Persuasion: Reflections on economics, data, and the ‘homogeneity assumption’

Abstract

This paper discusses issues to do with the empirical basis of modern economics and points towards the need to look more closely at the ‘homogeneity assumption’ that underpins much economic theory. It argues that severe problems currently prevent economics from becoming more persuasive to both students of economics and those outside the discipline. The issue involves the management of disciplinary boundaries, and excessive use of the ‘homogeneity assumption.’ Three areas of concern are explored. First is the literature on causes of growth, and the role of policy. The paper documents reasons to doubt the existence of robust relationships between growth and policy variables. Second is the ‘homogeneity assumption’ that different countries are usefully viewed as members of a single population. Third is evidence suggesting that an assumption of ‘normal’ maximizing behaviour has to be justified, not just assumed, and that regular deviations from the usual maximizing assumptions occur with gender and culture.

The paper argues that a central issue in economic methodology and pedagogy should be, as North implicitly argues, the negotiation of disciplinary boundaries: what economics can versus cannot explain. It suggests more explicitly basing the choice of explanatory models on empirics identifying where the model applies.

Keywords:

• homogeneity
• neo‐institutional economics
• robustness testing
• methodology
• policy advice
• experimental economics

ACKNOWLEDGEMENTS

This work draws upon a wide range of experience in policy advice and economic analysis, mainly relating to Vietnam. From this I owe much to many people too numerous to mention. For the work presented here, I thank for comments on earlier drafts: D. Gale Johnson, Adrian Wood, John King, Gus Edgren, Warwick McKibben, Doug J. Porter, Stefan de Vylder, Peter Robson and various anonymous referees, participants at an ADB seminar in Sept 2001, and the Experimental Economics ‘Global Club’ for happily sharing materials with me. I thank Valerie Tan for Research Assistance. The mistakes of interpretation remain mine.

Notes

The disputes about whether sub‐disciplines such as agricultural economics and development economics have anything theoretically specific to bring to the discipline support my position here.

I take as a working definition of the ‘homogeneity assumption’, the assumptions of ontological and epistemological universalism discussed by Kenny and Williams 2001: that is, that both the things economics studies, and how they should be understood, are the same through time and space. This is the same as the assumption that mass, as a characteristic of things, is the same here as on Mars, and should be analysed the same way in both places. See below for the relationship between these issues and the assumption of ‘instrumental rationality’.

Many economics departments, at least in Australia, are now situated within Faculties of Commerce rather than with the other social sciences. I do not wish to imply that the methodologies of other disciplines are, essentially, better guides to persuasion than those of economics: acceptance of ‘relativism’ does not really help people cope with intentional and rationalised choice, which is how policy remains understood and articulated, as practice.

Explanations of where these ‘theory‐mindsets’ come from is not something adequately explained by economics alone.

I would propose, thus, that typical questions asked of the presenter of an attempt to model, for example, the determinants of growth in a particular context, should include those that require the researcher to have considered whether variables like ‘inflation’ are indeed, conceptually and empirically, the same in his or her context as elsewhere, and to what extent they stand in causal relationships with other variables that are the same as those elsewhere: that is, that the ‘ontological and epistemological’ assumptions be explored – see below.

Goldfarb and Stekler 2000 seek to develop an empirical methodology for assessing changes in economists' views. They conclude, however, that further arguments are needed to provide a convincing explanation of the conflicting results in the literature (110). In their case study, they observe relatively little empirical relationship between empirical evidence and the acceptability of the rational expectations hypothesis (112). Compare this methodology with the more discourse‐based approach of Yonay 1998, in discussing the debates between neoclassical and institutional economists before WWII.

In my own teaching experience exposure to this work has a somewhat electric effect upon students both within and outside economics. This result is interesting.

This article was in fact preceded by another, Levine and Renelt 1992, which reached similar but less provocative conclusions (‘Almost all results are fragile … to small changes in the conditioning information set’ (942)). Later, in Easterly and Levine 1997, Levine used cross‐country regressions techniques to explore African growth processes, ‘adding‐in’ a ‘newly discovered variable’ (in this case ‘ethnicity’) to find that ‘… ethnic diversity helps explain cross‐country differences in public policies and other economic indicators’ (1203).

I am referred to McAleer, Pagan and Volcker 1985 questioning the original paper by Leamer 1983. Also Hendry and Mizon 1990. More recently, Hoover and Perez forthcoming examine what happens if the robustness criteria are relaxed. Perhaps not surprisingly, if the allowable range of variation is increased, more robust relationships can be found. Yet, what basis is there for judging what an allowable degree of significance is acceptable? We could chose to come back to Winch's position, and regard research as something that is human business, which may find the Truth, but can never be certain of it. To quote Hoover and Perez forthcoming: ‘We can never guarantee that the specifications selected by the general‐to‐specific approach are true. But the approach is part of a critical, indeed dialectical, methodology’ (19). Interestingly, this work concludes that ‘What is surprising is how few of the variables matter in the end and how much is left unexplained. The preferred regression explains only 42 percent of the variability of countries' growth experiences …’ (19). This could be taken to suggest that the rest of the story is ‘contingent’; as I argue below, this conclusion, like others, points to a possible consensus that articulates agreement over cause‐effect relations in terms of an amalgam of universalist and local relations: yes, we agree that these relations hold for all of us; yes, we agree that these other relations hold for the here and now, and not necessarily for others. This would seem a wise formula.

It seems to me that a proper awareness of the homogeneity assumption, refined and expressed in terms of statistical analysis ‐ i.e. that the sample is drawn from a single population ‐ is needed to make meaningful the statistical tests for parameter estimates. If it is unwarranted, and unless measures are taken to deal with this, significance results are literal non‐sense. Thus the risk would be that a technique, if one could be found, for securing robustness would remain founded upon sand: the robust results would reflect random order … leading to a range of conflicting but significant results, as in the case of the cross‐country regressions. Whether viewed as an issue in statistical analysis, or as a central assumption in the selection of analytical frameworks, the question of homogeneity is a valuable but (according to my own review of responses to Levine and Zervos under‐used) check upon the degree to which positions that purport to be based upon empirical work should be taken as persuasive.

The significance of the transition from laminar to turbulent flow is that wings often stall when this happens, which can kill people.

The work surveyed covers a wide range, and I have had to be selective in the ways we have used it for this review. I apologise for any apparent misrepresentation, which is unintentional.

See Kline 1980 for a discussion of alternative responses to situations where plausible assumptions lead to contradictory results.

Wood (xxxx: 68) here is arguing for investigative methods that focus more upon context and locality. Thus ‘…what is important for growth is not mainly how much you trade, but what you trade and how you manage your trade’.

Compare eg Panlilio 1963 for a discussion in a natural science.

See also Hausman 1992.

Here a caveat is in order. Such ‘differences in behaviour’ can be nested or non‐nested. That is, they can suggest different underlying models, along the lines of the comment from Camerer (1995) quoted above, or they can suggest that the parameters of the model are different in the different cases. In the simpler language of the engineer, Mars rocks are either not rocks at all, or they are, simply redder.

Yvonne Dunlop, personal communication. See Kidd and Shannon 1997 for an example. Standard methods assume lifelong employment, which is not what most women experience. In this case the difference is one of parameters, but if one reads Kidd and Shannon differently, especially their closing remarks about wage determination, they point to more fundamental issues.

They discuss three types of experiments: Public Goods, Ultimatum and Dictator. On reflection, and granted that in many human societies women bear contingent liability in the case of local ‘system breakdown’ (ie for the children in the case of family break‐up), it might be argued that their apparent selfishness when risks appear high is, rather, a qualitative shift in strategy reflecting a need to look after themselves so they can cope with the contingent liability.

Ethnographically, an interesting point here was the Chinese willingness to spend time exploring opponents' activities for signs of pre‐known strategies, referenced to classic stories and literature on war and its standard stratagems. This the Germans did not seem to do. The interactions were videotaped. In terms of the discussion of instrumental vs procedural rationality, this would seem to be a clear example of search for procedures, and selection of responses in consequence, where knowledge of the local meaning was essential to any persuasive analysis.

See for two stimulating studies, Lindauer and Pritchett 2002 and Yonay 1998.

For a fascinating study of the nature and relationship between interwar neo‐classical and institutional economics, see Yonay 1998. He argues that both were swept away by the mathematisation of economics as Samuelson's work became influential in the late 1950s. He also points to a far more widespread influence of institutional economic thinking than some other authors, and also to their concern to find new models to explain the economics data that was coming available (an endeavour that, he argues, failed).

Thus ‘institutions are unnecessary; ideas and ideologies do not matter; and efficient markets – both economic and political – characterise economics’ (2).