The alternative and more recent behavioural approach to industrial location theory focuses on the geography, growth and behaviour of the firm, not as an optimizing rationally-economic decision-making unit, but as one characterized by conflicting goals, limited levels of knowledge and control of its environment, irrationality of perception and behaviour, and so on. Its methodology and terminology, which is often derived from systems theory, is well illustrated by the essays in Hamilton's recent book (1974), all of which accept as a fundamental starting point ‘the domination of the location issue by the industrial organisation, firm or corporation: its goals, growth, size, age. production profile, organisation and behaviour' (Hamilton 1974, 13). This approach appears to be particularly influenced by the rise of giant multi-plant and indeed multi-national corporations, and the implications of their behaviour for industrial location change.

Without question, both the classical and behavioural schools have yielded valuable insights into real-world manufacturing location patterns and changes. This is probably rather more true of the behavioural school, with its greater reliance on actual firm surveys for hypothesis testing, than of more abstract 'desk-bound' classical theory where, as Smith (1971, 275) acknowledges, ‘very little progress has yet been made in ... direct application ... to real-world situations'. The main reason for this, in Smith’s view (1971, 276), is that classical theorists have been ‘more concerned with the construction of elegant theories of locational equilibrium, or with the fusion of location theory and production theory, than with providing a guide for empirical enquiry'. But additional factors are the very considerable data demands of classical theory testing, its basic preoccupation with static patterns rather than changes over time, and its stress upon the impact of transport costs on materials and products on manufacturing location. As later discussion shows, the latter influence would seem to be of minor and diminishing importance today, at least at the United Kingdom geographical scale. Moreover, of course, classical theory's normative economic basis, concerned as it is with long-run optimization of locational economic benefits, largely ignores variations in actual firm behaviour consequent upon environmental and other uncertainties, information availability, and varied goals and forms of company organization (Hamilton 1974, 4-5). For all these reasons, a recent study of the specific topic of manufacturing migration within developed countries such as Britain concluded that ‘as far as the movement of firms is concerned, traditional theories need to be replaced by a behavioural theory of movement' (Keeble 1974a, 7).

Unfortunately, however, the behavioural approach also suffers from major defects as a framework for evaluating national-scale manufacturing location change. Thus, for example, it is even more data-demanding than classical theory. Significantly, all the survey-based studies in Hamilton's 1974 volume are forced to confine their attention to only a relatively few firms, the largest sample, that of Townroe (1974), being of approximately 200 concerns. The latter survey's relatively large scale probably explains why it was able to ask only ‘a limited number of questions about post-move stability and change' (Townroe 1974, 292); while, conversely, the fact that ‘the conduct and detailed analysis of lengthy, relatively unstructured interviews is immensely timeconsuming' explains why Stafford's contribution was based on investigation of only six firms (Stafford 1974, 169). A sample of between twenty and 100 firms appears to be about average. The problem this raises, of course, in addition to considerable logistical difficulties of data acquisition, concerns the arguable validity of generalizing from small-sample findings to all manufacturing firms. The latter is clearly an essential requirement both for a general theory and comprehensive national-scale evaluation of actual manufacturing location change.

Another major problem with behavioural theory is that the more firms are treated as individuals, rather than in the aggregate, and the more detailed the information gathered about them, the more difficult it appears to be to identify behavioural regularities which might form the basis of a general location theory. This is illustrated, for example, by Townroe’s 1972 study of fifty-nine mobile companies, which attempted to consider ‘in some detail the procedures of locational decision making’ (Townroe 1972, 261). Observations for each firm of sixty-four separate variables measuring different characteristics of the firm and its locational search behaviour were therefore subjected to statistical testing (via chi square tests) and grouping (via principal components analysis), in a search for behavioural regularities. As Townroe (1972, 267) readily admits, specifically in relation to the latter, ‘the results were disappointing’, since ‘the patterns of behaviour which have emerged are not strong and distinct’. Rather, the results reveal ‘a great diversity of experience’ and attest ‘the complexity and heterogeneity of the industrial location decision making process in the fifty-nine surveyed companies’.

A third and last major problem is that, perhaps as a result of these difficulties, no well-defined single holistic behavioural location theory has as yet emerged from the welter of recent behavioural industrial location studies. Such initial attempts as Pred's behavioural matrix approach were, in that author’s own words, largely ‘only a verbal formalisation of the fairly obvious’ (Pred 1967, 121). Most of the recent conceptual models proposed are not much more than classificatory descriptions of the sequences through which firms may (or may not) pass in making some sort of locational decision.

This study is therefore located, methodologically, squarely in the longstanding tradition of empirical geographical inquiry. Its main aims are positive, not normative, in attempting to account for real-world patterns and changes rather than propound some ideal or optimal locational arrangement. It is a problem-oriented study, as much concerned with government policy implications and impacts as with academic investigation. On the other hand, it does perhaps break some slight new methodological ground, in two ways. First, it attempts deliberate fusion of what has elsewhere (Keeble 1971a, 40—2) been termed micro-level and macro-level empirical industrial location analysis. In particular, the latter is generally designed deliberately to test findings and hypotheses suggested by the former. Micro-level analysis may be defined as investigation of industrial location change by direct data acquisition, usually by questionnaire interviews, from samples of manufacturing firms. As noted earlier, this survey approach, while sometimes necessitated by the absence of original data from any other source, ‘also reflects the implicit or explicit view that the most logical method of explaining' locational change ‘is to ask the decision-makers actually involved' (Keeble 1971a, 40). However, micro-level analysis undoubtedly suffers from a number of defects. One is the possibility of biased samples, strictly random sampling methods rarely being used for logistical reasons. Another is the problem of post-facto rationalization, which leads firms to ascribe locational decisions to factors which in reality have become important only subsequent to the decision. The cloaking of supposedly ‘irrational' or personal locational motives by reference to conventional, but in reality less important, economic factors is another example. A further problem may be actual inaccuracy in the data provided, especially where this relates to past periods and is based on the interviewee's memory. Lastly, individual firms may well not be explicitly aware of environmental advantages or disadvantages which are in fact contributing to their in situ growth or decline relative to other areas, particularly if they have had no experience of operating elsewhere. Since the latter is true of the majority of manufacturing firms in most areas, individual ‘one-off' micro-level surveys of such ‘adoptive’ firms (see below) in a particular area may not yield particularly meaningful results.

These problems indicate the need for an alternative but complementary approach to empirical investigation of industrial location change. Macro-level analysis represents just such an approach. This usually involves the use of aggregate published or unpublished statistics, selected as measuring possible influences upon and rates of spatial manufacturing change, in some form of hypothesis-testing statistical analysis or model. Descriptive statistical manipulation of aggregate statistics may also be included here. Clearly, one major problem with this approach is the inferential leap from observed spatial association to possible causal influence. Nonetheless, the macro-level approach has been used fairly widely to investigate industrial location change in Britain in recent years, both by geographers (Keeble 1971a, 1972b; Chisholm and Oeppen 1973; Sant 1975a) and economists (Moore and Rhodes 1976). But in the present study, perhaps unlike some others, its use has been largely confined to testing hypotheses explicitly derived from micro-level findings. In this way, both types of analysis, with their particular defects, can provide interlocking testimony on the importance or otherwise of particular locational forces, with a resultant increased likelihood of valid results.

The other minor methodological contribution is through the use of entropy index and multiple regression analyses of industrial location patterns. While neither of these approaches is strikingly new, it is nonetheless true that no previous study has utilized the disaggregated entropy index formula adopted here to measure the changing spatial concentration or dispersion of industry at the national scale in Britain; while it can also be argued that no previous national-scale industrial location study in Britain has utilized multiple regression analysis with full regard to the technical problems involved. Thus for ample even Sant's stimulating recent multiple regression manufacturing movement study (Sant 1975a) appears totally to ignore the potential problem of residual spatial autocorrelation.

Of course, the use of such techniques as regression analysis does imply a particular conceptual view of manufacturing location change. In particular, it assumes that such change is spatially structured and guided by identifiable forces, which produce a non-random arrangement of manufacturing growth or decline. In this respect, therefore, the approach of the present study differs somewhat from that of workers such as Lever (1974b, 184) who has argued, admittedly at the narrower urban-region scale, that ‘shifts in the distribution of manufacturing ... are difficult both to describe and to explain because of the large number of variables involved’; and that ‘industrial location and relocation are probability processes and, consequently, stochastic rather than deterministic models should be used in describing them' (Lever 1972a, 22). Stochastic models, notably Markov-chain models, may indeed be of practical value for describing and forecasting short-run changes in manufacturing location patterns, as argued by Lever and Collins (1975). But they would seem to lack any great explanatory power, other than of a general ‘random-process’ kind. They also ignore the mass of evidence, as for example in this book, indicating that whereas individual location decisions may well incorporate a large random element (Keeble 1974a, 21). deterministic-type forces can indeed be identified as influencing manufacturing location dynamics in the aggregate. This important scale difference between the individual and the aggregate has been noted by Haggett (1965, 25-6), who concurs with Bronowski (1960, 93) that ‘a society moves under material pressure like a stream of gas; and on the average, its individuals obey the pressure ... (although) at any instance, any individual may, like an atom of gas, be moving across or against the stream’ (my italics). Of course at one extreme some individuals, in this case manufacturing firms, may consciously recognize and deliberately conform to the structural forces operating upon the space economy. Such firms have been termed ‘adaptive’ by Alchian (1950) and Tiebout (1957), in that they rationally adapt themselves to these environmental forces, and ‘make well thought out locational decisions based on relevant information’ (Pred 1967, 22). At the other extreme, however, there will be firms which do not possess the ability or information to perceive the nature and direction of these forces, and hence ‘react to their environment in relative ignorance, with the lucky ones being adopted by the system' (Pred 1967, 22). Such ‘adoptive' firms may thus initially spring up randomly in a variety of locations; but only those fortunate enough to have chosen a suitable location in terms of the underlying structural forces will survive and/or grow. This simple adaptive/adoptive dichotomy of course relates to two polar extremes, most firms probably being located somewhere along a continuum in between. But however firms may be classified, the underlying conceptual approach of the adaptive/adoptive viewpoint, which has attracted considerable attention from behavioural industrial location analysts (Smith 1971, 273: Townroe 1974, 305; Walker 1975), is that longer-run firm location, survival and growth is shaped by structural environmental forces which hence produce non-random spatial patterns of manufacturing change. This viewpoint is fully supported by, and forms the conceptual background to, the analyses presented in this book.

The focus of this study is then explanatory analysis of recent and current manufacturing location change in the United Kingdom, largely at interregional and inter-subregional scales. Chapter 2 describes the patterns of change, measured in particular by hitherto unpublished employment and floorspace data, both cartographically and statistically. Chapter 3 then looks at the question of inherited manufacturing structure as a control on change, while chapter 4 sets out the key centre-periphery and periphery-centre models, discussing inter alia such broad issues as the influence of market accessibility, innovation diffusion, labour availability, quality and cost, agglomeration economies, transport improvements, entrepreneurial and institutional factors, and government regional policy. Chapter 5 presents some of the main original explanatory findings of the study, based on multiple regression analysis. Chapters 6 and 7 are concerned with different components of aggregate spatial change, the former looking at the relative contributions of manufacturing firm births, deaths, expansion and contraction and, in particular, physical migration, the latter investigating recent locational change in four key manufacturing industries, iron and steel, clothing, motor vehicles and electronic engineering. In both cases, much of the analysis is based on hitherto unpublished statistics. The last two substantive chapters, 8 and 9, consider respectively government regional industrial location policy, its justification, historical development and impact upon the peripheral assisted areas, and the effect and effectiveness of regional planning with regard to manufacturing location at the within-region scale in the most prosperous region of Britain, South East England, together with adjacent East Anglia. The concluding chapter looks briefly at two potential future issues for manufacturing location in Britain, the impact of EEC membership and of North Sea oil and gas, before re-emphasizing some of the key findings of the book.