I am passionate about fulfilling the potential of applied statistical models and quality data visualization to inform policy makers and practitioners and improve the effectiveness of services - particularly in education. Since 2010 I have been exploring a number of different techniques to achieve these goals focusing on the methodology that best fits the needs of the audience. While working at the Wisconsin Department of Public Instruction, I have led a number of projects exploring this nexus of policy, practice, and evidence.
Major projects have included:
- Developing and deploying a statewide Dropout Early Warning System (DEWS) in Wisconsin
- Building training material in the R open-source statistical software DPI R Bootcamp
- Assembling a diverse set of techniques for analyzing administrative data
- Applying classification models to administrative records to determine natural groupings and patterns
- Visualizing education data in unique ways to explore temporal, spatial, and correlational relationships For public examples of this work please visit the Presentations page.
I also believe in building techniques that scale. A big part of this is building partnerships and sharing work. It is this problem of scale that drives my interest in open platforms and open science. This includes developing shared open source tools on GitHub, building communities of practice within and across state lines, and professionalizing the analysis of administrative data for use by practitioners.
Leveraging administrative records to inform policy and practitioners involves investigating the appropriate methods that meet the needs of the users. Analytical methods are about the producing information from data, and the method chosen helps determine what information can be provided. I am continually searching for methods that meet those needs. Linear regression is not the beginning and end of data analysis and has many limits that policymakers and practitioners inherently understand. Currently I am very interested in investigating the application of the following methods:
- Latent variable analysis including latent class analysis and structural equation modeling
- Bayesian data analysis that includes incorporating the judgment of subject matter experts with data
- Predictive modeling techniques including SVM, neural networks, and other classification algorithms
- Mixed effect models that incorporate nested and crossed random effects
- Variance decomposition
Once the analysis has been conducted, the challenge turns toward presenting the results in a way that is meaningful and actionable by educators. All too often important findings are buried in technical jargon, Greek letters, and incomprehensible tables filled with stars. Practitioners need visualizations that are engaging, meaningful, and make the complex workings of the analytical model accessible. I am interested in exploring the following approaches to this problem:
- Simulation studies to dynamically explore model parameters
- Multi-model inference to adjust models to different questions
- Interactive displays of model results that can be controlled by intuitive UI elements
A final piece of this puzzle is building partnerships around key initiatives. This includes building partnerships within governmental entities and partnerships between governmental entities and external organizations. Selecting and building partnerships requires extensive communication and genuine commitment - but such work is vital. Quality partnerships allow partners to achieve economies of scale and more comprehensive investigations of problems and solutions. I currently participate in a number of such partnerships and am committed to identifying such opportunities and pursuing them in alignment with the needs of policymakers and practitioners. The newest and biggest such partnership is DATA-COPE, an organization of state and local education agency data analysts across the United States.
Look for more information in the future as more of these projects roll out. Stay tuned!