5 Antecedents
Spatial econometrics may be seen in a number of different ways, as misguided, as an inward-looking endeavour for the specially interested, but these views are rarely placed in their context. In this section, we will sketch the antecedents of spatial econometrics, both providing views on the research context when it came into being, and on changes in econometrics over time. Grasping spatial econometrics today without insight into its antecedents is unnecessarily difficult, so taking time to explore how ideas came into being and developed is well-spent.
In the same way that Fujita, Krugman, and Venables (1999) begin their study of the spatial economy by looking at the antecedents of their subject, it is helpful to place spatial econometrics in its temporal and academic context. This context is sufficiently different from the contemporary setting that it may be hard to grasp the background for many of the features of spatial econometrics that came into being during its earlier years. Indeed, the ranges of topics that were studied in economics in the 1960’s and 1970’s differ markedly from those in focus today. If we can sketch the context within which spatial econometrics was created, and its methods developed, we should be able to illuminate choices made then which influence our understanding and application of spatial econometric methods.
Critics of the practice of spatial econometrics, such as Gibbons and Overman (2012), appear to overlook these antecedents, and consequently judge the potential of the field on a partial, perhaps anachronistic, understanding, viewing phenomena with a history in ahistorical way. Since we are attempting to provide an introduction to applied spatial econometrics, we need to throw light on the original motivations and concerns of the first scholars engaged in the field. Anselin (2010b) indicates clearly and repeatedly (Anselin 1988, 2006, 2010a) that we should acknowledge Spatial Econometrics by Paelinck and Klaassen (1979) of the Netherlands Economic Institute as our starting point, and so celebrated thirty years of spatial econometrics in 2009. This firm confirmation of the importance of Jean Paelinck’s contributions as scholar and community-builder is fully justified. We should then turn to the motivations given in Paelinck and Klaassen (1979) to indicate which contextual factors were of importance at that time, and the breadth of the academic communities with which they were in contact.
In a recent short commentary, Paelinck (2013, 11) recalled his conviction, expressed in 1967, that
early econometric exercises … relating only variables posessing the same regional index … were inadequate to represent the correct spatial workings of the economy, which would then be reflected in the policy outcomes.
A year before, Paelinck (2012, 36) pointed to salient isomorphisms linking spatial regression models, simultaneous equation models and input-output models; these were known of and discussed in the early formative period of spatial econometrics. We will return in subsequent chapters to the ways in which spatial regression models may be specified, but for now, a simple presentation of these isomorphisms as perceived in the early period is sufficient:
\begin{equation*} {\mathbf y} = {\mathbf A}{\mathbf y} + {\mathbf X}{\mathbf b} + {\mathbf \varepsilon}\end{equation*}
is a spatial regression model where {\mathbf A} is a matrix expressing the mutual first order spatial dependencies between regions — the similarity of this form and the Koyck distributed lag model is striking (Koyck 1954; Klein 1958; Griliches 1967);
\begin{equation*} {\mathbf A}{\mathbf y} + {\mathbf X}{\mathbf b} = {\mathbf \varepsilon} \end{equation*}
is a simultaneous equation model where {\mathbf A} is a matrix expressing the dependencies between the equations; and:
\begin{equation*} {\mathbf y} = {\mathbf A}{\mathbf y} + {\mathbf f} \end{equation*} is an input-output model where {\mathbf A} is a matrix of sectoral input-output coefficients, and {\mathbf f} is final demand.
Input-output models, simultaneous equation models, and the importance of policy outcomes were all known intimately at the Netherlands Economic Institute at this time, and elsewhere among applied economists. The isomporhisms flowed from the known to the unknown, from the stuff of contemporary research and policy advice to doubts about the calibration of aspatial models, and on to what became termed spatial econometrics. If we compare these topics with those described for Regional Science by Boyce (2004), we can see the outlines of research priorities at the time: including urban and regional models for planning, regional and interregional input-output models, transport and location models. During the 1960s and 1970s, many of these models were enhanced, matching needs for policy advice, to cover environmental questions, adding natural resources as inputs and pollution to outputs. Paelinck’s co-author in a key paper in spatial econometrics (Hordijk and Paelinck 1976), went on to work in environmental management and research.
Reading Paelinck and Klaassen (1979, viii), we see that the programme of research into the space economy undertaken at the Netherlands Economic Institute led first to the publication of Paelinck and Nijkamp (1975), and then to Klaassen, Paelinck, and Wagenaar (1979), published in the same year as Spatial Econometrics. All three books were published in the same series and appear to reflect the core concerns of economists at the Institute doing research on regionalised national macro-economic models. The direct link to Jan Tinbergen is evident in the account of the context of economic research in the Netherlands given by Theil (1964, 49). If we take Paelinck at his word, he and his colleagues were aware that an aspatial regionalisation of national accounts, of input-output models, or transport models, might prejudice policy advice and outcomes through inadequate and inappropriate calibration.
Klaassen, Paelinck, and Wagenaar (1979) is mainly concerned with model construction, while about a third of Paelinck and Nijkamp (1975) is devoted to input-output analysis. Both books show sustained concern for aggregate economic measurement, especially of national accounts data, intersectoral transactions, and many other topics. Considerable attention is also paid to the data collection units, be they sectors or regions. The need to attempt to define regions that match the underlying economic realities was recognised clearly, and a key part of Paelinck and Nijkamp (1975) is devoted to regionalisation, and the distinction between functional regions and homogenous regional classifications is made. In the motivation for spatial econometric models given in Paelinck and Klaassen (1979, 2–3), consumption and investment in a region are modelled as depending on income both in the region itself and in its contiguous neighbours, termed a “spatial income-generating model.” It became important to be able to calibrate planning models of this kind to provide indications of the possible outcomes of alternative policy choices, hence the need for spatial econometrics.
Economic planning was widespread in Europe at the time, and was also central in the development of Regional Science, in particular input-output models; as Boyce (2004) recounts, Walter Isard worked closely with Wassily Leontief. Operational and planning motivations for applied economics were unquestioned, as economists in the post-war period saw their role, beyond educating young economists, as providing rational foundations for economic policy. It is worth noting that Jean Paelinck participated actively in the Association de Science Régionale De Langue Française, becoming president in 1973–1976. The first president of the association was François Perroux, who had founded it with Walter Isard in 1961 (Bailly, Derycke, and Torre 2012).
Until the 1980s, it was not at all unusual to publish original results in other languages than English. French spatial economic research, for example Ponsard (1983), while making little impact in Anglophone countries, was widely used in teaching and research elsewhere (Billot and Thisse 1992). They contrast, though, the “word wizardry of François Perroux with the rigour of Claude Ponsard” Billot and Thisse (1992, 194), echoing the views expressed by Drèze (1964, 3–4) with regard to the work of Perroux. Even if we accept that “word wizardry” deserves more rigour and recasting in normative and empirically testable forms, it is also part of the context within which spatial econometrics came into being. A reading of Perroux (1950, a lecture given at Harvard University) is worthwhile, because it not only gives the reader a vignette of the context in the post-war period, but also provides a discussion of economic space, as opposed to “banal, unreflected space, mere position”, that has largely disappeared from our considerations.
The title of the journal: Regional and Urban Economics, Operational Methods, founded by Jean Paelinck in 1971, and which was renamed as Regional Science and Urban Economics in 1975. Boyce (2004, 43), points to the perceived importance of “operational methods”, a version of the term “operational theory and method” used in the title of Paelinck and Nijkamp (1975). Spatial econometrics does not seem to have come into being as a set of estimation techniques as such, as perhaps we might think today, but rather as an approach addressing open research questions both in space economy and in the enhancement of interregional models to be used in offering policy advice.
Were motivations of this kind common during the 1960s and early 1970s? Not only was the spread of Regional Science extensive and firmly established (Boyce 2004), but public bodies were concerned to regionalise economic measurement and policy advice (see, for example Graham and Romans 1971). In Britain, Environment and Planning was started in 1969 with Alan Wilson as founding editor and published by Pion; he was assistant director at the Centre for Environmental Studies at this time before moving to the University of Leeds. In a recently published lecture series, Wilson (2012, 35–36) cites Paelinck and Nijkamp (1975) as giving principles for contributions from economics to urban and regional analysis (see also Wilson 2000, 34). The papers presented at annual Regional Science meetings were published in a series by Pion; the first number in the series included contributions by Granger (1969) and Cliff and Ord (1969).
In a contribution to a panel session at the 2006 annual meeting of the American Association of Geographers (co-panelists Luc Anselin and Daniel Griffith), Keith Ord pointed to the continued relevance of Granger’s remarks at the meeting almost fourty years earlier (Ord 2010, 168–69); we will return to these concerns below. As noted by Bivand (2008, 1–5), communities of researchers working in and near mathematical and theoretical geography was more integrated in the pre-internet and pre-photocopier age than one might expect, with duplicated working papers prepared using stencils circulating rapidly between collaborating academic centres. Knowledge of the preliminary results of other researchers then fed through into rapid innovation in an exciting climate for those with access to these meetings and working papers.
There was considerable overlap between quantitative geography and regional science, so that work like Cliff and Ord (1969) is cited by Hordijk (1974), and was certainly known at the Netherlands Economic Institute several years earlier. Although it has not been possible to find out who participated in the August 1968 conference of the British and Irish Section of the Regional Science Association at which Cliff and Ord (1969) was read, it was not unusual for members of other sections to be present, and to return home with bundles of duplicated papers. Up to the 1990s, presenters at conferences handed out copies of their papers, and conference participants posted home parcels of these hand-outs, indexed using the conference programme.
Leslie Hepple was among the more thorough scholars working on the underpinnings of spatial econometrics prior to the publication of Paelinck and Klaassen (1979). His wide-ranging review Hepple (1974) is cited by Bartels and Hordijk (1977), again demonstrating the close links between those working in this field. We will be returning to the review paper, and to Hepple (1976), which studies methods of estimation for spatial econometrics models in some depth, building on and extending Ord (1975).
Hepple (1974), like Cliff and Ord (1973), saw no distinction between spatial statistics and the antecedents to spatial econometrics. Obviously, spatial econometrics was strongly influenced by the research tasks undertaken by regional and urban economists and regional scientists. As Griffith and Paelinck (2007, 211) point out, spatial statistics and spatial econometrics continue to share most topics of interests, with each also possessing shorter lists of topics that have been of less concern to the other. They advocate a “non-standard” spatial econometrics, which is inclusive to wider concerns. It seems appropriate in this context to mention the somewhat heterodox position taken by McMillen (2010), who draws attention to the crucial issue of functional form, which he argues may well lie behind observed spatial autocorrelation; we will return to this in later chapters.