The (physics class) Social Network

Title: Educational commitment and social networking: The power of informal networks
Authors: Justyna P. Zwolak, Michael Zwolak, and Eric Brewe
First author institution: Florida International University
Journal: Physical Review Physics Education Research 14 010131 (2018)

Attracting and retaining science, technology, engineering, and mathematics (STEM) majors is a key aspect of the 21st century knowledge economy. One way to do that is to focus on improving the overall educational experience for students rather than just their classroom experiences. To do so however requires understanding how students are integrated into the university environment and how that may affect their persistence in their STEM major. Presently, about one half of first time students who leave a university by the end of their first year do not return to college at any point in their lives so retaining these students is of utmost importance. The goal of today’s paper is to examine students’ networks within their university to see how they may be related to whether the student persists in their degree program.

Before delving into the details of today’s paper, it is important to understand what is meant by network and how these networks can be characterized. In physics education research, networks are usually discussed in terms of social network analysis (SNA), which can be used to understand how people interact with each other. In the sample network diagram in figure 1, the dots could be people at a cocktail party and a line between two dots means that those two people talked at the cocktail party.

Sample network.png
Figure 1: A sample network diagram. Here, the dots represent people while the arrows represent interactions. The tail of the arrow represents who initiated the conversation. Thicker arrows mean more interactions. Person “1” has a large indegree, person “2” has a large outdegree, and person “3” has a small closeness. (PERbites original image)

While there are many metrics to describe the network, the one most relevant to today’s paper is centrality, which is a measure of how important or influential a dot (representing a person) is to the network. In our cocktail party example, the host would most likely have a higher centrality than a typical guest. The host’s centrality could be calculated in many ways such as how many people talk to the host (indegree), how many people the host talks to (outdegree), and how many degrees of separation the host has to every other person at the party (closeness).

In today’s paper, the authors used social network analysis to study who students in a Modeling Instruction (MI) course at FIU interacted with and how the development of their network may influence their persistence as measured by whether the student took the second semester Modeling Instruction course. Overall, of the 273 students enrolled in the first Modeling Instruction course in either fall 2015 or fall 2016, 212 of them continued (“persisted”) on to the second semester course.

To collect the necessary data, the researchers gave students in the course a survey asking them to list who they worked with in class, how often they interacted with each of those people during the current week (once, a few times, daily), and who (if anyone) they worked with outside of class. Both the inside and outside of class responses could include the professor, teaching assistants, and learning assistants while the outside of class responses could include parents, friends who previously took the class, or other people in their major. The data were collected five times during the semester, at roughly 2-3 week intervals. The researchers expected to find that the out-of -lass interactions would be able to create a model that was more predictive of whether a student would go onto the next Modeling Instruction course than just creating a model from in-class interactions and demographic information.

So what did they find? First, based on recent research, they decided to only use the network information available at week 8 in the semester (roughly the halfway point), since the networks tend to almost be fully developed by this time. The network is shown in figure 2.

Social network figure 2.PNG
Figure 2: Social network diagram based on results from the survey in week 2 (left) and week 8 (right). Purple dots are students while grey dots are people not enrolled in the MI course. The size of the nodes correspond to closeness at week 8.  The tail of the arrow is the person who wrote that they work with the person at the tip of the arrow. Figure 2 in paper.

Next, they decided to first see if the different types of centrality (indegree, outdegree, closeness , and a fourth called eigenvector) were predictive of whether a student would continue to the next Modeling Instruction course. They found that all four types of centrality were statistically significant predictors of persisting in the Modeling Instruction course sequence by creating simple logistical regression models.

Next, the researchers added the effects of grades to the model. Since a student who does well in the first course is likely to take the second course and a student who fails the first course cannot take the second course, the researchers split the students into three groups: high grade students (A, A-, B+), intermediate grade students (B, B-, C+), and low grade students (C and below), and built models for each of them. When combining final course grades with measures of in-class network centrality and out-of-class network centrality, the researchers found the best predictive model to be the one that used out-of-class closeness, out-of-class outdegree, and the final course grade.

While the researchers were able to make a predictive model, they wanted to also create a model that could be used mid-semester to identify students who may not persist in the course sequence. The researchers decided to swap final grades for midterm exam grades since these two grades are significantly correlated. However, when doing this for the intermediate grade students, only out-of-class closeness remained a significant variable in the predictive model. Finally, the researchers were able to find a threshold value of the closeness: when a student had a closeness of above 0.14, they had a 92% chance of continuing to the next course. However, when the student had a closeness below 0.14, they only had a 63% chance of passing the course.

While this research only looks at a single course at a single university, it does show that out-of-class interactions may be an important part of retaining STEM students. However, there are a few limitations to this study. First, FIU has a very high (~90%) percent of commuter students so this may explain why out-of-class interactions may appear so important. Second, the Modeling Instruction course is highly interactive so there is less variability in the in-class interactions, which may make the out-of-class interactions appear more significant. As most physics courses are still taught in a traditional manner, the authors note that this study needs to be extended to traditional introductory courses.

So what can we take away from this paper? First, that in Modeling Instruction, students without high or low grades are more likely to persist in the course sequence if they are immersed in the social system of the university and are successful at reaching out to others for help in their course. Second, this study suggests promoting out-of-class interactions may help increase persistence in course sequences, which in the long run, may help increase the number of STEM degrees awarded.

Figures used under Creative Commons Attribution 4.0 International.

Leave a Reply

Your email address will not be published. Required fields are marked *