What is Sociology in the USA?

Background

Sociologists in the USA like describing sociology as the most heterogeneous social science of all, which is

…perhaps another way of saying that it has been less successful at institutionalizing itself as a discipline than its close relatives. Unlike economics, it does not have a core kit of analytical tools and models codified in textbooks and widely accepted as legitimate both inside and outside the field… Unlike political science, on the other hand, sociology does not have a well-defined empirical core to unify it, either.

Healy (2012, p. 88)

There are two ways of looking at this.

Glass Half-Full

Andrew Abbott (2001) claims sociology’s defining characteristic is “the fact that the discipline is not very good at excluding things from itself.” Some see this as something good because it turns sociology into some miniature version of all social science. In this view, the lack of “internal cohesion” in the discipline is interpreted as the cost we’ve paid in exchange for occupying a central place in the social science landscape (Moody and Light 2006).

Many celebrate the lack of a disciplinary core as providing some unique form of academic freedom.

This is the reason why many people claim they haven chosen to become sociologists.

I chose sociology because more than any other social science sociology would let me do what I pleased. If I went into sociology, I wouldn’t have to make up my mind what to do.

—Andrew Abbott

Why, then, did I choose sociology as an academic home?

Of all the available social sciences, sociology seemed to me to be the least disciplinary; it had the fuzziest boundaries. But even more significantly, sociology has valued its own marginal traditions in a way that other social sciences don’t.

—Erik Olin Wright

Glass Half-Empty

Others are less optimistic.

Presumably, we are not as interdisciplinary as we would like to believe.

…most sociologists don’t really get interdisciplinarity. Whether we acknowledge it or not, most of us have internalized a sociological supremacy that makes us believe our field’s insights are more important, more complete, more nuanced than those of other scientists (Healy 2017). This cultural background of intellectual superiority helps create what Lizardo (2014) called the “Comtean schema”—the implicit belief that all proper interdisciplinary research should take the institutional form of a subfield of sociology.

In a brilliant insight, Lizardo noted that sociologists create virtual “avatars” of other disciplines within sociology instead of working with their real-world counterparts. That is, rather than engage with economists, we create “economic sociology”; rather than engage with political science, we have “political sociology”; rather than engage with cultural evolution or cognitive science, we invent “the sociology of culture and cognition,” and so on. This fools us into thinking that we’re being interdisciplinary when, in reality, “[t]hese subdisciplinary avatars have been created by sociologists for sociological consumption” (Lizardo 2014:985).

Vaisey (2021, p. 1298)

Data

Note. The visualization provided below is provisional. I will need to email someone at the ASA and politely ask for permission to use the full directory.

When a sociologist joins the American Sociological Association (ASA), they have the options to choose up to four “areas of interest” for everyone else to see. For example, I chose (1) Theory; (2) Quantitative Methodology; (3) Organizations, Formal and Complex; and (4) Law and Society. And I would probably have chosen something different on a different day.

So, I created a dataset containing the areas of interest of over 1,000 individuals associated with 20 sociology programs in the USA. This dataset is provisional, but see Table 1 in case you are curious about the sample.

Each of these individuals chose anywhere between one and four areas of interest. Figure 1 highlights the ten largest areas of interest in blue. The largest is Stratification/ Mobility, which gives credence to the view that inequality defines the subject-matter of sociology. Among the top ten we also find what some may describe as sub-disciplinary avatars—Cultural Sociology, Economic Sociology, and Political Sociology—which gives credence to the view that sociology is a miniature version of all social science. Furthermore, the view that sociology is mostly concerned with race, class, and gender is also well represented in this top ten.

Note that some areas of interest like Demography and Medical Sociology are very well represented but don’t seem to

Figure 1: Areas of interest in the dataset. The largest 10 areas are highlighted in blue.

However, this doesn’t provide us with any information about the “structure” of the sociological areas of interest—i.e., the pattern of connections which some may describe in terms of coherency (or lack thereof).

This is what the next section is all about.

Two-Mode Networks

The direct ties between individuals and areas can be transmogrified—via simple matrix multiplication —into indirect ties among areas (Breiger 1974; Agneessens and Everett 2013). This idea should be familiar to everyone who chose Social Networks as area of interest.

transmogrification is not the “technical” term for this

To do this, we arrange the individuals and their corresponding areas in matrix form. Here, we have a matrix \(\mathbf X\) with 1112 rows (one for each individual) and 61 columns (one for each area). Each cell \(x_{ij}\) is either a \(1\) or a \(0\). Then we transform \(\mathbf X\) into a new adjacency matrix \(\mathbf A\) with each cell \(a_{ij}\) corresponding to the number of “shared” or “intersecting” individuals between areas \(i\) and \(j\).

\[ \mathbf{A} = \mathbf{X}^\top \mathbf{X} \tag{1}\]

The diagonal of \(\mathbf A\) contains the total number of individuals that chose each area, which corresponds to the same numbers in Figure 1.¹

Unfortunately, the resulting network, is very dense. It takes only one single individual with idiosyncratic tastes to create ties among four otherwise disconnected areas. As a result, the “structure” is one in which every area is seemingly connected to every other area.²

Because of this issue, researchers have looked for ways to “trim” uninformative ties. The intuition is straightforward. If the observed number of shared individuals between two areas ( \(a_{ij}\) ) exceeds the number we would expect to see in a random network (the null model), then we draw a tie between area \(i\) and area \(j\). In other words, we do a “significance test” for each tie in the network depicted by \(\mathbf A\).

If the links between individuals and areas of interest were random, the resulting network would consist entirely of isolated nodes. This is what would happen if there was no disciplinary structure.

Results

The results of this procedure are shown in Figure 2, which uses a significance level of 0.01 to draw ties between areas.

I used Zachary Neal’s (2022) backbone package to do all this.³

As we may have expected, there is a strong overlap between Stratification/ Mobility and Education; Demography and Family; Medical Sociology and Mental Health; Migration/Immigration and Latina/o Sociology; and so on.

The strong overlap between Cultural Sociology and so-called Theory deserves its own blog post.

Note. All isolate nodes have been removed. A bigger sample of individuals and their chosen areas would obviously result in more connected areas.

Figure 2: “Strong” Ties ( α = 0.01 )

Figure 3 uses a significance level of \(0.05\) to draw the ties between areas, which means we get to extract more of them.

The result is that we get to see more “structure.”

For example, the heterogeneity within Economic Sociology may be more easily understood by looking at adjacent areas: Social Networks, Political Economy, Science and Technology, and Organizations, Formal and Complex. The same might be said for other areas that look like “hubs” (e.g., Stratification/ Mobility, Political Sociology).

Futhermore, perhaps we may hypothesize that “triangles”—like the overlap between Medical Sociology, Aging/ Social Gerontology, and Demography—are formed by somewhat distinct underlying groups of sociologists.

Also, note that Quantitative Methodology now appears connected to three areas (Stratification/ Mobility, Demography, and Mathematical Sociology), but Qualitative Methodology is nowhere to be seen yet. This does not mean that one is more important than the other. It just means that the overlap between Quantitative Methodology and those three areas is stronger than what we would expect by chance; whereas the overlap between Qualitative Methodology and other areas is more compatible with a “null model” of random ties between areas.⁴

Figure 3: “Strongish” Ties ( α = 0.05 )

Finally, Figure 4 uses a significance level of \(0.06\) to draw ties, just enough for connections between smaller areas to start showing up (e.g., Marxist Sociology, Alcohol and Drugs). Religion is now connected to Cultural Sociology

Please keep in mind that this does not mean that these connections are somehow “weaker” than the other ones in any substantive way. Hypothesis testing is extremely dependent on sample size.

Figure 4: “Strongish” Ties ( α = 0.06 )

That’s all for now.

Sample

Universities and Individuals
	n	%
Northwestern University	84	7.55%
University of Michigan-Ann Arbor	76	6.83%
University of Chicago	72	6.47%
Harvard University	69	6.21%
University of Wisconsin-Madison	68	6.12%
Stanford University	66	5.94%
University of Pennsylvania	66	5.94%
Cornell University	65	5.85%
Brown University	62	5.58%
New York University	61	5.49%
Princeton University	59	5.31%
Columbia University	47	4.23%
UC Berkeley	46	4.14%
Ohio State University	45	4.05%
Duke University	42	3.78%
Yale University	42	3.78%
University of California-Los Angeles	41	3.69%
University of North Carolina-Chapel Hill	36	3.24%
Indiana University-Bloomington	33	2.97%
University of Texas-Austin	32	2.88%
Total	1,112	100%

Table 1: Number of individuals extracted from the ASA member directory

Citation

The backbone package automatically outputs the following suggested text:

Figure 2:

We used the backbone package for R (v2.1.2; Neal, 2022) to extract the unweighted backbone of a weighted and undirected unipartite network containing 61 nodes. An edge was retained in the backbone if its weight was statistically significant (alpha = 0.01) using the disparity filter (Serrano et al., 2009). This reduced the number of edges by 99%, and reduced the number of connected nodes by 68.9%.

Figure 3:

We used the backbone package for R (v2.1.2; Neal, 2022) to extract the unweighted backbone of a weighted and undirected unipartite network containing 61 nodes. An edge was retained in the backbone if its weight was statistically significant (alpha = 0.05) using the disparity filter (Serrano et al., 2009). This reduced the number of edges by 96.3%, and reduced the number of connected nodes by 34.4%.

Figure 4:

We used the backbone package for R (v2.1.2; Neal, 2022) to extract the unweighted backbone of a weighted and undirected unipartite network containing 61 nodes. An edge was retained in the backbone if its weight was statistically significant (alpha = 0.06) using the disparity filter (Serrano et al., 2009). This reduced the number of edges by 94.9%, and reduced the number of connected nodes by 21.3%.

Neal, Z. P. 2022. backbone: An R Package to Extract Network Backbones. PLOS ONE, 17, e0269137. https://doi.org/10.1371/journal.pone.0269137

Serrano, M. A., Boguna, M., & Vespignani, A. (2009). Extracting the multiscale backbone of complex weighted networks. Proceedings of the National Academy of Sciences, 106(16), 6483-6488. https://doi.org/10.1073/pnas.0808904106

References

Abbott, Andrew. 2001. Chaos of Disciplines. University of Chicago Press.

Agneessens, Filip, and Martin G. Everett. 2013. “Introduction to the Special Issue on Advances in Two-Mode Social Networks.” Social Networks 35(2): 145–47.

Breiger, Ronald L. 1974. “The Duality of Persons and Groups.” Social forces 53(2): 181190.

Healy, Kieran. 2012. “Sociology.” Pp. 90–122 in. New York: Blackwell.

Knoke, David, Mario Diani, James Hollway, and Dimitris Christopoulos. 2021. Multimodal Political Networks. Cambridge University Press.

Moody, James, and Ryan Light. 2006. “A View from Above: The Evolving Sociological Landscape.” The American Sociologist 37(2): 6786.

Neal, Zachary P. 2022. “Backbone: An R Package to Extract Network Backbones.” PLOS ONE 17(5): e0269137.

Opsahl, Tore. 2013. “Triadic Closure in Two-Mode Networks: Redefining the Global and Local Clustering Coefficients.” Social networks 35(2): 159167.

Vaisey, Stephen. 2021. “Welcome to the Real World: Escaping the Sociology of Culture and Cognition.” Sociological Forum.

Footnotes

1. This new matrix \(\mathbf A\) is sometimes called a two-mode or bipartite projection. The original network represented by \(\mathbf X\) goes by many names—e.g, affiliation network, bipartite graph, two-mode network, etc.

2. Knoke et al. (2021) [pp. 36] note three problems with using this sort of projection. First, it entails a loss of “identity information” on one of the two node sets—e.g., the cell entries in \(\mathbf A\) don’t reveal the specific individuals shared across areas. Second, it results in very dense networks, which leads to biases in network measurements—e.g., an artificially high number of triangles (see Opsahl 2013). Third, it obscures the generative process behind tie formation.

3. The threshold level here is arbitrary, this is a blogpost. The “disparity filter” algorithm was chosen because it was the fastest, this is a blogpost.
   
   The choice of algorithm and threshold level is rarely discussed in most publications of this sort, so I guess I’m sort of following tradition here.

4. The area of Ethnography (Anthropology)—which has about half the amount of individuals as Qualitative Methodology—has a connection to Urban Sociology, reflecting a long tradition of famous urban ethnographies.