Variance is a normalization metric which is useful when analyzing a check with multiple submissions. This will help you understand the level of disparity between submissions, this metric is mostly useful when using group similarity checks.
Let's take a class of 100 students for example, when uploading 100 submissions and running similarity checks, their respective peer scores will range between 0-100%. The highest peer score is taken as the submission score, then variance between all submissions is calculated (taking the maximum score subtracted by the minimum score in your group). We estimate based on this the level of concern in regards to code similarity displayed through the graph above.