Beyond Sectionality and Into Sizeness, or How Course Size Effects Grades: An Exploration of the Multiple‐Institution Database for Investigating Engineering Longitudinal Development Through Hierarchical Linear Models

George D. Ricco

Download Paper | Permalink

Conference: 2015 ASEE Annual Conference & Exposition
Location: Seattle, Washington
Publication Date: June 14, 2015
Start Date: June 14, 2015
End Date: June 17, 2015
ISBN: 978-0-692-50180-1
ISSN: 2153-5965
Conference Session: Examining "Big" Data
Tagged Division: Educational Research and Methods
Page Count: 12
Page Numbers: 26.280.1 - 26.280.12
DOI: 10.18260/p.23619
Permanent URL: https://peer.asee.org/23619
Download Count: 521

Request a correction

Paper Authors

biography

George D. Ricco Purdue University, West Lafayette

visit author page

George D. Ricco is the KEEN Program Coordinator at Gonzaga University in the School of Engineering and Applied Science. He completed his doctorate in engineering education from Purdue University’s School of Engineering Education. Previously, he received a M.S. in earth and planetary sciences studying geospatial imaging and a M.S. in physics studying high-pressure, high-temperature FT-IR spectroscopy in heavy water, both from the University of California, Santa Cruz. He holds a B.S.E. in engineering physics with a concentration in electrical engineering from Case Western Reserve University. His academic interests include longitudinal analysis, visualization, semantics, team formation, gender issues, existential phenomenology, and lagomorph physiology. He lives in romantic Spokane with his leporidae partner, Rochelle Huffington Nibblesworth.

visit author page

Download Paper | Permalink

Abstract

Exploration of the Multiple‐Institution Database for Investigating Engineering Longitudinal Development through Hierarchal Linear Models (HLMs) and clustering The MIDFIELD database includes complete student records of twelve institutions offering engineering degrees in the United States with a total student record of more than 1,000,000. These institutions enroll more than twelve percent of the nation’s total engineering students at any given time. While researchers have explored MIDFIELD using regression analysis, thus far only single‐level methodologies have been used. Hierarchical Linear Models, (sometimes called multi‐level models, nested models, or generalized mixed models,) provide a unique interpretive tool to probe a database such as MIDFIELD. Unlike ANOVA analysis variants, HLMs allow for robust analysis of incomplete data sets. For instance, in the case of a database of student grades, ANOVA methods require complete (or duplicate) recording of grades at each interval used for analysis. In other words, if one student has a missing grade for one semester, then that student cannot be used in the analysis. HLMs do not require completeness for strict convergence. More importantly, HLMs retain the nested structure of the data itself through the analysis. This aforementioned nested data structure may not necessarily be known a priori. Techniques of cluster analysis can be used to identify the presence of particular partitions in an arbitrary data set. The utilization of such methods can be paired with HLMs to provide a powerful, data driven framework for analyzing this phenomenon. In an expansion of a previous study, we focus MIDFIELD upon first‐ and second‐ year courses undertaken by most core curricula for our analysis: introductory physics; calculus; chemistry; computer languages; statics; and others. We then use a form of the null model in HLM in order analyze student grades. The null method allows us to glean an understanding of the variation of section grade distribution in these required courses through the construction of intercorrelation coefficients (ICCs) and through discussion of standard HLM regression coefficients. The work performed here, coupled with previous work, lead to a new MIDFIELD discussion of what grade distribution variations are observed at each of the partner institutions. We will be able to determine what effect the percentage of engineers enrolled in these first‐ and second‐year courses have been on section grade distribution. The effect of course size on grade distribution can also be studied. Finally, by employing the techniques of mixture modeling to examine clustering within our data and measures derived from information theory, we relate these results on section grade to fundamental outcomes in MIDFIELD, specifically graduation rates.

Citation
Format

Ricco, G. D. (2015, June), Beyond Sectionality and Into Sizeness, or How Course Size Effects Grades: An Exploration of the Multiple‐Institution Database for Investigating Engineering Longitudinal Development Through Hierarchical Linear Models Paper presented at 2015 ASEE Annual Conference & Exposition, Seattle, Washington. 10.18260/p.23619

TY  - CPAPER
AB  - Exploration of the Multiple‐Institution Database for Investigating Engineering Longitudinal Development                         through Hierarchal Linear Models (HLMs) and clustering            The MIDFIELD database includes complete student records of twelve institutions offering engineering degrees in the United States with a total student record of more than 1,000,000. These institutions enroll more than twelve percent of the nation’s total engineering students at any given time.  While researchers have explored MIDFIELD using regression analysis, thus far only single‐level methodologies have been used.  Hierarchical Linear Models, (sometimes called multi‐level models, nested models, or generalized mixed models,) provide a unique interpretive tool to probe a database such as MIDFIELD.  Unlike ANOVA analysis variants, HLMs allow for robust analysis of incomplete data sets.  For instance, in the case of a database of student grades, ANOVA methods require complete (or duplicate) recording of grades at each interval used for analysis. In other words, if one student has a missing grade for one semester, then that student cannot be used in the analysis.  HLMs do not require completeness for strict convergence. More importantly, HLMs retain the nested structure of the data itself through the analysis. This aforementioned nested data structure may not necessarily be known a priori. Techniques of cluster analysis can be used to identify the presence of particular partitions in an arbitrary data set. The utilization of such methods can be paired with HLMs to provide a powerful, data driven framework for analyzing this phenomenon.           In an expansion of a previous study, we focus MIDFIELD upon first‐ and second‐ year courses undertaken by most core curricula for our analysis: introductory physics; calculus; chemistry; computer languages; statics; and others. We then use a form of the null model in HLM in order analyze student grades. The null method allows us to glean an understanding of the variation of section grade distribution in these required courses through the construction of intercorrelation coefficients (ICCs) and through discussion of standard HLM regression coefficients. The work performed here, coupled with previous work, lead to a new MIDFIELD discussion of what grade distribution variations are observed at each of the partner institutions.  We will be able to determine what effect the percentage of engineers enrolled in these first‐ and second‐year courses have been on section grade distribution. The effect of course size on grade distribution can also be studied.  Finally, by employing the techniques of mixture modeling to examine clustering within our data and measures derived from information theory, we relate these results on section grade to fundamental outcomes in MIDFIELD, specifically graduation rates.  
AU  - George D. Ricco
CY  - Seattle, Washington
DA  - 2015/06/14
PB  - ASEE Conferences
TI  - Beyond Sectionality and Into Sizeness, or How Course Size Effects Grades: An Exploration of the Multiple‐Institution Database for Investigating Engineering Longitudinal Development Through Hierarchical Linear Models
UR  - https://peer.asee.org/23619
DO  - 10.18260/p.23619
ER  -