Eat, Drink, and Be Merry: The Spread of Obesity, Nicholas Christakis
Articles Blog

Eat, Drink, and Be Merry: The Spread of Obesity, Nicholas Christakis


– I’m Denise Anthony, I’m
Associate Professor and chair in the Sociology Department and I want to thank
the Rockefeller Center, the Rockefeller Center for Public Policy in cooperation with Sociology Today for bringing in our speaker, Dr. Nicholas Christakis. Dr. Christakis is Professor of Sociology at Harvard University. He is also Professor of Medical Sociology at the Harvard Medical School and he is an Attending Physician in
the Department of Medicine at Mount Auburn Hospital. Dr. Christakis is a doctor
of internal medicine and a sociologist. His unique perspective is both clinician and social scientist is shown
in his research on the social factors that affect health,
health care, and longevity. Evidence of his diverse approach can be found in the scores
of his publiciations that are published in
both top medical journals and also top social science journals. In his book, Death Fortold, Prophecy and Prognosis in Medical Care, Dr. Christakis examaines
the accuracy and role of prognosis in medicine,
and in particular, in end of life care. More recently, Dr.
Christakis has been exploring social networks and
health, more specifically, how ill health, disability,
health behaviors, which we’re gonna hear about today, health care, and even death in one person can influence the same phenomena among members of that
person’s social network. For example, he has explored in the past how ill health or death of a spouse can have cascading effects
for the other spouse. So in another work that
we’re gonna hear about today, Dr. Christakis is exploring
the dynamics of health and health behaviors in social networks. So join me in welcoming
Dr. Nicholas Christakis. (audience applause)
– Thank you. Okay so um. So given the size of this audience and the kind of informality
to which I’m accustomed, I’m happy to pause or take
questions at any time. Thank you Denise for the introduction. So I also wanna acknowledge
my collaborator, James Fowler, on this work. Today I’m gonna be discussing
some of our current work on the spread of health
behaviors within social networks. But as Denise already mentioned,
what we’re interested in is something quite broader than that. We’re interested in how
it might be that illness or death or healthcare use or
disability or health behaviors in one person might spread
and cause all those same outcomes in other people
to whom they’re connected. We’re interested in taking
seriously the contention that people are in
bedded in social networks of a substantial complexity, trying to disentangle that complexity, understand what it means, and trying to articulate a kind of non-biological spread of disease, a kind of sociological spread of disease, whereby
things might spread from person to person, but
not via that kind of customary mechanisms with which
we’re usually accustomed, whether it’s a spread of germs,
or a spread of second-hand smoke, or the kind of
biological or physical means with which we are usually at want to think about these kinds of things. Now, I’m gonna start with
some work that we’ve done in obesity, which we’ve
published more recently, but I’m also gonna speak
a little bit about smoking and eating behaviors, and then
actually talk a little bit about the spread of happiness
within social networks. To what extent is the
case that your happiness depends upon the happiness of the people around you, how might it
be that there could be ripples of happiness
through the kind of fabric of social networks that you might imagine. Now, as most of you probably know, the prevalence of obesity
has increased dramatically in the last 10 years, from
about 20-30% of Americans are now properly obese,
and about two thirds, 66% of Americans, are properly overweight, by conventional clinical standards. And numerous explanations
for this epidemic have been advanced, including
the change in the way we design our cities, to the
way we market food stuffs, which either increase the
amount of food we take in, or decrease the amount
of calories that we burn. But it has become in fact
trendy to speak of the obesity epidemic, and all this use of this sort of metaphorical expression led me and James Fowler to begin to wonder whether it was possible
to imagine that obesity was truly epidemic. Epidemic literally in
the sense of spreading from person to person. And so that’s what prompted
us to try to design the kind of study I’m about to show you, to find a source of data
that we could examine, and then analyze it with
a variety of methods that give us great confidence
that it is the case that weight gain in one person actually spreads to cause
weight gain in other people to whom they are connected. Because to the extent
that obesity is a product of personal choices and
voluntary behaviors, the fact that people are
embedded in social networks and are influenced by the
appearance and behaviors of other people who surround them means that weight gain in one individual should in fact influence
weight gain in others. And some of you may have
seen the press coverage of this article when
it came out in August, or July, this past summer, did some of you see this? It was amazing, I mean
a lot of the headlines were completely misdirected. They said, you know,
ditch your fat friends, which is not at all
(audience laughter) which is not at all what we were saying nor what we think should be done, and it was interesting both
in the newspaper coverage and the blog coverage
and a lot of the emails that we got, we got
something like 1,000 emails, couple of death threats, a
lot of people misunderstood what we were saying. There were two kinds, there
were a variety of narratives, but one of the bifurcations
of the narratives were a group of people who
said this is preposterous and outrageous, how could it possibly be that weight gain is
contagious, or spreads from person to person? And then another group
said, this is absurd, this is obvious, clearly
it is the case that human beings are influenced by the choices and behaviors of other human beings. And to me it was
fascinating to kind of watch the cleavage by which this work was seen and was interpreted, and
we’ll come back to that a little bit later, especially
if you’re interested. So this is a slide that shows
the network we’re gonna study. We painstakingly
reconstructed this network using archival data that had
previously not been used. The Framingham Heart
Study of course has been around for 60 years, but
there were these paper records in the basement that kept
track of who is who’s neighbor, or where did everyone
live, so we were able to reconstruct neighbor
ties, who is who’s friend, and we were able to
reconstruct friendship ties, family members, where did people work, so we were able to see that by dumb luck different individuals
worked at the same place, and then we were able to
reconstruct all of these ties for a long period of time. This graphic shows about
12,000 individuals. Every dot is a person. The red perimeter dots are women. This is a convention we’ll keep through the rest of the talk today. And the blue perimeter dots are men. The lines can indicate different
kinds of relationships, friendship ties, sibling
ties, spousal ties, neighbor ties, coworker ties, cousin ties, parent-child ties, and so forth and so on. Here I do no show
neighbor or coworker ties, and the kind of tie can
be indexed by the color, and then the arrow head can index the direction of the tie, which
is an important detail I’ll return to shortly. I might say that Denise is my friend. That’s a directed tie, or if
she’s my sister, there’s no sense in which there’s a direction, right? We’re siblings to each
other, that’s the kind of bi-directional tie. And we’re interested as I
said in studying what it means for people’s health, that they’re embedded in social networks of such
complexity more generally. One topic we chose I’m
not gonna discuss today, which is itself quite
interesting intellectually, is the following: If I were to ask you, where
would it be better for you if you lost your job, where
would you rather be located, would you rather be located in the, do we have a laser pointer or no? It’s okay. Where would you rather be located, would you rather be located
in the center of the network, or would you rather be located
here if you lost your job? Thank you. Where would you rather be located? – [Audience] Center. – [Nicholoas] Center, right? Because if you lose your job you might be much more likely, oh it doesn’t work on this- – [Voiceover] It doesn’t work well on- – [Nicholas] On this screen, yeah. You’d rather be in the
center because you’d be much more likely to
know somebody who could help you find a job, and similarly, most of you probably
have the sense that it’s good to have social support, being located centrally
in the network might confer some advantages to your health, you might be better of
being located in the center than in the periphery, okay? That’s the kind of question about your topological position, where are you located in this topology, in this social fabric? Which I’m actually not
gonna talk about today. Instead today I’m gonna
talk about the issue of how things spread within this network. Once you have this topology, once you define this social fabric, what might flow through it? Because if I were to ask you
where would you rather be in this network if you
were trying to avoid catching a lethal epidemic germ, where would you rather be? In the periphery, not
attached to anyone, right? So they’re different senses
in which different things spreading within the network might affect you differently and your
topological position might be important. Now these people in this network are observed repeatedly across 30 years, and these are the waves
at which they’re observed, roughly every four years, and they come in to be
examined by their doctors, there’s virtually no loss to
follow-up in this network. They’re weighed by their doctors,
their height is measured, their blood pressure and
all kinds of other physical examination measures are taken of them. They report how much they smoke, they took these measures of depression, the C-E-S-D, the standard
measure of depression, and a whole host of other
pieces of information were collected about them. And at inception in 1971,
the group that I’m gonna call are Egos, these are the source, the core individuals if the network, and then we’re
gonna look at what happens to them and they’re nested
within a larger network of 12,067 people. These egos who enter the study in 1971 upon entry have these attributes: they range in age from 6-70, they have an average age of 37, they’re about half and half men and women, they have the educational
and health distribution, that baseline that I just showed you. Now one of the first things we did in the obesity project is the following: we wanted to draw some
pictures to understand is it or is it not the case
that there are clusters of obese and non-obese
individuals within the network? So this is a graphic
that shows, again nodes, this shows the largest
connected subcomponent of the network, there’s just
2,200 people in this network. We make the node size
proportional to people’s BMI. So bigger people get bigger nodes. And then if their BMI is above 30 we also color the node
yellow, sort of give you a visual clue that this person
is overweight or obese, versus green, the person is not. And then we have two
different kinds of colors for the ties here. The purple ties are non-genetic ties, friendship or spousal ties, and the orange ties are
potentially genetic, namely familial ties. And if you squint here,
you might see that there’s some clustering of obese
and non-obese individuals within the network. Now I can look at these
images now because I’ve been looking at them for a long time, and I can look at this image
and see that there actually there is a tendency of some clustering, but we actually need a
more formal mathematical way of persuading you that
there is clustering, number one. Number two, persuading you that
any clustering that is there is more than would be
expected by chance alone, because by dumb luck if you
see people in a stadium, maybe you’re gonna have
clusters of women and clusters of men, and you
wanna know is that clustering exceeding what we would be
expected just by dumb luck. And number three, if there
is clustering and it’s more than can be expected by chance alone, we would like some way of telling you well how big are those clusters? Are they small, are they large, how far do they reach, right? If you get an obese individual
or a non-obese individual, ‘cus both cluster, does
it extend far or near into the distance? Now in order to do this, we had to exploit some
techniques that had been previously used by my
colleague Laszlo Barabasi whose a physicist that’s
been studying networks for a while and most
recently has been studying networks of genes, and so we are gonna use what’s called a topological permutation test. But I think I can walk
you through it and explain why that test is so valuable
in the current setting. Imagine that we have a situation in which we have a network of people, and 40% of the people are obese. And that the obesity
is randomly distributed through the network. So now we have this random BMI network, and in the center we
have an obese individual who is orange in color, and that person is
connected to five friends or siblings or social contacts, one degree removed in
the shell of this onion. You would expect an expectation
by dumb luck or alone two out of five of those
individuals, or 40%, to be obese, okay? If you go out two degrees removed, and now there are 15 people
in that shell of the onion, you would expect six of 15, again 40%, of the people to be obese. And if you go out three degrees removed, and there were 25 people in that shell, you would expect an
expectation of 10 out of 25, or 40% of the people to be obese. So at every degree
removed you have roughly, on average, the same from
every obese individual the same prevalence of obesity as is seen in the total population. On the other hand, if there’s clustering, if there’s not a random distribution, you’d get what you see on the left. So in the middle we again
an obese individual, but you go out one degree, you’d see three out of five, or 60% of those
person’s, the obese person’s friends would be, or social
contacts, would be obese. Two degrees removed, seven
out of 15, or about 50%. Three degree removed, eight
out of 25, or about 32% of the individuals would be obese, okay? What this means is it’s like
you have like these peaks and these valleys in
sociotopological space. You have this kind of
social network pattern that defines this kind of terrain and you have peaks of
individuals who are overweight, going down into valleys of
people who are normal weight, and if you pick me and I’m
underweight or overweight and you move a certain
degree removed from me, every step away you
remove from me contains some information about
the probability that these individuals would be obese or not, and you have to go three degrees removed or four degrees removed
before there’s no longer any association between
the body size of the person that’s my friend’s
friend’s friend’s friend, and my body size. And one can do this and then generate a variety of statistics that
tell us whether the amount of clustering that’s
observed exceed that which would be expected by chance alone. You can create these sampling replicates that allow you to do that. So when you do that what we find in fact is the following: This slide shows on
the y-axis the relative increase in the
probability of ego obesity. Ego is me, alter is Denise, conditional on given alter is obese, depending on the alter social distance, are we talking about
my friend or my spouse or my sibling, or my friend’s friend, or my friend’s friend’s
friend, one two three, up to six degrees removed of Kevin Bakin. So we get here one degree
removed and you can ignore the waves, the number of
waves that are shown there, that’s just the wave of the examination, what we find is that it alters distance one
from an obese individual have 40% excess risk of being obese than would be expected
due to chance alone. Alter’s distance two
have roughly 20% excess risk of being obese than would be expected due to chance alone. Alter’s distance three,
regardless of time, regardless of wave, have
roughly 10% excess risk of being obese than would be expected by chance alone, you have to get to four degrees removed before there’s no longer
any association between the weight status of a
person that far away from you and your own weight status. So what we’ve now shown
is there is number one, more clustering than
there would be expected due to chance alone in the network, and that this clustering
extends out to three degrees of separation. Your weight status is related to the weight status of people who you might not even know who are two or three
degrees removed from you. – [Man] Question. – Yeah. – [Man] On the cluster
diagrams, you took the 25 that’s given. Is there also a feedback into a characteristic like a health challenge or behavior that limits the size of that. Smokers might have
smaller social networks, does that factor- – Okay, so the way we do this, the way the statistics work is we observe a topology and we fix it, and then
we observe the prevalence of obesity and then we take that topology and randomly distribute obesity through that network and then we do that again and again a thousand times. And then we get a sampling distribution of the clustering of obesity around every obese individual in the random network, and we test the hypothesis
that the clustering, the observed network, is
more than would be expected by chance alone in this distribution. And since the topology is
fixed, it’s net of the thing that you’re describing. So whatever the propensity of attributes to contribute to the
formation of a particular structure of the network, we take that as a given. – [Man] Okay, and another question. So when you describe dumb luck, I mean, some of it’s dumb genetics, – Yes, – I’m in my brother’s social network, – Correct, that’s right. – and
– That’s right. – BMI is systematically
overstated for both of us because we’re big boned. – Right. – So…
(audience laughs) – Me too, I’m with you brother, I’m in the same group. Sign me up for that. – Right, so how does
your sampling technique take into account the genetic- – Okay in this, in a
minute I’ll show you things where we remove all the genetic ties, I’ll show you some other images, but everything I’ve shown you so far, all the estimation keeps them in, but that might confound
estimates at degree one, but that’s as far as it will reach because if you are Denise’s sister, if I’m your brother and
Denise is your sister, we take the shortest path link to me, so she doesn’t count as distance two, she only counts as distance one, so we make one hop to you my brother but then the next hop would be to your brother’s, to your wife, and then from your wife to her friend, so I’d be three degrees removed from your wife’s friend,
you see what I’m saying? Or your friend’s spouse or whatever it is, so at most it’s confounding just the one degree removed and we’ve
also done these tests removing the genetic ties. I’ll show you in a minute. – [Man] When you compare those you can sort of get a little bit of a peek
of nature versus nurture relative to the amount of
clustering you would see because people share the same genes, the social cluster that
you get because they share the same non-genetic linkages is big or small. – I’d have to think more
about that suggestion. We’ve done some other work, this came up in the talk earlier today too, we’ve done some other
work looking at the extent to which your weight status and the concordance of your weight status with the people to whom you’re tied affects the kind of
network that you are in. For instance, if you’re bigger,
do you have more friends than if you smaller, or if, one question, or if you’re weight is
becoming more discordant from your friend’s weight, are you more likely to sever
the tie with that person regardless, but that’s
another topic, I think, of conversation. Okay. So when we see the cluster,
and I’ll actually come back to this at then end if there’s
a little bit more time then, because I wanna, this topic’s
come up again and again today, and actually there’s a whole
set of questions that could be engaged, and once I
present some of the results on smoking, we’ll have more vocabulary, so we can talk a little
bit about what happens. – [Man] (inaudible) – Yeah, no no, it’s yeah. (laughter) Um, okay. So having observed the
existence of homophily, or rather, observed the
existence of clustering, now the question becomes, well what’s causing the clustering? And there are at least
three possible explanations. One is homophily, the
preferential tendency of like individuals to
befriend like individuals, birds of a feather to flock together. So more educated people hang out with more educated people, Dartmouth people hang out
with Dartmouth people, overweight people may hang out with overweight people, or underweight people
with underweight people, and so forth. So when you see clusters
of people that resemble each other, it might merely be a function of the preferential formation of ties between such individuals. A different possibility is confounding. So Denise and I, let’s
say know each other, we live near each other, and a McDonald’s open up near us, and the McDonald’s causes both of us to gain weight. Now you see me and Denise gaining weight, but it’s got absolutely nothing to do with me causing her to
gain weight whatsoever, it’s some third factor
that’s causing both of us to gain weight at the same time. And a third possibility
and for present purposes today one of the most interesting ones is this issue of
induction or peer effects, which is a question of
does Denise’s weight gain cause me to gain weight,
or her weight loss cause me to lose weight. And in fact everything we
looked at was symmetrical in terms of weight gain and weight loss, which was also interesting in
terms of the press coverage which went wild talking
about ditch your fat friends and stuff which is ridiculous but didn’t talk about how actually all our work talked about weight loss as well. In fact it was also very
interesting that the emphasis was always on how your friends are making you fat, and
a very American kind of perspective, never
considered that possibly you were making your friends fat, it was always inbound, you know, others were responsible, you had no responsibility for anyone else. Okay, so um, there are a
variety of sort of statistical and conceptual approaches
to getting a handle on distinguishing these things. Incidentally, there’s a very famous paper by Bruce Sacerdote here I think on whether or not having roommates or friends who are very good students improves your GPA, and what he did is he took advantage of a Dartmouth randomization of roommates during their freshman year and found that if you were randomized to have a roommate who was very studious, your GPA went up because you started copying
your roommate’s behavior. And so in a way, we’re after a similar kind of phenomenon. Okay, now the longitudinality
we have in the network, the fact that it’s repeated measures, gives us some opportunities
to begin to tease out some of the effects that I
was just asked about as well. Sorting out the difference
between the preferential formation of ties between
people and this kind of process of induction, or spread. This is a video animation
I’m gonna show you. In a minute I’m gonna put it into motion. It has the same conventions as before, the size of the node is
proportional to people’s BMI, bigger nodes are bigger people, and if the node is yellow
if the person’s BMI is above 30, here we only show non-genetic ties, only non-genetic, and the grey ties are spousal ties, so this man in blue is
married to this woman in red, and then the purple ties
are friendship ties, this woman in turn has this other woman as her best friend, she in
turn is married to this man, he in turn has this friend and so forth, and you can begin to trace out
these kinds of connections, these kind of threads in the social fabric through this network. In a minute, I’m gonna put
this network into motion, we’re gonna take daily
cuts through the network for 32 years, people are
gonna be born or die, and those will appear or disappear, ties will form or break
according to who befriends whom or who’s married to whom and the like, and people are gonna gain
weight, the nodes are gonna get bigger or smaller. But since this period of time includes the obesity epidemic in the United States, mostly you’re gonna see a sea of yellow arise over the 32 years that
this network is in motion. And finally, before I show
you this, I’m gonna tell you that we have spent, this
took half a million dollars to get to this point of making this video, and it was like one of the highlights of my professional career
because we had really wanted to show the following. When we first started this
work, I had this image that the obesity epidemic would be, that we’d be able to
find like patient zero. We’d be able to find
like an obese individual and that that person, that
there would be this spread of obesity through the
network of this person, and that it would be like
the physics experiments you did or you’re doing now in college or you did in highschool where you would take a table of water that’s very still and
you drop a pebble in it and you get this kind of ripple or wave, a concentric wave would spread out if you’ve ever been to a New England pond and you’ve thrown a rock in it, you know what I’m talking about, and it reaches the perimeter of the pond and then comes back and you get this interference pattern where you get these reinforcements of the waves and I kind of imagined
that there’d be this like, reinforced waves, that
there would be these peaks of obese and troughs of
non-obese individuals within this kind of
sociotopological space, that you have these kind
of reinforcement pattern within the network, which
as I said earlier today, is what the anthropologists
call culture, right? There’d be these areas within the network which would be people that sort of said it’s okay to be heavy, and other areas of the
network where people said it’s okay to be really thin, alright? So when I made this picture
I was really expecting to be able to see a single person who was gaining weight and then a kinda spread of
obesity out from this network, and this is what I wanted to see, and I want you to tell me if you see that when I show you this image. There’s one woman up here’s who’s gonna gain a lot of weight all of
a sudden, there she goes, and she’s gonna move to
the center of the network. And at the end, I can’t
freeze frame it, but you’re gonna see a cluster of obese
and non-obese individuals, but clustering’s more
apparent in this image than in the earlier image I’ve shown you, but it’s gonna cycle right
back to the beginning. Did you see this wave or not, of obesity spreading out? Why not? I looked at this, I was so depressed. (audience laughs) I like literally was waiting
to see this thing, you know, and finally the software
spit out this thing, it took two years to get
the money to do the project, two years to collect the data, and year to make it work, and literally five years from idea to looking at this video, and there it is and I don’t
see what I think I should see, and I’m really depressed
until I get a really obvious thought, what is the thought? Why do we not see that? Well the reason is that obesity is not a unicentric epidemic. It’s a multicentric epidemic and the proper analogy is not throwing a single pebble into the pond, but a whole handful of
pebbles into the pond. So you get a really choppy surface. This does not mean that a single pebble doesn’t cause the interference
patterns of the waves, it just means that if
you have many pebbles thrown into the surface at
once, it’s that much harder to discern what’s happening, to see it when you look at the image. And that’s why we’ve been using a variety of statistical techniques
to bring out the patterns, like earlier when I showed
you just the statics and I said we had to use those topological permutation tests to show you there’s more clustering, it’s a
similar kind of example, yeah. – [Voiceover] Could you play that again? – Yeah, sure. What’s amazing of course,
this is what we experience, this is what happens in our lives. We go through our lives
forming connections, people far from us are forming
and breaking connections, that might actually have
effects on us and we don’t even know it’s happening
beyond what is known as the social horizon, yeah. – [Voiceover] Quick question,
is there a relationship between age and obesity,
because you see individuals as they age.
– Yes. Yes, that’s correct. People tend to gain weight as they age. We could recreate this
image netting out the effective age and all
the statistical models I’m about to show you net that out, but age alone might explain the population prevalence of
obesity rising in this sample, but it doesn’t explain the
interpersonal effects that we, I’m gonna show you in a
minute, we’ll control for that. Okay, so our next step is to
specify variety of regression models to study this notion of induction, and here what’s gonna
happen is we’re gonna have a dependent variable in each model, which is ego obesity,
and independent variables which are lagged ego
obesity, alter obesity, lagged alter obesity, I’m
the ego, she’s the alter, ego age, gender, and education, and fixed effects of each wave, which combined with the age of inception account for the aging of the population over the course of time. We also have a variety of other lags so alter may change from t-minus one to t affecting ego’s weight at t plus one, but I’m just gonna show you the contemporaneous lags now, rather than further backed up lags, and the results are all the same. But in addition, before
I show you these results, I wanna highlight another detail. Let’s say I’m going to, you
can identify whether or not I’m a friend of Denise’s. One possibility is that I
say Denise is my friend, which would be an ego
nominated friendship, and Denise says I am her friend. That would be a mutual friendship. I nominate her, she nominates me. Another possibility is that Denise says that I am her friend, right? That would be an alter
nominated friendship. And a final possibility
is that I say Denise is my friend and she may have no idea who the hell I am, okay? That would be an ego nominated friendship and Denise has no idea who I am, that’s my high school experience. (audience laughs) So I think all these people are my friends and they don’t know who I am, okay, so let’s review. There’s mutual friendship,
we nominate each other. There’s ego nominated
friendship, I nominate her, and I know who she is. And there’s alter nominated
friendship, she nominates me, and I have no idea that
she’s nominated me. Everyone with me, the
three possibilities, okay? Which of those three do
you think is more likely, most likely that her
weight gain will affect me? Neutral, right, we’re very close friends, she gains weight, it’s most likely- next of those, which do you
think is the most likely? Ego, I nominate her, I
respect, I esteem her, I copy her, she does something. And finally? Alter. Alter, in fact, might
have no effect on me. I don’t even know who the hell she is. She’s nominating me as her
friend, I don’t know her, I’m not paying attention
to her, I have no idea whether she’s gaining weight
or anything of the kind, right? So in fact we might expect no effect of her weight gain on me. This directionality’s incredibly important for an econometric reason
known as identification, when I’m trying to identify the effect, as I’ll describe for you in just a moment. And what we find is the following: we find that for an ego perceived friend, and we did all these results
both with continuous BMI and with dichotomous obesity,
I’m just showing you the dichotomous obesity results, we find that if it’s an ego perceived friend becomes obese, it increases my risk of becoming obese by about 70%. If it’s a mutual
friendship, it’s about 171%, or if a mutual friend becoming obese nearly triples my risk of becoming obese, and an alter perceived
friend has no effect, okay? So if an alter becomes, if I, if Denise says I’m her
friend and she becomes obese, doesn’t affect me. So it seems to me like
this esteem is important, the closeness of the
relationship is relevant to this transmission of this obesity. Then we look at same sex
versus opposite sex friends. We find that concordant sex
individuals have an effect on each other, whereas
discordant effect individuals appear not to. Spouses have an effect on each other, siblings, and we see a
concordant opposite sex sibling kind of pattern, and immediate neighbors have no effect on each other. So if you’re immediate
neighbor becomes obese, it’s not associated with
whether you become obese. We have since looked at
coworkers which is not shown on the slide, and was
not in our published results, we find that if you have a
coworker who becomes obese, your risk of becoming obese
goes up, but only if you work in a small firm, you have to
be in a firm of six people or fewer, that may
contribute to relationship, you’ve got to actually
have interactions or know the person before it can affect you. Now this directional data
is especially important because it suggests that
confounding by unobserved factors is not the source
of the relationship. If it were the McDonald’s
that’s causing both me and Denise to become obese, that shouldn’t respect the
directionality of the tie, right? We should both be getting
obese regardless of whether I nominate her or she nominates me. So the fact that we see this
directionality of this effect, that it obeys a kind of sociological rule, increases our confidence
that some kind of causal thing is going on here,
rather than some kind of so called endogenous or
sort of a confounding kind of threat due to the
exposure to a McDonald’s or some kind of factor. In addition, the fact that
the effect is gendered among friends and spouses and siblings further supports the social nature of the effect at hand. If some unobserved environmental factor was contributing to both
ego and alter obesity, such as some unobserved local attributes or proximity to food, we would also expect weight change in
neighbors to be correlated but it is not. Also interestingly, apropos
of this neighborhood business, is the fact that the effect
decays in social distance, like I showed you earlier, your relationships with the weight status of the people you know is really intense one degree removed, less intense two degrees removed, even less intense three degrees removed, it decays in social
distance, but it does not decay in geographic distance. What this slide shows is
the lack of geographic decay and interpersonal
obesity effects of alters at distance one in our network. What we did is we took all the ego alter pairs that
are one degree removed from each other and we divided them into six equal groups based on how far apart they
lived from each other, okay? So these are people who live
let’s say in the same household next door, within a mile, within 10 miles, within 100 miles, whithin 1,000 miles. So one-sixth of the people were on average about 500 miles away from each other. And what we find is that the relationship between each ego and alter
obesity doesn’t matter, it doesn’t seem to matter
how far away you are. If you best friend is next
door or a thousand miles away, if they cross the
threshold and become obese, or cross the threshold
in the other direction and become non-obese,
it has an equal effect on your probability of
doing the same thing as if that friend were next door. So the effect of the
interpersonal friendship effect or social effect does not
decay with geographic distance. Taken together, we think the
findings that I’ve described so far support a role for
interpersonal induction in this phenomenon. Our model’s control
for baseline attributes of both ego and alter help me to account for the homophily, the tendency of individuals to form ties with each other. In addition, all our
models are marginal models, that is to say we look at, look at the tie between
me and Denis today, and you look at how my weight gained going forward is associated
with her weight gain going forward. So the other thing to
realize is that our results are not about who you’re connected to, whether they’re let’s
say thin or overweight, but rather what they are doing. You could be connected to a thin person who’s gaining weight,
that’ll contribute to your weight gain. So it’s not about the
status, the body status of the other person,
it’s not like you should quote “ditch your fat friends,” like the newspaper headlines said. If anything, it would be, well even that is a stretch, that you
should watch what the extent to which your friends are evincing certain kinds of behaviors. Now we also think that omitted variables are unlikely to be a big factor given the directional effects we observe. Plus if these omitted variables
were geographically based, for example, local wealth
or fast food or whatever, we would expect to see neighbored effects, which we don’t see, and
we would expect also distance effects, decay in distance, which we also don’t see. Now I should also mention
before going on that we’ve done other work
that looks at whether accounting for the spread
of a particular behavior, namely smoking, can help explain this. So here are the example
concern as the following: Well maybe what’s spreading
between me and Denise is not obesity as it were, but rather I’m copying
here smoking behaviors. She quit smoking and she gains weight, and I quit smoking and then I gain weight. And I can tell you that
after accounting for the spread of smoking cessation behavior within the network, and smoking does spread,
as I’ll show you in a bit, it doesn’t affect our obesity results. So looking at the spread of smoking doesn’t explain the spread of obesity. So what might explain
the spread of obesity? Well there’re a couple of broad classes. Some investigators,
some biologists actually believe that they’re
germs, literally viruses and bacteria of different
kinds that can spread from person to person and contribute to the obesity epidemic,
but what we’re interested in is not in the biological,
but rather in the social mechanisms by which obesity might spread, and there are at least
two or more, but at least two possible explanations. One is that alter’s appearance or behavior changes ego’s behavior. So Denise starts running,
and I copy her running and I start running. Or she stops smoking and I stop smoking. Or she says let’s go eat at the McDonald’s and watch the superbowl, and
I say great, let’s do that, and I copy the behavior,
so we jointly evince the same behavior. What spreads the two of us is a behavior. An alternative is that
alter’s appearance or behavior changes ego’s expectations or perceptions of norms, okay? So if my friend gains weight
and it changes my ideas about what an acceptable body size is and it make me start
gaining weight as well, here what spreads from person to person is not so much a behavior,
but rather an idea, a norm spreads from person to person. We see our results as being consistent with the latter explanation because first of all we
looked at one specific behavior, namely smoking, and that didn’t explain the effects that we observe. And second, the geographic
distance effects we speculate, these are speculations, support the idea of norms because we think norms might be better able to leap great geographic distances than behaviors are. You might see your brother once a year, and take at his, like I would
look at my brother Dimitri and see that he’s gained weight, and say oh my goodness, okay it’s okay for me to gain weight or I might form a
different idea about what an acceptable body size is, whereas no amount of
eating food with Dimitri on that particular day is
going to actually change my actual body size. So norms could spread with ephemeral or occasional contact, whereas behavior would
require more personal and constant contact we think. Now let’s look, any
questions at this point? Once you wind me up, I could just go. (audience laughter) Okay. So let’s segway a little
bit to look at smoking. This is a um, these images show smoking behavior clusters in the Framingham Heart Study Network. Again we make them node size proportional to how many
cigarettes you consume, more cigarettes bigger node, if you consume more than
30 cigarettes per day we color you yellow, 1971 you can see most of
the people are smokers, by 2000 you see that the people aren’t really smokers anymore, hmm? There’s been kind of a
disappearance of smoking, but there are few other
patterns that you could observe if you look more
carefully at this image. First of all, it seems
like by 2000 the smokers are more likely to occur at the periphery of their networks,
they’ve been kind of moved to the margins of the networks. And if you zero in on
this lower right hand kind of circled area over there and you look at who’s a smoker, you can see that they’re kind of occurring at the periphery of the network. They’re not evenly distributed throughout the network,
it’s like they been marginalized or pushed to the side. In addition, you can see
that, just visually perhaps, you can get the sense that um, that smokers are usually
in smaller subgroups than non smokers, so the cluster size is different from the
smokers and the non-smokers. Analyses similar to those
for obesity that we’ve done show that on average these
clusters of non-smokers and smokers stretched
out about three degrees, just like I showed you earlier. Now there are some additional more formal mathematical analyses you can do that are the following: The left hand graph shows the mean smoking cluster size and you observe mean smoking cluster size if you see the
top line is roughly flat across the waves of the study
over the 30 year period. That means the size of the
clusters that the smokers are in have stayed about
constant across time. First fact. The second fact is that the prevalence of
smoking has declined from about 65% to 20%
in this 30 year period. So the total number of
smokers has declined radically even if the cluster size has remained approximately the same. What would that tell you? Cause a way to square
those two observations. Yeah, what do we think
is happening is that whole groups of people
are quitting in droves. Entire clusters of smokers are dropping out of the network together. It’s not willy nilly,
people are just quitting to smoking on their own, at random, unconnected from other individuals. Local norms within the
social fabric change, militating against
smoking in one location, and people quit together. It’s as if your quitting
decision is not just about what you want to do, but about what the people near you, some of whom you, most of whom you don’t even know, what they’re deciding to do at that same time. So we think that what
this provides is evidence for people quitting in droves. The cluster size stays
relatively constant, even as the population
prevalence of smoking is declining. Now in addition, another
piece of information that’s also obvious is the, apparent from some analyses, is the following: You can compute something known as the eigenvector centrality, that’s the way of
summarizing mathematically how centrally located are
the people of the network, for example, if you look
at the network picture, individuals that are located in the center would have a higher number, higher centrality number, than individuals that are located at the periphery, okay? It’s the same kind of system that Google uses in its page rank system to identify which are the most salient or important nodes in the world wide web. And what we find is that the non-smoker
centrality in the blue at the top remains roughly constant, whereas the centrality
of the smokers declines across time. This just firms up the observation that the smokers are being pushed to the periphery of the network, right? They’re moving to the side, they’re being marginalized as it was,
not just statistically, but also sort of conceptually. Finally, analyses similar
to those that we’ve done for obesity show a dyadic spread in smoking cessation, but the dynamics of the spread of smoking were such that educated people both responded to alter smoking cessation more readily, and were more influential in getting
others to stop smoking. What we find in these
analyses is the following: If you I am deciding whether
or not to quite smoking, and Denise quits and
she’s highly educated, versus Denise quits and
she’s not highly educated, I’m much more likely to copy her if she’s more educated. And similarly, if Denise quits smoking, and now you’re looking at me, if I’m more educated,
I’m more sensitive to the stimulus of an alter quitting if I’m more educated than
if I’m less educated. It’s a kind of diffusion
of innovation like other kinds of things in our society where educated individuals are earlier adopters of the innovation
of smoking cessation. Now we’re beginning to explore other kinds of things,
exploiting other aspects of our data, for example, we’re
looking at food consumption, exercise, drinking, and
we’re beginning to examine some candidate mechanisms
for the spread of obesity. This is an image of banana
eating in the offspring cohort. Now we color you yellow
if you eat more than one banana per day,
otherwise we color you red. And if you look here you can see clusters of banana eaters and non-banana eaters, sometimes these clusters are
right next to each other, and they live very peacefully
without any conflict in the network
(audience laughs) in terms of banana eating
and non-banana eating, and we’re beginning to think about looking at this in more detail. We’ve also started looking
at such phenomena as the spread of happiness. Now emotional contagion is known to occur over dyads and over short time intervals, so many of you, for example, know that you might instinctively smile at someone in the subway, you know, a stranger smiles at you and you smile back, this is a very human, a very evolutionary adaptive kind of response. We also know from evolutionary biology that it makes sense for us to copy the emotions of the people near us so why would it make sense if you see a member of
your species that’s afraid, why would it make sense
for you to become afraid? – [Voiceover] It probably means
that something (inaudible) – Yeah, they’re seeing a predator and it’s good for you to,
oh my God you’re afraid, darn, you know, I better get afraid too and respond accordingly, okay? So there’s some evolutionary advantage to us copying. Emotions are a kind of
way of communicating one person to another, and
if you think about that, if you think about the
social role of emotions, reflect on this fact. Most of the emotions that we have we show. Why would we show our emotions if the purpose of them
was not to communicate information to other individuals, right? And that in turn begs a set of questions about social interaction. Why don’t we just feel our emotions, why do we demonstrate our anger or our joy or our fear or whatever
it is that we demonstrate? So it’s been shown that
happiness can spread over dyads and over short intervals of
time, it’s been shown before. Some other very clever experiments that have been done by um, by some economists that randomize waitresses who smile at their customers versus not smile at their customers, and then they would measure, because they’re economists, the amount of tips that the waitress got at the end of the day
(audience laughs) and this was a way of summarizing the effect of service with a smile, and I can’t remember the precise results, but it was like twice the
tips from just smiling when you serve your
customers versus not smiling. So there’s sort of iconic
industry of literature in this area which we wanted to extend by looking at whether
emotions could spread not just dyadically, but hyper-dyadically across greater social expanses, and also to see whether
emotions could last the spread could last
more than just a fleeting contagious smiling or tipping, but longer periods of time. And sure enough what we find is we find clusters of
unhappy and happy people in the Framingham Heart Study Network. Here if you’re unhappy, we color you blue, and if you’re happy we color you yellow, and in between we color you green. And again you can see clusters of happy and unhappy people within the network. You can also see that the unhappy people tend to be located at the
periphery of the network, right? They’re not central, they’re
kind of at the periphery of the network. And we find that the
happiness in an individual is associated with the
happiness of people, again, up to three degrees removed from them in the social network. So your happiness depends not just on the happiness of the people you know, but also on the happiness of the people that they
know, and also in turn on that happiness of the
people that they know! So it’s like you’re in
some kind of a groove, some kind of a happy space within the social network, and we imagined when we did this work that there might be some kind of ying and yang, some endless competition between
happiness and unhappiness in this kind of social fabric, blinking happiness, you know, and blinking unhappiness, a kind of battle at the margins between happiness and
unhappiness in the network. Similar analyses to those
that we’ve already done document that there is
a spread of happiness in the network, and we
also find that happy people have higher network centrality, that means that they
tend to be more likely to be in the middle of the network, have larger networks, and are located in larger clusters of happy people, so it’s not just that you know more people if you’re happy, but you know more people who are happy if you are happy. Now these results are even more remarkable considering that happiness
appears to require physical proximity and
because it also appears to decay with time. So there’s something known
the hedonic treadmill, this notion that in
order to make you happy I have to keep giving you
more and more good stuff. So at some point you
think, oh if I could just ace this examination, or
if I could just get a paper in the New England Journal,
I’d really be happy. Or if I could, you know, win the lottery, then I’d be happy. But then when we inquire
after you ace the exam or you get a paper in
the New England Journal, or you win the lottery, we
ask you a year or two later, you’re no happier than you were
before that thing happened. And in fact, even more
perplexing, if you survey people like my colleague Peter Youghal has at Michigan, survey
people who are paralyzed, who have lost an extremity, their happiness is not really much lower, if at all, than the happiness of people who haven’t suffered
those kinds of calamities. Now that’s a very counter-normative from a point of view of an economist or perhaps even any kind
of rational thinker, a counter-normative kind of proposition, that here you are, you
would look at someone, you would say if I asked
you how unhappy would you be if you became
paralyzed, you would say pretty unhappy. But then I go survey people who are paralyzed and they say I’m not unhappy. So what’s going on there? Do people adapt to their circumstances? Or if I said to you how happy would you be if you win the lottery? You’d say I’d be ecstatic,
but you go survery lottery winners and
they’re not, after a year or two years, they’re not
as happy as they were, as you would think they were. So we were curious to see if all these kinds of same
phenomena obtained here if Denise’s happiness makes me happy, well how long am I happy
after Denise gets happy? Is it, you know, for just
a month that I’m happy ‘cus she’s happy, or a
year, or what happens? And in fact what we find
is that I’m only happy for about a year and a half. (audience laughs) So if Denise becomes
happy, it makes me happy, but after a while, I’m
like okay what else is new? Denise is happy, enough already. And so you see that, you know, it takes, if she’s happy, I’m 40%
more likely to be happy for the first six months,
30% for the first year, and by a year and a half
or within the sampling era, so her happiness is no longer associated with my happiness. Interestingly, unlike the obesity results, we find that it matters how
far away Denise is from me. So if you look at the distance
between ego and alter, her happiness is only
associated with my happiness if she lives within two miles of me. Once you get beyond that, the ego alter pair, her
happiness is no longer associated with my happiness. And this, we think,
because for her happiness to cause my happiness, I have
to interact with her, right? I have to get, you know,
be happy, I have to have some kind of connection with
the person that is happy. So there are many conceivable, or documented examples of the spread of health related phenomena operating at the dyadic and we think hyper-dyadic level. These are some examples
that I’m summarizing here of health related phenomena many of which we’ve investigated, others of which are the work of others. There’s been a long tradition since 1858 of looking at the widower effect, death of one spouse increases the risk of death of the other. Illness in one spouse
could worsen the health of the other, last 25 years especially. You see a lot of people
who have caregiver burden, how my spouse’s illness
affects my own health status. Depression treatment
in parents could affect the health of their children. If you are thinking about
whether you aren’t sure, and you’re thinking about
whether you should provide for a medication that
might treat postpartum depression in women, you
should consider the possibility that not only would you benefit the women, but in treating them,
they might be more likely to vaccinate their children. So we’ve estimated you
could save 200 kids’ lives per year in this country by providing better treatment for moms after
they deliver their babies. Illness in children increases
the risk of death in parents, some nice work out of
Scandinavia that was published in the New England Journal
a few years ago now. Hip replacement in one spouse could reduce the disability in the other. Stroke prevention in
one person might benefit others, not just the person whom you prevent the stroke. Better terminal care reduces the risk of death in a spouse, a study we did of 200,000 couples show that if you take better care of me while I’m dying, you reduce my wife’s risk
of death during bereavement. Breast cancer in one person could prompt mammography in others,
sort of a well known anecdote story, a woman
gets breast cancer, all her friends go out and get mammograms. Heart attack in one person
could prompt weight loss, smoking cessation, exercise,
or aspirin use in others. This example raises the
possibility of some very perplexing kind of effects
when you begin to think about networks. It could be the case that
the following would happen: If I were to drop dead right now, I would lose, you know, 35
years of actuarial survival, but every one of you might
get the fear of God in you and start really taking
care of yourselves better. Exercising, eating right, quit smoking, take your vitamins, and so forth, and each of you might gain
a month, three months, a week, three weeks, five months, a year, and so forth of life, and
if I added all that up, that might exceed the
35 years of life lost because I died. It might actually be
really good for this group if I were to drop dead right now. (audience laughs) and not just ‘cus the
lecture would be over. So um, so this raises this notion of seeing us as embedded and interconnected
in these networks actually can raise some
paradoxical considerations. It also can raise some
other troubling ideas like the following: When organ transplantation
was first invented in this country, in the early 1950’s, there was a kind of a
bias towards giving the scarce organs to the family man, right? So you had a guy, he had
a wife and two children and he needed to support
them, he was given a preference to get this scarce organ in preference let’s say to a single guy, and this was seen as appropriate. After a while, however,
this was considered to be unethical, shouldn’t
be allowed, we really shouldn’t privilege someone based on this kind of circumstance, and there was actually a
whole fascinating history that my colleague Renee
Fox has written about in several books on organ transplantation and spare parts and
encourage to fail and others about how this sort of process
about allocation unfolded, but in any case it came
to be that eventually this was regarded as not a
proper way of allocating organs, but if you take what
I’ve told you seriously, you actually have to revisit that claim. You would have to say now wait a minute, if I give the organ to
this person, not only does this person benefit, but
all these other individuals benefit as well. We get more bang for our
buck from this scarce organ. So this might lead you
then to the conclusion that socially connected individuals
are more to be valued in our society than isolated individuals, which is a troubling kind
of place to find yourself in if you actually take seriously the things that I’ve been saying today. Seat belt use by one could save the lives of others, a beautiful
piece of work executed by colleagues of mine
at Washington University who showed that when you get into your car you should make everyone
buckle their seat belts, not just because if your
guy next to you buckles his seat belt it might save their lives, to hell with that, actually
it might save your life! Because if you’re in a
collision and they’re not buckled, their body
becomes a projectile and goes through the car,
hits you and kills you even though you’re buckled in your seat. Their body coming forward from the rear or from the side, and a very nice set of work looking at that, yeah. – [Voiceover] What about
the logic of other people in the car seat belt’s
unbuckled, you’re more likely to drive safely
because you’re afraid- – I really don’t buy that logic (audience laughs)
that’s like saying that’s like saying, you know, what would be an example, like you know if I really want to make
sure that you drive your car safely, I really should remove
the breaks or something, ‘cus then you’ll really be
careful when you’re driving ‘cus there’re no breaks in the vehicle, so um, although actually,
well we’ll come back to that, anyway, so alcohol, smoking, weight loss, or medication behaviors could spread, we saw some examples of that, and emotional states and
other health statuses might spread as well. So the fundamental fact
that people are embedded in social networks and connected to others has a number of consequences,
and this is my last slide. Health, health care, and
health behavior may have specific interpersonal collateral effects. Second, it means that something know as health externalities can arise. The classic example of
economic externalities is of course the industrial polluter, you’re a factory and
you’re making widgets, you pollute the environment, and the cost is born by the people that are downstream, and it’s not reflected
either in your balance sheet or in the price of the
widgets that you manufacture. So the transaction is
between you and the people that purchase the
widgets, but there’s this other person that’s harmed, the people who live around the factory, these externalities that bear the price of these externalities. Well the other classic
example is home improvement, you go to your house, you
work and you fix up the yard, or you paint your house,
and your neighbor’s property value goes up as a result. Strictly speaking, according
to economic theory, you should go and tax
your neighbor to acquire back some of this benefit
that you have caused. But what I’m interested
in is not in economic externalities, but rather
in health externalities. How is it what happen to me affects the health of others
to whom I am connected? And finally, it means
that social, clinical, and policy multiplier effects may result. And by multiplier effects
I mean that health related occurrences can arise as
a result of the existence of social networks and
can have cascade effects within social networks
that spread out from and back to the affected individual. Behavior changes,
clinical care, and policy maneuvers can have an important effects beyond the individuals
at whom they’re directed. In short, health, health
care, and health behavior are collective social
network based phenomena. Thank you. (audience applause) So wait, there was one
little, oh the question you asked about should you drive fast, there’s a very interesting idea that we’ve made all these improvements in the last 50 years,
there was some, I think there was a paper in the
American Economic Review or UJE, I can’t remember, on the following set of observations: Over the last 50 years, we’ve made huge improvements in highway
safety, we have mandatory seat belts, we’ve got
airbags, and set buckles, and all of these things,
but our fatality rate from motor vehicle accidents
hasn’t moved that much. So what does that tell you apropos the question I was asked? People are driving less
responsibly, that’s right. Because of all of these safety features, people are driving less responsibly, so people you still have the same ever, it’s just like a hedonic treadmilll again, we keep improving safety, but people then get more and more reckless as a result, so on average we never make much progress. Yeah, other questions. – [Vocieover] If we have any questions, please wait for the microphone. – [Man] Well, I guess two
things, I mean on that one, there are other benefits,
we get places quicker, we just
– I’m not saying we should dispense with safety
– Right. No no no, but the highway improvements, it’s just that the
benefit comes elsewhere. I was just wondering what
the role of television was in your thoughts about
the social networks? So for example, when you said I was worried about patient zero and I felt bad for ya
when I saw the graph, I was thinking Oprah, I mean God bless her with all of her weight
fluctuations and how influential she’s been, I mean television penetration was probably changing over the course of your sample, – But that should affect everyone in a secular way, right? It shouldn’t really,
you’re right, there’s no doubt television affects weight, – I think like, just
thinking of the regression, there’s an interaction on these contagion effects for whether you
have a television or not. I would argue that the
strength of those social ties that you measure person to person are muted when the person to person ties become a smaller fraction of all of the interactions they have, so this is, I was a little
kid in the 70’s and you know, I no longer had people
babysitters, I had the television babysitter.
– Yeah. – So you measuring these things, you’re gonna miss the fact that those social interactions,
no matter how many of them I have, are passed through this filter, which is the television, and I’m thinking- – You mean the ties
you have to celebrities or non-real people? So like Katie Couric’s
colonoscopy for example, everyone runs out and gets a colonoscopy, or Nancy Reagan’s mastectomy, everyone runs out and get a mastectomy because Nancy Reagan gets one, you mean? The effects of people who
really have relationships with their soap opera star? – You and I don’t know
each other, but then we become friends, we’re in this network. I have a television and you don’t, so you’re interaction with me becomes more important to you than my interaction with you because the television is where I have most of my interaction. So just thinking about
the way you lay this out, are the results the same if you have a TV versus if you don’t have a TV? – I don’t know, of course
we don’t have information about TV ownership, but
I’m not gonna answer the question that way, I
don’t, I’m not fully with you on your concern, I’d
have to think about it a little bit more, um, and I’m not sure you’re right, but you might be right, but I don’t, I have to think about it
harder before I can respond. – [Man] So it’s um- – But I will say one other
thing, we had a crazy, one guy that wrote to me
after a paper came out and I thought it was crazy at first, but now I’m not sure he was so cooky. This guy writes to me and he says I think the reason we’re getting fatter is because we have wide screen TV’s. I’m thinking what the
hell is this guy talking, I read further down in
his essay, and he says that a lot of the old
movies like I Love Lucy and all the programming was made for TV’s of a certain aspect ratio,
and then when they’re stretched out to fit the wide screen TV’s, all the people are made to look fatter and that this then changes our ideas about what the normal body size is because we’re seeing all
of these kind of like oompa loompa type, you know, stretched out kind of figures, on the television screen, so um, I don’t know if he’s right or not on that, but anyway, the TV example
made me think of this. Yeah. – [Woman] Thanks. I would like to ask,
at some slide, I’m not quite sure of all the academic
effects you looked at, like the emotional part, the obesity and the cigarettes kind of can be brought down to
the same basic mechanisms, so I would like to go
back to one of your slides where you were looking
at is it behavior or more change in norms that
causes change, for example, in friends, um, what I would like to think about is these are norms shared about what is the proper body size or what’s, you now, beautiful or not- – It wouldn’t have to be
shared, it wouldn’t have to be exclusively discussed, it
could just be witnessed. – Yes, witnessed. What I was wondering about if there’s actually an agreement to a more independent norm we share in society – Yes. – the decrease of difference, so if I have a friends who’s
gaining a lot of weight I might feel very uncomfortable
to be very slender and slim and just in order to decrease the difference between us I might be wanting to gain weight, so my question would be
– That’s a very good question. – If there’s a reference
to a more abstract norm, like kind of
– Equality. – Yeah, would it make
then sense to have a cross cultural study with
societies where probably those kind of norms are very different, or would that make sense? – There would be a lot of
effort, you’d have to spend a lot of blood to answer that question, like if we have to re-do
the whole study in another situation where we would
then, let’s say, if we found the difference, hang
the whole difference on that this is a more or
less equitable society, let’s say, the American society. I think your general point is excellent that there might be other kinds of norms regardless, not just norms
that drive the conversions, it’s not just that I’m copying your body size, it might be
that I wish to be similar to you, I’ll wanna dress alike to you and it’s not that I am
copying what you’re dressing, it’s that I feel like you and I should be dressed similarly or look similar or whatever. So and but our results,
you know, we’re not hanging our hat on the mechanism and even if your argument is true, it still is under the category of norms instead of shared behavior so it doesn’t subvert the general claim. What I thought you were gonna say which is also an interesting idea is the fact that even while we have claimed that these norms are important, many people will say well Nicholas, our society still privileges thinness, the super models are as
thin as they ever were and so how can that be? And the way I square that
circle is the following: I say that even though our ideology, I’d write this interesting
ideology in norms, even thoughour ideology is
still privileges thinness in our society, people’s
behaviors is much more influenced by the people they know, real people that they interact with, rather than abstract
ideological committments to people on the covers of magazines or on movies and the like, so I think that’s how we can make an argument about the spread of norms within a population,
even while the ideology might be different and
in fact you can make similar arguments about
all kinds of other things in our society, you
know, why do some people, you know we have a commitment to constitutional rights and to Habeas corpus and that’s like written in our laws and everyone
takes that very seriously, but then you might get
small groups of individuals who actually violate
that, and maybe they’re doing it under the influence
if you buy the whole Stanford Prison Experiment, they’re doing it because
of reference to the immediate population that
they’re interacting with, so local norms speak against some kind of broader commitment we might have. Yeah. – [Voiceover] So is that what’s happening in Washington? (audience and Nicholas laugh) – I set myself up for that a little bit. I mean, I think that,
I there is a kind of um sense in which people can act, would do things like in a riot that they would never do if they
were just by themselves, that kind of a mob
psychology can take over where people really behave badly because they think they
have a kind of anonymity or they think the rules have changed or everyone around them is behaving badly, that they would actually not do if they were acting in isolation, so I think these ideas
that we’re discussing today in the health setting
actually could be a broader relevance in our society. – [Denise] I just wanted to follow up on the norm mechanism
‘cus that’s the thing – Yeah we were talking about that earlier. – [Denise] Yeah, that I’m interested in and I think um, I don’t think this is
inconsistent to any of the things you said, but sometimes the norm comes not just from you adopting either the idea or the behavior, but
that other people then sanction you when you don’t follow it, and that kind of follows from this that it may not be so much that you and I wanna be similar, but other people might react to our dissimilarity, and that then reinforces or punishes certain kinds of behavior and I think um, it’s interesting to think
about how that might flow through for these kinds of behaviors. – That’s an excellent example and one of the famous examples of this is the notion of triadic closer, so if you imagine this person, this young woman in high school, A, and she has two friends, B and C, now the question is under which circumstance, and here… in circumstance number
one, the two friends B and C don’t know each other, and in circumstance number two, the two friends B and C know each other. In which of these two
circumstances do you think person A is gonna be more likely to obey and do what her
friends tell her to do? Two. Because these people could communicate with each other and enforce a norm on her that is much harder for them to do here where they can’t coordinate. This is a notion of triadic closure that the sociologist George Simmel introduced a hundred years ago, we’re actually looking at
this a little bit in Facebook in our work with Facebook networks, and it relates very strongly
to what Denise just said, but there was work done by Peter Bareman using the Add Health
data set that suggests that girls that are located
in this type of a situation are more likely to have suicidal ideation than young women who are
located in this circumstance, because here they get
less consistent messages from their friends, whereas
here they get a more consistent message from their friends, or less likely to be
buffeted by contrarian views, so there’s a kind of
coordination that occurs outside and it’s no just about the dyad, it’s about what other
individuals are speaking to the dyad. – [Voiceover] I think we
can take on more question. – [Woman] Mine is a very, is this on? It’s just a very specific question, I was wondering in
reference to your charts, and the studies that you did, did you study a specific age group or did it kind of fluctuate? – Uh, we only studied
adults, people had to age into our cohort had to be 21 or older, and but the age range was enormous. It was like from 20 to 70 or something, and then the age to
cross the next 30 years, but the cohort was
replenished with the birth of new members or the addition of the generation three cohort
that I talked about in 2002 or the Omni cohort in ’95. – [Voiceover] Any last questions? Okay, well thank you very much. (audience applause)

One thought on “Eat, Drink, and Be Merry: The Spread of Obesity, Nicholas Christakis

Leave a Reply

Your email address will not be published. Required fields are marked *

Back To Top