Networks and Attributes

In class yesterday, Ajay argued that the distinction between attributes and relations — so often made in intro-to-networks contexts — is not so clear. An attribute is a node-level variable, so centrality and any other node-level network concept is just another attribute. This is surely true, but it also a matter of semantics. Dan B prefers to use “attribute” in a more restricted way. An attribute is a characteristic of an actor themselves and not of their location in their social environment, such as their gender or attitude toward texting. In contrast, a node-level network property is a characteristic of a node’s position in the network, and can change if a tie is added or lost anywhere in the network. So gender and attitude are attributes while betweenness centrality is a node-level property. Joe said this distinction is problematic because network processes can cause attributes. For example, a person’s attitude is an attribute but it may have been formed as a result of contact with others with that attitude. But in my view the fact that attributes and network properties can cause each other is not relevant to the conceptual distinction between them; heat may cause fire but they are not the same thing.

The distinction here reminds me of the distinction between social networks and networks in general, which also came up in class. In my view, it is useful to restrict the term social network to just those cases where the nodes are entities, i.e., have agency. I can define a network in which the nodes are words and the ties indicate whether a given pair of words co-occurs in the same sentence in a given text. But I’d rather not call that a social network, and I certainly don’t want to apply social theory to it.


14 Responses to “Networks and Attributes”

  1. ajay mehra Says:

    Steve is correct that there is an arbitrariness to how the term “attribute” is/can be applied. He is also correct that Dan Brass was using it in class yesterday in the way it has historically been used in organizational studies. In that usage, there are two aspects to the notion of an attribute: the first has to do with whether the attribute is internal/external; the second with its relative fixity. Traditionally, the term has been used to refer to internal characteristics of people that are moreover relatively invariant/fixed. But there are complications here that have to do with more than mere semantics. Consider, for example, the personality trait of extraversion. The theory behind this attribute is that some people (extraverts) tend to be more sociable and gregarious– i.e., they tend to have/prefer more social connections (relative to introverts). This attribute, then, is a summary assessment of a person’s social inclinations averaged across a range of social settings. Furthermore, certain interactionists views in psychology suggest that the attribute of extraversion is not fixed but can vary (in a patterned, meaningful way) across social settings. So, extraversion, as an attribute, seems to have both an external/social aspect to it and a dynamic/changing aspect to it. Does that mean extraversion is not an attribute? If an attribute is conceptualized in such a way as to refer to some aspect of the node itself (and not having to do with their location in social networks), then what to do about “attributes” that reflect a tendency to be in certain social network positions? (Btw, even seemingly internal/fixed attributes, such as gender or race, tend to have a social aspect to them– these attributes are more social (external/dynamic) than some assume.)

    All this said, I don’t disagree with the gist of the distinction that Dan/Steve are drawing. I just think (a) the distinction is harder to draw than is being suggested; and (b) it is not merely a matter of semantics but a more fundamental challenge for the ongoing development of network theory and research.

  2. Steve Borgatti Says:

    I still think it is a semantic issue.

    “Attribute” has different, well, attributes, depending on the implicit contrast being invoked. It’s like “man”, which in one context means human and contrasts with animal. Once we set the context to human, man contrasts with woman. Once the context is male human, then man contrasts with boy. Once the context is adult male human, man contrasts with wimp, and so on.

    So in some contexts, “attribute” does carry the connotations of fixity and in others internality. But violations of fixity and internality are not relevant when the contrast is with network properties. In that context, it seems clear to me that “current emotion” is an actor attribute even though it is highly changeable and “social class” is an actor attribute even though class is highly socially constructed.

  3. ajay mehra Says:

    You are no doubt correct that the meaning of the word “attribute” is, like the meaning of all words, context dependent. But if actor attributes can be external/social and mutable and network characteristics (e.g., actor centrality) can be external/social and mutable, then just how the word attribute can be used to meaningfully distinguish one from the other becomes difficult to see.

    I guess what troubles me here is that when positioning individual attributes as a contrast to network characteristics the implicit implication is that network characteristics cannot be stable/internal and relatively fixed. But contemporary network research seems to suggest otherwise: consider Burt’s recent paper on the agency question in which he uses data from seven thousand people in a virtual world (each of whom played two or more roles) to show that people tended to have a “recurring network style” across the roles they played: people who built sparse networks in one role, did so when playing other roles, too (

    So, I get the gist of the (context-dependent) distinction that Dan/you are drawing. My point is that the distinction is not as sharp as it might appear at first blush; and that it might invite relative newcomers to treat as an absolute a distinction that is probably more a matter of degree.

  4. Joe Labianca Says:

    Interesting discussion on the attribute/relation distinction. But I guess I was a bit more disturbed by something else in class. Dan B likes to trot out the following graph and asking “who’s the most powerful person in this network?”

    A B
    \ /
    / \
    D E

    Two answers often emerge: 1. C because of the control C has over the network flows; and 2. it depends on what is flowing over the network.

    I’m a bit tired of Answer 1. It’s a purely structural response, taking no account of what is happening to C. What if C is a really stupid, rich guy walking through a bad neighborhood at 3am and is being mugged by A, B, D and E, all of whom are previously unacquainted, but all of whom see a great opportunity? Is C really powerful (that is, getting the others to do something they might not otherwise do)? Only if you believe that C getting his face in the way of the others’ fists makes for being in a powerful position. That doesn’t pass the grandmother test (my grandmother would tell me that I was reading too much and needed to go play more outside if I made that argument to her). Clearly a purely structural explanation with no accounting for the types of flows in the network is less than helpful.

    But I also see Answer 2, which is relational, as being inadequate. Yes, if toxic flows are coursing through the network, I wouldn’t want to be in C’s position. Big deal.

    The answers I’m drawn to combine the structural response with a relational response because they match the empirical reality of the world in which I live. I hope that as networks moves into marketing more, they don’t go overboard on the structural approach as much as I believe we have in management.

    • ajay mehra Says:

      For anyone in class not (yet) familiar with the structural/relational distinction (what a lot of distinctions we make): let me put in a plug for our very own borgatti and halgin’s recent essay (“On Network Theory,” in the journal Organization Science, 2011– preprint available here: Their paper takes up some of Joe’s concerns in their discussion of the “bond model” and the “flow model.” Enjoy!

    • Steve Borgatti Says:

      I think this particular version of structural versus relational is not a good one. Structural doesn’t mean ignore flow. In fact, when we take a flow perspective, we become very interested in structure because certain positions in the structure will receive things earlier, more certainly, less distortedly, etc., and at the network level certain structure transmit better than others. I would even go so far as to say that most of the mathematical machinery of network analysis (e.g., betweenness centrality) is based on trying to estimate expected flows given a certain structure.

  5. Dan Brass Says:

    As both Steve and Ajay noted, I was using “attribute” in the traditional OB sense to indicate personal characteristics of individuals (groups, organizations) and in the same sense as human capital, in contrast to social capital, as per Coleman. Attributes, like human capital, are owned by the individual, as opposed to network characteristics (such as centrality) which are jointly owned. Under this loose definition, relational demography would fall into the jointly owned catagory. The difficulty in using “attributes” to contrast network research with traditional management research is that “attributes” is included as a heading in UCINet and both individual attributes and network attributes such as point centralities can be entered for statistical purposes. I have no problem with this, but my use of the term “attributes” can create confusion in class. As a remedy, I can easily refer to individual attributes and network attributes, as I can’t think of another word to substitute for attributes that would not have the same definitional problems (e.g., characteristics).

    While some individual attributes such as personality and gender may be fixed, I don’t think that fixity is a key definitional necessity as attitudes change as does experience, skills, and behaviors such as style (traditionally thought of as human capital). I also think that the cause of the individual or network attribute should not be confused with the definition. Human capital creates social capital, just as Coleman famously noted that social capital creates human capital. Nor do we usually include cause in everyday definitions of words (with some exceptions). Regarding Burt’s recent “style” article, aggregating network positions across lots of networks and attributing a style to an actor (whether we think this is a legitimate way to measure style or not) suggests that “style” is human capital and could be considered an individual attribute, whereas one’s structural holes or centrality in a network is jointly owned (A has a hole only to the extent that B & C are not connected – so the hole is jointly owned by A, B, & C) and would be defined as social capital or a network attribute.

  6. Dan Halgin Says:

    I too have been thinking about our discussion of power in networks. Specifically, how does tie content influence C’s power/influence/clout in the star network? In some of our papers, and in our discussions, we tend to draw on Atkin’s backcloth/traffic distinction to introduce a typology of ties: Similarities->Social Relations-> Interactions-> Flows (I believe that this typology first appeared in the Borgatti et al 2009 Science piece). Thinking of the star network, does it matter what type of tie is the basis of the network and/or what is flowing through the network when thinking of C’s clout? Clearly if we think of positive flows, C is the most powerful. Even if we think of negative flows, C is the most “influential” and powerful (in one sense of the word) in that C has a direct effect on all other nodes in the network. It fails Joe’s Grandma’s test, but C is still the center of the action.

    But what if we step back and think in terms of social relations or similarities. Is C still the most powerful if the tie is “knows” (a social relation) or “attends the same event” (a similarity)? Without thinking in terms of what potentially flows through these ties, I’m not sure that we can say this. In fact, from a “ties as prisms” perspective node C might be considered the weakest due to ambiguous identity.

  7. ajay mehra Says:

    I’m fully aware that my adding yet another comment is going to confirm my status as one of those academics with nothing better to do than make a big deal of small distinctions. But let me boldly (blindly?) forge ahead:

    It seems to me that one of the issues that underlies much of the discussion thusfar has to do with tradeoffs that are fundamental to all theorizing. These basic tradeoffs have been discussed as “Thorngate’s impostulate” (see, e.g., the discussion by Weick of this impostulate: The basic gist of this impostulate is that no theory can be general, accurate, and simple at the same time (and still remain coherent– try it out on the theories you know and see if you buy the argument).

    The structural wing of network theory tends to maximize generality and simplicity but tradeoffs accuracy to some extent to pull this off. To enhance accuracy (by, for example, building in kinds of flows/relations instead of merely inferring them; or by building in “individual-attributes”) a theory must necessarily tradeoff some degree of generality. Just which tradeoffs one prefers that a theory make may be, to no small degree, a matter of taste rather than logical necessity.

  8. Dan Brass Says:

    Just to clarify, I use the ABCDE simple “star” network in presentations/class to illustrate the structural approach to social network analysis. Following a purely structural approach, a node is a node is a node and a tie is a tie is a tie. All that matters is the structural arrangement of nodes and ties. And the “star” network illustrates a strong structural constraint when compared to an “all-channel” clique where each node is connected to every other node (in which case it would be impossible to hypothesize which node is most powerful without some additional information). So, the centralized “star” serves it’s purpose. As we will discuss in class, the purely structural approach is probably the most often criticized approach as it does not take into account the content of the ties, the strength of the ties, the resources (human capital) of ego or the alters, the perceptions of the network, agency, the larger context/culture, etc. The example is simply meant to illustrate the pure structural approach; we will systematically add these other approaches as we progress through the readings. Despite the criticisims of the structural approach, it is interesting to note the two major approaches to social capital (Burt’s structural holes and Coleman’s closure) are primarily structural approaches. Aside from Granovetter’s “strength of weak ties,” the two structural approachs have dominated the research over the past ten years.

    As for Grandma’s test (God love her), note that the four alters are not connected (so no gang muggings on the central node). Note that the central node could possible allign with one of the alters (because there is a connection)to attack one of the other alters. In building any kind of gang/coalition, the central node is the most powerful. But, as I mentioned, it is only an illustration to convey the notion of the structural arrangement of nodes and ties.


  9. Joe Labianca Says:

    Steve — Dan asked who’s the most powerful. I don’t see how your comment relates.

    Dan B. — Note that I never said a gang mugging. In fact, I pointed out that the muggers don’t know each other, anticipating you’d say this. So, the central node could be a shop owner in Manchester, set upon by 4 individuals unknown to each other wanting to loot his store during a riot. The shop owner is in a weak power position. The only choices I see are to 1. get a gun (increase attribute-based power) or 2. get an ally to help fight back.

  10. ajay mehra Says:

    Here are links to two articles– about the effects of non-response on network analysis– that may be of interest given recent in-class discussion about the collection of network data:

    • ajay mehra Says:

      i meant to add that it was valdis krebs (who is a well-known SNA consultant) drew attention to these two articles during a recent exchange on “Socnet,” which is a listserv dedicated to discussion of social network analysis. for more details, check out the INSNA website (look under “publications and forums” for info. on the listserv):

  11. Tommy Flynn | Emory Says:

    It’s an interesting question, “At what point does society end and the individual begin?” Maybe there isn’t a clear barrier there, or maybe just a semi-permeable membrane. But, even in studying complex networks, we have to start by breaking them down to comprehensible parts, don’t we? We can’t look at every variable at once, at least not yet. Either way you use it, the term “attribute” doesn’t hold a lot of meaning until it’s clearly conceptualized. Which attributes matter, why, and how? Who knows, a specific centrality may end up being the holy grail of human attributes that,-once achieved-cures cancer and defies physics. I’ve started thinking of all attributes, network and otherwise (betweenness, age, gender, disease, outcome), as different variables in the same equation. That may be why my head is spinning. I would love to join your class, I could learn a lot from you folks. Cheers.

    Joe: The moment your looters decided to behave in the same way, at the same time, with the same purpose, the network structure changed. Their union (albeit unintentional) is the reason C lost her upper hand.

Leave a Reply

Fill in your details below or click an icon to log in: 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

%d bloggers like this: