Even the tools of predictive modeling are commonly applied to network data (e.g. correlation and regression)
Would running such tests require a need for assessing latent variables that emerge from network analysis? I will keep reading, but from what I know of SNA, it seems like you are only analyzing observable variables and it would be difficult to obtain a correlation from such unique variables. Am I way off here?
How well does a leader's perceived level of trust in his or her colleagues predict the number of alters to whom the person sent a collaboration tie in year 1, controlling for gender and the level at which the person works (district vs. school)? This question requires three vectors of independent variables (trust score, an indicator for gender, and an indicator for level) and one dependent variable vector (collaboration year 1 out-degree).
Oy. This (multiple linear regression) is much harder for me to wrap my brain around, but I'm going to give it a go (especially since I have so many variables and it might be useful for tying them together). Again, I have SP, CP, and discussion forum type. If I did the surveys, I'd also have perceived sense of belonging. How well does a student's perceived sense of belonging predict his/her level of CP in discussion posts, controlling for discussion forum type (for example, only looking at large group discussions)? That doesn't get the SP in there though, and may not actually be a great model of regression. :/ Help?
The question, therefore, is whether school leaders prefer to collaborate with those with whom they have collaborated in the past or with those that they have turned to discuss confidential issues.
I have next to no stats knowledge, so I'm going to try to extrapolate this out in regards to my own research to try to better understand it (hopefully!). In using my own research with SP (social presence) and CP (cognitive presence). I'm going to start with the varibles: levels of SP, levels of CP, discussion forum type. A question I have been asking is whether discussion forum type affects SP and/or CP. Modeling the question the same way as this one, it might be whether students are more likely to show higher CP with students they were in a programmatic small group discussion with versus just large group. I think this models this line of questioning, at least. This probably doesn't get the SNA part in. So, trying again... Are students more likely to respond to a student in a large group discussion that they formed a connection with in a programmatic small group discussion or a random small group discussion? This doesn't get the SP or CP working in there, but it gets the SNA. So part of what I'm studying. But, since I'm graphing SP as a weighted measure for SNA, maybe it could be whether students are more likely to demonstrate higher SP in an ensuing large-group discussion with students they were in small programmatic group discussion with in a previous module. Does that get all the parts working approporiately in a MR-QAP-procedure question??
So, let's say you are interested in the number of collaborative exchanges that occur between teachers from two different grade levels in a complete network of teachers within one elementary school. First, you count the number of times these types of exchanges occur in the observed network and then permute these relational data lots and lots of times. With each permutation, you calculate the number of times this type of tie (collaborative exchanges between teachers from two different grade levels) occurs and compare this result to the original observed network. After this process of permuting and comparing, you can see how often the results of these permutations are the same as the original observed results: The more often the results of the permutations are the same as your observed data, the more likely that the pattern of exchanges in the observed data was due to chance. If, however, the results from the observed data are so unlikely when compared to the results of the permutations, then you are to conclude that your results are not the byproduct of chance. Therefore, this result would be considered statistically significant.
In terms of my project (looking at racial and gender-based biases in communication between undergarduate students in an online class), then I could use this same rationale and process in order to make generalizations to a broader population?
If an individual opts out, this should mean that their name appears nowhere on the social network diagram (even if they are identified by another individual as being part of their social network). For instance, in the sample map, you can see that the map would be very disjointed if John and Holly opted out of the SNA.
Are we allowed to include nodes for John and Holly if they are identified by others, but without using their name? For example, we would refer to John as Anonymous 1 and Holly as Anonymous 2. Or would we have to exclude any data that involves them, regardless of anonymity?
For example, does a child benefit from having parents, teachers, neighborhood adults, and so on communicate with each other, or do these relations constrain that child?
This is such an interesting question! My area is in college students, so I don't have to worry so much about parental involvement, but for those in K-12, the assumption that this sort of close communication is good. I wonder if it's like the stroke victims where having more weak ties among the network is a good thing. Though I can't fathom why that would be the case in this situation.