Many of the points taken by Krugman (1995) in his account of the development of economic geography and regional science are well taken, and deserve to be received with more grace than has appeared in replies so far. Economics does not however share the institutional contexts and research traditions of geography, which, like economics, has both a disciplinary core and ``flavours'' extending in many directions. Several of these are manifestly present within this review, particularly medical and physical geography. At present, the clear focus of many quantitative and applied geographers is on geographical information systems and collaboration with disciplines like computer science and surveying. In spatial statistics, the key breakthrough occurred in 1973, with the publication of Cliff and Ord's Spatial Autocorrelation. Cliff has deepened his concern with epidemiological modelling in geography (Cliff and Haggett, 1996), while Ord, a statistician of note
, works from time to time with geographers such as Getis (Getis and Ord, 1996).
An intelligent description of the current setting and research traditions of quantitative geography has been written by Hepple (1998) in reply to an aggressive social-theoretic attack, in which all contact with correlation and regression is condemned for the links between Galton and Pearson and late nineteenth century eugenics. I feel that it is worth noting Hepple's approach. He is careful to express understanding for the criticism advanced, and proceeds to use the same methods as the social theorists in order to demonstrate that, with a more contextual reading admitting additional information from the period in question, one would have found that opinions were also divided. In the case of regression, Hepple advances a persuasive case for arguing that Yule, studying what we would now term social exclusion, made a more important contribution, and that consequently it would be premature to condemn quantitative methods as such on the basis of just some of their associations.
Indeed, it will be useful for our present discussion to cite Yule (after Hepple, 1998, p. 279):
The investigation of causal relationships between economic phenomena presents many problems of peculiar difficulty, and offers many opportunities for fallacious conclusions. Since the statistician can seldom or never make experiments for himself, he has to accept the data of daily experience, and discuss as best he can the relations of a whole group of changes; he cannot, like the physicist, narrow down the issue to the effect of one variation at a time. The problem of statistics are in this sense far more complex than the problems of physics.
Before we proceed to take up key issues raised in using the spatial `data of daily experience', a few words on a very few selected examples of empirical work in trade and location. The geography of innovation in the context of knowledge spillovers is a research area with substantial interest, but where opportunities for interaction with spatial statistics do not yet seem to have been exploited sufficiently (Jaffe, Trajtenberg and Henderson, 1993, Audretsch and Feldman, 1996). In work on dynamic externalities and growth in cities, Henderson (1997, p. 455) does admit that the residuals ``may be correlated for all counties within a metropolitan area'', and uses a simple ad hoc diagnostic. The study of the determinants of economic growth using cross-sectional regressions (Sala-i-Martin, 1994, Barro, 1997), despite technical sophistication, does not seem to have opened for the testing of hypotheses concerning residual or structural neighbourhood effects. In conclusion, attention can be fruitfully drawn to the work of Francophobe economists; Thisse (1997) sums up lucidly the indeterminacy of regional bounding, showing how processes like spillover render the construction of entities for empirical purposes problematic -- we will return to this issue again as the modifiable areal unit problem.