Visual Data Exploration as a Statistical Testing Procedure: Within-View and Between-View Multiple Comparisons.

A fundamental problem in visual data exploration concerns whether observed patterns are true or merely random noise. This problem is especially pertinent in visual analytics, where the user is presented with a barrage of patterns, without any guarantees of their statistical validity. Recently this problem has been formulated in terms of statistical testing and the multiple comparisons problem. In this paper, we identify two levels of multiple comparisons problems in visualization: the within-view and the between-view problem. We develop a statistical testing procedure for interactive data exploration that controls the family-wise error rate on both levels. The procedure enables the user to determine the compatibility of their assumptions about the data with visually observed patterns. We present use-cases where we visualize and evaluate patterns in real-world data.

Finding statistically significant high accident counts in exploration of occupational accident data.

INTRODUCTION: Finnish companies are legally required to insure their employees against occupational accidents. Insurance companies are then required to submit information about occupational accidents to the Finnish Workers' Compensation Center (TVK), which then publishes occupational accident statistics in Finland together with Statistics Finland. Our objective is to detect silent signals, by which we mean patterns in the data such as increased occupational accident frequencies for which there is initially only weak evidence, making their detection challenging. Detecting such patterns as early as possible is important, since there is often a cost associated with both reacting and not reacting: not reacting when an increased accident frequency is noted may lead to further accidents that could have been prevented. METHOD: In this work we use methods that allow us to detect silent signals in data sets and apply these methods in the analysis of real-world data sets related to important societal questions such as occupational accidents (using the national occupational accidents database). RESULTS: The traditional approach to determining whether an effect is random is statistical significance testing. Here we formulate the described exploration workflow of contingency tables into a principled statistical testing framework that allows the user to query the significance of high accident frequencies. CONCLUSIONS: Our results show that we can use our iterative workflow to explore contingency tables and provide statistical guarantees for the observed frequencies. PRACTICAL APPLICATIONS: Our method is useful in finding useful information from contingency tables constructed from accident databases, with statistical guarantees, even when we do not have a clear a priori hypothesis to test.