Mindsets and the learning environment: Using data science to unlock relationships between mindsets, college success, and context
In early 2017, the Mindset Scholars Network launched a new interdisciplinary initiative, called Mindsets and the Learning Environment, to explore how school and classroom environments shape students’ mindsets about learning. With funding from the Raikes Foundation, Overdeck Family Foundation, and the Joyce Foundation, the project’s aim is to rapidly generate scientific evidence about how schools and educators at all levels can convey messages to students that they can grow their ability, that they belong and are valued at school, and that what they are doing in school matters.
Eight research projects have been launched as part of the initiative. Seventeen different Network scholars are participating along with over a dozen external collaborators. The projects span a wide range of topics, from exploring how teacher practices cultivate learning mindsets and identity safety in K-12 classrooms, to the relationships between learning mindsets and neural processes throughout adolescent development.
This is a continuation of a series of eight posts in which we will hear from the leader of each research project to find out more about the questions they are exploring, what they are learning, and how their work is advancing the field of mindset science.
The next project we’re highlighting, A big biodata approach to mindsets, learning environments, and college success, is led by Mindset Scholar Sidney D’Mello. This project explores how students’ mindsets about intelligence and other psychological factors (e.g., grit, self-control) predict college success and the ways in which home, school, and neighborhood environments may influence these relationships.
Who are the members of the research team?
Sidney D’Mello is an associate professor at the University of Colorado Boulder and an expert in computer science and psychology. The team also includes Mindset Scholar Angela Duckworth, along with Margo Gardner, Stephen Hutt, Parker Goyer, Donald Kamentz, Chad Spurgeon, and Abigail Quirk. This group includes experts in computer science, data analysis, psychology, human development, and education.
As Sidney explained, having practitioners on the team helps ensure that they’re answering the “burning questions” of practitioners in the K-12 educational setting while contributing to foundational scientific knowledge.
“Instead of asking only what predicts [college success], we’re asking what predicts, how much does it predict, how generalizable is the prediction, and what can we do about it.”
About the data
This project uses data from the combined Common Application / National Student Clearinghouse dataset, a 6-year longitudinal dataset containing college applications from a national sample of about 300,000 U.S. students who enrolled in college in 2009. This dataset includes dozens of factors of students’ environments, experiences, and achievements, including family information (e.g., parent’s educational level), academic information (e.g., grades and standardized tests scores), and extracurricular activities (e.g., number of activities in Grade 12).
Data from a separate group of 1,700 high-school seniors was also analyzed to validate how psychological factors were coded in the current project. These students completed measures of growth mindset, grit, and self-control while their teachers provided similar ratings for their students. The researchers also obtained report card grades from school records.
What is the purpose of the project and how will it fit into the field of mindset science?
Creating a clearer picture of psychological factors that contribute to college success
While earning a college degree is becoming a requirement for more and more jobs, only about 40 percent of first-time, full-time, U.S. college students persist through school and graduate with a bachelor’s degree on time. What factors predict which students are more likely to graduate?
Traditional research on college completion has focused on factors such as socioeconomic status and high school academic preparation, but less is known about how psychological factors predict college outcomes. One of the main goals of this project is to explore how these factors, including learning mindsets, might increase students’ likelihood of graduating from college. Another key goal is to ascertain how successfully one can predict college success from biographical data and identifying the key predictors.
How do home, school, and neighborhood environments influence the way psychological factors shape college outcomes?
Because of the researchers’ commitment to providing practical recommendations to schools, they are not only exploring the factors that predict the likelihood of graduating from college, but also how contexts influence these factors. The team is exploring students’ home, school, and neighborhood environments to provide a more nuanced look at which factors predict what outcomes for what types of students and in what contexts. For example, while it has been established that participation in extracurricular activities is linked to college success, all students do not have the same opportunities to engage in these activities. However, if researchers are successful in identifying patterns in extracurricular participation that lead to success in specific contexts, they can make much more precise recommendations for extracurricular programming tailored to the properties of students’ environments.
Using novel data modeling approaches to explore how growth mindset and other psychological factors predict college success
One of the key contributions of this project to mindset science is the unique way the team is conducting its analyses. Sidney compares the team’s data analysis methods to different ways to predict weather outcomes. Traditionally, weather models have used theoretical principles from meteorology to build top-down models to predict what will happen in the future (e.g., when storms will happen and their severity). This is similar to the way traditional social science research models attempt to predict outcomes, using previous research and theory to generate hypotheses and then using outcomes to either confirm or refute their predictions.
By contrast, weather forecasting that leverages “data science” collects a vast amount of historical information on air, temperature, and humidity, and assesses the relationship between those measurements and actual past events, such as whether or not a storm occurred. Then, based on the historical data and past trends, scientists build a bottom-up forecasting model.
For this project, the team is combining both approaches. While not withholding the power of big data, they exercise restraint to avoid spurious findings. One important way they do this is by ensuring that their data-driven representations are grounded in theory and prior research. Another critical way is to design the models to generalize. This is done by using half the sample (~150,000 college applications from 2009) to explore common patterns in the experiences of successful college graduates when they were in high school. The findings are incorporated in a powerful predictive model that they are applying to the second half of the college applicants. If the model is accurate and generalizable, its predictions about which of this second group of students were ultimately successful in college should match those students’ actual graduation rates.
As Sidney explains, the team is interested in three ways of looking at how learning mindsets predict college success. “Instead of asking only what predicts,” Sidney said, “we’re asking what predicts [college success], how much does it predict, and how generalizable is the prediction.”
The results thus far indicate that the advanced models are quite accurate at prospectively predicting four-year college graduation, handily outperforming traditional models (i.e., about 16% better). The team is currently focusing on a subset of students where known factors, such as sociodemographics and test scores, made incorrect predictions with an eye for better understanding these specialized cases.
What are the next steps for the project?
The research team is continuing to explore the way their models predict college outcomes and how these relationships may vary for different subgroups of students. Additionally, the team is exploring how they can code the qualitative data in students’ 150-word responses describing a significant extracurricular activity or work experience. The idea is that these open-ended responses will provide a deeper understanding of students’ learning mindsets and motivation to better predict college success.