Transcript Public opinion leaderships analysis using

Contribution to
Wapor 68th Annual Conference
Buenos Aires, Argentina
June 16 - 19, 2015
"The Networks of Public Opinion: New
Theories, New Methods"
Public opinion leaderships analysis using
methods of social network analysis (SNA)
Miguel Oliva. Sociologist, Universidad Nacional de Buenos Aires; Professor Universidad Nacional
de Tres de Febrero. Phd. Candidate at UBA.
Professor in Maestría en Generación y análisis de Información Estadística and Maestría en
Investigación Social, Universidad de Bologna, Buenos Aires.
Introduction
This paper addresses the issue of the application of
formal models using methods of social networks
analysis (SNA) to political leadership analysis.
Study of political leadership today
Issues:
• How is leadership studied with polls?
• How can we study leadership and networks
with SNA?
• How can we use social network models (and
the idea of latent or potential networks
models) in these analyses?
Leaderships are important in politics.
¿How do we study today leaderships in public opinion
polls?
For political action, it is useful to identify influential leaders.
In 1944, the great methodologist Paul Lazarsfeld, published "The People's
Choice" (1944) and "Voting" (1954). There the "opinion leaders" were studied
within a network of relationships of direction of public opinion; constituting
competitive leadership circuits in the community. In such a network it is possible
to isolate and identify those individuals who are closest to occupy a central
position relations, the so called "opinion leaders"; they are of greater interest
and competence in the field of discussion, and representativeness within those
influenced.
Today, most adequate methodology for studying this subject, is the social
network analysis, whose techniques can detect individuals with greater influence
and centrality (nodes with greater centrality in terms of these theories).
Political leaders
• Political parties have leaders who deal with and respond to social
demands. Usually (but not always), the most important leaders of political
parties are the presidents, with a symbolic importance for the political
party.
• At the same time, and in the words of Luhmann (1991), the language has a
structural coupling with consciousness; and in that sense, there is
awareness (consciousness) because there is language, and vice versa.
• So, what politicians say is evaluated in relation to their actions (i.e., it is
analyzed to what extent their language is associated with his conscience, or
in other terms, the correspondence between what they say and what they
will do), in the same way as other social leaders (rabbis , priests, community
leaders).
• Political discussions between leaders of different political orientations tend
to reproach mismatch between saying and doing.
Political leaders in Latin America
In Latin America it has been very common to have strong
leaderships, in the nineteenth century (caudillos, etc.), and
populism in twentieth century.
Many of these leadership structures are observed leaders in various
forms of populism in Latin America.
Current methods of measurement of opinion about leaders in polls
are insufficient for analyzing these leadership structures.
Leaders evaluation
• When a leader is analyzed and rated, it is necessary to consider different types
of opinions, usually transmitted in semantic scales as very good, good, fair,
bad, very bad; they are usually summarized as positive (very good, good) and
negative (bad, very bad) image. Or, respondents are asked to evaluate the
leaders on a simple 0-10 scale (0=strongly dislike and 10=strongly like).
• Some individuals who know the leader A, do not have sufficient information to
evaluate him. Also we find people who do not know about the leader A. In
that regard, it is relevant to consider the knowledge and ignorance of the
leader; it is assumed that the greater the knowledge (visibility) of a leader, the
greater is his political capital or influence.
• One way of analyzing this information is simply to calculate a share of positive
assessments of the negative.
Evaluation of leaders in polls
Evaluation scale (1-10)
Respond Leader A Leader B Leader C
ent
1
0
1
2
2
10
9
5
3
5
6
6
4
6
7
9
5
7
8
1
Today we study leadership as the image of a
political leader. This is performed with usual
databases, with units of analysis, variables
and data.
• Evaluation scale: 0-10 scale
(0=strongly dislike and
10=strongly like).
• We can summarize means,
standard deviation (how
much heterogeneous is a
leader evaluation -polarization between those
who “love” or “hate” the
leader --.
• Correlations between
evaluations.
Matriz de Correlaciones: Imagen de líderes políticos de América Latina
TABARÉ VASQUEZ
FIDEL CASTRO
LUIZ IGNACIO LULA
HUGO CHÁVEZ
MICHELLE BACHELET
EVO MORALES
DA SILVA
R Pearson
TABARÉ VASQUEZ
FIDEL CASTRO
LUIZ IGNACIO LULA DA SILVA
HUGO CHÁVEZ
1
Sig. (2-tailed)
,213**
,423**
,231**
,449**
,354**
,000
,000
,000
,000
,000
N
22694
22694
22694
22694
22694
22694
R Pearson
,213**
1
,382**
,656**
,298**
,385**
,000
,000
,000
,000
Sig. (2-tailed)
,000
N
22694
22694
22694
22694
22694
22694
R Pearson
,423**
,382**
1
,415**
,531**
,565**
,000
,000
,000
,000
,000
N
22694
22694
22694
22694
22694
22694
R Pearson
,231**
,656**
,415**
1
,344**
,469**
,000
,000
,000
,000
,000
N
22694
22694
22694
22694
22694
22694
R Pearson
,449**
,298**
,531**
,344**
1
,580**
,000
,000
,000
,000
N
22694
22694
22694
22694
22694
22694
R. Pearson
,354**
,385**
,565**
,469**
,580**
1
,000
,000
,000
,000
,000
22694
22694
22694
22694
22694
Sig. (2-tailed)
Sig. (2-tailed)
Sig. (2-tailed)
,000
MICHELLE BACHELET
EVO MORALES
Sig. (2-tailed)
N
22694
**. Correlation is significant at the 0.01 level (2-tailed).
Here we see that the evaluation of Hugo Chavez is correlated with the evaluation of
Fidel Castro. Source: Latinobarómetro
• Data from network analysis has a different
format, capturing relations between nodes:
a
c
d
e
f
g
1
1
1
1
1
1
c
1
1
1
1
1
1
d
1
1
1
1
1
1
e
1
1
1
1
1
1
f
1
1
1
1
1
1
g
1
1
1
1
1
1
a
b
b
Political networks
Political social networks
•
•
•
•
Populist Latin-American political leaders created latent networks similar to religious
connections, which have long outlived their leaders (i. e. “peronismo”). The leader
builds connections, and the nodes connect through them.
Populism leadership usually aims to replace the lack of economic influence (for
instance workers or poor people) with political power.
These potential networks have also an intergenerational extent, and include central
nodes occupied by individuals (leaders) who are no longer alive.
For a social leader, sometimes death is usually a booster of the connections properties;
the connections are not interfered by vital decision of the leader, and so there is no
interference between the nodes connected by the central node. Everyone may
connect through a dead node, but the dead node will connect with no one.
“Peronismo” in Argentina is a network with a central (node), a non-living leader
(Perón). Maybe Hugo Chavez in Venezuela has developed a similar latent network. The
central node is a bridge of connections between other members of this network, and a
source of political power.
Individuals are connected through a leader.
This is a star network. Node a connects to b through “Lider”. In a similar way, in prophetic and monotheistic religions a
central node (deity) connects believers .
Where the ties among actors have been measured as a value (rather than just present-absent), the magnitude of the tie
can be suggested by using thicker lines to represent stronger ties, and thinner lines to represent weaker ties (as node f in
the graph). Sometimes ties are measured as negative-neutral-positive (-1, 0, +1) in the relation with the leader, as
grouped ordinal (5=very strong, 4=strong, 3= moderate, 2=weak, 1=very weak, 0=absent), full-rank order (10=strongest
tie of 10, 9=second strongest tie of 10, etc.), or interval (e.g. dollars, bolivares or pesos flowing from a central political
organization to a clientelistic political network participant).
MULTIPLE CENTRALITY MEASURES (OLD)
•
--------------------------------------------------------------------------------
Input dataset:
lider
(C:\Users\Usuario\Documents\UCINET data\lider)
Normalized Centrality Measures
1
2
3
4
5
6
7
•
•
a
b
c
d
e
f
Lider
1
2
3
4
Degree
Closeness Betweenness Eigenvector
------------ ------------ ------------ -----------16.667
54.545
0.000
40.825
16.667
54.545
0.000
40.825
16.667
54.545
0.000
40.825
16.667
54.545
0.000
40.825
16.667
54.545
0.000
40.825
16.667
54.545
0.000
40.825
100.000
100.000
100.000
100.000
Historically first and conceptually simplest measure is degree centrality, which is defined as the number of links incident upon a node (i.e.,
the number of ties that a node has). The degree can be interpreted in terms of the immediate risk of a node for catching whatever is flowing
through the network (such as a virus, or some information). In the case of a directed network (where ties have direction), we usually define
two separate measures of degree centrality, namely indegree and outdegree.
Betweenness is a centrality measure of a vertex within a graph . Betweenness centrality quantifies the number of times a node acts as a
bridge along the shortest path between two other nodes.
Political connections through leaders
An important political leader creates symbolic (badges, songs) and material
(money, resources, budget control) connections between his followers.
Sometimes, this membership to a political movement or organization creates a
meaning of life for the members and followers.
This type of network is usually not influenced by political marketing initiatives
and classical political advertising or campaigns.
Typically a populist political network with a central leader can be modeled as
centralized.
Also, we can measure the network centralization; a star network is centralized.
100 nodes star network
Even if the political leader is dead, and the node has not a physical existence, all the other
nodes and relationships between them are strictly empirical. Between two individuals
who share the same leader, a potential connection exists, mediated through the leader.
Political networks
A political network is usually not superimposable with other political network.
They are positional (a “political position”). If a person is leftist, he cannot be at
the same time right wing. If you are a democrat you cannot be at the same time,
republican.
This relates to the concept of spatial voting; and positions in politics. Also, in
Schmitt (1966) we find his best-known formulation, the distinction between
friend and enemy (Schmitt, 1996, p. 27); this concept is similar to that of non
superimposable networks.
In a similar way, religion networks are not superimposable. You cannot be
catholic and muslim at the same time.
Usually people who are not in political networks are “undecided” in voting
decision.
Coordination and power
In social activity, coordination and star like networks are usually
associated with power. A coordinated group with a central node or
leader, is more powerful.
The leader has a role in power construction. In some historical
circumstances, the main objective of the members of an organized
group is to be coordinated (adept, affiliates, voters, union
members); in this cases, the ideological or philosophical orientation
of the leaders, or if there are corrupt, may become less important,
regarding the main objective of the members: to be coordinated,
and to maintain a latent network cohesive in time.
In lationamerican history we can find “caudillos”, a term used to refer to a leader
whether political, military or ideological.
In a broad sense this term is used for any person making guiding others on any
terrain, but the word “caudillo” has a certain political connotation, usually used
as a reference to political leaders of the 19th and 20th centuries.
The emergence in the 19th century of numerous warlords in various South
American countries, was a social phenomenon called caudillismo (some of them,
José Gervasio Artigas, Ángel Vicente Peñaloza, Sergio Arboleda (Colombia).
• Social networks
Social networks
Social networks are interactions between nodes,
connected with different kind of social interdependency.
Formalizations and models can be applied to the
analysis and visualization of social networks, with
specific software (v.g. UCINET, Hanneman, 2008;
Pajek, Igraph – R language - ).
In networks analysis we can formalize social interaction
occurring simultaneously, or changes in these networks.
• Latent social networks
Potential or latent networks
A real connection is different from a potential connection. A potential or latent
network can be thought as a set of possible relations – i. e. the arrangement of
organizational roles in an institution is as a potential or latent network between it´s
members --.
Religions are potential or latent networks.
In prophetic religions, faithful are potentially connected through central nodes
(usually deities or gods).
We can also describe languages as potential networks. An individual who speaks a
language participates in a potential network of semantic conventions. If you are
Spanish speaking, you can potentially interact in different ways with other speakers
of that language. If you are not, you will be not included in that potential network.
If is the use of a language is more widespread, the corresponding potential
network is bigger. That is the reason for what we want to study english, more
often than African dialects.
Latent and saturated
Networks
In the Figure a network
of four nodes is
represented with
program UCINET (a, b,
c, d) that have all the
six possible
bidirectional
connections among
them. We call this
connection a saturated
connection -- it has all
the possible
interactions --. Usually
a latent networks has
this characteristic.
Individuals are usually engaged in various latent social networks.
Usually the opinion of an individual are influenced by his membership to
different potential social networks.
The opinions (and voting) usually are influenced by social networks inclusion.
Political networks tend to overlap with other latent social networks.
•
We detect two problems:
a) To be or not in a latent network.
b) Once in the latent network, to
interact or not to interact with the
other nodes.
Problems a) and b) are different:
•
a) A simple example would be to participate or not in a social
network (regardless of the connections you establish with
other members of that network). Once you are a registered in
Facebook or Twitter, you can potentially interact with all the
possible connections (although nobody interacts with all the
possible connections in this social network, simply because
they are too much). This is problem a).
Another example is to refer or not to certain object with a
specific and arbitrary word (such as “tree”, or “árbol”); that is
for example, being spanish - speaking or not.
In the same way, you may find people interacting or not with
political movements.
b) Once connected to the network (facebook in the example, or
political movements) you establish different real interactions
with other people in that social network. Or if you are English
speaking, you communicate with other people who speak
English.
Problem a: to be or not a supporter of a political movement
(peronista / no peronista), use or not a word (language),
exchange of goods in a network (money).
Examples of latent networks:
A religion with a monotheistic God can be understood as unidirectional interactions of the
nodes to a central node (God) which form a network. This central node usually has a
subjective, or spiritual existence. The effects are similar to those of node that has an
empirically verifiable existence. Connections between living believers in that God really
exist. The central node Deity (or God) creates potential connections between believers.
If you own money of a specific currency you can interact in different ways (i.e. buy
something) in a potential network – usually referred as a “market”--, in which all of the
nodes have a potential agreement to exchange goods or services using that currency. If
you do not own that currency, you are not included in this particular potential network.
Even there is no central node in this kind of network, we find that some people have
more money, or control more resources, and so they have more monetary interactions.
In the same way, social classes could be studied as latent networks, usually with less
organizational basis (i.e. there is no managers or chiefs of the working class, but there are
classist political organizations).
Usually, political networks reproduction should be coordinated and not conflictive
with other existing potential social networks.
Political leaders of mass parties in competitive democracies, are compelled or
required not to be competitive with the potential social networks that already exist;
they cannot be conflictive (or should have a condescending attitude) with social
networks already existing in a community as religion, monetary interchanges,
language, or racial - genetics networks.
Successful political leaders usually mount on existing social networks, rather than
creating new ones, since these networks usually develop in the history of a country,
and outlive the leaders short human life.
In a catholic country, it is conflictive to have a Muslim president (this is the problem
treated by Houllebecq in “Submission”, a story premised on voting between a FN
candidate and an Islamic candidate in France). Or a president with conflictive plans
with a network of monetary exchanges (abolish private property, i.e.), or in a
language speaking country an English speaking president, or in a country with racial
Anglo-Saxon predominance , a black president is a strange event.
A “Nation” can be understood also as a latent network. International ideas as
communism are sometimes conflictive with the national scope and latent network
(as we find in the peronist movement). In some cases, these latents networks have
an international scope, like communist, socialist, and other organizations with leftist
ideas. Usually in caudillismo and nationalistic populistic movements, the scope is
national, and with territorial references (which in it´s international scope usually
communism or socialism do not have).
The opinion of an individual are influenced by his membership to different potential
social networks. Opinions (and voting) usually are influenced by the social networks
inclusion of the individual.
10 nodes random network
Random networks do not describe the usual situation of a populist
leadership with a central node.
Random network 200 nodes
A complete random network of 200 nodes shows the complexity of
complete networks. Imagine networks of populations as 3 in CABA, 3
millions. There are impossible to study in a complete format.
Ego Nets
•
•
•
•
•
Because of this problem, we should use Ego network.
Surveys may be used to collect information on ego networks. We can ask each
research subject to identify all of the actors to whom they have a connection, and to
report to us (as an informant) what the ties are among these other
actors. Alternatively, we could use a two-stage snowball method; first ask ego to
identify others to whom ego has a tie, then ask each of those identified about their
ties to each of the others identified.
Data collected in this way cannot directly inform us about the overall
embeddedness of the networks in a population, but it can give us information on
the prevalence of various kinds of ego networks in even very large populations.
When data are collected this way, we essentially have a data structure that is
composed of a collection of networks. As the actors in each network are likely to be
different people, the networks need to be treated as separate actor-by-actor
matrices stored as different data sets.
Robert A. Hanneman (Department of Sociology, University of California, Riverside) and Mark Riddle (Department of Sociology, University of Northern
Colorado)
Introduction to Ego Network Analysis ©2008 Halgin & De
Network characteristics
These network models can show different characteristics.
Suppose to interact or not in a network it is equally probable (0.5,
or 1 / 2).
If each of the four individuals (a, b, c, d) has a probability of 1 / 2
to interact with the network (i.e., to be or not to be in a social
movement), the probability of a saturated connection is equal to
1 / 16 (0.0625).
The probability
decreases
exponentially.
With these
assumptions, the
probability of a
network of 7
simultaneous
interactions
(similar to Figure 2)
is 1 / 128,
0.007812.
Figure 2
Language
One of the specificities of human societies
is its coordination through generalization
of symbols, such as money (Parsons,
1982).
• In a language network the use of a
particular symbol to designate an
object, a symbolic coding of an object
is a connection; the coding is a
arbitrary word (“tree”, or “arbol”,
“arbre”, “Baum”) in a language.
• Use of the word "tree" in this network
corresponds to a 1, and the contrary not using this word - carries a 0.
• A spontaneous coordination of the use
of a word - an arbitrary code - for
addressing an object is very unlikely.
• Also, a word used by a greater number
of individuals, it more unlikely that a
less widespread word.
Suppose we have a dichotomic value of connecting to Facebook
or not. If we suppose a probability of connection, say, ½ (we do
not actually know this probability, but we know it is a number
between 0 and 1).
For 500 million users of the social network, we have a
probability of 1 / 2 ^ 500.000.000, a very low probability. This
kind of modeling shows how strange is social coordination.
Also, these figures suggest the fact that we need highly
sophisticated social technology for coordination of social action
(political leadership, language, money, and others).
Configuration of latent networks have different probabilities.
Conclusions
• Leadership studied with polls nowadays, can
be enriched by the analysis of networks with
SNA.
• By introducing the idea of social network –
and Latent or potential networks – Models,
the leadership analysis is enriched.
Conclusions
• Individuals are usually engaged in various latent social
networks: economic and monetary exchange, language,
religion. Social classes should be studied as latent networks,
usually with less organizational basis.
• The opinions (and voting) usually are influenced by social
networks inclusion.
• Political networks tend to overlap with other potential social
networks.
• Usually the opinion of an individual are influenced by his
membership to different potential social networks.
• Political networks are typically not conflictive with other existing
potential social networks.
• Political leaders of mass parties in competitive democracies, are
compelled to coordinate opinions and visions with the potential
social networks that already exist; usually they cannot be
conflictive with monetary interchanges, language, religions, or
racial - genetics networks already existing.
• Successful political leaders usually mount on existing social
networks, rather than creating new ones; these networks usually
are developed in long periods of the history of a country or
culture, and outlive the leaders short human life.
Conclusions
• A statistical model of a social phenomenon
may clarify a theory (explanatory function),
and also to represent a recurring process in
abstract form (Farraro, 1997, op. cit),
enriching the dominant systematic empiricism
of the social sciences (Willer, 1996).
Formalization helps us understand the nature
of the phenomenon.
Conclusions
• The development of models of SNA (Socio
centric or Ego centric ) and the collection of data
about networks is potentially useful, and may
improve the understanding of phenomena such
as power, public opinion, political leadership and
political movements.
• Including modeling, even without a direct
reference to empirical observable data, helps to
improve the understanding of social phenomena
and to increase the relevance of social science.
Thank you
[email protected]