.007* 
 (0.002) (0.002) (0.004) 
Special education exemption −0.002 −0.006 −0.008 
 (0.005) (0.007) (0.010) 
Alternative test exemption 0.016** 0.037*** 0.025 
 (0.007) (0.012) (0.017) 
Teacher portfolio exemption 0.001 0.016 0.044*** 
 (0.007) (0.010) (0.017) 
Twice retained exemption −0.004* −0.002 −0.003 
 (0.002) (0.003) (0.004) 
N 105,896 105,896 105,896 

Notes: Robust standard errors, clustered at the relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using a degree of 1, within 20 points of the promotion cutoff, and including all interacted controls, gender, and school and year fixed effects found in table 5, column 4. Each entry represents the difference in the estimated discontinuity for the given outcome at the cutoff between the maternal education level provided in the column and the baseline category (students who have mothers with less than a high school degree).

***p < 0.01; **p < 0.05; *p < 0.1.

Subgroup Analysis

Another interesting question in this context is whether the returns to SES on exemption receipt differ by student, teacher, or school characteristics in the third grade. This might take place, for instance, if high-SES parents are more likely to advocate for their children's exemption eligibility, yet their success in these efforts varies based on these attributes. Table 9 breaks down the analysis presented in table 7 by student race/ethnicity (first three columns), and by FRPL eligibility (the last two columns). All estimates are obtained using the fully interacted school fixed effects model given in column 4 of table 5. The findings are relatively consistent, with well-educated parents more likely to be granted an exemption regardless of student race/ethnicity and poverty, though the coefficients are imprecisely estimated in some cases.

Table 9.
Subgroup Effects of Scoring Below the Promotion Cutoff on Receiving an Exemption in Third Grade: Student Characteristics
WhiteBlackHispanicFRPL EligibleFRPL Ineligible
Below 0.228*** 0.226*** 0.233*** 0.234*** 0.282*** 
 (0.052) (0.047) (0.060) (0.032) (0.056) 
Maternal education high school × Below −0.007 0.031** −0.023 0.015 0.008 
 (0.020) (0.015) (0.021) (0.011) (0.025) 
Maternal education some college × Below 0.043* 0.045** 0.036 0.039** 0.058** 
 (0.025) (0.022) (0.030) (0.017) (0.027) 
Maternal education bachelor's+ × Below 0.055 0.060 0.058 0.064* 0.066** 
 (0.036) (0.044) (0.044) (0.035) (0.033) 
Mother foreign × Below 0.012 −0.057*** −0.050** −0.052*** −0.034 
 (0.032) (0.019) (0.020) (0.014) (0.023) 
Mother married × Below 0.001 −0.017 0.005 −0.0001 −0.006 
 (0.018) (0.017) (0.019) (0.011) (0.020) 
Mother's age at birth × Below 0.001 0.00002 −0.001 −0.0004 0.002 
 (0.001) (0.001) (0.002) (0.0009) (0.002) 
Black × Below    −0.009 −0.054** 
    (0.013) (0.023) 
Hispanic × Below    0.041** −0.023 
    (0.017) (0.026) 
Free or reduced-price lunch eligible × Below −0.006 0.020 0.064***   
 (0.017) (0.021) (0.023)   
N 35,740 40,629 23,261 72,928 31,041 
WhiteBlackHispanicFRPL EligibleFRPL Ineligible
Below 0.228*** 0.226*** 0.233*** 0.234*** 0.282*** 
 (0.052) (0.047) (0.060) (0.032) (0.056) 
Maternal education high school × Below −0.007 0.031** −0.023 0.015 0.008 
 (0.020) (0.015) (0.021) (0.011) (0.025) 
Maternal education some college × Below 0.043* 0.045** 0.036 0.039** 0.058** 
 (0.025) (0.022) (0.030) (0.017) (0.027) 
Maternal education bachelor's+ × Below 0.055 0.060 0.058 0.064* 0.066** 
 (0.036) (0.044) (0.044) (0.035) (0.033) 
Mother foreign × Below 0.012 −0.057*** −0.050** −0.052*** −0.034 
 (0.032) (0.019) (0.020) (0.014) (0.023) 
Mother married × Below 0.001 −0.017 0.005 −0.0001 −0.006 
 (0.018) (0.017) (0.019) (0.011) (0.020) 
Mother's age at birth × Below 0.001 0.00002 −0.001 −0.0004 0.002 
 (0.001) (0.001) (0.002) (0.0009) (0.002) 
Black × Below    −0.009 −0.054** 
    (0.013) (0.023) 
Hispanic × Below    0.041** −0.023 
    (0.017) (0.026) 
Free or reduced-price lunch eligible × Below −0.006 0.020 0.064***   
 (0.017) (0.021) (0.023)   
N 35,740 40,629 23,261 72,928 31,041 

Notes: Robust standard errors, clustered at relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using a degree of 1, within 20 points of the promotion cutoff, and including all interacted controls found in table 5, column 3, year fixed effects and gender. The coefficients on the interaction terms represent the difference in the discontinuity in exemption receipt at the cutoff by the listed student attribute, estimated separately for each student attribute listed in the columns. FRPL = free or reduced-price lunch.

***p < 0.01; **p < 0.05; *p < 0.1.

Tables 10 and 11 present the estimates obtained using various teacher and school subgroups of interest, respectively. It is important to note that it is difficult to make causal inferences in this exercise about the relationship between teacher/school characteristics and the returns to SES on receiving an exemption because of the nonrandom nature of student–teacher or student–school matchings. Nevertheless, table 10 breaks down the analysis by the third-grade teacher's race/ethnicity (first two columns) and experience (last two columns). The maternal education and maternal place of birth gaps in exemption receipt exist for all teacher subgroups yet seem to be slightly larger when the teacher is of the same racial/ethnic group as the student, or if the teacher is an early-career teacher with less than three years of experience.

Table 10.
Subgroup Effects of Scoring below the Promotion Cutoff on Receiving an Exemption in Third Grade: Third-Grade Teacher Characteristics
Same Race/EthnicityDifferent Race/EthnicityEarly CareerLate Career
Below 0.257*** 0.210*** 0.333*** 0.211*** 
 (0.043) (0.041) (0.049) (0.053) 
Maternal education high school × Below 0.012 0.019 0.019 0.027 
 (0.015) (0.014) (0.016) (0.018) 
Maternal education some college × Below 0.053*** 0.041** 0.048** 0.051** 
 (0.020) (0.018) (0.023) (0.024) 
Maternal education bachelor's+ × Below 0.101*** 0.029 0.085** 0.012 
 (0.030) (0.031) (0.038) (0.038) 
Mother foreign × Below −0.014 −0.059*** −0.055*** −0.034 
 (0.020) (0.015) (0.019) (0.021) 
Mother married × Below 0.0003 −0.007 0.018 −0.008 
 (0.015) (0.014) (0.017) (0.018) 
Mother's age at birth × Below −0.0004 0.0008 −0.002 0.002 
 (0.001) (0.001) (0.001) (0.001) 
Black × Below −0.022 −0.025 −0.048** −0.015 
 (0.0171) (0.018) (0.019) (0.019) 
Hispanic × Below −0.050** 0.046** −0.002 0.020 
 (0.024) (0.020) (0.023) (0.025) 
Free or reduced–price lunch eligible × Below 0.016 0.025 0.017 0.020 
 (0.015) (0.016) (0.018) (0.019) 
N 49,429 54,540 36,473 33,535 
Same Race/EthnicityDifferent Race/EthnicityEarly CareerLate Career
Below 0.257*** 0.210*** 0.333*** 0.211*** 
 (0.043) (0.041) (0.049) (0.053) 
Maternal education high school × Below 0.012 0.019 0.019 0.027 
 (0.015) (0.014) (0.016) (0.018) 
Maternal education some college × Below 0.053*** 0.041** 0.048** 0.051** 
 (0.020) (0.018) (0.023) (0.024) 
Maternal education bachelor's+ × Below 0.101*** 0.029 0.085** 0.012 
 (0.030) (0.031) (0.038) (0.038) 
Mother foreign × Below −0.014 −0.059*** −0.055*** −0.034 
 (0.020) (0.015) (0.019) (0.021) 
Mother married × Below 0.0003 −0.007 0.018 −0.008 
 (0.015) (0.014) (0.017) (0.018) 
Mother's age at birth × Below −0.0004 0.0008 −0.002 0.002 
 (0.001) (0.001) (0.001) (0.001) 
Black × Below −0.022 −0.025 −0.048** −0.015 
 (0.0171) (0.018) (0.019) (0.019) 
Hispanic × Below −0.050** 0.046** −0.002 0.020 
 (0.024) (0.020) (0.023) (0.025) 
Free or reduced–price lunch eligible × Below 0.016 0.025 0.017 0.020 
 (0.015) (0.016) (0.018) (0.019) 
N 49,429 54,540 36,473 33,535 

Notes: Robust standard errors, clustered at relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using a degree of 1, within 20 points of the promotion cutoff, and including all interacted controls found in table 5, column 3, year fixed effects, and gender. The coefficients on the interaction terms represent the difference in the discontinuity in exemption receipt at the cutoff by the listed student attribute, estimated separately for each teacher attribute listed in the columns.

***p < 0.01; **p < 0.05.

Table 11.
Subgroup Effects of Scoring below the Promotion Cutoff on Receiving an Exemption in Third Grade: School Characteristics
High PovertyLow PovertyHigh MinorityLow MinorityHigh Parental EducationLow Parental Education
Below 0.232*** 0.178** 0.216*** 0.193*** 0.226*** 0.206*** 
 (0.059) (0.076) (0.057) (0.068) (0.076) (0.052) 
Maternal education high school × Below 0.009 0.063* 0.017 0.047* 0.030 0.024 
 (0.016) (0.033) (0.016) (0.026) (0.033) (0.016) 
Maternal education some college × Below 0.056** 0.079** 0.051** 0.108*** 0.066* 0.086*** 
 (0.027) (0.037) (0.024) (0.033) (0.036) (0.027) 
Maternal education bachelor's+ × Below 0.089 0.106** 0.127*** 0.087* 0.105** 0.168*** 
 (0.059) (0.046) (0.045) (0.048) (0.044) (0.057) 
Mother foreign × Below −0.074*** −0.042 −0.045** −0.004 −0.052* −0.042** 
 (0.020) (0.029) (0.018) (0.042) (0.028) (0.021) 
Mother married × Below −0.016 0.004 −0.011 −0.014 −0.017 0.006 
 (0.018) (0.027) (0.017) (0.025) (0.027) (0.017) 
Mother's age at birth × Below 0.0003 0.003 −0.001 0.001 0.002 0.001 
 (0.001) (0.002) (0.001) (0.002) (0.002) (0.001) 
Black × Below −0.023 −0.069** −0.025 −0.070** −0.081*** −0.038* 
 (0.028) (0.031) (0.033) (0.032) (0.030) (0.022) 
Hispanic × Below 0.019 0.002 −0.022 0.093** −0.034 0.040 
 (0.032) (0.034) (0.034) (0.046) (0.031) (0.026) 
Free or reduced-price lunch eligible × Below 0.058* 0.037 0.050** −0.028 0.027 0.028 
 (0.031) (0.026) (0.024) (0.022) (0.024) (0.026) 
N 29,607 17,588 18,329 31,927 18,099 31,019 
High PovertyLow PovertyHigh MinorityLow MinorityHigh Parental EducationLow Parental Education
Below 0.232*** 0.178** 0.216*** 0.193*** 0.226*** 0.206*** 
 (0.059) (0.076) (0.057) (0.068) (0.076) (0.052) 
Maternal education high school × Below 0.009 0.063* 0.017 0.047* 0.030 0.024 
 (0.016) (0.033) (0.016) (0.026) (0.033) (0.016) 
Maternal education some college × Below 0.056** 0.079** 0.051** 0.108*** 0.066* 0.086*** 
 (0.027) (0.037) (0.024) (0.033) (0.036) (0.027) 
Maternal education bachelor's+ × Below 0.089 0.106** 0.127*** 0.087* 0.105** 0.168*** 
 (0.059) (0.046) (0.045) (0.048) (0.044) (0.057) 
Mother foreign × Below −0.074*** −0.042 −0.045** −0.004 −0.052* −0.042** 
 (0.020) (0.029) (0.018) (0.042) (0.028) (0.021) 
Mother married × Below −0.016 0.004 −0.011 −0.014 −0.017 0.006 
 (0.018) (0.027) (0.017) (0.025) (0.027) (0.017) 
Mother's age at birth × Below 0.0003 0.003 −0.001 0.001 0.002 0.001 
 (0.001) (0.002) (0.001) (0.002) (0.002) (0.001) 
Black × Below −0.023 −0.069** −0.025 −0.070** −0.081*** −0.038* 
 (0.028) (0.031) (0.033) (0.032) (0.030) (0.022) 
Hispanic × Below 0.019 0.002 −0.022 0.093** −0.034 0.040 
 (0.032) (0.034) (0.034) (0.046) (0.031) (0.026) 
Free or reduced-price lunch eligible × Below 0.058* 0.037 0.050** −0.028 0.027 0.028 
 (0.031) (0.026) (0.024) (0.022) (0.024) (0.026) 
N 29,607 17,588 18,329 31,927 18,099 31,019 

Notes: Robust standard errors, clustered at relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using a degree of 1, within 20 points of the promotion cutoff, and including all interacted controls found in table 5, column 3, year fixed effects, and gender. The coefficients on the interaction terms represent the difference in the discontinuity in exemption receipt at the cutoff by the listed student attribute, estimated separately for each school attribute listed in the columns.

***p < 0.01; **p < 0.05; *p < 0.1.

Finally, table 11 breaks down the analysis by school poverty level (top quartile in percent FRPL eligible versus bottom), school minority concentration (top quartile in percent white versus bottom), and school maternal education (top quartile in average maternal education versus bottom). The SES gaps are once again consistent across subgroups, albeit imprecisely estimated in some cases, suggesting that better-educated mothers are able to receive exemptions for their failing children at higher rates regardless of the school setting.

Identification Checks

A chief concern in any regression discontinuity analysis is the possibility of manipulation of the forcing variable around the cutoff (Urquiola and Verhoogen 2009). In this context, for example, we would be concerned about the ability of students just above the cutoff to serve as a counterfactual to students just below the cutoff if teachers or principals were able to manipulate students’ reading scores to push them over the promotion cutoff. Under this scenario, one would expect to see a discontinuity in the test score distribution around the promotion cutoff. It is important to note that because FCAT tests are scored objectively without teacher assistance this is unlikely, however. The formal test developed by McCrary (2008) is not appropriate in this case because it relies on local linear regression, which can lead to incorrect inferences when the running variable is discrete (Card and Lee 2008). We therefore present graphical evidence to dispel any concerns. Figure 5 shows the overall distribution of third-grade reading scores is smooth around the cutoff, with no indication of heaping observations around the cutoff.

Figure 5.

Distribution of Third-Grade Reading Scores

Figure 5.

Distribution of Third-Grade Reading Scores

The use of the regression discontinuity to provide a counterfactual to difference out also relies on the assumption that there are no discontinuities in other student characteristics associated with outcomes around the cutoff. Figure 6 addresses this concern by plotting the mean value of observable student characteristics against third-grade reading scores close to the cutoff, and shows that there are no discontinuities in observed student characteristics at the cut point.

Figure 6.

The Relationship Between Reading Scores in Third Grade and Student Characteristics

Notes: SAT10: Stanford Achievement Test.

Figure 6.

The Relationship Between Reading Scores in Third Grade and Student Characteristics

Notes: SAT10: Stanford Achievement Test.

Robustness Checks

To check the robustness of our findings to model specification, Appendix table A.2 presents the results of the interaction between our measures of SES and scoring below the promotion cutoff using various bandwidths and polynomial orders. The results in this table are obtained using the fully interacted model given in column 3 of table 5 with linear, quadratic, and cubic polynomial specifications and bandwidths of 5, 10, 20, and 50 points. Estimated differences at the retention cutoff are very stable across bandwidths and specifications, yet imprecisely estimated in some specifications.

5.  Conclusion

Our analysis relies on difference-in-difference framework at the retention cutoff to study whether Florida's test-based promotion policy is differentially enforced for students depending on their mother's socioeconomic background, specifically educational attainment. We find that the more educated a student's mother is, the less likely he is to be retained due to the Florida policy. Students are also more likely to be retained due to the policy if their mother is foreign born, if they qualify for free or reduced-price lunch, or if they are black. We also find that the socioeconomic differences in the effect of the policy are mainly driven by the fact that students from high-SES families are more likely to be granted an exemption even after we control for exemption eligibility. These findings are robust to various model specifications and hold true for different student, teacher, and school subgroups of interest.

Our results have important implications for public policy. Broad, universal educational policies are often implemented to address inequalities in outcomes for students of differing backgrounds by holding all children to the same standards. Although the allowance for exemptions in the Florida retention policy is in place in order to avoid retaining students for whom retention is seen to be harmful or inappropriate, it is important to understand whether an unintended consequence of this allowance is that children are being retained differentially based on their mother's education or other characteristics which should not impact whether or not the policy is appropriate for them. Although we cannot from our study discern the exact reason why Florida's retention policy is more strictly enforced for the children of less educated, poor, and foreign-born women, prior research findings indicating that parents of lower SES have less knowledge of their children's educational context and are less likely to intervene in school decisions led us to hypothesize that these same dynamics are at play in this context. This hypothesis is reinforced by the estimated socioeconomic differences in exemption receipt, especially the differences in teacher portfolio exemption receipt, for which the policy requires parents to actively engage with their child's teacher/principal.

It is important to note that we are not able to completely rule out the possibility that there are unobservable differences in students related to their mothers’ socioeconomic background and that influence whether teachers and school administrators grant students an exemption from the policy. We control for exemption eligibility using a wide array of student characteristics but it is still possible that the socioeconomic differences in exemption receipt might be driven by unobserved differences in reading skills (or cognitive abilities) between socioeconomic groups. We alleviate this concern by focusing on the students who scored right around the retention cutoff, examining the difference in retention probability for students of different backgrounds who are just impacted by the policy subtracting out any differences between those same groups of students who are just above the promotion cutoff. Differences by maternal education level are apparent but very small for children above the promotion cutoff, though these children would be subject to any inherent socioeconomic differences in schools’ or families’ desire to retain the student. Furthermore, when examining differences in retention probability during the two years before the policy was enacted, we also find much smaller differences. It therefore appears that the allowance for exemptions into Florida's test-based promotion policy has resulted in differential policy implementation by SES, allowing parents with greater knowledge, agency, and resources the ability to circumvent the policy in greater numbers and exacerbating any differences in retention that are present in the absence of the policy, although it cannot be ruled out that part of the greater utilization of student portfolios among students with parents of higher SES is due to teachers or principals recommending to these students’ parents that they request such an exemption.

Given that the test-based promotion policy was introduced under the assumption that students performing below the cutoff would benefit from being retained, it is unclear whether the greater number of more educated parents circumventing the policy is actually helpful or harmful for their children. Recent research on the academic effects of the early grade retention policy in Florida reveals significant short-term benefits compared with same-age peers that fade out over time (Schwerdt, West, and Winters 2015). These results suggest, from an academic standpoint, that children of lower SES who are being retained in higher numbers are not reaping long-term benefits academically from their retention despite the additional services they receive. They are, however, losing future earnings as a result of the increased time they spend in school. Furthermore, there is evidence that grade retention increases the likelihood of disciplinary incidents and suspensions and these negative effects are concentrated among black and economically disadvantaged students (Ozek 2015). This suggests that Florida's test-based promotion policy may in fact be harming, not helping, students from low-SES families.

Notes

1. 

In sociology literature, this is commonly known as the secondary effect in education (Boudon 1974). In particular, primary effects are all those, whether of a genetic or sociocultural variety, that are expressed through the association between students’ backgrounds and their actual levels of academic performance. Secondary effects are those that are expressed through the educational choices that children (or parents) from differing socioeconomic backgrounds make. In this paper, we investigate secondary effects by comparing students of different socioeconomic backgrounds with similar prior achievement levels, thus eliminating any differences due to primary effects.

2. 

For more information on the quality of the match between birth and school records, see Figlio et al. (2014).

3. 

During the two years prior to the implementation of Florida's mandatory grade retention policy, 24 percent of the state's third graders scored a Level 1 on the third-grade FCAT reading exam. By 2009, this percentage dropped to less than 15 percent.

4. 

For more detailed information on the Florida policy see www.fldoe.org/core/fileparse.php/7690/urlt/0070151-read_to_learn.pdf".

5. 

Until 2008, the state administered both the FCAT and the Stanford 10 to all students in grades 3 through 10.

6. 

The first group of exemptions is based upon the assumption that, despite their lack of reading proficiency, some students would be harmed by being retained and retention is not an appropriate educational strategy. The second group of exemptions, on the other hand, is for students who, in spite of their low FCAT scores, are actually proficient in reading and able to be successful in fourth grade.

7. 

For a student to receive a good cause exemption, the teacher must submit documentation to the principal, who then reviews the documentation and decides whether the student should be promoted. If the principal determines that the student should indeed be promoted, the principal makes the recommendation to the school district superintendent, who then approves or rejects the recommendation.

8. 

Greene and Winters (2009) examine whether Florida's retention policy led to disproportionate retentions by observable characteristics including race/ethnicity and FRPL eligibility. They find that racial minorities and economically disadvantaged students (as measured by FRPL eligibility) are more likely to be retained under the policy. These findings lend themselves to three possible explanations. First, different student groups might be retained at different rates because of differences in exemption eligibility. Second, students from different backgrounds tend to attend different elementary schools (or might be assigned to different teachers), and different schools (or different teachers) might be enforcing the retention policy differently. Finally, all else constant, school administrators/teachers might be enforcing the retention policy differently based on student SES. In this study, we explore the last possibility by explicitly controlling for exemption eligibility and school or teacher fixed effects, thus eliminating the first two explanations. Further, because we can match school records to birth records, we are able to study differential retention rates by family characteristics, such as maternal education, that more closely map to SES than the variables found in school records alone (such as those observed by Greene and Winters 2009). Additionally, Schwerdt, West, and Winters (2015) found stronger first-stage effects in their two-stage least squares analyses of Florida's retention policy for students who qualified for FRPL, providing evidence of greater compliance with the retention policy for these students.

9. 

During the two years before Florida's mandatory grade retention policy was enacted, less than 1 percent of students whose mothers had a bachelor's degree or more were retained, compared with more than 6 percent of students whose mothers had less than a high school degree.

10. 

Birth outcome data include birth weight, gestational weeks, congenital anomalies, and complications of labor and delivery.

11. 

Figlio et al. (2014) analyzed data from the Census Bureau's American Communities Survey to estimate that they would have expected an 80 percent match rate given in- and out-migration to and from Florida.

12. 

Because we only have access to birth records starting in 1992, we do not have a full cohort of students from the 2000 third-grade cohort (who would normally be born between 1 September 1991 and 31 August 1992). This results in our 2000 cohort consisting only of the approximately three-quarters of students we would expect to be born 1 January 1992 or after, therefore making the included students from the 2000 cohort slightly younger on average than students from full cohorts. We include this cohort in spite of this limitation in order to include more students and lend power in our comparisons between the before and after policy time periods. Because this cohort is from before the policy was enacted, they are not included in any regression discontinuity analyses.

13. 

Of the data used to create variables for our analyses, the following come from children's birth records: maternal education, maternal country of origin, maternal marital status, maternal birth date, child birth date, and child birth weight. Child race, FRPL status, disability, limited English proficiency, and all test score data come from educational records. Because maternal education and marital status may change over time, it is important to note that our measures are taken at the time the child was born, not at the time they entered third grade.

14. 

Observations were not included for three main reasons related to the design of, and questions put forth by, the study: (1) the student was not observed in the data the year after his first appearance as a third grader, rendering it impossible to know what grade the child was in the following year (1 percent); (2) the student did not have a third-grade FCAT reading score, rendering it impossible to know whether the child was subject to the policy (4 percent); and (3) data were missing on years of maternal education at birth (less than 1 percent). Observations were also not included if the child were born more than two years before or after the appropriate range of birthdates for first-time third graders for that cohort (less than 1 percent).

15. 

Indicators for the level of maternal education were created in the following way: eleven or fewer years of education reported is coded as less than a high school degree, twelve years of education is coded as a high school degree, thirteen to fifteen years of education is coded as some college, and sixteen or more years of education is coded as a bachelor's degree or more.

16. 

Results broken out by cohort year can be found in Appendix table A.1.

Acknowledgments

This research was supported by the U.S. Department of Education, Institute of Education Sciences, Multidisciplinary Program in Education Sciences, grant award no. R305B080027/R305B080027. We are grateful to Diane Schanzenbach, Jon Guryan, Lindsay Chase-Lansdale, Kirabo Jackson, Matthew Shirrell, Kelly Hallberg, Francie Streich, James Pustejovsky, and the rest of the Institute for Policy Research at Northwestern University for feedback on drafts, and to seminar and conference participants and discussants at AEFP and APPAM for helpful comments and suggestions. We would also like to thank Florida Departments of Education and Health for providing the data used in this analysis. All opinions expressed in this paper are those of the authors and do not reflect the views of the Florida Departments of Education and Health or our funders.

REFERENCES

Barg
,
Katherin
.
2012
.
The influence of students’ social background and parental involvement on teachers’ school track choices: Reasons and consequences
.
European Sociological Review
29
(
3
):
565
579
.
Boudon
,
Raymond
.
1974
.
Education, opportunity, and social inequality: Changing prospects in Western society
.
New York
:
Wiley
.
Cullen
,
Julie B.
,
Brian A.
Jacob
, and
Steven D.
Levitt
.
2005
.
The impact of school choice on student outcomes: An analysis of the Chicago Public Schools
.
Journal of Public Economics
89
(
5–6
):
729
760
.
Duncan
,
Greg J.
, and
Katherine
Magnuson
.
2011
. The nature and impact of early achievement skills, attention skills, and behavior problems. In
Whither opportunity? Rising inequality and the uncertain life chances of low-income children
,
edited by
Richard J.
Murnane
and
Greg J.
Duncan
, pp.
47
70
.
New York
:
Russell Sage Foundation Press
.
Figlio
,
David N.
,
Jonathan
Guryan
,
Krzysztof
Karbownik
, and
Jeffrey
Roth
.
2014
.
The effects of poor neonatal health on children's cognitive development
.
American Economic Review
104
(
12
):
3921
3955
.
Fiske
,
Edward B.
, and
Helen F.
Ladd
.
2000
.
When schools compete: A cautionary tale
.
Washington, DC
:
Brookings Institution Press
.
Greene
,
Jay P.
, and
Marcus A.
Winters
.
2007
.
Revisiting grade retention: An evaluation of Florida's test-based promotion policy
.
Education Finance and Policy
2
(
4
):
319
340
.
Greene
,
Jay P.
, and
Marcus A.
Winters
.
2009
.
The effects of exemptions to Florida's test-based promotion policy: Who is retained? Who benefits academically
?
Economics of Education Review
28
(
1
):
135
142
.
Greene
,
Jay P.
, and
Marcus A.
Winters
.
2012
.
The medium-run effects of Florida's test-based promotion policy
.
Education Finance and Policy
7
(
3
):
305
330
.
Hahn
,
Jinyong
,
Petra
Todd
, and
Wilbert Van
der Klaauw
.
2001
.
Identification and estimation of treatment effects with a regression-discontinuity design
.
Econometrica
59
(
1
):
201
209
.
Hastings
,
Justine S.
,
Thomas J.
Kane
, and
Douglas O.
Staiger
.
2006a
.
Parental preferences and school competition: Evidence from a public school choice program
.
NBER Working Paper No. 11805
.
Hastings
,
Justine S.
,
Thomas J.
Kane
, and
Douglas O.
Staiger
.
2006b
.
Preferences and heterogeneous treatment effects in a public school choice lottery
.
NBER Working Paper No. 12145
.
Haveman
,
Robert
, and
Barbara
Wolfe
.
1995
.
The determinants of children's attainments: A review of methods and findings
.
Journal of Economic Literature
33
(
4
):
1829
1878
.
Hsieh
,
Cheng-Tai
, and
Miguel
Urquiola
.
2006
.
The effects of generalized school choice on achievement and stratification: Evidence from Chile's voucher program
.
Journal of Public Economics
90
(
8–9
):
1477
1503
.
Holmes
,
Thomas C.
1989
. Grade level retention effects: A meta-analysis of research studies. In
Flunking grades: Research and policies on retention
,
edited by
Lorrie A.
Shepard
and
Mary Lee
Smith
, pp.
16
33
.
New York
:
The Falmer Press
.
Jacob
,
Brian A.
, and
Lars
Lefgren
.
2005
.
What do parents value in education? An empirical investigation of parents’ revealed preferences for teachers
.
NBER Working Paper No. 11494
.
Jacob
,
Brian A.
, and
Lars
Lefgren
.
2009
.
The effect of grade retention on high school completion
.
American Economic Journal: Applied Economics
1
(
3
):
33
58
.
Jimerson
,
Shane R.
1999
.
On the failure of failure: Examining the association between early grade retention and education and employment outcomes during late adolescence
.
Journal of School Psychology
37
(
3
):
243
272
.
Lareau
,
Annette
, and
Jessica McCrory
Calarco
.
2012
. Class, cultural capital, and institutions: The case of families and schools. In
Facing social class: How societal rank influences interactions
,
edited by
Susan T.
Fiske
and
Hazel Rose
Markus
, pp.
61
86
.
New York
:
Russell Sage Foundation Press
.
Lee
,
D. S.
, and
D.
Card
.
2008
.
Regression discontinuity inference with specification error
.
Journal of Econometrics
142
(
2
):
655
674
.
McCrary
,
Justin
.
2008
.
Testing for manipulation of the running variable in the regression discontinuity design
.
Journal of Econometrics
142
(
2
):
698
714
.
Ozek
,
Umut
.
2015
.
Hold back to move forward? Early grade retention and student misbehavior
.
Education Finance and Policy
10
(
3
):
350
377
.
Porter
,
Jack
.
2003
.
Estimation in the regression discontinuity model
.
Unpublished paper, Department of Economics, University of Wisconsin, Madison
.
Reardon
,
Sean F.
2011
. The widening academic achievement gap between the rich and the poor: New evidence and possible explanations. In
Whither opportunity? Rising inequality and the uncertain life chances of low-income children
,
edited by
Richard J.
Murnane
and
Gregg J.
Duncan
, pp.
91
116
.
New York
:
Russell Sage Foundation Press
.
Roderick
,
Melissa
, and
Jenny
Nagaoka
.
2005
.
Retention under Chicago's high stakes testing program: Helpful, harmful, or harmless? Educational Evaluation and Policy Analysis
27
(
1
):
309
340
.
Rose
,
Stephanie
, and
Karen
Schimke
.
2012
.
Third grade literacy policies: Identification, intervention, retention
.
Available
https://www.ecs.org/clearinghouse/01/01/54/10154.pdf".
Accessed 22 January 2019
.
Schwerdt
,
Guido
,
Martin R.
West
, and
Marcus A.
Winters
.
2015
.
The effects of early grade retention on student outcomes over time: Regression discontinuity evidence from Florida
.
NBER Working Paper No. 21509
.
Urquiola
,
Miguel
, and
Eric
Verhoogen
.
2009
.
Class-size caps, sorting, and the regression-discontinuity design
.
American Economic Review
99
(
1
):
179
215
.
Weininger
,
Elliott B.
, and
Annette
Lareau
.
2003
.
Translating Bourdieu into the American context: The question of social class and family-school relations
.
Poetic
31
(
5–6
):
375
402
.

Appendix

Table A.1.
Effect of Scoring Below the Promotion Cutoff on Retention in Third Grade—–By Cohort Year
200220032004200520062007
Below 0.603*** 0.551*** 0.644*** 0.605*** 0.464*** 0.388*** 
 (0.070) (0.083) (0.076) (0.087 (0.063) (0.071) 
Maternal education high school × Below 0.019 −0.070*** −0.031 −0.007 −0.013 −0.004 
 (0.023) (0.026) (0.027) (0.029) (0.022) (0.025) 
Maternal education some college × Below −0.023 −0.006 −0.092*** −0.027 −0.012 −0.050 
 (0.031) (0.035) (0.035) (0.038) (0.031) (0.035) 
Maternal education bachelor's+ × Below 0.010 −0.117** −0.043 −0.075 −0.026 −0.079 
 (0.047) (0.051) (0.052) (0.065) (0.042) (0.049) 
Mother foreign × Below 0.053* 0.025 0.087*** −0.011 0.015 0.013 
 (0.027) (0.030) (0.030) (0.035) (0.026) (0.028) 
Mother married × Below −0.004 0.029 −0.017 −0.068** −0.008 −0.029 
 (0.023) (0.024) (0.025) (0.029) (0.023) (0.024) 
Mother's age at birth × Below −0.0004 −0.002 −0.002 0.002 −0.00003 0.003 
 (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) 
Black × Below −0.018 −0.051* 0.032 0.033 0.016 −0.053** 
 (0.025) (0.029) (0.028) (0.032) (0.026) (0.026) 
Hispanic × Below −0.002 −0.006 −0.005 0.102*** −0.004 −0.065** 
 (0.033) (0.037) (0.033) (0.039) (0.029) (0.032) 
Free or reduced-price lunch eligible × Below 0.107*** 0.053* −0.0010 −0.010 0.034 −0.003 
 (0.025) (0.029) (0.029) (0.033) (0.024) (0.027) 
N 21,620 16,637 17,441 14,331 18,520 17,347 
200220032004200520062007
Below 0.603*** 0.551*** 0.644*** 0.605*** 0.464*** 0.388*** 
 (0.070) (0.083) (0.076) (0.087 (0.063) (0.071) 
Maternal education high school × Below 0.019 −0.070*** −0.031 −0.007 −0.013 −0.004 
 (0.023) (0.026) (0.027) (0.029) (0.022) (0.025) 
Maternal education some college × Below −0.023 −0.006 −0.092*** −0.027 −0.012 −0.050 
 (0.031) (0.035) (0.035) (0.038) (0.031) (0.035) 
Maternal education bachelor's+ × Below 0.010 −0.117** −0.043 −0.075 −0.026 −0.079 
 (0.047) (0.051) (0.052) (0.065) (0.042) (0.049) 
Mother foreign × Below 0.053* 0.025 0.087*** −0.011 0.015 0.013 
 (0.027) (0.030) (0.030) (0.035) (0.026) (0.028) 
Mother married × Below −0.004 0.029 −0.017 −0.068** −0.008 −0.029 
 (0.023) (0.024) (0.025) (0.029) (0.023) (0.024) 
Mother's age at birth × Below −0.0004 −0.002 −0.002 0.002 −0.00003 0.003 
 (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) 
Black × Below −0.018 −0.051* 0.032 0.033 0.016 −0.053** 
 (0.025) (0.029) (0.028) (0.032) (0.026) (0.026) 
Hispanic × Below −0.002 −0.006 −0.005 0.102*** −0.004 −0.065** 
 (0.033) (0.037) (0.033) (0.039) (0.029) (0.032) 
Free or reduced-price lunch eligible × Below 0.107*** 0.053* −0.0010 −0.010 0.034 −0.003 
 (0.025) (0.029) (0.029) (0.033) (0.024) (0.027) 
N 21,620 16,637 17,441 14,331 18,520 17,347 

Notes: Robust standard errors, clustered at the relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using the given polynomial degrees and bandwidths around the promotion cutoff, including all interacted controls found in table 5, column 3, year fixed effects, and gender. The coefficients on the interaction terms represent the difference in the retention discontinuity at the cutoff, estimated separately for each school year.

***p < 0.01; **p < 0.05; *p < 0.1.

Table A.2.
Effect of Scoring Below the Promotion Cutoff on Retention in Third Grade by Functional Form and Bandwidth
Linear
Score Range5020105
Below 0.530*** 0.530*** 0.517*** 0.494*** 
 (0.014) (0.018) (0.024) (0.033) 
Maternal education high school × Below −0.021*** −0.018** −0.016 −0.018 
 (0.006) (0.009) (0.011) (0.013) 
Maternal education some college × Below −0.051*** −0.044*** −0.050*** −0.053*** 
 (0.006) (0.006) (0.007) (0.006) 
Maternal education bachelor's × Below −0.010*** −0.072*** −0.047** −0.062** 
 (0.013) (0.015) (0.020) (0.022) 
Mother foreign × Below 0.021*** 0.038*** 0.029*** 0.032*** 
 (0.007) (0.008) (0.010) (0.010) 
Mother married × Below −0.009 −0.009 −0.014 −0.019 
 (0.006) (0.011) (0.016) (0.027) 
Mother's age at birth × Below −0.0002 −0.0003 −0.0016 −0.001 
 (0.001) (0.001) (0.001) (0.002) 
Black × Below −0.012* −0.007 −0.008 0.005 
 (0.007) (0.009) (0.011) (0.014) 
Hispanic × Below 0.002 −0.010 0.006 0.016 
 (0.007) (0.010) (0.010) (0.014) 
Free or reduced-price lunch eligible × Below 0.024*** 0.031*** 0.043*** 0.028** 
 (0.007) (0.009) (0.0128) (0.011) 
 Quadratic 
Below 0.514*** 0.529*** 0.516*** 0.461*** 
 (0.017) (0.024) (0.032) (0.020) 
Maternal education high school × Below −0.014* −0.012 −0.008 −0.011 
 (0.008) (0.011) (0.012) (0.014) 
Maternal education some college × Below −0.037*** −0.044*** −0.047*** −0.064*** 
 (0.007) (0.009) (0.008) (0.010) 
Maternal education bachelor's+ × Below −0.070*** −0.035 −0.054** −0.067* 
 (0.016) (0.022) (0.021) (0.030) 
Mother foreign × Below 0.032*** 0.028*** 0.033** 0.008 
 (0.008) (0.010) (0.012) (0.010) 
Mother married × Below −0.007 −0.021 −0.027 −0.026* 
 (0.010) (0.017) (0.023) (0.013) 
Mother's age at birth × Below −0.0001 −0.001 −0.001 0.001 
 (0.001) (0.001) (0.002) (0.001) 
Black × Below −0.003 −0.007 −0.001 −0.010 
 (0.009) (0.011) (0.016) (0.012) 
Hispanic × Below −0.011 0.008 0.017 0.018*** 
 (0.010) (0.012) (0.017) (0.005) 
Free or reduced-price lunch eligible × Below 0.033*** 0.040*** 0.021 0.021* 
 (0.009) (0.012) (0.012) (0.011) 
 Cubic 
Below 0.518*** 0.519*** 0.481*** 0.464*** 
 (0.020) (0.026) (0.017) (0.020) 
Maternal education high school × Below −0.009 −0.013 −0.023 0.006 
 (0.010) (0.013) (0.016) (0.033) 
Maternal education some college × Below −0.042*** −0.056*** −0.069*** −0.028 
 (0.008) (0.010) (0.011) (0.025) 
Maternal education bachelor's+ × Below −0.042** −0.067*** −0.089*** −0.091 
 (0.020) (0.021) (0.029) (0.056) 
Mother foreign × Below 0.040*** 0.025** 0.008 −0.033*** 
 (0.009) (0.011) (0.009) (0.009) 
Mother married × Below −0.013 −0.015 −0.031 −0.003 
 (0.014) (0.025) (0.025) (0.006) 
Mother's age at birth × Below −0.001 −0.001 0.001 0.005* 
 (0.001) (0.001) (0.001) (0.002) 
Black × Below −0.006 −0.0053 −0.009 −0.052*** 
 (0.011) (0.013) (0.012) (0.011) 
Hispanic × Below −0.007 0.009 0.019* 0.033*** 
 (0.011) (0.012) (0.011) (0.004) 
Free or reduced-price lunch eligible × Below 0.033*** 0.036*** 0.020* 0.065*** 
 (0.011) (0.011) (0.010) (0.008) 
N 279,091 105,882 52,385 26,218 
Linear
Score Range5020105
Below 0.530*** 0.530*** 0.517*** 0.494*** 
 (0.014) (0.018) (0.024) (0.033) 
Maternal education high school × Below −0.021*** −0.018** −0.016 −0.018 
 (0.006) (0.009) (0.011) (0.013) 
Maternal education some college × Below −0.051*** −0.044*** −0.050*** −0.053*** 
 (0.006) (0.006) (0.007) (0.006) 
Maternal education bachelor's × Below −0.010*** −0.072*** −0.047** −0.062** 
 (0.013) (0.015) (0.020) (0.022) 
Mother foreign × Below 0.021*** 0.038*** 0.029*** 0.032*** 
 (0.007) (0.008) (0.010) (0.010) 
Mother married × Below −0.009 −0.009 −0.014 −0.019 
 (0.006) (0.011) (0.016) (0.027) 
Mother's age at birth × Below −0.0002 −0.0003 −0.0016 −0.001 
 (0.001) (0.001) (0.001) (0.002) 
Black × Below −0.012* −0.007 −0.008 0.005 
 (0.007) (0.009) (0.011) (0.014) 
Hispanic × Below 0.002 −0.010 0.006 0.016 
 (0.007) (0.010) (0.010) (0.014) 
Free or reduced-price lunch eligible × Below 0.024*** 0.031*** 0.043*** 0.028** 
 (0.007) (0.009) (0.0128) (0.011) 
 Quadratic 
Below 0.514*** 0.529*** 0.516*** 0.461*** 
 (0.017) (0.024) (0.032) (0.020) 
Maternal education high school × Below −0.014* −0.012 −0.008 −0.011 
 (0.008) (0.011) (0.012) (0.014) 
Maternal education some college × Below −0.037*** −0.044*** −0.047*** −0.064*** 
 (0.007) (0.009) (0.008) (0.010) 
Maternal education bachelor's+ × Below −0.070*** −0.035 −0.054** −0.067* 
 (0.016) (0.022) (0.021) (0.030) 
Mother foreign × Below 0.032*** 0.028*** 0.033** 0.008 
 (0.008) (0.010) (0.012) (0.010) 
Mother married × Below −0.007 −0.021 −0.027 −0.026* 
 (0.010) (0.017) (0.023) (0.013) 
Mother's age at birth × Below −0.0001 −0.001 −0.001 0.001 
 (0.001) (0.001) (0.002) (0.001) 
Black × Below −0.003 −0.007 −0.001 −0.010 
 (0.009) (0.011) (0.016) (0.012) 
Hispanic × Below −0.011 0.008 0.017 0.018*** 
 (0.010) (0.012) (0.017) (0.005) 
Free or reduced-price lunch eligible × Below 0.033*** 0.040*** 0.021 0.021* 
 (0.009) (0.012) (0.012) (0.011) 
 Cubic 
Below 0.518*** 0.519*** 0.481*** 0.464*** 
 (0.020) (0.026) (0.017) (0.020) 
Maternal education high school × Below −0.009 −0.013 −0.023 0.006 
 (0.010) (0.013) (0.016) (0.033) 
Maternal education some college × Below −0.042*** −0.056*** −0.069*** −0.028 
 (0.008) (0.010) (0.011) (0.025) 
Maternal education bachelor's+ × Below −0.042** −0.067*** −0.089*** −0.091 
 (0.020) (0.021) (0.029) (0.056) 
Mother foreign × Below 0.040*** 0.025** 0.008 −0.033*** 
 (0.009) (0.011) (0.009) (0.009) 
Mother married × Below −0.013 −0.015 −0.031 −0.003 
 (0.014) (0.025) (0.025) (0.006) 
Mother's age at birth × Below −0.001 −0.001 0.001 0.005* 
 (0.001) (0.001) (0.001) (0.002) 
Black × Below −0.006 −0.0053 −0.009 −0.052*** 
 (0.011) (0.013) (0.012) (0.011) 
Hispanic × Below −0.007 0.009 0.019* 0.033*** 
 (0.011) (0.012) (0.011) (0.004) 
Free or reduced-price lunch eligible × Below 0.033*** 0.036*** 0.020* 0.065*** 
 (0.011) (0.011) (0.010) (0.008) 
N 279,091 105,882 52,385 26,218 

Notes: Robust standard errors, clustered at the relative reading score level, are given in parentheses. Discontinuity estimates are obtained parametrically using the given polynomial degrees and bandwidths around the promotion cutoff, including all interacted controls found in table 5, column 3, year fixed effects, and gender.

***p < 0.01; **p < 0.05; *p < 0.1.