4.7 Conclusion

The presence of missing values provides a barrier to getting analysis started, and if the proportion of missing is high, imputation can be unreliable. Restricting to only complete data might produce an analysis using little data. This paper has presented a new data structure, several new visual techniques, and an algorithm for handling large amounts of temporal missings. These tools facilitate exploring and understanding missing value patterns, and diagnosing the imputations in preparation for time series modeling. These are available in the R package, mists.

There are some natural next directions of this work. The literature review uncovered a lack of a comprehensive system for simulating different types of missing value patterns. Conceptualizing and developing a system would greatly help in understanding what is seen in practice. Much of the early work on exploratory methods for missings provided interactive graphical methods. The plots developed in this paper could be adapted and extended to incorporate interaction, as new technology arises that makes this easier.