Geography Reference
In-Depth Information
Table 17
.
2
Network statistics related to the pr-, co- and sm-networks
pr
P1
pr
P2
co
P1
co
P2
sm
P1
sm
P2
No. of actors
139
189
139
189
139
189
Share of largest
component
42.4%
49.7%
8.6%
31.2%
25.2%
32.3%
Share of isolates
30.2%
30.7%
51.1%
57.7%
54.7%
55.0%
Density
0.040
0.037
0.029
0.027
0.010
0.010
Mean degree
a
2.201
2.815
1.065
1.543
1.137
1.344
Network centralization
a
0.109
0.184
0.051
0.137
0.102
0.057
Clustering coei cient
2.452
2.191
3.634
2.833
0.856
0.648
Average distance
b
3.581
2.799
1.699
2.974
2.634
3.325
Note:
a
Networks have been dichotomized;
b
among reachable pairs.
we i nd a decrease in the
co
-network (0.029 to 0.027) and a rather constant value for the
sm-
network (0.010). The overall ef ect is dominated by the ef ects of cooperation, which
leads to a
pr-
network that is less dense in the second period (0.040 to 0.037). This result
suggests that the relative intensity of knowledge l ows decreased.
However, the density measure is not invariant to the size of the network. We there-
fore look at the mean degree as another measure of connectedness, which is dei ned as
md
5 S
i
51
d
i
/
g
. It states how many connections each actor on the average has. Here we
i nd an increasing connectedness in all three networks and thus an indication for an
increased l ow of knowledge and information among the actors.
Asking how this increased connectedness is distributed over the network, we can look
at the centralization index. The degree centrality of actor
i
is the number of its ties divided
by the number of possible ties
C
i
5
d
i
/ (
g
2 1) . The network centralization is then given
by
C
5 S
i
51
(
max
(
C
i
) 2
C
i
) / (
g
2 2) . For the
pr
-network, centralization increases from
0.109 to 0.184, which implies an increased importance of central actors for the l ow of
knowledge in the whole network. Looking at the other two networks, we observe a sharp
increase in the
co
-network (from 0.051 to 0.137) and a decrease in the
sm
- network (from
0.102 to 0.057). Connections through cooperation are therefore increasingly established
among or with the most central actors, while knowledge l ows through mobility are more
often found between more peripheral actors.
Another structural measure for a network is the overall clustering coei cient, where an
increase hints towards local coherence, that is, intensii ed grouping of actors in densely
connected clusters. This measure is calculated by averaging the clustering coei cients of
all actors within the network. These node-level clustering coei cients are calculated as
the density of the neighbourhood, that is, the network of actors directly linked to the
respective actor. For all three networks we i nd less clustering in the second compared to
the i rst period and thus less intense grouping of the actors. The average distance between
actors measures the ease of knowledge l ows and is often related to the rate of knowledge
dif usion (Cowan and Jonard, 2004). For the
co-
and the
sm
-network, we i nd an increase
whereas for the total
pr
-network this measure decreases. This seems to indicate that there
is a complementary relation between knowledge l ows via cooperation and via scientist
mobility.
