1 social network analysis tutorial rob cross university of virginia [email protected]
TRANSCRIPT
2
Social network analysis tutorial
Planning and Administering a Network Analysis
Visual Analysis of Social Networks
Quantitative Analysis of Social Networks
3
Planning and administering a network analysis
Formatting Data
Administering the Survey
Survey Design
Selecting an Appropriate Group
4
Social network analysis tutorial
Planning and Administering a Network Analysis
Visual Analysis of Social Networks
Quantitative Analysis of Social Networks
5
Organizational Network Analysis Software There are numerous network analysis software packages available.
We use the following.
• UCINET: Windows based tool which is used to manipulate and analyze the data. It includes a comprehensive range of network techniques. See www.analytictech.com
• NetDraw: Visualization software that creates pictures of networks. It can also incorporate attribute data into the diagrams. See www.analytictech.com
• Pajek: Sophisticated visualization software available from http://vlado.fmf.uni-lj.si
• Mage: Three dimensional drawing tool available from ftp://152.174.194/pcprograms/Win95_98_2000/
6
An Overview of UCINET
7
Transferring Data from Excel
8
Transferring Excel Matrix Data into UCINET
Step 1. Copy data from Excel
Step 2. Paste into spreadsheet editor in UCINET
Step 3. Save as “info,” etc.
9
Transferring Attribute Data into UCINET
Step 1. Copy data from Excel
Step 2. Paste into spreadsheet editor in UCINET
Step 3. Save as “attrib”
10
Opening Data in NetDraw
Step 1. File > Open > Ucinet dataset > Network Step 2. Choose network dataset (info.##h)
11
Opening Data in NetDraw
Step 1. Click - open folder iconStep 2. Click - boxStep 3. Choose network dataset (info.##h), then click OK.
12
Dichotomizing in NetDraw
Step 1. Choose “>=” and “4”
13
Using Drawing Algorithm in NetDraw
Step 1. Choose option on tool bar
Step 2. Choose = option on tool bar
14
Using Attribute Data in NetDraw
Step 1. Click - open folder icon AStep 2. Click - boxStep 3. Choose attribute dataset (attrib.##h), then click OK.
15
Choosing Color Attribute in NetDraw
Step 1. Select “Nodes” Step 2. Select “Region”Step 3. Place a check mark in the color box
16
Selecting Nodes in NetDraw
Step 1. Default is all groups selected. To remove one group, e.g. group 2, remove check from box
17
Selecting Egonets in NetDraw
Step 1. Layout > Egonets
Step 2. Choose egonet initials, e.g. BM
18
Changing the Size of Nodes in NetDraw
Step 1. Properties > Nodes > Size > Attribute-based
Step 2. Select attribute, e.g. gender
19
Changing the Shape of Nodes in NetDraw
Step 1. Properties > Nodes > Shape > Attribute-based
Step 2. Select attribute, e.g. hierarchy
20
Changing the Size of Lines in NetDraw
Step 1. Properties > Lines > Size > Tie strength
Step 2. Select minimum =1 and maximum = 5
21
Changing the Color of Lines in NetDraw
Step 1. Properties > Lines > Color > Node attribute-based
Step 2. Select attribute, then choose within, between or both
22
Deleting Isolates in NetDraw
Step 1. Select Iso option on the toolbar
23
Combining Relations in NetDraw
Step 1. Properties > Lines > Boolean selection
Step 2. Select relations, e.g. info and value
Step 3. Select cut-off operators and values, e.g. >= 4
24
Resizing and Re-centering in NetDraw
Step 1. Layout > Move/Rotate
Step 2. Select “Center” option
25
Saving Pictures in NetDraw
Step 1. File > Save diagram as > Bitmap
Step 2. Choose file name, e.g. “infoge4region”
26
The information seeking and information giving networks are both loosely connected. This represents an opportunity to improve knowledge re-use and leverage throughout the group.
I do typically seek information from this person
Density 5%
Cohesion n/a
Centrality 15
Density 5%
Cohesion 2.6
Centrality 12
Density 4%
Cohesion 2.6
Centrality 13
Density 5%
Cohesion n/a
Centrality 15
Network Measures Network Measures
Network Measures Network Measures
“From whom do you typically seek work-related information?”
I do not typically seek information from this person
“From whom do you typically give work-related information?”
I do typically give information to this person
I do not typically give information to this
27
Network Measures
Density = 3%Cohesion = 4.0Centrality = 3.1
= Location 2= Location 1
= Location 3= Location 4
Location
= Location 5= Location 6
= Location 8= Location 7
= Location 9= Location 10= Location 11= Location 12
Visual Data Display: Packing info in and allowing time for interpretation…
Information: “How often do you typically turn to this person for information to get your work done? Network includes responses to this statement of often to continuously (4,5&6).
28
Social network analysis tutorial
Planning and Administering a Network Analysis
Visual Analysis of Social Networks
Quantitative Analysis of Social Networks
29
Quantitative Analysis of Organizational Networks
Cross BoundaryAnalysis
Measures of Centrality
Measures of NetworkConnection
30
The survey data that we collect is usually valued data. Although we can use valued data in UCINET we prefer to take different cuts of the data. For example, we may want to examine the data where people only responded “strongly agree” to a question. To do this we dichotomize the data i.e. convert it to zeros and ones where one means strongly agree and zero means any other response.
Dichotomizing Valued Data
Step 1. Transform > Dichotomize
Step 2. Choose input dataset (info.##h)
Step 3. Choose cut-off op. and value (e.g. GE and 4)
Step 4. Specify output data set (infoGE4.##h)
31
Measures of Network Connection
Density• Shows overall level of connection within a network.• We can also look at ties within and between groups.
Distance• Shows average distance for people to get to all other people.• Shorter distances mean faster, more certain, more accurate
transmission / sharing.
Network Connection Centrality
Cross Boundary Analysis
32
Density
Number of ties, expressed as percentage of the number of pairs Dense networks have more face-to-face relationships
Low Density (25%)Avg. Dist. = 2.27
High Density (39%)Avg. Dist. = 1.76
Network Connection Centrality
Cross Boundary Analysis
33
Quantitative Analysis: Density
Step 1. Network > Cohesion > DensityStep 2. Input dataset “infoge4.##h”
Density of this network is 8%. Density of this network is 8%.
Network Connection Centrality
Cross Boundary Analysis
34
Distance
Average number of steps to reach all network participants Lower scores reflect a group better able to leverage knowledge
Short average distance Long average distance
Network Connection Centrality
Cross Boundary Analysis
35
Quantitative Analysis: Distance
Step 1. Network > Cohesion > DistanceStep 2. Input dataset “infoge4.##h”
Average Distance is 3.5 Average Distance is 3.5
Network Connection Centrality
Cross Boundary Analysis
36
Measures of Centrality
Degree Centrality: How well connected each individual is.
Betweenness Centrality: Extent to which individuals lie along short paths.
Closeness Centrality: How far a person is from all others in the network.
Network Connection Centrality
Cross Boundary Analysis
37
Degree Centrality
x
How well connected each individual is Technical definition: Number of ties a person has
y
Communication Networkdegree of X is 7
Seek Advice Networkin-degree of Y is 5
Network Connection Centrality
Cross Boundary Analysis
38
Closeness Centrality
How far a person is from all others in the network Index of how quickly information can flow to that person Technical definition: Total number of links along shortest paths
from the individual to each other individual
c
a f
d
b
e
g
h
ij
Closeness of F is 13
Network Connection Centrality
Cross Boundary Analysis
39
Betweenness Centrality
Extent to which individuals lie along short paths Index of potential to play brokerage, liaison or gatekeeping Technical definition: number of times that a person lies along the
shortest path between two others, adjusted for number of alternative shortest paths
c
a f
d
b
e
g
h
j
k
m
l
Betweenness of h is 28.33
Network Connection Centrality
Cross Boundary Analysis
40
Without the twelve most central people the network is 26% less well connected, reflecting a vulnerability in the group
Network Measures
Density = 5%Cohesion = 2.6Centrality = 12
Network Measures
Density = 3%Cohesion = 2.8Centrality = 9
Without 12 central people
“From whom do you typically seek work-related information?”
Responses of I do typically seek information from this person
41
Pulling People Dynamically From the Network…
42
Quantitative Analysis: Degree Centrality
Step 1. Network > Centrality > Degree
Network Connection Centrality
Cross Boundary Analysis
43
Quantitative Analysis: Degree centrality
Step 2. Input dataset “infoge4.##h”Step 3. Choose whether to treat data as symmetric. If you choose “no” it will calculateseparate figures for the people you go to and the people that go to you.
Network Connection Centrality
Cross Boundary Analysis
44
Quantitative Analysis:Degree Centrality
In-degree for HA is 7In-degree for HA is 7
Network Connection Centrality
Cross Boundary Analysis
45
Quantitative Analysis: Degree Centrality
Average in-degree is 3.7Average in-degree is 3.7
In-degree NetworkCentralization is 12%
In-degree NetworkCentralization is 12%
Network Connection Centrality
Cross Boundary Analysis
46
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00
175
302
111
279
105
308
47
26390
273
37
51
276
300
17615
22
240
177 160
139
101
43
74
316
234
30
117
231
192
143
57258
81
312
205257
195
188
255
315292
173
99
2
256
224
178
106
241
75
113
246 149
145116
78
191
140
222
202
118
242
193
54
296
89102148
19
6
248
32
35
295
230
270
91
223
201
45
3
198
163
164
209167
217 38
93
20634
61
174 211
303
112
144
265
1
187
7
69
212
155
5
299
10
189
26
247
16
27 153 216
243268
95
147
23237
170
301
311
266
249
119
28
52
29
92
169
100
82
12050
269
280
221
278
59
210
141
60
132
239
55
171
36
294
245
229
18548
39 220
275
131
2339 184
56
67
8
135
136
24
213190
196
127
158
264
286
272
183
133
281
197203199
44
53
87
244
14
314317
126
Opportunities exist to re-distribute relational load. Focus on ways to de-layer those in the top right quadrant (info access, decision rights, role)
while also better leveraging those in the bottom quadrant
# People Each Person Seeks Information From
# Pe
ople
Rec
eive
s In
form
atio
n Fr
om
High Info Sources
High Info Seekers
Integrators
“From whom do you typically seek work-related information?”
* Calculations based on people who responded to the survey only
47
0
10
20
30
40
50
0 10 20 30 40 50
BKA/BA/Research Analyst
Assoc/Know. Assoc
Speialist/Sr. Spec
Manager
EM/PKM
Assoc Principal
Partner
External
Admin/Assistant
Opportunities exist to re-distribute relational load. Focus on ways to de-layer those in the top quadrant (info access, decision rights, role) while also better leveraging
those in the bottom quadrant
# People Each Person gives Information To
# Pe
ople
Rec
eive
s In
form
atio
n Fr
om
High Info Sources
High Info Seekers
Integrators
48
Predicting Satisfaction
Social Network Level of Satisfaction:NeutralSatisfiedVery Satisfied
• There is a statistically significant relationship between Social OutDegree and Level of Satisfaction. (0.022)
• Correlation: 0.375
49
Showing performance implications can quickly get people’s attention…
HelpOut HelpIn KnowOut KnowIn KnbefOut knbefin SocOut SocIn Sat10 13 36 30 34 30 25 24 310 14 16 32 26 24 27 35 30 2 6 4 3 1 6 5 31 6 17 26 22 22 15 17 30 3 10 6 4 6 0 3 3
12 5 31 16 22 18 22 19 40 5 3 19 23 26 3 12 43 6 28 30 11 15 25 25 45 8 14 19 12 15 16 19 4
16 20 30 39 34 34 38 37 48 10 34 36 29 29 19 29 4
19 15 42 35 40 37 22 22 47 10 33 31 22 21 34 34 4
53 31 38 37 34 33 22 28 413 8 34 29 10 7 34 30 423 18 38 34 27 28 29 28 49 9 26 19 14 14 28 23 5
11 13 39 31 15 18 43 36 5
50
Cross-boundary Analysis
Density across boundaries: How connected are groups within themselves and with other pre-defined groups. This view can be used for different boundaries. We have used the following in our research:
• Function or other designation of skill or knowledge.• Geographic location (even if only different floors).• Hierarchical level.• Time in organization or time in department.• Personality traits.• Gender (interesting though may be inflammatory).
Brokers: Which individuals are the links between other groups. Brokers can be beneficial conduits of information but they can also hold up the flow of information.
Network Connection Centrality
Cross Boundary Analysis
51
Cross-boundary Analysis
Information Network: Density as related to practicePlease indicate how often you have turned to this person for information or advice on work-
related topics in the past three months (response of often or very often).
Healthcare Government IT Oil & Gas Pharmaceuticals IndustrialHealthcare 17% 0% 0% 7% 38% 0%Government 0% 17% 0% 0% 0% 10%IT 0% 0% 0% 0% 0% 6%Oil & Gas 4% 0% 0% 19% 3% 8%Pharmaceuticals 35% 0% 0% 1% 49% 0%Industrial 1% 9% 9% 12% 1% 8%
Network Connection Centrality
Cross Boundary Analysis
52
Density Across Practice
Step 1. Network > Cohesion > DensityStep 2. Input dataset “infoge4.##h”Step 3. Row Partitioning “Attrib col 3Step 4. Column Partitioning “Attrib col 3
Tip: Col 3 is the column that includes the practice attribute. You can selectdifferent columns for different attributes
Tip: Col 3 is the column that includes the practice attribute. You can selectdifferent columns for different attributes
Network Connection Centrality
Cross Boundary Analysis
53
Broker Categories
Coordinator - This person connects people within their group.Ego
A B
Gatekeeper - This person is a buffer between their own group
and outsiders. Influential in information entering the group.
A
Ego
B
Representative - This person conveys information from their
group to outsiders. Influential in information sharing.
B
Ego
A
Network Connection Centrality
Cross Boundary Analysis
54
Quantitative Analysis: Broker Metrics
Step 1. Network > Ego networks > BrokerageStep 2. Input dataset “infoge4.##h”Step 3. Partition vector “attrib col 2”
Tip: Col 2 is the column that includes the gender attribute. You can selectdifferent columns for different attributes
Tip: Col 2 is the column that includes the gender attribute. You can selectdifferent columns for different attributes
Network Connection Centrality
Cross Boundary Analysis
55
Additional Quantitative Analysis
Symmetrization & Verification
Scatter Plots
Combining Networks
QAP Correlation and Regression
56
Symmetrizing Data
Bill says he communicated with John last week, but John doesn’t mention communicating with Bill
Three options
• take the conservative option, and put no tie between John and Bill (minimum)
• take the liberal option, and put a tie between John and Bill (maximum)
• take the average, assigning a tie strength of 0.5 for the relationship between John and Bill (average)
Bill John
57
Symmetrizing Data (Continued)
Step 1. Transform > SymmetrizeStep 2. Input dataset “infoge4.##h”
Step 3. Symmetrizing method “maximum”Step 4. Output dataset “Syminfoge4.##h”
Tip: See previous slide for how to choose the most applicable symmetrizing method.
Tip: See previous slide for how to choose the most applicable symmetrizing method.
58
You have both “Give information to” and “Get information from” networks If A says they give info to B, then B must say that they get info from A
Verification of Asymmetric Data
Step 1. Tools > Matrix algebraStep 2. In the Enter Command box type “newinfo = average(transpose(infofrom),infoto)”Step 3. Enter
Tip: The new matrix “newinfo” cannow be used for various visual and quantitativeanalysis.
Tip: The new matrix “newinfo” cannow be used for various visual and quantitativeanalysis.
59
Scatterplots
Step 1. Create attribute file spreadsheet editor in UCINET. Each column is takenfrom the In-degree numbers in the Degree Centrality function.Step 2. Save as “Indegree”
60
Scatterplots (Continued)
Step 1. Tools > ScatterplotStep 2. File name “Indegree”Step 3. Choose X and Y axis
Step 4. To move initials – point and click Step 5. To save - File > Save as
61
Combining Networks In the picture to the left you can
see the information network.
In the picture below is the combined information and value network.
62
Combining Networks (Continued)
Step 1. Tools > Matrix AlgebraStep 2. In the Enter Command box type “infovalue = mult(infoge4,valuege4)”
Tip: The new matrix “infovalue” can now be used for various visual and quantitativeanalysis.
Tip: The new matrix “infovalue” can now be used for various visual and quantitativeanalysis.
63
QAP Correlation
Step 1. Tools > Testing Hypothesis > Dyadic (QAP) > QAP Correlations Step 2. 1st Data Matrix “InfoGE4”Step 3. 2nd Data Matrix “ValueGE4”
64
QAP Regression
Step 1. Tools > Testing Hypothesis > Dyadic (QAP) > QAP Regression > Original (Y-permutation) method
Step 2. Dependent variable “InfoGE4”Step 3. Independent variable “ValueGE4”
Adjusted R-Square of 0.214 indicates a moderate relationship between the two social relations.
Theprobability of 0.000 indicates that it is statisticallysignificant.
Adjusted R-Square of 0.214 indicates a moderate relationship between the two social relations.
Theprobability of 0.000 indicates that it is statisticallysignificant.