|
Blockchains,
shared secure data bases. People
can choose among them. There are
a diversity of ways people can vote – by coin, by person, by
representatives. Proof of
work, those who mine coins is
more like an oligarchy, where proof of stake is more like a democracy. He explores how
people sort using a model of Tiebout Sorting. The model: there are different institutional
arrangements spread across jurisdictions. Each jurisdiction has parties as well
as citizens. The citizens vote,
and then the parties adjust.
Citizens possibly move if they find a better jurisdiction. Both parties and
citizens have binary vectors of ideal points: smart contracts, transaction
fees, voting, energy consumption,
usability, speed. Each can
be zero or one. Results: He
then compares different institution and finds that democratic referenda
attracts more people than oligarchy as does direct competition and
proportional voting. Oligarchies
do not respond much to preferences.
Overall, he finds that proportional representation does best, followed
by direct competition and then democratic referenda and oligarch. He then runs a
free for all with all four types.
He finds that proportional representation does best and oligarchy does
worst. If he gets rid of the
oligarchies then proportional representation still wins by not as much. Last, he
compares a situation with multiple oligarchies and one of each of the
democratic institutions. He finds
that oligarchy again loses but is more stable. He finds that with a lot of
oligarchies and a handful of democratic institutions, then the oligarchies
can perform well because they cover the preference space.
She explores
the benefits of familiarity.
There are many hypothesis as to how familiarity may matter –
ease of prediction, group recognition, etc.. She examines population
structure The population structure hypothesis that
familiarity Model 1000
agents. . Agents have a set memory. They are biased towards people they
have interacted with frequently.
She adds a decay to familiarity. She considers
both random interaction and a spatial interaction model. She also considers a case where agents
gain fitness when they interact with familiar people and then adapts
familiarity bias, memory capacity, the threshold increases, and decay
decreases. Clustering is very
high in spatial but not in the random. Last, she asks
whether familiarity increases fitness. Initial Results She varies the
familiarity threshold and the familiarity decay. She finds in the non spatial setting
that familiarity does not create clusters in the random or spatial setting. She finds some clustering in the
spatial model. In the model in
which agents adapt, she finds that they adapt behaviors that allow for
familiarity to matter – large bias, longer memory, and not much of a
decay effect. She did not find any co-evoution of cooperation and bias. Post Disaster Migration Data: Data
on 18,000 people moving following Katrina. In most cases, most people 90% of
people return with two years. In
five Louisiana counties, approximately 50% return. In these 5 counties most effective
counties, 18,922 moved to 672 counties. Findings: People
with low credit scores went to Houston.
People with low credit scores and high credit scores went to different
locations within Houston.
Homeowners moved to a single neighborhood in Houston. People moved to
places with similar racial compositions as their original locations. Using gravity
model, modified gravity model, income or vacancy fails to explain much of
migration.
She studies
social norms and how they emerge.
Norms are patterns of behaviors and self-reinforced expectations. Shaking hands is a norm. Cooperating is a norm.
Model Agents
participate in a bargaining game.
Each can demand a share of $100. If the sum of demands is less
than $100, the money is divided.
An equity norm would be 50-50. In her simplified model,
players choose high ($70), medium ($50), or low (%30). Agents have a
memory of length L. An agent
chooses the action that gave the highest payoff in the last L periods. Results Individual Learning L =1,2 she
finds that M,H,L all equally likely L- 3-5, she
finds that there is either an equity norm or a split between H and L L= 6-10 much more
equity norm (M) Social Learning When groups are
less than size five, she finds less equity. When groups get larger, the equity
norm becomes more likely. Extensions She anticipates
experimenting with larger groups and making some groups more influential.
Her project
builds on the literature exploring foraging and step length in environments
with resource heterogeneity.
Collective foraging provides advantages over individual because of
less predation ,better information, and better decision making. Her model
explores how landscape effect movement which in turn affects
interactions. Also, landscape can
affect interactions Model She has three
level of clustering: Low, Medium, or High Three types of
movement: Ballistic, Levy, or Brownian. Initial model: Two types of
agents: signaler and non signaler. A signaler can
take targeted (prob p) or random steps (1-p). Later model All agents are
signalers with some probability. Results First, only
Brownian movements produce significant clustering. One of the main
takeaways is that when the food is more clustered, she gets more clustering. She finds that
when there is no competiton or cost there is no
clustering. However, as movement
costs, signal costs and
competition are present, then she gets more clustering She evolves
interaction level and movement.
She finds that highly clustered environment leads to higher
interactions. She finds that
optimal movement strategy converges to around two (Levy process) in many
cases. Niclas is analyzing whether we can see scaling
from a distribution perspective rather than a sum perspective. What is Scaling? Scaling is a
way to measure a size
amplification. If ten workers make ten chairs, twenty workers make
forty chairs, and thirty workers make ninety chairs, then productivity is
increasing nonlinearity. In this
example, if C is the number of chairs, and N is the number of people, then the
production function can be written C = (N/10)2. The two is the exponent. An exponent larger than one
implies super linear scaling. Here are some
examples of estimated exponents: Patents 1.27 Aids 1.23 Wages 1.12 Serious Crime
1.16 GDP 1.26 Scaling From a Distribution
Perspective Using
distributional data, he takes percentiles within cities and the fits a
scaling for each percentile. For
example, using income data, take
the person at the 68% level for
each city and look at the distribution of across cities. He looks at
firm turnover and income and finds that if you look at the exponents, they
are small, that is close to one, for all but the top 95% and up. He, therefore, finds that
scaling is largely (though not entirely) driven by the tails. He then
constructs a model and then asks if he can generate his randomly. Suppose that each city is drawn from
the same distribution. Larger
cities consist of more draws. He then asks if this would show
scaling. He finds that if you
look at the median, there is no scaling.
That is to be expected. Results He finds if the
power law has a low exponent and look at the sum, then you get a power
law. This occurs because if the
exponent is low, then the distribution does not have a mean. So the expected sum increases without
bound. He concludes
that there may be no inherent differences between the data generating
processes in larger cites.
Open source
programming is a collaboration, innovation network. Platforms such as Stack Overflow
and GitHub and GitLab.
These platforms lower the barriers for participation but building a
team is difficult. In this project
she investigates how teams assemble and who leaves. She examines the drop out rate for men
and women. This project is part
of a larger interest the value of diversity for collective outcomes. She considers cognitive diversity
(tenure) and identity diversity. She measures these using the Gini
Coefficient for tenure and the Blau Index for
identity. Data: She
uses a longitudinal data set of 23,493 GitHub projects and several hundred
thousand users? Most teams
consist of fewer than 11 members.
The data show 4.5% women, 78.2% men and 17.3% where gender could not
be determined. Findings: 75%
have no identity diversity. All
but 1% of those are all male.
The percentage of women who left was very high for teams with 3-5
people and only one woman. The
turnover and leaving the core Huge difference between teams of size 3-5 than
teams of size 6-9. In the
large teams, even though women do not leave, they are much more likely to
leave the core. Men are
very likely to leave the core when there is a lot going on. She finds much
higher proportions of women if there is less gender diversity. She is analyzing
choice as an emergent process.
She asks whether bees foraging can teach us about human decision
making.
Attraction effect: if you add D, that is near but worse
than A, people choose A. Similarity Effect: If
you add S, near but not dominated by A, people choose B Model She creates a
model where 200 honeybees search in two dimensional space for hives. She constructs a model in which
bees start in the explore state.
They then go to either A,B, or C based on the quality of the site plus
an error term. In her initial
model, once at a hive, a bee does not leave. She constructs
three search attributes of search: Recruitment:
Bees move with some probability based on the quality of the site and
the number bees. Evaporation.
The probability of hearing signal depends on distance. Switching where
bees can go back to exploring.
This produces eight
possible models. Results First, she
performs a feature sweep. She
considers all eight possible cases by turning on or off all three features. With just evaporation, she finds
random but stable not spatially contingent patterns. With switching, she finds unstable
patterns. With both, she gets
spatially contingent and unstable patterns. When she adds recruitment then she
finds stable spatially contingent patterns. She also finds a small attraction
affect. The Effect of Banning Discussion
Threads on Community Formation in Reddit Pamela Thomas, Computer Science,
University of Notre Dame. The research
explores whether a decision by Reddit to shut down a subreddit, a community dedicated topic succeeds or whether the
participants reorganize and if so, in an existing or new subreddit. Background: In 2015, if a
subreddit is not a safe platform or if people fear for their safety, then the
subreddit can be quarantined or banned. In 2015, the
subreddit fatpeoplehate was
shut down. Analysis. She first looks at whether people
posted to different subreddits before and after the ban. ŇFor example, fans of fatpeople hateÓ, had fewer posts after the ban than
prior. Attention Graph Model: Each person divides an attention across
sites. She first weights people
equally but later weights people by number of posts on fatpeoplehate. Analysis:
She investigates where people are likely to post. When she switches to a weighted graph,
she finds that there were two subreddits that were top 30 most popular for
weighted or unweighted. She
then compares the differences between the weighted rankings and the
unweighted rankings. She also
analyzes where people move from. Conclusions: The reformation of community ties
appears to be approximately linear.
Events, such as the end of Game of Thrones, can trigger reconnections
but overall, it appears that the ban worked. Outrage tends to dissipate. Model: She constructs a method for measuring
the extent of community re-formation. Who Joins the Alt-Right on Twitter The Alt-Right
community congregates on sites.
Who joins? Who is likely to?
Are people recruited? Research design: Look at networks over time. Investigate whether and how people
move in and out of the core, middle, and periphery. To do this, he creates a retweet
network and then does a K-core decomposition. Higher K means more connected Data: 672
days of people who belong to a community. Analysis:
he keeps track of how people move between quantiles by core
periphery-rank. So if
someone is in the top 20% in the first week, do they remain in the top 20%. He finds that
almost 80% of the people in the core stay in the core. |