Advances in technology have altered data collection and popularized large databases in areas including education. To turn the collected data into knowledge, effective analysis tools are required. Traditional statistical approaches have shown some limitations when analyzing large-scale data, especially sets with a large number of variables. This dissertation introduces to educational researchers a new data analysis approach called data mining, an analytic process at the intersection of statistics, databases, machine learning/artificial intelligence (AI), and computer science, that is designed to explore large amounts of data to search for consistent patterns and/or systematic relationships between variables. To examine the usefulness of data mining in educational research, one specific data mining technique--the Bayesian Belief Network (BBN) based in Bayesian probability--is used to construct an analysis model in contrast to the traditional statistical approaches to answer a pseudo research question about faculty salary prediction in postsecondary institutions. Four prediction models--a multiple regression model with theoretical variable selection, a regression model with statistical variable extraction, a data mining BBN model with wrapper feature selection, and a combination model that used variables selected by the BBN in a multiple regression procedure--are expounded to analyze a data set called the National Survey of Postsecondary Faculty 1999 (NSOPF:99) provided by the National Center of Educational Services (NCES). The algorithms, input variables, final models, outputs, and interpretations of the four prediction models are presented and discussed. The results indicate that, with a nonmetric approach, the BBN can effectively handle a large number of variables through a process of stochastic subset selection; uncover dependence relationships among variables; detect hidden patterns in the data set; minimize the sample size as a factor influencing the amount of computations in data modeling; reduce data dimensionality by automatically identifying the most pertinent variable from a group of different but highly correlated measures in the analysis; and select the critical variables related to a core construct in prediction problems. The BBN and other data mining techniques have drawbacks; nonetheless, they are useful tools with unique advantages for analyzing large-scale data in educational research.

Institutional efforts to organize the admissions process in several Mexican universities have led to the adoption of standardized instruments to measure applicants' initial academic qualifications for career programs. The University of Sonora, located in four campuses throughout the state, initiated the administration of a college level entrance exam in the fall of 1997. The Examen de Conocimientos y Habilidades Basicas (EXCHOBA), developed in 1991, is the instrument employed for aiding the academic and administrative agencies in making admissions and career placement decisions to date. Drawing on current practice, this project develops a model for investigating the alignment of the high school curriculum with the entrance examination by extracting and clarifying the academic standards that derive from the official curriculum. Through a series of statistical analyses on data from exam administrations, a working model for defining the standards along with the instruments' sub-tests is proposed. The basis for a system are then suggested to assist high school and university agencies and administrators to interpret the results with a clear set of procedures for making curricular and instructional decisions that will help improve the current rates of success in the different career programs at the institution. In particular, the results obtained will lead to a proposal to improve the academic advising and guidance programs that the Universidad de Sonora is currently implementing to improve student retention and graduation rates in its career programs.

The subjective nature of oral examinations often lead to reliability estimates that are lower than other types of examinations (i.e., written examinations). The potentially biasing individual attributes of examiners (i.e., experience) are of particular concern since the oral examination process depends specifically upon the quality of their assessments. In addition, traditional reliability estimation procedures are not always possible for some oral exams due to the utilization of incomplete measurement designs (i.e., one examiner per candidate) resulting from the inherent high costs and complicated logistics associated with large scale oral examinations. Consequently, the current study attempts to evaluate the quality of one such exam by developing alternative indicators of exam quality using a pre-existing data set. A series of examiner agreement variables were calculated for low, moderate, and high ability candidates and subsequently correlated with each other. A series of exploratory multiple regressions were also used to evaluate the potential impact of several examiner characteristics (experience, gender, specialty, variance of scale use, and fail rate) confined in the data set. Finally, a generalizability (G) study was conducted on a subset of the examination that utilizes a complete measurement design (i.e., two examiners evaluating the same candidate, and all examiners examine all candidates) for lower ability candidates. The G study was then followed by a decision (D) study to determine both the current level of dependability with two examiners, and how much the dependability of the process would improve by adding mure examiners. The results of the current study suggest that evaluating lower ability candidates is different and more difficult than evaluating higher ability candidates. Furthermore, systematic sources of error related to examiners appears to be less or a concern than previously anticipated. Finally, the results of the G-D studies suggest that the current dependability of evaluating lower ability candidates with two examiners could be greatly improved by adding additional examiners to the process.

No description available.

No description available.

No description available.

No description available.

No description available.

Venn diagrams were invented by John Venn in 1880 as an aid in logical reasoning. Since then, the diagrams have been used as an aid in understanding and organization for widely diverse audiences (e.g., elementary school children, business people) and widely diverse content areas (e.g., self-improvement courses, statistics). In this dissertation, Venn diagrams are used to illustrate and explain variable relationships. There are three main foci: (a) correlation and interaction, (b) variables and Venn diagrams, and (c) reliability and Venn diagrams. Confusion between correlation and interaction is explained, and the multicollinearity problem is illustrated using a Venn diagram composed of three circles and a horse-shoe shaped figure. Venn diagrams are presented for these variables: moderator; concrete and hypothetical intervening; component; traditional, negative, and reciprocal suppressor; covariate; disturbance; and confound. Venn diagrams are also used to differentiate among within-subjects, between-subjects, and reliability designs. Last, a detailed example, which assumes basic knowledge of classical test theory and generalizability theory, is presented to help illustrate, using Venn diagrams, the role of error variance in performance assessments. Evaluation based on comments of 13 American Educational Research Association, Division D, listserve members and 7 non-members was positive, and interest in the topic was shown by over 100 visits to the website where a portion of the dissertation was posted.

No description available.