The original title of this post was, “Remember that many-times-debunked claim of ‘contagion of obesity’? How’s it being cited nowadays?”, but that was misleading—see the followup. What I want to say is not that the claim was “debunked” but that the paper where it first appeared had statistical problems. This post is about how that particular paper was cited but, as its coauthor Nicholas Christakis helpfully pointed out in a comment, it did not at all engage with the later literature on the topic.
I happened to be talking with some students today about social network research—we’re doing some followup on our penumbra paper—and the topic came up of the controversial study by Nicholas Christakis and James Fowler from the 2000s on the contagion of obesity.
We covered the topic in this space back in 2010 and 2011:
Controversy over social contagion
Controversy over the Christakis-Fowler findings on the contagion of obesity
There we discussed the work of Christakis and Fowler; criticisms of that work by economists Jason Fletcher and Ethan Cohen-Cole, mathematician Russ Lyons, political scientists Hans Noel and Brendan Nyhan, and statisticians Cosma Shalizi and Andrew Thomas; and a response by the original authors, who wrote:
We do not claim that this work is the final word, but we do believe that it provides some novel, informative, and stimulating evidence regarding social contagion in longitudinally followed networks. Along with other scholars, we are working to develop new methods for identifying causal effects using social network data, and we believe that this area is ripe for statistical development as current methods have known and often unavoidable limitations.
The quick summary is:
1. Christakis and Fowler were doing interesting, innovative social science; they just went too far in the interpretation of their data. You know that saying, “high-risk, high-reward”? That’s what was happening here. There was potential high reward, but this study was ultimately a failure, except to the extent that failures can be useful too in helping us avoid certain dead-end paths in the future.
2. The claim of social contagion of obesity wasn’t supported by the data from the Framingham Health Study; the critics (Fletcher, Cohen-Cole, Lyons, Noel, Nyhan, Shalizi, and Thomas) were right.
3. There are social effects on attitudes and behavior, and they’re hard to study. As Christakis and Fowler wrote, this area is ripe for statistical development and also ripe for development in experimental design and data collection.
I was curious how this work is being cited, over 15 years later. Google Scholar lists 7000 citations. I searched for citations from this year, and here are the first few:
The first link above is to a book, and here’s the relevant passage:
The next is from a review article, which mentions the Christakis-Fowler paper as reference 31 here:
The next is a review on peer effects in “weight-related behaviours of young people,” which has this wrong summary:
Kinda makes me concerned about the rest of that literature review!
The next is a paper on “Community influence on microfinance loan defaults under crisis conditions,” which cites Christakis and Fowler not for their substantive claims but for a method they used:
It’s reference 40 in this next paper:
And it’s cited just a little bit too credulously as reference 15 here:
And then there’s this one:
They cited work of Brian “Pizzagate” Wansink! That’s not good. If this blog were a drinking game, everyone would have to take a swig right now.
Summary
The incorrect claim about contagion of obesity is out there, and just about nobody seems to be qualifying it with references to the critics. Sorry, Fletcher, Cohen-Cole, Lyons, Noel, Nyhan, Shalizi, and Thomas. Your hard work has come to (almost) naught.
P.S. In the comment section, Lyons shares some horrible examples of the persistence of the debunked claim an uncritical reference to the original, flawed article, in an exam for medical students. Your future doctor! This continuing episode says bad things about the scientific, media, and academic establishment demonstrates the challenges of citation and reference: if you want to discuss a topic, you should cite the original paper, but then you should cite the criticism, but then you should cite later literature on the topic, but then that’s a lot to chew on if you’re just trying to write an exam question . . .
This post made me think. I have cited studies that did experiments or analyses that I thought were poorly designed or had no chance of finding a signal in the noise. But they had kindly searched the literature, often on topics that I could not fully understand myself. I always thought that if I cited these studies and extracted a message from the study that might not be the main conclusion of the paper, that would be OK – not out of laziness, but because I am not able to find quality papers on a topic, but because they do not exist, or I just have not found any.
So my question would be, what do commenters think? Do they see problems with this kind of citation?
The question stems from our host’s comment that citing Wansink would be “not good”. As I understand it, the article in question only endorses the literature review of a particular Wansink paper, not Wansink and Chandon as researchers, or even the whole article they wrote. I may have misinterpreted our host’s words, but I still wonder about my dilemma above.
Raphael:
I would not cite anything by Wansink unless in the context of a discussion of serious scientific error. There is no way I would trust a purported “comprehensive review” by Wansink. In his research papers he has a demonstrated record of twisting or perhaps fabricating data to come to unsupported conclusions. I expect he would do the same thing in a review article. I looked up the Wansink and Chandon (2014) article, and it cites 25 items by Wansink in the bibliography, many of which are published research papers. My problem is not self-citation, it’s with citation of discredited work. This review article looks like an example of garbage-in, garbage-out, with the extra special feature that the collector of the garbage is the same person who produced it earlier.
I don’t have an easy solution to your dilemma. A starting point is to avoid trusting sources that we already have good evidence are untrustworthy.
Thank you for your reply.
Regarding Wansink: I looked up the article myself. My PubPeer plugin flags 26 articles cited in the paper! I did not go through them all – I am sure that Wansink’s work has been thoroughly discussed on PubPeer, and I would only touch this article with a very long pole, but not cite it. I have to say, the plugin is invaluable.
As for the dilemma of citing poorly designed research articles, I accept that it is a case-by-case decision. You are absolutely right to warn that we should not ignore flawed designs. I think you or someone else had named the fallacy of criticising one part of a whole and then assuming that the rest is fine. (The example was a newspaper article. A reader finds a mistake and gets livid about the poor research behind the article, but then accepts the rest of the paper as probably well researched.)
That’s “Gell-Mann amnesia”.
I’ve seen that term before, but wondered what it meant, and what famous physicist Murray Gell-Mann had done wrong to deserve to be tied to such a phenomenon.
Further reading (Wikipedia) reveals that the term was coined and defined by author Michael Crichton, who discussed it once with Gell-Mann and then named it after him to give it ironic credibility. It refers to an expert recognizing the errors in their own field but assuming articles about unfamiliar topics from the same source are correct.
Since I’m a climate scientist and Crichton is a global warming denier, I can easily recognize the errors in his denialism. Thus, in order to avoid Gell-Mann amnesia, I should avoid believing anything else he says. This necessarily includes his definition of Gell-Mann amnesia. Hence it doesn’t exist. So I may as well believe that it does exist. Hence it does exist. But to avoid it…
I have been aware that C&F continue to be generally cited uncritically. Moreover, I expected that from the start when I got involved. Nevertheless, I believe that the influence of our critical articles has been large, and the influence of mine is certainly much larger than I expected it to be.
Here is an interesting example of the uncritical acceptance of C&F’s work: https://www.khanacademy.org/test-prep/mcat/social-sciences-practice/social-science-practice-tut/e/is-obesity-contagious- It gives a practice question for the MCAT; note that this course has “the support of the Association of American Medical Colleges (AAMC).” (https://www.khanacademy.org/test-prep/mcat)
In this practice question, the “correct” answer is “There is almost no effect on the ego when someone in the same geographic proximity gained weight.” Of course, this accepts the null when there is insufficient evidence. I reported this error using the button there and received no reply. (I am ignoring the fact that all these statistics are unfounded to start with.)
Christakis still teaches these falsehoods and tells his Yale students that about 1/3 of the MCAT exams ask a question about C&F’s paper on obesity.
Here is an amusing event during my adventures to get my critique published. After failing to get it even considered at NEJM and the like, I rewrote and submitted it in July 2010 to “Statistical Science”. Three referee reports were uploaded Nov. 24–Dec. 24, 2010. The first two referees recommended it (“an important critique” and “well worth publishing”). The third referee said,
“The harsh critique of the C-F work offered by Lyons is largely accurate …[but this is not the] venue to publish extended critiques of the work of particular authors …[especially since] the C-F conclusions about obesity, smoking, and so on are just a light entertainment. They add nothing to the social sciences …”
The decision came on Feb. 19, 2011 and was to “Reject with resubmission”, making the tone more neutral and changing the focus, perhaps to “editorial decision making standards in medical journals.”
I’m sure Christakis will say “our evidence is weaker than we thought but our conclusions still hold” or whatever that classic quote is.
Anon:
I think he’ll just not respond at all, to the extent he can avoid responding. Recall the ladder.
I take that back (see Christakis’s detailed comment below) and I apologize to Christakis for assuming he would not respond.
This was a case where Bayesian inference led me astray—I was implicitly imputing his response from a distribution of past responses to criticisms by others, not appropriately weighting Christakis-specific information (that he does have a record of engaging with criticism).
Andrew: I think you can be more fair. It’s hardly appropriate to claim our landmark 2007 paper has been “debunked.” This is simply false. And the long comment I offered below (with 70 citations) is just some of the evidence supporting this. Furthermore, we have published several other responses to (sober-minded) critics in the past — and have in some cases profited from their comments. And we have done dozens of rigorous RCTs proving our ideas. You could acknowledge this better, in my view. I’d stack our level of rigor, honesty, and productivity against anyone’s.
Nicholas:
Yes, I agree. I’m right now in the midst of preparing a longer post, and I apologize for the inappropriate word “debunked.” I very much appreciate your patience in responding here and I do not want to mislead.
Andrew:
I just stumbled on this post in my twitter feed today. I must offer the opinion that I think this post is an incomplete summary of the very carefully expressed claims in the original 2007 paper, and also an inaccurate summary of the state of the literature. You also may want to look at our original exchanges on this blog from years ago, and to our published responses to prior critiques, including to some of the decades-old critiques (often inaccurate) that you mention.[1]
Many papers have replicated our findings of social contagion with respect to obesity (and the various other phenomena discussed in our original suite of papers), and many papers have evaluated the early methods we used (based on generalized estimating equations) and have supported that approach.
For instance, analyses by various scholars supported the GEE approach, e.g., by estimating how large the effect of unobserved factors would have to be to subvert confidence in the results.[2],[3],[4],[5] Other papers supported the findings in other ways.[6],[7],[8],[9],[10] This does not mean, of course, that this GEE approach does not require various assumptions or is perfectly able to capture causal effects. This is one reason the 2007 paper described exactly what models were implemented, was judicious in its claims, and also proposed certain innovations for causal identification, including the “edge-directionality test.” The strengths and limitations of the edge-directionality test for causal identification have subsequently been explored by computer scientists,[11] econometricians,[12] statisticians,[13] and sociologists.[14]
Work by other investigators with other datasets and approaches has generally confirmed the 2007 findings. Pertinent work regarding weight and related behaviors is quite diverse, including everything from observational studies to experiments.[15],[16],[17],[18],[19],[20],[21],[22],[23] Of course, as expected, work has also confirmed the existence of homophily with respect to weight. Still other studies have used experimental and observational methods to confirm that one mechanism of the interpersonal spread of obesity might indeed be a spread of norms, as speculated in the 2007 paper.[24],[25],[26],[27]
Of course, methods to estimate social contagion with observational data regarding complex networks continue to evolve, and continue to require various assumption, as ours did in 2007. As before, I would love to see someone offer a superior statistical method to observational data. And I should also clarify that the public-use version of the FHS-Net data (posted at dbGap) is not the same version as the one we based our 2007 analyses on (a constraint imposed by the FHS itself, in ways documented there); however, the original data is available via the FHS itself (or at least was). At times, this difference in datasets explains why other analyses have reached slightly different conclusions than us.
In our 2007 paper, we also documented an association of various ego and alter traits to a geodesic depth of three degrees of separation. We also did this with respect to public goods contributions in a network of hunter-gatherers in Tanzania[28] and smoking in the FHS-Net.[29] Other observational studies have also noted this empirical regularity with respect to information,[30],[31] concert attendance,[32] or even the risk of being murdered.[33] We summarized this topic in 2013.[34]
Moreover, we and others have observed actual contagion up to three degrees of separation in experiments which absolutely excludes homophily or context as an explanation for the clustering.[35],[36],[37] For instance, Moussaid et al documented hyper-dyadic contagion of risk perception experimentally.[38] Another experiment found that the reach of propagation in a subjective judgment task “rarely exceeded a social distance of three to four degrees of separation.”[39] A massive experiment with 61 million people on Facebook documented the spread of voting behavior to two degrees of separation.[40] A large field experiment with 24,702 villagers in Honduras showed that certain maternal and child health behaviors likewise spread to at least two degrees of separation.[41] And a 2023 study involved 2,491 women household heads in 50 poor urban residential units in Mumbai documented social contagion, too.[42]
In addition, my own lab, in the span from as early as 2010 to as recently as 2024 has published many demanding randomized controlled field trials and other experiments documenting social contagion, as noted above. For instance, my group published our first experiment with social contagion in 2010,[43] as well as many other experiments involving social contagion in economic games using online subjects[44],[45],[46] often stimulating still other work.[47],[48],[49]
Many other labs, in part stimulated by our work, have conducted many other experiments documenting social contagion. The idea of using a network-based approach to exploit social contagion to disseminate an intervention – so as to change knowledge, attitudes, or practices at both individual and population levels – has been evaluated in a range of settings.[50],[51],[52],[53],[54],[55],[56],[57],[58]
Finally, other rigorous observational and experimental studies involving large samples and mapped networks have explored diverse outcomes in recent years, beyond the examples reviewed so far. For instance, phenomena mediated by online interactions include phone viruses,[59] diverse kinds of information,[60],[61],[62] voting,[63] and emotions.[64] In face-to-face networks, phenomena as diverse as gun violence in Chicago,[65] microfinance uptake in India,[66] bullying in American schools,[67] chemotherapy use by physicians,[68] agricultural technology in Malawi,[69] and risk perception[70] have been shown to spread by social contagion.
The above, in my view, is a fairer and more complete summary of the impact, relevance, and accuracy of our claims about obesity in particular and social contagion in general. Work in the field of social contagion in complex networks, using observational and experimental studies has exploded since we published our 2007 paper.
Sincerely,
Nicholas
PS The convenient reason I have this material ready is that, as luck would have it, I recently wrote a (longer) review of this subject. :-)
[1] N.A. Christakis and J.H. Fowler, “Social Contagion Theory: Examining Dynamic Social Networks and Human Behavior,” Statistics in Medicine 2013; 32: 556-577.
[2] T.J. VanderWeele, “Sensitivity Analysis for Contagion Effects in Social Networks,” Sociological Methods & Research 2011; 40: 240-255.
[3] T.J. VanderWeele and O.A. Arah, “Bias Formulas for Sensitivity Analysis of Unmeasured Confounding for General Outcomes, Treatments, and Confounders,” Epidemiology 2011; 22: 42-52.
[4] K. S, Zachrison, T.J. Iwashyna, A. Gebremariam, M. Hutchins, and J. Lee, “Can Longitudinal GEE Models Distinguish Network Influence and Homophily? An Agent-Based Modeling Approach to Measurement Characteristics,” BMC Medical Research Methodology 2016; 16: 1-9.
[5] A.G. Steeg and A. Galstyan “Statistical Tests for Contagion in Observational Social Network Studies,” Proceedings of Machine Learning Research 2013; 31: 563–571.
[6] M.M. Ali, A. Amialchuk, S. Gao, and F. Heiland, “Adolescent Weight Gain and Social Networks: Is There a Contagion Effect?” Applied Economics 2012; 44: 2969–2983.
[7] A. Gonzalez-Pardo, R. Cajias, and D. Camacho, “An Agent Based Simulation of Christakis-Fowler Social Model,” Recent Developments in Computational Collective Intelligence 2014; 513: 69–77.
[8] Y. Qin, J. Ma, and S. Gao, “Efficient Influence Maximization Based on Three Degrees of Influence Theory,” In Web-Age Information Management: 16th International Conference, WAIM 2015, Qingdao, China, Proceedings 16 (pp. 465-468), Springer International Publishing, 2015.
[9] S. Aral, L. Muchnik, and A. Sundararajan, “Distinguishing Influence-Based Contagion from Homophily-Driven Diffusion in Dynamic Network,” PNAS: Proceedings of the National Academy of Sciences 2009; 106: 21544-21549.
[10] F.L. Pinheiro, M.D. Santos, F.C. Santos, and J.M. Pacheco, “Origin of Peer Influence in Social Networks,” Physical Review Letters 2014; 112: 098702.
[11] A. Anagnostopoulos, R. Kumar, and M. Mahdian, “Influence and Correlation in Social Networks,” Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 2008; 7-15.
[12] Y. Bramoullé, H. Djebbari, and B. Fortin, “Identification of Peer Effects Through Social Networks,” Journal of Econometrics 2009; 150: 41-55.
[13] C.R. Shalizi and A.C. Thomas, “Homophily and Contagion Are Generically Confounded in Observational Social Network Sudies,” Sociological Methods and Research 2011; 40: 211-239.
[14] W. An, “On the Directionality Test of Peer Effects in Social Networks,” Sociological Methods and Research 2016; 45: 635-650.
[15] A.A. Gorin, R.R. Wing, J.L. Fava, J.M. Jakicic, R. Jeffery, D.S. West, K. Brelje, and V.G. DiLillo, “Weight Loss Treatment Influences Untreated Spouses and the Home Environment: Evidence of a Ripple Effect,” International Journal of Obesity 2008; 32: 1678-1684.
[16] M.G. Barnes, T.G. Smith, and J. Yoder, “Economic Insecurity and the Spread of Obesity in Social Networks,” WSU School of Economic Sciences Working Paper 2010.
[17] H. Brown, A.R. Hole, and J. Roberts, “Going the Same ‘Weigh’: Spousal Correlations in Obesity in the U.K.,” Department of Economics, University of Sheffield, UK, 2010: Working Paper.
[18] K. de la Haye, G. Robins, P. Mohr, and C. Wilson, “Obesity-Related Behaviors in Adolescent Friendship Networks,” Social Networks 2010; 32: 161-167.
[19] J.G. Trogdon, J. Nonnemaker, and J. Pais, “Peer Effects in Adolescent Overweight,” Journal of Health Economics 2008; 27: 1388-1399.
[20] T.J. Halliday and S. Kwak, “Weight Gain in Adolescents and Their Peers,” Economics & Human Biology 2009; 7: 181-190.
[21] S.E. Carrell, M. Hoekstra, and J.E. West, “Is Poor Fitness Contagious? Evidence from Randomly Assigned Friends,” Journal of Public Economics 2011; 95: 657-663
[22] T.M. Leahey, J.G. LaRose, J.L. Fava, and R.R. Wing, “Social Influences Are Associated with BMI and Weight Loss Intentions in Young Adults,” Obesity 2011; 19: 1157-1162.
[23] A. Datar and N. Nicosia, “Assessing Social Contagion in Body Mass Index, Overweight, and Obesity Using a Natural Experiment,” JAMA Pediatrics 2018; 172: 239-246.
[24] B. McFerran, D.W. Dahl, G.J. Fitzsimons, and A.C. Morales, “Might an Overweight Waitress Make You Eat More? How the Body Type of Others Is Sufficient to Alter Our Food Consumption,” Journal of Consumer Psychology 2010; 20: 146-151.
[25] D.J. Hruschka, A.A. Brewis, A. Wutich, and B. Morin, “Shared Norms and Their Explanation for the Social Clustering of Obesity,” American Journal of Public Health 2011; 101: S295.
[26] M.C. Campbell and G.S. Mohr, “Seeing Is Eating: How and When Activation of a Negative Stereotype Increases Stereotype-Conducive Behavior,” Journal of Consumer Research 2011; 38: 431-444.
[27] M.A. Burke, F.W. Heiland, and C.M. Nadler, “From ‘Overweight’ to ‘About Right’: Evidence of a Generational Shift in Body Weight Norms,” Obesity 2010; 18: 1226-1234.
[28] C.L. Apicella, F.W. Marlowe, J.H. Fowler, and N.A. Christakis, “Social Networks and Cooperation in Hunter-Gatherers,” Nature 2012; 481: 497-501.
[29] N.A. Christakis and J.H. Fowler, “The Collective Dynamics of Smoking in a Large Social Network,” New England Journal of Medicine 2008; 358: 2249-2258.
[30] J.J Brown and P.H. Reingen, “Social Ties and Word of Mouth Referral Behavior,” Journal of Consumer Research 1987; 14: 350-362.
[31] J. Singh, “Collaborative Networks as Determinants of Knowledge Diffusion Patterns,” Management Science 2005; 51: 756-770.
[32] Y. Leng, X. Dong, E. Moro, and A. Pentland, “Long-Range Social Influence in Phone Communication Networks on Offline Adoption Decisions,” Information Systems Research 2023; online in advance
[33] A.V. Papachristos and C. Wildeman, “Network Exposure and Homicide Victimization in an African American Community,” American Journal of Public Health 2014; 104: 143-150.
[34] N.A. Christakis and J.H. Fowler, “Social Contagion Theory: Examining Dynamic Social Networks and Human Behavior,” Statistics in Medicine 2013; 32: 556-577.
[35] R.M. Bond, C.J. Fariss, J.J. Jones, A.K.E. Settle, C. Marlow, and J.H. Fowler. “A 61-million-person experiment in social influence and political mobilization,” Nature 2012; 489: 295-298.
[36] J.H. Fowler and N.A Christakis, “Cooperative Behavior Cascades in Human Social Networks,” PNAS: Proceedings of the National Academy of Sciences 2010; 107: 5334-5338.
[37] E. Airoldi and N.A. Christakis, “Induction of Social Contagion for Diverse Outcomes in Structured Experiments in Isolated Villages,” Science 2024; 384: adi5147.
[38] M. Moussaïd, H. Brighton, and W. Gaissmaier, “The Amplification of Risk in Experimental Diffusion Chains,” PNAS: Proceedings of the National Academy of Sciences 2015; 112: 5631-5636.
[39] M. Moussaïd, S.M. Herzog, J. E. Kämmer, and R. Hertwig, “Reach and Speed of Judgment Propagation in the Laboratory,” PNAS: Proceedings of the National Academy of Sciences 2017; 114: 4117-4122.
[40] R.M. Bond, C.J. Fariss, J.J. Jones, A.K.E. Settle, C. Marlow, and J.H. Fowler. “A 61-million-person experiment in social influence and political mobilization,” Nature 2012; 489: 295-298.
[41] E. Airoldi and N.A. Christakis, “Induction of Social Contagion Across Diverse Outcomes in Structured Experiments in Isolated Honduras Villages,” 2023; under review.
[42] M.A. Alexander, L. Forastiere, S. Gupta, and N.A. Christakis, “Algorithms for Seeding Social Networks Can Enhance the Adoption of a Public Health Intervention in Urban India,” PNAS: Proceedings of the National Academy of Sciences 2022; 119: e22120742119.
[43] J.H. Fowler and N.A Christakis, “Cooperative Behavior Cascades in Human Social Networks,” PNAS: Proceedings of the National Academy of Sciences 2010; 107: 5334-5338
[44] H. Shirado, F. Fu, J.H. Fowler, and N.A. Christakis, “Quality Versus Quantity of Social Ties in Experimental Cooperative Networks,” Nature Communications 2013; 4: 2814.
[45] H. Shirado, F.W. Crawford, and N.A. Christakis, “Collective Communication and Behavior in Response to Uncertain ‘Danger’ in Network Experiments,” Proceedings of the Royal Society A 2020; 476: 20190685.
[46] D.G. Rand, M. Nowak, J.H. Fowler, and N.A. Christakis, “Static Network Structure Can Stabilize Human Cooperation,” PNAS: Proceedings of the National Academy of Sciences 2014: 111: 17093-17098.
[47] K.R. McKee, A. Tacchetti, M. Bakker, J. Balaguer, L. Campbell-Gillingham, R. Everett, and M. Botvinick, “Scaffolding Cooperation in Human Groups with Deep Reinforcement Learning,” Nature Human Behavior, 2023: forthcoming.
[48] M. Tsvetkova and M.W. Macy, “The Social Contagion of Anti-Social Behavior,” Sociological Science 2015; 2: 36-49.
[49] D. Melamed and B. Simpson, “Strong Ties Promote the Evolution of Cooperation in Dynamic Networks,” Social Networks 2016; 45: 32-44.
[50] D.A Kim, A.R. Hwong, D. Stafford, D.A. Hughes, A.J. O’Malley, J.H. Fowler, and N.A. Christakis, “Social Network Targeting to Maximise Population Behaviour Change: A Cluster Randomised Controlled Trial,” The Lancet 2015; 386: 145-53.
[51] L. Beaman, A.B. Yishay, J. Magruder, and A.M. Mobarak, “Can Network Theory-Based Targeting Increase Technology Adoption,” American Economic Review 2021; 11: 1918-1943.
[52] M.A. Alexander, L. Forastiere, S. Gupta, and N.A. Christakis, “Algorithms for Seeding Social Networks Can Enhance the Adoption of a Public Health Intervention in Urban India,” PNAS: Proceedings of the National Academy of Sciences 2022; 119: e22120742119.
[53] J.H. Fowler and N.A Christakis, “Cooperative Behavior Cascades in Human Social Networks,” PNAS: Proceedings of the National Academy of Sciences 2010; 107: 5334-5338.
[54] R.M. Bond, C.J. Fariss, J.J. Jones, A.K.E. Settle, C. Marlow, and J.H. Fowler. “A 61-million-person experiment in social influence and political mobilization,” Nature 2012; 489: 295-298.
[55] A. Banerjee, A.G. Chandrasekhar, E. Duflo, and M.O Jackson, “The Diffusion of Microfinance,” Science 2013; 341: 1236498.
[56] D.G. Rand, S. Arbesman, and N.A. Christakis, “Dynamic Social Networks Promote Cooperation in Experiments with Humans,” PNAS: Proceedings of the National Academy of Sciences 2011; 108: 19193-19198.
[57] L. Muchnik, S. Aral, and S.J. Taylor, “Social Influence Bias: A Randomized Experiment,” Science 2013; 341: 647-51.
[58] E.L. Paluck, H. Shepherd, and P.M. Aronow, “Changing Climates of Conflict: A Social Network Experiment in 56 Schools,” PNAS: Proceedings of the National Academy of Sciences 2016; 113: 566-571.
[59] P. Wang, M.C. Gonzalez, C.A. Hidalgo, and A.L. Barabasi, “Understanding the Spreading Patterns of Mobile Phone Viruses,” Science, 2009; 324: 1071-1076.
[60] G. Lotan, E. Graeff, M. Ananny, D. Gaffney, and I. Pearce, “The Revolutions Were Tweeted: Information Flows During the 2011 Tunisian and Egyptian Revolutions,” International Journal of Communication 2011; 5: 31.
[61] S. Goel, A. Anderson, J. Hofman, and D.J. Watts, “The Structural Virality of Online Diffusion,” Management Science 2016; 62: 180-196.
[62] S.Vosoughi, D. Roy, and S. Aral, “The Spread of True and False News Online,” Science 2018; 359: 1146-1151.
[63] R.M. Bond, C.J. Fariss, J.J. Jones, A.K.E. Settle, C. Marlow, and J.H. Fowler. “A 61-million-person experiment in social influence and political mobilization,” Nature 2012; 489: 295-298.
[64] A. D. Kramer, J.E. Guillory, and J.T. Hancock, “Experimental Evidence of Massive-Scale Emotional Contagion Through Social Networks,” PNAS: Proceedings of the National Academy of Sciences 2014; 111: 8788-8790.
[65]A.V. Papachristos, A.A. Braga, and D.M. Hureau, “Social Networks and the Risk of Gunshot Injury,” Journal of Urban Health 2012; 89: 992-1003.
[66] A. Banerjee, A.G. Chandrasekhar, E. Duflo, and M.O Jackson, “The Diffusion of Microfinance,” Science 2013; 341: 1236498.
[67] E.L. Paluck, H. Shepherd, and P.M. Aronow, “Changing Climates of Conflict: A Social Network Experiment in 56 Schools,” PNAS: Proceedings of the National Academy of Sciences 2016; 113: 566-571.
[68] N. L. Keating, A.J. O’Malley, J.P. Onnela, S.W. Gray, and B.E. Landon, “Association of Physician Peer Influence with Subsequent Physician Adoption and Use of Bevacizumab,” JAMA Network Open 2020; 3: e1918586-e1918586.
[69] L. Beaman, A. BenYishay, J. Magruder, and A.M. Mobarak, “Can Network Theory-Based Targeting Increase Technology Adoption?,” American Economic Review 2021; 111: 1918-43.
[70] M. Moussaïd, H. Brighton, and W. Gaissmaier, “The Amplification of Risk in Experimental Diffusion Chains,” PNAS: Proceedings of the National Academy of Sciences 2015; 112: 5631-5636.
Nicholas:
Thanks so much for the careful, thoughtful, and informative comment. I posted it here along with an apology and reflections on my part, also I edited the title and some of the body of the above post.
All the papers that I looked for were available for me to read, yay!
The enduring “debate” here seems to be whether social contagion can be teased apart from background network autocorrelation. The question is interesting to me because it does not look intractable but may be out of reach due to the magnitude of work that would be needed to answer it.
I am imagining taking the obesity/contagion study data and applying the same methodology to all identifiable traits collected in the database, including ones like height and eye color that are not contagious, but also all the traits that could be contagious to some degree. Then take a list of the traits and ask a group of experts – say social scientists – unaffiliated with the work to rate each trait on its likelihood to be associated with contagion. Now calculate contagion effect size for each trait by the previously used method and see how well those numbers correlate with the expert ratings on contagion association. We should expect to see some clustering – network autocorrelation – for more or less everything due to latent homophily plus the fact that the real world is randomly clustery anyway. The effect sizes of the noncontagious traits effectively become the null. Ranked by effect size, would the top of the list be strongly skewed towards traits susceptible to contagion?
Matt: in case it helps in your efforts, we have published one prior discussion of the issue of contagion of mutable or immutable traits (as a falsification test), here: N.A. Christakis and J.H. Fowler, “Social Contagion Theory: Examining Dynamic Social Networks and Human Behavior,” Statistics in Medicine 32(4): 556-577 (February 2013)
In addition, this paper might be useful (and I am pasting the abstract below): A.J. O’Malley and N.A. Christakis, “Longitudinal Analysis of Large Social Networks: Estimating the Effect of Health Traits on Changes in Friendship Ties,” Statistics in Medicine 30(9): 950- 964 (April 2011)
We develop novel mixed effects models to examine the role of health traits on the status of peoples’ close friendship nominations in the Framingham Heart Study. The health traits considered are both mutable (body mass index (BMI), smoking, blood pressure, body proportion, muscularity, and depression) and, for comparison, basically immutable (height, birth order, personality type, only child, and handedness); and the traits have varying degrees of observability. We test the hypotheses that existing ties (i.e. close friendship nominations) are more likely to dissolve between people with dissimilar (mutable and observable) health traits whereas new ties are more likely to form between those with similar (mutable and observable) traits while controlling for persons’ age, gender, geographic separation, and education. The mixed effects models contain random effects for both the nominator (ego) and nominated (alter) persons in a tie to account for the fact that people were involved in multiple relationships and contributed observations at multiple exams. Results for BMI support the hypotheses that people of similar BMI are less likely to dissolve existing ties and more likely to form ties, while smoker to non-smoker ties were the least likely to dissolve and smoker to smoker ties were the most likely to form. We also validated previously known findings regarding homophily on age and gender, and found evidence that homophily also depends upon geographic separation.
For completeness, here also are citations to two prior published responses (from a decade ago) to some of the points raised by critics of the 2007 paper. To be clear, we agree with some of the points (e.g., the technical issue of how to account for tie dissolution) and disagree with others. I think the variety of models we brought to bear in the spate of initial papers of ours from 2007-2010, the methodological innovations (e.g., the directionality test for identification, the accounting for geographic location, etc.), the hard-to-obtain data, and the care with which we expressed our claims in the 2007 paper have stood the test of time. And our findings and methods have indeed been widely replicated (and not just cited).
J.H. Fowler and N.A. Christakis, “Estimating Peer Effects on Health in Social Networks,” Journal of Health Economics 27(5): 1386-1391 (September 2008).
N.A. Christakis and J.H. Fowler, “Social Contagion Theory: Examining Dynamic Social Networks and Human Behavior,” Statistics in Medicine 32(4): 556-577 (February 2013)
Furthermore, the work we did using observational methods has motivated a large literature in actual experiments regarding social contagion in complex networks, including many further contributions by my own group, starting early from 2010 onward. In these experiments, we can be very certain about the causal effects of one person on others to whom they are directly or even indirectly connected within the social fabric. There are more citations elsewhere in the exchanges we are having here, but here (again for completeness in the archive that your wonderful blog has become) is a representative selection:
J.H. Fowler, and N.A. Christakis, “Cooperative Behavior Cascades in Human Social Networks,” PNAS: Proceedings of the National Academy of Sciences 107(12): 5334-5338 (March 2010)
D.A. Kim, A.R. Hwong, D. Stafford, D.A. Hughes, A.J. O’Malley, J.H. Fowler, and N.A. Christakis, “Social Network Targeting to Maximise Population Behaviour Change: A Cluster Randomised Controlled Trial,” The Lancet 386: 145-153 (July 2015)
A. Nishi, H. Shirado, D. Rand, and N.A. Christakis, “Inequality and Visibility of Wealth in Experimental Social Networks,” Nature 526: 426-429 (October 2015)
M.A. Alexander, L. Forastiere, S. Gupta, and N.A. Christakis, “Algorithms for Seeding Social Networks Can Enhance the Adoption of a Public Health Intervention in Urban India,” PNAS: Proceedings of the National Academy of Sciences 119 (30): e22120742119 (July 2022)
E. Airoldi and N.A. Christakis, “Induction of Social Contagion for Diverse Outcomes in Structured Experiments in Isolated Honduras Villages” Science 384: eadi5147 (May 2024)
Again, while I disagree with some of your assessments, Andrew, I really appreciate the openness with which you engage in such topics. :-)