I have spent a number of days over the past few weeks grappling with my network data, trying to figure out how to run models that will tell me (theoretically) whether teachers’ behaviors are shaped by their network affiliations. More specifically, I’ve been randomizing participant IDs, developing variables, and adding variables to an influence model syntax for SPSS that someone else wrote.
*insert screeching halt sound here*
WHAT!? Let’s back up.
There are a few different ways I am using network data. My friends and colleagues are familiar with my general obsession with network graphics, but that’s only a small part (and, honestly, a somewhat misleading and distracting part) of analyzing networks. Don’t get me wrong. It’s fun. Check out, for example, this network graphic, which depicts the technology advice network at my research site:
This one, in particular, shows who teachers nominated as individuals they “consult” about technology and teaching. Nodes (dots, which represent teachers) are sized by in-degree (number of nominations received) and are colored based on an attribute variable (number of devices used in the classroom). The darker blue the node, the more devices used in that teacher’s classroom.
Setting aside the academic for a moment, these graphics are cool (I’m obsessed with them). But they can also tell me a lot about the social dynamics at my research site. I can see who is in the “center” (according to their colleagues) of the school’s tech network. I can see who receives the most nominations in the network, which can tell me a lot about where perceived expertise lies in the school. I can use this to inform interviews during data collection, to help me interpret interview and observational data afterwards, and to examine other networks (for example, the “close colleagues” network in the school). And all of that applies in the opposite directions, too.
But this is the icing on a very complicated cake. Beneath these graphics lie layers upon layers of statistical analysis possibilities — these are the layers of the cake that I have not discussed with my colleagues or in my conference presentations because I’m still trying to understand them. I’m good at graphics, and I picked up the graph theory elements of network analysis quickly. But my stats are shaky, and I don’t have much experience grappling with data (not to mention grappling with the sheer volume of data I collected for my diss).
Besides these badass graphics, one can also “create models” with network data, by which I mean, they can develop equations that predict the potential for one’s behaviors to be shaped by their social interactions. OR, equations that will predict how particular behaviors might lead one to choose particular types of friends or collaborators. OR, equations that will predict how individuals within a particular network cluster together into groups. And the list goes on and on.
The value of these models? I admit, I was skeptical when I first learned about this element of network science. I wasn’t in it for the quantitative models — I was in it for the power it would give my descriptions and analysis of teachers on a qualitative level. I wanted to know more about the people, not reduce their behaviors and relationships to mere numbers and equations. But over the past few months to a year of doing this work, I am beginning to see how these quantitative data reveal trends that I might have missed in my observations and conversations with teachers at the school. And these trends may help schools rethink how they lay out their schools, plan for professional development, or purchase equipment.
But in order to do this powerful analysis, I need to gain some basic programming skills. I have been staring at this screen all freakin’ day, folks…
…trying to make that list of errors at the bottom diminish. Trying to, more specifically, incorporate one more variable into a program someone else wrote (which means learning their coding scheme and inserting my own code into it). This is something my counterparts over in Ed Measurement can do in a few swipes on the keyboard, but I’m learning the hard way, with syntax reference sheets and help tabs open in the background.
But I’m learning a lot, and it’s worth it. By debugging this program to meet my needs, I’m forcing myself to relearn the math (and thus the theory) behind the models, I’m gaining valuable statistical analysis skills that I will use in future studies, and I’m thinking about how my quantitative findings (assuming I ever get any actual findings) line up with what I saw and heard at my research site this semester.
And this is all valuable work, even if I feel like I haven’t accomplished much of anything today.
Because I’m debugging my brain. And as it happens, this is an important part of dissertating.