Networks of Individuals or Individual Networks?

I read the first two chapters of Duncan Watts’ book, Six Degrees. In that book he covered a lot of the concepts that we discussed in class. However, one thing that Watts brought up a couple of times in the chapters was the unreliability of the data. For example, he talked about how normal social situations don’t actually reflect random social connections. Most of the people that we know are not random people from around the world, but rather people near where I live. Because our connections are not random, this makes modeling more difficult.

Another point that I thought Watts made really well was the difference between studying the network and studying the individuals. I was wondering what you guys thought was a better way to assess networks? Personally, I that the best way to assess networks is a combination of both methods. This was best shown through the story that Watts told about the power surges in the British electrical network due to people putting their kettle on during soccer halftimes. The behavior of the people is individual, but they are all part of a network.

Also, on a (slightly related) note: In COS 126, we were discussing this same concept and the professor showed us this website: It tells you the degrees of separation between Kevin Bacon and any other actor. Enjoy! — Cara

8 thoughts on “Networks of Individuals or Individual Networks?

  1. You raise an interesting point from the reading, Cara. The relationship between individual actions and how these actions appear as an aggregate has important implications for everything from economics (where the modeled actions of individual people and businesses together can sway markets) to politics (like how political structures operate and blocs of people vote).

    Like those subjects and like the tea kettle example you mention, Cara, I agree that the study of networks for the purpose of (1) understanding them and (2) developing policies that reflect them must incorporate study on both the individual level and the more macroscopic level. Unfortunately, this is much easier said than done. To incorporate part of what we covered in this course in the last couple of weeks, one way that could examine networks in light of both individual and collective actions would be to track a disease (or something like a disease) that spreads through a population. One example might be something like the World of Warcraft digital error that spread throughout the network; as a digital “plague” its spread related not only to the interaction between different individuals (the players in the game) but also the spread of information (as players signed on/off in response to the error’s progress). In a world without ethics, review boards, or privacy/security concerns, I would imagine that some simple computer virus that pings a location, name, and ISP to the virus’s creator would show how the program spread through networks much as a microbe would. Since diseases hop between clusters through certain links, so too could this sort of computer program. This is not to say I would support or encourage such a scenario; it just offers an intriguing what-if. A more benign and non-invasive approach might be to track the spread of a news story over a few days. For instance, the New York Times could break a story, certain people on Facebook post the link or comment on it, Friends of those people reply in some way, the story catches the attention of websites that hook into other networks, the story spreads more, CNN / Fox News / MSNBC pick up the story, and so on. If tracked in real time, the spread of this particular story could help map how a piece of information jumps between clusters of people.

  2. Well the problem with assessing networks comes from the collective action problem. For anyone whose every had to run a student organization, you know that it is extremely difficult to get a large amount of people to do a single activity at the same time. Unless some large incentive is given, people are more likely to just ignore wide-scale efforts. Thus, studying a network is, practically, a challenge. It’s also difficult to assess how said network behaves in a natural environment. Passive observation, such as assessing voter turnout after an election, obviously give the most insight, but the least control over the results produced. Active observation, such as releasing a poll into the field, inherently changes the metrics of the network, and thus produces a not entirely faithful result.

    So why do we assess networks? Well, to be honest, I’m a little skeptical of using big data to draw definitive conclusions. There was a NYTimes post about this not too long ago about how network assessment allows you to analyze things that are extremely common, but little else. ( I suppose then I lean more towards assessing the behavior of the individual. It’s harder to do, but the results are more accurate.

  3. I think what Watts showed is an relatively untouched area of study, or rather, approach in study, and that is the idea of dynamic networks. When Watts mentioned about the different levels of academic disciplines and the inability of the scholars to successfully communicate to different those of different disciplines, I realized how segregated our knowledge is. We have always split, categorized, studied individual subjects in isolation – this is rooted in our thinking, going all the way back to the scientific method. The problem of the deductive method or going bottom-up is that we cannot bridge and map all the relationships of components to understand the whole.

    For example, when we introduce a drug into the body, we may explain how certain chemical components bind to other components, but when it comes to understanding how that overall changes the health of an individual, we have no choice but rely on empirical testing and observation. We wave off the “system” (in this case, the human anatomy as a whole) as just too complex or unpredictable to understand. But maybe we we have just never quite approached this complexity in the right manner. This is where I think networks come into play.

    That everything is influencing everything else and constantly morphing itself is an idea that is disregarded across all my studies so far, an idea sacrificed for the sake of simplicity. But we have to return to this idea eventually, and as cmberger illustrates below, we are gaining the right technology to study it.

  4. I agree with you Cara, that the best approach is to integrate network and individual study methods. Just as Watts explains, individual behavioral and personality differences tend to cancel each other to provide a general trend. Yet, studying the individual behavior can be useful to figure out exactly why such trend in the network exists, as he shows with the example of the relations between electric surges and soccer championship in the U.K.. It seems like Watts is explaining the importance of such integration of the methods in his argument for the importance of more communication and integration of different academic fields. He explains the unexpected breakthrough in coming up with a novel theory on network by physicists working on electrons’ behavior but their lack of expertise in working with actual human beings with different personalities and interactional tendencies, and therefore asserts the importance of better cooperation between academic fields that had very little interactions with one another.

  5. Tianyuan raises an interesting point in her argument that knowledge is often too segregated. One of the most fascinating aspects about studying networks is that it is relevant in a variety of disciplines. Social interaction, in particular, is very difficult to model and generalize because of its randomness. Using quantitative network methods to attempt to model social networks is a great example of multi-disciplinary application of knowledge. On the other hand, there is danger in over extending concepts and theories across disciplinary lines. We have seen how some microeconomic principles are difficult to empirically examine due to irrational and unpredictable human behavior. On a much more disturbing note, we have seen Darwin’s concept of the “survival of the fittest” and many of his evolutionary principles used as excuses for countless persecutions and atrocities. To respond to Tianyuan’s specific example, psychoactive drugs are chemically absorbed in the same way for every individual but most empirical evidence shows that many individuals describe the effects of the drug differently. Therefore, trying to find some overarching description of the “system” of the anatomy as if it were a network could send us down a slippery slope, though I have to admit I’m not sure where it would lead.

  6. I agree that the theory behind network types can be practically useful in modelling network effects and is not merely an academic exercise. I recognize that (as @disqus_WxBShY1jVY:disqus reflects) the recent obsession with big data has come with a corresponding backlash challenging its usefulness, and as @ahanna points out, may therefore limit the practicality of network theory. However, knowing the various typology of network systems allows us to better understand how certain networks function, even if the data used to analyze those systems is faulty or misleading. Take, for example, the use of Google Search statistics to model the spread of the flu. Although the Google data overestimated the spread, among other problems, knowing the network paths and effects of pathogens, even without this data, can certainly aid health workers is directing their resources more appropriately. Pre-empting pathogen spread with targeted campaigning could imaginably be more useful and cost-effective than post-infection quarantine.

    One interesting possibility I considered from the models presented in lecture and reading is the concept of using node failure strategically. If you could identify and purposefully fail certain key nodes, you could much more easily isolate the spread of pathogens. This idea that one node could segregate the network (and that this node is not necessarily a large hub, but simply an important connection between two smaller networks of hubs) is a really useful concept in targeting relief efforts. Especially if a contagion became unmanageable (I’m think about the movie here) quarantining a certain state could be more effective and reasonable than shutting down the entire country.

    Of course, the opposite is also helpful; modelling the spread of information on online networks in crisis situations can help the government ensure information is received by the population in a timely fashion. Specifically, they can ensure that certain information nodes are protected in a disaster.

  7. One point that ahanna brought up that I find particularly insightful is the frequent passiveness of collective action, which throws another wrench into the attempt to model networks and the individuals who participate in them. For instance, Princeton students receive, on an intermittent basis, emails from students asking them to fill out a survey or contribute to some data set. Speaking from personal experience, unless (and sometimes even if) the student requesting information is a close friend, the students will often ignore these emails, no matter how quick the survey may be. However, these surveys are asked to be completed with no incentive for the students in question; an alternate view may be seen in student responses to psych studies on campus, which pay students up to $20 an hour for their data. These psych studies are often very popular among the students, no matter who is conducting them. This creates a hitch in modeling networks; if it is necessary to provide incentives for people to provide data, how much incentive is required? Does it skew the data, drawing from only certain demographics depending on where the line is drawn on incentives given to participants? Voluntary data submission is a difficult, though ideally the least intrusive way, to model networks and may perhaps need to be supplemented in order to obtain important information.

  8. I agree…incentivizing participation in studies is a crucial determinant of response rate and by extension the effectiveness of the study. Someone should do a thesis that’s a study of different studies that offer varying incentives for participation and try to model response rates of students.

    Of course, if we take for granted incentivizes are necessary to build effective network models, the experimental cost of network scientists skyrocket. Perhaps in trying to translate this into a workable research model, we can look at methods of incentivization that get the most bang for their buck. Are people more interested in taking part in studies when they offered a smaller amount of money that they can definitely earn? Or are they more interested in participating when there is a larger amount of money that will be raffled off? I would think the second method is more cost-effective, but perhaps less effective in response rate depending on how people view the probability of actually winning the money (maybe a good study-design would find some way of fooling people into thinking they can win).

Leave a Reply