ebook img

ERIC ED563073: The SAT® as a Predictor of Different Levels of College Performance. Research Report No. 2006-3 PDF

2006·0.55 MB·English
by  ERIC
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview ERIC ED563073: The SAT® as a Predictor of Different Levels of College Performance. Research Report No. 2006-3

Research Report No. 2006-3 The SAT® As a Predictor of Different Levels of College Performance Jennifer L. Kobrin and Rochelle S. Michel www.collegeboard.com College Board Research Report No. 2006-3 The SAT As a ® Predictor of Different Levels of College Performance Jennifer L. Kobrin and Rochelle S. Michel The College Board, New York, 2006 Jennifer L. Kobrin is a research scientist at the College Board. Rochelle S. Michel is a doctoral student at Fordham University. Researchers are encouraged to freely express their professional judgment. Therefore, points of view or opinions stated in College Board Reports do not necessarily represent official College Board position or policy. The College Board: Connecting Students to College Success The College Board is a not-for-profit membership association whose mission is to connect students to college success and opportunity. Founded in 1900, the association is composed of more than 5,000 schools, colleges, universities, and other educational organizations. Each year, the College Board serves seven million students and their parents, 23,000 high schools, and 3,500 colleges through major programs and services in college admissions, guidance, assessment, financial aid, enrollment, and teaching and learning. Among its best-known programs are the SAT®, the PSAT/NMSQT®, and the Advanced Placement Program® (AP®). The College Board is committed to the principles of excellence and equity, and that commitment is embodied in all of its programs, services, activities, and concerns. For further information, visit www.collegeboard.com. Additional copies of this report (item #060481783) may be obtained from College Board Publications, Box 886, New York, NY 10101-0886, 800 323-7155. The price is $15. Please include $4 for postage and handling. © 2006 The College Board. All rights reserved. College Board, Advanced Placement Program, AP, SAT, and the acorn logo are registered trademarks of the College Board. connect to college success and SAT Reasoning Test are trademarks owned by the College Board. PSAT/NMSQT is a registered trademark of the College Board and National Merit Scholarship Corporation. All other products and services may be trademarks of their respective owners. Visit the College Board on the Web: www.collegeboard.com. Printed in the United States of America. Figures Contents 1. Overall accuracy rates for total sample ....... 2 2. Overall accuracy rates for successful Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 students .................................. 3 3. Overall accuracy rates for unsuccessful Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 students .................................. 3 4. Accuracy rates for successful females......... 4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 5. Accuracy rates for successful males .......... 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 6. Accuracy rates for successful Asian American students......................... 4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 7. Accuracy rates for successful African American students......................... 5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 8. Accuracy rates for successful white students .................................. 5 Tables 9. Accuracy rates for successful Hispanic 1. Percentage of Students Achieving Various students .................................. 5 Levels of Success........................... 2 10. Accuracy rates for successful students 2. Summary of Results for Predicting attending very selective colleges ............. 5 Successful Students ........................ 3 11. Accuracy rates for successful students A1. Accuracy Rates for Logistic Regression attending moderately selective colleges ....... 5 Models for Total Sample.................... 8 12. Accuracy rates for successful students A2. Accuracy Rates for Logistic Regression attending least selective colleges ............. 5 Models by Gender ......................... 8 A3. Accuracy Rates for Logistic Regression Models by Racial/Ethnic Group ............. 9 A4. Accuracy Rates for Logistic Regression Models by College Selectivity............... 10 Introduction of predicting FGPAs of 3.5 or higher. The study also found that both ACT and high school grades are effective The SAT Reasoning Test™ (SAT®) is the most widely used predictors when success is defined as a FGPA of C (2.0) standardized test for college admissions. Research on the or better, but that in this case, high school grades were validity of SAT scores for predicting college performance actually a better predictor. has been conducted over several decades, going back Based on the same hypothesis offered by Noble and to the 1950s and continuing to this day. Fishman and Sawyer (2002), this study examined the validity of the Pasanella (1960) reviewed some of the earliest SAT SAT for predicting various levels of student success in predictive validity studies, and Young (2001) provides college, as measured by FGPAs. Logistic regression was the most recent review. Hezlett et al. (2001) performed used to predict the probability that a student would be a comprehensive meta-analysis of approximately 3,000 successful or unsuccessful in achieving a FGPA at various validity studies with more than 1 million students, and levels, based on that student’s SAT scores and high found that the SAT is a valid predictor of first-year college school grade point average (HSGPA). Although most grade point average (FGPA), with multiple correlations predictive validity studies employ multiple regression ranging from .44 to .62. as the main analytic procedure, this study employed Colleges typically use both high school grades and logistic regression. In multiple regression, the focus is scores on standardized tests such as the SAT to predict a on prediction of a value on the dependent variable. One student’s probability of success in college. One of the most prediction equation is produced for the sample used in persistent questions regarding the validity of the SAT is the analysis and prediction occurs on the full range of whether the SAT adds substantially to the prediction of values of the dependent variable. In this study, however, college success after high school grades are taken into the primary research question focused on whether the account. To address this question, the majority of SAT SAT or HSGPA is a better predictor of FGPA for students predictive validity studies examine the relationship of with high FGPAs as compared to students with lower SAT scores and high school grades to first-year college FGPAs. Therefore, it was necessary to dichotomize the grades. Many of these studies find that SAT scores make dependent variable to make the distinction between low a substantial contribution to the prediction of FGPA, and high levels of FGPA, and use logistic regression to and that using admissions test scores such as the SAT predict the probability of success at each level. in combination with a measure of high school grades produces higher validity coefficients than using either measure alone (Camara, 2005). Method Although high school grades and SAT scores are highly correlated, these two indicators measure slightly different constructs and are expected to be related to Analyses were conducted using data from 30 colleges college achievement in different ways. The SAT measures that agreed to participate in a multiyear validity study reasoning ability and educational achievement related to sponsored by the College Board. The sample consisted successful performance in college. High school grades of approximately 34,000 students who entered college measure educational achievement, but also measure in the fall of 1995. The sample was 53 percent female noncognitive factors such as effort, attendance, conformity, and 47 percent male, and approximately 75 percent and motivation (Stiggins, Frisbie, and Griswold, 1989). Based white/Caucasian, 10 percent Asian American, 5 percent on the research of Goldman and others (e.g., Goldman and African American, 5 percent Hispanic, 1 percent Native Hewitt, 1975; Goldman, Schmidt, Hewitt, and Fisher, 1974; American, and 4 percent other or unknown. The gender Goldman and Widawski, 1976), Noble and Sawyer (2002) distribution in the sample was very similar to the 1995 hypothesized that while average college grades are more population of college-bound seniors; however, the ethnic likely to reflect noncognitive components, high college distribution differed somewhat. The sample had fewer grades are more likely to reflect cognitive achievement and African American and Hispanic students and more white less likely to reflect noncognitive factors. Based on this students than the 1995 population. hypothesis, Noble and Sawyer expected that predictions The sample was of higher ability than the population of moderate first-year college grades would be better of college-bound seniors, with a mean SAT verbal score when based on high school grades, while predictions of 577 (compared to 504 in the population) and a mean of high FGPAs would be better when based on college SAT mathematics score of 585 (compared to 506 in the admissions test scores (in their case, the ACT). They used population). The standard deviations of SAT verbal and logistic regression to study how well the ACT and high mathematics scores were slightly smaller in the sample school grades predicted various levels of FGPAs (e.g., 2.0 (91 for SAT-V and 93 for SAT-M) than in the population or higher, 3.5 or higher). Their results indicated that ACT (113 and 112 for SAT-V and SAT-M, respectively). The composite scores did a better job than high school grades mean self-reported high school grade point average for the  Table 1 sample was 3.5 with a standard deviation of .53 (compared Percentage of Students Achieving Various Levels of to a mean of 3.19 and standard deviation of .66 in the Success population). The mean FGPA for the sample was 2.8 with a standard deviation of .72 (comparison means for the FGPA Greater than or Equal to: population were not available). Subgroup N 2.00 2.50 3.00 3.25 3.50 3.75 Females 8,09 89.7 73.4 47.3 3.4 7.8 6.7 Logistic regression analyses were performed six times Males 5,860 84. 65. 39.5 6.4 5.5 6.6 on each subgroup using different levels of FGPA as the Native criterion variable. The criterion variable was whether the 65 8.5 55.8 30.6 8.5 9.4 .6 American student earned a FGPA greater than or equal to 2.0, 2.5, Asian 3.0, 3.25, 3.5, or 3.75. The regression models included the 3,36 89. 73.3 49.4 34.4 0.3 8.0 American following predictors: 1) HSGPA only; 2) SAT verbal and African ,776 74.5 48.9 .8 .8 5. .7 math scores only (SAT); and 3) SAT and HSGPA. The American logistic regression analyses were performed on the group Hispanic ,60 8.0 6.3 34. 9.7 .0 3.4 as a whole, and separate analyses were conducted on each White 5,60 88. 7. 45. 30. 7.5 7. gender and racial/ethnic group. Other 74 87.6 7.8 47.4 34.3 9.7 6.6 Because selective colleges may place more weight on TOTAL N 34,069 9,67 3,689 4,870 9,894 5,706 ,67 cognitive components in the grading of students, separate TOTAL % ****** 87. 69.5 43.6 9.0 6.7 6.7 analyses were also performed on subgroups based on the selectivity of the college they attended. The information example, Asian American students were more likely than used to calculate an index of selectivity for the colleges other subgroups to obtain FGPAs at or above all of the and universities in the sample was taken from the College criterion levels, while African American students were Board’s 1995 College Handbook. The selectivity index is least likely to obtain each of the six criterion levels. the ratio of the number of students who were accepted to Figure 1 and Appendix Table A1 show the accuracy the college/university compared to the number of students rates for the various logistic regression models for the who applied to that college/university. Three subgroups of total sample. The accuracy rate represents the proportion students based on their colleges’ selectivity were created: of test-takers who were correctly classified, either as Level 1 included students attending the most selective successful or unsuccessful.2 An overall summary of the colleges (those accepting less than half of their applicants). results is presented in Table 2. This table shows whether Level 2 included students attending colleges with average the SAT or HSGPA was a better predictor of successful selectivity (those accepting between 50 and 75 percent of their applicants), and Level 3 included students attending 100 the least selective colleges (those accepting more than 75 HSGPA SAT Both percent of their applicants). There were 4 Level 1 schools, 11 Level 2 schools, and 11 Level 3 schools in the sample.1 80 Results 60 The results focused on the accuracy of the logistic regression models to correctly classify the individuals as 40 successful or unsuccessful in their first year of college. Table 1 gives the percentage of students in the sample by gender and racial/ethnic group achieving the various levels 20 of success as defined by their FGPA. At the lowest level of success, approximately 87 percent had FGPAs at or above 2.0; while at the highest level of success, about 7 percent had FGPAs at or above 3.75. The percentages varied for 0 2.00 2.50 3.00 3.25 3.50 3.75 the different gender and racial/ethnic subgroups. For Figure 1. Overall accuracy rates for total sample. 1 The 1995 College Handbook did not have the information necessary to compute the selectivity index for 4 of the 30 institutions in the sample. The average selectivity is about .68, meaning about 68 percent of those who applied were admitted. The most selective school had an index of .22, admitting only 22 percent of those who applied, and the least selective school in the sample had an index of .95, admitting 95 percent of those who applied. 2 The accuracy rates for the total sample and each subgroup examined in this report can be found in the Appendix to this report.  Table 2 Summary of Results for Predicting Successful Students FGPA Criterion Level Subgroup 2.0 2.5 3.0 3.25 3.5 3.75 Total Sample SAT HSGPA SAT SAT SAT --- Females SAT HSGPA HSGPA SAT SAT --- Males SAT SAT HSGPA SAT SAT --- Native American = HSGPA SAT --- --- --- Asian American = SAT HSGPA SAT SAT --- African American SAT SAT SAT SAT --- --- Hispanic SAT SAT HSGPA SAT --- --- White SAT HSGPA SAT SAT SAT --- Other SAT SAT HSGPA HSGPA HSGPA --- Attending Very Selective Colleges = HSGPA SAT SAT SAT --- Attending Moderately Selective Colleges SAT HSGPA SAT HSGPA SAT SAT Attending Least Selective Colleges SAT HSGPA HSGPA HSGPA HSGPA SAT Note: An “=” indicates that the SAT and HSGPA were equally effective predictors; an “---” indicates that neither measure was an effective predictor, or that there were not enough cases to evaluate the models. students within each subgroup and for the total sample at success-criterion levels, with the exception of the 2.5 each of the six FGPA criterion levels. and 3.75 levels, the SAT is slightly more accurate than At two of the highest levels of success (i.e., FGPAs HSGPA in predicting the successful group, but less greater than or equal to 3.25 and 3.5), the SAT was found accurate than HSGPA in predicting the unsuccessful to have slightly greater predictive power than HSGPAs. group. Overall we see that as the success-criterion level Interestingly, at both the lowest and highest level of success, increases, the percentage of correct classifications for the accuracy rate of either the SAT or HSGPA alone is the the unsuccessful group increases and the percentage of same as the accuracy rate of the SAT and HSGPA combined. correct classifications for the successful group decreases. At two levels of criterion success (FGPAs at or above 2.5 This finding is directly related to the decrease in the to 3.0), HSGPA is better than the SAT when used as the number of cases as the success-criterion level increases, sole predictor of FGPA. Neither the SAT nor HSGPA was as well as a decrease in the variability of HSGPA and successful at predicting the highest criterion level, 3.75. SAT scores as the successful group becomes more Although overall the model is most accurate at the elite. At the highest criterion level (3.75 FGPA), the 3.75 level, there is a difference in the models’ accuracy accuracy rate for both the SAT and HSGPA was zero in classifying the successful and the unsuccessful for predicting successful students and 100 percent for individuals. As shown in Figures 2 and 3, at most predicting unsuccessful students. 100 100 HSGPA SAT Both HSGPA SAT Both 90 90 80 80 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 2.00 2.50 3.00 3.25 3.50 3.75 2.00 2.50 3.00 3.25 3.50 3.75 Figure 2. Overall accuracy rates for successful students. Figure 3. Overall accuracy rates for unsuccessful students. 3 Figures 4 and 5 show the accuracy rates for the three d100.0 e logistic regression models for successful females and males, ssifi90.0 HSGPA rmesapleesc tmivaeyly b. eT fhoue nrde siun ltAs pfpoern duinx sTuacbclees sAfu2l. Ifne malal lelos giasntidc ctly Cla7800..00 SBAotTh regression models, females are correctly classified at higher orre60.0 rates than males for the lowest two FGPA criterion levels C ul 50.0 (2.0 and 2.5), while males are correctly classified at higher sf s rates than females for FGPAs of 3.0 and higher. There is ce40.0 c u a sizable difference in the accuracy rates for females and e S30.0 males who were successful and unsuccessful. With only ag20.0 nt one exception, in all models and all FGPA criterion levels, e successful females were correctly classified at higher rates Perc10.0 0.0 than males. The logistic regression model with HSGPA as 2.0 2.5 3.0 3.25 3.5 3.75 First-Year College Grade Point Average the sole predictor (Model 1) results in higher accuracy rates Figure 4. Accuracy rates for successful females. for successful females at the 2.5 and 3.0 criterion levels, while the model with the SAT as the sole predictor (Model 2) results in higher accuracy rates for successful females at the 2.0, 3.25, and 3.5 criterion levels. Conversely, the highest accuracy rates for unsuccessful females are produced by Model 2 (SAT only) at the 2.5 and 3.0 levels and Model 100.0 d 1 (HSGPA only) at the 2.0, 3.25, and 3.5 levels. Model 2 e produces higher accuracy rates than Model 1 for successful ssifi90.0 HSGPA males at all but two criterion levels (3.0 and 3.75). y Cla80.0 SBAotTh The classification rate for females using the SAT alone ctl70.0 e is equal to or slightly higher than using HSGPA alone at all orr60.0 C FGPA criterion levels. For males, the use of the SAT alone ul 50.0 produces an accuracy rate that is equal to or slightly higher ssf ce40.0 than that of HSGPA alone for all but two FGPA criterion uc levels (2.5 and 3.0). The combination of HSGPA and the SAT e S30.0 g a20.0 (Model 3) increases the accuracy rate over either predictor nt e used alone in many but not all cases. Overall for both females erc10.0 P and males, Model 3 increases the accuracy rate at all FGPA 0.0 2.0 2.5 3.0 3.25 3.5 3.75 criterion levels except for the lowest (2.0) and highest (3.75). First-Year College Grade Point Average For successful students, Model 3 increases the accuracy rate Figure 5. Accuracy rates for successful males. at the 3.0 through 3.5 levels; for unsuccessful students, the two-predictor model increases the accuracy rate over either of the single-predictor models only at the 2.5 criterion level for females, and at the 2.0 level for males. Figures 6–9 show the accuracy rates for successful students by racial/ethnic group (Native American students 100.0 are not included due to the small number of students in this ed fi study). The full results for both successful and unsuccessful ssi90.0 HSGPA a students are displayed in Appendix Table A3. For Native Cl80.0 SAT y Both American students, the SAT model has a higher accuracy ctl70.0 e rate than the HSGPA model at the 2.5 and 3.0 levels, while orr60.0 C both models are equally effective for the 2.0 level. For Asian ul 50.0 American students, both models are equally effective across ssf ce40.0 all FGPA levels with no difference larger than 0.5 percent. For c u African American students, the HSGPA model has a higher e S30.0 g a20.0 accuracy rate than the SAT model at the 2.5 FGPA level, but nt e the SAT model has a much higher accuracy rate for predicting erc10.0 P success at FGPA levels between and including 3.0 and 3.5. At 0.0 2.0 2.5 3.0 3.25 3.5 3.75 the highest level of success (3.75 and higher), both models First-Year College Grade Point Average have the same accuracy rate. The HSGPA model is slightly Figure 6. Accuracy rates for successful Asian American better than the SAT model in predicting Hispanic and white students. 4 100.0 d 100.0 e d Correctly Classifi67890000....0000 HSBAoStTGhPA orrectly Classifie67890000....0000 HSBAoStTGhPA uccessful 4500..00 ccessful C4500..00 ge S30.0 e Su30.0 a20.0 g ent nta20.0 Perc100..00 Perce10.0 2.0 2.5 3.0 3.25 3.5 3.75 0.0 First-Year College Grade Point Average 2.0 2.5 3.0 3.25 3.5 3.75 First-Year College Grade Point Average Figure 7. Accuracy rates for successful African American Figure 10. Accuracy rates for successful students attending students. very selective colleges. 100.0 100.0 d d e e ssifi90.0 HSGPA ssifi90.0 HSGPA y Cla80.0 SBAotTh y Cla80.0 SBAotTh ctl70.0 ctl70.0 e e orr60.0 orr60.0 C C ul 50.0 ul 50.0 sf sf s s ce40.0 ce40.0 c c u u e S30.0 e S30.0 ag20.0 ag20.0 nt nt e e erc10.0 erc10.0 P 0.0 P 0.0 2.0 2.5 3.0 3.25 3.5 3.75 2.0 2.5 3.0 3.25 3.5 3.75 First-Year College Grade Point Average First-Year College Grade Point Average Figure 8. Accuracy rates for successful white students. Figure 11. Accuracy rates for successful students attending moderately selective colleges. 100.0 100.0 Correctly Classified67890000....0000 HSBAoStTGhPA Correctly Classified67890000....0000 HSBAoStTGhPA ntage Successful 23450000....0000 ntage Successful 23450000....0000 Perce100..00 Perce100..00 2.0 2.5 3.0 3.25 3.5 3.75 2.0 2.5 3.0 3.25 3.5 3.75 First-Year College Grade Point Average First-Year College Grade Point Average Figure 9. Accuracy rates for successful Hispanic students. Figure 12. Accuracy rates for successful students attending least selective colleges. 5

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.