Dr. Jennifer Gibson has over fifteen years of experience conducting social science research for government and private sector clients. Jen has conducted market research studies; evaluated selection and assignment systems; and conducted research on recruitment and worker well-being. As Vice President of Advanced Analytics, Jen and her team consult with other divisions within Fors Marsh Group on research involving human subjects; data management; appropriate estimation, interpretation, and presentation of statistics; complex survey design and analysis; data science and machine learning techniques; questionnaire design and psychometrics; quantitative and qualitative research methods; experimental design; and research synthesis. She also supervises Research and Development efforts and business process management.
We get excited about living our values at Fors Marsh Group (FMG), whether it’s engaging with our B Corp community or applying ourselves to critical social issues such as exercising the right to vote, improving worker safety, and promoting healthy behavior. FMG employees are also passionate about blending social science and data analysis to solve real-life problems so, naturally, we jumped at the chance to have it all and compete in the 3rd Annual Society for Industrial-Organizational Psychology (SIOP) Machine Learning Competition.
This year’s SIOP Competition posed a classic problem to the competitors: Create an open-source science solution that simultaneously balances employment fairness and staffing needs. Since employers are always looking to fill jobs with the strongest candidates, and it’s illegal for U.S. employers to discriminate based on an applicant’s race, gender, religion, national original, physical or mental disability, age, sexual orientation, or gender identity, the SIOP competition posed a real issue that organizations face in hiring. FMG seized the opportunity to tackle this real-world challenge, assembling a multidisciplinary team of data enthusiasts with expertise in psychology, computer science, statistics, sociology, and data science. In the spirit of FMG’s philosophy to use the right tool for the job, we used R and Python—two popular science tools for machine learning and data science.
Our mission was to use machine learning to recommend which job applicants should be hired for entry-level retail positions from a pool of candidates. SIOP shared multiple incomplete data sets for Team FMG to draw from to create a full data science solution to solve the problem, as well as descriptions of the data and how they were collected. For some applicants, we were equipped with information about how they answered questions on the pre-employment tests. For others, our team knew whether they were hired, how well they did on the job, whether they were a member of a group protected from discrimination by federal law, or whether they had already left the job. We had only one example question for each section of the pre-employment tests, and we were not given the correct answers to the test questions. The first question we asked each other was, “What should we do with all this data”? Which was quickly followed by, “And what should we make of the unknowns?”
Not unlike our daily work using data science to help clients take on some of society’s most pressing issues, Team FMG’s approach was to pore over the scenario description, brainstorm as a group to come up with a plan, explore the data, test different methods, and repeat. We used some tried-and-true analyses, such as predictive models, as well as modern data science tools, like XGBoost, in our efforts to create a system that would minimize selection bias and would successfully identify the best job candidates. Each team could submit multiple sets of hiring recommendations during each stage of the competition, continuously working to top their previous best score.
For several months, Team FMG held the top position in the SIOP public leaderboard. By the time we entered the last days of the competition, our team had followed many leads we found in the data, debated being led by theory versus letting the data speak for itself, and had gotten to know the scenario by heart. With great anticipation, and a little relief, we submitted our final hiring recommendations and began our wait for the final standings.