What are some effective ways to visualize data with missing values?

One interesting aspect of visualizing missing values is how it can help identify patterns and clusters of missingness in the data. Heatmaps and bar charts are effective techniques for visually representing the occurrence of missing values across different variables. These visualizations can help identify any potential relationships between missingness and other observed variables, allowing for a better understanding of the missing data mechanism. Moreover, visualizing missing values can aid in deciding which strategies, such as imputation or deletion, may be most appropriate for handling the missingness. For example, if there is a pattern of missingness related to other variables, regression imputation is a suitable method.

Missing values can be like a canary in the coal mine. I always review data prior to the missing data as those value could be suspect. If the missing data are time-stamped, that provides a starting point to examine other data collected at that time to check if there are any problems elsewhere.

Depending on how the missing data information is going to be used, I'd employ different techniques. If the missing values are to be corrected/filled, I'd use a symbol to highlight those data points in the visualization. If the missing values are to be reported, I'd create a metric and build a chart with it. If the users need to know the presence of missing values but are most interested in the data points that are actually present, I'd include a small note giving the presence and quantity of the missing data point. I'd also have a filter that allows to include or exclude the missing data points, and by default it would exclude those values. If the missing values are not relevant at all, I'd mask them from the vizualisation.

Missing values is not always a bad thing. Let's say you have an HR system that counts injuries. If that system generated a NULL for that month surely that's a positive thing. We should not be hasty to remove null values or outliers. First we should really take a solid grasp of the data and the business context behind it before deciding whether to explore removal techniques.