|
Computational Modeling: Beyond Regression in Evaluation
|
| Presenter(s):
|
| Nicole Cundiff,
Southern Illinois University at Carbondale,
karim@siu.edu
|
| Nicholas G Hoffman,
Southern Illinois University at Carbondale,
nghoff@siu.edu
|
| Alen Avdic,
Southern Illinois University at Carbondale,
alen@siu.edu
|
| Abstract:
Model comparison will be conducted comparing multiple regression models to neural network backwards propagation models using existing evaluation data in order to assess the predictive ability of both types of models. The comparisons being made will attempt to find better ways to model predictive relationships in evaluation data that could be used by clients to assist in decision making. Currently, regression models are the most popular way to test prediction; however it is a strict modeling technique that requires many assumptions to be met, including linearity and homoscedasticity. Feed forward backwards propagation networks do not have to meet these restrictions as do parametric tests, and therefore give more accurate and generalizable models from the data. Problems with interpretation of data will be discussed as well as information on the use of neural networks in evaluation. Additionally, a brief discussion on how to explain findings to clients will be included.
|
|
Social Network Analysis for Evaluation: Open and Closed Approaches
|
| Presenter(s):
|
| Carl Hanssen,
Hanssen Consulting LLC,
carlh@hanssenconsulting.com
|
| Abstract:
This paper is based on work from two evaluations of Math Science Partnerships funded by National Science Foundation (NSF). One goal of the Milwaukee Mathematics Partnership (MMP) is to strengthen school-based networks and build distributive leadership in schools. Similarly, a goal of the Life Sciences for a Global Community (LSGC) teacher institute is to develop a national network of high school teacher leaders.
Both evaluations incorporate social network analysis as a tool for exploring network development. This review contrasts the open and closed approaches used in the MMP and LSGC projects, respectively. An open approach allows participants to identify any individual who they consider part of their professional network—doing so enables monitoring the expansion of networks over time. The closed approach limits network participants to a defined list of individuals and enables monitoring of network changes for a specific set of individuals (e.g., institute participants).
|
| |