Board 411: Training Socially Responsible and Engaged Data Scientists: Lessons from Four Student Cohorts

Valentina Kuskova, Nitesh V. Chawla, Sugana Chawla, Danielle Wood, Ann-Marie Conrado2024 · DOI

TL;DR

The Community-Engaged Educational Ecosystem Model (C-EEEM) successfully bridges the gap between STEM education and local community needs by placing students in immersive, project-based internships. Data from four cohorts across three Midwestern cities shows that this model significantly boosts STEM confidence and regional "place attachment," with the most profound impacts seen in high school students and underrepresented groups.

What Problem Does This Paper Address?

Legacy industrial cities in the American Midwest face a dual challenge: a "brain drain" of educated workers and a persistent lack of diversity within STEM fields. Traditional engineering education often feels disconnected from real-world social issues, which can lead to lower retention rates, particularly among women, underrepresented minorities (URM), and students from low socio-economic status (LSES) backgrounds.

The authors address whether a community-engaged "ecosystem" can simultaneously improve technical skills, professional identity, and the desire for students to remain in and contribute to their local regions. They explore how to replicate a successful pilot program (at the University of Notre Dame) across different institutional contexts (University of Louisville and Youngstown State University).

Methods

The study evaluates the Community-Engaged Educational Ecosystem Model (C-EEEM), which is structured as a hierarchy of four nested layers:

  1. Collaborative Infrastructure: Sustained networks between universities and community organizations.
  2. Community-Identified Projects: Real-world challenges (e.g., affordable housing, sustainability) defined by local partners.
  3. Supportive/Diverse Learning Environment: A network of diverse mentors and targeted recruitment.
  4. Immersive Internship: The core experience where students work with high autonomy on interdisciplinary teams.

To measure impact, researchers used a retrospective pre/post survey design. Unlike a standard pre-test, this approach asks students after the internship to reflect on their abilities both before and after the experience. This is considered more sensitive for measuring "dispositional shifts" in unfamiliar settings. Data was analyzed using Paired-Samples T-Tests to determine statistical significance and Cohen’s D to measure effect size.

Datasets

The paper aggregates two years of implementation data (2022 and 2023) from three distinct sites:

  • South Bend-Elkhart, IN: The original pilot site (University of Notre Dame).
  • Louisville, KY: A replication site (University of Louisville).
  • Youngstown, OH: A replication site (Youngstown State University).

The dataset includes survey responses from a diverse pool of high school and college students, categorized by gender, race, socio-economic status, and education level.

Key Insights and Ideas

  • The "Gateway" Effect: The program is most transformative for high school and early college students. For these younger cohorts, the C-EEEM acts as a "gateway" experience that satisfies the psychological needs of autonomy, relatedness, and competence.
  • Equity-Driven Impact: While the program benefited all participants, the effect sizes were consistently larger for underrepresented groups (women, URM, and LSES). This suggests that community-engaged learning is a powerful tool for closing the "confidence gap" in STEM.
  • Place Attachment as a Metric: Success isn't just measured in code written or data cleaned; it’s measured in "place attachment." The program successfully convinced students that they could have a meaningful career and life within their home regions.
  • Scalability through Flexibility: Replication is not about "copy-pasting" a program. Each site maintained a unique approach based on local institutional supports (e.g., Youngstown’s partnership with a redevelopment non-profit), yet the core elements of the model produced similar positive impacts.

Key Results

The quantitative results showed "highly statistically significant" changes across all major constructs:

  • STEM Confidence: The overall effect size for applying learning to real-world challenges was Large (Cohen’s D = 1.002).
  • Women in STEM: Showed particularly large gains in data analysis confidence (0.847) and overall technical self-efficacy (1.10).
  • URM Students: Demonstrated a large effect size in feeling a connection to their specific region (0.871) and understanding how positive change happens in communities (0.890).
  • High Schoolers: This group saw the highest gains, with a massive effect size of 1.287 for understanding community change and 1.19 for confidence in applying technical skills.

Limitations and Future Work

The authors acknowledge that implementation differences between sites (like staff transitions or varying levels of institutional support) can influence outcomes. Currently, the South Bend pilot site still disproportionately influences the aggregate data due to its maturity.

Future Work: * Standardized Curriculum: The researchers plan to implement common projects across all three sites (e.g., measuring tree canopy or walkability) to hold the "curriculum" constant. This will allow them to better isolate how different institutional cultures and regional settings affect student outcomes. * Long-term Tracking: While the paper shows immediate post-internship shifts, long-term longitudinal data would be needed to confirm if these students actually stay in the region or persist in STEM careers over several years.

Summary updated

Abstract

Her expertise is advanced data analytics, especially time-series analysis and networks.Kuskova has extensive corporate and
BibTeX
@misc{kuskova2024board,
  author = {Valentina Kuskova and Nitesh V. Chawla and Sugana Chawla and Danielle Wood and Ann-Marie Conrado},
  title = {{Board 411: Training Socially Responsible and Engaged Data Scientists: Lessons from Four Student Cohorts}},
  year = {2024},
  doi = {10.18260/1-2--47000},
}

Loading related content...

Figure