Search

This paper introduces a new decision support tool, data mining, in the context of knowledge management. The most striking features of data mining techniques are clustering and prediction. The clustering aspect of data mining offers comprehensive characteristics analysis of students, while the predicting function estimates the likelihood for a variety of outcomes, such as transferability, persistence, retention, and success in classes. Compared to traditional analytical studies that are often hindsight and aggregate in nature, data mining is forward looking and is oriented to individual students. A real life project presents the work of data mining in predicting the possibility of return to school for every student currently enrolled at a community college in Silicon Valley. The project applies neural network techniques and two rule induction algorithms, C&RT and C5.0, to choose the best prediction followed by a clustering analysis using TwoStep. The list of students who are predicted as less likely to return to school by data mining is then turned over to faculty and management for direct and indirect intervention. The paper also discusses potential applications of data mining in higher education. The benefits of data mining are its ability to gain deeper understanding of the patterns previously unseen using current available reporting capacities. In addition, prediction from data mining allows the college an opportunity to act before a student drops out or to plan for resource allocation with confidence gained from knowing how many students will transfer or take a particular course. An appendix discusses Cross Industry Standard Procedures for Data Mining. (Contains 4 tables and 10 references.) (Author/SLD)

Statistical clustering is critical in designing scalable image retrieval systems. This paper presents a scalable algorithm for indexing and retrieving images based on region segmentation. The method uses statistical clustering on region features and IRM (Integrated Region Matching), a measure developed to evaluate overall similarity between images that incorporates properties of all the regions in the images by a region-matching scheme. Compared with retrieval based on individual regions, this overall similarity approach: (1) reduces the influence of inaccurate segmentation; (2) helps to clarify the semantics of a particular region; and (3) enables a simple querying interface for region-based image retrieval systems. The algorithm has been implemented as part of an experimental SIMPLIcity image retrieval system and tested on large-scale image databases of both general-purpose images and pathology slides. Experiments have demonstrated that this technique maintains the accuracy and robustness of the original system while reducing the matching time significantly. (Contains 41 references.) (Author)

This study investigated cluster analysis as a method to identify common school score profiles. Scores from more than 100,000 students were used to create school level subscores in reading and mathematics. Four scores for constructing and extending meaning in fiction and nonfiction reading passages were analyzed in reading. Seven subscores were analyzed in mathematics. For reading, the resulting profiles grouped about 70% of the students into a consistently flat (high, average, or low) profile. For mathematics, the resulting profiles grouped only about 7% of the students into a consistently flat profile. Of the other profiles, six had one low or high score, seven had two low subscores, and one had three low scores. While the reading profiles later had clear instructional implications, the profiles from mathematics were less straightforward. More research should be done to validate the identified profiles and their usefulness in instruction. (Author/SLD)

This study examined the nature and extent of the applicability of Corno and Mandinach's (1983) self-regulation model in the new context of activities of searching the Internet. Data were collected from observations of 80 undergraduate subjects while they searched the Internet for information, as well as from survey information regarding their perceptions of their use of self-regulation strategies and processes. Cluster analysis methodology was used to isolate four independent groups of subjects, with labeling consistent with the four forms of cognitive engagement of the model (i.e., self-regulated learning, task focus, resource management, and recipient). The results provide evidence of the application of the cognitive and metacognitive strategies and processes of the Corno and Mandinach model by students while navigating the unstructured Internet. In addition, three patterns of student Internet searching emerged from the data. (Contains 12 references and 4 tables.) (MES)

Each year, millions of state and federal education funds are spent on career development interventions for America's teenagers. However, no uniform nomenclature, definition, and organization of these interventions exists in the professional literature. This hampers communication among and between practitioners, policymakers, and researchers. The lack of a comprehensive organization of these interventions makes quality research and program evaluation very difficult. To help remedy the problems of naming, defining, and organizing in the field of career development interventions, a study was conducted to: (1) identify comprehensive lists of career development interventions that occur in America's secondary schools; and (2) create a taxonomy of the identified interventions. Through consultation with career guidance practitioners and scholars across the country, as well as through examination of research articles, grant reports, and program manuals, a comprehensive list containing 44 interventions was established. These 44 interventions were then rated on five variables (time, mode, control, place, and size) by a random sample of the membership of the Guidance Division of the Association for Career and Technical Education. These ratings were then grouped empirically by means of cluster analysis, producing a four-cluster solution. The four clusters of career development interventions were titled work-based interventions; advising interventions; introductory interventions; and curriculum-based interventions. (Contains 44 references.) (KC)

This paper focuses on clustering a World Wide Web site (i.e., the 1998 World Cup Soccer site) into groups of documents that are predictive of future user accesses. Two approaches were developed and tested. The first approach uses semantic information inherent in the documents to facilitate the clustering process. User access history is then used to reorganize the clusters to better indicate access patterns. This method was found not to be an effective solution to the problem. Hence a second method based on hierarchical clustering of trail information was developed. This method is shown to be far more effective than the first method. (Contains 11 references.) (MES)

One set of approaches to the problem of clustering with dichotomous data in cluster analysis (CA) was studied. The techniques developed for clustering with binary data involve calculating distances between observations based on the variables and then applying one of the standard CA algorithms to these distances. One of the groups of distances that are designed for binary data is known collectively as matching coefficients. There are several incarnations of matching coefficients, but all take as their main goal the measurement of response similarity between any two observations. Thus, distance and similarity come to express the same concept with respect to the observations. This study examined four measures of association that are common to four previous studies. Using Monte Carlo simulation, cluster analysis was conducted using the four distance measures. Under the conditions of this study, the four measures performed very much the same in terms of correctly classifying individuals into two clusters based on dichotomous variables. Another interesting result is that clustering solutions were virtually identical for samples of size 240 and 1,000. (Contains 6 tables, 6 figures, and 12 references.) (SLD)

The effects of a learner-centered approach on self-regulated learning were studied, and a cluster analysis was used to categorize students based on their goal orientation and to relate goal orientation to the self-regulated learning strategies used by the learner. Participants were 114 sixth and seventh graders from 2 multiage classrooms in an urban middle-class school that espoused a learner-centered approach to education. The study used a variety of methods, including the Self-Regulated Learning Interview Schedule and the Patterns of Adaptive Learning Survey, to elucidate the self-regulated learning strategies and achievement goal orientations of these students. The self-regulated learning strategies used most often were seeking, organizing, and transforming information, seeking social assistance from teachers, and goal-setting and planning. Less often used were rehearsing and memorizing, self-evaluation, and record-keeping and monitoring. With regard to achievement goal orientation, these students were most oriented toward developing new skills, the intrinsic value of learning, developing their understanding, and improvement. Findings suggest that self-regulated learning strategy use may be affected by motivation goal orientation. Also, some differences were noted between the two classes with respect to task orientation. (Contains 1 table and 20 references.) (SLD)

This report discusses the outcomes of a study that used cluster analysis to identify groups/clusters of children (n=158) with severe emotional disturbances receiving services at a rural, mental health center in Southeastern Ohio. Extensive data were gathered at intake that included information regarding child and family characteristics and the child's presenting problem. Within this large amount of data, numerous "yes" or "no" survey questions for the parent or guardian of a child were used to assess risk factors. Eight questions were selected due to their perceived salience in predicting subsequent child functioning and mental health service utilization. The selected questions asked if the child had ever been arrested, hospitalized for psychiatric purposes, sexually abused, or physically abused. Questions also asked whether a child had been a sex perpetrator, run away from home, used drugs or alcohol, or attempted suicide. Five groups of children were identified using the eight risk factors present at intake. The groups were labeled physically abused, high runaway risk, high delinquent risk, low risk, and sexually abused. Not only were the groups different in terms of their initial risk factor presentation, but also groups differed in terms of their level of functioning. (CR)

This study used structural modeling methodologies to analyze student outcomes at Prince George's Community College (PGCC) in Largo, Maryland. Path analysis and cluster analysis were used to analyze data on 2,643 first-time entrants to PGCC in fall 1990 over the course of 6 years. Path analysis revealed the centrality of student attitude factors (motivation, flexibility, academic gamesmanship) on student career success, as compared to the lesser impacts of social background, college preparedness, and other factors. It also highlighted the importance of the formative early semesters in setting student trajectories. Cluster analysis identified several varieties of success-prone students, as well as four varieties of low-achievement students. The results are discussed in light of Vincent Tinto's (1993) model of retention, which posits that success-prone college entrants will be those who, during their first year of study, learn how to become students, adopt academic and institutional identities, and integrate themselves into the college community. (MDM)

Most studies of child and adolescent personality have identified three main types (resilients, overcontrollers, and undercontrollers) and further subtypes using Q sort methodology. This study investigated the generalizability of a personality typology by using cluster analysis on the Big-Five scale with adolescents between 12 and 16 years of age. The goals were to: (1) demonstrate the same three personality types in terms of their configuration of scores on the Big-Five personality dimensions; (2) demonstrate subtypes within each of the resilient, undercontrolled and overcontrolled adolescent boys; and (3) validate the types and subtypes with self-reports on perceived relational support, adjustment, delinquency, substance use, and bullying. Participating were 3,284, 12- to 18-year-olds in the Netherlands. Data were obtained through questionnaires and sociometric techniques. The findings indicated that the Big-Five personality measure was able to replicate the three personality types of Overcontrollers, Undercontrollers, and Resilients found in earlier studies. Using a cluster analysis, all participants were classified. The subtypes of Impulsive and Antisocial, Undercontrolled, and Communal and Agentic Resilient adolescents were also identified. Adolescents representing specific personality types and subtypes revealed distinct configurations of adaptation in their self-descriptions of problem behaviors, addictive behaviors, and in peer evaluations of acceptance, rejection, and reputation. (Contains 20 references.) (KB)

Philosophical analysis indicates that underlying much of the Western scientific world view is the metaphysical presupposition of duality, the claim being made that the world is made sense of in terms of either/or and in terms of polarities (e.g., light versus dark). By way of contrast, no concept is more important in Asian philosophical and religious thought than that of nonduality. The basic ideas of quantum physics are not so much difficult as that they are strange. In some situations, electrons that are usually referred to as 'particles' may exhibit 'wave-like' behavior. Both matter and radiation can be viewed as having a dual (wave-particle) nature. In an empirical study of student thinking, the powerful heuristic metaphor of the map is used to construct graphic representations of United Kingdom Advanced level students' understanding of quantum physics. The nature of students' understanding is represented by their construction of groupings of ideas in a personal psychological space with underlying dimensions providing a co-ordinate system for their perceptions. The relationships between students' conceptions of quantum phenomena at the level of the population group are investigated using a structured questionnaire and multivariate analytical techniques (Multidimensional Scaling, Cluster Analysis, and Factor Analysis). A novel quantitative methodology is used to probe students' qualitative implicit understanding. Findings confirm the primacy of dualism in student thinking. Contains 51 references. (Author)

This paper describes the methodology used by The College of St. Rose (New York) to identify peer institutions (similar small private liberal arts colleges) as part of a comparative compensation survey. To identify peer institutions, researchers used a hybrid approach involving both objective identification of institutions and administrator knowledge of similar institutions. Cluster analysis using 33 institutional descriptors and a 16-cluster solution was applied to schools listed in major compendia about higher education for an objective listing. Descriptors covered the categories of membership, size, quality, price, and finance. The resulting list was submitted to the College administration to ensure the list reflected the mission and objectives of the College. Limited resources, practical constraints and political realities of the institution were the motivating factors influencing this approach. The study resulted in the identification of 21 peer institutions in New York State. Institutional researchers are urged to carefully select institutional descriptors in such an effort, recognize the limited statistical techniques available for data analysis, and be sensitive to the political agendas of all parties involved. (Contains 15 references.) (DB)

Two important sets of performance indicators for institutions of higher education have become established in the United Kingdom: research quality ratings and teaching quality ratings. The research quality ratings and, to a lesser extent, the teaching quality ratings influence the level of government funding provided to higher education institutions. This paper considers the correlations between the two ratings and the possible consequences of policies which reshape the higher education sector by concentrating research resources in a limited number of institutions. Comparisons are made between quality assurance/assessment approaches in the United Kingdom and the United States, finding that U.S. higher education is much larger, more heterogeneous and has less government control than U.K. higher education, While the U.S. system of colleges and universities is generally unranked (by those responsible for accreditation), the UK system includes rankings. Use of use various analytical approaches to compare teaching and research ratings for both systems concluded that it remains unclear whether the measurement standards will lead to improvements in teaching and research. (Contains 38 references.) (Author/DM)

The most popular methods for setting passing scores and other standards on educational tests rely heavily on subjective judgment. This paper presents and evaluates a new procedure for setting and evaluating standards on tests based on cluster analysis of test data. The clustering procedure was applied to a statewide mathematics proficiency test administered to 818 seventh-grade students in a small urban/suburban school district. Content area subscores were derived from the test specifications to serve as clustering variables. Subsequent course grades in mathematics were used to validate the cluster solutions and the stability of the solutions were evaluated using two random samples. The three-cluster (K-means) solution provided relatively homogeneous groupings of students that were consistent across the two samples and were congruent with school mathematics grades. Standards for "intervention,""proficient," and "excellent" levels of student performance were derived from these results. These standards were similar to those established by the local school district. The results suggest that cluster analytic techniques may be useful for helping set standards on educational tests, as well as for evaluating standards set by other methods. Suggestions for future research are provided. (Contains 2 figures, 7 tables, and 23 references.) (Author/SLD)

Portfolio assessment has become a widely cited technique for obtaining a rich demonstration of the skills that students have acquired, and it is often implied that portfolio assessment provides different types of information from that obtained from standardized tests. This paper provides support for the contention that portfolio assessments do indeed provide information about student capabilities that is different from that available from standardized tests. This support is given in the context of a portfolio assessment model called PASSPORT designed for use with students in grades 9 through 12. PASSPORT is a portfolio assessment system for collecting and presenting evidence of student knowledge and skills in the areas of English language arts, mathematics, and science. Students produce portfolios of their work through a loose set of guidelines. Each portfolio contains five work samples and a cover letter that explains why samples were included and what they tell about the student's capabilities. To compare the content assessed by PASSPORT and the American College Testing Program (ACT) Assessment, a multidimensional item response theory model (MIRT) was applied to the mathematics and language arts scores from the portfolio and the multiple choice test. The MIRT analysis was followed by a cluster analysis of similarities of the dimensions assessed by the items. Comparison of the content and cluster analyses shows clear differences in the types of skills and knowledge that are assessed. The portfolio provides evaluations of student performance on major writing tasks and on mathematics skills related to data analysis and problem solving, while the standardized test provides information about the details of the writing process and the rules for manipulating mathematical expression. (Contains two figures, six tables, and nine references.) (SLD)

Standard statistical methods are used to analyze data that is assumed to be collected using a simple random sampling scheme. These methods, however, tend to underestimate variance when the data is collected with a cluster design, which is often found in educational survey research. The purposes of this paper are to demonstrate how a cluster design affects the standard error statistic and the subsequent analyses, and to present practical techniques to analyze data from cluster designs correctly. A heuristic example is given to illustrate how to compute the variance estimate for a cluster design and the corresponding design effect. Simulation data is then used to examine variance estimation results from one- and two-stage cluster designs, respectively. Both a formula approach and a jackknife resampling approach are used in obtaining variance estimates. It is shown that, for 150 observations sampled from a population of 1,000, using a 2-stage cluster design, the actual variance can be underestimated by a factor of 3 if the standard statistical method is used. The underestimated variance or standard error statistic will lead to unwarranted statistical significance in hypothesis testing, or a narrow confidence interval in parameter estimation. Consequently, misleading conclusions can be made based on these inappropriate analysis findings. (Contains 1 tables and 11 references.) (Author/SLD)

Context-dependent items are traditionally analyzed independently, creating a situation in which the potential local item dependence effects among these items may cause a biased estimation of examinees' abilities. This study investigated the local item dependence effects on testlets in the tryout version of a statewide science assessment by a Rasch partial credit model. Cluster sampling combined with stratified sampling was used. Data were analyzed in five different configurations to study the relationships between context-dependent items at the individual item level and at the testlet level. It is shown that local dependence effects may be controlled and a better fit for testlet calibration can be obtained by employing the Rasch partial credit model for some, but not all testlets. (Contains 2 figures, 11 tables, and 35 references.) (Author/SLD)

To track student progress and develop a model of the total academic process at Prince George's Community College (PGCC), the Office of Institutional Research and Analysis acquired 5 years of background, behavioral, and outcome data for a cohort of first-time students entering the college. The academic process was broken down into three divisions: input (environmental variables measuring pre-college student attributes and college-external personal circumstances), through-put (process variables measuring student progress), and output (outcome variables measuring fulfillment of educational objectives). Exploratory modeling was used to create a structured academic process. The research first used regression analysis and created 10 new academic process factor scales, plus 7 student background indicators. Next, a model was created to explain student outcomes through the application of causal path analysis. The last phase of research was to produce a second model complementary to the path model but less abstract. Cluster analysis was applied to the cohort data, resulting in 10 clusters describing student attributes and behaviors. Though the path and cluster models were highly tentative, valuable insights were offered, including the importance of personal motivation, taking student career differences seriously, and creating academic support programs. Appendices include cluster data tables and methodology of social background variable construction. (YKH)

This paper analyzes the mission statement of a small liberal arts college: (1) to establish a rationale for examining a mission statement from a postmodern perspective, which will aid in the understanding of the context within which the mission statement and accompanying goals were developed and produced; (2) to clarify the multiple voices and readings of the rhetorical act, by identifying the ultimate terms used in the act and the clusters associated with them; and (3) to identify conflicts among various voices in the statement. The paper examines the statement and goals by deconstructing the text, context, and readers of the rhetorical act. Using Kenneth Burke's method of discovering ultimate terms and cluster analysis, the paper identifies terms directed at specific readers. And, through the enlightened lens produced by the deconstruction and cluster analysis, the paper discusses the conflicts of power existing in the mission statement. It points out that the college studied developed a statement of their "new" mission and goals as an introduction to the revised general education curriculum package intended to "take students into the 21st century." The paper finds that, in presenting a text to serve multiple readers, the message is diluted, occupying a dual position of change and tradition. The paper notes that those involved in the production of the rhetoric of higher education must realize the rhetorical implications that offer them either success or failure as they present themselves to the multiple voices that constitute modern society. Contains 29 references. (NKA)

This article illustrates how a cluster analysis can be conducted, validated, and interpreted. Data normed for a behavioral assessment instrument with 14 scales on a nationally representative sample of U.S. school children were utilized. The discussion explores the similarity index, cluster method, cluster typology, cluster validity, cluster structure, and prediction of cluster membership. The Behavior Assessment System for Children (BASC) form that assessed 6- to 11-year-old students with the Teacher Rating Form (TRS-C) was used with a sample of 1,228 elementary school children. The clustering method involved a two-step procedure: a Ward hierarchical analysis followed by an iterative cluster partitioning via a K-means analysis. As illustrated, the following steps are suggested for cluster analysis: (1) select the study units (children); (2) choose the system of response variables; (3) decide how to measure the response variables; (4) select the similarity index; (5) select the cluster method; (6) determine the initial cluster typology; (7) provide some evidence of cluster validity; (8) interpret final cluster typology; (9) describe final cluster structure; and (10) develop a classification rule for new units. (Contains 2 figures, 9 tables, and 48 references.) (SLD)

Multidimensional scaling and cluster analysis were used to describe and categorize tasks from six graduate disciplines including academic psychology, applied psychology, English literature, journalism, physics, and electrical engineering. A sample of task descriptions was constructed through interviews with four or five graduate students from each of these disciplines. A rating instrument was designed to describe task goals and to evaluate whether tasks were well or ill structured with respect to various aspects of problem definition and problem solution. Graduate faculty (three to five per discipline) used the rating instrument to characterize a sample of tasks from their disciplines. The scales were found reasonably reliable and were useful in identifying and describing task cluster and how such clusters varied both within and across disciplines. A cluster of short-term problems that were posed by someone other than the student was found in every field, although the other characteristics of this cluster of tasks varied with discipline. In all disciplines except physics, a cluster of complex tasks emerged that was characterized as having multiple objectives that needed to be satisfied. The cluster of complex tasks found in physics was not described clearly by the scales. Problem-finding was an important task characteristic in the social sciences and humanities, but not in the physical sciences. The relevance of multidimensional scaling and clustering to test design is discussed. An appendix contains examples of task descriptions. (Contains 8 figures, 15 tables, and 18 references.) (Author/SLD)

This paper reports on the first stage of the construction and refinement process underway in pursuit of developing a brief composite instrument to measure teachers' attitudes toward computers. Items (n=284) from 14 questionnaires assessing teacher attitudes toward computers were completed by 118 educators in Texas in 1995. Internal consistency reliabilities for the 32 Likert-type and Semantic Differential subscales included in the battery ranged from 0.41 to 0.96, with 27 of the 32 indices falling in the "respectable" range of 0.70 of higher. Forty-five of the correlation indices computed for the subscales were 0.70 or higher, indicating that many subscale pairs shared half their variance or more in common. A higher-order factor analysis of the 32 subscales indicated that 4 higher-order attributes probably exist among the 32 subscales, and a cluster analysis of the subscales produced 6 major clusters. The next step will be to gather data from a greater number of educators to derive a stable factor structure for the original items and an eventual shorter form. Appendix A lists the TAC subscales, and Appendix B presents the test. (Contains 2 figures, 7 tables, and 23 references.) (SLD)

A Monte Carlo study compared the usefulness of six variable weighting methods for cluster analysis. Data were 100 bivariate observations from 2 subgroups, generated according to a finite normal mixture model. Subgroup size, within-group correlation, within-group variance, and distance between subgroup centroids were manipulated. Of the clustering methods examined, the flexible average algorithm with beta equal to -.15 or -.20 gave the best recovery. Of the remaining methods, that of J. H. Ward yielded the best recovery, followed closely by beta-flexible linkage and the EML algorithm of the Statistical Analysis System. In the absence of variable weights, negative within-group correlation resulted in much poorer recovery for all clustering algorithms. The ACE weighing method of D. Art, R. Gnanadesikan, and J. R. Kettenring proved preferable overall. Clustering with Mahalanobis distance based on the pooled within-group covariance matrix indicated that knowing the correct covariance method would yield improved recovery over the ACE method approximately 10% of the time. Two appendix figures provide weighting data. (Contains 8 figures, 7 tables, 2 appendix figures, and 40 references.) (Author/SLD)

The term differential item functioning (DIF) refers to whether or not the same psychological constructs are measured across different groups. If an item does not measure the same skills or subskills in different populations, it is said to function differentially or to display item bias. A multilevel approach to DIF is proposed. In such a model, the dependency between observations due to cluster effects is explicitly taken into account. Results of a multilevel logit model and of a multilevel logistic regression model are compared with results of analogous unilevel models. The procedure is illustrated with data from a national assessment of geography performed with respect to gender bias. Each of the 294 items was answered by an average of 2,161 respondents. Analysis supports the use of multilevel models, which have the advantage of accounting for cluster effects in data from a hierarchical population. DIF does seem to be more stable according to multilevel models than to unilevel models. Five tables and four figures present analysis results. Appendixes provide parameter estimates. (Contains 19 references.) (SLD)

A study determined differences among the levels and types of career indecision within and between two cross-cultural groups of undergraduate students from two counseling centers. The groups were 88 University of Pittsburgh students and 79 Universidad del Pacifico, Peru, students. The Indecision Scale of the Career Decision Scale (CDS) was used to examine levels and types of career indecision. Cluster analysis was performed to determine groups with similarities in indecision. One-way analyses of variance found differences within levels of indecision. Two-way analyses of variance compared each of the cluster levels of the two groups. Pearson Product Moment Correlation Coefficients and t-tests were used to establish differences between age and sex, but none were found. Findings identified four defined clusters in the levels of indecision per group. Data from the Pittsburgh group showed a higher level of indecision per cluster than the group from Peru, suggesting that cross-cultural characteristics may influence the decision-making process for undergraduate students. (Appendixes contain 12 references and item-item reliability tables from the CDS for Pittsburgh and Peru.) (YLB)

A study was conducted at the Community College of Philadelphia (CCP) to examine the course-taking patterns of 94 graduates of the associate in arts (AA) curriculum, 1,957 graduates of the association in general studies (AGS) curriculum, and 99 graduates of the associate in science (AS) curriculum. Using a computer-based approach to transcript analysis, the study identified the average number of courses in various disciplines taken by graduates of each of the three curricula. Study findings included the following: (1) an overall look at course clusterings across the three curriculum pointed to five general career directions for graduates: liberal arts transfer, teacher preparation, health preparation, chemical technology and pre-pharmacy, and engineering science; (2) the common courses taken by graduates across all three curricula included an English writing and a research course; Introduction to Psychology; Introduction to Sociology; General Biology; one of three English communication courses; and a first course in either French or Spanish; (3) AS graduates completed a number of second-level mathematics and science courses as part of their study; however, there was minimal completion of second-level general education courses in humanities or the social sciences; (4) AA graduates completed second-level language study as part of their curriculum requirement, and AGS graduates completed second-level psychology, sociology, and biology courses, most likely as a requirement for the nursing curriculum; (5) the AGS graduates were the most likely of the three groups to have completed developmental English courses; (6) a number of students in each curriculum had what appeared to be specific career goals, but there was also evidence that several engaged in career exploration in their course selections. Recommendations concerning curriculum reform and core requirements are included. (KP)

Monte Carlo studies investigated effects of within-group covariance structure on subgroup recovery by several widely used hierarchical clustering methods. In Study 1, subgroup size, within-group correlation, within-group variance, and distance between subgroup centroids were manipulated. All clustering methods were strongly affected by within-group correlation; negative correlation yielded much poorer recovery. Smaller effects were found for the interaction of clustering method and within-group variance. Study 2 separated effects of direction of correlation from the direction of differences in the subgroup centroids. Subgroup size, within-group correlation, direction of the vector separating subgroup centroids, and distance between subgroup centroids were manipulated. Superior recovery was associated with within-group correlation that matched the direction of subgroup separative. Overall, the EML algorithm of the Statistical Analysis System yielded best recovery, followed closely by Ward's method, average linkage, and a version of the beta-flexible algorithm. Several alternative measures are discussed. Six tables and eight figures present analysis data. (Contains 52 references.) (Author/SLD)

Inclusion of irrelevant variables in a cluster analysis adversely affects subgroup recovery. This paper examines using moment-based statistics to screen variables; only variables that pass the screening are then used in clustering. Normal mixtures are analytically shown often to possess negative kurtosis. Two related measures, "m" and coefficient of bimodality "b," are also examined. A Monte Carlo study compared the screening measures to no selection, De Soete's (1988) ultrametric weights, and the forward selection procedure of Fowlkes, Gnanadesikan, and Kettenring (1988). Screening based on kurtosis degraded recovery and is not recommended. In contrast, screening on "m" or on "b" improved recovery over both no selection and forward selection, and screening performed as well as ultrametric weights. Combining screening with ultrametric weights performed extremely well. All methods were found to be somewhat sensitive to other types of error. Screening variables appears to be a viable alternative to both ultrametric weights and forward selection. The potential advantages and disadvantaged of screening are considered. Eleven tables and six figures illustrate the analyses. An appendix provides supplemental detail. (Contains 51 references.) (Author/SLD)

The relationship between career indecision subtypes and ego identity development was examined in a study of 423 college students (aged 18-26) who were attending undergraduate psychology classes at five southeastern universities and colleges and who had not yet decided upon a career. The students were divided into the following four cluster groups on the basis of their scores on the Career Factors Inventory: ready to decide, developmentally undecided, choice anxious, and chronically indecisive. A profile analysis technique that included a multivariate analysis of covariance was used to identify relationships among the four cluster groups and the first five stages of Erikson's psychosocial model of identity development--trust, autonomy, initiative, industry, and identity. The ready-to-decide group scored significantly higher on the trust, industry, and ego identity subscales than the developmentally undecided, choice-anxious, and chronically indecisive groups did. The ready-to-decide group also scored highest on the autonomy subscale. On the initiative subscale, however, no significant differences were found between the scores of the ready-to-decide and developmentally undecided groups. Career counseling interventions were suggested for each of the four career decision groups. (Contains 31 references.) (MN)

Groups from two Michigan-based science education projects, funded through the Kellogg Foundation Science Education Initiative, used networking conferences with cluster evaluation processes for their program assessment format. Cluster evaluation is categorized as a type of naturalistic evaluation, and it includes negotiated common cluster outcomes, collaborative data collection and analysis procedures, regular networking conferences, technical assistance to individual projects, and cooperative derivation and dissemination of findings. Networking conferences (NC) are special gatherings held semi-annually of project directors, evaluators, other key project staff, foundation staff, invited guests, and cluster evaluators. The Science Education Initiative aims to improve science teaching and learning with a long-term goal of improving the scientific literacy of Michigan citizens. The specific purposes of the cluster evaluation networking conferences included the following: (1) conduct strategic planning for, exchange ideas about, provide direction to discuss issues and problems emerging from, and review and analyze data and findings of the cluster evaluation; (2) share lessons learned with other projects; (3) learn about current and developing issues in science education and science curriculum, instruction, and assessment topics directly pertinent to projects; (4) formally and informally share science education curriculum materials and instructional strategies; and (5) visit project sites. (Contains nine references.) (JB)

PG-TRAK90 is a cluster-based geographic marketing system designed by Maryland's Prince George's Community College (PGCC) to maximize educational marketing objectives. To create it, United States Census Bureau files containing over 200 demographic, housing, and lifecycle variables for 172 tracts in Prince George County (PGC) were reformatted into marketing-style indicators and subjected to a statistical sorting technique known as cluster analysis. The result was a typology of PGC neighborhoods sorted into 22 clusters based on geographic, socioeconomic, and ethnic categories, including the mostly white, upscale business executives in the "exurbs"; Black Middle America; and the inner suburban mix of young Black and Hispanic families, among others. To determine how well PGCC draws students from across this demographic spectrum, clusters-in-the-county were compared to clusters-in-the-student-body revealing that as of 1990, the student body closely mirrored the county demographics. An analysis of market penetration by cluster revealed that upscale groups sent proportionally more students to PGCC than downscale clusters. Credit versus non-credit course markets were also analyzed for 1985-90, indicating similar proportions of enrollment by cluster in both types of courses. PG-TRAK90 is incorporated directly into PGCC's student recruitment efforts and will provide critical market analysis and contact targeting. (KP)

The major innovation of the American Psychiatric Association's "Diagnostic and Statistical Manual III" (DSM-III) was the explicit specification of criteria for determining whether an individual's problems qualify for a particular diagnosis. However, neither the choice of child diagnostic categories in DSM-III nor the choice of criteria to define each category was based on empirical findings. In this paper, an approach to taxonomy is developed that involves quantitative analyses of standardized assessment data to identify groupings of problems that tend to co-occur, as reported by each type of informant. Primarily by using factor analysis and principal-components analysis, syndromes of co-occurring problems are derived. The overall approach is called "Multiaxial Empirically Based Assessment and Taxonomy." Eight cross-informant syndrome constructs are derived from parent-, teacher-, and self-reports. A computerized profile is developed for scoring an individual child in terms of the eight cross-informant syndromes normed for that child's sex and age and the type of informant. Cluster analyses of profiles, performed in 1993, are being used to identify children who share patterns of syndrome scores. Fourteen figures present analysis data. (Contains 4 references.) (SLD)

Various methods used to assess the content of a test are reviewed, and a new procedure designed to improve on these methods is presented. The two tests considered are a professional licensure examination, the auditing section of the Uniform Certified Public Accountant Examination, and an educational achievement test, a nationally standardized social studies achievement test. Previous methods have generally been empirical, using factor analysis or multidimensional scaling (MDS) to analyze the inter-item correlation matrix derived from examinee responses, or subjective, using the data provided by subject matter experts (SMEs) to determine whether items represent content areas that the test purports to measure. A method has previously been proposed that uses MDS to discover dimensions obtained from the analysis of ratings by SMEs of the similarity of items comprising a test. This study expanded that method by using 2 groups of SMEs (15 for each test) to evaluate the content of the 2 tests studied. Correlation and cluster analyses results suggest that the content structure of a test can be evaluated adequately by analyzing item similarity data provided by SMEs. Results further suggest that the MDS procedure should be used to supplement analyses of item relevance data rather than replace them. Six figures and 18 tables present analysis findings. (Contains 23 references.) (SLD)

An ethnographic study assessed the family constellations of ninth graders in an urban public school through cluster analysis. Survey data were collected from a random sample of 120 high school freshmen and were used to compute 4 variables for each student: (1) presence of the mother in the home; (2) presence of the father in the home; (3) presence of other adults in the home; and (4) presence of other children in the home. The familial configurations of subjects were extended maternal (n=7), nuclear family (n=26), mother with children (n=46), orphaned (n=13), only child (n=21), and adopted (n=7). Family constellation, determined by cluster analysis, and gender served as independent variables in a general linear model used to assess student attendance at school and grade point average. Results indicate that the educational performance of males and females differs with respect to the types of familial configurations in which they find themselves. Females appear to perform most poorly in family units characterized by mother's presence and father's absence, but males perform best in groups where the mother and/or maternal relations were present. Contrary to expectations, males in this group of students perform best when the father is absent from the family configuration. This ethnographic approach appears promising for revising familial configuration effects within heterogeneous and transient populations. Two tables present study data. (SLD)

This paper presents a methodology for developing a framework that can help decision makers distinguish intensity of needs and regional funding differences within a state. A cluster analysis was used to distinguish need in Georgia's 159 counties. Four clusters that identified the distribution of goods or services to those in need were identified: (1) low educational achievement, sparsely populated; (2) high unemployment, rural; (3) urbanized; and (4) "typical" rural. A detailed discussion of the methodology describing the unit of analysis, selection and preparation of variables, factor analysis, and cluster model determination is provided. Advantages of the typology are that it organizes many variables into a small number of groupings and uses inductive techniques that limit researcher bias. Two suggestions for determining the distribution of resources are offered: to use a weighted per-capita figure, or to use a per-capita basis with individual agencies establishing a separate fund to be allocated according to need. Finally, the framework helps agencies to focus on the critical needs of counties. Six tables and a map of Georgia counties are included. (Contains 17 references.) (LMI)

In contrast to the mass-related verbal flow description given in part I, this report focuses on the cooperative interaction of textual agents and objectives in the production of information flows. Perspective Text Analysis (PTA) is used with the purpose of establishing their physical and functional significance in a non-mass based description of text building behavior. The most important feature of part II is the double aspect of the methodological approach to text building. One aspect of text building is an elaboration of Gibson's methodology into the study of a language-specific pickup of ecological invariants, exploring the abstract projections of ecological optics onto language spaces and the way in which projected optical flow fields constrain the coupling of perception and action, i.e. locomotor activities. The other refers to physical conditions as provided by the famous Visual Cliff experiments and their theoretical significance in an explanation of the established temporal morphologies. These have been represented within a topological space. This space is conceived of as a collection of points that are connected by affinity relations determining the As and Os of the AaO schema. It is shown that each point can be represented by a different number, and that the concentration of these points in the topological space ultimately becomes helical. It is also demonstrated that self-references and self-organization have significance for the embedding of the perceived topological properties of the experimental environment into text. The results of the analysis show that the ensemble of texts macroscopically is dominated by highly similar flow-field properties. Three figures, one table. An attachment contains cluster analyses. (Contains 15 references, 3 tables, and 1 figure.) (Author)

This study assessed whether the Academic Misconduct Survey (AMS) can distinguish groups of teacher education students relative to their propensity toward various forms of misconduct. Focus was on identifying clusters of persons within each school who had a propensity toward the various academic misconduct constructs. The AMS includes 41 Likert-type items that measure academic misconduct across 5 constructs. In the spring of 1991, the AMS was administered to 90 undergraduate teacher education students enrolled in foundations of education classes in 1 large urban state university, 1 small rural state university, and 1 small private religious college in the southern United States. Data from 15 students in each school were used in the analyses, for a total sample of 45 students. Separate Q-technique factor analytic procedures were conducted for each sample. Several identifiable person factors emerged across these various analyses. Similarities in results across the three samples suggest that the results are due to systematic differences in people rather than spurious correlations attributable to sample bias. Students across the three schools give higher ratings to items pertaining to quasi-misconduct, cheating on tests and assignments, and use of illegal resources. There were fewer occurrences of use of illegal resources. Individuals across schools gave lower ratings to items related to subtle manipulation and bold manipulation. Six tables and an appendix listing AMS items are included. (Author/RLC)

Three important methods areas of multivariate analysis that are not always thought of in terms of latent variable constructs, but for which latent variable modeling can be used to great advantage, are discussed. These methods are: (1) random coefficients describing individual differences in growth; (2) unobserved variables corresponding to missing data; and (3) variance components describing data from cluster sampling. An educational achievement dataset of longitudinal observations on secondary mathematics achievement (the National Longitudinal Study of American Youth) is described as a motivating example. It is shown that all three topics can be simply expressed in terms of latent variable modeling that fits into existing and generally available structural modeling software. This approach makes possible a connection between psychometricians and other methodologists interested in latent variable modeling. Interesting extensions of these statistical analyses are discussed. One table presents missing data patterns.

This handbook presents a process for the assessment of college curricula and student learning in general education and liberal learning. The key concepts of the Coursework Cluster Analysis Model (CCAM) are described and explained. Some of the general guidelines on interpreting the results of CCAM analyses are presented as well as key decisions regarding reform of the curriculum, although revision of the assessment program and enhancement of the student advising system will have to be made on each campus employing the CCAM model. Specific chapters discuss CCAM and the assessment of student learning, the research that went into developing CCAM, the considerations for planning for the assessment of student learning, a general overview of the methods and procedures in using CCAM, determining student learned abilities, and the detailed procedures for conducting the coursework patterns analysis. Appendices provide studies of the CCAM analysis and the job control language required to perform the CCAM procedures. Contains 50 references. (GLR)

A statewide survey of 536 randomly selected Virginia secondary school teachers of academic subjects explored practices and opinions concerning various aspects of classroom testing and grading. The main focus was on the use of preestablished percentage scales as a basis for evaluating students. The responses suggested that large proportions of teachers hold opinions and pursue practices contrary to what many measurement specialists would recommend. Cluster analysis identified a small group of teachers whose opinions were largely consistent with what measurement specialists would recommend. This group differed from five other cluster groups in that it contained a disproportionate number of mathematics and science teachers and its mean years of out-of-state experience was substantially greater than that of any other group. Opinions and practices characterizing each of the other groups were extremely diverse. Some groups held views that might be considered inconsistent or self-contradictory. These findings led to the recommendation that the study questionnaire be adapted for administration to groups undergoing inservice training in measurement to facilitate focusing instruction according to the characteristics of each group. Seven tables and one figure present study findings. Four appendixes contain the survey instrument, a follow-up request, and proposed instruments for other research and inservice training. (Author/SLD)

Approximately 600 teachers in 32 public elementary schools and secondary schools from 1 urban southeastern school district participated in a study conducted to identify prototypes of school climate concerning decisional participation cultures. A 19-item decisional participation measure was used, and the median score of the teachers at each school was computed on each item. A Q-technique factor analysis was conducted to isolate the prototypes. Three school prototypes were identified. The first cluster of schools might be characterized as involving climates that were student-interactions-targeted. The second cluster involved school climates that were policy-focused. The third cluster involved climates that focused on instructional-formalities. Related variations in participation patterns were also analyzed. Three tables, 4 figures, an appendix of schools sorted into groups, and 32 references are included. (Author/SLD)

As part of a research program investigating the interlanguage lexicon of Kenyan learners of English, based on the lexical field of locomotion, a card sorting task was used to produce raw similarity measure data. This was subjected to cluster analysis and multidimensional scaling (MDS). This paper suggests that cluster analysis/MDS of perceived similarities of selected verbs in a sorting task may indicate possible lines of investigation of the lexical knowledge of different groups of learners in comparison to that of native speakers. (Contains 25 references.) (Author)

This document describes how the main principles of Perspective Text Analysis are implemented in the PC-system PERTEX, concentrating on the main steps of the analysis. The analysis starts with normal text and ends in a topological representation of the mentality that the text presents. The text material is processed in the following main steps: (1) coding of function words by means of a special dictionary; (2) design and coding of blocks according to the AaO (Agent-verb-Objective) paradigm; (3) supplementation of A- and O-dummies; (4) generation of A/O matrices; (5) cluster analysis based on generated matrices; and (6) topological presentation of outcomes. PERTEX gives an integration of all the steps in the analysis, and the user is offered numerous comprehensive functions for automatic coding and control of syntax. By a multilingual design, PERTEX can operate on texts in different languages. The user can select different menu-languages for the interaction with PERTEX. The technical output of the system is illustrated in the appendix with the complete 17-page printout from analysis of a classic text. (Author/SLD)

This report presents the findings of the application of the Differential Coursework Patterns (DCP) cluster analytic model to the tests and transcripts of a combined sample of 1987-88 and 1988-89 graduating seniors at Evergreen State College (Washington). The project found that enrollment in different patterns of programs was consistently associated with gains in the general learned abilities of undergraduate students. Other findings indicated that the development of general learned abilities did not have an exact one-to-one relationship with departmental categories; that the development of general learned abilities was not confined to the lower division; and that there was little formal monitoring and description of the curriculum in terms of general learned abilities. The report is divided into seven sections, each describing one of the following research components: the theoretical basis for the differential coursework hypothesis including the development of the cluster analytic model as a means of testing the hypothesis; the procedures and methodology used; the major characteristics of the two samples of graduating seniors whose transcripts and test scores provided the data analyzed; characteristics of the programs found on the student transcripts in terms of the criterion variables; the intended curriculum of the institution, its goals, curriculum organization, and structure; the findings; and a summary, conclusions, and recommendations. Contains 199 references. (Author)