HE major contention of this paper is that T American railroad investment behavior in the period 1897-1914 is best understood by emphasizing the role of financial factors, in spite of the fact that the accelerator theory from its inception has been most successfully applied to railroad investment in precisely this period [3, 11, 12]. (This was also true in Tinbergen (17) although Tinbergen preferred a model using profits as the explanatory variable.) This argument is also in direct contradiction to the recent explanation of railroad investment behavior in this period advanced by Kmenta and Williamson [10] . The argument rests upon significant changes in railroad finance which occurred at the beginning of the period. These permitted easier access to funds from sources external to the individual companies and likewise permitted more internal funds to be used for capital formation. This encouragement of investment due to easier financing was ended, however, by the Panic of 1907. These financial changes are sufficient to vitiate any form of the accelerator explanation of investment behavior which depends upon a fixed accelerator coefficient and a fixed lag structure of investment response to changes in demand for the period 1897-1914 as a whole. The argument is tested by comparing the effectiveness of alternative regression equations in explaining the investment behavior indicated by new estimates of railroad capital formation prepared for this study. Tihe new estimates are designed to remedy the defects existing in those published by Ulmer [18] . The results of the testing may be summarized most easily by reference to the Kmenta-Williamson hypothesis that investment behavior in American railroads is best explained by models appropriate to the various phases of the industry's life cycle. The model they chose to explain investment behavior in the phase 1897-1914 is described by the equation I = a, + a2 (RX.K_.) where I = net investment in 1929 dollars (Ulmer's estimates), X = output in 1929 dollars, and K = net capital stock (first of year). One conclusion of this paper is that this particular model is inferior to a model emphasizing the costs of financing investment. Of more general significance is the fact that models of investment behavior which incorporate financial variables and which have performed satisfactorily for more recent periods in the railroad and other industries also do quite well in the period 1897-1914. This disputes the notion that it is more efficacious to take separate phases of industry life cycles in order to explain investment behavior.

T HE basic purpose of this paper is to show that an improved understanding of longswing mechanisms in economic-demographic interactions may be attained with cross-spectral analysis even though spectral analysis has, to date, cast only doubt on the existence of such long swings. The superiority of crossspectral analysis to simple estimation of power spectra stems from its emphasis upon examining relationships among economic and other variables (Granger, 1966). For exploration of the long-swing hypothesis, spectral analysis suffers from a further disadvantage its results are sensitive to the form of the data analyzed (levels, rates of growth, or deviations from trend), while cross-spectral analysis is not (see section II). Application of both spectral and cross-spectral analysis to economic and demographic data in Sweden, the United Kingdom, and the United States indicates (1) results of spectral analysis are generally similar across the three countries and negative with respect to the long swing hypothesis; and (2) cross-spectral analysis shows quite different long-swing mechanisms at work while giving some positive confirmation to the Easterlin model of long swings for the United States.