My goal for this week was to progress with my research. Two research projects were at the forefront of my mind. First it was to do a more throughout analysis for an upcoming conference presentation. Secondly, with a team of collaborator we decided to finally get that paper done. The other we abandoned, deciding that our question was answered by another researcher, and ex-colleague to be exact.
The aim of these research-tagged post is twofold: First, to share with other career and citizen researchers at all seniority levels the steps I have taken to analyze my data. Maybe that helps to some degree towards opening up researcher. Secondly, to let go of some steam and organize my thoughts. If you are deep in R code and data it is sometimes necessary to get out of it, to catch some breath before doing another dive. This line is specifically for all those capstone students who soon will be finding themselves in exactly this stage.
Vacancies and skills: Why are certain skill categories mention in certain vacancies ?
If you read more of my research posts, you’ll realize that networks are my thing. So in a way I’m happy that this week was full of social network analysis. I’m getting my head into a new analytical method thanks to my research into academic vacancies and skills: ERGM for bipartite graphs. I’m half excited about it, half banging my head. At the end of the day, it’s still an ERGM.
Admittedly, I haven’t been able to produce any results. This might sound like I have made no advances. But, it’s good to always look at things in two ways. These are my outcomes for this week for this project:
- I have not produced an ERGM model that converged. I’m stuck with a singular hessian matrix.
- I have worked on model degeneracy and I think I got this sorted.
- I have created two network graphs. One with isolates and one without. They are not yet beautiful, but they paint the first picture
The reason why I have renewed motivation to work on this project is that I do need to present some findings in 30 days. The results I included in the conference abstract are descriptive. Secondly, I’m helping a colleague analyze his network data (Collaboration among developers) which could also be analyzed using the bipartite ERGM. For the project on collaboration among developers, we are currently thinking about doing two analysis: Valued ERGM and longitudinal social network analysis.
Key sources that guide my work
I have a couple of key articles for this research which describe the method and provide some insight into what theories can be used. Compared to other researchers, I’m switching research topics. Hence I am not that familiar with the literature on the academic labour market. Most of what I have read so far comes from an economic perspective. Regarding the method, I draw inspiration from Bubbling up good ideas by Bryan Stepehens and colleagues.
Why do innovators from different background exchange information?
When I started to work in academia, I had a position as junior researcher in an innovative large-scale interfaculty project. Our team was task to experiment with educational innovations. Later on, academic and administrative staff were given the opportunity to apply for seed money for small-scale innovation project that in some way improve the educational experience of students. With a group of colleagues we researched information exchange among these innovators. The research question explores information exchange in interdisciplinary academic teams.
It falls partly within the research of science of team scientists. However, as these teams work on an educational innovation, and not a scientific project, it has a different angle. Keep in mind that in many cases, teaching and education is secondary for scientists at universities, and hence spending extra time on secondary activities that not necessarily pay off for your promotion might not make sense.
How I analyzed the data
For this project, we have many teams with different sizes. We eliminated all teams with too may missing values and then relating information exchange to dyadic variables such as physical proximity, hierarchical differences, and knowledge sharing attributes. I’m using multiple linear regression quadratic assignment procedure, MRQAP, for this. Very brief, this is linear regression adapted to network data. Network data violates the assumption of independence, thus traditional regression can’t be applied. I could do ERGM, but a) all our data is valued and would have to be binaries for ERGM, and b) we wanted to keep the paper accessible for an audience not familiar with network methods.
While I do have some results, I still need to control for group membership. This makes the results not valid, and hence I’m not sharing them. I do know now that the code runs. It’s bug free.
This post was written on a playground, in between martial art practice session of my kids.
Call to researchers: Are you looking for a place to clear your head and write down preliminary insights about your research ? Contact me for a spot.