Interactive Statistical Modeling with XGms

Abstract

This chapter intends to clarify how statistical models can be used in the interactive analysis of experimental data. It is discussed how maximum likelihood estimation can map experimental data into model parameters, and how confidence intervals on these parameters play a key role in drawing inferences. One frequent inference is whether or not similar data collected in two different experimental conditions require a significant change in model parameters, hence indicating that the conditions are not equivalent. Another frequent inference is whether a simple or a more complex model needs to be adopted for describing observed data. It is argued that the user, who has insight into the problem underlying the data, should be allowed to play an active part in selecting models and specifying the inferences of interest. Hence, the need for interactive visualization is identified and the program XGms that has been developed for this purpose is discussed. The fact that the statistical models implemented within XGms are geometrical in nature simplifies their visualization and interpretation.