Quantitative methods III: Strength in numbers?

Abstract

In this third and final report on quantitative methods, I focus on academic community: what we do, what we call ourselves, and why this is a matter of importance for the entire discipline of geography, but especially quantitative human geographers. I first highlight the increasingly diverse ways in which quantitative methods community is produced and manifested, before turning to the shifting, ever-expanding, and overlapping names and labels used to define this group. I argue that, although there is ample evidence that the quantitative methods community is thriving, there is also a growing disconnect from the sub-discipline of quantitative human geography.

I Introduction

Much of what we purport to cover in quantitative methods reports such as this one deal in models and diagnostics, their philosophical foundations, and the evolving infrastructure that supports and facilitates the production of new methods and research. Underpinning all of this are people and social mechanisms that enable group cohesion—that is to say: community. Diffusion and uptake of new analytical approaches relies on traditional outlets such as journal and conference papers, but also newer mechanisms such as open course materials and software packages. All of these efforts are community driven and community reinforcing. The “we” of quantitative methods is readily apparent in our outputs: papers, software, books, and data services. On that basis, our presence is strong. Perhaps it has never been stronger.

Our “we” is rapidly evolving. Although our community appears externally cohesive when viewed from the perspective of outputs, it is increasingly internally fractured. In this report, my third and final in the series, I cap off my intention to consider “themes of flux and continuity in quantitative methods” (Franklin, 2022: 690) with a focus on community in quantitative methods. In many ways this builds directly on my previous two reports (Franklin 2022, 2023), which addressed the more mainstream quantitative issues of uncertainty and theory. Our community is in flux in ways that echo past experience but also defy it—and this is particularly evident in what we do and what we call ourselves.

In the context of this report, community is defined by practice: the what and how of quantitative research that binds researchers together, but also, especially, the modes of information sharing, building of norms, and construction of a collective identity (Wenger, 1999). Community in this sense echoes the ways in which many, including Barnes (2004) and Johnston et al. (2019), have written retrospectively of the ‘quantitative revolution’. The punch cards and mainframes are gone, replaced by R and Python, GitHub, Docker, and Bookdown. Gone, too, is the illusion of a monolithic, unified quantitative human geography, replaced by urban analysts, geographic data scientists, geospatial data scientists, and GeoAI practitioners. This report is not, however, written retrospectively; it is written from the trenches—although no battle has yet ensued. ¹ Writing in medias res will inevitably miss part of the bigger picture, as well as the final resolution of the story, but it captures other important, albeit less stereotypically academic, aspects of community: the quotidian and practical. Developing, testing, and applying quantitative methods requires not only technical but also social infrastructure.

Community is a timely topic. It’s apparent in the number of valuable and novel ways in which quantitative methods knowledge is currently produced and shared—including open source software and teaching materials, handbooks and textbooks instructing on the application of methods in different programming and computing environments, communal ways of sharing knowledge and providing professional support, and new sections of journals focused on code and data. Although fundamental to progress in quantitative methods, these activities, which help build and maintain our community, are rarely in the spotlight.

In addition, although quantitative methods are flourishing, it doesn’t necessarily follow that quantitative human geography is also flourishing. The past several years have witnessed the emergence of several new sub-fields that seek to carve out and redefine the space that, previously, was largely occupied by quantitative geography. What we see is less a community of practice and more a multiplicity of communities of practice, which introduces new complications to the task of reporting on progress in quantitative methods.

In what follows, I first outline several of the ways in which community is performed and maintained in quantitative methods before turning to a discussion of community identity and how this connects to an array of emerging sub-fields associated with quantitative methods. I then close with some thoughts on how the quantitative community is changing and the opportunities and challenges this evolution presents.

II Celebrating our community of practice

The headline areas of focus in quantitative methods, which have remained constant for the last couple of decades, are to a large extent characterized by an increasing emphasis on openness and what Fotheringham (1998) might have termed the “tools” that we employ to improve our models and methods. Chief amongst these are data, reproducibility, and software—all of which serve as the enabling platform for quantitative methods research and community.

Access to, and availability of, data is a necessary input for empirical quantitative research. At the same time, the advent of the Big Data and Data Science eras has not only drawn renewed attention to the promises and perils of data, but has also created demand for new methods and analytical approaches, as well as new opportunities for applied research. It’s therefore little surprise that there’s been growing interest in documenting the increased range and characteristics of spatial data available (Arribas-Bel, 2014) or that geographers have heralded a new epoch in data-driven geography (Miller and Goodchild, 2015). At the same time, much has been written about the shortcomings of new forms of data (e.g., Graham and Shelton, 2013; Kitchin, 2013; O’Sullivan, 2018). This is well-trodden ground; what’s important here is the way in which these dialogues have served to open-up data—with all the attendant caveats—to help make it a true public good for the quantitative methods community.

Concern with reproducibility is everywhere, once you look for it (a very small sample: Brunsdon and Comber, 2021; Chen and Poorthuis, 2021; Goodchild et al., 2021; Kedron et al., 2021; Nüst and Pebesma, 2021). On the face of it, reproducibility is about the degree of robustness and confidence that we can have in quantitative, computational, or analytical research (Brunsdon, 2016). As methodological approaches and data have become more complex, awareness of the benefits of openness has increased (Paez, 2021). From an alternative perspective, reproducibility is about sharing and community: providing transparent explication of decisions that have been made regarding data, variables, software environments so that others know what we’re doing. As Brunsdon and Comber (2021) note, reproducibility goes hand in hand with open data and code, as well as collaboration and community. Similar points are raised in Singleton et al. (2016), although from the viewpoint of geographic information science and not explicitly quantitative methods.

Nowhere are community needs more center stage than where software is concerned. Development of free and open software packages is nothing new (Anselin, 2012): GeoDa, and its antecedent, SpaceStat, (Anselin, 2003) and GWR (Fotheringham et al., 2002) are only a few examples of applications that have existed for many years. There are three developments that arguably make recent years different. The first is the widespread adoption of programming languages, notably R and Python, and use by quantitative researchers, which has helped further shift the balance away from closed, proprietary software, whose operations tended to be more obscured and automatic, defying collaboration and documentation. The second is the development of software packages, aimed at every step of the analytical cycle, from data ingestion, to model estimation, to visualization (see, e.g., Anselin and Rey, 2022; Bivand, 2021). Packages are generally produced by community members, for community members, with no expectation of payment or other rewards (e.g., counting towards academic promotion), although this may be changing with increased attention to the academic contribution that code and software make (Arribas-Bel et al., 2021). Third is the increased emphasis on coding as community (Rey, 2019, 2023), which originated outside geography, which highlights not only how groups of researchers produce tools for free and general consumption, but also how the practice of doing so helps to reinforce community.

There’s also longstanding tradition of producing textbooks, handbooks, and other instructive materials. Myriad volumes have been written over the decades that not only introduce quantitative methods, but also assemble bodies of knowledge, and provide rallying points for community. These include classics in the field on quantitative geography (Fotheringham et al., 2000), geocomputation (Longley et al., 1998), spatial analysis (Fotheringham and Rogerson, 2008, 2013), and geographic analysis (O’Sullivan and Unwin, 2003). They also include the influential (in their time) “Concepts and Techniques in Modern Geography” (CATMOG) series of booklets that provided overviews of geographical methods (Gregory, 1983; Johnston et al., 2019).

The above is a non-exhaustive list that serves to illustrate the strong community quantitative methods has long rested on and to emphasize how things have changed in recent years. CATMOG, for example, is no longer updated, but lives on in perpetuity online, freely accessible and hosted by the Royal Geographical Society’s (RGS-IBG) Quantitative Methods Research Group (QMRG, 2023). Textbooks and primers continue to be published (e.g., Brunsdon and Singleton, 2015, on geocomputation, or Singleton et al., 2017, on urban analytics) but often today focus on implementation tools (i.e., code) alongside the methods. For example, Lovelace, Nowosad, and Muenchow’s (2019) primer on geocomputation stresses the geocomputation “with R.” Similarly, Rey, Arribas-Bel, and Wolf’s (2023) recent volume on geographic data science emphasizes the “with Python,” Kopczewska (2020) highlights the “R” with spatial statistics, and Moraga (2019) R for geospatial health data. Notably, although published as books, many are free and open online, increasing community access and also allowing the methods and software to be live objects that can change and evolve. In parallel, across the quantitative methods community it has become much more common to publish teaching and training materials free, open, and online (e.g., Arribas-Bel, 2021; Kolak, 2023; Paez, 2022; Reades and Rey, 2021). This represents a marked sea change in how quantitative methods is done and taught and typifies the positive ways in which quantitative methods is a communal and collective effort.

III Lamenting our eroding community identity

So far, so good. The production and dissemination of quantitative methods is healthy and robust and this is at least partly due to practices that have enhanced community. What about group identity, though? A key component of community is solidarity in identity, however, construed. On this front, quantitative methods—and quantitative geography—is undergoing a massive splintering, one which seems to have recently picked up pace and shows little sign of slowing. To work in quantitative methods is increasingly to identify with geocomputation, geographic data science, urban analytics, urban data science, geoAI, and the list goes on. There are also related longstanding group identities around geographic information science (GIScience), regional science, critical GIS (O’Sullivan, 2006; Thatcher et al., 2016), and computational social science (Torrens, 2010). The past months alone have yielded multiple job advertisements, workshop calls, and other references to geospatial data science and spatial data science. Meanwhile the territory of human geography appears to have been ceded (not that it was a battle!) by quantitative methods. ² What is going on?

Opinions differ on the history and story of quantitative methods in geography (Barnes, 2004; Gregory, 1983; Johnston et al., 2019), especially around timing and impact, whether it was in fact a “revolution” and the ending of the so-called revolution (Burton, 1963). Typically, phrases such as, “intellectual breaks and ruptures” (Barnes, 2004: 568) are employed and paradigm shifts are invoked, with the general sense that, when it occurred, the emergence of quantitative geography was a wave that swept the discipline, monolithic in character, and eventually receded. One exception is Macmillan (1997), who argued that quantitative geography was heterogeneous from the start and its eventual recession was overstated. Certainly, quantitative human geography has long been characterized by diversity, both internally and externally. Internal to the discipline, quantitative human geography spans transport, population, and cities, for example. Externally, quantitative human geography is almost inextricable from regional science, in the North American context at least (Franklin, 2021).

There’s precedent for fracture—or at least substantial muddying of the identity waters—within quantitative methods. Quantitative geography has shared the analytical stage with GIScience and geocomputation for over 20 years, both of which overlap considerably in membership and research purview. Goodchild (1991) wrote of geographical information systems transcending technology and focusing more on “generic issues of geographic information” (p. 198). In contrast, Murray (2010) characterizes GIS as a core method within quantitative geography. Wilson (2021), writing about the origins of GIScience, appears to root the field in quantitative geography. However, GIScience and quantitative geography are treated as separate sub-fields in geography. In both the United Kingdom and the United States, for example, GIScience and quantitative methods are represented by separate groups within the RGS and American Association of Geographers (AAG). In this journal, too, quantitative methods and GIScience have separate streams of progress reports.

For its part, geocomputation could be considered a 21st Century moniker for quantitative geography, although Openshaw (2014) links it to “computational science” and views it as encompassing all of quantitative human geography and more. Many appear to define geocomputation as a combination of computation and geography (Fotheringham, 1998; Harris et al., 2017)—“doing geography with computers,” as O’Sullivan puts it (Harris et al., 2017: 598). Macmillan (1997) adopts a similar stance, tying geocomputation directly to quantitative geography. ³ However, just to complicate things, Miller (2002) in assessing geocomputation, writes that it appears to be mostly GIS. In sum, an argument can be made that much of GIScience and, certainly, geocomputation is quantitative geography with a different name. This much is evident: over the past 25 years there has been a computational turn in geography that straddles the borders of quantitative methods and this has spurred the creation of new labels and identities. These additional identities, GIScience and geocomputation, also potentially resonate outside the discipline in ways that quantitative geography does not.

It doesn’t stop there, however. We are currently witnessing an explosion of identities and sub-groups associated with quantitative methods. These include urban data science (Kang et al., 2019), urban analytics (Batty 2019a, 2019b; Boeing et al., 2021; Singleton et al., 2017), geographic data science (Arribas-Bel and Reades, 2018; Harris et al., 2017; Singleton and Arribas-Bel, 2021), and GeoAI (Li et al., 2020). Individually, any of these emerging fields can be seen as a natural successor, or rebadging, of quantitative geography—or geocomputation or GIScience. And, individually, the justification for each is compelling: the need for novel approaches that leverage the potential of data and data science in a geographical context. GeoAI accentuates computational resource and deep learning, positioning itself as “data intensive GIScience” (Li et al., 2020). Urban analytics and city science emphasize the urban, as well as systems and modeling perspectives (Batty 2019b). In the same vein, urban data science and geographic data science focus on the interface between data science and location, whether urban or more general. Much of it sounds like the same wine in different bottles.

IV Fractals and factions

The question here—if there is a question—is whether, when placed alongside quantitative human geography, GIScience, and geocomputation (already arguably a crowded field), this emerging plethora of new sub-fields damages disciplinary coherence and identity. There are benefits to carving-out new disciplinary corners and niches, including external signaling (Kang et al., 2019; Boeing et al., 2021), competition for scarce resources within a university context, and the natural evolution of disciplines. From this viewpoint, rebadging and renaming are rational and contribute to the long-term health and viability of quantitative methods. There are, however, diminishing returns to establishing multiple sub-fields that each aim to accomplish a similar goal. For example, there is little additional benefit that accrues to there being a geographic, a spatial, and a geospatial data science. One will likely suffice. Moreover, the embrace of the potential of artificial intelligence and data science in quantitative geography is not new (Couclelis, 1986) and to ignore this is to create false historical breaks and intimate paradigm shifts that may not be occurring.

Expanding and overlapping sub-fields also has implications for community in quantitative methods and quantitative human geography. At the individual level, many will realistically hold multiple identities, as quantitative human geographers, but also as urban or geographic data scientists. It is possible, and even desirable and unavoidable, to be active members of multiple communities and this has little disadvantage for quantitative methods. Indeed, from a methodological perspective there are considerable advantages to exposure to advances occurring outside geography. From a disciplinary perspective, however, the splintering of identities, and the increased distance from geography, creates a risk of decreased internal group cohesion, wider disciplinary alienation, and, paradoxically, potentially decreased external visibility, as we present as many very small sub-fields, rather than one large, albeit heterogenous, group. Moreover, the emphasis on tools—AI, data science, computation—although advantageous in the short-term, also creates an association between quantitative methods and implementation, which is overly narrow and underbounds the true extent of our reach. Lastly, it sets a precedent of additional schism with the advent of every new analytical approach; if current trends continue, imagine where we will be in 10 years.

Quantitative researchers will be fine. What about teaching within our discipline? On this front, the impact of quantitative methods identity shifting away from geography (and towards data science or analytics) may have debilitating longer-term effects. There have been extensive debates, at least in the UK context, regarding the role of quantitative human geography in, especially, undergraduate curricula. One example is the exchange between Johnston et al. (2014) and Cresswell (2014), along with other commentaries, including O’Sullivan (2014), around the extent to which a textbook, and undergraduate teaching more generally, mischaracterised, and rendered invisible, quantitative geography in geographic education. Some departments, in recent years, have made a deliberate move towards teaching geographic data science or urban analytics—one concrete way that quantitative methods community and identity affect the discipline as a whole—creating additional distance between quantitative methods and geography. Another indicator is the extent to which quantitative human geography has been relegated to methods or tools, where teaching is concerned. How often are quantitative human geographers teaching urban or economic geography? By identifying with tools and analytics, quantitative methods effectively shunts itself even further away from human geography, pigeonholing quantitative human geographers as methods people.

V Conclusions: Strength in numbers! ⁴

Community is important. It’s the ecosystem within which quantitative methods are developed, applied, and shared. Community is also the source of the raw material—interactions, sharing of knowledge, intellectual stimulation and support—that nurtures careers and collegiality and defines norms for training, research, and teaching. How one assesses the quality of the quantitative methods community likely depends on one’s position and perspective. However, there are a few general points that can be made.

First, there are no losers in terms of how quantitative methods community is practiced. Whether on the research or teaching side, at both the disciplinary (geography) and sub-field (quantitative geography) levels, increases in sharing and openness provide strong benefits in terms of exposure to new approaches and methods, quality of outputs, group cohesion, and costs. We have further work to do on inclusivity, but we have laid a strong foundation for success.

There are also numerous “branding” benefits that accrue both internally, within university settings, and externally, to the wider world and especially industry and government, in how quantitative methods is identified. Specifically, re-labeling and rebadging quantitative methods as “data science” or “analytics” may help increase visibility and attention from university leadership and help position geography programs to attract scarce resources for teaching and training in quantitative methods. And increases in resources and student numbers for quantitative methods often benefits entire geography departments; the destinies of quantitative and non-quantitative geographers are intertwined. Externally, pitching quantitative methods as “data science” may help with graduate employment and validation in interactions with industry or government.

The current, and seemingly unprecedented, ⁵ fragmentation of the quantitative methods community identity, however, is unlikely to be collectively beneficial. That is to say, identifying and naming a new sub-field within quantitative methods may generate a return (e.g., publication, promotion, grants, new hires, students) at the individual or team level, but as more new labels crowd the landscape, differentiation becomes more difficult, especially for those, such as industry or university leaders, who were already a bit confused about what geospatial or geographic or spatial data science are all about. At the community level, then, visibility and leverage are diminished.

And what about quantitative methods’ relationship to quantitative human geography? Here is where I see the greatest potential for worry. ⁶ There exists already a longstanding gap between human geography and quantitative human geography. In spite of Barnes’ (2009) assertion that the split between quantitative and critical geography is artificial, it’s not uncommon for quantitative human geography to be omitted from characterizations of the discipline (e.g., Schurr et al., 2020; Franklin et al., 2021). Although speaking of GIScience, both O’Sullivan (2006) and Thatcher et al. (2016) refer to GIS and human geography as separated by a “fence” (O’Sullivan) or “on opposite sides of the table” (Thatcher et al.). This separation is often echoed in descriptions of quantitative geography and human geography—us and them—and is often cast as a slight on the part of human geographers, who have excluded their quantitative kinfolk from their midst. However, if we don’t identify ourselves as quantitative human geographers, I am not sure we can complain! My point is that, while quantitative methods does not suffer when community identity shifts away from quantitative geography, quantitative human geography surely does.

I have painted an equivocal picture of our contemporary quantitative methods community. This reflects my current mood, as a geographer who employs quantitative methods to answer human geography sorts of questions. On the one hand, where practice is concerned, I laud my community for its openness and inclusivity. We should keep up the good work. On the other hand, perhaps we should stop coming up with new names for ourselves and, instead, devote a share of that energy to vocally promoting quantitative human geography and all that we have to offer to our discipline and to the world.