Now You See Me Now You Don't: A Framework To Detect and Reduce Proxy Capacity
Organizations: Upstart Network, Inc., USA
Abstract
In regulated domains such as lending, the two most salient fairness concepts are disparate impact and disparate treatment: disparate impact is generally defined by differential rates of adverse outcomes between demographic groups, while disparate treatment is generally defined by the use of demographic information or proxies to issue decisions. There already exist established methods to quantitatively detect and mitigate disparate impact; however, there has been less research on doing the same for disparate treatment. From a practitioner's perspective, the manner in which a conditional independence test is formulated and operationalized to identify proxies is nuanced and underexplored. As Machine Learning (ML) systems increase in scope and application in regulated domains, practitioners need a defensible framework to identify proxies at scale. Current disparate treatment frameworks typically rely upon some combination of quantitative and qualitative review of models' inputs to identify proxies. Features that are identified as proxies are then typically removed from the model in practice, and predictive performance degrades in favor of fairness through unawareness. Here, we propose a two-part framework that reduces the disparate treatment risk of a feature and keeps the predictive performance that removal of the feature loses. First, we propose a quantitative test for proxy capacity that identifies whether some of a feature's values reveal membership in a particular demographic group. Second, we propose a pre-processing methodology that adjusts those proxy features to reduce their proxy capacity. We show empirically on synthetic and public datasets that this framework identifies features that proxy for demographic information and reduces their proxy capacity while the adjusted feature retains task-relevant signal (i.e. improves predictive performance versus removal).
Figures & tables
| untouched | soft proxy (MLP) | soft proxy (XGB) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AUC over removal | AUC over removal | AUC over removal | ||||||||
| proxy | ||||||||||
| 0.8 | 0.39 | +0.015 | +0.065 | 0.25 | +0.013 (88%) | +0.056 (86%) | 0.22 | +0.007 (45%) | +0.060 (91%) | |
| 1.2 | 0.54 | +0.009 | +0.059 | 0.26 | +0.002 (21%) | +0.046 (77%) | 0.24 | +0.006 (61%) | +0.060 (102%) | |
| 1.6 | 0.66 | +0.013 | +0.058 | 0.43 | +0.006 (45%) | +0.050 (86%) | 0.26 | +0.009 (71%) | +0.052 (90%) | |
| 2 | 0.83 | ✓ | +0.012 | +0.051 | 0.45 | +0.012 (104%) | +0.045 (88%) | 0.24 | +0.008 (67%) | +0.043 (85%) |
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
| distributional proxy test | aggregate measures | ||||||
| candidate value | implied BF | MI | V corr dcor | AUC | |||
| sex Male ( , ) | |||||||
| relationship Husband | 19,716 | 1.000 | 1.000 | 2,502 | 0.225 | 0.58 | 0.80 |
| marital-status Married-civ-spouse | 22,379 | 0.889 | 0.886 | 3.8 | 0.101 | 0.43 | 0.73 |
| relationship Wife | 2,331 | 0.001 | 0.000 | 0.00 | 0.055 | 0.32 | 0.57 |
| sex Female ( , ) | |||||||
Algorithmic Fairness
Algorithmic fairness studies how automated decisions distribute errors, opportunities, and harms across groups, and how to measure and reduce disparities.
See all 455 papers