2. Background: Voting in the Colombian Congress
Two chambers, House (~165)
and Senate (~100):
Votes occur on floor and on 7
committees (~20 members)
60% reelection rate
Numerous small parties
Members are independent in
practice
3. A Peculiar thing about the Colombian Congress
“In this Congress we are free to
vote however we want and do
whatever we want and the party
never punishes us for it ”
- Omar Yepes (Colombian Representative)
4. Parties not organizing voting behavior
Survey of Legislators in Lat Am (PELA) MDS Scaling of votes House 2010-2014
5. More Evidence of Parties not organizing
Cosponsorship Colombia
Lower House 2010-14
Cosponsorship Network France
Lower House 2012-2016
https://f.briatte.org/parlviz/parlement/
Cosponsorship USA
Lower House 2015-2016
(from Neal 2018)
6. Political Science: without parties, voting is not stable
● Political Science focuses on how parties are needed to avoid "chaos"
● Most influential theories see party leadership as constraining voting
● Parties structure the choices available in legislatures
● This literature does not account for contexts where parties do not serve this
purpose and decentralized relationships are the basis for legislative
organization.
POLISCI EXPECTATIONS: VOTING COALITIONS IN COLOMBIAN CONGRESS
WILL BE UNSTABLE!
7. Research Question
● Consistency in patterns of behavior can emerge without external constraints
like parties
● There are dynamics in which interactions between individuals become
reinforced, resulting in synchronized actions
● If parties do not constrain members, are legislators voting independently
bill-by-bill? Or are there detectable groups of legislators that vote
together across time?
8. Methods
● Get votes on bills from members of the Colombian House of Representatives
for the 2010-2014 Congressional Period
● Build a Dynamic Co-Voting Network
● Apply community detection algorithm to Co-voting network at different
“snapshots” in time
● Track community membership over time
9. Data
● Scraped voting data
from Congreso Visible
● Cleaned Data.
Selected votes from
House from 2010-2014
● Result: dataset of
1,325 voting sessions
for 168
● 195,632 individual
votes
10. Time-stamped co-voting edge-list
● Three different types of
votes:
○ (1) No vote
○ (2) Yes vote
○ (0) Abstention
● Create network for every
voting session
○ Nodes are legislators
○ Build a link between them
if they vote the same way
○ Add time stamp that
reflects when voting
happened
11. Multi-layer Co-Voting Network
● Flatten the temporal dimension by
binning data into smaller static graphs
called snapshots
● Create static snapshots that represent 2
months of voting
● All of the times every pair of legislators
voted in the same way during that
two-month period are added resulting in
weighted links
● Multi-layer weighted undirected network
with 20 layers (one for each snapshot)
13. Community Membership Evolution
● Community detection and matching across snapshots to track the evolution of
the co-voting over time
● Test: Louvain algorithm for community detection on each of the layers
● Create a temporal community similarity matrix
15. How similar are these communities over time?
● Jaccard Similarity Index:
compares members for two
sets to see which members
are shared and which are
distinct.
● Communities at different
points in time appear to share
members
● Suggests consistency rather
than randomness
16. Next Steps:
● Compare and Evaluate Community Detection methods. Benchmark
comparisons (method evaluation)
● Explore Community Evolution with mainstream Community Tracking methods:
He, Ziwei, Etienne Gael Tajeuna, Shengrui Wang, and Mohamed Bouguessa. "A Comparative Study of Different Approaches
for Tracking Communities in Evolving Social Networks." In 2017 IEEE International Conference on Data Science and
Advanced Analytics (DSAA), pp. 89-98. IEEE, 2017.
17. Next Steps:
● Explore attributes that might drive community membership
Party
Region/constituency
Length of
tenure
Bill Sponsorship
Committee Membership
Biographical
data