Testing the limits of knowledge

I was so inspired and encouraged by the responses to my last question, I’d like to give the format another try.  This time, I have certainty about the literatures within which the question belongs, but a host of questions about measurement.  Network Org Theorists: This one’s for you.

Basically, wunderkind Mark Pachucki and I have network data that links samples (pre-recorded pieces of music, adapted into new songs) to the rap singles that employ them.  This figure here (see mess, below) links performer-to-performer (“target” rap artists on the left; “source” multi-genre artists on the right), but we can break it up all manner of ways.  Seriously…name a variable, and we have it.  Record label-to-label?  Yup.  No problem.  Want that in terms of month and year of release onto the charts?  Done and done.


We know the data evince the same power law distribution of rewards found in science and the arts. In science, the large inequalities in the distribution of rewards (attributed to the Matthew Effect) have been linked to practices of preferential attachment, where high-status scientists are more likely to be cited, sought for collaborative endeavors, published, or funded. Thus, the process is self-reinforcing (Moody 2004: 216).  Scholars have found evidence of preferential attachment in project-based creative labor markets as well; Faulkner and Anderson (1987) find high densities of market transactions between actors with similar career trajectories, and severe attrition for low-status, while qualified, personnel. In research largely situated afield of cultural and artistic spheres, it has been shown that in economic markets generally, producers look to other producers for signals as to how to behave, and uncertainty as to status and product quality influence reward distribution (cf. White 1981, Zuckerman 1999 [Whoot!], Gould 2002, Podolny 2005). Generalized to cultural markets, this implies that some cultural actors in a network of relations will be more susceptible to preferential attachment than others by virtue of perceived status.

Social network “stars”—those with a disproportionate number of colleagues and/or resources—act as “‘pumps’ for ideas that are then quickly circulated” (Moody 2004: 236) among their peers (see also Fredkin 1988; Collins 1998; Owen-Smith 2001).  This “network influence model” suggests that individuals exchange ideas, questions, methods, and techniques, and in “structurally cohesive networks [this] should generate consensus, at least with respect to problems and methods” (Moody 2004: 214).  Thus, stars are likely to be influential sponsors of new ideas and practices that are later globally legitimated (Crane 1972; Newman 2001).  In networks of artists, we find evidence that network stars act as “pumps,” spreading and legitimating new global ways of acting, thinking, and producing (White and White 1965; Anheier et. al 1995; Becker 1982; Faulkner 1983; Fine 1992; Giuffre 1999; Kadushin 1976; White 1993).

We have some evidence of stars in our dataset.  We’re using a measure of “betweenness centrality” in the current analysis.  Just as it sounds, this is a measure of the degree to which any given actor (e.g., an artist) lies “between” two others, acting as a bridge.  If a given (hypothetical) rapper Beta uses the same sample as rapper Alpha, and Beta shares a different sample with rapper Gamma (and there are no shared samples between Alpha and Gamma), then Beta lies “between” the two other rappers.  The rappers with high betweenness scores thus share samples with many alters, who do not otherwise share samples with one another.  Here are the results for the seven highest betweenness scores for rappers:


We can make some simple conclusions here: Public Enemy is a more commonly used bridge than De La Soul.  EPMD more than Tribe Called Quest.  And we can do the same with the artists being sample—which we call “source artists:”

And we can make some simple conclusions about these folks, too.  For example: James Brown is more commonly sampled in combination with samples of other artists than Sly and the Family Stone.

However, and here is where it gets interesting: why are these folks high, and other rap artists and source artists are lower?  I have already noted (Lena 2004) some substantive reasons why rappers sample—the music is “better” or more fitting with the sub-genre style of the rap group; the sample is easier to dredge out of the song’s mix; the source artist or song is associated with some significant historical, political, or economic event; and so forth.  But more interesting now are the formal reasons for selection, and these we know very little about.

For example, we know from other studies that some innovations* have reputational value only or mainly for the first to introduce it to the network.  Once introduced, the innovation is either not diffused, or diffused widely, without consequence for social status.  I like to call these “quick burns.”

Also, we know to expect hoarding.  Hoarding can be positive, as when cliques (here, sub-genres or record label groups) circulate the innovation among close peers, and develop techniques to prevent further spread.  Hoarding can also be negative, as when a clique refuses to adopt an innovation used by neighbors.

Finally, there would be some individual-level echoes of these trends, as when an individual refuses to adopt an innovation, or refuses to share one.  These decisions may be correlated with success or failure of that artist.

We need to figure out how to distinguish these causes, and our large data set suggests we’re going to have to make some assumptions (as compared to launching a huge, matching, qualitative study).  We don’t even think we’re going to be able to stick with the betweenness centrality measure to study “stars” since rappers with the highest bridging scores are more likely to be engaging in more overall sampling.**

The question for you is really: How far can we push these data in order to explain “quick burns”, “hoarding” (both positive and negative), and refusal?  Mark suggests that documentation of failed diffusion of innovations is still a contribution (he cites Kaufman & Patterson’s article on Cricket as an important exception to a lacuna).  If we push further, he suggests we might look at an array of hypothesized predictors (e.g., sales, age of record label, # of label mates in the Top 100 chart) in the weeks following the adoption of a sample by an artist.  At the very least, ruling out these explanations for diffusion of a sample (or lack thereof) may be of value.  So, what say you, jury?

* I’d like to please set aside for now the question of when innovations are innovative.  That is, for now I’d like you to assume that innovations are just new things, not new & valuable/important/resonant things.  We’re just using, as Mark says, the “vernacular of diffusion studies.”

**On this issue: we did check betweenness results against indegree (the indegree of a rapper is the number of adjacent rappers, by virtue of shared samples) and the correlation is pretty high (.7616 with a Cramer’s V and Rajski scores suggest that betweenness captures indegree (.7629) better than indegree captures betweenness (.5858)).  Thus, we can preliminarily conclude that popularity is a function of sampling, not of other reasonable measures like sales success or duration on the chart.

Written by Jenn Lena

November 11, 2008 at 1:20 am

Posted in uncategorized

9 Responses

Subscribe to comments with RSS.

  1. This makes up for that “rap” video from/about the LHC, several times over.



    November 11, 2008 at 1:56 am

  2. Have I got a model for you! It razzles, it dazzles, and it can efficiently summarize correlates of diffusion parameters in massive diffusion datasets!
    What you do is you feed data on several dozen innovations (including failed innovations) into the hopper and tell it some substantive variables describing each of these innovations and period effects that may cross-cut them. You can even attach the traits of the initial adopter as a trait to the innovation. You start the model running — and presto bango! It tells you how each of these substantive variables is related to the asymptote of adoption, contagious diffusion, and exogenous diffusion!
    Not only that, but it’s my understanding that you and Mark have an in with the inventor of this fantabulous thingamajig.
    Rossman, Gabriel, Ming Ming Chiu, and Joeri M. Mol. 2008. “Modeling Diffusion of Multiple Innovations via Multilevel Diffusion Curves: Payola in Pop Music Radio.” Sociological Methodology 38:201-230.
    Currently you have to run the results in one program then filter the results through a spreadsheet (we provide a template) however lately I’ve been procrastinating my other work by writing a Stata ado file that will provide intelligible results directly in Stata (think of it as an xt syntax for a Bass diffusion model). Depending on how much I hate my other work this could be done in just a few weeks.



    November 11, 2008 at 5:49 am

  3. I must admit, I’m not well read at all in the social networks literature, but I’ve got some ideas that may be helpful in terms of explaining “quick burns” and “hoarding.”

    First, all but one of the artists you list as having high betweenness scores are from NYC. The only one who isn’t is Redman, and he’s from Newark, NJ which is right next door. In my opinion, this is highly relevant because hip-hop was born in NYC (Bronx). As such, it is the key legitimator of all things hip-hop. If a NYC artist held in high regard samples another artist, then others may be compelled to follow suit and sample the same source.

    Second, Public Enemy, EPMD, Redman, and LL Cool J are (or have been) on the Def Jam label together. They undoubtedly have shared studio time together, and I know that some of them have collaborated to release some classic material. Further, A Tribe Called Quest and De La Soul were both part of the Native Tongues collective, so they definitely recorded together (even though they were signed to different labels). In short, it might be interesting to look at the bridges between members of the same label, same city, and across regions.

    Third, it might be kind of cool to look at the production credits on the songs used in the analysis. I suggest this because there are definitely cases of producers working with multiple rappers found in the diagram. For instance, Erick Sermon produced records for EPMD, Redman, LL Cool J, and numerous other artists. Also, the Bomb Squad production team who produced most of Public Enemy’s early work, also produced on Ice Cube’s “Amerikkka’s Most Wanted,” “Death Certificate,” and “The Predator.” In this case, the record producers seem to act as a bridge between NYC and LA rappers, as they have a hand in the work of each. Therefore, it might be interesting to compare quick burn/hoarding among producers, and then compare it with rappers.

    I don’t know if that was helpful or not, but it sure was fun for me to think through it.l


    Mark Bissonnette

    November 11, 2008 at 7:23 am

  4. Excellent points by Mr. Bissonnette. However, his one misstatement brings up a point about the data set. The Bomb Squad (noted for their heavy use of samples) did not produce Ice Cube after “Amerikkka’s Most Wanted.” The reason? Copyright enforcement after the Biz Markie/Gilbert O’Sullivan case. This is why the org chart is dominated by artists ca. 1987-89. — after that point, it became much less feasible (and profitable) to engage in heavy use of samples, and eventually the practice fell somewhat out of vogue (in come the Neptunes.)

    It would be interesting to use additional stylistic data such as words/minute, rhymes/minute, use of key words or phrases (or general word frequency.)



    November 11, 2008 at 7:04 pm

  5. Gabriel: Mark and I will be in touch, you rockstar, you!

    Mark Bissonnette: Certainly, New York’s centrality in rap “history” has an influence on artistic practices. The counterveiling force, of course, is the desire of artists from other places to carve out trademark sub-genre practices in order to make their music sonically distinct and exciting. I would immodestly refer you to both my 2004 Poetics piece and 2006 Social Forces piece, both of which speak (in part) to this issue.

    Label, geographic region, and city data are include among our variables so we’ll look into each of those, for sure!

    Tim is correct to point out the importance of copyright law on sampling patterns. There are some other field-level changes that impact the data, including a major adjustment in how SoundScan gathered point-of-sale and performance data, after 1992.

    I do not have production credit data–is there a reliable, digital source?


    Jenn Lena

    November 11, 2008 at 7:15 pm

  6. Prof. Lena: I did some searching and came across a message board for music engineers/producers. There was one post in which everyone was lamenting the fact that there is not an “imdb-type” database for music producers. One website was recommended as “the best available,” and I’ve posted the URL below:

    Unfortunately, this is a community-built database so the reliability of the information will most likely be an issue. I will say, though, that I was impressed with the way this site accounts for numerous misspellings of producer names. It does a better job of accounting for typographical errors…

    Prof. Tim: Good point about the copyright laws. I was contemplating it earlier today, and had an idea–when artists can’t get sample clearances they can still put their sample-heavy music on the market by “leaking” songs to mixtape DJs. If I’m right about that (and I may be), is sampling in hip-hop music really on the decline, or is it just less prevalent in official releases? Mixtape data would be needed to answer this question decisively, however I doubt that it exists.


    Mark Bissonnette

    November 12, 2008 at 6:50 am

  7. By the way, my second-to-last sentence should read, “If I’m right about that (and I may or may not be)…” Just a typo that I wanted to address immediately because it comes off pretentious as it is currently written.


    Mark Bissonnette

    November 12, 2008 at 7:07 am

  8. To echo Jenn’s thanks, these are useful suggestions, Gabriel, Mark B., and Tim! For me, a more general attraction of this project has been to use the case of rap artists to think about how networks and culture evolve and mediate one another. In addition to Jenn’s status questions, there are additional dimensions we might productively focus on:

    a) “within-label/geography micro-cultures”, as Mark B. points out;
    b) “within-artist-sample-cultures” over time;
    c) interaction between artist preferences (signified by use of samples) and audience preferences (as signified by Billboard ranking);
    d) genre development as cultural change enabled and constrained by artist network evolution

    Each of these framings have different implications for understanding the relationships between actors and the cultural content they co-construct over time… I wonder, do readers have thoughts about one path yielding greater returns to knowledge in the field?

    (and it does bear remembering that from one perspective, our “network” as conceived right now is more of an imagined community of artists connected by virtue of their commonly-referenced scraps of musical culture. This, as opposed to us having concrete knowledge that certain artists have some kind of well-defined social relationship (i.e. friendship via name-generators), or one more loosely defined, i.e. collaborator, labelmate, guest on each others’ albums, etc. via liner notes…)


    Mark P.

    November 13, 2008 at 8:05 pm

  9. Not sure which one will yield the greatest returns, but as I was thinking about the project I had a couple of thoughts:

    (1) If thinking about the “within-label micro-culture” it would be interesting to see if the rappers who are sampling soul/funk/blues artists are on labels that own the masters to the original works. If labels own the masters to the older stuff, I would presume that the rappers on that label would face less difficulty in clearing samples. For instance, Motown has a hip-hop division. Do rappers on the label sample Motown artists more than other rappers?

    (2) Is there a difference between the success of songs/rappers that incorporate vocal samples more prominently than production samples? What about songs/rappers that incorporate production samples more heavily? It seems to me that vocal samples would be more recognizable to the average listener. Could the use of vocal samples appeal explicitly to the average fan of the sampled artist? Production samples, I would think, are harder for the average listener to pick up, especially when there are multiple tracks/samples layered onto one another in the newly created rap songs. Therefore, could production samples appeal to the average listener in a much less overt manner?

    Just some thoughts. I hope they help.


    Mark Bissonnette

    November 21, 2008 at 9:33 pm

Comments are closed.

%d bloggers like this: