5 min read – cross-posted on Big Data & Society’s blog.
This post presents the paper Epistemic clashes in network science: Mapping the tensions between idiographic and nomothetic subcultures (45 min read).
My interest for networks was passed to me by one of my teachers, Franck Ghitalla, who had just read Albert-László Barabási’s best seller book Linked. Like many others, I was intrigued by the discovery of the scale-free network, a new and exotic structure that scientists started to find in every aspect of the world, from genetics to economy, from the power grid to terrorism, and to love. Or at least, that is what Barabási claimed.
The scale-free network is special because a few nodes get most of the links, while the rest is poorly connected. The number of links follows a power law, a distribution already known to be intriguingly pervasive in physics. From there, Barabási and other researchers went on a quest to theorize a universal law of complex networks, using the scale-free model as a foundation. But as the emergent field of network science consolidated, the apparent simplicity of the situation faded away. More accurate measurements challenged the pervasiveness of the power law. Models required more sophistication. Power laws were found in non-scale-free contexts. Scale-freeness became more difficult to assess in empirical situations. Yet the pervasiveness of more or less scale-free networks remained: Network scientists had found something, but what? The field adopted flexible umbrella terms such as “complex network” and “heavy-tailed distribution” to account for the diversity of empirical cases. Network science persisted as a field, but the prospect of theorizing a universal law had lost momentum. The scale-free model was productive despite constant criticism, but now, two decades later, some want to let it go to the benefit of a more experiment-driven approach. They meet a fierce resistance.
Epistemic clashes in network science narrates and reflects on the two main disputes on scale-freeness. The first dispute started in 2005 with a pre-print contesting the characterization of the power law, discussed in blog posts first, and later in academic publications. A compromise position gradually emerged, giving the impression that a consensus had formed. Network scientists believed that the problem originated in a disciplinary divide between statisticians and physicists, and that a better mutual understanding had solved it. However, a second dispute started again in 2018, once again as a pre-print contesting the pervasiveness of scale-free networks, discussed on social media then in academic publications. It came as a surprise to many network scientists, who considered the case closed.
This repeated failure of network scientists to agree on the facts established by their own field motivated me to analyze their exchanges as a controversy, drawing a methodological inspiration from Bruno Latour. My material consists of a set of 40 academic and non-academic publications that I selected, coded and analyzed. I synthetize, document and illustrate the dynamic of the controversy by focusing on a reduced set of actors and claims. I then propose my own interpretation about the persistence of a disagreement, arguing against the commonly accepted idea that it roots in a disciplinary divide.
To social science scholars, I offer a guided tour of network science, a field that is probably less well known than social network analysis. My account should make clear that the field is not as unified as it may seem from a distance. I put a particular effort into providing a nuanced perspective on the epistemological commitments of physicists such as Barabási. Their nomothetic approach to knowledge postulates the existence of universal laws of nature – finding them being the purpose of science. Evelyn Fox Keller criticized Barabási’s “faith” in “the traditional holy grail of universal ‘laws,’” but I argue that researcher’s beliefs are of no relevance here. The postulated existence of laws acts as an epistemic device, driving methods and shaping scientific claims; laws, here, are not a knowledge, but a way to know. Barabási’s papers, contrary to his general public books, do not claim the existence of universal laws. Their scientific validity does not depend on their existence, because to postulate is not to believe; that is why the approach is scientifically effective – robust. Social science scholars may find instructive to follow the resistance of network science’s nomothetic claims to their constant criticism. Although the nomothetic approach might be ultimately losing its dominance on the field.
To network scientists, beyond a recap of a controversy they already know about, I offer a better explanation of why the controversy reopened recently. The commonly accepted idea that it is rooted in a disciplinary divide is not satisfying. Indeed, bridging the disciplinary gap did not solve the problem; on the contrary, Aaron Clauset, a researcher who actively worked at it, co-authored the pre-print that reopened the controversy. I borrow Peter Galison’s concept of trading zones to develop a better diagnosis. I argue that since the inception of network science in the late nineties, theorists have been trading their models for the results of experimentalists. Theorists offered the scale-free model, postulating the pervasiveness of the power law; they needed experimental evidence of this pervasiveness. Experimentalists needed theoretical material to design their experiments; they employed the scale-free model and produced enough empirical evidence to ground the claim to pervasiveness of scale-free networks. This trade was beneficial to both sides, but it was not symmetrical. I argue that the controversy arose when experimentalists (e.g. Clauset) pushed for their own program to characterizing power laws independently of the scale-free model. This program is problematic because it breaks the exchange. Indeed, without the scale-free model, the experimental results do not benefit theorists such as Barabási; worse, they might challenge existing models and lead to new theories. I suggest that the situation is controversial because the balance of power is shifting in network science: from a dynamic where theory leads to experiments (a theory-driven program), the field is moving to experiments calling for new theories (experiment-driven program).
Read the paper online: