the structure of mixing

“Do people mix at mixers?” This is the question that Paul Ingram and Michael Morris address in a paper with one of the more ingenious study designs I’ve seen. The paper, “Do People Mix at Mixers? Structure, Homophily, and the ‘Life of the Party,’” appears in a recent issue of the Administrative Science Quarterly. Using electronic name tags that allow the researchers to track the movements and interactions of attendees of a business school mixer, Ingram and Morris are able to figure out which attendees shared face time and how long their interactions lasted.

The results indicate that people are more likely to interact at parties with people they already know well, and with people they are indirectly connected through third party ties. Thus, if you and I share a common friend but do not share a direct pre-mixer tie, we’re much more likely to interact at a party than two completely disconnected people would. The results then provide evidence that structural conditions shape patterns of interaction in a conversational setting. Surprisingly, the results do not indicate that people are more likely to interact with similar others. One might expect the homophily principle to be in play at mixers – that you will seek out others who share similar values, characteristics, etc. But this doesn’t appear to be the case. The latter non-finding is especially surprising because of the vast amount of evidence indicating that friendships and other type of associations tend to form among people with similar demographic characteristics. Ingram’s and Morris’s findings suggest that homophilous attraction, at least in a conversational setting, may be based more in structural conditions than in preferences for likemindedness.

Thus our result supports McPherson, Smith-Lovin, and Cook’s (2001) claim for the primacy of structure as a cause of homophily and derives from just the sort of dynamic analysis they call for as necessary to separate confounded accounts of the origins of network ties.

Score a point for the structuralists in the room.

Written by brayden king

May 9, 2008 at 3:19 pm

Posted in brayden, networks, sociology

8 Responses

Subscribe to comments with RSS.

  1. ” Ingram’s and Morris’s findings suggest that homophilous attraction, at least in a conversational setting, may be based more in structural conditions than in preferences for likemindedness.”

    Could this be an artifact of the research design? For example, I might expect high school students to match via race, but the average (Columbia b-school?) mixer has people who are more willing to go cross-race/gender. Maybe there aren’t quite enough blacks to get many black-white contacts? An older observation on friendships is that if you have a low % of type X, it’s easy to split them up so they have to go cross category. Ie, in a schooll of 100 students with only 10 blacks, it’s easy to split them up when compared to the same school with 30 blacks, where you can form cliques.


    Fabio Rojas

    May 9, 2008 at 3:52 pm

  2. I meant “black-black contacts,” not “black-white” contacts.


    Fabio Rojas

    May 9, 2008 at 3:52 pm

  3. Sorry Brayden, but I fail to see exactly how this article scores any points for network structuralist theory. First, the main prediction of the theory, which would be that the pre-mixer network structure (at a minimum triadic embededness effects) strongly determines the structure of encounters during the mixer is not borne out: “The number of friends in common in the pre-mixer network does not affect the likelihood of encounter” (572). Second, the structural context in which the dyad is embedded does not affect engagement, but individual characteristics and pre-mixer liking (a subjective not “structural” feature, at lest in Mayhew’s and Wellman’s accounting) does: “Apparently once two people meet, it is their characteristics and pre-mixer relationship, not the trajectory of their recent experience at the mixer, that predict whether their conversation persists, so hypothesis 1c is not supported for engagement. People who had strong pre-mixer liking relationships conversed for longer when they engaged each other at the mixer as predicted in hypothesis 1a, but pre-mixer mutual friends did not influence engagement, counter to hypothesis 1b” (575).

    Thus it seems to me that Ingram and Morris actually misstate their main finding in the discussion section when they say that “preexisting network structure operated at the mixer much as it does in more mature relations: encounters and engagement were much more likely with pre-mixer friends than with strangers.” But it is not the pre-existing “network structure” that affected encounters, but dyadic closeness. Most network theorists–i.e. Wellman 1988–would not call this a structural effect (in fact this is closer to Homan’s dynamic theory of liking leading to interaction leading to more liking, leading to more interaction, etc. see Simon 1952).

    In fact most structural effects–once again the effects of the structural context in which the dyad is embedded net of subjective “tie strength”–were null. So score nothing for the structuralists (the only structural effects that they find have to do with the “emergent”, ephemeral structure of the mixer itself, which is something that a more situational structuralism of the Harrison White, David Gibson type might be happy with, but I doubt that this is enough for “hard core” structuralists of the Blau lineage). The subsidiary conclusion that “mixers don’t work” because people don’t mix at them, seems to me to be both unsurprising (all of the sociological theories “structuralist” or otherwise point toward the no-mixing hypothesis) and of little theoretical interest (although it might be of practical interest for the companies that are wasting their money throwing these parties).

    Finally, I find it puzzling that Ingram and Morris ended up interpreting their null finding of homophily in this particular social situation as implying support for the idea that “structural homophily” trumps “choice homophily” in all types of encounters and situations. First, it is clear that the correlation between the socio-demographic characteristics of dyads that we find in surveys is spurious. This in itself is not surprising. The million dollar question is what is the relevant “Z” variable that accounts for the association between ego and alter’s sociodemographic characteristics. The “structural” hypothesis following Feld’s (1982) influential argument is that it is “socially structured interaction foci” that produce this association. As most of the lucky few that have sat through Miller’s class can clearly recognize, this explanation is “Blausian” and “Mayhewian” in that the preferences of individuals are thought of as secondary (they could be considered either weak or random) in producing the empirical effect.

    However, what most “structuralists” of this type (notice that the “structural” element here is not even network structure anymore but “foci”), don’t acknowledge, is that there are other “Zs” that could be producing both the choice of which foci to belong to and which friends to select. One of these things, as noted by both Carley (1991) and Mark (1998) is shared cultural knowledge (my favorite empirical example is Long 2003 on book clubs). In other words, Ingram and Morris are presuming that for all of the high liking dyads in their data, this “liking” is the spurious by-product of some “foci-effect” in the past to which individual selectivity (and cultural compatibility) among the dyads either contributed nothing or happened after these people were “thrown together” in the same social focus. I think that their findings here don’t warrant that conclusion (people simply just don’t get thrown together into social situations), and that given their null results they could have gone either way (the structuralist or the more culture-friendly route) but because of the general location of Columbia in the Northeast–conspicuous in the people who show up in the title footnote–they were swayed toward structuralism (a focus effect!). But the main point is that empirically this question is as wide open right now as it was before I had my coffee this morning.



    May 9, 2008 at 4:59 pm

  4. the only structural effects that they find have to do with the “emergent”, ephemeral structure of the mixer itself, which is something that a more situational structuralism of the Harrison White, David Gibson type might be happy with

    Given their location in Blau space, this makes sense doesn’t it?



    May 9, 2008 at 7:24 pm

  5. Omar, I think there’s a greater theoretical contribution here than you estimate. A lot of the current research on longitudinal networks is about emergent processes, but many of these papers involve simulations since it is so difficult to get real data on network emergence. Thus, if you think about a mixer as a setting where network emergence takes place, the setting is both theoretically and practically interesting. In fact, that you don’t find a lot of pre-mixer network effects only emphasizes how important emergent characteristics of networks are in shaping future patterns of interaction.

    BTW, Omar pointed out that I didn’t clearly report one of their findings. Pre-mixer friend networks did not have a statistically significant effect in the sample of all dyads. However, pre-mixer contacts did have an effect if you isolate only those dyads that did not have any kind of direct pre-mixer relationship. I believe that effect is substantively significant. Among people who have had no prior contact, structural connectedness increases the likelihood of interaction, thus closing the triad. That’s a fairly “real” structural effect isn’t it?

    Thanks for the correction.



    May 9, 2008 at 7:33 pm

  6. I didn’t mean to give the impression that I didn’t find anything of theoretical interest in this paper. In fact there’s a lot of theoretical interest here, I just thought that the finding that they highlight the most (the “mixers don’t mix” one) is actually the least interesting.

    The fact that they replicate a closure effect is very neat, and the fact that the situational structure of encounters matters more than the pre-existing network is even neater (Goffman would be happy). However, even though they do find a closure effect among previously unacquainted dyads, it is important to point out that the sharing pre-mixer friends did not lead to increased engagement. So there is an effect of shared friends on brief “meetings” but it does not lead to sustained interaction, which means that the process that would lead to mature friendships cannot simply be thought of as a produced by way of a simple structural balance mechanism among actors with either neutral or random preferences.

    So the question still remains: what explains the existence of those pre-mixer strong liking ties? My sense is that one of the things that this paper actually shows is that (because of the high inertia of previous subjective dyadic closeness) the usual structuralist mechanisms (including Feld’s “interaction foci” one!) don’t seem to do a good job of explaining how do those ties come about. Or that a pure structural theory that at the same time does not tell us how exactly is it that people find each other “likable”–and thus willing to engage in sustained interactions–and not superficial encounters produced by situational exigencies–won’t do the job.



    May 9, 2008 at 11:33 pm

  7. Man, Omar beat me to all the good critiques… :-)

    As I was reading this, I just kept thinking of a scatterplot showing a very strong association between cultural similarities and the probability of associating and thinking of this paper as “zooming in” to one little section of it. If you zoom in too close to any section of a scatterplot, things start to look pretty random. Not sure if this makes sense in words, but it’s making a great picture in my mind!

    Anyway, like Omar said, the question is: what explains the existence of pre-mixer ties? My guess is that it’s not (in any interesting way, at least) the effect of any previous mixers, except to the extent that self-selection into the mixer is at work.


    Steve Vaisey

    May 10, 2008 at 12:21 pm

  8. These analyses don’t actually test whether common pre-mixer friends affect encounters or engagement because they include current and cumulative shared ties in the models. Current and cumulative shared ties likely mediate the effect of pre-existing ties. (That is, A and C don’t know each other, but they both know B. At the mixer, each sees B independently, then at some point after B realizes both are there, B introduces them. It is more likley that B will introduce two friends he/she has just interacted with than two friends that have not been seen.) The effect of shared pre-existing ties absent the current and cumulative ties effects in the model is where pre-existing ‘structure’ would be evident.

    In addition, we shouldn’t make too much of these findings. As with most network studies, this is a case study of a single network. This is especially notable in this instance, given the emergenet nature of interactions and encounters in such a compressed timeframe (how much could one expect structure to influence interactions in a maximum of 80 minutes). It’s impossible to determine from one case whether the absence of homophily effects is a general phenomenon of mixers, a product of the relative homogeneity of attendees that Steve noted, or simple Type II error.


    David Schaefer

    May 12, 2008 at 7:42 pm

Comments are closed.

%d bloggers like this: