“wwit'cm \ m u . - ..“ SELECTION CRITERIA AND PROCEDURES FOR PREDICTING THE SUCCESS OF INDUSTRIAL EDUCATION DOCTORAL APPIJCANTS Dissertation for the Degree Of Ph. D. MICHIGAN STATE UNIVERSITY CHARLIE HARRIS - 1976 LIlRA RY WW U?“ . . i This is to certify that the t ,sis en*‘fled SELECTION CRITERIA AND PROCEDURES FOR PREDICTING THE SUCCESS OF INDUSTRIAL EDUCATION DOCTORAL APPLICANTS presented by Charlie Harris has been accepted towards fulfillment of the requirements for Ph.D. degree in Education Major professor Date May 6, 1976 0-7639 IIIIIII IIIIIIIIII II III III III III III II ABSTRACT SELECTION CRITERIA AND PROCEDURES FOR PREDICTING THE SUCCESS OF INDUSTRIAL EDUCATION DOCTORAL APPLICANTS BY Charlie Harris The selection of a potential doctoral student is usually based on all the estimates of student quality that are available. The purposes of this selection study were: 1. To investigate the selection criteria and pro- cedures being used in certain industrial education depart- ments. 2. To test the effectiveness of certain predictor variables (criteria) that industrial educators in selected Big Ten universities can use in selecting industrial edu- cation doctoral students. 3. To contribute to the deve10pment of a selection model that allows industrial educators in selected Big Ten universities to predict the success of potential industrial education doctoral students. The sample included seventy-five former doctoral students who were identified by seven industrial education Charlie Harris department representatives from selected Big Ten univer- sities. Fifty-four of these subjects were graduates and twenty-one were drop-outs. Fourteen independent variables were used as the initial set of predictors to discriminate between students who had graduated from or dropped out of the doctoral pro- gram. Success was the criterion variable for this study. Instruments for collecting data included the personal interview and the questionnaire. An interview was con- ducted with seven industrial education department repre- sentatives to collect data pertaining to the selection criteria and procedures used in their departments. These data assisted in choosing the initial set of predictor variables. The data collected on the fourteen predictors were compiled from the questionnaires. The procedures for collecting data included tele- phone calls and letters to department representatives from the seven departments. Data were screened for missing values before statistical application; the overall mean for a variable was substituted for missing values. The SPSS RAO stepwise discriminant analysis method was employed to test the hypotheses; the discriminant analysis classification equation was considered to be a selection model and a procedure for validation. The decision to reject or accept the null hypoth- esis was based on the chi-square statistics. The null Charlie Harris hypothesis was rejected since the chi-square value was significant (x2 = 34.23791; df = 14; p < .002). Rejecting the null hypothesis indicated that the alternative was accepted. By accepting the alternative hypothesis, the researcher implied that the following predictors were relevant for the selection of industrial education doctoral students: 1. Number of years taken to complete the master's program 2. Graduate Record Examination Quantitative score 3. Age at the time of application to the doctoral program 4. Overall undergraduate grade-point average 5. Marital status at the time of application to the doctoral program 6. Overall master's grade-point average 7. Miller Analogies Test score 8. Number of dependents at the time of application to the doctoral program 9. Number of publications listed on the application for the doctoral program 10. Years of relevant professional education work experience 11. Undergraduate grade-point average for the last two years Charlie Harris 12. Last employment before admittance to the doctoral program 13. Master's grade-point average for courses taken in industrial education 14. Graduate Record Examination Verbal score A further inspection of the F-to-enter from the computer output sheet revealed that the first nine pre- dictors were most effective with the selection-process (x2 = 34.680; df = 9; P < .000). Therefore, these nine variables would be best suited for developing a selection model. It must be stressed, however, that these variables only accounted for 37.25 percent of variance in the criterion variable. This implies that 62.75 percent of variance could be found in other predictor variables. Finally, variables number one, three, five, and nine had not officially been used with the selection pro- cess; therefore, their inclusion in a selection model would be an improvement over what had been done in the past. SELECTION CRITERIA AND PROCEDURES FOR PREDICTING THE SUCCESS OF INDUSTRIAL EDUCATION DOCTORAL APPLICANTS BY Charlie Harris A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Secondary Education and Curriculum 1976 ACKNOWLEDGMENTS I am grateful to many people who assisted with the completion of this dissertation. The guidance committee members have been motivating and understanding, and the chairman of the committee, Dr. George Ferns, has gone beyond what was required. An additional thanks goes to the seven Big Ten department representatives who assisted with the collection of data. Without the interest and cooperation of these peOple, the study could not have been completed. Finally, my family has been a dominant factor in providing spiritual leadership when it was needed. Special motivating forces were the memory of my deceased sister, Susie Harris, who was always encouraging and understanding; and the persistent assistance of my wife, Eula Harris. ii TABLE OF CONTENTS LIST OF TABLES I O O O O O O O O O O O 0 Chapter I. II. III. IV. STATEMENT OF THE PROBLEM . . . . Background and Need for the Study . . Purposes and Objectives of the Study Hypotheses . . . . . . . . . . Definition of Terms . . . . . . Delimitations of the Study . . Theory and Design for the Study Overview of the Study . . . . . REVIEW OF LITERATURE . . . . . . Introduction . . . . . . . . . Selection Studies in Industrial General Selection Studies . . . Summary . . . . . . . . . . . . RESEARCH DESIGN AND METHOD . . . Introduction . . Sources of Data . The Population . . . The Sample . . . . . Variables for the Study Predictor Variables . . . Criterion Variable . . . Data Collection . . . . The Instrument . . . The Procedures . . . Hypotheses . . . . . . Data-Analysis Procedures Summary . . . . . . . . . . . EVALUATION OF DATA . . . . . . . Introduction . . . . . . . . . Results of Interviews With Seven Department Representatives . iii Education Page I...- . OGQO‘GU‘IH l-‘ g..- N NNHH bCNN b) \D mbbbfiébbbbbbw NmmflGO‘UINNl-‘OOO U'IUIU'I hub-b Chapter Selection Criteria . . . . . Selection Process . . . . . . Features of Doctoral Degree Pr grams Availability of Subjects . . .Data Collection . . . . . . . Description of the Sample . . . Interpretation of the Hypotheses DeveloPment of a Selection Model smary O O O O I O O I O I I O V. SUMMARY AND CONCLUSIONS . . . . . APPENDICES Summary . . . . . Conclusions . . . Discussion . . . Recommendations . A. PERSONAL INTERVIEW DATA . . . . . B. QUESTIONNAIRE FOR DATA COLLECTION C. DATA FROM COMPUTER OUTPUT SHEET . BIBLIOGMPHY O O O O O O O O O O O O O 0 iv 103 Table . 2.1 3.1 4.1 4.2 4.3 4.4 4.5 CO4 C.S LIST OF TABLES Summary of Related Literature . . . . . . . . Distribution of Sample by Department . . . . Summary of Selection Criteria .,. . . . . . . Mean for Each Continuous Predictor Variable . Group Centroids . . . . . . . . . . . . . . . F-to-Enter Summary Table . . . . . . . . . . Discriminating Power of.Nine Predictor variables I O O O O O O I O O O O O I O O 0 Summary of Means for Graduates and Drop-Outs Standard Deviations for Predictor Variables . Standard Discriminant Function Coefficients . Unstandardized Discriminant Function Coefficients . . . . . . . . . . . . . . . Summary of Entry Criteria and F-to-Remove . . Summary of Variables in the Function . . . . Page 35 42 56 66 7o 72 73 86 98 99 100 101 102 CHAPTER I STATEMENT OF THE PROBLEM Background and Need for the Study The selection of a potential doctoral student is usually based on all the estimates of student quality that are available. The findings of a study by Heiss in 1970 implied that there is a body of research on estimates of student quality, such as early culture and motivation, personality, interest, grade-point average, class rank, and scores on the Graduate Record Examination or the Miller Analogies Test. These predictors and others often provide some degree of accuracy for forecasting the success of students.1 Most industrial education doctoral programs specify that to obtain a degree the candidate should: (1) satisfy a residence requirement, (2) master the requirements of a series of courses and seminars, (3) successfully complete a written and an oral qualification examination, (4) secure the approval of a faculty committee on the choice of a research topic and the method to be used in the study, and (5) write and defend the results of the research in a form 1Ann M. Heiss, Challenges to Graduate_Schools (San Francisco: Jossey-Bass, Inc., Publishers, 1970): pp. 92-93. '/ _/ 1 approved by the faculty committee. Each of these require- ments should contribute to the student's scholarly develop- ment; therefore, any independent or dependent variable used as part of a selection procedure should reflect these requirements.2 The modern industrial educator has many job require- ments or expectations. The concern of colleagues, stu- dents, administrators, and even members of the community tends to have a great influence upon the impact of these expectations, some of which include: 1. Teaching undergraduate and graduate industrial education courses 2. Advising students in the industrial education department 3. Selecting students who will succeed in industrial education 4. Serving on local, state, and national committees 5. Conducting or encouraging research to improve the quality of industrial education 6. Attending local, state, and national conferences and conventions on industrial education 7. Acting as a consultant for other educational institutions and in industry 8. Performing administrative functions 21bid., p. 109. Chaplin and others, in their national status study of 1974, provided an insight into how industrial education professors spend their time: Of the responding institutions, 81 percent indi- cated that staff members spent 70 percent or more of their time teaching. Only five institutions indi- cated that less than 39 percent of faculty time went to teaching. Nearly 75 percent of the colleges reported that less than 5 percent of faculty time was alloted to research. . . . Approximately 75 percent of the colleges reported that less than 10 percent of staff time went to administration, 10 percent to departmental and university meetings, 5 percent to national profes- sional organizations, and 5 percent to assisting master's and doctoral graduates. Of possible concern is the fact that less than 10 percent of staff time was given to student counseling. The national status study revealed the proportion of time professors allocate to many of their job require- ments, such as selecting, teaching, and advising doctoral students. These expectations demand much of the profes- sor's time and effort; therefore, the results of the present study should enable the professor to use his time most profitably by: l. Becoming more accurate in selecting doctoral stu- dents while minimizing the number of predictor vari- ables used in the selection process. 3Jack Chaplin, Ronald Todd, and John Gradwell, "Industrial Arts Teacher Education: Myths and Realities,” ManlSociety/Technology; A Journal of Industrial Arts Edu- cation 34 (Winter 1974): 93-94. 2. Becoming more adept at advising students concerning their probability of graduating from or dropping out of the industrial education doctoral proqram. 3. Focusing on the possible relationship between inde- pendent and dependent variables that are used in selecting industrial education doctoral students. Investigating the current selection criteria and procedures and testing the effectiveness of predictor vari- ables used in the selection process is one way of attempt- ing to develOp a model for selecting industrial education doctoral students within Big Ten universities. The results of this study could add precision to the selection of suc- cessful industrial education doctoral students. In addition, a review of the literature revealed that few recent studies have dealt with the selection of industrial education doctoral students; therefore, the paucity of research generates interest in conducting this study. Finally, interviews conducted with several indus- trial education professors revealed that they are inter- ested in knowing more about common elements of the selec- tion criteria and procedures being used by other Big Ten departments. They are also concerned with gaining insight about students who completeor drOp out of industrial edu- cation doctoral programs. The results of this study may provide information about the relationship among selection criteria and procedures. Purposes and Objectives of the Study This study was designed to carry out the following purposes and objectives: 1. To investigate the selection criteria and pro- cedures being used by certain industrial education depart- ments. 2. To test the effectiveness of certain predictor variables (criteria) that industrial educators in selected Big Ten universities can use in selecting industrial edu- cation doctoral students. The predictor variables are: (1) overall undergraduate grade-point average, (2) under- graduate grade-point average for the last two years, (3) overall master's grade-point average, (4) master's grade-point average of courses taken in industrial educa- tion, (5) years of relevant professional education work experience, (6) Graduate Record Examination Verbal score, (7) Graduate Record Examination Quantitative score, (8) the Miller Analogies Test score, (9) the number of pub- lications listed on the application for the doctoral pro- gram, (10) age at the time of application to the doctoral program, (11) number of years taken to complete the master's degree program, (12) number of dependents at the time of application to the doctoral program, (13) marital status at the time of application to the doctoral program, and (14) last employment before admittance to the doctoral program. 3. To contribute to the development of a selection model that allows industrial educators in selected Big Ten universities to predict the success of potential industrial education doctoral students. Hypotheses The hypotheses relate to a set of predictor vari- ables that can discriminate between industrial education students who will graduate from.or drOp out of the doctoral program. A null hypothesis is applicable, which implies an alternate hypothesis. Null Hypothesis: There is not a set of independent variables (predictors) that can discriminate between industrial education students who will graduate from or drop out of the doctoral program (criterion). Alternative Hypothesis: There is a set of independent variables (predictors) that can discriminate between industrial education students who will graduate from or drop out of the doctoral program (criterion). Definition of Terms Criterion variable--The criterion variable refers to a dependent variable that assumes scores from indepen- dent variables are "criterion measures" for the criterion variable. Success is the criterion variable for this study. For classification purposes, success relates to two categories: (1) students who graduate from the industrial education doctoral program and (2) students who drop out of the program. Predictor variables--Predictor variables refer to independent variables that provide information about the criterion variable. Former doctoral students--Former doctoral students are those who graduated or dropped out and had an indus- trial arts and/or trade and industrial education major during enrollment in the doctoral program. DroErouts--Drop-outs are former students who sur- passed the number of years departments allow to complete degree requirements and/or former students defined by departments as drop-outs for other reasons. Industrial education--Industrial education is a generic term that refers to industrial arts, vocational- industrial, and technical education. Department representatives--Department representa- tive refers to a chairman or coordinator of an industrial education program within the Big Ten universities selected for the present study. Delimitations of the Study This study has two major limitations, which relate to: (1) variables to be used in the selection of doctoral students and (2) the population for the study. The vari- ables include predictor and criterion variables, which are associated with three sources: (1) variables that are currently used by Big Ten departments in the selection process, (2) predictor and criterion variables that have been reported in related literature studies, and (3) vari- ables that have been suggested by professors in selected Big Ten departments. The pOpulation for the study included subjects from seven departments within Big Ten universities that had industrial education doctoral programs. The pOpulation included one more department than university because one university had two separate industrial education depart- ments, and each of the other universities had one such department. Originally, it was anticipated that data would be collected on ten doctoral graduates and five drop-outs from each of the industrial education programs. However, using only former students as subjects restricted the sample range; nothing could be done about this since the departments did not have information on applicants who had not been admitted to the doctoral program. Theory and Design for the Study Research is a systematic process that provides solutions to problems. These problems usually take the form of questions or hypotheses. Generally, the three types of research that educators often use are descriptive, historical, and experimental.4 4Ralph H. Jones, Methods and Techniques of Educa- tional Research (Danville: The Interstate Printers and Publishers, Inc., 1973), p. 7. Since the present study has most of the character- istics of descriptive research, it is logical to classify 5 Such research can determine present it as descriptive. conditions, and can provide for the description, analysis, and investigation of specific problems. Descriptive research can determine trends, so that predictions can be made about the future. In this study the personal interview and the ques- tionnaire were used to collect data. The personal inter- view was used to elicit information from the industrial education department representatives of Big Ten depart- ments, whereas the questionnaire was used to record data pertaining to subjects in the study. ‘ A The statistical method for this study was the SPSS RAO stepwise discriminant analysis. Such analysis was used to test the effectiveness of predictor variables that can be used to classify students as graduates or drop-outs of the industrial education doctoral program. The RAO stepwise discriminant analysis classification equation is apprOpriate for categorizing industrial education doctoral applicants as graduates or drOp-outs. The same classifi- cation equation is suited for a validation procedure. The study had three purposes, each of which required a different treatment. The purposes and their treatments are: 51bid., p. 6. 10 1. To investigate the selection criteria and pro- cedures being used by certain industrial education depart- ments. This purpose was accomplished by conducting per- sonal interviews with industrial education department representatives from seven Big Ten universities. 2. To test the effectiveness of certain predictor variables (criteria) that industrial educators in selected Big Ten universities can use in selecting industrial edu- cation doctoral students. The questionnaire was the instru- ment used to collect data pertaining to variables. However, the SPSS RAO stepwise discriminant analysis method is apprOpriate for testing the effectiveness of the variables. 3. To contribute to the development of a selection model that allows industrial educators in selected Big Ten univer- sities to predict the success of potential industrial education doctoral students. A visual inspection of the F-to-enter from the computer output reveals the most effective predictor variables for classifying doctoral students. Overview of the Study The remainder of this research includes a Review of Literature, Research Design and Method, Evaluation of Data, and Summary and Conclusions. The review of litera- ture encompasses selection studies in industrial education and general selection studies. Described in the Research Design and Method chapter are the sources of data, the 11 variables for the study, the data-collection instruments and procedures, the hypotheses, and the data-analysis procedures. The fourth chapter, Evaluation of Data, includes a report of the results of interviews with seven depart— ment representatives, a description of the sample, an interpretation of the hypotheses test, and a discussion of a selection model developed for the study. The final chapter, Summary and Conclusions, relates a summary of the study, conclusions, discussion, and recommendations for further research. CHAPTER II REVIEW OF LITERATURE Introduction Numerous studies have been focused upon the Selection of students; however, many of them do not per- tain to the industrial education doctoral student. This review of literature was undertaken to investigate the strengths and the weaknesses of existing selection studies. The studies represent the major areas of research that pertain to the selection of students during the past two decades. The relevant studies fall into two main cate- gories: (1) selection studies in industrial education and (2) general selection studies. The studies in indus- trial education include those that will help one compre- hend principles employed in selecting industrial education students. The general studies include those that will help with the overall organization of the study and with the choice of predictor and criterion variables. Selection Studies in Industrial Education In 1951 Belman and Evans studied the selection of undergraduate students transferring from other schools 12 13 within Purdue University into the trade and industrial education curriculum.1 Data were obtained from orientation tests, student records, and other test scores of 107 students. Six tests were employed during the selection of students for the trade and industrial education curriculum: the Purdue Adaptability Test, an English orientation test, the How to Supervise Test (Form M), grades earned before transfer, grades earned in the course entitled Introduction to Trade and Industrial Education, and a question Belman and Evans considered important. The response to the question was considered relevant if the student indicated the option of transferring to the trade and industrial education cur- riculum was definite.2 The criterion variable was grade-point index after transfer to the trade and industrial education cur- riculum. This variable was correlated with predictors by using the multiple correlation coefficient method. The investigators concluded that the scholastic success of students entering an industrial education cur- riculum could be predicted if certain data were available. This conclusion was based upon correlations (P < .01) for several predictors with the criterion. The correlations __l 1H. S. Belman and R. N. Evans, "Selection of Stu- dents for a Trade and Industrial Education Curriculum," Journal of Educational Psychology 42 (Winter 1951): 56. 2Ibid. 14 were: (1) Purdue Adaptability .4120, (2) Purdue English .2995, (3) How to Supervise .2893, (4) Purdue mathematics .3441, and (5) index before transfer (GPA) .4982. These correlations probably would have been lower if students with low scores had not been eliminated during the selec- tion for other schools within the university. Thus, the Spread of the sample range had been restricted before selection for the trade and industrial education curric- ulum.3 Benson completed a study at Wayne State University in 1958 that was designed to (1) identify objective and subjective factors used in the selection of doctoral candi- dates specializing in industrial education; (2) establish, insofar as possible, the importance of measurable factors in predicting success for advanced graduate work; and (3) identify methods, techniques, or procedures that had been used to help students successfully complete degree requirements.4 Data were collected from student records, research literature, and personal interviews. Information from each student's record and the research literature available 3Ibid., p. 58. 4William A. Benson, "Measurable and Observable Factors in the Selection Retention of Doctoral Candidates With Special Implication for Industrial Education" (Ann Arbor, Michigan: Xerox University Microfilm, #ED 11611, 1958). p. 8. 15 at Wayne State University was related to objective and subjective factors previously used in the selection of industrial education doctoral students. The personal interview was used to collect data pertaining to methods and procedures used by nine department chairmen to aid students in the completion of degree requirements. The criterion variable for the study was success. Data for the criterion came from students who had gradu- ated or withdrawn from the degree program. Some pre- dictors utilized in the study were undergraduate grade- point average, graduate grade-point average, Miller Analo- gies Test, and the Graduate Record Examination Verbal and Quantitative scores. Discriminant analysis was the statistical method applied. This analysis revealed no significant differ- ence between the undergraduate grade-point averages of the successful and unsuccessful groups. However, scores achieved on the Miller Analogies Test and the Graduate Record Examination by the successful group were signifi- cantly higher than scores achieved by the unsuccessful group. The latter two tests provided a basis for dis- tinguishing between potential industrial education students who would complete degree requirements and those who would become drop-outs. From these results it can be inferred ¥ 51bid., p. 9. 16 that systematic and continuous evaluation of the process of selecting doctoral students is needed. A study conducted by Johnson in 1949 was designed to (1) determine current admission practices in industrial arts teacher education programs throughout the nation, (2) discover background interests of students, (3) develop profiles of students and other graphic analysis of the findings of various factors that would be helpful in deter- mining the probable potential of students as prospective industrial arts teachers, and (4) offer suggestions and recommendations for improving a selection and guidance program at the State Teachers College in Cheyney, Pennsyl- vania. The researcher intended that the results of the study he made available to other schools with similar programs.6 Johnson used a questionnaire as a data-collection instrument. He collected information on a sample of forty freshman and SOphomore industrial arts students who were enrolled from 1947 through 1949. Mean and frequency dis- tributions were the statistical methods employed to deter- mine the effectiveness of tests used in the selection 7 - process. ‘ 6Rufus C. Johnson, "A Study of Selection and Guidance Procedures for Students in the Program of Indus- trial Arts Teacher Education at the State Teachers College" (Ann Arbor, Michigan: Xerox University Microfilms, #ED 11616, 1949). P. 9. 7Ibid., pp. 4-5. 17 Actually, the findings of Johnson's study added little to the selection process. However, it was suggested that entrance tests should be administered to all students; The results of these tests could be used in planning pro- grams to meet individual needs.8 A study reported by Jarvis in 1953 was concerned with student survival factors at Stout Institute. The intent of the study was (1) to determine the relationship that might exist between entrance test, high school rank, selected high school subjects offered at college entrance, and scholarship in freshman-level technical courses; and (2) to determine whether these entrance tests, high school rank, and selected high school subjects helped to identify students who would complete the requirements for a 8.8. degree in four years.9 Data were collected on 393 entering freshman indus- trial arts or vocational-industrial education students who had not taken college work before the entrance examination. The information was obtained from college entrance tests and high school records of students who had enrolled during the school years 1947-48 through 1950-51.10 8Ibid., pp. 142-41. 9John Jarvis, "Student Survival Factors in the Stout Institute: A Statistical Study of High School Records, Entrance Test Scores, College Course Grades, and Other Mea- sures With Relation to Survival in the Graduation by a College Teacher Training Type-Male Student" (Ann Arbor, Michigan: Xerox University Microfilms, #ED 11615, 1953). p. . 10Ibid., pp. 7-9. 18 The multiple regression method was employed with predictor and criterion variables; test measures were used as predictors of the criterion. Whether a student gradu- ated or failed to graduate was the criterion variable. This variable was called success, for statistical applica- tion purposes. The study revealed that tests administered to enter- ing students at Stout Institute were of little value in predicting the success of students enrolled in technical courses in industrial arts or vocational-industrial edu- cation. Students who later graduated from Stout Institute could not be identified by their scores on: (1) the American Council on Education Psychological Examination for College Freshmen or (2) the Cooperative English and Myers-Ruch School Progress Test, Form AM. Actually, there was a sig- nificant difference between the graduate and the nongradu- ate in terms of high school rank. This was the only pre- dictor variable that provided relevant information for the selection process. These findings suggested that it would probably have been better to include predictor variables that were not based on formal testing.11 In 1963 Torres completed a study that was concerned with determining the relationship between intellectual variables and first- and third-semester achievement of industrial arts students in the industrial, the general, 11Ibid., p. 97. 19 and the total academic program at Long Beach State Col- lege. The study also sought to determine the extent to which the intellectual variables could be used to predict achievement.12 Two hundred male junior college students who had enrolled at Long Beach State College and majored in the industrial arts program between 1957 and 1960 were included in the study. These students had to have completed the college entrance tests and three or more consecutive semesters of full-time study at the college.13 Data pertaining to predictor and criterion vari— ables were collected from the testing office, the office of the registrar, the records and admissions office, and from industrial arts department records at the college. The predictor variables were the results of the Owen- Bennet Test of Mechanical Comprehension, Form CC; the Minnesota Paper Form Board Test, Series MA; the Cooperative English Ability Test, Form AA; and the junior year college grade-point average. The criterion variables were grade- point average for the first and third semesters in indus- trial arts, the first- and third-semester grade-point 12Leonard Torres, "A Study of the Relationship Between Selected Variables and the Achievement of Indus- trial Arts Students at Long Beach State College" (Ann Arbor, Michigan: Xerox University Microfilms, #ED 11620, 1963), p. 6. 13Ibid., p. 33. 20 average for all general courses, and the grade-point average for the total academic program at Long Beach State College.14 The regression equation and the multiple correla- tion were the statistical methods applied; significant relationships were found between predictor and criterion variables. A statistical method was not used to select predictor variables for the study; therefore, this could have had some influence on the relationship between vari- ables. The review of selection studies in industrial edu- cation provided some clues about the weak and strong points to consider when studying the selection of doctoral stu- dents. Additional insight was obtained by reviewing other selected studies; these research efforts are discussed in the following section. General Selection Studies Chase, Ludlow, and Pugh completed a study at Indiana University in 1964, which was designed to describe characteristics of master's degree students in the school of education and to investigate the utility of admissions tests and personal history data as predictors of success.15 ¥ l4Ibid., pp. 25-27. 15Clinton I. Chase, Glenn H. Ludlow, and Richard C. Pugh, Predicting Success for Master's Degree Students in Education (Bloomington: Bureau of Educational Studies and Testing, 1964), p. l. 21 A questionnaire was used to collect data pertaining to predictor and criterion variables from about one thou- sand subjects. Predictor variables for the study were scores on the Cooperative English Test, the Concept Mas- tery Test, and the numerical ability portion of the Differ- ential Aptitude Test. Data on sex, experience, race, and previous institution attended were also used as predictors of the grade-point average. The multiple correlation coefficient was the statistical method employed in the study.16 Data were divided into four subgroups for statis- tical application. The groups were men, women, Native Negro, and Native White. The Concept Mastery total for men was correlated .30 with grade-point average. However, the Cooperative English Reading and Differential Aptitude Tests were correlated .64 for women. Both of these corre- lations were large enough to improve the prediction of grade-point average over selection by chance.17 Data pertaining to the Native Negro group were analyzed using the vocabulary score from the reading comprehension test. The correlation with grade-point average was .52; this test alone seemed to predict grade- point average as well as a variety of other tests.18 Madaus and Walsh conducted a study at a New England university during 1965. The research concerned the k 16Ibid., p. 4. 17Ibid., p. 20. 18Ibid. 22 predictive efficiency of the Graduate Record Examinations, and used Graduate Record Examination scores of beginning graduate students who had been involved in an educational testing program from 1961 through 1963. Data pertaining to predictor and criterion variables were collected for 569 students, by department. The criterion variable for the study was grade-point average at the end of a semester of graduate study. The multiple regression analysis was used to select predictor variables in the order of their contribution to the selection process.19 Correlations on the subjects' Graduate Record Examination Verbal and Quantitative scores were .19 and .18, respectively. The correlation of Graduate Record Examination scores with grade-point average ranged from 0 to .69 for all departments (P < .01). The low correla- tions could have been a result of the use of a single- category predictor variable. The use of a statistical method to select predictor variables to be applied in the study was a step toward determining the relevancy of pre- dictor variables that were being used in selection studies. Chase completed a study in 1960 that utilized the records of undergraduate students from Hunter College ‘_1 19George F. Madaus and John J. Walsh, "Departmental Differentials in the Predictive Validity of the Graduate Record Examination Aptitude Test," Educational and Psycho- lggical Measurement 25 (Winter 1965): 1106. 2°Ibid., p. 1107. 23 who later earned a doctoral degree, and compared their validity with a random sample of records of the whole student body.21 During the fall of 1957 the Office of Scientific Personnel of the National Research Council forwarded Hunter College a list of its graduates who had been awarded the doctorate degree by other institutions between 1936 and 1956. Other data for 294 of these students were collected from their records at Hunter College. The result was that the average of the undergraduate records of the doctoral group was higher than the average of the records of a random sample of nondoctoral graduates, with a differ- ence of .48 (P < .001).22 Another objective of the research was to determine whether the undergraduate records of students who later received a doctoral degree revealed significant differ- ences in terms of the particular disciplines in which the degrees were received. The records included grades for courses in biological science, arts and humanities, psy- choloqy, social sciences, and education. The only sig- nificant difference was found in education.23 The average of cumulative indices of students who later received a doctorate degree in education was different ‘ 21Edith B. Chase, "A Study of Undergraduate Records of Graduates From Hunter College Who Later Earned Doctor- ates," Journal of Experimental Education 29 (Fall 1960): 59. 22Ibid., p. 54. 23Ibid., p. 59. 24 from that of the noneducation group, with a significant correlation of .47 (P < .001). The cumulative and first- term indices (grades) were found to be useful as predictors of achievement in graduate school. The correlation between first-term grades and cumulative average was .86 for the doctorate degree recipients and .71 for the random sample. A significant correlation (.66) was reported between high school and college averages of the future doctoral recipi- ents.24 Finally, the average of the grades earned in the future-doctorate-related major area was higher than the general average. Thus, a review of the academic perform- ance of a potential doctoral student during undergraduate studies might reveal the area in which he is most likely to succeed as a doctoral student. A study completed by Kooker in 1971 was designed to predict the grade-point average of potential doctoral students enrolled in a required statistics course in the school of education. Data were collected on sixty-nine students identified by the counseling center at North Texas University, even though not all of them had been screened for graduate school. The predictor variables applied in 24Ihid. 25 the study were the Watson-Glaser (W-G) and the Miller Analogies Test (MAT).25 The Watson Glaser Thinking Appraisal, Form AM was administered to students while they were taking the sta- tistics course. This appraisal was designed to measure the students' capacity to comprehend statistics. The criterion measure represented the performance of students on three tests administered during a semester; this score was obtained by totaling the three scores and dividing by the highest total in the class and converting the quotient to a percentage.26 The multiple correlation coefficient and the Pearson correlation coefficient were the statistical methods employed with predictor and criterion variables. The correlation between the MAT and the test scores was not significant (.21). However, the correlation between the W-G and the test scores was .37, which was significant (P < .01). The multiple correlation coefficient was used to correlate the MAT and the W-G with the criterion; it did not reveal a significant increase over the correlation between the W-G and the test scores.27 25Earl W. Kooker, "The Relationship Between Perform- ance in a Graduate Course in Statistics and the Miller IAnalogies Test and the Watson-Glaser Thinking Appraisal," figurnal of Psychology 77 (Spring 1971): 166. 26Ihid. 271bid., p. 167. 26 Actually, the MAT did not reveal a significant multiple correlation coefficient (.38) when used with the W-G. The Watson-Glaser Thinking Appraisal accounted for 14 percent of the variance in test scores. Thus, includ- ing in the study students whose applications had not been screened for admittance purposes allowed for less restric- tion of the sample range than in many previous selection studies. Many studies have revealed a low relationship between performance on the MAT and performance in graduate school when grade-point average was the criterion. This might imply that the MAT should always be employed with other predictor variables when selecting students. During 1969 Mehrabian reported a study that con- sidered the relationship among a series of predictor vari- ables that could be used with the selection of students in graduate psychology programs. The study further char- acterized ability factors based on the criteria employed in selecting candidates for graduate school and the rela— tionships between various selection criteria and graduate performance.28 Data were collected from the admissions files of 260 potential UCLA graduate psychology students. Using a regression analysis, an admissions committee used the data to assess the validity of thirteen predictor variables. —; 28Albert Mehrabian, "Undergraduate Ability Factors in Relationship to Graduate Performance," Educational and Psychological-Measurement 29 (Summer 1969): 409. 27 Some of the variables used for projecting the academic success of graduate students were Graduate Record Examina- tion scores, grade—point averages, Miller Analogies Test, number of mathematics and logics courses taken, rating of the department the student had attended as an undergradu- ate, the amount of research experience as an undergraduate student, sex, and grade-point improvement in the last two years.29 Six of the thirteen factors accounted for 75 per- cent of the total variance; the order of their importance was: Graduate Record Examination and Miller Analogies Test percentiles; research orientation; grade-point aver- ages (overall, junior, and senior years); sex; grade-point improvement in the last two years; and mathematical training.3O Finally, the collection of data on students before admittance to the graduate psychology program was an asset of the study because it helped to eliminate some of the restriction on the sample range. Furthermore, the deter- mination of predictor variables in the order of their contribution to the selection process could have given an indication of predictors that could have been eliminated from the selection process because they had little predic- tive value. 291hid., p. 411. 3°Ibid., p. 414. 28 A study reported by Merenda and Reilly in 1971 investigated the effectiveness of a set of predictor vari- ables in determining the success of graduate students. Data were collected on seventy-five students admitted to graduate study in psychology at the University of Rhode Island between 1964 and 1968.31 Predictors for the study were total undergraduate grade-point average in psychology, overall undergraduate grade-point average (based on a 4.00 grading system), Graduate Record Examination Verbal score, Graduate Record Examination Quantitative score, Graduate Record Examination Advanced score, and the rating of the college at which the baccalaureate degree had been earned. The last was a sub- jective rating by the instructor. The criterion applied was success, as depicted by the following three categories: (1) students who had earned degrees or were working toward eaning a degree without delay, (2) students who had earned degrees or who had been delayed, and (3) students who had failed to earn degrees because of scholastic failure or for other reasons. The first category accounted for forty of the subjects, whereas the second and third categories accounted for nineteen and eighteen subjects, respectively. ¥ 31Peter F. Merenda and Raymond Reilly, "Validity of Selection Criteria in Determining Success of Graduate Students in Psychology," PsychologyAReport 28 (Winter 1971): 265. 32Ibid., p. 261. 29 Discriminant analysis was the statistical method employed with the data. Among the three criterion cate- gories there was a trend for the scores on the six pre- dictor variables to be higher in the first category and lower in the third. The best predictor variables were total undergraduate grade-point average, Advanced Graduate Record Examination, and grades earned in undergraduate psychology courses. The Graduate Record Examination Verbal score assumed almost one-half of the weight for undergradu- ate grade-point average; undergraduate college rating assumed less weight but was significant. On the other hand, the GRE Quantitative Test assumed a slighly negative weight.33 One advantage of this study was that it included a discriminant analysis method that could determine the probability of an applicant's membership in a criterion group. The results of the calculation of a discriminant equation could be used to predict the category into which a subject would fall. The subject was predicted to belong to the category whose discriminant equation revealed the highest probability. This method has not often been used ‘with the selection process; therefore, the initiation of inew studies using the discriminant analysis should provide new insight for the selection process. 33Ibid., p. 263. 30 Miller's 1973 study was concerned with the impor- tance of admissions criteria to future performance in gradu- ate school. Data from students' academic records were col- lected for five predictor variables that were to be used with the admission of behavioral science students. Pre- dictor variables used in the study included GRE Verbal score, quality rating of undergraduate institution, grade- point average in sociology coursework, Graduate Record Examination Quantitative score, and undergraduate grade- point average.34 Originally, 118 students were considered for the study, but the triadic scale eliminated thirteen subjects because of insufficient data. Some predictor variables' values could not be obtained from the students' academic records; missing values were compensated for by using the overall mean for the population.35 The actual admission decision was determined by the use of a multivariate analysis of mean differences to determine significant group differences on the performance criteria. This instrument did indicate a significant pre- dictive relationship for the five predictor variables. The multiple R for the five predictors was .56, with a 34John J. Miller, "The Graduate Admission Process in Two Behavioral Science Departments at Michigan State University" (Ph.D. dissertation, Michigan State University, 1973): PP. 37-55. 351hid., pp. 53-61. 31 coefficient of determination equal to .32. The ANOVA F-test value was .0005. Finally, the study revealed a way to account for missing variable values by using the overall mean, which is discussed in Chapter III. This approach appeared to be an equitable way to account for missing variable values. A study designed to reveal the usefulness of under- graduate grades and the Miller Analogies Test (MAT) in predicting several measures of "success" in the graduate psychology program at the University of Michigan was under- taken by Platz, McClintock, and Katz in 1959. Data were collected from the records of 124 graduate students from 1950 through 1955. The Miller Analogies Test and under- graduate performance were used to select the population. The major predictor variables for the study were overall undergraduate grade-point average; undergraduate grade-point average in science, mathematics, and psychol- ogy courses; the Miller Analogies Test score; and an objec- tive comprehensive examination. Three measures were also used for defining "success"; they were grade-point average in graduate courses, marks on the preliminary doctoral examination, and a faculty rating.37 36Arthur Platz, Charles McClintock, and Daniel Katz, "Undergraduate Grades and the Miller Analogies Test as Predictors of Graduate Success," American Psychologist 14 (Summer 1959): 285. 37Ibid., p. 286. 32 The multiple correlation coefficient and the Pearson product-moment correlation were used for computational purposes. Some of the relationships were as follows: (1) The correlation between grades in gradaute courses and the combined predictors of undergraduate science grades and scores on the comprehensive achievement examination in psychology was .60; (2) The correlations between prelimi- nary and graduate grades and faculty ratings of potential scientific contribution were .63 and .60, respectively; (3) The combination of science grades and scores on the objective comprehensive examination taken when entering the university revealed a correlation of .60; and (4) The correlation of the Miller Analogies Test, the undergradu- ate science grades, and graduate performance was .52.38 This study related how grades earned in subject areas could be used in the selection process. Several categories of predictor variables were applied; this seemed to be more logical than using just one predictor category. One category might indicate a few things about a student's potential, but several could reflect many aspects of the student's potential. In a selection study they conducted in 1969, Roscoe and Houston used a combination of Graduate Record Examination scores and four criterion variables. The study was concerned with determining the relevancy of the Graduate —‘ 38Ibid., p. 288. 33 Record Examination as a selection standard for doctoral students at Colorado State College.39 The sample was restricted to doctoral students who had graduated (231) or who had been dismissed (21) from the program during a recent three-year period. The dismissed students had to have completed a minimum of thirty quarter hours of doctoral work.40 The predictor variables applied in the selection study were the Graduate Record Examination Verbal and Quan- titative scores. Thus, the criteria represented grade- point average in doctoral studies, graduation versus dis- missal from the program, normative judgment analysis, and the ipsative judgment analysis. Data were collected from the students' records in an attempt to develop new criteria for selection purposes.41 The multiple correlation coefficient was the sta- tistical method employed; in each case the predictor vari- ables were significantly related to the criterion variables. Actually, better results probably could have been realized if more dismissed students had been included in the sample. Twenty-one dismissed students were too few to include with 231 who had graduated. A ratio of one dismissed student 39John T. Roscoe and Samuel R. Houston, "The Pre- dictive Validity of GRE Scores for a Doctoral Program in Education," Educational and Psychological Measurement 29 (Summer 1969): 508. 4olhid., p. 507. 411bid., p. 508. 34 for every three who had graduated would have been a more logical sample. Summary In summary, many aspects of past selection studies had to be considered in undertaking a study that would assist in the selection of potential industrial education doctoral students. Table 2.1 displays a summary of related literature, indicating for each study the statistics, data- collection method, sample size, number of predictor vari- ables, number of criterion variables, and the year in which the study was completed. The survey of related literature revealed some common elements of the selection studies; the present study was based on some of those mutual characteristics. The dominant statistical method was the multiple correla- tion coefficient, which was used in eleven of the fourteen studies discussed in this chapter. Multiple correlation was not the most apprOpriate method for the present study, though, since there was a desire to test the ability of a set of predictor variables to discriminate between gradu- ate and drOp-out students. Two studies in the review of literature used discriminant analysis, and two used the Pearson correlation coefficient as statistical methods. Discriminant analysis was ideal for testing the effective- ness of predictor variables in a selection process; there- fore, it was chosen as the statistical method to be used 35 whoa no oomnuuuhcmz N z 3ow>uoucH N H cesHoo oHnmoHHmmd N x 03» no one N m ouwmccowumoso N o mammamcm ucmcfiswuomwo D an ousumuouwa mo 3ow>om N Am ouooou ucoosum N mm cowumaouuoo acmumom N m umos N a com: N z newumaouuoo mamwuasz N m: "hon coumnom $3 3050 a «no x mm... x x a moonom .Hm um mmooos mmma m x vNH x x x NHMHm mnma oocmeu0wuom x maa x x Madam: sedans Ahma mmmoosm x E. x a mocmuwz moma mmooosm x mom x x :mwnmunmz Huma umo mama ¢m0 x ov x x concnoo $3 330.5 x .uooo m x x x x x condom mcm>m Hmma dmo x Sea x x x w cgamm x m gm H o 9 mm 2 40 m m: oouoameou onwm spasm moon 6303.8 mason nofisd mnouoflooum noduooaaoo puma mowumwumum .ousumuouwa unawaou mo aumsfidmII.H.~ manna 36 in the present study. The Pearson correlation coeffi- cient was not adaptable for testing the hypothesis because it only allowed for testing the relationship between one predictor and the criterion variable. The mean was employed with one study, and it showed potential for the present research. In the studies reviewed, data were collected by using test results, student records, personal interviews, questionnaires, and a review of literature. Student records tended to be dominant and most promising, whereas tests were second in frequency of use. The personal interview and the questionnaire were often employed as data-collection instruments. Therefore, personal inter- views, student records, and questionnaires were chosen as the means of obtaining data in the present study because they were economical and could provide the neces- sary data. The review of literature displayed varied sample sizes, ranging from forty to over a thousand subjects. These varied sample sizes tended to be appropriate for each study. Predictor and criterion variables also varied in number. Most studies reported in this chapter used many predictor variables; however, a few did use only one or two variables. Actually, predictor variables that were often successfully used with selection studies were: 37 (1) overall undergraduate grade-point average, (2) under- graduate grade-point average for the last two years of study, (3) undergraduate and master's grade-point averages in the major area, (4) first-term grade-point average (undergraduate), (5) Miller Analogies Test score, (6) Miller Analogies Test score percentiles, (7) Graduate Record Examination Verbal score, (8) Graduate Record Exami- nation Quantitative score, (9) high school rank, (10) rele- vant work experience, (11) number of mathematics and logics courses taken, and (12) age of the student. Just one criterion variable was employed in most of the related studies; the dominant criterion was grade-point average. Grade-point average was not a suitable criterion for this study since there was no interest in knowing how accurately quality of scholarship could be predicted. Success was the second dominant criterion employed, and was chosen as the criterion for this study because it allowed a testing of predictors used in classifying industrial education doctoral students as potential graduates or drop-outs. The rationale for variable selection is discussed in Chapter III. Related studies dated from 1949 through 1973. In these research efforts were found many consistent patterns Nnorthy of consideration for future selection studies. :HOwever, a restriction of the sample range was one area of tweakness in most selection studies; this occurred because 38 most subjects were students who had been selected for or who had completed programs. Therefore, most of the studies did not consider students who had applied and were not admitted or those who had been dismissed from doctoral programs. Finally, the strong and the weak areas in these studies helped determine the research design and method for the present study; these subjects are develOped more fully in Chapter III. CHAPTER III RESEARCH DESIGN AND METHOD Introduction The Research Design and Method chapter includes discussions on the sources of data, variables, data col- lection, hypotheses, and data-analysis procedures. The section on sources of data comprises an identification of the population and the sample. A discussion of variables for the study includes comments on predictor and criterion variables. Covered in the data-collection section are the instruments and the procedures used. The hypotheses relate to a set of predictor variables that discriminates between industrial education doctoral students who will graduate from or drop out of the doctoral program. A section on data-analysis procedures includes the sequences for: (l) preparing data for statistical analysis, (2) testing hypotheses, and (3) visually inspecting independent vari- ables that are used in predicting the criterion variable. A summary highlights the main components of the chapter. Finally, this chapter displays a plan for accomp- lishing the following three purposes of the study: 39 40 1. To investigate the selection criteria and procedures being used by certain industrial education departments. 2. To test the effectiveness of certain predictor variables (criteria) that industrial educators in selected Big Ten universities can use in selecting industrial education doctoral students. 3. To contribute to the deve10pment of a selection model that allows industrial educators in selected Big Ten universities to predict the suc- cess of potential industrial education doctoral students. Sources of Data The Population The population included former students identified by representatives of seven industrial education depart- ments within selected Big Ten universities. The Big Ten Records Book 1972-73 was used to identify the Big Ten universities.l Subsequently, seven industrial education departments were distinguished by using the Industrial Education Directory_for 1974-75.2 lMichael McClure and Jeff Elliott, Big Ten‘Records Egok 1972-73 (Chicago: Big Ten Service Bureau, 1972), p. 50 2Industrial Education Directory (Cedar Falls, ifibwa: The Wolverton Printing Company, 1975), pp. 19-58. 41 The Sample The sample included former doctoral students iden- tified by the seven industrial education department repre- sentatives. Former doctoral students were defined as those who had had an industrial arts or a trade and indus- trial education major while enrolled in the doctoral pro- gram and who had graduated from or drOpped out of the program. The two categories of former students included: (1) the ten most recent doctoral graduates and (2) the five most recent doctoral drop-outs--former students who had gone beyond the number of years allowed to complete the industrial education doctoral program or defined by departments as drOp-outs for other reasons. The original plan called for a total of seventy doctoral graduates and thirty—five drop-outs. However, not every department had accessible data on the number of subjects needed. Table 3.1 reveals the sample distribu- tion for each department. Seventy-five former students were included in the study, fifty-four of whom were gradu- ates and twenty-one drOp-outs. Finally, the last year of attendance for graduates ranged from 1969 through 1975, whereas the year in which official drOp-out occurred ranged from 1967 through 1975. 42 Table 3.l.--Distribution of sample by department. Department Graduates Drop-Outs Total 1 9 5 14 2 10 5 15 3 10 5 15 4 2 5 2 6 10 2 12 7 10 0 10 Total 54 21 75 Variables for the Study Predictor Variables Predictor variables refer to independent variables that provide information about the criterion variable. The determination of predictor variables to be used with the selection study was a difficult task, since it was the value of predictor variables that provided a basis for pre- dicting the criterion variable. The review of literature revealed two procedures for selecting predictor variables; these procedures were classified as "rational" and "statistical." Rational solutions were determined logically; statistical solutions were mathematically based . This study employed both procedures . ‘ 3Donivan J. Watley, "Factors That Influence the Ehalection of Predictor Variables in Multiple Regression," Skillege and University 39 (Fall 1973): 72. llll Ill D'II .‘ ll ll 1. ‘I‘ ’1')... I'll .1 ill 43 Predictor variables in the following categories were originally considered as a basis for choosing vari- ables for the study: (1) predictor variables that were being used in the selection process by the seven Big Ten departments, (2) predictor variables employed in related literature studies, and (3) a few variables suggested by professors in selected Big Ten departments. The follow- ing predictor variables were selected for the present study: 1. 2. 10. 11. Overall undergraduate grade-point average Undergraduate grade-point average for the last two years Overall master's grade-point average Master's grade-point average of courses taken in industrial education Years of relevant professional education work experience Graduate Record Examination Verbal score Graduate Record Examination Quantitative score The Miller Analogies Test score Number of publications listed on the applica- tion for the doctoral program Age at the time of application to the doctoral program Number of years taken to complete the master's degree program 44 12. Number of dependents at the time of applica- tion to the doctoral program 13. Marital status at the time of application to the doctoral program 14. Last employment before admittance to the doctoral program All fourteen predictor variables had to be assigned at least one code before the statistical method could be employed. However, variables such as last employment before admittance to the doctoral program, marital status at the time of application, Graduate Record Examination score, Miller Analogies Test score, and grades required more than one code, as explained below. The place of last employment before admittance to the doctoral program required the use of the following three additional codes: (1) employed by a post-secondary education system (PSED=1); (2) employed by an elementary or secondary education system (ESED=2); and (3) employed by business and industry, military, or other organization (IBMO=3). Marital status at the time of application to the doctoral program was another variable that required the use of other codes; they were: (1) married at the time of application (YES=1) and (2) not married at the time of application (NO=2). Codes were also associated with raw scores on the Graduate Record Examination and the Miller Analogies Test. 45 The additional codes were needed because departments only required an applicant to report scores on either the Gradu- ate Record Examination or the Miller Analogies Test; therefore, data on subjects were usually available for only one of the two tests. The codes used with subjects who had missing raw scores on the Graduate Record Examination or the Miller Analogies Test were: (1) Graduate Record Examina- tion Verbal score missing (GREV=4), (2) Graduate Record Examination Quantitative score missing (GREQ=4), and (3) Miller Analogies Test score missing (MATS=4). The value of grades may vary from one department to another or from one professor to another, but all grades used in this study were considered equivalent for research purposes. Furthermore, grades had to be converted to a numerical grading system for computational purposes; 4.00 was used as the maximum and 0.00 the minimum for the grading scale. Criterion Variable A criterion variable refers to a dependent variable that is assumed to be predictable from independent vari- ables. Scores obtained from independent variables were considered "criterion measures" for the criterion variable. Some of the related studies in Chapter II employed more than one criterion variable, but the present study used a single criterion--"success"--as represented by two cate- (gories. The categories were: (1) students who had 46 graduated from the industrial education doctoral program and (2) those who had dropped out of the industrial educa- tion doctoral program. This criterion was appropriate since the study was concerned with selecting variables that could discriminate between students who would gradu- ate from or drop out of the doctoral program. A coding system had to be employed with the cri- terion variable before the statistical application could be completed. The coding system associated with the criterion variable (success) was: graduates=l and drop- outs=2. Data Collection The Instrument A personal interview was planned and conducted with department representatives in the seven industrial education departments in selected Big Ten universities. The purpose of the interview was to gain insight into the current selection criteria and procedures employed by those departments. An eighteen-item questionnaire was used to col- lect data on subjects pertaining to predictor and cri- terion variables. The questionnaire provided data rele- vant to two of the research purposes: (1) to test the effectiveness of certain predictor variables (criteria) that industrial educators in selected Big Ten universities 47 can use in selecting industrial education doctoral stu- dents, and (2) to contribute to the development of a selec- tion model that allows industrial educators in selected Big Ten universities to predict the success of industrial educa- tion doctoral students. The Procedures The data-collection procedures varied. A tele- phone call was made to each of the seven department rep- resentatives requesting an appointment for a personal interview. After the appointment had been arranged, a letter containing an additional explanation of the inter- view was mailed to each representative. The letter was an outline for the interview, which sought to gain insight into the current selection criteria and procedures, and to obtain recommendations on new variables that might be applied in the selection process. The interview sessions were tape-recorded for future reference.4 A few months after the interviews, a letter and questionnaires were mailed to each department. The ques- tionnaire was designed to collect data on each subject pertaining to predictor and criterion variables. The department representative either provided data on subjects from department files or channeled the letter and ¥ 4A c0py of the personal interview schedule can be found in Appendix A. 48 questionnaire to another source for data-collection pur- poses. A stamped self-addressed envelope was mailed with the letter and questionnaire to facilitate the return of raw data.5 Questionnaires for each department were color coded to allow for follow-up on instruments that were not returned within three weeks. The coding system also was used to identify the two categories of subjects. Hypotheses The hypotheses were related to a set of predictor variables that can discriminate between industrial education students who will graduate from or drOp out of the doctoral program. A null hypothesis was applicable, which implies an alternative hypothesis. Null Hypothesis: There is not a set of independent vafIables (predictors) that can discriminate between industrial education students who will graduate from or drOp out of the doctoral program (criterion). Alternative Hypothesis: There is a set of independent variables (predictors) that can discriminate between industrial education students who will graduate from or drop out of the doctoral program (criterion). Data-Analysis Procedures The data-analysis procedures included: (1) pref paring the data for statistical analysis and (2) testing the hypotheses. Personal interviews and questionnaires * 5A copy of the questionnaire can be found in Appendix B . 49 were used to collect data pertaining to industrial educa- tion doctoral departments and subjects included in the study. Data collected through personal interviews were collated and screened for predictor and criterion variables that could be used in the study. After the personal interviews had been completed, data on subjects were col- lected by means of questionnaires. This information was related to the list of predictor and criterion variables. Before the statistical method could be employed, these data had to be screened for missing variable values. The mean for a variable was used as a substitute for miss- . 6 ing values. The SPSS RAO stepwise discriminant analysis method was used to test hypotheses related to predictor and cri- terion variables. This method was a subprogram of the Statistical Package for the Social Sciences (SPSS). Nie and others explained the stepwise procedure as follows: The stepwise procedure begins by selecting the single best discriminating variable according to a user determined criterion. . . . A second discrimi— nating variable is selected as the variable best able to improve the value of the discrimination criterion in combination with the first variable. The third and subsequent variables are similarly selected accord— ing to their ability to contribute to further discrimi- nation. 6Norman H. Nie and others, Statistical Package for .Ebe Social Sciences (New York: McGraw-Hill Book Company, 1975), p. 456. 7 Ibid., p. 436. 50 Finally, the result of the statistical method pro- vided a listing of predictor variables in accordance with their ability to predict the criterion variable. The discriminant analysis method also provided a classifica- tion equation that arranged subjects into groups. Nie and others reported that: . . . By classification is meant the process of identifying the likely group membership of a case when the only information known is the8case's values on the discriminating variables. . . . Even though the selection study was not statis- tically validated, the classification equation was con- sidered to be a predictive model. This model related to one purpose of the study: to contribute to the development of a selection model that allows industrial educators in selected Big Ten universities to predict the success of potential industrial education doctoral students. The relevant computer output for this study was: 1. The Wilks lambda was an inverse measure of the discriminant power in the original predictor variables that had not been removed from the discriminant function. Lambda was converted to a chi-square (x2) statistic to test for statistical significance. 2. Centroids were defined as the mean discriminant scores for each group on the function. When there was only