 
Jay P. Greene
June 5, 2006 

About the Authors His research was cited four times in the Supreme Court's opinions in the landmark Zelman v. SimmonsHarris case on school vouchers. His articles have appeared in policy journals, such as The Public Interest, City Journal, and Education Next, in academic journals, such as The Georgetown Public Policy Review, Education and Urban Society, and The British Journal of Political Science, as well as in major newspapers, such as the Wall Street Journal and the Washington Post. Jay Greene is the author of Education Myths (Rowman & Littlefield, 2005). Greene has been a professor of government at the University of Texas at Austin and the University of Houston. He received his B.A. in history from Tufts University in 1988 and his Ph.D. from the Government Department at Harvard University in 1995. He lives with his wife and three children in Fayetteville, AR. Marcus A. Winters is a Doctoral Academy Fellow at the University of Arkansas. He has performed several studies on a variety of education policy issues including highstakes testing, charter schools, and the effects of vouchers on the public school system. His oped articles have appeared in numerous newspapers, including The Washington Post, USA Today, and the Chicago SunTimes. He received his B.A. in political science with departmental honors from Ohio University in 2002. 

This study uses a widely respected method to calculate public high school graduation rates for the nation, for each state, and for the 100 largest school districts in the United States. We calculate graduation rates overall, by race, and by gender, using the most recent available data (the class of 2003). Among our key findings:


Introduction The unreliability of official public high school graduation rates is well known. It is so well known that last year, the National Governors Association (NGA) released a report that stated: “Unfortunately, the quality of state high school graduation and dropout data is such that most states cannot accurately account for their students as they progress through high school.”[1] In response, fortyfive state governors signed an agreement to implement an improved, standard calculation of the fouryear high school graduation rate. One might think that the battle has been won—that there is no longer a need for independent estimates of graduation rates, such as those that we have produced in the past and that appear in this report. But there are several reasons that we continue to need these independent estimates of public high school graduation rates. It will be many years before most states develop the data systems to accurately track students and compute graduation rates. In the interim, we will continue to need reliable estimates of graduation rates. The governors have pledged to take reasonable steps to improve graduation rate calculations until systems are in place to track individual students over time. But to ensure the proper implementation of both the immediate and longterm reforms, we will need independent estimates to verify the official statistics. We would not have recognized the need for improvement of official graduation statistics had it not been for independent estimates; and we will not know that they have, in fact, improved unless we continue to produce those independent estimates. We also continue to need reasonable independent estimates of public high school graduation rates because not everyone has accepted that the independent estimates are more reliable than official statistics. Even though most of the nation’s governors concede the point, Lawrence Mishel of the Economic Policy Institute has taken a firm stand in support of the official results and against the independent estimates.[2] Mishel’s argument is that independent estimates rely upon enrollment and diploma counts from the U.S. Department of Education’s Common Core of Data (CCD). How can we be sure, he asks, that those counts are reliable? In addition, he observes that two highquality government surveys, the Current Population Survey (CPS) and the National Educational Longitudinal Survey (NELS), produce graduation rate results that are similar to each other and significantly higher than the independent estimates based on CCD. Mishel speculates that the CCD counts may be unreliable but offers no support for his speculation. We have good reason to believe that the CCD enrollment and diploma counts are reliable. CCD establishes standards and procedures for states to collect and report enrollment and diploma data. If states do not meet those standards or follow those procedures, their data are not reported. It should not be difficult for states to track enrollment and diplomas. Enrollment counts are based on schools taking attendance, which schools are very good at doing. One reason schools are likely to keep accurate attendance is that enrollment counts are the basis for school funding by state and federal governments. Further, because attendance determines how much money state and federal governments allot to schools, these higher levels of government are inclined to check and ensure the accuracy of attendance figures. Similarly, diploma counts are likely to be accurate because it is easy for schools to count diplomas and it is easy to verify the numbers. At the very least, schools have to know how many diplomas should be printed and distributed. Mishel specifically questions our estimates of the entering ninthgrade class enrollment, which he claims are distorted by the tendency for those enrollments to be inflated because of students being held back in that grade. It is possible to run a simple check to see if our estimates of ninthgrade enrollment are on target. Using the official CCD enrollment counts, we estimate that 3,635,420 students entered the ninth grade in public school in 1999. According to the U.S. Census—in a number derived from its CPS—there were 3,892,340 fourteenyearolds in the nation in June 1999. According to the National Center for Education Statistics (NCES), 835,328 students attended private high schools (in 2001), which, divided by four, suggests that there were 208,832 ninthgraders in private school. If we subtract the private school ninthgraders from the fourteenyearold population, we are left with a difference between the number of fourteenyearolds and our estimated ninthgrade entering class of 48,088 students, or 1.3 percent. It would seem that the enrollment counts that we use are accurate. Enrollments and diplomas are easy to count accurately, and the actors have incentives to ensure that the counts are accurate—a simple check helps confirm that; on what basis does Mishel believe otherwise? He simply has more faith in graduation rates computed from CPS and NELS surveys than in those derived from CCD enrollment and diploma counts. Essentially, Mishel is arguing that we ought to believe the results from samples more than results from the population. This is exactly the opposite of standard social science practice. Normally, we expect some degree of error whenever we survey a sample drawn from a population. If we have concerns about the sample, we check the characteristics of the sample against known characteristics of the population from which the sample was drawn to ensure its validity. In this case, however, Mishel is suggesting that we ought to check the accuracy of the characteristics of the population against the characteristics in samples. Samples always involve some degree of random error, but CPS and NELS have additional, known biases for the purpose of calculating graduation rates. The NELS and CPS surveys both overstate graduation rates because they have difficulty finding and following marginalized and disadvantaged people, such as dropouts. Phillip Kaufman (the primary author of previous government calculations of graduation rates that used CPS) indicated that such a coverage bias probably exists. Specifically, dropouts are less likely to be reached by sample surveys (that is, they are “undercovered”). In a report for the Harvard Civil Rights Project, Kaufman estimated that if we made the reasonable assumption that 50 percent of those undercovered by the CPS were dropouts, we would end up with a completion rate of 80.4 percent.[3] If we then excluded GED recipients from that estimate, we would get much closer to the estimate of a 70 percent graduation rate that we and others suggest. In other words, the systematic sampling biases of CPS and NELS make their graduation numbers higher and less reliable than those derived from population counts. We can do a simple check on Mishel’s “true” graduation rates derived from CPS and “confirmed” by NELS. If Mishel is correct in saying that the true graduation rate is in the neighborhood of 90 percent,[4] there should have been about 3,678,300 diplomas awarded in 2003 from public and private high schools. According to CCD, there were only 3,062,000 diplomas given out that year. If Mishel is correct, CCD would have to have missed more than 600,000 diplomas in its count. Is it more likely that CPS and NELS suffer from a sampling bias due to the difficulty of finding dropouts, or that school systems undercounted the number of diplomas they awarded by more than 600,000, making those schools appear less successful than they actually were by nearly 20 percent? Until official graduation statistics produce more reliable estimates, it is clear that we will continue to need independent estimates of graduation rates. Those independent estimates will also help ensure progress toward improved official statistics. What’s New in This Report? While this report builds upon a foundation of previous reports, there is much that is new. First, this report contains graduation rate estimates for the class of 2003, the most recent year for which data are available. Unfortunately, CCD enrollment and diploma counts are being released with greater time lags. However, since graduation rates tend not to change dramatically in short periods of time, this study provides a valuable snapshot of the performance of public schools today. Second, in this report we are able for the first time to break out graduation rates by gender. Observers have long suspected that the graduation rate for boys is significantly lower than that for girls. CCD now contains enough information to allow us to estimate graduation rates using our method for boys and girls separately. Third, this report contains graduation rates for each of the 100 largest school districts in the country. We previously reported rates for these districts in a 2001 report, “High School Graduation Rates in the United States,” with results for the class of 1998. But in the last few national reports, we did not release results for districts. The district results in the 2001 report were based on enrollment and diploma information gathered from districts and states. After releasing that report, we had concerns about the reliability and consistency of those counts, so we refrained from producing district graduation rates in subsequent national reports. For this report, we believe that we have addressed those concerns by relying only on district information gathered from CCD. Because of the uniform standards and procedures enforced by CCD, we feel confident once again to report district results. It is important to note that no comparisons ought to be made between the district results for the class of 2003 and our previously reported district results for the class of 1998. Because those earlier results may not be reliable and were not computed using the same method as the current report, no conclusions should be drawn about any change in graduation rates for the districts. In this report, there is no need to discuss issues that we have covered in previous reports. For example, if readers are interested in our thoughts on why graduation rates are important, how officially reported rates are often mistaken, why GEDs ought not to be included in graduation rates, and other related issues, we would urge them to peruse our report “Public High School Graduation and CollegeReadiness Rates: 1991–2002.”[5] Summary of Results Though they are consistent with previous evaluations, the results reported in this paper are certain to raise many eyebrows. Overall, we estimate that only 70 percent of the students in the class of 2003 earned a high school diploma. This figure represents little change from our estimate of a 71 percent graduation rate for the class of 2002 and a 72 percent graduation rate for the class of 1991. We discovered that about 78 percent of white students and 72 percent of Asian students graduated high school, but little more than half of Hispanic and AfricanAmerican students took home a sheepskin: 53 percent and 55 percent, respectively. Further, in each racial category that we evaluate, females graduate at a higher rate than males, with a particularly large difference for Hispanic and AfricanAmerican students. An already low 58 percent and 59 percent of Hispanic and AfricanAmerican females graduated from high school in 2003; only 49 percent and 48 percent of males in these categories earned a diploma. Our districtlevel results suggest that high school graduation rates are a particular problem in our nation’s most populated school districts. For example, only about 43 percent of the 1.1 million students in New York City public school district graduate from high school. The calculations are similarly disturbing for most of the nation’s largest school systems. None of the nation’s ten largest school systems, which over 8 percent of U.S. public school children attend, graduates more than 60 percent of its students.[6] As with the nation as a whole, larger school districts uniformly graduate far fewer minority and male students than white and female students. Data and Method To calculate graduation rates for each state and several school districts, we utilize enrollment and diploma data reported by NCES, the statistical arm of the United States Department of Education. We acquired enrollments over several years by grade, race, and gender from NCES’s Common Core of Data (CCD). Unlike in previous years, diploma counts for the class of 2003 were not made publicly available, so those data were obtained from the restrictedaccess data file of the CCD.[7] The advantage of using CCD information on enrollments is that these figures are the enrollments that the states officially report to the federal government under uniform guidelines. Thus, we can have confidence that the data are accurate and comparable among the states. The disadvantage of using CCD, however, is that the data lag to the point where the most recent graduation rate calculation available is for the class of 2003. However, what is gained in the quality of the data reported likely more than outweighs the timing of the data, especially considering that high school graduation rates tend not to change substantially in a short time span. The method for calculating graduation rates is straightforward. The method takes the form: We must estimate the number of students who enter the ninth grade in 1999 instead of simply taking the reported ninthgrade enrollment in that year because researchers agree that the ninthgrade enrollment number is inflated by students repeating ninth grade. What is often referred to as the “ninthgrade bubble”—the tendency for ninthgrade enrollments to be exceptionally high compared with other grades—likely occurs because the ninth grade is the first that students must pass by earning a minimum number of credits. Thus, ninthgrade reported enrollments reflect the many students who are repeating the grade. To estimate the cohort’s ninthgrade enrollment, we cannot simply substitute the cohort’s eighthgrade enrollment because a large number of students who attend private school in the eighth grade enter public school in the ninth grade (there are far fewer private high schools, and they tend to be more expensive). Further, we cannot use only the cohort’s tenthgrade enrollment because by that time, students have already begun to drop out. To estimate the entering ninthgrade cohort for the class of 2003, we take the average reported enrollments of students in the eighth grade in 1998, ninth grade in 1999, and tenth grade in 2000.[8] The resulting “smoothed” figure provides a reasonable estimate of the entering student cohort. A large percentage of states failed to report enrollments by gender, especially in 1998, our cohort’s eighthgrade year. All but two states, however, reported high school diploma counts by gender for the spring of 2003.[9] In order to include as many states as possible in our calculation, we adopted a strategy for estimating the gender enrollments in eighth, ninth, and tenth grades—which was implemented for all states in the gender calculations. Nearly all states reported enrollments by race and overall for each of the years necessary to calculate graduation rates.[10] To estimate the enrollment by race/gender, we simply took each state’s enrollment by race and multiplied it by the percentage of fourteenyearolds in the state of that race who were male or female according to the U.S. Census in the summer before the cohort’s ninthgrade year. For example, in Arkansas in 1998, there were 26,433 white students in the eighth grade. According to computations using census data, 51.711 percent of white fourteenyearolds in Arkansas in the summer of 1999 were male. Therefore, we estimate that Arkansas had about 13,669 (or 26,433 x .51711, with rounding) white male students in the eighth grade in 1998.[11] To calculate the population change at the state and national levels, we use population estimates by age, race, and gender reported by the United States Census.[12] We take the difference between the number of seventeenyearolds in the population during the summer of 2002 (the summer before the cohort’s twelfthgrade year) and the number of fourteenyearolds in the population during the summer of 1999 (the summer before the cohort’s ninthgrade year). We then divide the resulting change in population by the number of fourteenyearolds in 1999 to get the percent increase (or decrease) in the area’s population of students in the cohort’s age group. We use a different population change computation for graduation rates by school district because population estimates by age are not readily available at the school district level. We use districtlevel enrollments as a substitute for the age populations and make the reasonable assumption that, on average, transfers in and out of a high school are equal for each grade in the school. We take the difference between the number of students in grades nine through twelve in 2002 (the cohort’s twelfthgrade year) and the number of students in grades nine through twelve in 1999 (the cohort’s ninthgrade year) and divide the resulting figure by the number of students in grades nine through twelve in 1999. This produces an estimate of the percent change in the district’s enrollment while the cohort was in high school. We then adjust the estimated ninthgrade cohort by the change in the population while the students were in high school. This produces the projected graduating cohort—the number of students who could possibly graduate with the class of 2003. Finally, we take the number of diplomas that were actually given out in the spring of 2003 and divide it by the projected graduating cohort. The result is the estimated high school graduation rate. Though this method tends to produce reliable estimates of graduation rates, it can be distorted when there are particularly small cohorts or when population changes are extraordinarily large. For this reason, we adopt and apply consistent rules for excluding cohorts for which we do not have adequate information.[13] We do not report graduation rates for cohorts of students less than or equal to 200 or when the cohort’s population change is 30 percent or greater. We also exclude any case where the cohort is less than or equal to 2,000 and the population change is 20 percent or greater. However, though we do not report graduation rates in areas with these cohort or populationchange levels, their enrollments and populations are included in the state and national calculations. It is important to clarify that the method in this paper is not a fouryear ontime graduation rate. Though the method does follow high school enrollments through four sequential grades, students who take longer than four years to graduate are estimated into the calculation as well. Such students would exit our cohort; however, they would likely be replaced by students in the previous cohort class who have also taken longer to graduate. For example, if a student who entered the ninth grade in 1999 took five years to graduate (that is, graduated with the class of 2004), he would not receive a diploma in the spring of 2003 and thus would not be included in our calculation. However, if another student entered the ninth grade in 1998 (the expected graduating class of 2002) and also took five years to graduate, that student would receive a diploma in 2003 and would thus be included in the graduation rate calculation. As long as there are not dramatic yeartoyear differences in the number of students who take longer than four years to graduate, these students should replace each other in the calculations, and any distortion should be quite limited. Thus, the result of our estimates can be thought of as the graduation rate for the class of 2003, not the ontime graduation rate for that class. Unlike many other high school graduation rate calculations, the estimates using the above method can be manipulated to interpret the high school dropout rate as well. The high school dropout rate is found by subtracting the high school graduation rate from 100. That is, a graduation rate of 70 percent implies a dropout rate of 30 percent. Other graduation rate estimates (including nearly all official government calculations) contend that the dropout rate is different from simply 100 minus the graduation rate. They produce far lower dropout estimates where many nongraduates are classified in ways other than as dropouts. However, this practice is contrary to both logic and the public’s understanding of the information that a high school graduation rate conveys. For the purposes of our calculation, a student is either a high school graduate or a high school dropout: the student earns a diploma or does not. Thus, our calculation is less confusing than many other methods, and it matches what the public and policymakers expect from a graduation rate. The above calculations were performed to produce graduation rates in total, by race, gender, and race/gender for the nation, each state, and each of the 100 largest school districts in the United States for which data were available. An Example of a StateLevel Graduation Rate Calculation An example of our calculation will illustrate the method: let us calculate the total graduation rate for New York State. First, we estimate the number of students who entered the cohort in ninth grade. In New York, the enrollment in eighth grade in 1998 was 200,097, ninth grade in 1999 was 252,864, and tenth grade in 2000 was 217,734. The average of these enrollments is 223,565, which is the estimated number of students who entered the cohort in the ninth grade. Note that the ninthgrade enrollment is much higher than either the eighthgrade or tenthgrade enrollment: this is the “ninthgrade bubble” referred to previously. Next, we compute the change in New York’s population of the cohort’s age group. In June 2002, there were 261,326 seventeenyearolds in New York; and in June 1999, there were 233,701 fourteenyearolds in the state. The difference in these populations is an increase of 27,625 children. We then divide this difference by the number of fourteenyearolds in 1999 (27,625 divided by 233,701) to get a population change of about 12 percent. We then combine our estimated ninthgrade class with the population change to produce an estimated number of students who could graduate from high school among the entering cohort. We take the estimated number of entering ninthgraders in 1999 (223,565) and multiply this number by 112 percent (100 percent plus the 12 percent population increase in the state). This produces a potential graduating class of 249,992 students. Finally, we calculate the state’s graduation rate by dividing the number of diplomas that were distributed in New York in the spring of 2003 (143,818) by the estimated number of students who could graduate in the cohort (249,992). This produces an estimated graduation rate of 57.5 percent for the state of New York for the class of 2003. An Example of a DistrictLevel Graduation Rate Calculation Since the method varies slightly, it is useful to illustrate our calculation of the districtlevel graduation rates with another example: let us calculate the total graduation rate for Los Angeles. The enrollment in Los Angeles in the eighth grade in 1998 was 45,053, ninth grade in 1999 was 58,834, and tenth grade in 2000 was 46,664. The average of these enrollments is 50,183, which is the estimated number of students who entered the ninth grade in 1999. Again, note the bubble in the ninthgrade enrollment. We next calculate the population change using the school district’s high school enrollments during the cohort’s ninth and twelfthgrade years. In 2002, the cohort’s twelfthgrade year, in Los Angeles there were 68,802 students in the ninth grade, 49,109 students in the tenth grade, 38,387 students in the eleventh grade, and 27,253 students in the twelfth grade, which totals 183,551 students in the high school grades. In 1999, the cohort’s ninthgrade year in Los Angeles, there were 58,834 students in the ninth grade, 46,971 students in the tenth grade, 36,825 students in the eleventh grade, and 28,369 students in the twelfth grade, which totals 170,999 in all high school grades in the school district. We take the number of students in high school in 2002 (183,551) and subtract from it the number of high school students in 1999 (170,999) to get an increase in the population of 12,552. We then divide this figure (12,552) by the number of high school students in 1999 (170,999) to get a population increase of 6 percent. Next, we adjust the estimated entering ninthgrade class by the increase in the Los Angeles school district’s population. We take the estimated ninthgrade cohort (50,183) and multiply it by 106 percent (100 percent plus the 6 percent population increase) to get an estimated potential graduating cohort of 53,150 students. Finally, we divide the number of regular diplomas that were granted by the Los Angeles school district in the spring of 2003 (27,563) by the number of students we estimated could potentially graduate in the cohort (53,150). This produces an estimated graduation rate of 51 percent for the Los Angeles school district in 2003. Results The results of our statelevel and national calculations of graduation rates overall, by race, gender, and race/gender are reported alphabetically by state in Table 1. The national overall graduation rate is about 70 percent, which is in line with calculations from previous years. Nationally, about 78 percent of white students and 72 percent of Asian students graduated with a regular diploma in the class of 2003, compared with the much lower estimates of 53 percent for Hispanic students and 55 percent of AfricanAmerican students. Female students graduated at a rate of about 72 percent, compared with males at about 65 percent. The race and gender gaps in high school graduation also held when evaluating by race/gender. At only 48 percent, AfricanAmerican male students reported the lowest graduation rates of any subgroup nationally, while white female students had the highest graduation rate, at 79 percent. The disparity between male and female graduation rates was much higher for AfricanAmerican (females, 59 percent; males, 48 percent) and Hispanic (females, 58 percent; males, 49 percent) students than for Asian or white students. Table 2 ranks the states by overall high school graduation rate. The table shows that graduation rates differed substantially among the states. New Jersey had the highest overall graduation rate (88 percent) and was followed by Iowa, Wisconsin, and North Dakota, each at 85 percent. The lowest overall graduation rate was in South Carolina (54 percent), followed by Georgia (56 percent) and New York (58 percent). Some states fared well overall but had low graduation rates for certain populations of students. For example, Wisconsin ranked third in the nation for overall graduation rate mostly because it had the highest graduation rate for white students. However, of the thirtythree states for which the necessary information was available to calculate graduation rates for AfricanAmerican students, Wisconsin ranked thirtysecond. Conversely, Texas ranked thirtysixth in the nation in overall graduation rate but had the fifthhighest graduation rate for AfricanAmerican students among the thirtythree states for which adequate information was available. Graduation rates overall and for each subgroup for the 100 largest school districts (and a few other districts of interest) are reported in order of the district’s total enrollments in 2002 in Table 3, and alphabetically in Table 4.[14] The appearance that larger school districts have lower graduation rates is confirmed by a simple Pearson’s correlation, which finds a negative correlation between total enrollment and total graduation rate of 0.32. However, one should be very cautious in making a conclusion about the role of district size on graduation rates from such a calculation, since this does not account for differences in the populations of students educated in these districts. Larger districts, for example, could have a much higher proportion of minority students, which might lead to lower overall graduation rates. Table 5 ranks the 100 largest school districts by their overall graduation rate. Among the 100 largest school districts, Davis (UT) has the highest graduation rate, at 89 percent, followed by Ysleta (TX) at 84 percent and East Baton Rouge Parrish (LA) at 83 percent. The lowest graduation rate of the nation’s 100 largest school districts was in San Bernardino (CA), at 42 percent; Detroit (MI) was also at 42 percent, and the nation’s largest school district, New York City, at 43 percent. Conclusion The graduation rate estimates for the class of 2003 reported in this paper confirm that far fewer students graduate high school than is often realized. It is important for policymakers and the public to understand that only about 70 percent of all students and a little more than half of Hispanic and AfricanAmerican students graduate from high school. While it is not the place of this report to provide guidance on how to improve high school graduation rates, these results do suggest that there is a graduation problem that needs to be addressed. Another interesting finding in this report is the difference in high school graduation rates between males and females. Females graduate at higher rates for each racial subgroup analyzed in this report, but the gender gap in high school graduation is particularly large for Hispanic and AfricanAmerican students. The reasons for this gap should be addressed in future research. Finally, our calculation of high school graduation rates for the 100 largest school districts suggests that the graduation problem is centered primarily in the nation’s largest school districts. Only one of the nation’s ten largest school districts in the nation—where more than 8 percent of all students attend school—graduates more than 60 percent of its students. We are not able in this report to define the reasons for such low graduation rates in our nation’s largest school systems; but clearly, if the public is to improve high school graduation rates, it would do well to focus its efforts on the education provided in these urban areas. Endnotes  
(click on the tables for a larger version) 
Department of Education Reform University of Arkansas 201 Graduate Education Building Fayetteville, AR 72701 http://www.uark.edu/ua/der  Ph: 479/5753172 Fax: 479/5753196 edreform@uark.edu 