adjudicating between networks and taste

I just read Frédéric Godart and Ashley Mears’ article in Social Forces, “How Do Cultural Producers Make Creative Decisions?: Lessons from the Catwalk.” The article is based on both ethnographic work and an interesting network analysis of fashion design firms “linked” by contracts with the same models.

While designers use a language of authenticity in which they claim to use judgment and inherent taste to select models, the article reveals that there is an “option” system in which designers can place options on contracts with models. These options are free and do not bind designers, but they are communicated. This, the authors reason, is a signaling mechanism that allows designers to seek models that other designers like them are using, thereby adjusting their taste to the preferences of significant network alters.

The claim is plausible, and of course appealing to a sociologist. We’d always rather explain social patterning in terms of social structure (networks) than in terms of shared preferences (tastes), because the latter just feel more like a sociological explanandum than an explanation. But in this case I don’t see it as demonstrated. The null hypothesis, IMHO, ought to be that designers really do use their own taste, their immediate judgment, to select models. We could expect observed patterns that taste to be the result both of exogenous taste with selection bias (people with similar taste select into the market) and of learning (people learn what “good” taste is from years in the business). In other words: the fact that designers could use the options system as a signaling mechanism to adjust their selections of models according to the networks is not in itself evidence that they do use it that way.

I bring this up not to pick on the article, which was interesting, fun to read, and innovative, but more to ask about the general question. Network approaches, however appealing, often seem to suffer from selection effects problems. Maybe networks are the result of the selection of network positions because of properties of the individuals selecting in–again, not that I think that’s the case, but it’s a suitable null hypothesis for the general claim of network effects, that is, that social structure, represented as a social network, has effects beyond whatever can be explained by properties of the people in that network.

Author: andrewperrin

University of North Carolina, Chapel Hill

6 thoughts on “adjudicating between networks and taste”

  1. Andrew: This is a huge problem in network analysis, and has been picked up as an important topics by many researchers. Right now, there are persuasive models suggesting that it could be a serious problem. Also, some folks in economics, I believe, are running experiments to test such issues, while statisticians are doing simulations to see how biased model estimates can be. The dust has not settled on this issue.


  2. Tom Snijders and co. have developed a fairly sophisticated approach ( to parsing influence and selection effects. However, it requires some pretty extensive assumptions and at least three waves of social network data.

    Most papers using the method seem to find that both processes occur simultaneously, but usually selection effects are larger. However, these models always assume influence *only occurs between linked nodes, and anticipatory influence (whereby actors change their behavior to match those with whom they want to be linked) is chalked up as selection. So there is probably more influence going on than is reflected in these models.


  3. It seems to me that it is useful to keep in mind Tilly’s distinctions between methodological individualism, phenomenological individualism, holism and relational realism. Fashion itself would be holism, a macro structure that influences individual action, the tastes of the individual designers would be individual. It is difficult to see how individual action can influence the whole industry. Whose taste would it be? There is the flip side, which is: how can an individual act independently within the dictates of the industry?

    Social network are often applied as networks of individual as economic choices, but Tilly, White and others have emphasized the higher level abstractions such as roles, showing that roles are a consequence of social position and not individual characteristics. These relations become “real” to us, thus “relational realism.” In this view even “persons” are the consequences of relations and not individual discrete objects. The relational realist view is anti-essentialist.

    Perhaps one unfortunate wording in the article is “information sharing.” The article is in the relational realism camp and is about how the superlative “beauty” of the top fashion models is not an essential characteristic but socially constructed. Talking of “information sharing” often leads us to think of computers, which are methodologically individual (economic).

    Far too many articles on social networks these days do not go past methodological individualism, asking about the choices nodes make in their own best interest to create the networks we observer. Godart and Mears are asking a more interesting question: how can our relations create things that are so obviously attributes such a superlative beauty?

    The question that has been posed, about whether is is better to take the individualist position or the relational position is not going to be answered here, but both Tilly and White argue that “it depends.” It is disquieting to some to think that the same thing holds for other things besides fashion and often what seems to be exceptional merit is the product of social forces. However, even if this paper does not prove the relational explanation, it certainly puts the individualistic one into question and that is a valuable contribution.


    1. I do understand the claim that “even ‘persons’ are the consequences of relations and not individual discrete objects.” But this is a claim, not something demonstrated here and certainly not appropriate as an assumption!


  4. Thanks to Andrew Perrin and Scatterplot for commenting on our Social Forces article. We think there is an interesting discussion to be had here, so we would like to reply.

    Based on Perrin’s reading of our article, we would seem to think that the choice of fashion models by designers is based solely on an exchange of information that happens through an “options mechanism,” in which designers can discover other designers’ choice of models for the catwalk. From here Perrin thinks we’ve got selection bias, which forms the basis for his broader critique of social network analysis.

    But in fact, the options mechanism is only half the story in our paper. We begin by noticing a huge convergence in designers’ tastes for models, though in interviews they swear by their own gut instincts when it comes to choosing a model for their catwalk. It’s a great puzzle: if models are chosen according to personal taste, how does this collective convergence happen? Herbert Blumer, in a wonderful and rare article on fashion in 1969, made similar observations about fashion buyers in Paris: they rely on personal feelings when making aesthetic choices, but everyone happens to feel similarly about at the exact same moment. Blumer reasoned that if sociologists could unlock how this “collective taste” happens, we could figure out how cultural change works. Sociologists didn’t have powerful computer technology to work with in 1969; today, we think social network computing has great potential for cultural sociology.

    So let’s return to the catwalk. Through interviews and ethnography, we discover that producers’ tastes take shape through TWO mechanisms: 1) socialization in a professional and highly sociable environment which is localized in cities, and 2) this fascinating options mechanism that allows, but indeed does not necessarily mandate, fashion houses to coordinate their choices of models. Perrin writes that people with similar taste may select into the market; just as likely they learn what “good” taste is from years in the business. We’re on board with that, in fact we argue that this is the case in fashion. But this still can’t explain how this collective convergence happens, in which some fashion houses seem to vie for the same handful of models out of a huge supply pool. The ethnography takes us this far. We know that designers use gossip and options because we’ve observed it in practice, but up to now we don’t know the broader significance of these mechanisms. Based on the literature, we hypothesize that there is a status structure to it.

    We test this with a social network analysis of one season of data from, and find that high status houses converge around the in-demand group of models, while lower status houses tend to choose less popular models, a clear core-periphery structure. This tells us that a status structure is tied to designers’ seemingly personal taste and so-called gut instincts.

    But does this just reflect selection bias? There is no selection effect in our database from since it is exhaustive (at least for the given season). There is indeed, as Perrin points out, a selection of people with similar tastes into the fashion industry, starting with the taste for fashion itself. But this cannot account for subtler taste differences when it comes to choosing a fashion model. We explain that people do learn what “good” taste is through experience in the industry, namely through their social networks. But again, when it comes to choosing the “right” model for the catwalk, such choices happen collectively and are pattered along a status structure. Network analysis helps us see the big structural picture; ethnography helps us figure out how people work and express themselves.

    As for the more general discussion about the merits of social network analysis vs. individualism, we follow a long tradition of theory (from Simmel, Bourdieu, White, Wellman, Burt, Latour, and Stark) in believing that indeed tastes emerge and are shaped by social networks, and in turn shapes them.

    We are now working on a follow-up paper with an extended dataset (20 seasons form, with additional data on fashion houses, modeling agencies, and even models’ height and eye color!). We’re pleased our paper has attracted attention and we will make sure to integrate these comments in the next one.


    1. This is a great response, and I appreciate the care the authors put into it! I certainly wouldn’t claim that the observed effect is only selection bias, just that I also don’t think it’s only a network effect. Put another way, I suspect that networks such as this one work, in part, through discouraging people with different tastes from joining them in the first place.


Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.