21_human_network.md

Human Network

• How many potential networks can you form with n of 30?

  • total number of potential edges = n choose 2 or 435
  • total = "we note that each possible friendship could either be switched “on” or “off,” and so there are 2 possibilities for each friendship. Thus, the number of possible networks is 2 × 2 × · · · × 2, with 435 entries."

• friendship paradox:

  • sociologist Scott Feld in 1991
  • on average, people have fewer friends than their friends
  • Friendship paradox = selection bias
    • people w/ lots of friends are overrepresented
  • consequences:
    • Popular people disproportionately set perceptions and determine norms of behavior. This can include modal man a woman dates, etc.
    • rich gets richer: people with friends more likely to get more friends.
    • Misperceptions about popularity of some things. As Shane Frederick (2012) states in a study about our tendency to overestimate other people’s willingness to pay for things: “Customers in the queue at Starbucks are more visible than those hidden away in their offices unwilling to spend $4 on coffee.”
    • "We disproportionately fly on the most heavily booked flights, eat at the most popular restaurants, drive on the busiest roads and at the busiest times, go to parks and attractions at the most crowded times, and attend the most crowded concerts and movies. These experiences bias our perceptions as well as our perceived social norms"

• small world

  • stanley milgram: sent out letters to people in kansas and nebraska and asked them to get a folder to people in MA. they were told names + towns of recipients. directions = “If you do not know the target person on a personal basis, do not try to contact him directly. Instead mail this folder…to a personal acquaintance who is more likely than you to know the target person…it must be someone you know on a first-name basis.”
  • Out of the 160 folders that started in Nebraska, 44 made it to the final target—27.5 percent. The median number of steps was five and the range was from two to ten, with an average just above five.

• descriptive stuff on human networks

  • in 2013: "According to a recent Pew Research Center survey, adult users on Facebook in the U.S. have an average of 338 friends and more than half of all adult users have over 200 friends."
    • 2011: "Indeed, 99.9 percent of Facebook’s more than 700 million active users are in a single giant component."
      • this has consequences for potential diffusion of info. + disease
  • 2012: "despite the sparsity of Facebook’s network, the paths between users are extremely short. Perhaps astoundingly, the average distance between any two active users is only 4.7 links."

• when networked:

  • externality: for infectious diseases, vaccines subsidized + encouraged by both gov. and corp. as infecting others = externality
  • The most popular can be among the first to hear new news but also the first exposed to new infections.
  • small n can lead to large consequences: "it has been estimated that as few as 3 percent of people infected with Ebola may have spread more than half of the cases in an outbreak in Sierra Leone."
  • overconfidence: "Your friends’ opinions are partly based on your past opinions—so part of the “new” information you get from repeatedly talking to your friends are echoes of your beliefs. As your friends begin to confirm your beliefs, you can end up being overconfident. It is quite natural: you feel better about an opinion if other people agree with it. Even if you know a lot about the structure of the network, filtering out your echoes is hard."
    • correlation neglect. demarzo, vayanos and zweibel 2003, glaeser and sunstein 2009

• returns on friendship with plausible identification strategy: - "Ron Laschever, an economics historian, had a great idea: he examined a military draft, which assigns people into relatively small groups for extended periods of time and strong friendships develop. As Laschever found, a 10 percent increase in the employment rate among a former soldier’s company mates in 1930 corresponded to a 4 percent increase in the chance that the soldier was employed."

  - "David Marmaros and Bruce Sacerdote took advantage of the random assignment of freshmen at Dartmouth College to see how their eventual employment fates, ... . Upon graduating, a student’s employment was correlated with the fraction of their hall mates (students in rooms in the same hall in their dormitory) from freshman year who were employed: changing their hall mates from unemployed to employed led to a 24 percent increase in the chance that a student was employed compared to the average student. Moreover, each extra dollar that those hall mates earned led to a 26-cent increase in the students’ earnings, above that of the typical student"
  

• weak ties - Granovetter found that only a sixth of jobs that came via the network were from strong ties, with the rest coming via medium or weak ties; and with more than a quarter coming via weak ties.

• social proof:

  • "For instance, weather patterns help us measure how many people will see a movie because others saw it. Suppose it happens to be blisteringly hot in Chicago, but pleasantly temperate in New York, on a summer day when a movie opens. More people than usual will go to the movies in Chicago, while fewer than usual will go to the movies in New York. ... Indeed, for each extra hundred people who went to a movie on that first weekend, another fifty go the second weekend, and another thirty in the third weekend."

• social pressure? assigning someone to monitor helps people save. emily breza and arun chandrasekhar

Applications

• Microfinance:

  • Comparing a village whose initial seeds were at the bottom end of the eigenvector centrality to one whose initial seeds were at the top of the scale led to a tripling of the participation in microfinance, on average.

  • Our estimate for the spread of microfinance was that news tended to spread for roughly three iterations—not really traveling much beyond the friends-of-friends-of-friends.

  • Also, some topics inspire people to talk with everyone they know about it, and other topics inspire people less. In the case of microfinance, our estimate of the frequency with which one household would tell one of its friends at each iteration was 1/5.

• Medici:

  • they have more connections than any other family, almost double the number of marriage and business connections of either of their connections
  • out of a fuller data set of ninety-two elite families, more than half of the families married to the Medici were married to at most two other families, while more than half of the families married to the families opposing the Medici were married to more than four other families.
  • For instance, just over half of the shortest paths between other pairs of families in the marriage network pass through the Medici, while the Strozzi lie on roughly one in ten. 1 in 10 for Albizzi. Second highest betweenness centrality for guadagni --- 1/4

• Infectious Diseases

  • Ebola’s basic reproduction number (in the absence of intervention) has been estimated to be just over 1.5 in Guinea and Liberia, but closer to 2.5 in Sierra Leone. This difference stems from differences in population densities, which affect the average number of people that a person has contact with per day, with Sierra Leone’s being more than 60 percent higher than that in Guinea and Liberia.

  • The measles’ reproduction number, in contrast, is much higher than Ebola’s since instead of spreading via blood and saliva, it spreads via airborne particles and has a reproduction number from 12 to 18 depending on local population densities and interaction frequencies.

  • Diseases such as diphtheria, mumps, polio, and rubella, are intermediate, in the 4 to 7 range.

  • 1929: "Herbert Soper, a statistician who studied the fluctuations of many diseases over time. He noted that measles outbreaks in Glasgow had patterns that could be explained by school sessions."

  • spanish flu spread ~ ww1 troops returning

  • Measles would eventually make their way to Hilo, Hawaii, in 1848 via a U.S. Navy frigate, the Independence, coming from Mexico.32 Whooping cough and flus would make their way to Hawaii that same winter, starting a series of epidemics together with the measles that would eventually conquer roughly a quarter of the native population.

  • most popular first to get infected. christakis and fowler: "groups: one group of a few hundred students who were picked at random from the population, and another group of a few hundred students who were named by others as a friend. As we know from the friendship paradox, the students named as friends should have higher degree than those picked randomly. .. found, those named as a friend by others had the flu on average two weeks earlier than the random group of students."

  • "it has been estimated that as few as 3 percent of people infected with Ebola may have spread more than half of the cases in an outbreak in Sierra Leone."

  • travel bans delay but don't prevent an outbreak.

    • H1N1: "As Mexico had some of the first H1N1 flu cases in 2009, many of the travel alerts that were issued that spring mentioned Mexico. That led travel to and from Mexico to drop by around 40 percent in the late spring of 2009. Travel changes delayed the spread by a few days."
    • Ebola with lower R_0, it has been effective

  • quarantine can backfire. Polio is transmitted from human feces to other humans orally: so, having open sewers near children is a deadly mix. The quarantining led sanitation to deteriorate and more children to become exposed; and children who developed fevers for other reasons were shut in with others who had polio, with deadly consequences.

  • Solutions for flu like pandemics:

    • vaccination
    • quarantining
    • use antiviral to shorten infection + lower chance of transmission

• 2008 Financial Collapse:

  • argument: lehman brothers should have been bailed. "Reserve Primary Fund held more than three quarters of a billion dollars in Lehman Brothers debt."

  • misaligned incentives:

    • "In a review of bankruptcy costs,12 Ben Branch, an economist at the University of Massachusetts-Amherst, estimates that claimants on a firm typically get 56 percent of the (book) value of the bankrupt firm before it became insolvent.13 Typical recovery rates on bonds are in the 40 to 50 percent range, and even if the debt is secured or has priority in bankruptcy that still might only be 70 percent."
      • "These externalities mean that incentives are not aligned: Lehman Brothers not only continued to underestimate the risks of a heavy exposure to subprime"...worried mostly about own profits than damage to others.

  • fewer players + concentrated so fewer opp. to diversify:

    • "In 1980 there were over 14,000 commercial banks recognized by the FDIC (Federal Deposit Insurance Corporation) in the U.S., yet by 2016 that number was just over 5,000"

    • The ten largest banks in the world—the top four of which are in China—had assets of almost $26 trillion in 2016. in 2016 and the world total GDP was just over $75 trillion.

    • In 1990 the five largest banks in the U.S. accounted for roughly 10 percent of the total assets in domestic banking industry but by 2015 they accounted for almost 45 percent.

  • regulation is hard:

    • Even when one regulates one market, investors move their money to other institutions that are not covered by the regulation. For example, after the Great Depression banks were prohibited from providing interest on checking accounts. ... Thus, savings and loans and other financial institutions began to offer checking accounts and easy movement of funds together ... This eventually led to more than one thousand savings and loans (out of just over three thousand in the U.S.) to go bankrupt between the late 1980s

  • bottom line: not clear how much systemic risk is there in fin. systems.

• caste

  • A recent comprehensive study of cross-caste marriages, however, found that only 5 percent of Indians marry outside of their caste
  • such a small village, people are fifteen times more likely to exchange favors within caste groupings than across them,
    • borrow kerosene
  • It has been estimated that between 87 to 90 percent of the money that people gave or lent to others in some villages like these went to someone within the same caste, and that risks were far from being shared well across castes

• Race

  • Roland Fryer found that fewer than one percent of whites in the United States marry blacks even though blacks make up more than 10 percent of the population. Similarly, fewer than 5 percent of blacks marry whites even though whites make 60% of the pop.

• Res. segregation

  • "One study estimated the thresholds for “white flight” by examining how large a percentage of minority populations needs to move in in order for many whites to leave. They found evidence of thresholds of between 5 and 20 percent—so that whites exited even when small numbers of minorities moved in. This suggests that those moving out actually had preferences to be in a large majority, which would be a much more extreme situation than our Schelling example."

  • post 2004 blasts, segregation between the Arab and Spanish population increased by more than 5 percent in the following two years—a large amount considering how infrequently people move.

• women-men salary gap partly explained by lower referral rates for women? - "Lori Beaman, Niall Keleher, and Jeremy Magruder, in a study in Malawi, found that men tend to recommend other men even if they know well-qualified women, while women’s tendencies to refer other women are not enough to overcome the effect."

Networks 101

• degree = how many connections or links (relationships) that a person has in some network

• degree centrality = If someone has 200 friends and someone else has 100 friends, then the first person is twice as central according to degree centrality.

• eigenvalue centrality = person's centrality is proportional to their friend's centrality. extends degree centrality.

• betweenness/brokerage centrality = For each node in the network, betweenness centrality keeps track of the fraction of shortest paths between all other pairs of nodes that pass through that node.

• diffusion centrality = how far will information about X spread if A is the first person to heard the info.

• “path-connected,” or “con.nected” for short: each person can reach every other via paths in the network

• schelling: small biases in pref. can have large consequences for segregation

  • classic example = pref. to live where at least a small fraction of neighbors of same type
  • reality: see above

Misc.

• Inequality and inter-generational mob.:

  • For instance, in Figure 6.1 (intergenerational earning elasticity), Germany comes in at .32 while the U.S. has a rate of .47, almost one and a half times the rate of Germany. Yet, if you ask people whether they agree with the statement “If I work hard, I can improve my life,” then you find that almost 84 percent of Americans agree while only 62 percent of Germans agree.

  • one sees that the correlation in educational attainment between parents and children in Denmark is less than 10 percent, while in the U.S. it is almost 50 percent

  • great gatsby curve: inequality/intergenerational mobility curve. name by judd kramer

  • In the U.S., children with well-to-do parents are more than two and a half times more likely to graduate from college than children from poor families.

  • Someone from the wealthiest one percent in the U.S. is 77 times more likely to go to an Ivy League school than someone from the poorest 20 percent.

  • Figure 6.7 shows that more than 90 percent of children born in 1940 earned more than their parents did—there was a booming middle class and growing prosperity and incomes for almost all Americans were rising from one generation to the next. Of those born in 1980 only 50 percent would earn more than their parents. The numbers are even worse when looking just at males, as nearly all born in 1940 were doing better than their parents at age thirty, but of those born in 1984, only 41 percent were—so 59 percent were doing worse than their fathers.

    • GS: not sure how we know the outcomes for people for 1980

  • "In 1970, almost two out of three people lived in neighborhoods (GS: not sure what that is) in which the median income differed by no more than one fifth from the U.S. median income. By 2009, that fraction had dropped to just two out of five.

• The service sector is also rapidly growing in Asia, and is now the leading employer in China.

• Moving to opp.:

  • Much smaller than I had thought: "The estimated benefit for an eight-year-old child of moving to a low-poverty neighborhood via such a voucher is three hundred thousand dollars in lifetime earning."

• In China the fraction of the labor force working in agriculture dropped from 60 percent in 1990 to 28 percent in 2015, and in India over the same time period it dropped from over 60 percent to around 45 percent.

• Education:

  • striking: "The percentage of twenty-five- to twenty-nine-year-olds who have a bachelor’s degree rose from just over 20 percent in 1974 to just under 30 percent by 2014."

  • In 2016, the average quoted price (inclusive of room and board) for a four-year private college in the U.S. was $44,000, but the average paid was only $26,000. Looking at public four-year universities in the U.S., the average net tuition and fees paid by the highest quartile is $6,330, while by the lowest quartile it is $–2,320.

• Misinformation

  • "Sam Wineburg and some of his colleagues from the Stanford History Education Group42 tested whether students from middle school through college were able to infer the reliability of various information that they saw on the Internet. They were given tasks such as answering whether a posted picture of wilted daisies really offered proof of fallout from a nuclear disaster, discerning which stories on a site were news and which ones were advertisements, and discerning the reliability of articles on sites of professional medical organizations with varying motivations. The report summarizes the findings by saying, “When thousands of students respond to dozens of tasks there are endless variations….However, at each level—middle school, high school, and college—these variations paled in comparison to a stunning and dismaying consistency."

• News:

  • People in news: "In the year 2000, an estimated 56,400 people were employed in newsrooms at U.S. newspapers. By 2015, the number was down to 32,900. For instance, television network news staffs have dropped by half since the 1980s, as have news magazine staffs since 1985.

  • Even the fifty all-news local radio stations that existed in 1980 dropped to thirty by 2010 and reach only a third of the U.S."

  • In the United States, reporters (or citizens) can request information under the Freedom of Information Act. Between 2005 and 2010 the number of requests for information covered by the Freedom of Information Act dropped by almost 50 percent.

• In a study of sushi restaurants in Los Angeles (a town with a good reputation for sushi), a team of researchers (Willette et al. [2017]) gathered 364 fish samples from twenty-six restaurants, in a variety of neighborhoods, price ranges, and ratings levels, over a four-year period. Using analysis of the DNA of the fish samples, they checked whether the fish that they were served matched what it was sold as. Every single restaurant sold them at least one piece of sushi that was not as advertised. Overall, 47 percent of the sushi was mislabeled!

  • For instance, 89 percent of red snapper and 93 percent of what was sold as yellowtail were something else, while only 8 percent of the mackerel was mislabeled.

• partisanship

  • "However, after 1990 partisanship began a sudden and steep increase. For example, by counting how many times various snippets of words appear in one minute of a typical congressional speech from 1870 or 1990, a listener could be about 55 percent sure of the speaker’s political party. So, one minute of terminology would provide a slight clue to a person’s political leaning, but only 5 percent more than flipping a coin. However, by 2008 the terminology used by the different parties had diverged. By counting snippets of words used, and not the actual points made, one could be 82 percent sure of a speaker’s political leanings. This is after only one minute of speech. After four more minutes of tracking words used in a 2008 speech, we can be more than 95 percent sure of a speaker’s political party, while listening to speeches from 1990 or before leads only to a 65 percent certainty."