Research summary
Quantitative analysts must make many decisions when estimating models. Some decisions are easy because they are guided by theory and data. But others are more difficult. For example, how should missing data be handled? Should robust standard errors be used, and if so, what kind? Scholars have tried to answer these questions by using simulations to uncover the properties of different models under a range of conditions. But these simulations can never capture the diversity of contexts, models and data that analysts face.
Dr. Dave Armstrong, Canada Research Chair in Political Methodology, is developing a new statistical framework for making these challenging decisions. He and his research team will use this framework to create custom simulations in R, an open-source statistical computing language. These simulations will help analysts determine the best methods to solve missing data problems, handle heteroskedasticity (the unequal variance of errors in a regression model), determine the strength of instrumental variables, and establish the appropriate level-2 degrees of freedom in multilevel models.