Dr. Nicholas A. Christakis has an interesting piece on Edge.org right now. He’s also done some cool research on a variety of subjects with social networks as the focus.
Here I present a critique of his dialog on Edge.org. I eagerly await the actual publication of his Facebook.com-based research papers http://christakis.med.harvard.edu/pages/pubs/pub-sn_ihe.html. In the meantime I’ve researched his publicly available papers (such as this one: http://christakis.med.harvard.edu/pdfs/077.pdf) and read his Edge.org piece several times. He’s consistent in his approach and vocabulary across his publications.
My goal in commenting on the dialog isn’t to add more noise or to be an anti-academic ranter (nor is it altruistic!). Using social networks as a data source for understanding behavior is very useful for improving media, business and our lives. Unfortunately in our collective business, academic, social and political rush to make use of all this data, our vocabularies and approaches are all over the place and the conclusions drawn from all the research do not yet provide much practical value. The interest in this research has exploded and now is the time to coagulate it all.
I’m excited that Dr. Christakis is out of the gate and hopefully I can help us benefit from his work, refine it, and build from it. I personally have several reasons to value is work – UChicago connection, health care (my family was one of the main subject of SiCKO), and social network studies (that’s what I do!).
That said, here we go.
“For me, social networks are like the eye. They are incredibly complex and beautiful, and looking at them begs the question of why they exist, and why they come to pass. Do we need a kind of just-so story to explain them? Do they just happen to be there, for no particular reason? Or do they serve some purpose – some ontological and also pragmatic purpose? “
Complexity is only a weak connection between the development of the eye and the development of social networks. That is, stating two things are similar because they are complex provides no value in understanding either. It’s an anchor for saying “hey, here’s something that seems like it should have a purpose because it’s seemingly so well suited to what it does and the other thing over here has that same feeling to it.” Fine, I get it. Unfortunately, I think using it as the headline to an article and lead in paragraph gets the reader linking the two subjects. The evolution of the eye is such an abused ID vs evolution metaphor it’s best to not recall it.
Though the passive appeal to purpose is not helpful, at least Dr. Christakis does ask whether they serve a purpose or they “just happen to be there.” He clearly understands that purpose itself is not likely to lead to understanding the social networks anymore than its helping in uncovering the workings and “origin” of the eye.
The eye and social networks are emergent properties of selection by consequences. We can eliminate the purpose discussion right now. When a scientist goes down the road of uncovering “purpose” it leads only to further linguistic logic, not to actual descriptions of relationships between variables.
“The amazing thing about social networks, unlike other networks that are almost as interesting – networks of neurons or genes or stars or computers or all kinds of other things one can imagine – is that the nodes of a social network – the entities, the components – are themselves sentient, acting individuals who can respond to the network and actually form it themselves. “
The sentient and acting qualities of the entities of social networks is hardly unique or amazing. No doubt human behavior and social interaction is complicated, but there’s no mysterious free will or free creator aspect to any of it. Other network entities like neurons, genes, stars, computers and particles all operate more or less under selection by consequences (the differences is in what “unit” is selected and what stimulus the unit can attend to.)
Social networks seem amazing to us, I suppose, because of their complexity and our inability to talk about about human behavior/social behavior without appealing to the “thinking” (sentient, acting) man. Social networks become more measurable and understandable when we stop trying to measure mentalistic concepts and stop analyzing the behavior in terms of some unique quality of mankind.
“I began to see in a very real way that the illness of the person dying was affecting the health status of other individuals in the family. And I began to see this as a kind of non-biological transmission of disease – as if illness or death or health care use in one person could cause illness or death or health care use in other people connected to him. It wasn’t an epidemic transmission of a germ; something else was happening. This is a very basic observation about what I now call “interpersonal health effects, but as I began to have more and more clinical experience with such patients, I began to broaden the focus. I became interested not just in dyadic transmission of illness and illness burden, but also hyper-dyadic transmission.”
Interpersonal health effects? Hyper-dyadic transmission? These are big phrases more simply stated as behavior between two or more people. Why add a higher level language construct when simply cataloging, measuring and describing the behavior and stimulus does the trick?
Dr. Christakis can do without the big phrases as he does in his following paragraph:
“For example, one day I met with a pretty typical scenario: a woman who was dying and her daughter who was caring for her. The mother had been sick for quite a while and she had dementia. The daughter was exhausted from years of caring for her, and in the course of caring, she became so exhausted that her husband also became sick from his wife’s preoccupation with her mother. One day I got a call from the husband’s best friend, with his permission, to ask me about him. So here we have the following cascade: parent to daughter, daughter to husband, and husband to friend. That is four people – a cascade of effects through the network. And I became sort of obsessed with the notion that these little dyads of people could agglomerate to form larger structures. “
Great! Here we can actually dig into the behavior of the people, the consequences, and the web of feedback stimilus.
Interestingly you find that nothing is actually transferred nor spread between people. There’s no unit of illness that is transferred. It is misleading to suggest there’s a “nature of contagion within networks” when people do not actually exchange a contagion.
He somewhat agrees with that in saying “What spreads from person to person is a behavior, and it is the behavior that we both might exhibit that then contributes to our changes in body size. So, the spread of behaviors from person to person might cause or underlie the spread of obesity.” But really it’s not behavior itself that spreads. Nothing is spread at all. An entity responds to its environment and the consequences to its own behavior. An entity does not catch behavior or even mirror behavior. If an entity’s behavior is reinforced, it will continue. Social networks have a variety of ways in which participants reinforce or extinguish behavior via consequences (humiliation, praise, points, money, jobs, credibility, reputation, pictures… and so on), they have no power to transmit behavior. What is behavior? What is the unit of behavior?
Saying behavior spreads is like saying time flies. Time isn’t anything. Time is a word we use to say “we’re going to count the frequency of events relative to other events.” Behavior is a similar concept. (I’m going to need to follow up on this or flat out delete it later as it may not be useful to anyone but me.) The take away here is that if you can’t define something and literally see it transmit from one entity to another, the concept of spreading is kind of moot. You can transmit a virus (literally watch the virus go from one host to another). You can’t transmit behavior. Behavior is what an entity does. Other entities and the environment either reinforce the entity to keep emitting the behavior or to extinguish it. If many individuals emit the same behavior in succession it may appear to be “spreading” but really the entities are likely responding to the same consequences in similar ways. What’s the harm in thinking of it spreading? The harm is that one starts looking for the transmission medium (remember “ether” in early physics!?) or other mental constructs to explain casual chains.
Here we see that play out:
“So we can begin to think about combining a broad variety of ideas. Some stretch back to Plato, and thinking about well-ordered societies, the origins of good and evil, how people form collectives, how a state might be organized. In fact, we can begin to revisit ideas engaged by Rousseau and other philosophers on man in a state of nature. How can we transcend anarchy? Anarchy can be conceived of as a kind of social network phenomenon, and society and social order can also be conceived of as a social network phenomenon. “
Dr. Christakis is going back to Plato for insight? Hey, I like Plato as much as the next intellectual but I don’t ever look back to him for present behavioral insight no more than I look to Aristotle to describe gravity to me.
Yes, you can permute philosophical ideas to social network phenomenons. Who cares?
How do we transend anarchy?”
What does that mean?
Well ordered-societies? The origins of good and evil?
Really? We’re not past that yet? When are we actually going to get down to talking about how people behave?
“This is how I began to think about social networks about seven years ago. At the time when I was thinking about this, I moved from the University of Chicago to Harvard, and was introduced to my colleague James Fowler, another social scientist, who was also beginning to think about different kinds of network problems from the perspective of political science. He was interested in problems of collective action – how groups of people are organized, how the action of one individual can influence the actions of other individuals. He was also interested in basic problems like altruism. Why would I be altruistic toward somebody else? What purpose does altruism serve? In fact, I think that altruism is a key predicate to the formation of social networks because it serves to stabilize social ties. If I were constantly violent towards other people, or never reciprocated anything good, the network would disintegrate, all the ties would be cut. Some level of altruism is required for networks to emerge.”
Altruism is another word that provides no explanatory power. Take any definition of altruism you like. and you still end up no where. Lack of violence does not equal altruism. No entity takes one for the team. No entity is selfish. These are personifications and metaphors. Really, leave out selfishness and altruism and purpose and the analysis proceeds more smoothly.
Selection by conquences describes the social interactions accurately without all of the linguistic and mentalistic scaffolding. Dr. Christakis layers on economics, topology, sociology, nuerobiology and as many other ologies and ics as is possible to explain behavior. Let’s get it back to basics!
“Again, the study of social networks is part of this assembly project, part of this effort to understand how you can then have the emergence of order and the emergence of new phenomena that do not inhere in the individuals. We have, for example, consciousness, which cannot be understood by studying neurons. Consciousness is an emergent property of neuronal tissue. And we can imagine similarly certain kinds of emergent properties of social networks that do not inhere in the individuals – properties that arise because of the ties between individuals and because of the complexity of those ties. “
We’re getting closer here. The claim that consciousness is an emergent property of neuronal tissue is out of place though. Consciousness as a concept isn’t needed here and is so ill defined it describes (relates) nothing.
Just as I think we’re finally moving on to the actual relationships between variables in networks, Dr. Christakis describes the spread of obesity.
“To us, it is a very, very fundamental observation that things happening in a social space beyond your vision – events that occur or choices that are made by people you don’t know – can cascade in a conscious or subconscious way through a network and affect you. This is a very profound and fundamental observation about the operation of social life, which we initially examined while looking at obesity. We found that weight gain in a variety of kinds of people you might know affected your weight gain – weight gain in your friends, in your spouse, in your siblings and so forth. Moreover, people beyond those to whom you were directly tied also influenced your weight, people up to three degrees removed from you in the network. And, incidentally, we found that weight loss obeys the same properties and spreads similarly through the network.”
Obesity in an individual is a probability/possibility that depends on the reinforcement of healthy eating and exercise behaviors in media, in our friends, and in our families combined with genetics, epigentics and food/water supply all mashed into a web of contingencies. A social network study may highlight that web, but it’s not any particular property of the network itself.
His other paper has the same faulty logic that the network layout is a causal agent. (N.A. Christakis and J.H. Fowler, “The Spread of Obesity in a Large Social Network Over 32 Years,” New England Journal of Medicine 357(4): 370-379 (July 2007) MS#077)
In Spread of Obesity study he observes:
“Although connected persons might share an exposure to common environmental factors, the experience of simultaneous events, or other common features (e.g., genes) that cause them to gain or lose weight simultaneously, our observations suggest an important role for a process involving the induction and person-to-person spread of obesity.
Our findings that the weight gain of immediate neighbors did not affect the chance of weight gain in egos and that geographic distance did not modify the effect for other types of alters (e.g., friends or siblings) helps rule out common exposure to local environmental factors as an explanation for our observations.”
Ruling out neighbors’ effects does not rule common exposure to local environmental factors. You are more likely to go to work, school, shopping and church (local environment) with your mutual friends rather than your neighbors. How is local environment defined? When discussing obesity one must include the common exercising, eating and stress inducing environments, not simply the local neighborhood or grocery store.
If Dr. Christakis is best anchored in his language of social networks that’s ok as long as we all get an accurate understanding of the relationships between the variables he’s studying. Unfortunately, in this dialog the terminology generates relationships between words, not between people and their behavior and consequences. Less efficiently, he’s simply repackaging (almost as though it were NEW!) well known aspects of behaviorism, evolution and economics.
“We are interested not in biological contagion, but in social contagion. One possible mechanism is that I observe you and you begin to display certain behaviors that I then copy. For example, you might start running and then I might start running. Or you might invite me to go running with you. Or you might start eating certain fatty foods and I might start copying that behavior and eat fatty foods. Or you might take me with you to restaurants where I might eat fatty foods. What spreads from person to person is a behavior, and it is the behavior that we both might exhibit that then contributes to our changes in body size. So, the spread of behaviors from person to person might cause or underlie the spread of obesity.
A completely different mechanism would be for there to be not a spread of behaviors, but a spread of norms. I look at the people around me and they are gaining weight. This changes my idea, consciously or subconsciously, about what is an acceptable body size. People around me who start gaining weight reset my expectations about what it means to be overweight or thin, and this is what spreads from person to person: a norm. It is a kind of meme (but it is not quite a meme) that goes from person to person. “
Copying behavior? Norms? Memes?
Run from these explanations! They add more layers of language. So now to explain behavior I need to understand genetics, memes, norms, contagions! Ugh.
Dr. Christakis’ conclusion:
“In our empirical work so far, we have found substantial evidence for the latter mechanism, the spread of norms, more than the spread of behaviors.”
Okay, so now we’re looking for the spread of norms. (another word for values). Why introduce a world like “norms” to replace “values”? And whether you call it norms or values, it still isn’t anything that is spread. I hate to beat a dead horse, but by using a spreading metaphor as a transmission method we move further from what is actually going on.
We are reinforced by what we value. What we value can be altered by what others value (Super Size Me!) and our environment (If I only have access to junk food, I come to value it).
Really, we do not need all the extra terminology and models. Indentifying values, uncovering environmental variables, measuring behavior rates, and plotting schedules of reinforcement is the data needed. The extra intervening variables (memes, norms, mirror neurons, contagion) do not predict anything and do not improve the explanatory accuracy.
Dr. Christakis points to his work on Facebook data. I, too, think it’s a neat source of social data, but it should not constituite serious data for things like obesity, health, privacy and other complicated subjects. Facebook is flush with noisy and commercialized information. A lot of what people put online is not what they’d do if you met them, what they’d put in medical records, what they have in their photo albums at home, how they answer an anonymous survey and so on. In other words, trying to suss out universal networking theories from a commercialized, college focused social networking site is probably not great.
Specific to my point let’s look at his statements on social ties:
“We have trawled through this large social network and grabbed information about people in the network, and their social ties, as is available on Facebook – for example, information having to do with their tastes, with the people with whom they appear in photographs, and so on. For example, a person might have an average of 100 or 200 friends on Facebook, but they might only appear in photographs with 10 of them. We would argue that appearing in a photograph constitutes a different kind of social tie than a mere nomination of friendship”
Most photos posted to facebook are done in batches – usually within the initial sign up process or the “honeymoon” period when people are still excited about signing up. Photos people post online are representative of who they are around physically most of the time and in picture taking settings. The composition of your friends list is highly biased towards who is active on Facebook, who you might be in school or work with and so on. It really isn’t great at suggesting the particulars of social ties in any real world kind of way.
So much of what you do on Facebook is heavily influenced by user interface and the software system. Certain things are easier than others or more obvious AND the interface has changed constantly (how you post photos, how you set privacy, what’s set by default). Also impacting use of Facebook is the savviness of the user. There are far more people not using Facebook in this world than those that are. Facebook skews far younger and savvier than the population at large, so the stuidies that come out are biased to that group (and that is important when discussing behavior!) http://www.quantcast.com/facebook.com/demographics
Again, it is a great source of data and certainly has value, but you need to have many secondary sources to back up conclusions based on online network data. Also, what you don’t get access to is all the behavioral data – emails, alerts, pokes, system alerts, click throughs, ad response, eye tracking, referrers…
There are two other authors connected to Dr. Christakis’ dialog. There isn’t much meat to their statements. Rushkoff’s metaphor of a “media virus” is pretty shallow and is another case of language complication. One day I may go in for a bigger response. Rushkoff is a neat dude producing lots of cool stuff so this statement isn’t indicative of his quality of thinking.
I have another pretty detailed presentation from my pal, Dan Goldstein (www.decisionsciencenews.com) from the London School of Business that is tangential to all of this (I will post with permission soon). A lot of the vocabulary is the same and the data is impressive. It’s missing a key part too! WHAT IS REALLY GOING ON AT THE ENTITY (PEOPLE) LEVEL? Dr. Goldstein and Dr. Christakis agree that the topology of the network is hugely important to understanding how fast, when, what, who spreads ideas, data, values, norms (whatever you want to call it!). However, there’s no WHY inherent in the network makeup. What is reinforcing to social network participants? How does reinforcement work? What behaviors can we reliably measure on the social networks? What data should we ignore? How is behavior reinforced on the network? How we tie online and offline behavior together?
That’s my task. Filling that in is my contribution.