Insights in Experimental Data

Empirical researchers turn to statistics to assist them in drawing conclusions, also called inferences, from their collected data. Often, this data is experimental data, i.e., it consists of (repeated) measurements collected in one or more distinct conditions. The observed data can hence be summarized into histograms that specify how frequently measured values occur in the distinct conditions. The purpose of statistical analysis can therefore be reformulated as characterizing or modeling the change in such histograms across conditions. While existing statistical programs (such as SPSS or R) offer a wide range of statistical methods for studying and characterizing such changes, they assume familiarity with statistical terminology and offer little or no insight into how statistical methods work and into the (model) assumptions they make. This lack of insight can lead to erroneous use of such methods, however.


We propose that it is possible, and even advantageous, for users to understand up to a certain degree the statistical modeling that is applied to their data. This insight can of course not be based on an understanding of the mathematical algorithms involved in those statistical analyses. The claim is instead that insight can be accomplished through well-chosen visualizations of both the data and the models used to represent the data, especially if users can interactively explore such visualizations. In order to validate this proposed approach, it was necessary to develop an entirely new program for interactive statistics, called ILLMO (Interactive Log-Likelihood MOdeling). Run-time versions of the program for both Mac OS and Microsoft Windows can be downloaded, together with some supporting material, from the project website http://illmoproject.wordpress.com.