In the reconstruction of the model that we have undertaken above, no inconsistencies exist between Aukrust’s long-term model and the steady-state model in growth-rate form. However, economists and econometricians have not always been precise about the steady-state interpretation of the system (3.10)-(3.13). For example, it seems to have inspired the use of differenced-data models in empirical tests of the Scandinavian model—Nordhaus (1972) is an early example.
With the benefit of hindsight, it is clear that growth rate regressions only superficially capture Aukrust’s ideas about long-run relationships between price and technology trends: by differencing, the long-run frequency is removed from the data used in the estimation; see, for example, Nymoen (1990: ch. 1). Consequently, the regression coefficient of, for example, Aae, t in a model of Awe, t does not represent the long-run elasticity of the wage with respect to productivity. The longer the adjustment lags, the larger the bias caused by wrongly identifying coefficients on growth-rate variables with true long-run elasticities. Since there are typically long adjustment lags in wage-setting, even studies that use annual data typically find very low coefficients on the productivity growth terms.
The use of differenced data clearly reduced the chances of finding formal evidence of the long-term propositions of Aukrust’s theory. However, at the same time, the practice of differencing the data also meant that one
avoided the pitfall of spurious regressions (see Granger and Newbold 1974). For example, using conventional tables to evaluate ‘t-values’ from levels regression, it would have been all too easy to find support for a relationship between the main course and the level of wages, even if no such relationship existed. Statistically valid testing of the Norwegian model had to await the arrival of cointegration methods and inference procedures for integrated data (see Nymoen 1989a and Rpdseth and Holden 1990). Our evaluation of the validity of the extended main-course model for Norwegian manufacturing is found in Section 5.5, where we estimate a cointegrating relationship for Norwegian manufacturing wages, and in Section 6.9.2, where a dynamic model is formulated.