More on Elusion Testing in Relativity
The Tip of the Night for March 12, 2019 discussed elusion testing in Relativity. Elusion testing is performed with active learning to determine when assisted review (TAR) misidentifies relevant documents as being non-responsive.
After an elusion test is run, the results are accessed by clicking 'View Elusion Results' on the elusion card on the active learning tab. The results show three key stats:
1. Elusion Rate - the percentage of documents coded as relevant in the re-review of the elusion sample. Any documents skipped during the re-review will be regarding as relevant documents.
2. Eluded Documents - the estimated number of eluded documents, based on how representative the sample set was (the confidence level), and the percentage by which the elusion rate in the sample may vary from the full set (the margin of error).
3. Pending Documents - documents not submitted to the model, including those selected for the elusion sample, and user selected documents coded during the elusion test.
A choice must be made to either complete the project or resume the project. If the results are not acceptable the model will be unlocked and documents coded after the elusion test was run will be submitted to the model.
The Review Statistics section of Active Learning will list the results for each elusion test:
1. Rank cutoff - the predicted relevancy value
2. Discard Pile Size - uncoded documents below the rank cutoff.
3. Skipped - documents in the re-review that are not coded for the elusion test.