A study at Leicester University and Exeter University has found that the traditional view that decision makers only act in their own interest is incorrect – in some instances, decision makers act in the best interest of their team, often at their own expense. Classical game theory predicts that people act out of their individual self-interest while team reasoning theory suggests that people look after the interest of their team, not always their own selfish interests. Their findings show that team reasoning predicts decision making more powerfully than orthodox game theory in some games.

Specifically, their findings from two experiments: lifelike vignettes (Experiment 1) and abstract games (Experiment 2) with certain structural properties provide evidence in support of collective preferences and team reasoning. Most players preferred team-reasoning strategies over the unique Nash equilibriua strategies supported by traditional game theory.

In both the games studies, the pair of players who followed team-reasoning strategies received higher payoffs than those who chose Nash equilibria. In all cases, a player motivated by individualistic payoff maximization could have obtained a higher payoff by choosing differently.

In traditional game theory, a strategy is dominant if it earns a player a larger payoff than any other, regardless of what other players do.

The experimental design can be found here.

John Lamberth, in his paper Driving while black: A Statistician proves that prejudice still rules the road (1998) discussed the idea that police officers pulled black drivers more often than white drivers.  In his research based on data from Maryland State police searches he discovered that, “although blacks were searched with greater frequency than whites, drugs were found on approximately twenty-eight percent of blacks stopped, and on approximately twenty-eight percent of whites stopped” (Ramirez 2003).  Some officers may act on what is known as “circumstantial correlation,” the belief that “the likelihood of criminal activity increases when people of color are in certain circumstances” (Ramirez 2003). 

This is similar to the case we discussed in class where “self-fufilling expectations equilibria” exist.  According to Lamberth, there are some officers who search black drivers more frequently than white drivers because they have asymmetric information that black people are more likely to be involved in crime, at least under certain circumstances.  Because they are inclined to believe this, they continue to search drivers of color until their expectations are fulfilled, even though there is little evidence to support their belief.  While comparing people to cars may seem like a stretch, this situation is similar to that in which everyone believes that a car for sale is low quality, so no one will bid a high price for it, and because no one will bid a high price then no good cars go on the market and only low quality cars (lemons) are available for sale.  If some police officers believe that blacks are more likely to be involved in criminal activity, then they will be suspicious even of their interactions with innocent blacks, and so they will likely emphasize the interactions they have had with blacks who actually were found with contraband, which as was demonstrated earlier, is the same percentage as whites who were found with contraband.

 http://www.counterpunch.org/drivingblack.html

For more information see:
Defining racial profiling in a post-September 11 world.
Ramirez, Deborah A.; Hoopes, Jennifer; Quinlan, Tara Lai
Pg. 1195(39) Vol. 40 No. 3 ISSN: 0164-0364
June 22, 2003
 

8 weeks ago, the financial markets faced a great problem when news spread that Bear Stearns was running out of cash. On Thursday, March 13th, lenders refused to let Bear Stearns borrow any more cash to cover their assests. News spread like wildfire. Overnight, complete confidence in Bear Stearns ability to back investors evaporated. By Friday morning, investors were selling everything they had in Bear Stearns. In one day, the value of Bear Stearns dropped 47%.

The news caused a giant information cascade which sent the $8 billion dollar company into a deadly free-fall. Investors panicked, and every stock market was feeling the pain. Only two days after near record gains, the stock market was back to where it was, on the verge of a big recession. The information cascade was so severe that the Federal Reserve feared the worst, a recession the size of 1987. To stop the building crisis, the Federal Reserve lent J.P. Morgan the required cash to buy Bear Stearns and back all of its investments. The result, a much calmer market, and an eventual upswing in investor confidence.

The information cascade was the fear that such a large bank could go bankrupt overnight. Investors panicked and ignored their signals. Bear Stearns was not in as much trouble as investors believed, but they still ran to liquidate their assests in Bear Stearns. When this happened, the fear became a reality when Bear Stearns faced one of the biggist liquidity problems in decades. Only when the Federal Reserve stepped in did the problem calm down and investors began reading the signals. Since the buyout of Bear Stearns by J.P. Morgan, both the Dow Jones and NASDAQ are up. From March 17th to April 1st, the Dow Jones has seen a 2.7% increase, with the NASDAQ seeing a 9.3%. Once the cascade stopped, investors regained their confidence, used their own signals to make decisions, and everyone started trading again.

Sources:
http://www.washingtonpost.com/wp-dyn/content/article/2008/03/14/AR2008031401617.html
http://finance.aol.com/quotes/the-bear-stearns-companies-inc/bsc/nys
http://www.washingtonpost.com/wp-dyn/content/article/2008/03/11/AR2008031100893.html

http://www.nytimes.com/2008/03/02/business/02view.html

“How a Bubble Stayed Under the Radar”

–New York Times 

If the price of a good drastically strays from its inherent value, while buyers are buying and sellers are selling at considerably high levels, the market is most likely experiencing an economic bubble. This behavior is no different from the recent housing bubble, in which the market’s buyers ignored their personal intuition to buy houses at prices too high. So, why do otherwise rational people make these irrational decisions? “Were all these people stupid? It can’t be.”

This seemingly irrational behavior can best be attributed to an information cascade. When someone is entirely surrounded by people buying real estate, the temptation to comply with herd behavior sways that person to disregard their better judgment to purchase a house too. The herd behavior of the housing market’s buyers and sellers fueled this economic bubble, resulting in damaging effects.  This housing bubble led to extremely detrimental effects, with the personal consequence of a poor investment and the large-scale consequence of a misallocation of resources. According to Robert J. Shiller of the New York Times, the most harmful force driving this housing bubble was the weakness of the information that prompted the cascade. “The information obtained by any individual – even one as well-placed as the chairman of the Federal Reserve – is bound to be incomplete.” The information prompting a cascade, undoubtedly incomplete, begins the informational cascade; with the decision are made individually and sequentially, if one person makes the wrong decision, all of the following people in the information cascade will follow. And so, the recent housing bubble burst out of control, with each person blindly basing his or her decision on the actions of others.

This is part of my short paper.

After Korea got its independence from Japan in 1945, the country was divided and controlled by U.S. and Soviet Union. Starting from that time, two counties were walking totally different paths until now. There was a Korean War, or WW3 in 1950 for three years and each country’s scars became deeper. It had been more than half century, but they still could not have reconciliation. I would like to first examine the relation between South and North Korea in historical view and analysis in Game theory that we have discussed in class. The historical relation can be organized from zero-sum play to prisoner’s dilemma, and finally deer hunting game in future.

Ofcom (Office of Communications) published their findings on the emergence of social networking sites in Britain. I was amazed to find how quickly social networking sites are catching on, especially among the 8-17 year old audience. Almost half of the children who have access to the Internet have their own social networking profile, while only about 22% of adult users age 16 and over have their own profile. In addition,it is not uncommon for adults to have profiles on more than one site (average is 1.6).

The report also categorizes social networkers into 5 different groups:

• Alpha Socialisers – mostly male, under 25s, who use sites in intense short bursts to flirt, meet new people and be entertained.
• Attention Seekers – mostly female, who crave attention and comments from others, often by posting photos and customising their profiles.
• Followers – males and females of all ages who join sites to keep up with what their peers are doing.
• Faithfuls – older males and females generally aged over 20, who typically use social networking sites to rekindle old friendships, often from school or university.
• Functionals – mostly older males who tend to be single-minded in using sites for a particular purpose.

I find that these definitions mostly apply to social networking sites like MySpace, where people are more inclined to be-friend total strangers. For the most part, users of Facebook (mostly college students and the generations of recent graduates) tend to use the site to stay in touch with friends and colleagues they know in real life. Also, the format of Facebook doesn’t really allow for much customization as on MySpace, thus making it harder for users to lure others by their profile’s visual appeal.

Another interesting finding from their research was the following:

“Some teenagers and adults in their early twenties reported feeling ‘addicted’ to social networking sites and were aware that their use was squeezing their study time.”

I feel that this is a growing trend, especially now, with the introduction of Facebook Chat. Once you’re logged into Facebook, you can see all the other users who are currently online. Just over the past two days, I’ve seen online counts above 60. Either the students at Cornell have very little to study (doubtful) or we’re all just getting a little too comfortable on Facebook.

If you wanted to synchronize your watch with all your friends and all of you did not have access to a universal time source, how would you proceed?  The first step would probably be figuring out a method in which to communicate with all of your friends in order to relay the exact time displayed on your local watch.  But, as you would quickly discover, communicating simultaneously with all of your friends would be very difficult.  Additionally, each person would take a variable amount of time before they were able to adjust their watch appropriately.  In the end, the times between you and your friends’ watches would remain out of sync!
This dilemma of synchronizing time is actually relevant to the synchronization of nodes in a distributed and connected network. Building a model of the situation with nodes, edges and times in a graph of the network helps to prove and illustrate some interesting facts—something that we have been doing in Networks with other network models.  According to a paper authored by two MIT professors it is impossible to synchronize processes in a network completely, unless strong assumptions about the network are made.  But, it is possible to prove that synchronization can only be achieved to within a certain value and no better, no matter what algorithm is used.
The Lundelius-Lynch model assumes a distributed and completely connected system of processes that can only communicate by sending messages to one another.  As we learned in class, a completely connected graph is one where every node  is connected to the others via a direct path.  Within each process there is a physical clock, which is not controlled by the process but can be read by the process.  It is important to note that the physical clock cannot be set and continually advances. Additionally, the entire network is defined to contain n processes, which can be represented as vertices in our complete communication graph.  Also, each process is allowed to know the positions of other nodes in the network and the size of the network itself.  Further, a process can take an indeterminate amount of time to read its incoming messages.  Thus, some processes read their messages immediately while others wait until they are ready to process the message, adding an element of uncertainty.  We can represent the edges between nodes as having a message time plus an for uncertainty.  After analysis in the paper, it is shown that time cannot be synced completely across all nodes!

We could relate this process to the spread and mutation of an epidemic (something we have touched on in class), and find bounds for how much the disease could transform.  How good does a vaccine have to be so that it could accommodate at least the minimum ‘sync’ of the disease?  The network could be represented as a proximity network like we did in class, where sending a “message” or disease with an uncertain mutation would not be a decision, just like sending a message when syncing is required.  Along the same lines, what happens if the system of nodes we consider is faulty?  For example, what if there is a probability, just like in the spread of disease, that a node does not receive the message (perhaps disease).  The introduction of a probability and faulty systems are potential fields of explorations, and our knowledge of networks provides a strong foundation to embarking on solving these problems.

Referenced Paper:

ti.tuwien.ac.at/ecs/people/schmid/Mypapers/MS06.pdf

BBC article: http://news.bbc.co.uk/1/hi/technology/7370956.stm

The Internet faces a couple of significant hurdles as technology and its popularity sprint towards the future: the first is the amount of traffic across it, which has exploded in recent years; the second is the omnipresent debate about network neutrality. The two issues are intricately related, but both demand a solution, and soon.

Depending on to whom you talk (cue Orwell’s rolling in his grave), the Internet may or may not be about to buckle under its own swollen girth: companies like AT&T and Sprint, who provide most of the Internet’s major infrastructure, claim that their resources will be overwhelmed by a wave of streaming video and file-sharing, pushing Internet usage to hitherto unimaginable heights. This collapse could come within as few as three years. Cisco insists that traffic will only increase at a rate of 50% to 60%– manageable under the current rate of infrastructure improvement.

The other facet of the debate concerns network neutrality, the principle that ISP’s not discriminate against traffic based on content. Providers argue that the cost of infrastructure now outweighs their user fees, and that they should be able to charge users for access to certain services. Streaming video is a prime example—it consumes a tremendous amount of bandwidth, slowing down performance for a large portion of the network. In forming an opinion on the subject, the average user must weigh their unwillingness to pay for a service that really ought to be free, and maintaining an acceptable level of performance

                In terms of network traffic, the effects are most noticeable locally (though this is a result of the ISP’s distribution system more than anything else). If the information flood becomes truly excessive, however, it’s possible that the entire network might become overloaded.

 The article below discusses the general importance of network effects for many businesses and how it can be dangerous to rely on such a phenomenon because it can be very difficult to come by. What I found interesting was a small comment saying how Google’s success did not rely on network effects, but rather an amazing technology service. It then mentioned that Google does use network effects in a small way because as more and more people use the service, Google can keep track of more search data and history to enhance everyone’s experience when the use Google and improve the probability that they find what they are looking for. Many companies have these small network effects that enhance their user’s experience not directly because there are more user, but because the increased number of users has allowed the company to get more data and deliver a more accurate and satisfying product. For example, as more people started using mail delivery companies, the companies were able to make shipments faster because there were more people so this naturally led to more equipment and technology which ultimately leads to a better service. Another example would be consulting companies because as the company gets more and more customers, they become more experienced and are able to give better solutions to their clients. This is why the best consulting companies will always be around, since they will always get customers and continue to improve their services. One last example is about taking courses. Imagine that you are in a certain course. What you learn is really not dependent on how many people there are in the class. But as more and more people take the course, the course tends to improve throughout the years and becomes of better educational value to the students.

http://technology.timesonline.co.uk/tol/news/tech_and_web/article3809879.ece

Background Information: The degree of separation in a network is equal to the average length of the shortest path between pairs of nodes. Extracting the degrees of separation for a large network is computationally demanding. The computation involves averaging the degrees of separation of each individual node in the network. Thus, the time required to compute the degrees of separation depends number of nodes. In fact, the computational demands grow quadratically as the number of nodes increase. In order to circumvent the demanding calculation, it is possible to estimate the degree of separation by first selecting a small sample of nodes, and then averaging the degrees of separation for this limited set. Therefore, by sacrificing some accuracy, we a able to save a lot of computation time.

Experiment: I investigated exactly how much accuracy is lost by the estimation method described above.

Method: I created a virtual network of 1000 people using the simulation described in my first post. I ran the simulation until the network’s giant component fully developed (to see how this was done, see my second post). Then I calculated the degree of separation for various sample sizes. I also calculated the exact degree of separation by using the entire population as the sample. The result is summarized on the plot below:

The actual value of the degrees of separation, as shown by the blue line,was 9.66. The blue dots are estimates made by various sample sizes. As expected, as I increased the sample size, the sample points converged to the real value. The space between the green lines represent an area of 95% accuracy. As seen on the graph, even when we limited our sample to as little as 0.02% of the population (20 people), we obtained a value that is 95% accurate. The sample size required for accuracy levels of 99% and 99.5% are also shown on the graph.

Conclusion: Using samples as small as 0.02% of the population, I was able to estimate the degrees of freedom with an accuracy of 95% . Therefore, I support the practice of taking a small sample of nodes into account when computing the degrees of freedom in a large network.

My simulation source code can be found here (Visual C++ 2005).