Abstract

Each year our college system gives twice as many bachelor’s degrees (BAs) to African American females as males. Large gender gaps favoring women also exist for other groups. Indeed, on average our college system gives 35% more BAs to females than males across all racial and ethnic groups. Traditional explanations for this pattern have focused on what happens in college. In this brief we note that at least part of the story can be found in differences in high school graduation rates. While the differences in high school graduation rates are far smaller, the variations by race and gender follow the same pattern as those found in college. We find that our high schools are graduating about 23% more African-American females than males each year and our estimates suggest about a 4% difference that favors females across all ethnic groups.

In order to conduct this analysis we build on methods of calculating public high school graduation rates used in previous Urban Institute work and enhance them by adding in graduates from private high schools.

---

Introduction
On average, males do far better than females in the labor market. Consequently differences favoring females in college graduation are not necessarily a cause for concern. However, the evidence we have seen suggests there should be concern for minority males who are not doing well in the labor market and who are graduating at rates far lower than their female counterparts.

In this brief we present new evidence on graduation rates from high school and show how these data map onto college graduation by race and gender. While the gender differences we find are much smaller in high school than in college, the gender differences across races are similar and raise additional concerns, especially for minority males.

College Graduation
As Table 1 shows, large gaps exist in the number of college graduates by race and gender.

In 2003-2004 over 80,000 African-American females received BAs in the United States. Only half as many degrees were given to their male counterparts. The differences are far smaller for other racial/ethnic groups and are even reversed for nonresident aliens. However, a large and substantial gap still remains overall favoring females who received around 800,000 BAs compared to males who received only about 600,000.

Table 2 shows that the advantages for females are even larger when one considers other types of post-secondary degrees including associate’s and master’s degrees. The only exception is for PhDs and other professional degrees where men still outnumber women slightly, but PhD degrees represent less than 4 percent of all degrees awarded in the U.S.

Males also outnumber females in PhDs and professional degrees within racial and ethnic subgroups with one exception: African American females receive more PhDs and professional degrees than their male counterparts by fairly wide margins (see Table A.1).

High School Graduation
The gap between males and females in college graduation may well be due in part to conditions of the college experience. In addition, however, we see evidence that while the gender differences in high school graduation are far smaller than for college graduation, many of the patterns in the data are similar—the overall gap favors females, the gender gap is far larger for blacks than for whites, and the gap is somewhere in between for students of other races and ethnicities. Table 3 presents numbers of high school graduates based on public school data from the 2000-2001 school year and private school data from 2001-2002. Using these data, we estimate that on average, 23 percent more black females graduate from high school than black males. For whites, the ratio of male to female high school graduates is almost exactly 1 (0.995, to be exact). For students of other races and ethnicities, the ratio is 1.09, meaning that about 9 percent more females graduate than males. Overall females have about a four percent advantage compared to males.

High School Graduation Rate Calculation Enhancement
Previous research has looked at public high school graduation rates by race and gender (Orfield et al, 2004; Greene and Winters, 2006) using methods similar to those used in the “Wall Chart” in the 1980s (Ginsburg et al, 1988). Those estimates are all based on the number of graduates divided by the number of 9th graders four years earlier, with various types of adjustments [1]. These methods have four weaknesses. First, they estimate graduation rates of 9th graders and thus omit students who drop out before 9th grade. Second, they include students held back in 9th grade, increasing the size of the denominator meaning that the rates may be biased downwards [2]. Third, they omit private school students. Finally, they may be biased by migration between public and private schools [3].

In this brief we present a way of calculating national high school graduation rates that overcomes all of these difficulties. This method uses reports of the total number of graduates (public and private) divided by an estimate of the population of potential graduates. This Simple Graduate Ratio (SGR) method has been used in some earlier work (Chaplin, 2002) and has been reported for a number of years by the U.S. Department of Education (National Center for Educational Statistics, 2001). It overcomes the problems of students dropping out before 9th grade or being held back in 9th grade by using an estimate of the total population as the denominator instead of an estimate of the number of first time 9th graders. It allows for estimates that account for private school students and overcomes the problem of migration between public and private schools by including graduates from both types of schools in the numerator. This is the first report to present estimates based on the SGR method by race and gender.

Method
The main reason previous research has not between able to provide very precise estimates of high school graduation rates, including private schools by race and gender, is that the only data containing information on the universe of private school graduates is the Private School Survey (PSS) and these data do not include numbers of graduates broken down by these categories [4]. In order to overcome this limitation we simulate private school graduates by race and gender using information that is available in the PSS on the total number of graduates, the numbers of students enrolled by gender, and the numbers of students enrolled by race/ethnicity.

In order to do this simulation we need to have data that enables us to estimate the relationships between the numbers of graduates by race and gender and the data we have in the PSS. To estimate these relationships, we use data from the CCD for public school districts that appear similar to private schools based on their size and racial and gender make-up. Public school districts with characteristics similar to those of private schools may not have identical graduation rates so these estimates may be off somewhat. However, we suspect that the remaining differences in graduation rates (after controlling for race, gender, and school size) are likely fairly small. In addition, private school graduates comprise only 12.7% of the total graduates in our sample. Thus, these remaining differences affect small fractions of our total estimates.

We used a four-step method to estimate graduates by race and gender in private schools. First we identified a set of districts similar to private schools based on their size [5]. This was done to help ensure that the patterns seen in the public school districts would be more similar to the patterns in the private schools. Second, we created baseline estimates of graduates by race and gender for each public school district in our sample based on their overall racial/ethnic and gender make-up and their total number of graduates, variables that are available in both the CCD and PSS. This was done to give us a good starting point for our final estimates. Third, we used this variable and a number of others that we had in both the CCD and PSS in a regression to predict the actual number of graduates by race and gender. This third step is designed to deal with the fact that our baseline estimates are likely to be off because graduation rates vary substantially by race and gender within each school and by the racial and gender make-up of the school. Finally, we used the results of this regression to predict graduates by race and gender in each private school in our sample.

Step 1: The subset of districts used in the regression was a random subset of districts chosen to resemble the distribution of private school enrollment by a pre-chosen set of size categories. All districts had to have students enrolled in 12th grade. Within this set of districts we randomly selected districts based on 9th grade enrollment categories so that the number of districts in each category roughly matched the number of private schools in that same size category. For example, any district with 1 to 20 students in grade 9 was automatically included in the sample, districts with between 21 and 30 students in 9th grade had a 20% chance of being included, districts with 31-40 such students had a 30% chance, and districts with 41-53 such students had a 75% chance. No district with more than 727 9th grade students (the maximum number in the PSS) was included in our sample [6]. This resulted in a subset of 2,622 districts whose distribution of 9th grade students looked roughly similar to that for private schools based on the size categories we chose.

Step 2: Our baseline estimates of graduation numbers by racial-ethnic group and gender took the following form:

= total graduates in district*(fraction of students of race r in district) * (fraction of students of gender g in district)

Step 3: We then ran a regression of actual graduates by race and gender in the district a_rg on our baseline estimate () with controls for the fraction of that gender (f_g), the fraction of that race (f_r), and interactions of these variables with our baseline estimate. Thus, our regression took the following form:

This regression was run separately for each racial/ethnic and gender category. We used three race/ethnicity groups: Black non-Hispanics, White non-Hispanics and Others where Others includes students identified as Asian, Hispanic, or Native American.

The r² values for our regressions ranged between .75 for white females to .97 for black females. Nearly all coefficient estimates were statistically significant at the p < .0001 level. Three coefficient estimates(4 for white females,3 for other males, and 3 for other females) failed to reach statistical significance at the p < .10 level. These regression results can be found in Table A.2.

Step 4: These models were then used to predict the number of graduates by race and gender at private schools using analogous variables at the school level and the coefficients from the relevant model [7]. These numbers were then totaled with the original public school numbers to obtain national totals for each race-ethnicity and gender group.

To estimate graduation rates using the SGR method by race and gender, the number of estimated graduates was divided by an estimate of the size of the 17-year-old population based on 2000 Census data. We did not use 18-year-olds because they include recent immigrants to the U.S. who would not have had a reasonable chance to graduate from high school in the U.S. The 17-year old population includes fewer immigrants than the 18-year old population and those immigrants that are there would have a better chance of attending U.S. high schools [8]. We use data on public school graduates from the class of 2000 and private school graduates from the class of 2001. These data are provided in the 2000-2001 CCD and 2001-2002 PSS respectively.

Limitations
This new method of calculating high school graduation rates has limitations. First, because it requires combining public and private school graduates, it does not allow for estimation of separate rates by school type. Separate rates can be calculated using these data, but not without making strong assumptions regarding migration between the public and private schools. Second, we assume that the ratios of graduates by race and gender are similar between public districts and private schools, conditional on our control variables. On the other hand, it should be noted that the method allows graduation rates to differ by race and by gender (separately) within schools and by school type across schools, both unconditionally and also conditional on the variables in the model.

Results
Table 3 presented the numbers of graduates by race and gender using this method. Table 4 presents the graduation rate estimates using the method described above and two other methods of estimating high school graduation rates that have been used in recent reports.

The first column contains results from Orfield et al (2004) [9]. These are estimates of on-time graduation rates. Thus, it is not surprising that the rates are somewhat lower than those in the remaining two columns, which are intended to estimate overall graduation rates including students who took 5 or more years to graduate, as well as those who graduated on-time (generally in 4 years) [10]. The first and second columns contain estimates of public school graduation rates out of students who enter 9th grade while the last column contains an estimate of the total graduation rate including public and private schools starting at birth. Some students drop out before 9th grade so one might expect the first two columns to be larger [11]. At the same time, private school graduation rates appear to be higher than public school graduation rates, especially for minorities (Grogger and Neal, 2000). This may explain why the last column has the highest estimates overall and for each subgroup [12]. What is particularly relevant for this paper is that the ratios of the female to male graduation rates remain fairly constant regardless of the methods used. The ratio is largest for African-Americans, ranging from 1.23 to 1.31 and smallest for Whites, ranging from 1.06 to 1.09. Thus, the same basic story remains—females are graduating from high school at higher rates than males and the difference appears to be much larger for African Americans than it is for Whites.

All three methods of estimating high school graduation rates used in Table 4 suffer from using a numerator (graduates) that includes students who may not be in the denominator (9th graders 4 years earlier or 17-year-olds 1 year earlier). Thus, none of these methods is ideal for estimating cohort-specific graduation rates, though they may still be better than many alternatives. In the absence of major changes in cohort sizes, graduation rates, and the rates at which students are held back, these mismatches should more or less cancel each other out for the SGR method. In contrast, the non-SGR methods can produce biased estimates even with a constant rate of students being held back because student retention inflates their denominators. This could affect the gender ratios if males are held back at different rates than females. As shown above the gender ratios are fairly similar across methods with the exception that the Greene method has a lower ratio for Blacks.

Like the other methods used in Table 4, the SGR method can be affected by immigration in two important ways, both of which relate to comparisons with alternative datasets that have been used to estimate high school graduation rates by the U.S. Department of Education. First, the SGR method will underestimate the graduation rate of a cohort of students who were in the U.S. in 8th grade if many 17-year-olds enter the U.S. after 8th grade and end up not graduating from high school. This is relevant because the U.S. Department of Education uses data from the National Longitudinal Survey of 1988 to estimate such graduation rates and ends up with estimates around 82% (Mishel and Roy, 2006), over ten percentage points higher than any of the estimates in Table 4. However, only about 2.5 percent of 17-year-olds in the U.S. are foreign born and immigrated to the U.S. in the last 4 years and many of them do graduate from high school [13]. This suggests that graduation rates estimated using the SGR method are unlikely to be off by more than one or two percentage points because of immigration compared to the true graduation rate of a cohort of 8th graders (assuming fairly stable graduation rates over a period of time). However, if immigration rates differ by gender they could have a larger impact on the gender ratios of graduates. The NELS data suggest an overall female to male gender ratio of around 1.03 (McMillen and Kaufman, 1996), substantially lower than the ratios in Table 4.

The SGR method may also not match up with graduation rates of 25-34-year-olds such as those often reported based on Current Population Survey data, again because of immigration—but in this case largely because of people who immigrate to the U.S. after the age of 17. However, since immigrants tend to have lower rates of high school graduation (Laird et al, 2006) this would suggest that the CPS estimates should produce lower graduation rate estimates [14]. Instead the CPS estimates are much higher, suggesting that other forces must explain the differences, the major one of which is that the CPS estimates generally include GED certificates which are excluded in the calculations above [15]. In any case, CPS estimates also suggest that more females graduate from high school than males with overall ratios of 1.03 to 1.04, very similar to those from the NELS data, but much lower than those shown in Table 4 (Kaufman et al, 2000 and Laird et al, 2006) [16]. In future work we hope to investigate whether immigration rate differences by gender could help to explain differences in the gender ratios of high school graduates reported by the NELS and CPS data compared to those based on CCD data.

Conclusion
Previous work has shown large differences in graduation rates by gender and race in both high school and college that favor women, especially among African-Americans. We confirm these gaps and show that they exist for almost all types of college degrees and that they are found even when one adds in private schools to the calculation of high school degrees. The fact that the gaps emerge in high school suggests that important factors are affecting gender gaps before students enter post-secondary education. Given the continued success of men in the labor market it is not clear if these gender gaps are important for society as a whole. However, the labor market problems of African American men may be related to their relatively poor high school graduation rates, so at least for this particular subgroup earlier educational interventions are likely to be important.

---

Footnotes (click on a footnote number to return to the paper)

1. For example, the method used by Swanson in Orfield et al (2004) can be written as graduates over 9th graders 4 years earlier times the ratios of 12th graders in the current year divided by 12th graders in the previous year, 11th graders in the previous year over 11th graders in the year before that and 10th graders 2 years ago over 10th graders 3 years ago.
2. The method used in Orfield et al (2004) is designed to estimate on-time graduation rates but may be biased up because it includes all graduates in the numerator, including those who did not graduate on time. This upward bias is offset to some extent by the inclusion of students held back in 9th grade in the denominator. The method used by Greene and Winters (2006) averages grades 8-10 enrollment instead of just using grade 9 enrollment in the denominator. Since students can be held back in all three grades, this method may still produce some downward bias..
3. The potential importance of this type of migration was stressed in recent work by Engberg and Gill (2006).
4. Relatively imprecise estimates of private school graduation rates by race and gender can be obtained from survey data that cover only a fraction of the population. Based on analyses of public school graduation rates these data may produce biased estimates (Warren 2005; Greene and Winters 2006; Swanson and Chaplin 2003) though this possibility is the source of some debate (Mishel and Roy, 2006).
5. We could not use data on public schools because the CCD does not report on graduates by school—only by district.
6. The remaining cut-points for the size categories were 80, 122, and 184.
7. Estimates were not possible for Arizona, Kentucky, New Hampshire, South Carolina, Tennessee, Vermont, and Washington because these states did not provide race information to the CCD for the 2000-2001 school year.
8. Indeed, in 2000 the high school graduation rate of immigrants age 18 who arrived after 1996 was over 30% compared to only 11% for those age 17.
9. Footnote 35 in that report suggests that those rates were calculated without AZ, ID, NH, and VT. When we calculate graduation rates using our method and that of Orfield et al (2004) using state-level data and the same set of states for both methods we get numbers that are very similar to those shown in Table 4, especially for the gender ratios.
10. The definition of on-time varies and often allows Special Education students additional years.
11. About 5 percent of 25-29 year olds report having dropped out before 9th grade but many of these people may be recent immigrants (Mishel and Roy, 2006).
12. See Appendix A.3 for a breakout of estimated numbers of graduates between public and private schools.
13. This is based on unpublished Urban Institute tabulations of March 2004 Current Population Survey Data, available upon request.
14. More precisely, Hispanic immigrants have lower graduation rates. Non-Hispanic immigrants actually have higher graduation rates but the bulk of immigrants are Hispanic so the overall graduation rate of immigrants (people born outside of the U.S.) is well below the U.S. average.
15. The Department of Education used to present estimates of high school graduation rates without GEDs in their annual dropout reports (Kaufman et al, 2000) but stopped doing so because of concerns about the large mismatches between the total numbers of GEDs reported in the CPS and the numbers reported by the GED testing service (Laird et al, 2006).
16. The ratios are 1.03 or 1.04 depending on whether or not GEDs are included. Interestingly, it appears that males are more likely to get GEDs than females (Kaufman et al, 2000).

---

References

• Chaplin, Duncan. “Tassels on the Cheap.” Education Next, 2002, pp. 24–29.
• Engberg, John and Brian Gill (2006) “Estimating Graduation and Dropout Rates with Longitudinal Data: A Case” RAND Working Paper, WR-373-PPS, July.
• Ginsburg, A. L., J. Noell, and V.W. Plisko (1988) “Lessons from the Wall Chart,” Educational Evaluation and Policy Analysis, 10(1):1-12.
  
• Greene, J. P., & Winters, M. A. (2006) Leaving boys behind: Public high school graduation rates. New York: Manhattan Institute for Policy Research [Online]. Retrieved from www.manhattaninstitute.org/pdf/cr_48.pdf on June 6, 2006.
  
• Grogger, J. and D. Neal (2000) "Further Evidence on the Effects of Catholic Schooling." Brookings/Wharton Papers in Urban Economics 1: 151-193.
  
• Kaufman, P., Jin Y. Kwon, Steve Klein, and Christopher D. Chapman (2000) Dropout Rates in the United States: 1999 (NCES 2001-022), U.S. Department of Education. National Center for Education Statistics. Washington, DC: U.S. Government Printing Office, November.
  
• Kaufman, P., Alt, M. N., & Chapman, C. (2005) Dropout rates in the United States: 2001 (NCES 2005-046). U.S. Department of Education. National Center for Education Statistics. Washington, DC: U.S. Government Printing Office [Online]. Retrieved from http://nces.ed.gov/pubs2005/2005046.pdf on June 6, 2006.Kaufman, P. et al. 2005.
  
• Laird, J., Lew, S., DeBell, M., and Chapman, C. (2006). Dropout Rates in the United Sates: 2002 and 2003 (NCES 2006-062). U.S. Department of Education, Washington, D.C. National Center for Education Statistics, June.
  
• McMillen, Marilyn M. and Phillip Kaufman (1996) Dropout Rates in the United States: 1994, (NCES 96-663) U.S. Department of Education. National Center for Education Statistics. Washington, D. C.
  
• Mishel, Lawrence and Joydeep Roy (2006) Rethinking High School Graduation Rates and Trends, The Economic Policy Institute, Washington, D.C.
  
• National Center for Education Statistics (2001) Digest of Education Statistics, NCES 2001-34, U.S. Department of Education, Washington, D.C.
  
• Orfield, G., D. Losen, J. Wald, and C.B. Swanson, (2004) Losing our future: How minority youth are being left behind by the graduation rate crisis. Joint release by the Civil Rights Project at Harvard University, the Urban Institute, Advocates for Children of New York, and the Civil Society Institute [Online]. Retrieved from www.urban.org/UploadedPDF/410936_LosingOurFuture.pdf on September 10th, 2006.
  
• Swanson, C.B., and D. Chaplin (2003) “Counting High School Graduates When Graduates Count: Measuring Graduation Rates Under the High Stakes of NCLB.” Paper presented at the National Economics Association Annual Meeting, Washington, DC, January.
  
• Warren, John Robert (2005) “State-Level High School Completion Rates: Concepts, Measures, and Trends.” Education Policy Analysis Archives, 13(51): 1-38.
  

---

Table 1 (click here to return to the paper)

Table 2 (click here to return to the paper)

Table 3

Table 4