Thank you for your interest in our research, originally presented at ICML 2022: Workshop on Spurious Correlations, Invariance and Stability.
We plan to have our code cleaned up for open sourcing in the next month.
In the meanwhile, please check out our open source demo, where you can test almost any BERT-like model (hosted on Hugging Face) for selection bias induced spurious correlations for gender using your choice of input text, W
variable, x-axis values, and the plotted degree of fit at https://huggingface.co/spaces/emilylearning/spurious_correlation_evaluation
Please find the arXiv version of our paper at https://arxiv.org/pdf/2207.08982.pdf