The Bias Detection Report makes Bias Detection queries for a selected set of variables across all clusters in a model and returns the results in an easy to digest interface for review. The Bias Detection Report is accessed through the Model Actions page for any Supervised Clustering Model.
Generating a Bias Detection Report
Starting the Bias Detection Report requires the selection of several parameters:
Inter-cluster Alpha: The threshold value under which a test’s p-value will be considered significant for testing the distribution of the test Variable between clusters.
Intra-cluster Alpha: The threshold value under which a test’s p-value will be considered significant for testing the distribution of the test Variable within a cluster.
Variable: The data elements which will be tested for correlations in the test.
Visualization: The Cluster Visualization whose Cluster Labels will be displayedin the test results.
When all parameters have been selected, click on Generate Report to initiate the process.
The report may take several seconds to generate depending on the number of clusters in the model and the variables selected for testing.
Reading a Bias Detection Report
The report starts with a metadata summary of the Model tested and on Report generation.
The first chartgraphs the p-values of all tests executed, sorted by the variable tested, on a logarithmic scale.
Alpha-value thresholds are drawn on the chart and labeled. Values below the inter-cluster alpha are labeled “Elevated Risk”. Values below the intra-cluster alpha are labeled “Critical Risk.”
The space below the results graph has a tabbed selection to browse detailed test results and find information on the tests.
Summary of Findings:
Under this tab you will see a table with the test results for each cluster/variable combination.
How To Interpret Findings:
Here you will see a description of how to read the results of each test.
Overview of Tests:
This is where you will find a description of each of the statistical tests used in the Bias Detection Report.
This shows a summary of findings for each cluster, followed by the detailed results of each of the statistical tests executed in the Report.