In 1996, the Connecticut Supreme Court ruled in Sheff v. O’Neill that severe racial and economic isolation among students in the public schools of Hartford, Connecticut, about 90% of whom are black or Hispanic, violated their constitutional rights to an adequate education. As a result of this ruling, many magnet schools were established in the Hartford region that aimed to attract students both from Hartford as well as from the surrounding suburbs where most students are white, thus achieving integration. A school is considered integrated if no more than 75% of its students are black or Hispanic, and acceptance to magnet programs is based on a lottery application process. Because the magnet schools are of much better quality than Hartford’s non-magnet public schools but still generally of poorer quality than suburban schools, demand for a magnet education is higher among Hartford residents than among suburban residents, and some magnet schools have difficulty attracting white and Asian students. These magnet schools then seem reluctant to accept additional minority students from their waitlists after the initial lottery selection because they may lose their “integrated” status [1]. In February 2018, a lawsuit named Robinson v. Wentzell was filed challenging the lottery application system and integration “quotas” of the Hartford magnet school system as discriminatory towards black and Hispanic students. It was determined in March 2019 that this case would be settled in federal court [2]. In February 2019, another lawsuit, Connecticut Parents’ Union v. Wentzell, was filed challenging a law that required all magnet schools throughout Connecticut to adhere to Hartford’s integration standards. In this paper, we investigate whether attendance at magnet schools has a demonstrable relationship with the performance of Connecticut students in contrast to their peers at traditional public schools. In particular, we investigate whether the level of integration of the school, considering both racial integration and economic integration, in addition to schools’ magnet status or lack thereof, is associated with differing levels of achievement, especially among students of low socioeconomic status. Several previous studies similar to the one we are undertaking have been carried out in the past. The 2018 study “Income Segregation between School Districts and Inequality in Students’ Achievement” by Ann Owens investigates the relationship between income segregation between school districts nationwide and the achievement gaps between black and white students in these districts. Using an ordinary least squares regression approach, she found that greater income segregation between districts located in the same metropolitan statistical area has a statistically significant association with greater gaps between the achievements of black and white students in these districts in both mathematics and reading [3]. A similar nationwide study on the relationship between racial segregation and achievement gaps between black and white students was performed by the National Center for Education Statistics (NCES), based on data from the National Assessment of Educational Progress (NAEP). This study found that higher levels of black student density within a school are associated with a statistically significant decline in the academic achievement levels of both black and white students [4]. The study did not, however, find that the black-white achievement gap present in schools with high black student density significantly differs from the black-white achievement gap present in schools of lower black student density [4]. Besides these nationwide studies, several studies have particularly focused on education in Connecticut. The CT Mirror, for example, has done a large amount of investigation into Connecticut magnet schools and students’ achievement there. While they did not perform any advanced statistical analysis, they compared the outcomes of magnet and non-magnet students. They found that in 2017-2018, a mere 20% of low-income students in Hartford’s traditional public schools performed at grade level and 56% were multiple grade levels behind on Connecticut’s English exam. In contrast, 35% of low-income students who attended magnet schools operated by Connecticut’s Capitol Region Education Council (CREC) performed at grade level and only 38% were multiple grade levels behind. 

Although the CT Mirror did not test for statistical significance, their finding seems to show a strong positive relationship between attendance at magnet schools and better academic performance among low-income students in the Hartford region. Possibly the most rigorous study to have been done so far specifically focusing on correlations between magnet school attendance and Connecticut students’ achievement was the result of a commission from the Connecticut State Department of Education: “Evaluating the Academic Performance of Choice Programs in Connecticut: A Pretest-Posttest Evaluation Using Matched Multiple Quasi-Control Comparison Groups”. This study followed students from 2010 to 2012 to determine whether attendance at magnet schools and other choice programs might be associated with closing the achievement gap on the standardized Connecticut Mastery Test (CMT) between urban students and their non-urban peers. They found that magnet schools operated by Regional Education Service Centers (RESCs) were most successful in closing achievement gaps between urban and non-urban students in grades [5]. Between grades 6-8, however, the study found that public charter school attendance was associated with the strongest proficiency gains (73.3% of charter students were proficient on the 2010 CMT, while 81.3% were proficient in 2012) [5]. Attendance in other choice programs did not produce statistically significant results. Our research differs from the studies we have described above in several important ways. First, in contrast to the two studies mentioned first, our research will specifically focus on Connecticut. In addition, whereas the CT Mirror’s study focused on low-income students in the Hartford region, we will focus on students from across the state and from various demographic groups. Moreover, we will explicitly investigate certain relationships that the CT State Department of Education’s research did not focus on, such as the level of integration in Connecticut schools and whether this might be correlated with some of the positive achievement gains by students in magnet schools. 

The data used in our research is publically available, found on the Connecticut State Department of Education’s website (Edsight) [6]. We use data on the achievement of students in grades 3, 4, 5, 6, 7, 8, and 11 in the school years from 2014 to 2018 on Connecticut’s newly adopted standardized test, the Smarter Balanced assessments in English language arts (ELA) and mathematics. (Data from Grade 11 was only available for the 2014-2015 school year.) Data on the number and demographics of teachers comes from EdSight as well. Data on student demographics and school location and magnet status comes from NCES’s Common Core of Data for the 2014-2017 school years and from the Connecticut Department of Education for the 2017-2018 school year [7]. For the sake of consistency, if a school were reported with magnet status in any of the school years 2014-2018, it was counted as a magnet school for all four years. (After the 2014-2015 school year, magnet schools drawing students from only one district were no longer officially reported as having magnet status, but for our study we continued to regard them as such.) Percent of students at various levels of standard achievement (Level 1 = “Not Met”, Level 2 = “Approaching”, Level 3 = “Met”, and Level 4 = “Exceeded”) is reported by year, grade, school, and race/ethnicity.  

Table 1:  Breusch-Pagan Tests – MetStandard [Grade, t] = Xb + u [Grade] + e [Grade, t].

Table 2: Summary Statistics – English Language Arts, Non-Magnet Schools.

Table 3: Summary Statistics – English Language Arts, Magnet Schools.

Table 4: Summary Statistics – Mathematics, Non-Magnet Schools.

Table 5: Summary Statistics – Mathematics, Magnet Schools.

Table 6: Black / African-American students, ELA achievement.

Table 7: Hispanic / Latino students, ELA achievement.

Table 8: Black / African-American students, mathematics achievement.

Table 9: Hispanic / Latino students, mathematics achievement.

Thus, in the tables of summary statistics, the horizontal titles represent races/ethnicities for which achievement data was obtained and the achievement level percentages represent the average achievement of members of these races/ethnicities for all schools, grades, and years in which data was obtained. The race/ethnicity and free/reduced-price lunch eligibility percentages reported represent the average student demographic makeup of their schools. It ought to be noted that much data was censored for privacy reasons. The Edsight system does not display count values less than 5. Additionally, if that rule caused the censure of just one value in a set, the system censors the second-lowest count value in that set as well. Thus, for example, if 2 students in a grade were in the Level 1 percentile, 7 in Level 2, 12 in Level 3, and 15 in Level 4, the system would only display the values for Levels 3 and 4. The value for Level 1 would have been censored because it was below 5, while the value for Level 2, as the second-lowest value, would also have been censored in order that the Level 1 value could not be calculated (since the total number of students in the grade would also be available in the data). For this reason, there is not much data about the performance of black, Hispanic, and white students in schools where they form a small minority of their grades, potentially causing biased estimations. All analysis was performed in Stata SE 15.1. To model our data, we had to choose whether to use a pooled Ordinary Least Squares (OLS) model, a fixed-effects model, or a random-effects model. We defined Grade as our panel variable and performed the Breusch-Pagan Lagrange multiplier (LM) test for random effects on the model.

where Met Standard was the percent of students who scored at Levels 3 or 4, Black Hispanic Students was the total percentage of black and Hispanic students in a school, FRPL Eligible was the total percentage of students eligible for free or reduced-price lunch programs (FRPL), Magnet was a dummy v

English Language Arts: Scatter plots such as the ones in (Figure 1) suggest that there is negative correlation between higher concentrations of black and Hispanic students and the achievement of students in both magnet and non-magnet schools (Tables 2-5). This negative correlation was only very slight for white students in magnet schools and was less pronounced for black students than for Hispanic students in both magnet and non-magnet schools. This suggests that Hispanic / Latino students are the ones most negatively impacted by segregation, while the impact on African-American students may be smaller, but still present. Running an ordinary least squares (OLS) regression model with robust standard errors, the percentage of students of black and Hispanic race/ethnicity was not shown to have a statistically significant correlation with the achievement of black / African-American students when it was the only independent variable (regression 1 in Table 6). However, magnet status alone was significant at the 5% confidence level (regression 2). When these were considered together, both coefficients were slightly larger and, again, only the school’s magnet status was statistically significant at the 5% confidence-level (regression 3). When another dummy variable, Meets Integ Standard, was introduced that indicated whether the school met integration standards (no more than 75% of the student body black or Hispanic), its regression coefficient was shown to be statistically significant at the 1% confidence level (regression 4). This indicates that all else equal, schools that met integration standards had, on average and holding all else equal, 2.93-percentage points more of black students perform at levels 3 or 4 in ELA than non-integrated schools did. The results changed significantly when a third dummy variable, Magnet Integration, was introduced that tested whether a school had both magnet status and was meeting integration standards. This variable was not statistically significant, but its inclusion resulted in much higher coefficient values for the variables indicating a school’s magnet status and whether or not it met the integration standard (regression 5). Controls for the percentage of students in the school who were eligible for free or reduced-price lunch, the school’s location, the student-teacher ratio, and the percentage of staff of black or Hispanic race/ethnicity were included for all regressions. The results seem to indicate that for black students, attendance at either a magnet school or an integrated school was associated with higher ELA achievement, without it being necessary for a magnet school to meet the integration standard. However, it is very important to realize that all of these regression results had very low R2 values, indicating that there are many other variables that could not be controlled for that likely had a much greater impact on ELA achievement levels of black students in different school environments. For Hispanic / Latino students, in contrast to black students, both the percentage of black and Hispanic students in their schools and whether the school had magnet status were shown individually to have statistically significant associations at the 1% confidence level with the number of Hispanic students who performed at or above grade level (regressions 1 and 2 in Table 7). A school’s magnet status was shown to have a stronger relationship with achievement than the school’s racial/ethnic composition (on average, 4.386-percentage points more Hispanic students reached ELA achievement levels 3 and 4 in magnet schools than in non-magnet schools), but the R2 value was slightly lower for this regression (2) than in regression 1. When these two variables were considered together, both regression coefficients were shown to be statistically significant at the 1% level, with a school’s magnet status again shown to have a much greater association than racial/ethnic composition (regression 3). When whether a school met Connecticut’s integration standard was tested rather than the percentage of students who were black or Hispanic, both this factor and magnet status had statistical significance at the 1% level (regression 4). In this regression, schools that met the integration level had, on average and holding everything else equal, 5.249-percentage points higher of Hispanic students reaching levels 3 and 4 in ELA than schools that did not meet the integration standard. When the Magnet Integration dummy variable was included in the regression, however, this interaction was not found to be statistically significant (regression 5). Rather, whether the school met the integration standard and the school’s magnet status were statistically significant at least at the 5% level and had slightly higher values than in regression 4. All of these regressions had R2 values of at least 0.39, and all included various controls.

The scatter plots shown in (Figure 2), similar to the results obtained for English language arts data, suggest that there is negative correlation between higher concentrations of black and Hispanic students and the mathematics achievement of students in both magnet and non-magnet schools. Again, this negative correlation was the smallest for white students in magnet schools and black students in non-magnet schools. Compared to when ELA achievement was tested, a school’s racial/ethnic composition appeared to be more strongly correlated with the mathematics achievement of white and African-American students and more weakly associated with the achievement of Hispanic students. As when we investigated ELA achievement, we investigated mathematics achievement with an ordinary least squares (OLS) regression model, taking into account robust standard errors. With this model, the percentage of students of black and Hispanic race/ethnicity was not shown to have a statistically significant correlation with the percentage of African-American students at a school who met or exceeded the standard in math (regression 1 in Table 8). However, the school’s magnet status alone did have a statistically significant relation with the mathematics achievement of black / African-American students (regression 2). When both of these factors were considered together, only the magnet status variable was statistically significant at the 5% confidence-level (regression 3). When another dummy variable, Meets Integ Standard, was introduced that indicated whether the school met integration standards, its coefficient was shown to be statistically significant at the 1% confidence level (regression 4). This indicates that all else equal, schools that met integration standards had, on average and holding all else equal, 3.436-percentage points more of African-American students perform at levels 3 or 4 in mathematics, compared to mathematics achievement levels of black students in non-integrated schools. This variable’s coefficient was slightly higher when a third dummy variable, Magnet Integration, was introduced that tested whether a school both had magnet status and was meeting integration standards (regression 5). This dummy variable was not statistically significant, but the coefficients on the variables independently testing for a school’s magnet status and integration status were both statistically significant at the 1% confidence level. These results seem to indicate that for black students, both integrated schools and magnet schools are associated with higher achievement, but the presence of both together may not be as significant. All regressions incorporated various controls but still suffered from low R2 values. This suggests that there are many other variables that were not controlled for and likely had a much greater impact on the mathematics Achievement levels of African-American students in different school environments. For Hispanic / Latino students, as with black students, only whether the school had magnet status, not the percentage of students of black or Hispanic ethnicity, was shown individually to have statistical significance at the 1% confidence level (regressions 1 and 2 in Table 9). A school’s magnet status was shown to have a rather strong association with mathematics achievement (on average, 3.73-percentage points more Hispanic students reached mathematics achievement levels 3 and 4 in magnet schools than in non-magnet schools). When a school’s magnet status and its percentage of black or Hispanic students were considered together, again only the school’s magnet status was shown to be statistically significant at the 1% confidence level. Next, when whether a school met Connecticut’s integration standard was tested rather than the percentage of students who were black or Hispanic, both regression coefficients had statistical significance at the 1% level (regression 4). In this regression, schools that met the integration level had, on average and holding everything else equal, 3.865-percentage points higher of Hispanic students reaching levels 3 and 4 in mathematics than schools that did not meet the integration standard. When the Magnet Integration dummy variable was included in the regression, however, this interaction was not found to be statistically significant (regression 5). Rather, whether the school met the integration standard and its magnet status were both significantly significant at the 1% level and had higher values than in regression 4. All regressions included various controls. It is important to note, however, that all of these regressions also had R2 values of less than 0.28.

The results gained from this study have many different implications, although they only indicate correlation and do not necessarily imply causation. First, they show that for black / African-American students, since R2 values were always quite small, attendance at magnet schools and/or attendance at integrated schools may not be the factors that account for the greatest variation in their achievement in either ELA or mathematics. Future studies might improve upon the results of this study by incorporating more controls and investigating the potential impact of other factors. These factors might include the amount of funding that schools receive and what that money is used for, as well as parents’ education levels and involvement in their children’s education. Unfortunately, these factors were beyond the scope of this paper. Several other controls were, however, included in this paper, including the percentage of students in a school who were eligible for free or reduced-price lunch as a measure of the economic well-being of a school’s community as well as the percentage of a school’s educators who were of African-American or Hispanic race / ethnicity. (It was thought that the presence of role models from minority racial / ethnic backgrounds might have a positive influence on the achievement of students of minority racial / ethnic backgrounds). For Hispanic / Latino students, attendance at magnet schools and attendance at integrated schools appeared to account for much more of the variance in ELA achievement levels than in mathematics achievement levels. Additionally, these factors were generally more closely associated with ELA achievement than with mathematics achievement. It is not clear why this might be so. Nevertheless, the regressions also showed that for both ELA and mathematics, whether a school met Connecticut’s integration standard had a slightly greater correlation with the achievement of Hispanic students than did the school’s magnet status, even controlling for whether a school were located in a city, suburb, or rural area. It would be interesting to investigate this result further by including more controls, such as the amount of funding each school receives and what this money is used for, since these factors vary widely between districts. Another important note is that magnet schools have various specialty areas. While many focus on improving achievement in science, technology, engineering, and mathematics (STEM), others are oriented towards performing or visual arts or other fields where standardized test scores may not be a main priority. For this reason, perhaps further studies ought to investigate black and Hispanic magnet student achievement taking into account the main academic focus area of these schools. Teacher quality is another important factor that likely accounts for much variation in students’ achievement levels but which we were unable to include in our regression analysis. Unfortunately, it is likely that our results contain additional error due to missing data for situations in which there were only a few students falling into various achievement categories and/or various racial/ethnic groups. Perhaps having full access to the data could have provided more accurate results. In addition, self-selection bias may be present because in order to attend a magnet school, a student or the student’s parent/guardian must decide to fill out an application. There are various factors which could result in the neediest students being less likely to enroll in magnet schools, such as concerns over support for English language learners or students with specific disabilities. Moreover, magnet schools often are farther away than the neighborhood schools that students would normally attend, and families without a car might choose not to try to enroll their students in magnet schools due to difficulty reaching their children if there were ever an emergency at the school.

In conclusion, this study seems to indicate that both whether a school met Connecticut’s integration standard and whether a school had magnet status were positively and statistically significantly associated with the ELA and mathematics achievement of African-American and Hispanic / Latino students. Whether the school met Connecticut’s integration standard was slightly more strongly correlated with the academic achievement of black and Hispanic students than magnet status was. However, accurate conclusions from the regression results about the achievement of African-American students in both ELA and mathematics and the mathematics achievement of Hispanic students cannot be drawn because of the low R2 values of these regressions. Further analysis must be done with more control variables included to draw more accurate and representative results.

1. Thomas RJ. Do magnet schools need white students to be great CT Mirror. 2018; 15.
2. Megan, K. Judge rules magnet school discrimination case belongs in federal court. CT Mirror. 2019; 15.
3. Owens, A. Income segregation between school districts and inequality in students’ achievement. Sociol Edu. 2018; 91: 1-27.
4. National Center for Education Statistics. School composition and the black-white achievement gap. National Assessment of Educational Progress. 2015.
5. Mooney, R. Evaluating the academic performance of choice programs in Connecticut: A pretest-posttest evaluation using matched multiple quasi-control comparison groups. Connecticut State Department of Education. 2015.
6. Connecticut State Department of Education. 2017; 14.
7. National Center for Education Statistics. Elementary Secondary information system. 2019.

Figure 1: English Language Arts Achievement by Race/Ethnicity and Magnet School Status.

Figure 2: Mathematics Achievement by Race/Ethnicity and Magnet School Status.