We’re excited to announce that pymetrics has been acquired by Harver!
Read MoreFor the latest blogs, please visit Harver.
pymetrics’ Head of Policy + I/O Science, Kelly Trindel Ph.D., discusses how AI can help optimize how you use good people data.
Watch the full session here.
When using data to make people decisions, there are a few things to consider. First, it’s important to build custom assessments using bidirectional measures. This is because soft skills are not inherently “good” or “bad”, meaning while a candidate may not fit one role, they will fit another. Secondly, we must utilize data sources distributed across demographic groups. Leveraging soft skills for candidate assessments allows for a more equitable representation of the applicant pool, as portrayed by the graph below:
Another consideration we must make is to proactively check for bias and choose the least biased model for evaluating candidates. Optimize your model for both fairness and efficiency. Relying on this type of data means there is less potential to be biased by human observations. Lastly, validate and monitor. It is crucial to understand what is being measured, and how it is related to the work you are doing. Always look for ways to improve your candidate evaluation.
Validity means that the assessment or evaluation is accurately measuring what it is designed to measure. The first pillar of understanding what is being measured is comprehending the data points being utilized in the algorithm and the constructs they represent. The second pillar to validity is understanding how the test is related to the job. Connecting data to the job function should be rational and straightforward. The third and final aspect of validation is to document changes and outcomes, and verify over time.
Below are some best practices for both maximizing the benefits of good people data and reducing any potential harm:
1. Optimize for fairness - choose the least biased option that aligns with hiring needs
2. Demonstrate job relevance - consider what is being measured and how it assesses job-relevant criteria
3. Inform candidates - candidates should be privy to how the assessment relates to the position they are applying for
4. Audit outcomes - the outcomes should be constantly audited and improved if need be
Hiring decisions have a significant impact on both your organization and the applicants. That is why it is crucial to choose a HR tech provider that offers internal processes that uphold your ethical responsibility. So what should you verify with your vendor? Ensure the provider has fairness in training data. The data should be trained to assess for fairness across demographic groups. As for the algorithms, they should be transparent and tested for bias, both pre and post deployment. In addition, the manner in which the data is used in relation to the job should be validated by subject matter experts. All providers should be including user consent and data ownership, as well as external auditing.
To watch this session on-demand, please click here.
