Unit 2 Case Study: Collaboration Ties over Time

1. Prepare

Our Unit 2 Case Study: Collaboration Ties over Time revisits the research of Dr. Alan Daly and centers around the impact No Child Left Behind reform efforts on school and district leadership networks. In this unit we move beyond visual depictions of networks from our previous SNA case study and learn to describe networks using common network-level measures. Our analyses are guided by research conducted by Daly and Finnagan @daly2011ebb who examined the ebb and flow of social network ties between district leaders under high-stakes accountability. As we work through each step of the data-intensive workflow [@krumm2018], we will focus on the following topics:

Prepare: Prior to analysis, we’ll take a look at the context from which our data came and load various R packages for data science and network analysis.
Wrangle: In section 2 we make use of the {readxl} and {tidygraph} packages to import and prepare relational data stored as an adjacency matrix.
Explore: In our prior case study, we indirectly used the {igraph} package on which {tidygraph} depends. In section 3, we use {igraph} directly to calculate a range of network-level measures to describe and compare collaboration networks over time.
Model: While we don’t do any modeling of our data in this case study, we dive a little deeper into work of Daly and Finnagan @daly2011ebb to see how modeling can be used to examine longitudinal network data.
Communicate: Finally, we prepare a simple “data product” consisting of a data visualization and/or table that highlights some key findings from our analysis.

1a. Review the Research

Daly, A. J., & Finnigan, K. S. (2011). The ebb and flow of social network ties between district leaders under high-stakes accountability. American Educational Research Journal, 48(1), 39-79.

Abstract

Recent scholarship suggests the importance of school district offices in supporting reform. These studies provide strategies for building relations between central offices and sites in order to improve change efforts. However, what is frequently overlooked is that organizational reform efforts are socially constructed. Therefore, examining the underlying reform-related social networks may provide insight into how relational structures support or constrain efforts at reform. This longitudinal case study draws upon social network analysis and interviews to examine the reform-related knowledge, advice, and innovation network structures of central office and site leaders in a district facing sanction for underperformance and engaging a districtwide reform. Findings indicate that over time, the networks increased the number of superficial interactions, and more frequent exchanges remained unchanged, resulting in a centralized network structure.

👉 Your Turn ⤵

Take a look at the article The Ebb and Flow of Social Network Ties Between District Leaders Under High-Stakes Accountability located in the lit folder of your files pane and answer the following questions:

What research questions were the authors of this study trying to address and why did they consider these questions important?
- “In what ways have reform-related social networks between district and site leaders evolved over a 2-year period? How do district and school leaders perceive the evolution of the social networks and its effect on the exchange of reform-related resources within the district?” (Daly & Finnigan, 2011, p. 41).
- The researchers considered these questions important because they were trying to determine the ways in which a district marked “INI” (in need of improvement) responds to the pressures of the No Child Left Behind Act (NCLB). As a result of NCLB, districts across the country suffered sanctions as a result of their low performance. According to Daly and Finnigan, the school and district leadership then became responsible for implementing reforms to attempt to improve their schools and districts (2011). They researchers believed that collaboration and connectivity between district and school leaders could potentially affect the reform process and could potentially serve as an indication of the success and longevity of the reforms (Daly & Finnigan, 2011).
How were participants selected for this study and what was the rationale?
- The researchers wanted a district that was implementing changes in response to NCLB sanctions and with whom they could work for an extended period of time. In the words of the authors, “We selected La Estasis as it typifies many of the districts across California that serve primarily students of color from low socioeconomic communities, has a pattern of underperformance, and is engaging in an intentional districtwide reform in an effort to exit INI” (Daly & Finnigan, 2011, pp. 50-51). Within that district, the authors specifically wanted the leadership to participate in the study because this research focused on the connections/ties between district leadership, who are considered agents of change. Therefore, teachers were not included in the study: district leadership (superintendent, assistant superintendent, etc.) and school-based leadership (principals and assistant principals) were included (Daly & Finnigan, 2011, p. 52).
How did the authors collect social network data for this study and what specific information about actors and ties were collected?
- The authors used a mixed-methods longitudinal study. Primarily the social network data were collected using surveys. However, they augmented this data with interview data from a select number of actors. They did not, however, conduct interviews with all 43 participants (8 interviews total). The survey was administered online and included a combination of “social network relations and demongraphic questions” (Daly & Finnigan, 2011, p. 51). The SNA questions included frequency questions in which participants were asked to rank the frequency with which they sought out other administrators using a scale of 1-5, with 1 being never and 5 being 1-2 times per week (2011, p. 51). Specifically, the SNA questions were related to advice-seeking behavior, and the researchers tailored these questions to the specific reforms being conducted in their district. For example, one question was ‘’Please select the administrators in the La Estasis network to whom you go to for advice related to implementation of the [literacy] reform effort and at what frequency?’’ (Daly & Finnigan, 2011, p. 51).

1b. Load Libraries

Recall that R packages, sometimes referred to as libraries, are shareable collections of R code that can contain functions, data, and/or documentation and extend the functionality of R. Before loading an R package, you must first make sure it has been installed in your environment. You can always check to see which packages have already been installed and loaded into RStudio Cloud by looking at the the Files, Plots, & Packages Pane in the lower right hand corner.

Core SNA Packages

tidygraph 📦

The {tidygraph} package is a huge package that exports 280 different functions and methods, including access to almost all of the dplyr verbs plus a few more, developed for use with relational data. While network data itself is not tidy, it can be envisioned as two tidy tables, one for node data and one for edge data. The {tidygraph} package provides a way to switch between the two tables and uses dplyr verbs to manipulate them. Furthermore it provides access to a lot of graph algorithms with return values that facilitate their use in a tidy workflow.

ggraph 📦

Created by the same developer as {tidygraph}, {ggraph} – pronounced gg-raph or g-giraffe hence the logo – is an extension of {ggplot} aimed at supporting relational data structures such as networks, graphs, and trees. Both packages are more modern and widely adopted approaches data visualization in R.

While ggraph builds upon the foundation of ggplot and its API, it comes with its own self-contained set of geoms, facets, etc., as well as adding the concept of layouts to the grammar of graphics, i.e. the “gg” in ggplot and ggraph.

igraph 📦

Both {tidygraph} and {ggraph} depend heavily igraph network analysis package. The main goals of the igraph package and the collection of network analysis tools it contains are to provide a set of data types and functions for:

pain-free implementation of graph algorithms,
fast handling of large graphs, with millions of vertices (i.e., actors or nodes) and edges,
allowing rapid prototyping via high level languages like R.

New Packages

readxl 📦

The {readxl} package makes it easy to get data out of Excel and into R. Compared to many of the existing packages (e.g. gdata, xlsx, xlsReadWrite) readxl has no external dependencies, so it’s easy to install and use on all operating systems. It is designed to work with tabular data.

Since one of our data wrangling steps in the next section is importing network matrices stored in excel files, this package will come in handy.

👉 Your Turn ⤵

Use the code chunk below to load the tidyverse, igraph, tidygraph, ggraph, and readxl packages that we’ll be using in this case study:

library("readxl")

library("ggraph")

## Loading required package: ggplot2

library("tidygraph")

## 
## Attaching package: 'tidygraph'

## The following object is masked from 'package:stats':
## 
##     filter

2. Wrangle

In general, data wrangling involves some combination of cleaning, reshaping, transforming, and merging data[@wickham2016r]. As highlighted in Estrellado et al. @estrellado2020e, wrangling network data can be even more challenging than other data sources since network data often includes variables about both individuals and their relationships. For data wrangling in this unit, we’re keeping it simple since working with relational data, especially in the form of matrices, is a bit of a departure from our working with rectangular data frames.

2a. Import Data

One of our primary goals for this case study is to examine how collaboration networks among school and district leaders changed over time. To do so, we’ll need to import two Excel files from the Social Network Analysis and Education companion site. Both files contain edges stored as a square matrix (more on this later) for the first and third year of a study examining the impact of national reform efforts [@daly2011ebb].

These files are included in the data folder of your R Studio project. A description of each file from the companion website is linked above and each data file is linked below:

year_1_collaboration.xlsx This adjacency matrix reports on “collaboration” ties among 43 school leaders in year 1 of a three-year study. This is a directed valued (weighted) network measured on five-point scale ranging from 0 to 4, with higher values indicating more frequent collaborations (1–2 times/week). These data are used throughout Chapter 9.
year_3_collaboration.xlsx This adjacency matrix reports on “collaboration” ties among 43 school leaders in year 3 of a three-year study. This is a directed valued (weighted) network measured on five-point scale ranging from 0 to 4, with higher values indicating more frequent collaborations (1–2 times/week).

Recall from above that our relations, or edges, are stored as a valued adjacency matrix in which columns and rows consist of the same actors and each cell contains information about the tie between each pair of actors. In our case, the tie is a directed and valued “arc” where the value indicates the frequency of collaboration.

Let’s use the read_excel() function to import the year_1_collaboration.xlsx file, add an argument setting the column names to FALSE since our file is a simple matrix with no header or column names, and assign the matrix to a variable named year_1_collaboration:

R Studio Tip: Type ?read_excel into the console and check the arguments section to examine the different arguments that can be used with this function.

year_1_collaboration <- read_excel("data/year_1_collaboration.xlsx", 
                            col_names = FALSE)

## New names:
## * `` -> ...1
## * `` -> ...2
## * `` -> ...3
## * `` -> ...4
## * `` -> ...5
## * ...

Before importing our Year 3 file, let’s quickly inspect the R object we just imported to see what we’ll be working with.

year_1_collaboration

## # A tibble: 43 × 43
##     ...1  ...2  ...3  ...4  ...5  ...6  ...7  ...8  ...9 ...10 ...11 ...12 ...13
##    <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
##  1     0     0     3     0     0     0     0     0     0     0     0     0     0
##  2     0     0     0     0     0     0     0     0     0     0     0     0     0
##  3     4     0     0     0     0     0     0     0     0     0     0     0     0
##  4     0     0     0     0     0     0     0     0     0     0     0     0     0
##  5     0     0     0     0     0     0     0     0     3     0     0     0     0
##  6     3     0     0     0     0     0     0     0     0     0     0     0     0
##  7     0     0     0     0     0     0     0     0     0     0     0     0     0
##  8     0     0     0     0     0     0     0     0     0     4     0     0     0
##  9     0     0     0     0     4     0     0     0     0     0     0     0     0
## 10     0     0     0     0     0     0     0     4     0     0     0     0     0
## # … with 33 more rows, and 30 more variables: ...14 <dbl>, ...15 <dbl>,
## #   ...16 <dbl>, ...17 <dbl>, ...18 <dbl>, ...19 <dbl>, ...20 <dbl>,
## #   ...21 <dbl>, ...22 <dbl>, ...23 <dbl>, ...24 <dbl>, ...25 <dbl>,
## #   ...26 <dbl>, ...27 <dbl>, ...28 <dbl>, ...29 <dbl>, ...30 <dbl>,
## #   ...31 <dbl>, ...32 <dbl>, ...33 <dbl>, ...34 <dbl>, ...35 <dbl>,
## #   ...36 <dbl>, ...37 <dbl>, ...38 <dbl>, ...39 <dbl>, ...40 <dbl>,
## #   ...41 <dbl>, ...42 <dbl>, ...43 <dbl>

As you can see, we have a 43 x 43 “tibble” or data table representing our collaboration ties. In Section 2, we’ll learn how to restructure this data into a formal matrix and then convert to a network class R object so we can explore our data through network visualization and analyses.

👉 Your Turn ⤵

Use the code chunk below to import the Year 3 collaboration network located in data folder of your R project and then inspect your newly created object to make sure it imported as expected.

year_3_collaboration <- read_excel("data/year_3_collaboration.xlsx", 
                            col_names = FALSE)

## New names:
## * `` -> ...1
## * `` -> ...2
## * `` -> ...3
## * `` -> ...4
## * `` -> ...5
## * ...

year_3_collaboration

## # A tibble: 43 × 43
##     ...1  ...2  ...3  ...4  ...5  ...6  ...7  ...8  ...9 ...10 ...11 ...12 ...13
##    <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
##  1     0     2     4     3     1     1     1     3     1     1     1     1     1
##  2     0     0     1     0     0     0     0     1     0     0     0     0     0
##  3     4     1     0     1     2     2     2     3     2     2     2     2     2
##  4     3     2     0     0     0     0     0     1     0     0     1     0     0
##  5     1     0     2     0     0     1     1     2     3     0     1     1     2
##  6     3     0     0     0     0     0     0     1     1     0     0     0     0
##  7     2     1     1     1     2     1     4     2     1     4     1     1     1
##  8     3     2     4     1     3     3     2     0     3     3     2     3     3
##  9     2     0     2     0     4     2     0     3     0     1     2     2     1
## 10     2     1     3     2     2     1     4     4     3     0     3     2     2
## # … with 33 more rows, and 30 more variables: ...14 <dbl>, ...15 <dbl>,
## #   ...16 <dbl>, ...17 <dbl>, ...18 <dbl>, ...19 <dbl>, ...20 <dbl>,
## #   ...21 <dbl>, ...22 <dbl>, ...23 <dbl>, ...24 <dbl>, ...25 <dbl>,
## #   ...26 <dbl>, ...27 <dbl>, ...28 <dbl>, ...29 <dbl>, ...30 <dbl>,
## #   ...31 <dbl>, ...32 <dbl>, ...33 <dbl>, ...34 <dbl>, ...35 <dbl>,
## #   ...36 <dbl>, ...37 <dbl>, ...38 <dbl>, ...39 <dbl>, ...40 <dbl>,
## #   ...41 <dbl>, ...42 <dbl>, ...43 <dbl>

2b. Create a Tidy Graph

As noted above, before we can begin exploring our data through through network visualization and analysis, we must first restructure our data into a formal matrix and then convert to a network class R object required by the igraph, tidygraph, and ggraph packages.

However, one thing you may have noticed from our inspection of the data above is that our data table is missing the names of the school leaders in our network, which are needed for determining dyads in our network. R has packages for creating random names to help anonymize data, but to keep things simple, we’ll just assign the numbers 1-43 as names for our rows and columns.

rownames(year_1_collaboration) <- 1:43

## Warning: Setting row names on a tibble is deprecated.

colnames(year_1_collaboration) <- 1:43

You may have seen a warning stating: Setting row names on a tibble is deprecated. You can ignore that for now but it’s basically telling us these functions are old we we need to use newer function or our code will some day stop working.

Again, let quickly inspect our year_1_collaboration data table to see if this worked:

year_1_collaboration

## # A tibble: 43 × 43
##      `1`   `2`   `3`   `4`   `5`   `6`   `7`   `8`   `9`  `10`  `11`  `12`  `13`
##  * <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
##  1     0     0     3     0     0     0     0     0     0     0     0     0     0
##  2     0     0     0     0     0     0     0     0     0     0     0     0     0
##  3     4     0     0     0     0     0     0     0     0     0     0     0     0
##  4     0     0     0     0     0     0     0     0     0     0     0     0     0
##  5     0     0     0     0     0     0     0     0     3     0     0     0     0
##  6     3     0     0     0     0     0     0     0     0     0     0     0     0
##  7     0     0     0     0     0     0     0     0     0     0     0     0     0
##  8     0     0     0     0     0     0     0     0     0     4     0     0     0
##  9     0     0     0     0     4     0     0     0     0     0     0     0     0
## 10     0     0     0     0     0     0     0     4     0     0     0     0     0
## # … with 33 more rows, and 30 more variables: 14 <dbl>, 15 <dbl>, 16 <dbl>,
## #   17 <dbl>, 18 <dbl>, 19 <dbl>, 20 <dbl>, 21 <dbl>, 22 <dbl>, 23 <dbl>,
## #   24 <dbl>, 25 <dbl>, 26 <dbl>, 27 <dbl>, 28 <dbl>, 29 <dbl>, 30 <dbl>,
## #   31 <dbl>, 32 <dbl>, 33 <dbl>, 34 <dbl>, 35 <dbl>, 36 <dbl>, 37 <dbl>,
## #   38 <dbl>, 39 <dbl>, 40 <dbl>, 41 <dbl>, 42 <dbl>, 43 <dbl>

As you can see, we now have names for our rows and columns corresponding to each school leader in our network. So we can see that leader 1 collaborated with leader 3 and with a frequency of “3” but obviously did not collaborate with 2 as indicated by the “0” in the corresponding cell. Note that leader 3 also indicated that they collaborated with 1, but on a slightly more frequent basis as indicated by the “4” in the corresponding cell.

Convert to Matrix Object

Now that we have names included for our rows and columns, we need to convert our data table, or tibble, to a formal matrix class object. To do so is relatively simple using the as.matrix() function built into R.

year_1_matrix <- as.matrix(year_1_collaboration )

The word “class” and “object” have been used quite a bit in this case-study and warrant a brief explanation. Classes and objects are basic concepts of Object-Oriented Programming environments like R. An object is simply a data structure that has some methods and attributes. Everything in R is essentially an object. A class is just a blueprint or a sketch of these objects. It represents the set of properties or methods that are common to all objects of one type.

Let’s use the class() function on the year_1_collaboration and year_1_matrix to see the types of objects we just created:

class(year_1_collaboration)

## [1] "tbl_df"     "tbl"        "data.frame"

class(year_1_matrix)

## [1] "matrix" "array"

Great! We can now see that our year_1_matrix is formally an object of the “matrix” class.

Convert to Graph Object

Our final step before we’re able to begin exploring our data is to convert our matrix to a network object recognized by the {igraph} and {tidygraph} packages. The as_tbl_graph() function can easily convert relational data from all common network data formats such as matrices, network, phylo, dendrogram, data.tree, graph, etc.

Run the following code to convert our matrix to directed network graph and save as a new object called year_1_network:

library("ggraph")
library("tidygraph")
year_1_network <- as_tbl_graph(year_1_matrix, directed = TRUE)

Again, let’s inspect our new object:

year_1_network

## # A tbl_graph: 43 nodes and 82 edges
## #
## # A directed simple graph with 3 components
## #
## # Node Data: 43 × 1 (active)
##   name 
##   <chr>
## 1 1    
## 2 2    
## 3 3    
## 4 4    
## 5 5    
## 6 6    
## # … with 37 more rows
## #
## # Edge Data: 82 × 3
##    from    to weight
##   <int> <int>  <dbl>
## 1     1     3      3
## 2     3     1      4
## 3     3    24      3
## # … with 79 more rows

As you can see, our year_1_network provides a range of information about out network including network size, type, number of components, and a preview of the node and edge lists that it created. The node and edge lists are treated just like a typical data frame and can now be used with other tidyverse packages and functions to create new actor-level network variables like degree, reciprocity, and centrality measures.

👉 Your Turn ⤵

Use the code chunk below to complete the following data wrangling steps for our Year 3 collaboration network from the Excel file you imported earlier:

Add row and column names to the data frame;
Convert the resulting data frame to a matrix class object and check to confirm that it is indeed a matrix;
Convert the matrix to a network graphs object with directed edges;
Inspect the Year 3 collaboration network and answer the questions after the code chunk.

rownames(year_3_collaboration) <- 1:43

## Warning: Setting row names on a tibble is deprecated.

colnames(year_3_collaboration) <- 1:43

year_3_matrix <- as.matrix(year_3_collaboration)

class(year_3_collaboration)

## [1] "tbl_df"     "tbl"        "data.frame"

class(year_3_matrix)

## [1] "matrix" "array"

year_3_network <- as_tbl_graph(year_3_matrix, directed = TRUE)

year_3_network

## # A tbl_graph: 43 nodes and 1314 edges
## #
## # A directed multigraph with 1 component
## #
## # Node Data: 43 × 1 (active)
##   name 
##   <chr>
## 1 1    
## 2 2    
## 3 3    
## 4 4    
## 5 5    
## 6 6    
## # … with 37 more rows
## #
## # Edge Data: 1,314 × 3
##    from    to weight
##   <int> <int>  <dbl>
## 1     1     2      2
## 2     1     3      4
## 3     1     4      3
## # … with 1,311 more rows

Section 2 Questions:

What is the size of the Year 3 collaboration network?
- Our tibble graph is 43 x 1314
How many components are in this network?
- This is a directed multigraph with 1 component (decrease from Year 1, which had 3 components due to two outliers).
How many nodes and edges variables are in our tidy graph and what does each variable (e.g., weight) represent?
- There are 43 distinct nodes (1-43) within our tidy graph; each represents a unique actor within the network.
- There are 1,314 edges within the tidy graph. Each represents the connection that one actor “sends” to another. The weight represents the frequency with which the actor states that they connect with the alter (receiving actor). For example, a weight of 4 indicates frequent connection: 1-2 times per week.
How does collaboration in Year 3 compare to Year 1 and what inferences might you already be able to make just based on the size of the two networks?
- There are significantly more edges in the Year 3 data. From this, we can infer that the school leader collaboration was successful at least in establishing more connections between individuals. Furthermore, since the nodes remained the same from Year 1 to Year 3 (43), we can infer that these connections expanded within the same network population and were not a result of the network itself growing.

3. Explore

Relevant to our case study of school leader collaboration, Carolyn notes in SNA and Education @carolan2014 that:

… to address questions that probe a network’s dynamics (i.e., its change over time), it is necessary to first figure out what the network looks like at one point in time, what is commonly referred to as the network’s topography.

In order to describe a network’s topography, social network researchers utilize an array of algorithms and indices to describe both visually and numerically different characteristics of social networks. These algorithms and indices rely on both simple and somewhat sophisticated mathematical computation.

In this section we revisit the {tidygraph} and {ggraph} packages for creating sociograms and computing basic node and edge-level measures, and then introduce some key functions from the {igraph} package for obtaining complete network-level measures.

3a. Visually Describe the Network

One of the defining characteristics of the social network perspective is its use of graphic imagery to represent actors and their relations with one another. To emphasize this point, Carolyn (2013) reported that:

The visualization of social networks has been a core practice since its foundation more than 90 years ago and remains a hallmark of contemporary social network analysis.

Network visualization can be used for a variety of purposes, ranging from highlighting key actors to even serving as works of art. Katya Ognyanova’s also excellent tutorial on Static and Dynamic Network Visualization with R helps illustrate the variety of goals a good network visualization can accomplish.

Recall from our previous SNA case study examining Twitter networks surrounding the Common Core State Standards that we introduced the dead simple plot() function built into R to quickly generate a sociogram showing the nodes and edges in our network.

Let’s supply our year-1-network object as the first argument of this function and see what it produces:

plot(year_1_network)

While not the most aesthetically pleasing plot, it definitely is functional. As indicated by our tidy graph output from section 1, we can see that there is one large network component and two additional components consisting of isolated school leaders with no collaboration ties.

If this had been a larger network with more nodes and edges, however, the utility of the basic plot() function would evaporate quickly. For example, try this function out with the Year 3 network in the console and see what it produces.

Fortunately, the {ggraph} package includes a plethora of plotting parameters for graph layouts, edges and nodes to improve the visual design and readability of network graphs.

Note: If you’re having trouble seeing the plot you can run the code in the console and see the plot in the Viewer pane or click the little arrow/window icon in the output of the code chunk to open the plot in a new window and resize.

Sociograms with ggraph

One thing to keep in mind when building a network viz with the {ggraph} package is that just like it’s {ggplot} counterpart for making more traditional charts and graphs, there is a minimal code template for producing a basic plot. Specifically, ggraph requires 3 main functions

ggraph() takes care of setting up the plot object along with creating the layout of the graph (default = “stress”).
geom_node_*() functions and their associated arguments add and modify the nodes of the network plot.
geom_edge_*() functions and their associated arguments add and modify the edgez of the network plot.

Like the basic plotting function above, ggraph has a similar autograph() function for automatically generating a simple sociogram to get a quick snapshot of your network. Run the following code supplying our year_1_network object as the first argument for this function:

autograph(year_1_network)

As you can see, this produces a very basic plot that includes our network nodes and edges with minimal styling.

Within this function, however, are the three core functions noted above that can be modified though different arguments or by adding additional layers of functions to change the visual aesthetics of your sociogram to highlight key characteristics of the network, like edge weight or central actors.

Let generate the same basic plot above but explicitly call our ggraph, node, and edge functions:

ggraph(year_1_network) +
  geom_node_point() +
  geom_edge_link()

## Using `stress` as default layout

We won’t spend a lot of time on polishing our graph here. We’ll save that effort for when we are ready to communicate our findings to a specific audience, but lets make some quick functional and aesthetic changes: Run the following code to:

Change the graph layout to the Fruchterman-Reingold (FR) layout using the layout = "fr" argument to help highlight areas of clustering;
Change the size of our nodes using the size = 3 argument to make them more visible;
Add color to our edges based using the aes(colour = weight) argument to highlight more frequent collaborations; and,
Add the theme_graph() layer to remove the gray background.

ggraph(year_1_network, layout = "fr") +
  geom_node_point(size = 3) +
  geom_edge_link(aes(colour = weight)) +
  theme_graph()

That is perhaps slightly better, but before sharing with an education partner, I would likely do quite a bit more clean up to highlight key findings specific to network measures that we’ll explore in the next section. For example, I might change the size of nodes for school leaders based on measures of centralization to highlight more central actors in the network and add arrows to indicate directionality of the ties since this is a directed network.

👉 Your Turn ⤵

Use the code chunk below created a simple sociogram for the Year 3 collaboration network and answer the questions that follow:

ggraph(year_3_network, layout = "fr") +
  geom_node_point(size = 3) +
  geom_edge_link(aes(colour = weight)) +
  theme_graph()

How does the Year 3 collaboration network compare to Year 1? What are some similarities and differences?
- The Year 3 collaboration network obviously has significantly more edges (roughly 16 times as many). Therefore, there are far more lines, making it far more difficult to see. However, the number of nodes remains the same (43). Interestingly, even though the edges have multiplied tremendously, there seem to be proportionately far more less frequent connections (i.e. weights of 1 or 2) than in the Year 1 network. In the Year 1 network, the majority of the connections appear to be more frequent (i.e. weights of 3 or 4). This indicates that although the connections increased over the two years, the quality of those connections seems to be more superficial, on average, than those connections reflected in the Year 1 data.
How might you modify this graph to make it more insightful and useful? For ideas check out these network viz tutorials here, here and here.
- I would like to spread out the nodes more so that the connections/edges are clearer. Even when I open this plot in a separate window and zoom in, it is difficult to see all of the edges.
- Furthermore, rather than having the weights as a gradient of blue, I would have each of the four weights as a distinct color. This would make it easier to tell the rough proportions of each weight within the edges.
- Finally, I would increase the size of the nodes based on the number of “ties” they make with other actors.

3b. Describe the Network Mathematically

This section include some of the most common measures used to describe the features of complete networks, as well was .

Network-level structural measures are those that are calculated from the entire network. Therefore, they provide an excellent snapshot of the network’s structure—the pattern of relations among the network’s actors.

In addition to providing an overall summary of the network, these measures provide a concrete means to compare relations such as whether the friendship, social support, and advice seeking networks within the same fraternity, for example, are comparable. In our case study, for example, were are

1. Network Size

One simple but important property of a social network is its size, or the number of nodes and edges in a network. As Carolan notes:

Size plays an important role in determining what happens in the network— what resources are exchanged among actors, for example… Size affects other network measures, but on a conceptual level, it influences the structure of relations, as actors only have so many resources and capacities for creating and maintaining ties with others.

As we saw in the previous section, the year_1_network output provided us with basic information about our networks size, including the number of nodes and edges in our network. The {igraph} package also has some basic functions for retrieving the number of nodes and edges in a network.

Run the following code chunk to use the gorder() and gsize() functions obtain the number of vertices and edges respectively:

library("igraph")

## 
## Attaching package: 'igraph'

## The following object is masked from 'package:tidygraph':
## 
##     groups

## The following objects are masked from 'package:stats':
## 
##     decompose, spectrum

## The following object is masked from 'package:base':
## 
##     union

gorder(year_1_network)

## [1] 43

gsize(year_1_network)

## [1] 82

2. Centralization

A key structural property of complete networks, and often the focus of network studies is the concept of centralization, or the extent to which relations are focused on one or a small set of actors. Centralization is an important concept in network because, as Carolan notes:

Central actors likely wield a disproportionate amount of influence on the network. Therefore, high centralization provides fewer actors with more power and control.

Centralization can be based on a number of different measures, but degree is one of the most common. Degree refers to the number of ties an actor either sends (out-degree), receives (in-degree), or in the case of a non-directed network or both sent and received in a directed network, simply just “degree” for all actors to which one is connected.

The centr_degree() function in {igraph} provides a centralization score based on the metric selected, as well as the degree for each actors in the network. Let’s first take a look at the centralization measure based on number of actors individuals have ties with regardless of directionality:

centr_degree(year_1_network, mode = "all")

## $res
##  [1] 5 0 5 3 3 3 4 8 4 7 4 1 0 4 4 3 2 5 4 4 5 4 3 6 4 4 4 3 8 4 4 3 3 6 3 3 4 3
## [39] 6 2 3 4 2
## 
## $centralization
## [1] 0.05102041
## 
## $theoretical_max
## [1] 3528

The first variable $res provides node-level centrality scores. So for example, the first school leader in our network is connected to 5 other actors in the network. The $centralization variable provides our actual centrality score, which is quite small suggesting a very decentralized network. And the final $theoretical_max provides the maximum theoretical graph level centralization score for our graph based on the number of vertices.

Let’s also take a quick look at centralization scores for in-degree and out-degree:

centr_degree(year_1_network, mode = "out")

## $res
##  [1] 1 0 2 2 1 3 2 2 1 3 3 1 0 3 2 2 1 2 2 3 3 2 1 3 2 1 3 3 2 2 2 2 2 3 1 3 0 3
## [39] 2 1 2 2 1
## 
## $centralization
## [1] 0.02602436
## 
## $theoretical_max
## [1] 1806

centr_degree(year_1_network, mode = "in")

## $res
##  [1] 4 0 3 1 2 0 2 6 3 4 1 0 0 1 2 1 1 3 2 1 2 2 2 3 2 3 1 0 6 2 2 1 1 3 2 0 4 0
## [39] 4 1 1 2 1
## 
## $centralization
## [1] 0.09745293
## 
## $theoretical_max
## [1] 1806

Again, we still have a very decentralize graph, though slightly more centralized around in-degree.

While our primary concern is complete network-level measures so we can compare Year 1 & 3 networks, obtaining node-level measures is also very useful for helping to describe the network, particularly for visualization. For example, we might want to adjust the size of nodes in a network based on degree, so those actors with more connections stand out more prominently.

The {tidygraph} package has an unique function called activate() that allows us to treat the nodes and edges in our network object as if they were a standard data frame that we can then apply standard tidyverse functions to like select(), filter(), and mutate().

The latter function, mutate(), can be use to create new variables for node-level measures of degree, in-degree, and out-degree using the centrality_degree() function in the {tidygraph} package.

Run the following code to add each actor’s degree to our Year 1 collaboration network and assign the output year_1_network again so the results are saved:

year_1_network <- year_1_network |>
  activate(nodes) |>
  mutate(degree = centrality_degree(mode = "all"))

year_1_network

## # A tbl_graph: 43 nodes and 82 edges
## #
## # A directed simple graph with 3 components
## #
## # Node Data: 43 × 2 (active)
##   name  degree
##   <chr>  <dbl>
## 1 1          5
## 2 2          0
## 3 3          5
## 4 4          3
## 5 5          3
## 6 6          3
## # … with 37 more rows
## #
## # Edge Data: 82 × 3
##    from    to weight
##   <int> <int>  <dbl>
## 1     1     3      3
## 2     3     1      4
## 3     3    24      3
## # … with 79 more rows

Now we could adjust the size of more central actors in our sociogram from above like so to highlight the full range of collaboration ties in our network:

ggraph(year_1_network) +
  geom_node_point(aes(size = degree, color = degree)) +
  geom_edge_link(aes(color = weight)) +
  theme_graph()

## Using `stress` as default layout

3. Density

Since we know the number of ties, or edges, in our network, we can examine the overall density of our network, which is directly linked to network size. As Carolan explains:

Density refers to the number of ties in the network reported as a fraction of the total possible number of ties… The closer this number is to 1.0, the denser the network.

The igraph packages has two different function for calculating the density of a network, edge_density() and graph.density(), but both yield the exact same result. The edge_density() function is the more recent function and should be used, but let’s try them both out:

edge_density(year_1_network)

## [1] 0.04540421

graph.density(year_1_network)

## [1] 0.04540421

As you can see, our collaboration network in Year 1 has a VERY low density, which suggests that collaboration in this network is quite limited and may impact the flow of information, resources, and innovations among school leaders.

4. Reciprocity

Reciprocity is an important measure because it reveals the direction through which resources in networks flow and whether or not it flows in both directions. The tendency towards reciprocity in most networks, sometimes referred to as the “norm” of reciprocity, has been acknowledge by ancient writers such as Cicero who stated:

There is no duty more indispensable than that of returning a kindness… all men distrust one forgetful of a benefit.

At the network-level, reciprocity is defined as reciprocity is a measure of the likelihood of vertices in a directed network to be mutually linked. Using the reciprocity() function, we can directly calculate this measure:

reciprocity(year_1_network)

## [1] 0.1707317

As illustrated by the reciprocity measure above, the tendency towards reciprocated collaboration ties is, surprisingly, very low in our Year 1 network.

And if we wanted to highlight this lack of reciprocity in a network visual, we could use the edge_is_mutual() function to create a new edge variable called reciprocated that indicates whether the tie is mutual or asymmetric.

year_1_network <- year_1_network |>
  activate(edges) |>
  mutate(reciprocated = edge_is_mutual())

year_1_network

## # A tbl_graph: 43 nodes and 82 edges
## #
## # A directed simple graph with 3 components
## #
## # Edge Data: 82 × 4 (active)
##    from    to weight reciprocated
##   <int> <int>  <dbl> <lgl>       
## 1     1     3      3 TRUE        
## 2     3     1      4 TRUE        
## 3     3    24      3 FALSE       
## 4     4    29      3 FALSE       
## 5     4    41      4 FALSE       
## 6     5     9      3 TRUE        
## # … with 76 more rows
## #
## # Node Data: 43 × 2
##   name  degree
##   <chr>  <dbl>
## 1 1          5
## 2 2          0
## 3 3          5
## # … with 40 more rows

As shown in the output above, the tie between school leader 1 and 3 is reciprocated, but the tie between 3 and 24 is not.

Let’s now add this new variable to our sociogram to highlight reciprocated ties in our Year 1 network:

ggraph(year_1_network) +
  geom_node_point(aes(size = degree)) +
  geom_edge_link(aes(color = reciprocated)) +
  theme_graph()

## Using `stress` as default layout

As you can see, there only appear to be seven dyads with reciprocated ties!

5. Transitivity & Clustering

Transitivity, also known as the “friend of a friend” phenomenon, focuses on triads, or any “triple” of actors, rather than dyads like reciprocity. Just like a networks tendency towards reciprocity, there is also a tendency toward transitivity. For example, in the figure below Teacher A collaborates with Teacher B and Teacher B collaborates with Teacher C, then Teacher A is more likely at some point to collaborates with Teacher C and “complete” the triangle and less likely to collaborate with Teacher E since they do now share a connection:

As Carolan points out, establishing a network’s transitivity is important, as it is theoretically connected to actors’ tendencies to divide into exclusive subgroups or cluster over time, especially around positive relations such as friendship. While the algorithms behind network transitivity are not trivial, the defining network transitivity is rather straightforward:

Transitivity measures the probability that the adjacent vertices of a vertex are connected. This is sometimes also called the clustering coefficient.

The transitivity() function provides a convenient means to calculate a network-level transitive measures:

transitivity(year_1_network)

## [1] 0.2435897

For our particular network, a relatively small value of 24.4% is not too surprising since this is the start of the research study and since the density of our network overall is very low.

6. Diameter & Distance

Whereas transitivity focuses on the importance of certain configurations of triads, diameter and distance are more straightforward. A network’s diameter simply refers to the longest path between any two actors. A related measure is the average path length, which measures the mean distance between all pairs of actors in the network.

The two functions for computing diameter and average path length are diameter() and mean_distance() respectively. Let’s apply both of these to our year_1_network object:

diameter(year_1_network)

## [1] 42

mean_distance(year_1_network)

## [1] 4.858216

Carolan notes that both diameter and average path length are important network-level structural properties. Similar to diameter and distance measures for our Year 1 network above, he notes that a network with a large diameter and small average path length suggests a structure in which there are parts of the network that some network actors may be unable to access.

👉 Your Turn ⤵

Use the code chunk below to calculate the following measures for the Year 3 collaboration network, then answer the questions below:

Size
Centralization
Density
Reciprocity
Transitivity & Clustering
Diameter & Distance

#YOUR CODE HERE

#size
gorder(year_3_network)

## [1] 43

gsize(year_3_network)

## [1] 1314

#centralization 
centr_degree(year_3_network, mode = "all")

## $res
##  [1] 76 36 80 41 68 37 73 81 65 64 75 50 56 66 73 35 68 81 38 67 62 76 52 71 58
## [26] 45 55 80 33 69 37 55 72 76 48 53 74 77 54 50 68 60 73
## 
## $centralization
## [1] 0.2423469
## 
## $theoretical_max
## [1] 3528

#centralization in-degree and out-degree
centr_degree(year_3_network, mode = "out")

## $res
##  [1] 42 11 42 15 35  6 43 42 33 41 42 20 25 39 41  4 42 42  8 42 31 42 21 38 27
## [26] 19 21 43  4 43  8 26 42 42 22 26 41 42 26 25 42 26 42
## 
## $centralization
## [1] 0.2962348
## 
## $theoretical_max
## [1] 1806

centr_degree(year_3_network, mode = "in")

## $res
##  [1] 34 25 38 26 33 31 30 39 32 23 33 30 31 27 32 31 26 39 30 25 31 34 31 33 31
## [26] 26 34 37 29 26 29 29 30 34 26 27 33 35 28 25 26 34 31
## 
## $centralization
## [1] 0.2009967
## 
## $theoretical_max
## [1] 1806

#centralization mutations
year_3_network <- year_3_network |>
  activate(nodes) |>
  mutate(degree = centrality_degree(mode = "all"))

#centralization visualization
ggraph(year_3_network) +
  geom_node_point(aes(size = degree, color = degree)) +
  geom_edge_link(aes(color = weight)) +
  theme_graph()

## Using `stress` as default layout

#density
edge_density(year_3_network)

## [1] 0.7275748

graph.density(year_3_network)

## [1] 0.7275748

#reciprocity
reciprocity(year_3_network)

## [1] 0.7837423

#reciprocity visualization 1
year_3_network <- year_3_network |>
  activate(edges) |>
  mutate(reciprocated = edge_is_mutual())

#reciprocity visualization 2
ggraph(year_3_network) +
  geom_node_point(aes(size = degree)) +
  geom_edge_link(aes(color = reciprocated)) +
  theme_graph()

## Using `stress` as default layout

#transitivity
transitivity(year_3_network)

## [1] 0.9013133

#diameter & distance 
diameter(year_3_network)

## [1] 4

mean_distance(year_3_network)

## [1] 1.277962

How do network-level measures for Year 1 and Year 3 collaboration compare? What are some similarities and differences?
- One of the differences that jumped out at me right away was the significant increase in reciprocated ties. Obviously, since the number of ties is so much greater in Year 3, there would be an increased quantitatively. However, there is also a significant increase proportionally. Year one had a reciprocity value of 0.1707317. Year 3 had a reciprocity value of 0.7837423.
- Both Year 1 and Year 3 have a node size of 43. This is because the number of actors did not change from Year 1 to Year 3. I wonder how the researchers would have responded if the number of leadership positions had changed throughout the longitudinal study (perhaps related to reform efforts; perhaps unrelated). In my district, the district titles and personnel with those titles seems to shift every year. It seems unlikely to me that we would manage to retain the same personnel model for three consecutive years.
- The centralization score increased from 0.05 in Year 1 to 0.24 in Year 3. This shows that a few actors in the center of the plot probably exert more influence over the general timbre of the district conversation.
- The mean distance shrank substantially from Year 1 to Year 3 from 4.85 to 1.28, indicating that every actor within the network is more easily available to every other actor within the network. It does not, however, necessarily provide any insight into the quality of these connections.
What do these measures suggest about changes in collaboration among school leaders over time? In your response, cite some specific measures from Year 1 & 3 as evidence for your conclusions.
- These measures suggest that collaboration among school leaders increased over the longitudinal study. I am particularly interested in the dramatic increase in the reciprocity value. Low reciprocity can indicate more of a hierarchical network structure—i.e., “I turn to you for advice because I see you as an expert, but you do not turn to me for advice.” Higher reciprocity indicates that individuals are turning to one another on a more equal plane, seeing each other as collaborators and repositories of advice/wisdom. To me, this can lead to more effective transmission of ideas and attitudes. If those attitudes are positive and optimistic about the reforms, it can lead to more productive implementation of reforms. If the attitudes towards the reforms are negative, however, this increase in reciprocity can actually be counterproductive in terms of implementing reforms.
- Additionally, as I mentioned above, the consistency in nodes is interesting to me because it indicates that the managerial structure of the district did not change within the two years. This suggests that the hierarchy of the district remained intact during this time, which lends credence to a theory that the reform-related knowledge became more centralized during this time because the “ruling parties” so to speak became more established. It’s possible that as the leadership became more entrenched in their positions, they became more insular as a result.
- Finally, there is the noticeable shift towards centralization of knowledge, which pushes school leaders to the periphery of the network. This suggests that although interactions between leadership increased significantly, there was not necessarily an increase in knowledge exchange, since most of the knowledge was being circulated amongst a few individuals at the center of the network in positions of district leadership. Within this centralization, school leaders are not only cut off from knowledge being exchanged at the district level, but also from knowledge that other principals might gain.

4. Model

A developing area of network research is the advancement of statistical models and tools to examine network change over time [@carolan2014]. These models are used to empirically test if changes in key structure measures are statistically significant, and also explain potential mechanisms driving those changes. Stochastic actor-based models, for example, are a family of models that have the purpose of representing network dynamics on the basis of observed longitudinal data. These stochastic actor-based models allow users to test hypotheses about these tendencies toward reciprocity, homophily, and transitivity over time and to estimate parameters expressing their strengths while controlling for other tendencies (“confounders”).

👉 Your Turn ⤵

What model(s) did Daly and Finnagan use in their longitudinal case study to examine changes in reform-related knowledge, advice, and innovation network structures?
- Daly and Finnegan used stochastic actor-based statistical models to examine changes in reform-related knowledge, advice, and innovation network structures. Specifically, they used the Simulation Investigation for Empirical Network Analysis software package (SIENA) Version 3.17 (2011, p. 54) to model density, reciprocity, homophily, triadic closure, and centrality (2011, p. 54). They then used the UCINET software package to determine QAP correlations between knowledge, advice, and innovation, specifically examining correlations between knowledge and advice, advice and innovation, knowledge and innovation (2011, p. 54).
- Finally, they looked specifically at the knowledge network using UCINET and examined it for density, centrality, external-internal index, and overall network structure (Daly & Finnegal, 2011, p. 55).
What findings about network changes (list 2-3) did you find most interesting and why?
- It was interesting to me that the knowledge network became more centralized—specifically around district leadership, with “marginalized actors” in the form of “site administrators, who existed on the periphery of the network with little interaction between one another or district leaders” (Daly & Finnegan, 2011, p. 67). Although we are beyond the days of NCLB, during the five years that I lived and worked in this county, my district has had very little consistency. Each year (sometimes each month! I don’t exaggerate), there is some new “reform” initiative put forth that becomes mandatory. Although this particular study looked at leadership and not teachers, I know firsthand that very few teachers regularly communicate with district leadership. They are much more likely (and generally encouraged) to communicate with their site leaders. Yet, teachers and principals are those who interact with students on a daily basis. It is bizarre that reform-based knowledge would be centralized in the district leadership when they are not the individuals principally charged with enacting the reform. In my own experience, lack of transparency, poor communication, and a sense of being given instructions with little training or explanation, leads to teachers shutting their doors, ignoring the reform, and carrying on with business as usual. Alternatively, it leads to principals enforcing changes that they don’t fully understand, which leads to confusion, backtracking, and mixed messaging. This in turn causes teacher resentment, which generally does not improve student learning. Centralizing knowledge of reform efforts at the district level would, it seems, inevitably lead to employee dissatisfaction, confusion, and reluctance to implement prescribed changes. This does not seem like a positive environment for students.

5. Communicate

The final step in our workflow/process is sharing the results of our analysis with wider audience. Krumm et al. (2018) have outlined the following 3-step process for communicating with education stakeholders findings from an analysis:

Select. Communicating what one has learned involves selecting among those analyses that are most important and most useful to an intended audience, as well as selecting a form for displaying that information, such as a graph or table in static or interactive form, i.e. a “data product.”
Polish. After creating initial versions of data products, research teams often spend time refining or polishing them, by adding or editing titles, labels, and notations and by working with colors and shapes to highlight key points.
Narrate. Writing a narrative to accompany the data products involves, at a minimum, pairing a data product with its related research question, describing how best to interpret the data product, and explaining the ways in which the data product helps answer the research question.

👉 Your Turn ⤵

For your final Your Turn, your goal is to distill our analysis from above into a simple “data product” designed to illustrate key findings about changes in the collaboration network over time. For the purposes of this task, imagine that your audience consists of state-level education policy makers with limited background in SNA and adapt the following steps accordingly:

Select. Select a few network-level measures that you think would be interesting or relevant for your target audience and that help answer our research questions.
Polish. Create and polish a data visualization and/or data table to communicate your selected findings.
Narrate. Write a brief narrative (2-3 paragraphs) to accompany your visualization and/or table that includes the following:
- The question or questions are guiding your analysis;
- The conclusions you’ve reached based on your analyses and findings;
- How your audience might use this information;
- How you might revisit or improve upon this analysis in the future.

Use the code chunk below create a polished table or visualization(s) and write a brief narrative in the space that follows.

Data Visualization or Table

# YOUR CODE HERE
centr_degree(year_1_network, mode = "all")

## $res
##  [1] 5 0 5 3 3 3 4 8 4 7 4 1 0 4 4 3 2 5 4 4 5 4 3 6 4 4 4 3 8 4 4 3 3 6 3 3 4 3
## [39] 6 2 3 4 2
## 
## $centralization
## [1] 0.05102041
## 
## $theoretical_max
## [1] 3528

centr_degree(year_3_network, mode = "all")

## $res
##  [1] 76 36 80 41 68 37 73 81 65 64 75 50 56 66 73 35 68 81 38 67 62 76 52 71 58
## [26] 45 55 80 33 69 37 55 72 76 48 53 74 77 54 50 68 60 73
## 
## $centralization
## [1] 0.2423469
## 
## $theoretical_max
## [1] 3528

year_1_network_final <- year_1_network |>
  activate(nodes) |>
  mutate(degree = centrality_degree(mode = "all"))

year_3_network_final <- year_3_network |>
  activate(nodes) |>
  mutate(degree = centrality_degree(mode = "all"))

ggraph(year_1_network_final, layout = "fr") +
  geom_node_point(aes(size = degree)) +
  geom_edge_link(aes(color = weight)) +
  theme_graph()

ggraph(year_3_network_final, layout = "fr") +
  geom_node_point(aes(size = degree)) +
  geom_edge_link(aes(color = weight)) +
  theme_graph()

Narrative

Background and Research Questions:

For the purposes of our study, we sought to determine how collaboration networks among district and site-based leadership changed over a longitudinal study based in a California school district. This district was seeking to implement district-wide reforms in response to being dubbed an “In Need of Improvement” district under the No Child Left Behind (NCLB) Act. It is our belief that district-wide reforms are more effectively implemented if there is a cohesive reform vision, and communication of this vision, among district and site-based leadership. Furthermore, in order to enact this vision, knowledge must also be distributed among the change-makers expected to enact the reform. Therefore, we examine the following questions in our research:

RQ1: In Year 1, who are the influential actors within the network (measured in terms of centrality within the knowledge network)? In Year 3, who are the influential actors within the network (measured in terms of centrality within the knowledge network)?

RQ2: Do the influential actors change over the two-year period? If so, do they grow in number or shrink in number? [This is of particular interest given that the number of actors remains constant at 43.]

Since we are looking at the knowledge network, we are particularly interested in centrality because this determines if a few actors within the network are more influential than others. Given that this is a district-wide reform initiative, we would hypothesize that evenly distributed knowledge would lead to more uniform (and perhaps more effective?) implementation of reform initiatives. However, if knowledge of the reforms is limited to a few actors, this may be reflected in heterogeneous (and perhaps ineffective?) implementation across school sites.

Conclusions:

By comparing Year 1 centrality to Year 3 centrality, two distinct differences emerge. First, the degree of centrality at the node-level dramatically increases from Year 1 to Year 3. In Year 1, the highest node-level centrality value was 8; in Year 3 it is 81–a ten-fold increase. Second, the number of most-centralized actors (defined for simplicity as actors with the highest relative degree of centrality for their year, i.e. 8 for Year 1 or 80/81 for Year 3) changed from two in Year 1 to four in Year 3. Although this was a small numerical increase, it doubled in size. Furthermore, within a network of only 43 actors, 4 people is slightly less than 10%. Given the dramatic increase in overall centrality within the network from Year 1 to Year 3, this indicates that these four individuals have a substantial amount of influence within the network.

Furthermore, although we were not investigating the frequency (weight) of the ties within the nodes, the color-coded of the edges indicates that in Year 3, although the ties multiplied in quantity, the majority of these ties received a frequency score of 1 or 2, calling into question the depth of these relationships.

RQ1: In response to our first research question, we found that the following actors were most influential within Year 1 and Year 3, respectively. Given the much higher centrality score of the network in Year 3, however, we can infer that the actors listed in Year 3 have substantially more influence within the knowledge network than they did in Year 1.

Year 1: Actor 8 (node-level centrality score of 8)

Actor 29 (node-level centrality score of 8)

Year 3: Actor 3 (node-level centrality score of 80)

Actor 8 (node-level centrality score of 81)

Actor 18 (node-level centrality score of 81)

Actor 28 (node-level centrality score of 80)

Interestingly, Actor 8 remained influential from Year 1 to Year 3, whereas Actor 29’s influence (relatively) decreased significantly. In Year 3, Actor 29’s node-level centrality score was only a 33, which is actually the lowest in the entire network.

RQ2: Yes, the number of influential actors increases from 2 to 4. However, as stated above, although this doubles in size, it is still a remarkably small portion of the network population to be holding so much influence. This is distinction is particularly stark when examining the node-level centrality values in Year 3. For example, the actor with the highest node-level centrality score (81) is 48 “ties” higher than the lowest node-level centrality score (33).

Audience Implications:

These data are relevant for state-wide policy makers seeking to understand why a particular district has responded to reform initiatives in the manner that they have. If the district in question was unsuccessful in implementing required reforms, it could be hypothesized that this was because a significant amount of reform-related knowledge was housed in less than 10% of the network population. Conversely, if the required reforms were implemented successfully, it could be hypothesized that these four actors, as repositories of knowledge, were effective in implementing district-wide change despite their small number.

Future Analysis:

Future analysis would benefit significantly from data confirming or denying whether or not the district in question was able to implement reforms satisfactorily and if these reform initiatives were successful in improving student learning. Based upon the centrality scores and the fact that (according to Daly & Finnegan, 2011), most school-based leaders were on the fringes of the centrality network, I would anticipate that the reforms did not trickle-down to the students as intended. As education professionals, it is vital that we remember that no reform can truly be considered “successful”–whether it achieves district-wide implementation or not—unless it improves student outcomes. Therefore, this research would benefit significantly from data on student learning at the end of Year 3.

Furthermore, since most reforms are actually implemented at the school- and classroom-level, network data analyzing teacher-principal knowledge exchange would be useful. With the limited data available, it seems that most of the knowledge is housed at the district level, with some of it being distributed to site-leaders. It would be useful to know how this knowledge is then distributed to teachers—if at all. The current data are useful in analyzing data across leadership networks, but it raises more questions than it answers. If the leadership is so centralized, this suggests that teachers are largely ignorant of reform-related knowledge. From the outside, this seems like a fundamental flaw in the district’s reform initiative. However, further data are needed to confirm this hypothesis.

🧶 Knit & Check ✅

Congratulations - you’ve completed the Unit 2 case study! To share your work, click the drop down arrow next to the ball of yarn that says “Knit” at the top of this markdown file, then select “Knit top HTML”. Assuming your code contains no errors, this will create a web page in your Files pane that serves as a record of your work.

Once your file has been knitted, you can publish this file online using RPubs (see screeshot below), or share the HTML file through another means.

Unit 2 Case Study: Collaboration Ties over Time

ECI 589 Social Network Analysis and Education

Dr. Shaun Kellogg

January 30, 2022

1. Prepare

1a. Review the Research

Abstract

👉 Your Turn ⤵

1b. Load Libraries

Core SNA Packages

New Packages

👉 Your Turn ⤵

2. Wrangle

2a. Import Data

👉 Your Turn ⤵

2b. Create a Tidy Graph

Convert to Matrix Object

Convert to Graph Object

👉 Your Turn ⤵

3. Explore

3a. Visually Describe the Network

Sociograms with ggraph

👉 Your Turn ⤵

3b. Describe the Network Mathematically

1. Network Size

2. Centralization

3. Density

4. Reciprocity

5. Transitivity & Clustering

6. Diameter & Distance

👉 Your Turn ⤵

4. Model

👉 Your Turn ⤵

5. Communicate

👉 Your Turn ⤵

Data Visualization or Table

Narrative

🧶 Knit & Check ✅

References