For questions relating to missing data problems, which can involve special data structures, algorithms, statistical methods, modeling techniques, visualization, among other considerations.
When working with data in regular data structures (e.g. tables, matrices, arrays, tensors), some data may not be observed, may be corrupted, or may not yet be observed. Treatment of such data requires additional annotation as well as methodological considerations when deciding how to impute or use such data in standard contexts. This becomes a problem in data-intensive contexts, such as large statistical analyses of databases.
Missing data occur in many fields, from survey data to industrial data. There are many underlying missing data mechanisms (reasons why the data is missing). In survey data for example, data might be missing due to drop-out. People answering the survey might run out of time.
Rubin classified missing data into three types:
- missing completely at random;
- missing at random;
- missing not at random.
Note that some statistical analysis is only valid under certain class.
