w—w—w—\ -—A—_ m THE EFFECT OF CERTAIN SUECULTURAL SACKGRDUND FACTORS ON THE PREDICHON OF GRADES AT THE UNIVERSITY OF :‘diCHlGAN Thesis for the Degree 0f Ed. D. MECHWAN STATE UNNERSWY Roy E. Haliaciay 1966 LIBRA R Y Michigan State University TH as" This is to certify that the thesis entitled THE EFFECT OF CERTAIN SUBCULTURAL BACKGROUND FACTORS ON THE PREDICTION OF GRADES AT THE UNIVERSITY OF MICHIGAN presented by Roy Eldon Halladay has been accepted towards fulfillment of the requirements for Doctor's degree in Education MM Maj r professor Date MILE—.— -“ u A..——_ ._—.__ The co turned from labeled "0p. 0f "selecti' the shear w. 0f higher e. °f 30m ins need to gel “on has ca t° be used the? my th for “tutti °f aptitude arrive at t With t of applicat on the“ fa The charges dQSIgued by ABSTRACT THE EFFECT OF CERTAIN SUBCULTURAL BACKGROUND FACTORS ON THE PREDICTION OF GRADES AT THE UNIVERSITY OF MICHIGAN 0-0 by Roy E? Balladay The Problem The colleges and universities of this country have, in this decade, turned from general adherence to a philosOphy on admission typically labeled "open-door? to acceptance, in varying degrees, of a philosOphy of "selective admission." This change has come about as a result of the shear weight of numbers of students seeking admission to institutions of higher education and the inability, or disinclination, on the part of some institutions to meet the demand for facilities. The resulting need to select, from multiple applicants, those to be granted admis- sion has caused many colleges and universities to seek out measures to be used in predicting the probable success of the applicants that they may then grant acceptances to those with the highest expectancy for retention. Mbst of these colleges have turned to some combination of aptitude tests, achievement tests, and secondary school record to arrive at this prediction. With the increased use of test scores as a part of the consideration of applications to college have come criticisms related to over-emphasis on these fiactors and their "fairness" to certain groups of applicants. The charges have been leveled, for instance, that such tests are designed by persons of middle-class cultural background and hence are not aPPml ainority 1 that some saphistics admission At th that since particular prevent ed that stand: of making ' b~‘legrom'td: This s tern for "f rural Educa the ability Data ft Classes of 1 "MW of p man‘l‘ural varsity from of less than are located 1 resource, f0! Roy E. Balladay not apprOpriate for groups which depart from this "norm," i.e., the minority groups and the culturally disadvantaged. Also charged is that some students, such as those from.small, rural schools, are not saphisticated test takers and are therefore disadvantaged in the admission decision. At the same time, there is another group of critics who charge that since the secondary school record is so poorly defined, and particularly since the college-going population is so mobile as to prevent admissions officers from always knowing the high school, that standardized aptitude and achievement tests are the only way of making "fair" decisions about students from diverse educational backgrounds. The Study Design, This study was designed to investigate one facet of this con- cern for "fairness": the relative effect of either small school- rural educational background or large school-urban background on the ability to predict academic performance in college. Data for the study was collected from the 1962-65 freshman classes of the College of Literature, Science, and Arts of the Uni- versity of Michigan. Three groups were defined and studied. A small-rural group included 101 students who had entered the Uni- versity from secondary schools in Michigan with graduating classes of less than 100 students from towns of under 15,000 population which are located in areas of the state dependent primarily on the natural resources for their economy. These were upper peninsula and upper- Roy E. Balladay lower peninsula towns. A large-urban group of 256 students was selected from comprehensive secondary schools with enrollments well in excess of 500 students and from industrially oriented towns in Michigan of over 15,000 population. A third group of 495 students was used as a control, or comparison, group and consisted of’a random sampling of every sixth person in the freshman class of 1964-65. variables included in the study were: the Scholastic Aptitude Test of the College Entrance Examination Board, the English Composi- tion Test of the College Board, an average of College Board Achieve- ment Tests, the Secondary School Percentile Rank, and the University Freshman Grade-Point Average. The Study Results The results of the study suggest that: l. The means of the Scholastic Aptitude Test, the Achievement Tests, and the Preshman.Grade-Point Average are significantly lower for students from small-rural schools than they are for students. from large-urban schools or from the freshman class taken as a whole. 2. The ability of the Scholastic Aptitude Test and the Achieve- ment Tests to predict Freshman Grade-Point Averages is not signifi- cantly different for either the small-rural group or the large-urban group than it is for the freshman class as a whole. 3. The mean Secondary School Ranks differ significantly between groups, but in an order inverse to the order of means on the Scholastic Aptitude Test, the Achievement Tests, and the University Freshman Grade-Point Average. That is, the small-rural group tend to present the highest ranks and the control group the lowest. h. in the a] academic between I r (rare 1) 5. equation Percenti] in this 1 Test-Vert School Pe were: 21 6. for :11 s from anal Students t0 over.p ab°ut the predictio th°39 nea Predictio the fa rth Roy E. Halladay 4. There is a significant difference between the study groups in the ability of the Secondary School Percentile Bank to predict academic success at the University of Michigan. The correlations between University Grade-Point Average and this variable are: r(rural) - .191, r(urban) - .441, and r(control) - .320. 5. Combining the predictor variables in a multiple regression equation tends to compensate for the variation in Secondary School Percentile Rank, but does not canpletely make up for the variability in this factor. The multiple correlations for Scholastic Aptitude Test-Verbal, Scholastic Aptitude Test-Mathematical, and Secondary School Percentile Rank as predictors of Freshman Grade-Point Average were: 1(rura1) - .399, [(urban) - .551, and R(control) - .479. 6. It would appear that the use of a single prediction equation for all students does not predict in a similar manner for students fro-tamall-rural or large-urban secondary schools as it does for the students in general. More specifically, such an equation would appear to over-predict for students from small-rural schools. For students in about the tap five per cent of classes in large-urban schools, a single prediction equation would seem to be an equivalent predictor, but for those nearer the middle of their class in these schools, the single prediction equation appears to over-predict by’a margin which increases the farther down the ranking the student appears. THE EFFECT OF CERTAIN SUBCUUTURAL BACKGROUND FACTORS ON THE PREDICTION OF’GRADES AT THE UNTVERSITT’OT NICHTGAN BY 0 Roy Eléonalladay A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF EDUCATION College of Education 1966 TABLE OF CONTENTS LI S T or TABI‘ES ‘0 O O O O O O O O O O O O O O O O O O O C 1v Chapter I. II. III. IV. m mom 0 O O O O O I O O O O O O I O O O O O O 1 The Decade of Revolution The Concern About Fairness Rural-Urban Factors and School Size as Daterrents to College Admission Need for Research The Prospectus mum LITERATURE O O O O O O O O O O O O O O O O 11 Previous Reviews of the Literature Studies of School Comparisons Rural-Urban Comparisons Studies of Academic Prediction Summary SOURCE AND CLASSIFICATION OF DATA. . . . . . . . . 28 Source of Data The Scholastic Aptitude Test The Achievemant Tests The Secondary School Record The College Grade-Point Average Classification of Dnta ANALYSIS AND INTERPRETATION OF THE DATA. . . . . . 40 Summary of the Data Differences in loans Differences in Correlations of Predictor Variables with the Criterion Differences in Standard Errors of Estimate of Variables Comparison of the Multiple Regression Prediction Equations CONCLUSIONS AND RECOMMENDATIONS. . . . . . . . . . 57 Conclusions Recommendations 11 111 Chapter Page APPENDIX . . . . . . . . . . . . . . . . . . . . . . . . 66 BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . 80 Table 1. LIST OF TABLES Page Standard Errors of Measurement and Reliability Coefficients for the College Entrance Examina- tion Board Scholastic Aptitude Test and Achieve- ment Tests . . . . . . . . . . . . . . . . . . . 30 Frequency Distributions of Zero-Order Correla- tions of ESR (or BSA), SAT-V, and SAT-E with Freshman Grade Point Average (Semester er Year) for 271 Groups in Four-Year and Two-Year Colleges, Exclusive of Engineering Groups. . . . 31 Frequency Distributions of Zero-Order Correla- tions of HSR (or HSA), SATBV, SAT-M with Freshman Grade Point Average (Semester or Year) for 63 Liberal Arts Groups, Classified According to Sex. 32 Multiple Correlation of HSR (or ESA) and SAT with Freshman Grade Point Average (Semester or Year) for 231 Groups in Four-Year and Two-Year Colleges, Exclusive of Engineering Groups. . . . . . . . . 33 Multiple Correlation of BSR (or BSA) and SAT with Freshman Grade Point Average (Semester or Year) for 41 Liberal Arts Groups, Classified According to so: 0 O O O O O O O O O O 0 O O O O O O O I O O 34 Summary of Sample Size, Means, and Standard De- viations for Rural, Urban, and Control Groups for all var1ables O O O O O O O O O O O O O O O O I O O 41 F Ratios Resulting From Analysis of Variance Between Rural, urban, and Control Groups . . . . . 41 Results of the Newman-Keuls Tests of Significance for Pairs of Variables from Rural, Urban, and Cont r0 1 Groups I O O O O O O 0 O O O O O O O O O O 43 Correlation of SAT-V, SATBI, Secondary School Per- centile Rank, English Composition Test, and Achieve- ment Test Average with Freshman Grade-Point Average for Rural, urban, and Control Groups . . . . . . . 45 iv Table 10. 11. 12. 13. 14. 15. 2A. 3A. Page Tests of Significance Between Pairs of Zero- order Correlations of Predictor Variables with Freshman Grade-Point Average for Rural, urban, and Control Groups. . . . . . . . . . . . . . . . 46 Cbmparisons of Means and correlations with Freshman Grade-Point Average for Secondary School Percentile Rank for Rural, urban, and control Groups. . . . . . . . . . . . . . . . . . 47 Standard Errors of Estimate for Predicting Freshman Grade-Point Average by SAT-V, SATBM, Secondary School Percentile Rank, English composition Test and Achievement Test Average for Rural, Urban, and Control Groups. . . . . . . 48 Results of the Gulliksen Test for Differences in Standard Errors of Estimate Between the Predictor Variables and Criterion for Rural and Control Comparisons and Urban and Control Comparisons . . 49 Stepdwise Regression constants and Coefficients and the Resulting Multiple Correlations and Standard Errors of Estimate of the Best-Weighted Combina- tions of Predictors with the Criterion for Rural, Urban, and control Groups . . . . . . . . . . . . 50 Predicted Freshman Grade-Point Averages using Regression Equations with Variables, SAT-V and Secondary School Percentile Rank for Rural, Urban, and Control Groups and Differences Between the Control Group and Rural and Urban Groups. . . 53 APPENDIX Frequency Distributions, Means, and Standard De- viations of the Verbal Section of the Scholastic Aptitude Test for Rural, urban, and Control arm” 0 O O C O O O O O O O I O O I O O O O O O 7 1 Frequency Distributions, Means, and Standard Deviations of the Mathematical Section of the Scholastic Aptitude Test for Rural, urban, and Control Groups. . . . . . . . . . . . . . . . . . 72 Frequency Distributions, Means, and Standard De- viations of the Secondary School Percentile Rank for Rural, Urban, and Control Groups. . . . . . . 73 vi Table Page 4A. Frequency Distributions, Means, and Standard De- viations of the English Composition Test for Rural, urban, and Control Groups. . . . . . . . . 74 5A. Frequency Distributions, Means, and Standard De- viations of the Achievement Test Average for Rural, Urban, and Control Groups . . . . . . . . . . . . 75 6A. Frequency Distributions, Means, and Standard De- viations of the University Freshman Grade-Point Average for Rural, urban, and Control Groups. . . 76 7A. Analysis of Variance Summary Table for the Scholastic Aptitude Test-Verbal . . . . . . . . . 77 8A. Analysis of Variance Summary Table for the Scho- lastic Aptitude Test-Mathematical . . . . . . . . 77 9A. Analysis of Variance Summary Table for the Second- ary School Percentile Rank. . . . . . . . . . . . 77 10A. Analysis of Variance Summary Table for the English Composition Test. . . . . . . . . . . . . . . . . 78 11A. Analysis of Variance Summary Table for the Achieve- ment Test Average . . . . . . . . . . . . . . . . 78 12A. Analysis of Variance Summary Table for the Univer- sity Freshman Grade-Point Average. . . . . . . 78 13A. Correlation Matrix. . . . . . . . . . . . . . . . 79 LIST OF FIGURES Figure Page 1. COmparison of the Regression Equations for Pre- dicting the Freshman Grade-Point Average from the Secondary School Percentile Rank for Rural, Urban, and Control Groups. . . . . . . . . . . . . . . . 54 2. Comparison of the Regression Equations for Pre- dicting the Freshman Grade-Point Average from the Secondary School Percentile Rank when the SAT-V Equals 500 for Rural, urban, and Control Groups . 55 3. Comparison of the Regression Equation for Predict- ing the Freshman Grade-Point Average from the Secondary School Percentile Rank when the SAT-V Equals 750 for Rural, Urban, and Control Groups . 55 CHAPTER I THE PROBLEM THE DECADE OF REVOLUTION The period 1960-1970 may well go into the history books as the decade of revolution in education. This period is not only being marked by sweeping methodological and tech- nological changes in the process of education but those changes are coming at a time when the post-World War II baby boom is flooding the educational market place. The massive increase in numbers, caused on the one hand by population in- creases and on the other by an increased interest in and need for education at all levels by the populace, has forced educa- tional institutions to review and revamp their educational processes. This change has been no more pronounced nor no more up- setting to the people of the nation any place than it has in the movement of young people from secondary school to college. One of the overt signs of this change has been a turn to the use of externally administered tests as major implements of selection of applicants to college. As some evidence of this shift, in 1960 the College Entrance Examination Board admin- istered 661,335 Scholastic Aptitude Tests (used by colleges and universities as one predictor in admission), while in 1 2 1965, 1,269,442 took the same test.1 Several factors have contributed to this turn to selec- tive admission and the use of aptitude and achievement tests in the selection process. Already mentioned is the increase in numbers of college-age youth and the increased percentage of secondary school graduates going to college. The nation's colleges have been hard pressed to provide sufficient dormi- tory space, classroom facilities, and faculty to meet the de- mand. As a consequence the colleges, first the small private ones and more recently and in increasing numbers the large and the public, have been forced to select their freshmen classes from a number of applicants which exceeds the capacity of the institutions. The colleges are forced into selection. This process, more often than not, takes the form of selecting those students who exhibit, through past achievement and through tests of aptitude and achievement, the highest poten- tial for academic success in college. Aside from numbers, however, at least three other fac- tors have made the use of tests and other objective measures as predictors of academic success a regular part of the admis- sion process. First, the increased percentage of secondary school graduates desiring to go to college has resulted in greater heterogeneity. Since it is difficult for colleges, particularly small ones, to meet the educational needs of 1Unpublished reports, Educational Testing Service, Princeton, New Jersey, 1965. 3 such a range of abilities they tend to select those for whom they can best provide. Secondly, the educational revolution referred to has resulted in an increased diversity in the quantity and quality of secondary education. This range of curriculum offerings, methods of teaching, and quality of fac- ulties forces colleges to look for new standards of compari- son between schools. Gone are the days of the Carnegie Unit. It is no longer possible to compare students on the basis of the number of units of preparation they have had in secondary school. Thirdly, modern society is mobile. Whereas the col- lege bound once tended to go to college near home, this fac- tor now has little bearing on college selection. The result is that college admissions officers can no longer be expected to know the secondary schools from which they may be drawing their candidates. Without this familiarity, they must rely more heavily on standardized tests to help them make Judgments about the relative quality of preparation of individual candi- dates. THE CONCERN ABOUT FAIRNESS The greater reliance on tests in the admission process has frequently caused papular opinion to question the "fair- ness” of these instruments in the decision-making process. Protests over what is sometimes imagined to be over reliance on such factors are common.2 There are frequent criticisms 2There are many examples of these criticisms. One of the ination lov soc imagina‘ taking, and so ( l revolutj nensions crininat well for geograph economic T “0 leans be raise border, fifty co Stranded the Span under qu is not 0 such to . :xalples .' Herr. 4 of the Scholastic Aptitude Test of the College Entrance Exams ination Board, claiming that it is "unfair” to students from low socioeconomic areas, from rural areas, candidates who lack imagination, candidates who are not saphisticated about test taking, candidates from schools where few go on to college, and so on. Much concern about "fairness" is related to the social revolution taking place in America. This concern has many di- mensions. Do the objective measures used in admissions dis- criminate against race? Do these measures predict equally well for applicants regardless of the school attended, or the geographic area of the country where they live, or the socio- economic status of their parents? This concern for "fairness" to minority groups is by no means limited to the negro race. Similar questions could be raised on behalf of the French-Canadians along the Maine border, the eight million "old Americans" in the two hundred fifty counties of the Southern Appalachias, the more or less stranded people in the Lakes States cut-over areas, the Indians, the Spanish-Americans, and other foreign language groups. Another area in which the validity of such tests comes under question is in that of international education. This is not only a decade of revolution in education within our such to which the reader might refer and which cites other examples is: Hillel Black, They Shall Not Pass (New York: In. Morrow and Company, 1963), p. 334. ovn CC cation This I the as States the ed univer tional tests aPtitu Occupa econom these 5 own country but it may well be the period in which higher edu- cation gains a firm foothold in the international setting. This may be accomplished in two ways; on the one hand, through the assistance with educational problems offered by the United States of America to other countries, and, on the other, through the education of foreign nationals in American colleges and universities. In either case, wherever use is made of educa- tional tests, the problem of cultural influences on these tests will arise and will need to be understood and dealt with. RURAL-URBAN FACTORS AND SCHOOL SIZE AS DETERRENTS TO COLLEGE ADMISSION Studies relating the effects of cultural background on aptitude test scores have isolated the following factors: race, occupation, rural-urban factors, school differences, and socio- economic class differences.3 The interrelated complexity of these problems prevents their study in toto. This study deals with a combination of these factors, i.e., the rural-urban aspect combined with size of school, be- cause it is believed that this may be a significant variable in determining access to higher education in the United States. Over 70 per cent of the nation's public schools are 3Charles M. Lucas, Survey of the Literature Relating to the Effects of Cultural‘Background'on Aptitude Test SéfiFes, A"Report"fo the Research COmmiffee o? the COIIege Entrance 'EXaminatIon‘Board,‘June3U__I953T_“PFEpEEEETBy_EHfiEEfIEEEI 15sting Serviée (Research‘BuIIefin 53- 13, June 30,1953), p. 3. locatet these I nearly three-c availa] have S! gas fox rural-t school library riculun lack of travel. use of an 1n0r detemi varYing Such st ‘ight t 6 located in small towns or rural areas.4 Over 50 per cent of these schools are of such a size as to make quality programs nearly impossible or definitely improbable. Whereas about three-quarters of the city schools have accelerated curricula available for superior students, approximately half that many have such provisions available in the other schools.5 Flana- gan found in his studies of the American high school that the rural-urban factor correlates highly with the size of the school and such related items as the number of books in the library, the summer school policy, the offerings of the cur- riculum, the experience of teachers, the education of teachers, lack of guidance programs, lack of homogeneous grouping, and travel.6 NEED FOR RESEARCH With the growing interest in, and concern about, the use of tests in the college admission process, there arises an increased need to explore the validity of these tests for determining the admissibility to college of candidates with varying backgrounds. The action, based upon the results of such studies, might take one of two directions. One action might tend to be tied to the philosophy that cultural bias 4John C. Flanagan et. a1., A Survey and Follow-Up Study of Educational Plans and'DéEIEIons‘Ih‘RelatIOnTto Aptituae Pat- fErns: Studies of the American High school (PittsburghzfiThe ‘Unlversity of Pittsfiurgfi, 1952), pp. 2-35. 51bid., pp. 2-46. 61bid., pp. 6-23. should 1 0: the ¢ that thc nay redx which t1 does n01 ' it be u: and how such bit to dupl: consistc predict: 1 port e11 7 should be eliminated from all such tests wherever possible. On the other hand, Anastasi7 and Turnbull,8 have pointed out that the elimination of "cultural differentials" from.a test may reduce its validity for the prediction of most criteria, which themselves are culturally loaded. What follows, however, does not detract from the basic problem. It is necessary that it be understood how differences, if they exist, come about and how they relate to the criterion. Knowing if and where such biases in tests exist will allow test developers either to duplicate them in parallel tests in order to make results consistent, or to eliminate them if this results in a better predictive instrument or is desirable on some other basis. 4 Recent surveys of the literature report studies that sup- port either side of the argument about whether or not cultural differences res'u'lt in differences on various tests of aptitude.9 Two fairly recent studies, for instance, have attempted to re- late the differences on the Scholastic Aptitude Test of thethnp lege Entrance Examination Board resulting from certain of these 7A. Anastasi, "Some Implications of Cultural Factors for Test Construction," Proceedin s 1949 Invitational Confer- ence on Testing Problems ° Iii‘Sirvice, 1 , pp. 8w1111a-.w. Turnbull, "Influence of Cultural Background on Predictive Test Scores," Proceedi 1949 Invitational Con- ference on Testis Problems (PFIncefon, l.3.: Educational Test- ng erv ce, , pp. 4. 9The reader is referred specifically to two such sur- veys: Lucas, loc. cit. and David E. Levin, The Prediction of Academic Performance (New York: Russell Sage Founaififin, 1935). culture found ‘ prediC' ling, 4 could I considc ing wh: used 11 exactl; rural g acaden: seeks 1 or urb: dictiV1 ‘ry Sc] 3 rang 8 cultural factors to the criterion of grades in college. Schulm: found that the Scholastic Aptitude Test scores neither over- predict nor under-predict for various socioeconomic classes.10 Wing, on the other hand, found that gains in predictions could be made if cultural background factors are taken into consideration.11 The problem that exists, therefore, is one of determin- ing which cultural factors, if any, affect the results of tests used in the selection of students for colleges and, further, exactly how they affect these tests. THE PROSPECTUS This is a statistical study of the relative effect of rural and urban academic backgrounds on the ability to predict academic performance at the University of Michigan. The study seeks to determine if there is anything inherent in the rural or urban educational environment, reflected in the usual pre- dictive factors of aptitude and achievement tests and second- ary school achievement, which, in some way, may cause the pre- diction of success to be influenced in an abnormal manner. The study will attempt to answer the following basic 10D. C. Schultz, The Relationship Between College Grades and Aptitude Test Scores for Differenf”SOcIoeconomic Groups (Princeton: Educational Tesfihg‘Service, 1953). 11Cliff w. Wing, Jr. and Virginia Ktsanes, The Effect of Certain Cultural Background Factors on the PrediCtion of‘ Studenf”Grades InCCOIlege (unpubliShedfireport to the College 'EntranceExaminationBoard, 1960). questions: (1) Is the Scholastic Aptitude Test of the College En- trance Examination Board biased as a predictor of academic success in college for small school-rural or large school- urban type applicants? (2) Are the Achievement Tests of the College Entrance Examination Board biased predictors of academic success in college for either small school-rural or large school-urban type applicants? (3) Is the secondary school achievement record, as a predictor of academic success in college, affected by the cul- tural nature and size of the secondary school? (4) Do the combination of predictive factors (Scholas- tic Aptitude Test-Verbal, Scholastic Aptitude Test-Mathemat- ical, achievement tests, and high school record) predict as well, and in the same way, for applicants from small-rural secondary schoohsand for applicants from large-urban schools as they do for the applicant group as a whole? The data to accomplish this study have been drawn from the freshman classes of the University of Michigan's College of Literature, Science, and Arts for the years 1962-65. Fur- ther, the populations studied are drawn from secondary schools within the state of Michigan that are readily classified as small-rural or large-urban. There is a primary limitation in this study design which should be clarified here. Since the University of 10 Michigan is quite selective in its admission process, the ability range of the students in the study is limited to the more able. Hewever, even though the results will be strictly applicable only to the University of Michigan, it should be possible to draw some implications which may be of importance to the educational community at large. CHAPTER II RELATED LITERATURE PREVIOUS REVIEWS OF THE LITERATURE The literature contains scores of studies concerning the effect on test scores of cultural background, and many complete and adequate summaries of work done in this general area also appear in the literature. One of the most extensive in this area was a survey prepared by Charles M. Lucas of the Educational Testing Serv- ice for the Research Committee of the College Entrance Examin- ation Board.1 Lucas not only reported on the general points of view regarding cultural influence on test scores and on the research and development of culture-free tests but also report- ed on studies of specific socioeconomic influences on tests. In particular, Lucas reported on studies of racial comparisons, occupational group comparisons, rural-urban comparisons, school comparisons, and socioeconomic class comparisons. Lucas re— ported that methodological shortcomings, of one type or an- other, as well as differing experimental designs, rendered many of the results of the studies be reviewed not strictly ISurvey of the Literature Relating to the Effects of CulturaljBackground In Aptitude‘TesffiScores (Princeffin: Edu ca- ona ng erv ce, 11 conparab seeningl reported cultural on intel this inf instance: non-exis A ing test Levin to: to this 1 Ported t] denic pe] the high by Lavin: tional 1, Ported b1 mentionec Lam, co: Status 8.: the Socic inVerSQ E \ 2 “38611 § 31 12 comparable. Notwithstanding both these difficulties and the seemingly contradictory findings of a few of the studies, he reported that the bulk of the evidence seemed to indicate that cultural background, on the average, tends to influence scores on intelligence and scholastic aptitude tests. The effect of this influence, as reflected in group tendencies, was, in many instances, shown to be quite marked and in others, practically non-existent. A very recent, and complete, review of studies concern- ing test scores and cultural background was done by David E. Lavin for the Russell Sage Foundation.2 Of special interest to this study is Lavin's review of thirteen studies which re- ported that socioeconomic status is positively related to aca- demic performance. That is, the higher one's social status, the higher his level of performance. These studies, reported by Lavin, indicate that this relationship holds for all educa- 3 Of special interest, also, are six studies re- tional levels. ported by Lavin whose findings contrast with the previously mentioned thirteen studies reporting a positive relationship.4 Lavin concludes that the relationship between socioeconomic status and academic performance is positive through most of the socioeconomic status range, but at the upper levels it is inverse.5 2The Prediction of Academic Performance (New York: The Russell BageFOundation, 1965). 31bid., p. 125. 41bid., p. 152. 51bid., p. 126. .I JIIII {I I'll: All III‘ 1". ll 11...." 11...! I l 'J. .I. 13 Other reviews and summaries specific to socioeconomic background have been published by Lorimer and Osburn,6 Neff,7 8 9 Herrick, and Jenes. STUDIES OF SCHOOL COMPARISONS A number of studies have reported results of comparing test scores of groups of individuals with various types of school background. Commonly considered school characteristics are size of school, location of school in good or poor socio- economic neighborhood, and public or private school. The Project Talent survey indicated a very high rela- tionship between size of high schools and whether or not they were rural or urban.10 However, school size in itself may not necessarily be colinear with rural-urban location if one takes 6F. Lorimer and F. Osborn, "Variation in Certain Intel- lectual Development Among Groups Classified by Occupation or Social Status," Chapter VIII of Dynamics of Population (New York: Macmillan and Company, 1934), pp.C157L176. 7W. S. Neff, "Socio-Economic Status and Intelligence: A Critical Survey," Psychological Bulletin, 1958, 35, 727-757. 8J. E. Herrick, "What Is Already Known About the Rela- tion of the I.Q. to Cultural Background," Chapter II of Intel- ligence and Cultural Differences by Eells, K. et a1. (ChiEEEE: University of Chicago Pfess, 1957). "_"_‘ 9H. E. Jenes, "Environmental Influences on Mental Devel- opment,” Chapter XI of Manual of Child Psychology, ed. Leonard Carmichael (New York: JbEn WiIey and:SOns, Inc., 1946), pp. 528-632. 10John C. Flanagan, et al., A Survey and Follow-Up Study of Educational Plans and Decisions inIReIation to Aptitude Pat- terns: Studies of tfie Kierican HigE ScfiooI (PittsBurgH: Univer- BIEy OI IiEEsBurgH, 1962): PP- 2‘35- III-Ii!“ llrlll‘vl‘lvil'lllfl‘uldl‘ll‘l Ill '1 I." [lill 14 into account the occurrence of large, rural, consolidated schools and relatively small, decentralized, urban schools. For this reason studies which have separated these factors are reviewed here, but the present study has been designed to control both factors. Feder11 divided entering college freshmen into two groups according to size of secondary school attended. He found that students from large schools scored higher on qual- ifying examinations than those from small schools. In a sim- ilar study, Smith12 divided secondary school students into groups by size of school where small schools were defined as having enrollments of less than 250 and large schools more than 250 students. It was reported that mean scores were sig- nificantly higher in mental, history, and English tests for students from the larger schools than those from smaller schools. In another study, Allison and Barnett13 divided college students into three groups according to the size of secondary school they had attended. The groups consisted of 282 students from small high schools (enrollments less than 11D. D. Feder, "Factors Which Affect Achievement and Its Prediction at the College Level," JOurnal of the American Association of Collegiate Registrary, 1910, IE 107-118; Edu- catiOnaI Abstracts, 1940, 5, No. 292, 76. 12W. R. Smith, "Test Results Reveal Advantages of Larger Schools” (Publ. Educ. Pa., 1930, 5, No. 5, 13), Psychological Abstracts, 1930, 12, No. 2141, 231. 136. Allison and A. Barnett, "Freshman Psychological Examination Scores As Related to Size of High Schools," JOurnal of_Applied Psychology, 1940, 23, 651-652. 15 150), 435 students from schools of intermediate size (enroll- ments of 150 to 500), and 283 students from large schools (en- rollments of 500 and over). There was considerable overlapping, yet differences in the mean Q (quantitative) and L (language) scores of the American Council on Education Psychological Ex- amination were found to be statistically significant. All differences were in favor of the larger schools for all com- parisons. Allison and Barnett reported a correlation of .372 between gross score and size of high school enrollment. Find- ings in greater detail were as follows: Critical Ratio Critical Ratio of Q Score of L Score High School Comparison Differences Differences Small and Intermediate 3.73 5.59 Small and Large 7.59 10.48 Intermediate and Large 4.40 4.97 RURAL-URBAN COMPARISONS Many investigators have seized upon the possibility of making comparisons of mental ability between rural and urban children. The majority of these seem to show that the mental scores of urban subjects tend to be higher than those of rural subjects. 14 Klineberg made a study of ten to twelve year old 14O. Klineberg, "A Study of Psychological Differences Between 'Racial' and National Groups in Europe," Archives of Psychology, 1931, No. 132. 16 European boys by comparing their results on a performanceetype test. He found the difference between the mean of the urban sample of 300 and the rural sample of 700 to be over eight times its standard error. In terms of overlapping, only 30 per cent of the rural children reached or exceeded the median of the urban group. Koch and Simmons15 compared the mean scores for urban and rural children between the ages of eight and fifteen years. Tested were 294 urban and 326 rural chil- dren of native American stock as well as 270 urban and 180 rural children of Mexican stock. The test administered was either the National Intelligence Test or the Detroit Intelli- gence Test. Results indicated a superiority in favor of the urban group to the extent shown in the following table: Range of Difference Between Means For Range of Critical Comparison EachAge Legel Ratio Urban American and Rural American 0-23 .2-5.4 Urban Mexican and Rural Mexican 17-28 3.8-10.1 Many other reports of rural-urban comparisons in mental ability have appeared in the literature. Predominant findings have been that lower scores in mental tests appear to be as- sociated with a rural environment. Among these are studies 15H. L. Koch and R. Simmons, "A Study of the Test Per- formance of American, Mexican, and Negro Children," Psycholog- ical Monograph, 1926, 35, No. 5. 17 17 Pressey,18 and Irion and by Pintner,16 Pressey and Thomas, Fisher.19 Pintner reported the median index of intelligence for 154 rural school children to be 10 per cent below the median of an urban group. Pressey and Thomas found only 27 per cent of their rural sample tested to be above the norm for urban children. In another paper, Pressey reported that only 22 per cent of 183 rural children, ages six, seven, and eight years, scored above the median for age as determined from urban children. Irion and Fisher found that the median score for their sample of 361 rural children, between the ages of ten and sixteen years, was ten points below the urban norm on the basis of the National Intelligence Test. Fewer studies have been done on the relation of rural- urban differences on scores of tests of aptitude than on tests of intelligence. Those which have been done, however, gener- ally show much the same relationship as those attributed to intelligence tests. Turnbull, reporting on studies at the 16R. Pintner, "A Mental Survey of the School Population of a Village," School and Society, 1917, 5, 597-600. 17S. L. Pressey and J. B. Thomas, "A Study of Country Children in a Good and Poor Farming District by Means of A Group Scale of Intelligence," Journal of Applied Psychology, 1919, 8, 534-539. 18L. W. Pressey, "The Influence of Inadequate Schooling and Poor Environment Upon Results of Tests of Intelligence," JOurnal of Applied Psychology, 1920,‘4, 91-96. 19T. Irion and F. C. Fisher, "Testing the Intelligence of Rural School Children," American School Master, 1921, 14, 221-223. '- 18 Educational Testing Service,20 showed that results of a study of the first Army-Navy College Qualifying Test indicated that differences were significantly in favor of the urban sample, especially in the verbal area. Furthermore, the analysis of variance performed on this Army-Navy sample showed that dif- ferences between groups depended upon individual items rather than upon the type of test material. In a study of students at the University of Kansas,21 Smith found that percentile scores of the American Council on Education Psychological Examination (ACE) showed a slight tendency to decrease with distance from urban centers. Also studying results of the ACE,22 Nelson found an urban group of 580 college entrants to be significantly higher in average total score than a rural group of 466. By contrast to these studies, Frederickson, Olsen,and Schrsder studying the first semester grades at Kenyon College23 could find no clear trend in the relationship between size of community and tendency for better achievement in college than had been indicated by other 20W. W. Turnbull, "Influence of Cultural Background on Predictive Test Score," Proceedings 1949 Invitational Conference on Testing Problems (PriEE5t5Ei_EHEEEfi5fiEI‘TEEfifiE‘SEFViEET‘_"' 1950) , 29:34. 21M. Smith, "An Urban-Rural Intellectual Gradient," Sociological Society Research, 1943, 2:, 307-315. 22C. W. Nelson, "Testing the Influence of Rural and Urban Environment on ACE Intelligence Test Scores," American JOurnal of Sociology, 1942, 1, 743-751. 23H. Frederickson, M. Olsen, and W. B. Schrader, Predic- tion of First Semester Grades at Kenyon College, 1948-1949 (Princeton: EducatIOnaI”Testing ServiCe,C195O). 19 predictive measures. The studies of rural-urban differences in both intel- ligence and scholastic aptitude yield evidence in favor of linking higher scores with urban residence. Statistically significant differences between rural and urban samples have often been reported. Overlap of some distributions has also been found. Two possible explanations for these score dif- ferences stem from a number of studies. Environmental differ- ences may operate so as to insure superior test performance for the group which happens to be closest culturally to the pOpulation on which the test has been standardized. Secondly, selective migration may Operate so that the less intelligent remain in the rural setting while the mentally superior gravi- tate to the cities. One study supporting the first of these hypotheses was done by Shimberg.24 He made an investigation into the validity of norms with reference to urban and rural groups by construct- ing two information tests, standardizing one on a rural group, and the other on an urban group. When both tests were admin- istered to 4,812 rural children in grades three to twelve, and to 962 urban children in grades four to seven, the urban group was found to be one year retarded on the basis of the test standardized on the rural group and vice versa. Differences 24M. Shimberg, "An Investigation into the Validity of Norms with Special Reference to Urban and Rural Groups,“ Ar- chives of Psychology, 1928-29, 16, Ser. 104. "' 20 in both instances were significant to the extent that the ratios of the differences to the standard error ranged from 5.56 to 9.33. In another study, Wheeler25 tested 3,252 children in the Eastern Tennessee mountain area and compared the results with a similar study he had made ten years earlier. He found about the same rate of decline in I.Q. with increase in age from six to sixteen years (about two points each year). How- ever, the 1940 mountain child was found to be mentally superior to his 1930 prototype at all ages and all grades, an average increase in I.Q. of ten points. Wheeler also noted that dur- ing the decade intervening between his two studies, there had been definite improvement in the economic, social, and educa- tional status of this mountain area and that improved test scores are apparently associated with improvement in environ- mental conditions. In support of the explanation that rural-urban differ- ences on intelligence and aptitude tests exist because the more intelligent individuals migrate to the larger cities is a study by Gist and Clark.26 They followed up 2,544 rural Kansas secondary school pupils thirteen years after being 25L. R. Wheeler, "A Comparative Study of the Intelli- gence of East Tennessee Mountain Children," Journal of Educa- tional Psychology, 1942, 33, 321-334. 26H. p. Gist and c. D. Clark, "Intelligence as A Selec- tion Factor in Rural-Urban Migration," American Journal of Sociology, 1928, 42, 284-294. 21 tested with Terman group tests of intelligence. Of those lo- cated, roughly 38 per cent were living in urban centers and 62 per cent in rural areas. On the average, the migrants to urban centers were those who had scored higher, when tested thirteen years previously, than those who had remained in a rural environment. Also, those who had migrated to large cities surpassed those who had migrated to smaller cities. 27 28 Papers by Sanford and Mauldin also lent support to the selective migration hypothesis. STUDIES OF ACADEMIC PREDICTION The literature reviewed up to this point bears directly on studies of the effect of culture on certain of the varia- bles which are considered in predicting academic success in college. Specifically, this chapter has dealt so far with the influence of rural-urban environment and size of secondary school on tests of intelligence and scholastic aptitude. Atten- tion will now be turned to studies which relate directly to the correlation of these factors with college grades and their bearing upon the prediction of academic success in college. Wing and Ktsanes compared the relationship between col- lege grades, scores on the Scholastic Aptitude Test of the 27G. A. Sanford, "Selective Migration in a Rural Alabama Community," American Sociological Review, 1940, 5, 759-766. 28W. P. Mauldin, "Selective Migration from Small Towns," American Sociological Review, 1940, 2, 748-758. 22 College Entrance Examination Board, and secondary school per- formance measures for students from varying social-class back- 29 The two sets of cultural grounds at Tulane University. background factors were (1) father's occupation, as an indi- cator of social class, and (2) urban or rural home background. The results indicated that the working class and rural men do not do as well in college as can be expected on the basis of their Scholastic Aptitude Test scores and secondary school rank in class; and that the upper-class and city men do better than can be expected on the basis of their Scholastic Aptitude Test scores and secondary school rank. The authors concluded that small but consistent gains in prediction are likely to occur if cultural background factors are taken into considera- tion. In direct contrast to these findings were those of Schultz earlier.30 Schultz studied 1700 male students in seven colleges who took the Selective Service College Qualify- ing Test in May or June of 1951 and later took the Scholastic Aptitude Test. In comparing the regression equations for various status groups the differences were found not to be statistically significant and to be small and inconsistent. 29Cliff w. Wing, Jr. and Virginia Ktsanes, The Effect of Certain Cultural Background Factors on the Predictiofiwof' Sfudefif;Grades in COIIege (New Orleans, Louisiana: Tulane University,’1960). 30D. G. Schultz, The Relationship Between College Grades and Aptitude Test*ScoreSIIOr‘Different‘SoCio-economic Groups (PrinceTon,T.J.: Educational TesfingTerfice, 19535. 23 It was concluded in this study that scholastic aptitude test scores predict grades equally well, and neither over-predict nor under-predict for all socioeconomicciasses among college students, and that this equivalence was not altered after a period of attendance. In this study Schultz does admit to a very inadequate representation of the lowest socioeconomic levels. Shaw and Brown performed a study which related college performance to the size of the town from which the students had come.31 Using Fisher's variance ratio technique, they found that there was a significant difference in achievement as measured by college grades between groups from various size communities, with a higher comparative percentage of achievers coming from larger communities. In the same study no signifi- cant difference was found between the results of achievement tests taken by the same students. The conclusion of the Shaw- Brown study, which has relevance for this study, is the con- clusion relating the results to the differing value systems of the various size communities. Although it is evident that the authors failed to explore other hypotheses, and have little evidence to support this conclusion, they do raise a question which should be explored by further research. Washburne conducted a study on two college campuses 31Merville C. Shaw and Donald J. Brown, "Scholastic Underachievement of Bright College Students," Personnel and Guidance JOurnal, 1957, 32, 195-199. 24 which attempted to relate socioeconomic factors, including urbanism, to college performance.32 Although the results are clouded by quite different results from the two campuses, one in the southwest and one in the northeast, it was concluded by the study that correlation between the degree of urbanism and college performance was positive on the lower and of the urban scale but very low on the upper end. They set a lower limit of 500,000--the population of metrOpolitan areas, as the point at which the heterogeneity of students made them indistinguishable in regard to college performance. In con- trast again, another study found that while urban students are higher in aptitude than rural students, they were no different in academic performance.33 Hewever, as the authors explain, the rural students tended to be registered in colleges of agriculture, and urban students in business or arts and sci- ences colleges; consequently, it is difficult to interpret the results of this study because the criterion is not the same. It is difficult to find studies which adequately sepa- rate the variable urbanism (or ruralism) from that of school size. The suggestion coming from most studies is that if school size does have a relationship to college performance, 32Norman F. Washburne, "Socio-economic Status, Urbanism and Academic Performance in College," JOurnal of Educational Research, 1959, 53, 130-137. 33William B. Sanders, R. Travis Osborne, and Joel E. Greene, "Intelligence and Academic Performance of College Students of Urban, Rural, and Mixed Backgrounds," JOurnal of Educational Research, 1955,‘4g, 185-193. 25 it is likely a result of differences in facilities, teacher competance, and the like. Small high schools are probably found more frequently in rural areas and their facilities and teacher salaries are likely to be inferior. Two studies examine the relation between size of secondary school and academic performance in college and find opposite results. Altman, studying this question at Central Michigan College, reports no significant difference in the performance in col- lege for students from varying size secondary school classes§4‘ Hoyt, on the other hand, in a rather thorough study of 884 freshman students at Kansas State University, found that when ability is held constant a given secondary school rank will tend to over-predict the achievement of the student from the small secondary school and under-predict the achievement of the student from the larger school.35 SUMMARY Individuals who attend larger secondary schools and schools located in higher socioeconomic areas, as compared with those who attend smaller secondary schools and schools in communities of lower social status have been found to score 34Esther R. Altman, "The Effect of Rank in Class and Size of High School on the Academic Achievement of Central Michigan College Seniors, Class of 1957," JOurnal of Educa- tional Research, 1959, 52, 307-309. 35nonald p. Hoyt, "Size of High School and College Grades," Personnel and Guidance Journal, 1959, 31, 569-573. 26 higher on intelligence tests. Overlapping of score distribu- tions seems to be consistently present. Factors other than size, location, and type Of school very likely operate. Among these are socioeconomic factors and factors related to selec- tion and motivation. Studies of rural-urban differences in intelligence and scholastic aptitude yield evidence in favor of linking higher scores with urban residence. Statistically significant dif- ferences between rural and urban samples have often been re- ported. Overlap Of score distributions has also been found. Two possible explanations for these score differences stem from a number of studies. On the one hand, environmental dif- ferences may Operate to insure superior test performance for the group which happens to be closest culturally to the popu- lation on which the test has been standardized. Secondly, selective migration may Operate so that the less intelligent remain in the rural setting while the mentally superior gravi- tate to the cities. Studies relating rural-urban backgrounds and size of secondary school class to the prediction of academic perform- ance in college are contradictory and inconclusive. It seems apparent that the inability to separate the many factors re- lated to the general term "socioeconomic status" may be a con- tributing factor in the contradictory results of studies. Where secondary school size has been studied, it is not clear to what extent the influence of such factors as the quality of 27 the schools or the native ability of the populations has been controlled. Similarly, rural-urban findings are ambiguous. A number of factors, either singly or in combination, could account for the results. As in studies of school size, such things as intelligence, social-class levels, and quality Of schools may or may not be Operating. At present, the research findings do not allow an assessment of the possibilities. CHAPTER III SOURCE AND CLASSIFICATION OF DATA SOURCE OF DATA Several factors had a direct bearing on the source of the data. Since the primary Objective of the study was to (:Ompare the effect of rural versus urban backgrounds on the ability of the Scholastic Aptitude Test of the College En- trance Examination Board to predict academic success, it was necessary to locate a college or university where the Scholas- ‘tic Aptitude Test is required of all applicants and where both Iniral and urban sub-pepulations could be clearly identified. lflae University of Michigan was chosen for this purpose since t11ey have required the Scholastic Aptitude Test as part of ttie admissions credentials since 1962. The availability of a. number of classes with test data proved helpful since the Ixarcentage of applicants from rural secondary schools is com- pauratively small and it was necessary to combine several classes t<> obtain a sufficient sample size. THE SCHOLASTIC APTITUDE TEST The Scholastic Aptitude Test (which shall hereafter be referred to as the SAT) offered by the College Entrance Examin- at ion Board and administered by Educational Testing Service, is 28 29 a three-hour objective test Of verbal and mathematical skills. The verbal sections are designed to measure ability to under- stand the relationships among words and ideas and reading com- prehension. This portion has separately timed sections such as sentence completion, reading comprehension, analogies, and antonyms arranged in approximate order of difficulty. The mathematical section is designed to measure abilities closely related to college-level work in the liberal arts and engineer- ing by getting answers to such basic questions as: (1) How well has the testee mastered elementary mathematics? (2) HOw well can he apply what is already known to new situations? and (3) How can he use what is known in insightful and non-routine ways of thinking?1 Scores on the SAT and on the College Board's Achievement Tests are reported on a scale ranging from 200 to 800. The re- ference group was all twelfth grade students who took the tests in April, 1941. The mean standard rating for the fixed refer- ence group was set at 500 and the standard deviation of the rat- ings at 100.2 Standard errors of measurement and reliability coefficients for the SAT and Achievement Tests may be found in Table 1. 1For a full description of the Scholastic Aptitude Test the reader is referred tO A Description of the Scholastic Apti- tude Test, prepared and prfiduced annuaIIyTTOr the College En- trance EXamination Board by Educational Testing Service, Prince- ton, New Jersey. 2Henry S. Dyer and Richard G. King, College Board Scores: Their USe and Inter retation (Princeton: Educatiofiil‘TéSting Service;_l9555, pp. IUI, I02. 30 TABLE 1.--Standard errors of measurement and reliability coef- ficients for the College Entrance Examination Board Scholastic Aptitude Test and Achievement Tests3 Test S.E.a r Scholastic Aptitude Test-Verbal 32 .90 Scholastic Aptitude Test-Mathematical 34 .88 American History and Social Studies 31 .90 Biology 35 .88 Chemistry 30 .91 English Composition 37 .87 EurOpean History and World Cultures 31 .90 French 26 .93 German 20 .96 Hebrew 21 .96 Latin 34 .89 Mathematics Level I (Standard) 36 .87 Mathematics Level II (Intensive) 32 .90 Physics 30 .91 Russian 23 .95 Spanish 28 .92 AL aThe Standard Errors of Measurement for all tests are computed from the application of Kuder-Richardson Formula 20. The validity of the Scholastic Aptitude Test as a pre- dictor of college grades has been studied frequently in many colleges. JOhn HOwell prepared a compendium of validity study results for the College Entrance Examination Board showing a wide range of validity coefficients.4 HOwell's results are summarized in Tables 2 and 3. 3College Board Score Reports: A Guide for Counselors and (Admissions Officers (Princeton: EducainnaIITesfing ServICe, 1965), p. 46. 4John Howell, A Compendium Of College Board Validity Study Results, 1958-1964, an unpuBIIsHea'repoft to the COIIege Entrance‘EXaminationCBOard, 1965. 31 .mmmH..bnmom,GOHmmcHEdelmosmesmoonHOD on» OPIPHOQOH mommHHnsmss cm .voma ImmoH .nHHsnos sessm ssHsHHs> ensom ouoHHoo Ho ssHssoesoo s .HHosom nsos seams .mmsonm new epmummow QmHHO>O was» museum «o O>staoxms mm. mm. He. mm. mm. ow. we. on. on. mm. mm. we. ansos Hsm hm eHH «OH Hem mm vHH «OH Hem mm eHH moH Haves H H so.uoo. H a H oo.:me. m H. m m m H m «H.aoH. SH 6 m h «H m oH H H mH.umH. on e NH eH sH m z o m m H em.:ou. mm s OH ”H mm o mH mH e H m m em.umm. He mH m mH oe e «H mm 0H s m m em.uom. me o mH Ha ow mH oH sH on s 0H sH om.ucm. em OH «H mH em oH HH SH as e sH eH ev.nov. mm H mH m we OH nH HH mm m «H om mv.nms. mm m «H w es s on m we HH HH oH em.aom. NH . NH mH e «H H me m cm oH om.unm. s H m w H o H mm m Hm s so.ueo. m m eH w m m mo.umo. m m e H es.cos. ms.ucs. em.nom. as mass 2 s as use: a s as use: 3 s nsoHosH IOHHO stash >uswmsaoxo .mmmeHHOO HdmmlOBv can unchlhsow sH museum Hum hon Andes HO Hepmoaomv Owsse>s peace Ocnum assmmohm mpHB_SIBls canon ouoHHoo Ho asHenoaaoo < .HHosom shoe scams .mnsouw New Opmndmom deHo>o «as» mason» no O>HmsHOde wN. Hm. mN. no NH Hm wN. ON mm. mm Nw. NH Hm. Hm NM. ON mm. Vm. Nv. we NH Hm mv. ON sdHeos Hence HvomwmevaH HH HNMNNH HMMHVHVFNVIH HMHHMIOQ‘H IH [‘V‘H NH OH HH HNWQOH NCGH INHNH V‘mr-HDV‘bfi‘Nu-I IVHOCONN HIHN HI-l IN ”V‘mmmb-QNN I60 COHNNHHN N [HMHMHQON lr-l HHHHNflCDV' I vo.loo. mo.lmo. vH.IOH. mH.lmH. VN.ION. mN.ImN. vm.lom. mm.lmm. vv.lov. mv.lmv. wm.Ion. amt-mm. v0.loo. am.lmm. Vb.l°b. mb.lnh. ¢N.Iom. NP BBQ: >IH¢m we use: a Hd undon occhm meanness mvHs.=|94w .>lb¢m .AoH also «on a one as unsoumusmams sH.o sH.o eH.o o -'and cen.e sv.« sv.n pH.« .<.n.o sensuous ne.sH an.sH no.vH e no» one use use van seen->4 «his .no< e«.nu e«.e« se.oH a can use men one Han «eon .dnoo .u-n ee.H Ho.“ Ho.« cHe.«u tee.uu ne.en «s.ae e«.«o He.«e nuns oHHonooaon Ho.n« Ho.e« s«.eH e the use «on can can HaoHuunonounneem «n.oH «n.aH ee.eH H can oHn men can use Hdnuo>1n nulm.uvnm.o . “\sns neonenenudn nasal museum Honusoo oss .ssnab .Hsnna loan noHoanso no enHun non coneoHuH-sH- «o sauce anomnnunhon one no eaHsuonnn.e manna 44 The difference in means between the Rural group and the Urban group for Secondary School Percentile Bank are not significant at the 5 per cent level. Since the standard analysis of variance technique de- ponds to some extent on the assumption that samples are norm- ally distributed and because it is obvious from inspection of the data that the Secondary School Percentile Bank does not fit this condition, a further test of the differences in means of this variable, which does not depend on this assumption, was carried out. This non-parametric, or distribution-free, compar- ison test is a simplification of the Rank test. It is arrived at by merely classifying all scores as being above, or not above, the median of the combinedsamples.3 A contingency table is then set up and a chi-square computed. In this case chi-square for the Rural-Control comparison is 7.132 and for the Urban-Control comparison 4.786. Both of these results are significant at the 5 per cent level. This non-parametric test, therefore, bears out the results of the somewhat questionable analysis of variance that the difference in means between the Rural and Control groups and between the urban and Control groups are significant at the 5 per cent level. DIFFERIISUS II CORRELATTOIS OT PRIDICTOR VARIABLES 'ITR‘THI CRITERION Of primary concern in.thds study is the degree to which 3Belen I. talker and Joseph Lev, Statistical Inference (New York: Henry Holt h 00., 1953), p. 435?“ 45 each of the predictor variables tends to agree with the cri- terion of Freshman Grade-Point Average and the relationship of these correlations between groups. It can be seen, from.Tablo 9, that these zero-order correlations range from .191, the correlation for Secondary School Percentile Rank in the Rural group, to .540, the cor- relation of the Achievement Test Average in the Urban group. It should be noted here that, contrary to the order of the values of the means of the variables, the correlations, in every instance except the SATblathematical, found the Urban group with the highest values and the Rural group, again with the exception of SAThlnthomatical, with the lowest values. In the case of SAT-Mathematical, the correlation for the Rural group was .337, for the Urban group .305, and for the Control group .272. TABLE 9.-—Correlatien of SATbV, SATBI, Secondary School Per- centile Rank, English Composition Test, and Achievement Test Average, with.rreshman Grade-Point Average for Rural, Urban, and Control groups Predictor Rural urban Control SAT-Verbal .296 .439 .402 SAT-lath. .337 .305 .272 Percentile Rank .191 .441 .320 Eng. Comp. Test .319 .448 .409 Ach. Test. Avg. .390 .540 .484 The question next becomes one of considering whether or not these correlations are enough different that the groups can be considered to have come from different populations in 46 regard to that particular variable. To test for this differ- ence in correlation the statistic Z - Zr - z _\/r 1 _..JL_. is used. This variable, introduced by R. A. Fisher, is normally distributed and, hence, its significance may be determined by reference to the nornal probability curve at z - .95.4 The results of these tests are given in Table 10. TABLE 10.-Tests of significance between pairs ef zero-erder correlations of predictor variables with Freshman GradedPoint Average for Rural (r), Urban (u), and Control (c) groups Variable Z(r,c) Z(u,c) Z(r,u) SAThVerbal 1 07 0 58 1.37 SAThlathematical 0 59 0 30 0.27 Percentile Rank 1.26 1.82‘ 2.36‘ Eng. Comp. Test 0 96 0 48 1.22 Ach. Test Avg. 0 99 0 94 1.54 Z - 1.645 .95 l"Significant at 5 per cent level It can be seen that the only significant differences in correlation are between the variable Secondary School Percentile Rank and Freshman Grade-Point Average of the Urban-Control comparison and of the Rural-urban comparison. Since the only variable which shows significant differ- ences in either neans or correlation with;rreshnan Grade-Point Average is the Secondary School Percentile Rank, it nay be of 41bid., p. 256. 47 interest to summarize those relationships. This is done in Table 11. TABLE 11.-Comparisons of loans and Correlations with Freshman Grade-Point Average for Secondary School Percentile Rank for Rural (r), urban (u), and Control (c) groups Statistic r vs c u vs c r vs u lean 1* an -.. Correlation ——- it an *‘Significant at the 5 per cent level The reader is cautioned, at this point, about the degree of confidence which may be placed in the conclusions regarding the Secondary School Percentile Rank. It will be remembered that the distribution of this variable is very skewed toward the upper end of the rankings. Even though the means have been compared by the use of a non-parametric test, it is still quite conceivable that this non-normality is causing the differences on this variable to appear greater than, in fact, they really are. There is still another cause for concern in this regard. Since relatively small numbers of students apply to the Univer- sity of Michigan from.the kind of schools which have been de- scribed as snall-rural, the differences may be due, in part at least, to the differences in sampling between the various groups in the study. oh f _‘.’ he“ i ii 'Illmllvlll I... .lpgl. I 48 DIFFERENCES IN STANDARD ERRORS OF ESTIIATE OF VARIABLES The standard errors of estimate for predicting the cri- terion Freshman Grade-Point Average for each of the variables are given in Table 12. TABLE 12.--Standard Errors of Estimate for predicting Freshman Grade-Point Average by SAT-V, SAT-I, Secondary School Percen- tile Rank, English Composition Test, and Achievement Test Average for Rural, urban, and Control groups Predictor Rural urban . Control SATAVerbal .705 .583 .693 SAT-Hath. .695 .618 ..728 Percentile Rank .724 .582 .717 Eng. comp. Test .699 .580 .691 Ach. Test Avg. .679 .546- .662 It is of interest at this point in the interpretation of the data to determine if these differences between groups are greater, as a whole, to an extent which cannot be attributable to chance. A method, proposed by Gulliksen and lilks, was used to test for the significance of these differences.5 To accomplish this test a pooled error of estimate (8.8.13) is obtained and the absolute value of the statistic G‘ is ob- tained by the formula 6‘ - R1J log. (S.E.1J) - 81 log. (8.E.1) - leog. (8.E.J). This statistic is distributed as Chi-square with one degree of freedom. The results are shown 5n. Gulliksen and s. s. um, "Regression Tests for Several Samples," Psychometrica, 1950, _]_._5_, 91-114. in Table 13. TABLE 13.-Resolts of the Gulliksen Test for Differences in Standard Errors of Estimate between the predictor variables and criterion for Rural and Control comparisons and urban Control comparisons Variable R vs. C U vs. C SATHVerbal 0.02365 2.46891 SAT-Bathematical 0.09124 2.21241 Percentile Rank 0.02124 3.53904 Eng. Comp. Test 0.25906 2.62996 Ach. Test Avg. 0.06600 2.96200 Since Chi-square for one degree of freedom is 3.8 at the 5 per cent level of confidence, it is obvious that none of these differences is significant. COMPARISON OF'THE IUUTIPLE REGRESSIOI PREDICTION EQUATIONS The next step in the interpretation of the data pre- sented is to look closely at the differences in prediction of the Freshman Grade-Point Average resulting when all or part of the predictor variables are used in a standard multi- ple regression equation. These stepdwise regression equations are presented in Table 14. The multiple correlations are first tested with the F ratio statistic, r - 32 (n-n-1)/ n (1.32), where n is the total number of cases and I the number of variables. The re- sults are r(rural) - 11.032, F(urban) - 11.667, and r(control) - 38.365. Since F.95 - 2.30, all of these results are we. «no. eemoc.o eaooo.eu Haooe.ou ounce.o sopse.o «meme.~u ee. wee. moaoo.o eeooe.o mpoce.o veeao.o venem.~u ee. ape. ebeee.e menoo.e emana.c ennnm.nn be. new. manoo.o neouo.e oveen.uu «a. can. eeeuo.e oeneo.e noun-to on. see. «smoo.o meeoo.on anooe.on munee.e Hanuo.o nuanm.an en. Hen. Hanco.o emeoo.o eenoe.e mvmne.o enmnm.nu on. one. «ecoo.o anaco.o meamo.c mam~e.nu on. we». unmee.o eenmo.o penem.au an ”we. enneo.e moeoe.nu aunts ee. one. sauce.e, «mooo.e eunoo.o amoeo.e mmnae.o oevoe.nc so. «no. ae~oo.o Hence.o eeooo.o eueao.e owaan.nu me. man. amnee.o meaoe.o sp-o.o v~ame.on on. one. emuoo.e maeeo.o memee.oi as. no“. meauo.o ounm~.o Hanan .m.m e .22 .noe .980 one .793 p.93 .13 eaten-8 queue noses» Hosanna one .nennb .Hensm non soaneodno one ands nsoaowoenn no user—33:8 weaned-tuna... on» no Salaam no been e358 83 e... 3033330 esoauasm usnvunnon as» one nauseounneoo one nuneunnoo scannenueu ennniaoomtt.v~ Manda 51 significant, and we can reject the hypothesis that the multi- ple correlations of the entire population from which each of these samples is drawn is equal to zero. In order to make some Judgment about the comparability of the results obtained when predicting Freshman Grade-Point Average using separate regression equations for the various groups, a method proposed by Gulliksen and Wilks was followed.6 The method considers tests for three hypotheses regarding the populations from which the different groups are drawn: (a) RA, the hypothesis that all standard errors of estimate are equal; (b) St, the hypothesis that all regression lines are parallel, (assuning HA), and (c) at, the hypothesis that the regression lines are identical, (assuming Rb). Since the variables used in the selection of students at the University of Iichigan are Secondary School Rank, SATBV and SAT-fl, because the Gulliksen-Wilks procedure is computa- tionally laborious, and since including the English Composi- tion Test and the Achievement Average would add very little to understanding of the relative usefulness of these prediction equations, it was decided to restrict the comparisons to these multiple regression equations which included only the varia- bles Secondary School Rank and SAT scores. In comparing both the Urban-Control groups and the Rural-Control groups, it was found that the standard errors of estimate do not differ significantly at the 5 per cent 61bid. 52 level of confidence. For the Urban-Control comparison the test statistic GA - 3.132 and for the Rural-control compari- son GA - 0.04, Chi-square for one degree of freedom is 3.8. loving on to test Rh, significant differences are found indicating that it must be considered that the re- gression lines are not parallel. This also curtails the Gulliksen-Wilks test of comparison since it is obvious that BC, the hypothesis that the regression lines are identical, is false. Since it appears obvious at this point that the vari- able Secondary School Percentile Rank is the primary force acting on the prediction equations to cause them.to be dif- ferent, and because all other variables vary in the same di- rection and to a comparable degree for the three groups being studied, it seems reasonable to look at the regression equa- tions for predicting Freshman Grade-Point Average based on Secondary School Percentile Rank alone. These equations are depicted in Figure 1. It seems plausible to assume that multiple regression equations based on all variables, or on some combination of the other variables with Secondary School Rank, would have the same configuration as the lines of Figure 1. As a check on this assumption, a sampling of Predicted Freshman Grade-Point Averages was computed. These predicted grades were based on the regression equations using the Secondary School Percentile Rank and SAT-Verba1.(See Table 14.) The results of these computations are to be found in Table 15. 53 TABLE 15.-~Predicted Freshman Grade-Point Averages using re- gression equations with variables SAT-V and Secondary School Percentile Rank for Rural, Urban, and Control groups and dif- ferences between the Control group and Rural and Urban groups PREDICTED FRESHMAN G.P.A. DIFFERENCES RANK SAT-V *Rural Urban Control CSR C40 60 500 1,561 1,263 1.648 .087 .385 550 1.681 1.389 1.805 .124 .416 600 1.799 1.515 1.961 .162 .446 650 1.918 1.641 2.118 .200 .477 700 2.037 1.767 2.274 .237 .507 750 2.156 1.803 2.431 .275 .538 70 500 1.731 1.591 1.856 .125 .265 550 1.851 1.717 2.013 .162 296 600 1.969 1.843 2.169 .200 .326 650 2.088 1.969 2.326 .238 357 700 2.207 2.095 2.482 .275 .387 750 2.326 2.221 2.639 .313 418 80 500 1.001 1.919 2.064 .163 145 550 2.021 2.045 2.221 .200 .176 600 2.139 2.178 2.377 .238 .199 650 2.258 2.297 2.534 .276 .237 700 2.377 2.423 2.690 .313 .267 750 2.496 2.549 2.847 .351 .298 90 500 2.071 2.247 2.272 .201 .025 550 2.191 2.373 2.429 .238 .056 600 2.309 2.506 2.585 .276 .079 650 2.428 2.625 2.742 .314 .117 700 2.547 2.751 2.898 .351 147 750 2.666 2.877 3.055 .389 .178 100 500 2.241 2.575 2.480 .239 -.095 550 2.361 2.701 2.637 .276 -.064 600 2.479 2.834 2.793 .314 -.041 650 2.598 2.953 2.950 .352 -.003 700 2.717 3.079 3.106 .389 .027 750 2.836 3.205 3.263 .427 .058 «mm! “Min'f 'Vr's 2 A A - " ia'h‘ A"?! l N PREDICTED PRESRMAN G.P.A. H 6O 70 80 90 100 SECONDARY SCHOOL PERCENTILE RANK FIGURE 1.-Comparison of the regression equations for predict- ing the Freshman Grade-Point Average from the Secondary School Percentile Rank for Rural, Urban, and Control groups For further clarification of the relationship of the regression lines representing each of the study groups, those lines have been presented graphically for SAT-V scores of 500 and 750 in Figures 2 and 3. It becomes clear from Table 15 and Figures 2 and 3 that if a single prediction equation (the Control group) is used to determine the eligibility for admdssion of all students, the academic average for rural students of the type included in this study will be over-predicted. The amount of this over- prediction is minimized somewhat by the addition of the vari- able SATBV to the Secondary School Percentile Rank, but the difference is still substantial. Table 15 indicates that this 55 U N CONFROL GROUP —} w%/_// RURAL GROUP 3 PREDICTED FRESHHAN G.P.A. p..- 1 60 7O 80 90 100 SECONDARY SCHOOL PERCENTILE RANK FIGURE 2.-Comparison of the regression equations for predict- ing the Freshman Grade-Point Average from the Secondary School Percentile Rank when the SAT—V equals 500 for Rural, urban, and Control groups COMLOL GROUP —- \ Puma-rm mam G.P.A. E E 9 § H ,3 60 7O 80 90 100 SECONDARY SCHOOL PERCENTILE RANK FIGURE 3.-Comparison of the regression equations for predict- ing the Freshman Grade-Point Average from the Secondary School Percentile Rank when the SAT-V equals 750 for Rural, Urban, and control groups ...IJ .+ lyix‘flw a i ".1 u' 56 over-prediction will be as much as .42 of a grade-point at the upper ability level where many of the University of lich- igan students will be found. comparing, in a similar way, the prediction of grades for students from Large-Urban schools, it is clear that, ex- cept for students at about the 95 to 100 percentile rank, a single prediction equation developed for all students will over-predict their academic average. This over-prediction becomes greater as the percentile rank lowers. Using only the rank as a predictor, a single equation appears to ever-predict by .72 of a grade-point at the 50th percentile rank. The use of the SATAV as an added variable reduces this over-predic- tion since it can be shown that at the 50th percentile the difference ranges from .51 at SATAV equals 500 to .63 at SATAV equals 700. q ILI .V-bs‘ta.’ Ifl‘nlnlnhnul t a CHAPTER V CONCLUSIONS AND RECOMMENDATIONS CONCLUSIONS From the analysis and interpretation of the data in Chapter IF, it is now possible to draw some conclusions from the study. Before doing so, however, the reader's attention is drawn to certain limitations of the study. This is a study of the prediction of academic success at the University of lichigan. Although the results may have implications outside the population of students represented in the study, the uniqueness of the population must be considered. Since the University admits only the very able students, the range of abilities included in the study is quite restricted. Whether or not the conclusions reached could be applied to the full range of abilities is clearly not answered by the study. A more ideal study design (and a recommendation for further study) would have drawn its sample from a college or univer- sity which does not restrict admissions. Such a condition 'will be most difficult to obtain since institutions which have such "open-door" policies typically do not require tests for admission. The data might be obtained if some college or uni- versity was willing to administer such tests to all or a ran- dom sampling of those admitted soon after arrival. 57 58 A related limitation is represented in the variable Secondary School Percentile Rank. Most statistical procedures used in a study of this sort depend upon the assumption of normal distribution. All variable distributions in this study except the Secondary School Percentile Rank, are reason- ably close to normal. Although certain non-parametric pro- cedures are applied to this variable, the conclusions about it are open to some question. 0n the other hand, this is a statistic being used at the university of Michigan in its ad- missions operation, and to analyze that process it is impor- tant that the data be used as it is in the actual process if conclusions about its effect on the prediction of grades are to be adequately drawn. A third limitation may be found in the fact that there are a varying number of sample cases from variable to vari- able and from group to group. Although this difference is small and believed to have a negligible effect on the con- clusions it should be given consideration when comparing zero-order correlations in particular. The conclusions which seem to be suggested by the study are as follows: 1. The population of. students from Small-Rural high schools appears to be lower in level of ability on the SATéV, the SAT-I, the English Composition Test, the Achievement Test Average and the population of students as a whole. At the same time students from Large-Urban schools seem to be 59 enough like the total population, on all of these variables, to be considered to have come from the total population. 2. 0n the variable Secondary School Percentile Rank, the population of Small-Rural students appears, again, sig- nificantly different from the population as a whole, but in this case they tend to present higher ranks than the students representing the entire class. The students from Large-Urban schools also appear to differ significantly in class rank and tend to present higher ranks than the population as a whole. Interestingly, the Large-Urban population does not seem to differ significantly from the Small-Rural population on this variable. The reader is again reminded that the derivation of the samples and their unusual distribution makes this con- clusion less secure than it would be if these questions about the data did not exist. 3. It can be concluded from 1 and 2 that students of equal ability as measured by SATHV, SAT-fl, Achievement Tests, and by their Freshman Grade-Point Average tend to be ranked higher by the Small-Rural schools than those in Large-Urban schools and higher in Large-urban schools than in all schools taken together. This is probably to be expected since (a) the Small-Rural schools have less competition for position in class standing than do larger schools and schools which tend to send large numbers to college, (b) in this study, a per- centile rank was used as a measure of class standing and this makes no allowance for size of class, and (c) the Large-Urban group has missing from its population, as a result of the . 60 sampling procedure used, the more homogeneous academically- oriented college-preparatory high schools which tend to be the source of a high percentage of the total freshman class. Since the Large-Urban schools in this sample are more compre- hensive in nature, the college-bound in the schools tend to rank higher in their classes than would those in schools where more emphasis is placed on preparation for college. 4. Between the Small-Rural group and the population as a whole there are no significant differences in either the zero-order correlations or the errors of estimate between predictor variables and the criterion. 5. Between the Large-Urban group and the population as a whole there are no significant differences in either the zero-order correlations or errors of estimate except for the variable Percentile Rank. Here it was found that the Large- Urban group has a higher correlation of Percentile Rank with Freshman Grade-Point Average with a somewhat lower error of estimate. 6. From 4 and 5 it can be concluded that all variables except rank-in-class may be considered comparable predictors of academic success in college regardless of whether the school is Small-Rural or Large-Urban. Further, keeping in mind the concern for sampling and non-normality, it can be concluded that the rank in class presented by the Large-urban comprehen- sive secondary school is a better predictor of academic success in college than are either the ranks presented by Small-Rural schools or by the schools in general. An explanation for this 61 might be that in the Small-Rural schools, as has already been indicated, the small number of college-bound students tend to rank high. In the total freshman class, however, but not rep- resented by either of the samples studied, are large numbers of students from out of the state where only the top ranking students are accepted. Also not included in these samples are students from the suburban, college-oriented communities where the schools have high percentages of very able students. 7. The use of a multiple prediction equation, as op- posed to depending only on rank in class appears to improve_ the University of lichigan's ability to predict academic suc- cess. This is especially true for students from Small-Rural secondary schools where the increase in correlation was much greater than for other students. Public institutions in gen- eral would do well to consider this aspect seriously. Public colleges have been reluctant to use tests in the admission process but have rather predominantly relied on secondary school rank-in—class. They have done this under the flag of Democracyh-being fair to high school graduates by not being "selective" through the use of tests. It would appear that they will be more "unfair" to students when using only rank— in-class as a predictor than they would be if they added stand- ardized aptitude and achievement tests to their admission cri- teria. The tests, as evidenced by this study, 2232 to equal- ize the disparity in ranking students. 8. The slopes and/or intercepts of the multiple pre- diction equations are enough different that for some students 62 in both the Small-Rural schools and the Large-Urban schools, the use of a single prediction equation may over-predict the academic average he can be expected to make during his first year in the University. Specifically, it appears that the Freshman Grade-Point average for students from Small-Rural schools is fairly uniformly over-predicted, possibly by as much as a half a grade point. Students from Large-Urban schools, who rank in the upper 5 per cent of their class, seem not disadvantaged by the use of a single predictor but students who rank nearer the middle of their class may be over-predicted by as much as three-fourths of a grade-point. It is now possible to turn to the questions posed as the basis for this study. The results may be summarized as follows: 1. The results of the study suggest that, the Scho- lastic Aptitude Test of the Cbllege Entrance Examination Board is ngt_biased as a predictor of academic success in college for applicants from either Small-Rural or Large-Urban type schools. 2. The results of the study suggest that, the Achieve- ment Tests of the Cbllege Entrance Examination Board are not biased predictors of academic success in college for appli- cants from either Small-Rural or Large-Urban type schools. 3. The secondary school achievement record, at least as represented by a secondary school percentile rank is, ap- parently, affected by the cultural nature and size of the high school. 63 4. The combination of predictors used in the predic- tion of academic success do not appear to predict as well, or in the same way, for applicants from Small-Rural secondary schools or for applicants from Large-Urban secondary schools as they do for the applicant group as a whole. RECOMMENDATIONS There are certain recommendations which can be sug- gested to the University of Michigan and certain other recom- mendations for further study. Those recommendations directed to the University of Michigan deal directly with the use of the predictor variable Secondary School Percentile Rank. This variable, as it is now used, holds the potential of misleading those concerned with making Judgments about the future academic success of applicants. Every attempt should be made to ameliorate this circumstance. This is not to say that admissions counselors do not take this factor into ac- count now as a subgective factor in the admissions process. There are some statistical applications, however, which might improve prediction for these students. One way might be to use, as a measure of secondary school achievement, the grade average obtained from courses which could be considered strictly academic. A second approach, might be to make an allowance for the size of graduating classes. Such a tech— nique was described on page 36 of this study. A third way, 64 might be to add an adjustment factor to the predicted grades of students from special secondary schools. In order to accomplish the latter the University is encouraged to develop experience tables for other clearly defined groups i.e., the academically oriented secondary schools, private schools, large city schools, out-of-state schools, etc. There is a limit to this subdividing, of course, but since this seems to be such a key factor in admissions of- forts in this direction would appear to be well worth the re- sults. Although this study substantiates the hypothesis that there is a lack of bias in the tests of the College Entrance Examination Board for students of rural or urban cultural backgrounds, it does not answer what may be a doubt about other cultural factors. Other studies need to be carried out to see if the hypothesis regarding the lack of cultural bias can be extended to other factors. Finally, this study clearly suggests that a primary stumbling block to the improvement of prediction of academic success in college is the unreliable assessment of the appli- cant's secondary school achievement. Reliance on rank-in- class, as a.measure of past achievement, appears to be quite ”unfair" to some candidates even when this factor may be com- bined with such equalizing factors as aptitude and achieve- ment test results. It may be "unfair" to applicants, such as have been identified in this study as Small-Rural type applicants, to assume on the basis of high school rank that 65 they can compete successfully with applicants from other schools with comparable class ranks. Such an assumption, coupled with what must be assumed to be great social adjust- ments could result in failures, which might be tragic for some individuals. It must be assumed that errors of a dif- ferent nature occur when over-reliance on secondary school rank-in-class is part of the admission decision for appli- cants from such special groups as private schools, or public schools with high-average academic ability. This study was based on the assumption that there was probably a tie between ruralism and small schools. The study suggests fairly clearly that ruralism at least, is not a con- tributing factor to the relative prediction of success. It seems clear, too, that the size of the school is only impor- tant in that it leads to misunderstanding about the relative ability of an applicant. Size of school does not, apparently, either add to or detract from the applicant's real ability to do college-level work. What is needed, then, is additional research into ways of comparing the achievement record of applicants from various size and types of secondary schools in order not to either over-predict or under-predict college achievement on this basis. liiill'il" I. APPENDIX APPENDIX A MICHIGAN HIGH SCmOLS CLASSIFIED AS SHALL-RURAL Post Office Alba Alpha Amasa Arcadia Atlanta AuGres Baldwin Baraga Barryton Bear Lake Beaverton Bellaire Bergland Blanchard Boyne City Boyne Falls Brethren Bremley Britten Buckley Carney Cedarville Central Lake Champion Channing Chassell Chatham Crystal Falls Custer De Teur Dollar Bay Elk Rapids Ellsworth Evart Ewen Fairview Felch Frankfort Frederic Freesoil Garden 66 Bi h School Alba High School Mastodon Twp. 3.8. r Hematite Twp. B.S. 3 Arcadia B.S. Atlanta B.S. AuGres-Sims B.S. Baldwin E.S. Baraga B.S. Barryton B.S. Bear Lake H.S. Beaverton R.S. L Bellaire B.S. Bergland B.S. Blanchard H.S. Boyne City B.S. Boyne Falls B.S. Brethren B.S. Bremley H.S. Britten H.S. Buckley H.S. Carney B.S. Cedarville 3.8. Central Lake B.S. Champion B.S. Channing H.S. Chassell 3.8. Chatham H.S. Crystal Falls B.S. Custer H.S. De Tour B.S. Dollar Bay H.S. Elk Rapids H.S. Ellsworth H.S. Evart B.S. Ewen B.S. Fairview B.S. Felch H.S. Frankfort B.S. Frederic H.C. Freesoil B.S. Garden B.S. Post Office Grand Marias Grant Grayling Gwinn Hale Harbor Springs Harris Hart Hermansville Hesperia Hillman Indian River Jehannesburg Kalkaska Kingsley Kingston Lake City Lake Linden L‘Anse LeRoy Luther Mackinaw City Mancelona Manton Maple City Marenisco Marion Mass McBain Merritt Mesick Hichigamme Mio Morley Nahma National Mine Newaygo Northport Norway Onaway Onekama Ontonagon Painesdale Paradise Pellston Pentwater Perkins Posen Powers 67 High School Burt Twp. H.S. Grant H.S. Grayling H.S. Gwinn H.S. Hale H.S. Harbor Springs H.S. Bark River-Harris H.S. Hart H.S. Hermansville H.S. Hesperia H.S. Hillman H.S. Inland Lakes H.S. Jehannesburg H.S. Kalkaska H.S. Kingsley H.S. Kingston H.S. Lake City H.S. Lake Linden L'Anse H.S. LeRoy H.S. Luther H.S. Mackinaw City H.S. Mancclona H.S. Manton H.S. Glen Lake H.S. Marenisco H.S. Marion H.S. Mass H.S. McBain H.S. Merritt H.S. Mesick H.S. Hichigamme H.S. Mio H.S. Morley-Stanwood H.S. Nahma H.S. National Mine H.S. Newaygo H.S. Northport H.S. Norway H.S. Onaway H.S. Onekama H.S. Ontonagon H.S. Jeffers H.S. Whitefish Twp. H.S. Pellston H.S. Pentwater H.S. Perkins H.S. Posen H.S. Powers-Spalding H.S. Post Office Rapid City Rapid River Remus Republic Rock Rockland Roscommon Rose City St. Ignace Stambaugh Stanton Suttons Bay Trenary Trout creek Tustin Vanderbilt Vestaburg Vulcan Wakefield Walkerville Watersmeet Weidman White Cloud White Pine Whittemore Wolverine 68 ‘Eigh School Rapid City H.S. Rapid River H.S. Remus H.S. Republic H.S. Rock H.S. Roger Clar H.S. Gerrish-Higgins Rose City H.S. LaSalle H.S. Stambaugh H.S. Stanton H.S. Suttons Bay H.S. Trenary Trout Creek H.S. Tustin H.S. Vanderbilt H.S. Vestaburg H.S. Vulcan H.S. Wakefield H.S. Walkerville H.S. Watersmeet H.S. Weidman H.S. White Cloud H.S. White Pine H.S. Whittenore H.S. Wolverine H.S. H.S. APPENDIX B MICHIGAN HIGH SCHOOLS CLASSIFIED AS LARGE-URBAN Post Office Adrian Allen Park Battle Creek Benton Harbor Berkley Dearborn Heights Dearborn Heights Detroit Detroit East Detroit Perndale Flint Flint Garden City Grand Blanc Grand Haven Grand Ledge Grosse Pointe Hamtramck Hazel Park Highland Park Holland Inkster Lansing Lapeer Lincoln Park Madison Heights Marshall Melvindale Midland Milford Monroe Mount Clemens Mt. Morris Muskegon Mona Shores Muskegon Heights Niles Oak Park Owosso Plymouth 69 EEEh School Adrian H.S. Allen Park H.S. Battle Creek H.S. Benton Harbor H.S. Berkley H.S. Riverside H.S. Haston H.S. Redford Union Lee M. Thurston H.S. East Detroit H.S. Ferndale H.S. Beecher H.S. Ainsworth H.S. Garden City H.S. Grand Blanc H.S. Grand Haven H.S. Grand Ledge H.S. Grosse Pointe H.S. Hantramck H.S. Hazel Park H.S. Highland Park H.S. Holland H.S. Inkster H.S. waverly H.S. Lapeer H.S. Lincoln Park H.S. Madison Heights H.S. Marshall H.S. Melvindale H.S. Midland H.S. Milford H.S. MOnroe H.S. Mount Clemens H.S. Mt. Morris H.S. Muskegon H.S. Mona Shores H.S. Mnskegon Heights H.S. Niles H.S. Oak Park H.S. Owosso H.S. Plynouth H.S. 70 Post Office Portage Port Huron River Rouge Rochester Romulus Roseville Saginaw St. Claire Shores St. Claire Shores St. Claire Shores St. Joseph Southfield Southgate Southgate South Haven Sturgis Taylor Trenton Troy Walled Lake Warren Warren Wayne Wyandotte Ypsilanti gig School Portage H.S. Port Huron H.S. River Rouge H.S. Rochester H.S. Romulus H.S. Roseville H.S. Douglas MacArthur H.S. Lake Shore H.S. Lakeview H.S. South Lake H.S. St. Joseph H.S. Southfield H.S. Southgate H.S. Schafer H.S. South Haven H.S. Sturgis H.S. Taylor Center H.S. Trenton H.S. Troy H.S. Walled Lake H.S. Fitzgerald H.S. Lincoln H.S. Wayne H.S. Theodore Roosevelt H.S. Ypsilanti H.S. APPENDIX C DATA SUMMARY The Scholastic Aptitude Test-Verbal TABLE lA.--Frequency distributions, means, and standard deviations of the verbal section of the Scholastic Aptitude Test for Rural, urban, and Control groups — f 4 Rural urban Control Score % Below % Below ‘1; Below Intervals N Interval N Interval N Interval 750-799 0 100 1 100 4 99 700-749 4 96 I4 94 31 93 650-699 6 90 36 80 60 81 600-649 13 77 48 61 94 62 550-599 26 51 49 43 106 40 500-549 19 33 52 23 78 25 450-499 14 19 25 13 82 8 400-449 14 5 26 3 24 3 350-399 3 2 7 0 15 0 300-349 2 0 O 0 0 0 250-299 0 0 0 0 1 0 Total N 101 256 495 Mean 537 568 569 S. D. 87 88 87 71 72 The Scholastic Aptitude Test-Mathematical TABLE 2A.--Frequency distributions, means, and standard devia- tions of the mathematical section of the Scholastic Aptitude Test for Rural, urban, and Control groups Rural Urban Control Score '1. Below % Below % Below Intervals N Interval N Interval N Interval 750-799 0 100 11 96 23 95 700-749 6 94 33 83 60 83 650-699 12 82 32 70 81 67 600-649 14 68 48 52 103 46 550-599 21 48 50 32 89 28 500-549 18 30 38 17 61 16 450-499 16 14 28 6 43 7 400-449 12 2 12 2 21 3 350-399 1 l 4 0 11 1 300-349 1 0 0 0 2 0 250-299 0 0 0 0 l 0 Total N 101 256 495 Mean 555 598 602 S. D. 91 94 97 73 The Secondary School Percentile Rank TABLE 3A.--Frequency distributions, means, and standard devia- tions of the Secondary School Percentile Rank for Rural, urban, and Control groups Rural Urban Control Score % Below ‘5 Below ‘1» Below Intervals N Interval N Interval H Interval 95.00-99.00 52 49 118 54 189 62 90.00-94.99 26 23 62 30 117 38 85.00-89.99 8 15 44 13 73 24 80.00-84.99 10 5 18 5 60 11 75.00-79.99 3 2 8 2 25 6 70.00-74.99 2 0 4 1 l6 3 65.00-69.99 0 0 2 0 9 1 60.00-64.99 0 0 0 0 2 1 55.00-49.99 0 0 0 0 1 1 50.00-54.99 0 0 0 0 1 1 45.00-49.99 0 0 0 0 0 l 40.00-44.99 0 0 0 0 2 0 35.00-39.99 0 0 0 0 0 0 30.00-34.99 0 0 0 0 l 0 Total N 101 256 496 Mean 92.61 92.23 89.72 S. D. 6.73 6.85 9.26 Al F. . (so ..,.....1.- . am.x..F,i.‘ . - C. ||ll|lllll|lcl 74 The English Composition Tost TABLE 4A.--Frequency distributions, means, and standard devia- tions of the English Composition Test for Rural, Urban, and Control groups Rural Urban Control Score % Be low '1. Below % Below Intervals N Interval N Interval N Interval 800-800 0 100 1 100 2 100 750-799 0 100 2 99 7 98 700-749 1 99 6 96 28 91 650-699 3 95 27 84 38 82 600-649 11 83 43 66 59 68 550-599 19 60 47 45 86 47 500-549 24 33 47 25 89 26 450-499 l6 14 26 13 70 9 400-449 6 7 20 5 28 3 350-399 5 1 8 1 10 0 300-349 0 l 3 0 1 0 250-299 1 0 0 0 0 0 Total N 86 - 230 418 Mean 531 558 563 S. D. 78 89 92 75 The Achievement Test Average TABLE 5A.--Frequency distributions, means, and standard devia- tions of the Achievement Test Average for Rural, Urban, and Control groups Rural Urban Control Score % Below % Below $.Below Intervals N Interval N Interval N Interval 800-800 0 100 0 100 1 100 750-799 0 100 1 100 4 99 700-749 0 100 6 97 19 95 650-699 4 96 25 86 50 83 600-649 13 82 46 67 70 68 550-599 18 63 56 43 108 44 500-549 24 37 51 22 112 19 450-499 26 10 36 7 61 5 400-449 9 0 12 2 19 1 350-399 0 0 3 0 4 0 300-349 0 0 1 0 0 0 250-299 0 0 0 0 0 0 Total N 94 237 448 Mean 534 567 573 S. D. 65 77 77 76 The University Freshman Grade-Point Average TABLE 6A.--Frequency distributions, means, and standard devia- tion of the University Freshman Grade-Point Average for Rural, Urban, and Control groups Rural Urban Control Score % Below % Below ‘1. Below Intervals N Interval N Interval N Interval 4.00-4.00 0 100 4 98 4 99 3.75-3.99 0 100 4 97 16 96 3.50-3.74 4 96 8 94 26 91 3.25-3.49 2 94 18 87 23 86 3.00-3.24 8 86 20 74 44 77 2.75-2.99 7 79 27 68 66 64 2.50-2.74 12 67 36 54 54 53 2.25-2.49 12 55 42 38 79 37 2.00-2.24 22 33 47 20 79 21 1.75-1.99 9 24 17 13 39 13 1.50-1.74 7 17 20 5 30 7 1.25-1.49 8 9 7 2 8 5 1.00-l.24 2 7 2 2 5 4 0.75-0.99 1 6 0 2 4 3 0.50-0.74 4 2 2 1 6 2 0.25-0.49 0 2 l 0 2 1 0-0.24 2 0 1 0 7 0 Total N 100 256 492 Mean 2.17 2 47 2.47 s. n. .74 :65 .74 APPENDIX D ANALYSIS OF VARIANCE SUMMARY TABLES TABLE 7A.--Ana1ysis of variance summary table for Scholastic Aptitude Test-Verbal Source of Variation d.f. Sum of Squares Mean Square Fa Groups 2 90,232 45,116 5.91 Within-groups 849 6,484,961 7,638 Total 851 6,575,193 a F.95(2,849) - 3.00 TABLE 8A.--Ana1ysis of variance summary table for Scholastic Aptitude Test-Mathematical Source of Variation d.f. Sum of Squares Mean Square Fa Groups 2 181,430 90,715 8.81 Within-groups 849 7,710,070 10,293 Total 851 7,891,500 a - F.95(2,849) 3°00 TABLE 9A.-Ana1ysis of variance summary table for Secondary School Percentile Rank _J r w J L Source of Variation d.f. Sum.of Squares Mean Square Fa Groups 2 1,435 717.5 10.36 Within-groups 850 58,894 69.3 Total 852 60,329 “F.95(2,850) ' 3-00 77 78 TABLE 10A.-Analysis of variance summary table for the English Composition Test Source of Variation d.f. Sum of Squares Mean Square Fa Groups 2 72,619 36,309 4.52 Within-groups 731 5,867,512 8,026 Total 733 5,940,131 aF 3.00 .95(2,731) ' TABLE 11A.-Analysis of variance summary table for the Achieve- ment Test Average Source of Variation d.f. Sum of Squares Mean Square F3 Groups 2 115,349 57,675 10.04 Within-groups 776 4,457,724 5,744 Total 778 4,573,073 ”F.95(2,776) ' 3-°° TABLE 12A.-Ana1ysis of variance summary table for the univer- sity Freshman Grade-Point Average Source of Variation d.f. Sum of Squares Mean Square F“ Groups 2 8.057 4.028 7.91 Within-groups 845 430.748 .509 Total 847 438.805 a - F.95(2,845) 3'00 I Jill ’lll llll.‘ I'll-l I. I." 11' u INTERCORRELATIONS OF ALL VARIABLES TABLE l3A.-Correlation Matrix APPENDIX E Fr. Eng. Ach. Fr.G.P.A. Rural 1.000 0.191 0.296 0.337 0.319 0.390 Urban 1.000 0.441 0.439 0.305 0.448 0.540 Control 1.000 0.320 0.402 0.272 0.409 0.484 Rank Rural 1.000 0.149 0.238 0.116 0.164 Urban 1.000 0.289 0.290 0.410 0.442 control 1.000 0.200 0.184 0.283 0.271 Urban 1.000 0.471 0.675 0.699 Control 1.000 0.414 0.660 0.674 SAT-M Rural 1.000 0.327 0.583 Urban 1.000 0.392 0.622 Control 1.000 0.343 0.569 Eng.Comp. Rural 1.000 0.695 Urban 1.000 0.815 Control 1.000 0.798 Ach.Avg. Rural 1.000 urban 1.000 Control 1.000 79 BIBLIOGRAPHY BIBLIOGRAPHY BOOKS Black, Hillel. They Shall Not Pass. New York: Wm. MOrrow and Co., 1963. Chauncey, Henry and JOhn E. Dobbin. Testing: Its Place in Education Today. New York: Harper and Row, 1963. Davis, A. Social Class Influencegpon Learning. Cambridge, Mass.: Harvard University'Press, 1948. Herrick, J. E. "What is Already Known About the Relation of the I.Q. to Cultural Background," Chapter II of Intelli- ence and Cultural Differences by Eells, K. et aI. CEIcago: niversity“61’Chicago‘Press,71957. '__—__ Jones, H. E. "Environmental Influences on Mental Develop- ment," Chapter XI of Manual of Child Psycholo y, ed. Leonard Carmichael. New YorE: JoEE WIIey and Sons, Inc., 1946. Klineberg, O. Negro Intelligence and Selective Migration. New Yurk: CEIfimbia Univers y ess, . Lavin, David E. The Prediction of Academic Performance. New Yerk: Russell Sage Foundation,1965? Lindquist, E. P. Design and Anal sis of Experiments in Psy- chology and Education. Baston: HEugEfon IIIIIIn 05., Lorimer, F. and r. Osborn. Dynamics of Population. New York: MacMillan and company, 1934. McNenar, Quinn. Psychological Statistics. New Yerk: John Wiley and Sons, . Walker, Helen M. and Jeseph Lev. Statistical Inference. New York: Henry Holt and Co., 1953. Wiser, B. J. Statistical Principles in Experimental Design. New York:‘MEGF333HTII_CETT'I962. ‘Jr "' YOung, Robert and Donald Veldman. Introductory Statistics for the Behavioral Sciences. 80 81 ARTICLES AND PERIODICALS Allison, G. and A. Barnett. "Freshman Psychological Examina- tion Scores As Related to Size of High Schools," JOurnal of Applied Psychology, 1940, 24, 651-652. Altman, Esther R. "The Effect of Rank in Class and Size of High School on the Academic Achievement of Central Michi- gan College Seniors, Class of 1957," Journal of Educa- tional Research, 52, 1959, pp. 307-309. Cattell, R. B. "A Culture-free Intelligence Test," Journal of Educational Psychology, 1940, 31, 161-179. Coffman, William E. "Developing Tests for the Cultmfll; Dif- ferent," School and Society, 1965,‘g§, 430-433. Davis, A. "Socio-economic Influences on Learning," Phi Delta Feder, D. D. "Factors Which Affect Achievement and Its Pre- diction at the College Level," JOurnaI of the American Association of-Collegiate Registrar , IHIU,'15;”IU7:II8; EfiucatIOnaI Ab§tracts, I940, E, No. 292, 76{- Gist, N. P. and C. D. Clark. "Intelligence As A Selection Factor in Rural-Urban Migration," American Journal of Sociology, 1928,‘42, 284-294. Guiliksen, H. and S. S. Wilkes. "Regression Tosts for Sev- eral Samples," Psychometrica, 1950, 12, 91-114. Haller, Archie O. and Sewell, Wm. H. "Farm.Residence and Level of Educational and Occupational Aspiration," American Journal of Sociology, 1957,‘§2, 407-11. Havighurst, R. J. "Using the I.Q. Wisely," N.E.A. Journal, 1951, 42, 540-541. ’ Hoyt, Donald P., "Size of High School and College Grades," Personnel and Guidance Journal, 1959'.§Z! 569-573. Irion, T. and F. C. Fisher. "Testing the Intelligence of Rural School Children," American School Master, 1921, 2.3., 22 1.223 e Klineberg, O. "A Study of Psychological Differences Between 'Racial' and National Groups in Europe, Archives of Psy- chology, 1931, No. 132. 82 Koch, H. L. and R. Simmons. "A Study of the Test Performance of American, Mexican, and Negro Children," Psychological Mauldin, W. P. "Selective Migration From.Sma11 Towns," Ameri- can Sociological Review, 1940, 5, 748-758. Neff, W. S. "Socio-Economic Status and Intelligence: A Crit- ical Survey," Psychological Bulletin, 1938, 35, 727-757. Nelson, C. W. "Testing The Influence of Rural and urban En- vironment on ACE Intelligence Test Scores," American Sociologist, 1942, 1, 743-751. Pintner, R. "A Mental Survey of the School Population of a Village," School and Society, 1917,.5, 597-600. Pressey, L. W. "The Influence of Inadequate Schooling and Poor Environment Upon Results of Tests of Intelligence," Journal of Applied Psychology, 1920,'3, 91-96. Pressey, S. L. and J. B. Thomas. "A Study of Country Children in a Good and Poor Farming District by Means of A Group Scale of Intelligence," Journal of Applied Psychology, 1919, 8, 534-539. Sanders, William B., R. Travis Osborne, and Joel E. Greene, "Intelligence and Academic Performance of College Students of Urban, Rural, and Mixed Backgrounds," Journal of Educa- tional Research, 39, 1955, pp. 185-193. Sanford, G. A. "Selective Migration in a Rural Alabama Com- munity," American Sociological Review, 1940, 5, 759-766. Shaw, Merville C., and Donald J. Brown, "Scholastic Under- achievement of Bright College Students," Personnel and Guidance Journal, 36, 1957, pp. 195-199. Shimbug, M. "An Investigation Into the Validity of Norms with Special Reference to Urban and Rural Groups," Archives of Psychology, 1928-29, 19, Ser. 104. Smith, M. "An Urban-Rural Intellectual Gradient," Sociological Society Research, 1943, 21, 307-315. Smith, W. R. "Test Results Reveal Advantages of Larger Schools," Psychological Abstracts, 1930, 12, No. 2141, 234. Washburne, Norman E. "Socio-economic Status, Urbanism and Academic Performance in College," Journal of Educational ‘Research, 53, 1959, pp. 130-137. 83 Wheeler, L. R. "A Comparative Study of the Intelligence of East Tennessee Mountain Children," JOurnal of Educational Psychology, 1942, 33, 321-334. OTHER SOURCES Anastasi, A. "Some Implications of Cultural Factors for Test Construction," Proceedin s 1949 Invitational Conference on Testing Problems,—Pr nceton, N. 3.: EducatiOnal Test- ng erv ce, , 3-17. College Entrance Examination Board. A Description of the Scholastic Aptitude Test. Princeton, N.J.:;E3ucat10nal listing Service, 1965. College Entrance Examination Board. College Board Score Re- rts: A Guide for Counselors and Idiissions Uiiicers. inceton, 3.3.: EducationalTosting Service, 1965. College Entrance Examination Board. A Description of the Colle e Board Achievement Tests. Ptinceton, N.J.: Educa- ona esting Service;“1965. Dyer, Henry S. and Richard G. King. Colle e Board Scores: Their Use and Interpretation. Princetomm ing Service, . Flanagan, John C. et al. A Surve and Follow-up Study of Edn- cational Plans—mechEFiE—W terns: Studies of tEe Kierican HIEE SchoI, PittsfiurgE: ‘TE37Uii?3FEiti‘ET‘Pittifififgfifi'1962. Frederickson, N., M. Olsen, and W. B. Schrader. Prediction of First Semester Grades at Ken n colle e 194 - . ace- Haggard, E. A. "Influence of culture Background on Test Per- formance," Proceedin s 1949 Invitational Conference on Tost- ing Problems, Ptinceton, N. 3.: Educationai Testing Service, ’ O Howell, John. A Compendium of Celle e Board Validit Stud Results 1958:1964. An unpuinsEea report to tit CSIIege Entrance Eiiiination Board, 1965. Lucas, Charles M. Surve of the Literature Relati to the Effects of CulturaI Background on Iptituae Test Scores, Princeton, N.J.: Eaficationai Testing Service, I953. 84 Mollenkopf, Wm. G. A Study of Secondary School Characteristics as Related to Test Scores,‘P?inceton, N.J.: Educational Testing Service, 1953. Schultz, D. G. The Relationship Between Colle e Grades and Aptitude Test'ScOres for Differentisocioeconomic Grou s, PtiiCeton, N.J.: Educational‘Tisting Service, 1953. Turnbull, Wm. W. "Influence of Cultural Background on Test Performance," Proceedings 1949 Invitational Conference on Testing Problems,‘Princeton, N.J.: Educationai Testing erv ce, , 9-34. Wing, Cliff W., Jr. and Virginia Ktsanes. The Effect of Cer- tain Cultural Background Factors on the Pfediction or Student GradesiiCEIIege. UnpuEIisEea Report to tEe CEIIege Entrance Eiamination Board, 1960. Jul“ 1'. I ‘ I. All. 1' 1|!