A classic that wasn’t: Statistical Geography and paths only later taken

I Introduction

Science is a cumulative activity, with each generation of scholars standing on the shoulders of those who preceded them and with its progress very largely preserved through the published record – the books, chapters and articles reporting material that others build on. And yet there are many examples of publications that fail to attract the attention they probably deserved. Potentially important material may not be immediately taken up and will only get wide appreciation either when ‘discovered’ at some time after its publication (what some call ‘sleeping beauties’: Fang, 2018) or when ‘re-invented’ by others, perhaps in ignorance of the earlier source. In human geography, for example, Duncan (1974) has shown that recognition of Hägerstrand’s pioneering work on spatial diffusion was delayed in part because it did not fit into the dominant paradigm of geographic research at the time of its initial publication and in part because recognition of its innovative nature was only fully realised by those seeking a paradigm shift during his visit to Seattle in 1959/60: before then he was a ‘productive isolate’. After that visit he received wide recognition and enjoyed a very considerable reputation for his work on diffusion and subsequent studies.

Such a substantial oversight of a potentially important source also occurred with a book entitled Statistical Geography: Problems in Analyzing Areal Data by three American sociologists (Duncan et al., 1961: henceforth DCD), which attracted very little attention from geographers. Disciplinary myopia meant that, as far as citation analysis can show, relatively few geographers accessed it – certainly in the decade after its publication – let alone followed up some of its important suggestions regarding quantitative analyses of spatial data. According to Google Scholar^© it had received 476 citations by early 2018 (accessed 26 April 2018), of which only 70 were by individuals readily identified as geographers.

That failure is somewhat surprising given the ferment of change that characterised human geography at the time of its publication. According to Burton (1963) the discipline had ‘undergone a radical transformation of spirit and purpose’ in the preceding decade, that he considered best portrayed as the ‘quantitative revolution’ which ‘had reached its culmination in the period from 1957 to 1960, and is now over’. That conclusion may have been presumptuous; indeed, in retrospect there probably never was a complete revolution as that term is generally understood – the overthrow of an established order. Nevertheless, there was a rapidly-growing increase in the application of statistical methods to geographical data (Johnston et al., 2018), in the UK as well as in the USA, and a book with that title should have attracted considerable attention among those seeking to promote the revolution/paradigm shift. But it did not – even though one of the revolution’s leaders, Brian Berry (pers. comm.), was a colleague of Dudley and Beverley Duncan at the University of Chicago when the book was being written and encouraged them to give it a title including the word geography; the Duncans worked in the university’s Population Research and Training Center and were not closely linked to the Department of Sociology, home of the world renowned ‘Chicago School’ of urban sociology.

II The book

As with many books, DCD’s subtitle is more informative about its contents than is the title; Statistical Geography does not cover the entire field then being pioneered by geographers (there is nothing on point and line patterns, for example, and no mention of theoretical models of location such as central place theory) and its entire focus is on data for areal units. This reflects its origins in a series of studies on ‘Natural Resources and Regional Economic Growth’ undertaken for the Resources for the Future Inc. thinktank which involved exploring a set of methodological problems associated with the available data for such studies – most of them collected by the US Bureau of the Census and similar bodies. The authors’ Preface indicated their ‘manifestly incomplete’ knowledge of the geographical literature and their only becoming aware of some of it when the work was nearly complete, but also their belief that the problems they identified did not apply just to ‘any unique body of subject matter which representatives of a particular discipline are best equipped to investigate’ (p. vi). They expressed the hope that ‘geographers may learn from our efforts’; the argument we develop here is that very largely they did not, and important issues raised were only appreciated later – in some cases much later – in apparent ignorance that DCD had identified them as important, indeed crucial to the proper quantitative analysis of areal data.

The book is in three unequal parts: a short section (25 pages) on ‘Preliminaries’; a slightly longer one (29 pages) on ‘Areal Units and Areal Data’; and a much longer one (115 pages) on ‘Analysis of Areal Data’. The first identified four main ‘perspectives on areal differentiation’ within the extant literature: chorography, or areal differentiation, including regionalisation (Richard Hartshorne and Preston James are both quoted); analyses of areal distributions; analyses of spatial structures; and explaining areal variation. Use of regions in some analyses attracts criticism (pp. 140–1):

…some investigators think of regional differentiation as playing a role in the explanation of areal variation. In our opinion, this view is to be accepted only with grave reservations. In common parlance, of course, we talk as if ‘regions’ constitute an influence on social and economic phenomena…[but this may be] only a pseudo-explanation, at best a clue to the discovery…of some heuristic value for an investigator familiar with conditions prevailing in the region

The book’s core is in its third section, which identified four main objectives in the analysis of areal data (pp. 60–1):

The aggregation of data to obtain a datum for a territorial universe;

Most attention is given to the fourth of these.

A wide range of issues is raised, few of which were immediately taken up by geographers – see, for example, the discussion of compositional effects (pp. 99ff). The problems of ecological inference are covered, for example (pp. 69ff.; for a recent review of this topic see Gnaldi et al., 2018), and there are hints regarding both shift-share analysis (p. 63) and trend surface analysis (p. 134). We pay particular attention in the next section to three major issues facing the analysis of areal data that were discussed in some detail, two of which were realised by geographers within two decades of DCD’s publication, but without any reference to it; the full import of the third was not realised for several decades.

III The hints not taken

In the first decades of human geography’s ‘quantitative revolution’ it was generally accepted that standard statistical procedures deployed in other social sciences – notably but not only those comprising the general linear model – could be applied to geographical/spatial data without any particular problems. In 1956, however, Reynolds raised a number of specific issues regarding the geographical analysis of statistical data, in a brief paper that received little attention, with only 20 citations according to Google Scholar^©: Barnes (1998: 216) termed it an ‘anodyne review of the potential of statistics in geography…which included a few mild cautionary remarks’. Reynolds claimed that geographers needed to develop ‘distinctive tools’ specific to their discipline, but Garrison (1956: 428) responded that the ‘present methods of statistical inference are applicable to spatial-type problems’ such as correlation and regression, that ‘In short, there is ample evidence that present tools are adequate to our present state of development. No type of problem has been proposed that could not be treated with available tools’. He concluded that ‘The logical methods of science are universal. If assertions that geography is somehow different are accepted without proof, we may lose the benefits derived from the findings of others’ (p. 429). Indeed, that geography was not considered different is indicated by Hepple (2001: 385) noting that Yule ‘in the 1890s…constructed both the theory and application of multiple regression analysis, using geographical’ rather than individual biometric data (see also Denis and Docherty, 2007).

By the 1970s, however, it was clear to some that spatial data of various types needed bespoke methods and developments were put in train. Statistical Geography had earlier contained hints, in some cases strong hints, of those needs, but they were not taken up.

IV Selective progress

Science progresses through the accumulation of knowledge. Central to this process is the dissemination of new findings and arguments through printed media, in some cases enhanced by presentations at conferences, symposia, seminars and other events. Individuals and groups hear/read what others have done and build their future work on those foundations. But, as noted above in reference to Duncan’s (1974) study of Hägerstrand’s influence, some published works may not have a substantial impact even though they are innovative in one or more ways. Statistical Geography is one such work: it has a large number of citations but, despite the noun in the title, only some 15 per cent of them from geographers. Further, few of those items in which DCD was cited take up any of the major issues facing spatial analysis that they raised: most just make relatively bland statements about them – as in Simmons’ (1967: 389) footnote that ‘Problems of analysing areal data are discussed’ in that book. ²¹

So why did Statistical Geography attract so little attention among geographers, especially in the years immediately after its publication when such a title should have been a magnet for the small, but rapidly increasing, number of geography’s practitioners for whom the topic was central to the changes they wished to make to their discipline? In his exploration of the delayed recognition of Hägerstrand’s work on spatial diffusion Duncan (1974) suggests three possible causes of such a situation: (i) general social resistance, perhaps linked to linguistic or political barriers; (ii) what he terms a ‘paradigmatic effect’, with the work being largely ignored because it did not fit into the discipline’s established practices; and (iii) a ‘Matthew effect’ – science is practised in interacting communities, groups of individuals who share common interests (research subjects, methods, etc.), and the work of outsiders – especially those with few if any links to one or more of those communities – may be ignored because of the lack of contacts. (Granovetter’s (1973) work on social networks is relevant in this last case. Each scientific community is characterised by strong ties among its members, which sustain its activity. But one or more members may have weak ties with extra-community members, and these can be the source of new ideas. Without such weak ties, however, the community may either not recognise or just ignore such work.)

The first of those possibilities does not apply to DCD. The authors were American, Dudley and Beverly Duncan were by then well-established scholars working at a prestigious university, ²² and their book was produced by a reputable publisher, based largely on work for a well-known think-tank. With regard to the second possibility, Statistical Geography clearly did not fit into the then-established geographical paradigm (as clearly set out by both Hartshorne (1939, 1959) and James and Jones (1954): both are cited by DCD). But, as Burton (1963) had argued, the book was published at a time of considerable ferment within geography, with a growing number of its scholars seeking to change its practices fundamentally – indeed to ensure a revolution (Johnston and Sidaway, 2016). But citation analysis suggests that most of those involved in promoting that tradition, within which the adoption of the rigour of statistical analysis was a central goal, didn’t find stimuli within DCD. ²³

To some extent, that is understandable. Although the adoption of quantification was core to the revolutionaries’ agenda, most of their attention in the early years of their activity was not on areal analysis. Following the lead set by Schaefer (1953 – not cited by DCD), among others, their focus was on spatial analysis, on point and line patterns, much of it set within the context of central place theory. As exemplified by Bunge’s (1962) pioneering case for this ‘new geography’, areal analysis was not central to its concerns and so the methodological issues raised by DCD, with most of their empirical attention on a single substantive topic (regional economic growth) that was also peripheral to much contemporary geographical study, were not relevant. They became so – as illustrated by Simmons’ (1967) paper; indeed it could be argued that within little more than a decade areal analysis had supplanted spatial analysis as the preferred/practised methodology of much quantitative human geography, but by then DCD had attracted little attention and was not widely discussed within geography’s relevant research communities.

Simon Duncan’s paradigmatic argument has some relevance to an appreciation of DCD’s limited impact on human geography in the 1960s, therefore. So too does his ‘Matthew effect’ argument. DCD’s three authors were sociologists with few direct contacts with geographers (apart from Brian Berry, the only other geographers acknowledged in their Preface are Harold McCarty and Arthur Robinson, both of whose published works are cited in the bibliography, along with those by 12 other geographers). These were ‘weak ties’, in Granovetter’s (1973) terms, but they seem to have contributed little to the emerging geographical practices. Geography was not widely recognised as a social science in the early 1960s: few other social scientists developed relationships with geographers then and few geographers sought inspiration from their work – something that changed very substantially when there was a shift away from the geometric focus of spatial analysis towards areal analysis and the study of spatially varying population profiles (as in social area analysis: Berry and Rees, 1969: Johnston, 1969, 1971). ²⁴ That sociologists in the United States were well ahead of geographers in their adoption of relatively sophisticated quantitative approaches is well illustrated by the contents of Hagood’s (1941) textbook and its revised edition (Hagood and Price, 1952).

The history of scholarly disciplines often has a Whiggish connotation, emphasising continuity as the present builds on the past, with occasional interruptions – paradigmatic shocks. Those histories focus on the publications that have most impact. Citation analyses show– on that metric at least – that many publications have little or no impact and so do not appear in the written histories (or the unwritten ones, such as those presented to students). Does that matter; is it of value to future disciplinary adherents to know about such pieces, even if some think they might/should have been important? As exemplified here, if the ideas that they encapsulate are important they will surely be taken up by others a little later – progress may be delayed, but not substantially so. This was certainly the case with both spatial autocorrelation and MAUP: DCD’s identification of their importance to areal analysis was not followed up, but the former became a major research issue within a decade and the latter within two decades, although in that case resolution has never been reached. But the third exemplar deployed here – multilevel scale issues – went totally unrecognised for four decades, and when independently discovered then (by Tranmer and Steel, 2001) it again went largely disregarded until a further re-invention some 15 years later. ²⁵

For a variety of reasons, good and bad and including those advanced here, some pieces and arguments are likely to go largely unrecognised, increasingly so as the volume of published work continues to expand exponentially. Researchers have to be selective in what they read and take account of, and to a large extent they work within the established parameters of their own sub-discipline (or set of practices and concerns that might be trans-disciplinary) and the formal and, especially, informal networks to which they belong. Some may, occasionally if not frequently or regularly, cast their net more widely and, perhaps serendipitously, encounter something that they can bring into, and potentially change, their paradigm. They are the ‘weak ties’ without which much science may atrophy, but strong intra-community ties dominate scientific practices and it may be that no firm links are made with a body of literature – even a single item – that could transform a set of practices. It is perhaps also relevant that, despite DCD, areal analysis did not come to occupy a major role in sociology in the 1970s–1980s either (or, for that matter, political science); if it had, there may have been more cross-fertilisation with geographers, but the growing emphasis on survey analysis by sociologists which geographers didn’t immediately share meant that there was a major unbridged methodological gulf between the exponents of quantitative analysis in the two disciplines (see Converse, 1987). ²⁶

Scientific practices, as illustrated here, are socially constructed and are conducted within relatively closed communities. Occasionally those communities are challenged by new ideas, that may be independently developed internally or, more likely, are introduced from an external source. The latter may be instrumental in changing a discipline, or at least one or more of its parts, but if their introduction is not undertaken, or is at best partial and not then followed up by other practitioners, the impact may be slight. That was the case with DCD’s Statistical Geography: despite its relevance to their cause, geographers seeking to change their discipline in the 1960s didn’t identify it as valuable material to advance their agenda, and its authors, although hoping that ‘geographers may learn from our efforts’, did very little to advance that. ²⁷ In the history of statistical geography, Statistical Geography played a very minor role – and geography changed anyhow!

There is a growing literature on what is termed ‘counterfactual history’ (Evans, 2016; see also Fearon, 1991, and Tetlock and Belkin, 1996), which addresses ‘what if’ questions – conjectures on what did not happen, or which might have happened, that assist understanding of what did happen (Black and McRaild, 2007). These might also be addressed to the history of academic disciplines and their practices, allowing us to appreciate better the paths taken by exploring the paths not taken and why. Just as the quantitative revolution might have taken a rather different form – lacking its early commitment to logical positivism, for example – if its proponents had been influenced by the writings of Crowe (1938) and Jones (1956) on laws and tendencies, might its practices have developed differently if DCD had been widely read and built-upon in the 1960s? Crowe and Jones were geographers who published in geography journals, but those papers had little impact; DCD were not geographers but they published a book with geography in the title. The lack of impact for Crowe and Jones suggests a failure of bonding social capital within geography – is it the same now? And DCD’s weak impact suggests an absence of bridging social capital between disciplines six decades ago – raising the same question. A Romanian sociologist has represented his discipline as characterised by ‘multi-paradigmaticity, scattered cumulativity and multi-localized ignorance’ (Rusu, 2012), a representation that has been applied to the contemporary situation in human geography (Johnston et al., 2014). Its wider application, alongside the concepts of bonding and bridging capital and that of strong and weak ties between communities, could shed much light on disciplinary history – of what has been and might have been. ‘History is written by the winners’ is a much-challenged aphorism that has more than a ring of truth when the history of geography – or the nature of progress in geography – is being debated; exploring the discipline’s ‘might-have-beens’, the ‘losers’ who had little or no impact, could illuminate how and why it has progressed along its chosen paths.