Towards Data-Driven Affirmative Action Policies under Uncertainty

Corinna Hertweck,Carlos Castillo,Michael Mathioudakis

DOI: https://doi.org/10.48550/arXiv.2007.01202

2020-07-02

Abstract:In this paper, we study university admissions under a centralized system that uses grades and standardized test scores to match applicants to university programs. We consider affirmative action policies that seek to increase the number of admitted applicants from underrepresented groups. Since such a policy has to be announced before the start of the application period, there is uncertainty about the score distribution of the students applying to each program. This poses a difficult challenge for policy-makers. We explore the possibility of using a predictive model trained on historical data to help optimize the parameters of such policies.

Computers and Society,Machine Learning

What problem does this paper attempt to address?

This paper aims to solve the problem of how to design effective affirmative action policies in the university admission system in the presence of uncertainty, in order to increase the admission rate of students from under - represented groups. Specifically, since these policies need to be announced before the start of the application period, there is uncertainty about the score distribution of the applying students, which poses a challenge to policy - makers. The paper explores the possibility of using prediction models trained on historical data to optimize the parameters of such policies.

### Main research questions
1. **Affirmative action policy design under uncertainty**: How can affirmative action policies be designed to effectively increase the admission rate of under - represented groups without knowing the specific information of the applying students?
2. **Application of prediction models**: How can prediction models trained on historical data be used to optimize the parameters of affirmative action policies to reduce the admission rate gap?
3. **Policy effect evaluation**: How can the effects of affirmative action policies under different strategies be evaluated, especially in terms of the balance between reducing the admission rate gap and maintaining the overall admission quality?

### Research methods
- **Data sources**: The paper uses data from the University of Chile admission system, covering all applicants and university programs from 2004 to 2017.
- **Prediction model**: A multi - label probability classifier was constructed to predict the application behavior in 2017. The model was trained on 2016 data and generated multiple possible application sets.
- **Benchmark method**: Directly use historical data to calculate the best affirmative action policies in the past few years.
- **Objective function**: An objective function was defined, combining the equality of admission rates (SPD) and the average grades of admitted students (utility). The form of the objective function is:
\[
o_b = (\mu_0 - \mu_b) + \lambda\cdot|SPD_b|, \quad \lambda\geq0
\]
where \(\mu_0\) is the average grade when no affirmative action policy is implemented, \(\mu_b\) is the average grade when an affirmative action policy is implemented, \(\lambda\) is a weighting parameter, and \(SPD_b\) is the statistical fairness difference after implementing the affirmative action policy.

### Experimental results
- **Overall effect**: The method based on the prediction model performs better in reducing the admission rate gap, with smaller errors and less fluctuation.
- **Application in specific projects**: For projects that have historically had unequal admission rates, the prediction model method also shows better results.
- **Advantages of conservative strategies**: The prediction model method is more conservative and can avoid the problem of over - correcting the admission rate gap.

### Conclusions
By comparing the method based on the prediction model and the method based on historical data, the paper finds that the prediction model method achieves a better balance between reducing the admission rate gap and maintaining the overall admission quality. Although the method based on historical data may also be effective in some cases, the prediction model method has more advantages in dealing with uncertainty. Future research can further explore the impact of the announcement of affirmative action policies on students' application behaviors.