Results
Visualization
A visual examination of the networks in Figure 21.1 highlights differences not only between data sets (e.g., Frat, Ham) but also between networks in the same data set. For example, the Frat behavioral network has a very dense core, which does not seem to exist in the recall network. Similarly, the Tech recall network is almost separated into two distinct components, which is much less obvious when looking at the Tech behavioral network.
Figure 21.1. Visualization of four pairs of networks.
Table 21.2. Descriptive statistics of four networks for behavior and recall
|
FRAT |
HAM |
OFF |
TECH |
|||||
|
behavior |
recall |
behavior |
recall |
behavior |
recall |
behavior |
recall |
|
|
Size |
58 |
44 |
40 |
34 |
||||
|
Density |
0.12 |
0.064 |
0.022 |
0.0074 |
0.13 |
0.13 |
0.13 |
0.12 |
|
Average degree |
6.79 |
3.65 |
0.95 |
0.38 |
4.95 |
5.15 |
4.47 |
4.12 |
|
Clustering |
0.61 |
0.35 |
0.68 |
0.58 |
0.25 |
0.38 |
0.46 |
0.43 |
|
Centralization |
46 |
13 |
17 |
9 |
8 |
8 |
11 |
12 |
|
QAP correlations |
0.29 |
0.46 |
0.32 |
0.34 |
||||
Preliminary Statistical Analysis
The key statistics presented in Table 21.2 provide little evidence of systematic differences between the behavioral and recall networks. For the Frat and Ham data sets, the recall network is much sparser than the behavioral network, whereas the densities are similar for the two networks in the Off and Tech data sets. QAP correlations confirm the moderate level of correspondence in the global structure of the networks that Bernard et al. (1979) found at the actor level.
