^* Revised version of a paper presented at the annual meeting of the American Association for the Advancement of Science, Chicago, Ill., Dec. 29, 1959. The authors are indebted to the Social Science Research Council for financial assistance, and to David H. Johns, Robert W. Kern, and the staff of the Operations Analysis Laboratory and the National Opinion Research Center of the University of Chicago for their aid in statistical work involved. Professor Samuel K. Gove of the University of Illinois kindly supplied some of the data later published in his volume, Illinois Votes: 1900–1958 (Urbana, 1959), to supplement our computations.

¹ How to Predict Elections (New York, 1948), ch. 10.

² Lubell, Samuel, The Future of American Politics (New York, 1952), p. 35.Google Scholar

³ Key, V. O. Jr., “A Theory of Critical Elections,” Journal of Politics, Vol. 17 (1955), pp. 1–18.CrossRefGoogle Scholar

⁴ Ibid., p. 4.

⁵ The value of cooperation between history and other social sciences has been pointed out, for example, by Challener, R. D. and Lee, M. Jr., in “History and the Social Sciences: The Problem of Communications,” American Historical Review, Vol. 46 (1956), pp. 331–338 CrossRefGoogle Scholar; and by Benson, Lee in “Research Problems in American Political Historiography,” in Komarovsky, M. (ed.), Common Frontiers of the Social Sciences (Glencoe, Ill., 1957), pp. 113–183.Google Scholar

⁶ Gosnell, H. F., Grass Roots Politics (Washington, 1942).Google Scholar

⁷ Op. cit., pp. 5–6.

⁸ This method is used in Key, V. O. Jr., and Munger, F., “Social Determinism and Electoral Decision: The Case of Indiana,” in Burdick, E. and Brodbeck, A. J. (eds.), American Voting Behavior (Glencoe, 1959), p. 458 n. 23.Google Scholar It has also been demonstrated for a longer period in Blustein, Milton J., “Voting Tradition and Socio-Economic Factors in the 1936 Presidential Election in Illinois,” M. A. thesis, University of Chicago, 1950, p. 50.Google Scholar

⁹ Butler, D. E., The British General Election of 1955 (London, 1955), pp. 202–203.Google Scholar This value is low in comparison with the United States, but comparable with Illinois figures.

¹⁰ For evidence that Illinois participated only moderately in the 1896 reorientation, see Nixon, H. C., “The Cleavage Within the Farmers' Alliance Movement,” Mississippi Valley Historical Review, Vol. 15 (1928), pp. 22–23.CrossRefGoogle Scholar The dispersion of shifts at each of these two transitions may have been influenced by the presence of a third-party vote at the prior presidential election. Third-party votes will be considered in more detail below. The data presented in Figure 1 are intended only to show that Illinois manifests the same sort of reorientation as the nation.

¹¹ “A Theory of Critical Elections,” op. cit., p. 12.

¹² For discussion of the principles of factor analysis see, for example, Cattell, R. B., Factor Analysis (New York, 1952).Google Scholar The specific variant with which we are concerned here resembles the P-and O-techniques (Cattell, pp. 102–107). Application of these methods to analysis of election statistics has been proposed by Baggaley, A. R. in “Patterns of Voting Change in Wisconsin Counties, 1952–1957,” Western Political Quarterly, Vol. 12 (1959), pp. 141–144.CrossRefGoogle Scholar Principal component analysis, a closely related procedure, is discussed in Kendall, M. G., A Course in Multi-variate Analysis (New York, 1957).Google Scholar

¹³ In “Social Determinism and Electoral Decision: The Case of Indiana,” op. cit., Key and Munger point out the significance of traditional partisan alignments in Indiana counties over periods of thirty years or more. It appears that when electoral alignments change, as in a critical election, a large stable component of the party vote still remains.

¹⁴ It should be noted that weighting all counties equally in this analysis leads to a stress on rural areas. In principle this bias can be remedied by studying urban units as well, as Gosnell, has done in Machine Politics, Chicago Model (Chicago, 1937).Google Scholar

¹⁵ This procedure, carried out on the Univac, was initially proposed by Saunders, D. R. in “Practical Methods for the Direct Factor Analysis of Psychological Score Matrices,” Ph.D. dissertation, University of Illinois, 1951.Google Scholar The Univac program was initially written by M. Woodbury and W. Tuianski at Remington Rand Univac, and rewritten by F. K. Bamberger at the Operations Analysis Laboratory, University of Chicago. Mathematical details are given in an appendix to this paper.

¹⁶ Sources for these data were Blue Book of the State of Illinois (Springfield, biennial), and Official Vote of the Slate of Illinois Cast at the General Election (date), compiled by the Secretary of the State of Illinois (Springfield). In the elections of 1918 and 1920 the Democrats failed to nominate a Congressional candidate in one District; as a measure of the potential Democratic vote in those counties, data for Trustees and Congressman-at-large in the same year were substituted.

¹⁷ Wooddy, C. H., The Case of Frank L. Smith (Chicago, 1931), pp. 371–372.Google Scholar Our figure is further adjusted to reduce the effects of local popularity of one Trustee candidate in a county. The vote for Treasurer departed from that for other offices in 1904 and 1916, when Len Small was the Republican candidate.

¹⁸ This procedure is described in the appendix.

¹⁹ Sources for county characteristics are: foreign-born, Fourteenth Census of the United States Taken in the Year 1920 (Washington, 1922), Vol. II, pp. 1335–1336 Google Scholar; rural-farm, Fifteenth Census of the United States: 1930 (Washington, 1933), Vol. III, Pt. 1, pp. 614–620 Google Scholar; Catholic, Census of Religious Bodies: 1936 (Washington, 1941), Vol. I, pp. 743–744 Google Scholar; 1930 wet vote, Blue Book of the State of Illinois 1931–1932 (Springfield, 1931), pp. 852–853. Similar correlations are given by Blustein, op. cit., p. 36.

²⁰ Ogburn, W. F. and Talbot, N. S. found that in 1928 the prohibition issue was most closely correlated with the vote and Catholicism next, for 173 counties of the North and West: “A Measurement of Factors in the Presidential Election of 1928,” Social Forces, Vol. 8 (1929), pp. 175–183.CrossRefGoogle Scholar Religion may have been of relatively temporary significance, however, in relation to other variables shown in Table I. Blustein (op. cit., pp. 43, 50–51) shows that the Smith vote had a relatively low correlation with both earlier and later elections, supporting the notion that it involved temporary rather than lasting reorientations of the vote. It should also be noted that our correlations, weighting each county equally, treat the Catholic population as consisting disproportionately of German Catholics in rural counties.

²¹ The increase in the Republican percentage from 1912 to 1916 (R) correlated +.93 with the 1912 Progressive vote (P), and the regression slope of R on P was +.85; from this we infer that the 1912 Progressives largely returned to the Republican party in 1916. The appeal of Theodore Roosevelt to the urban Catholic vote in 1904 was referred to by Kohlsaat; see below, n. 38. See also Mowry, George E., Theodore Roosevelt and the Progressive Movement (Madison, 1946), p. 280.Google Scholar

²² Sullivan, a Catholic, was several times head of the Democratic delegation to the National convention, but was not universally popular with downstate Democrats. In 1914 he failed to carry White County, home of State Chairman A. W. Charles and a normally Democratic county. See Townsend, W. A. and Boeschenstein, C., Illinois Democracy (Springfield, 1935), vol. I, p. 319 Google Scholar and passim. Sullivan's part in the campaign of 1908 will also be discussed below.

²³ Biographical information on the Republicans involved appears in Wooddy, The Case of Frank L. Smith, op. cit., passim, and on the Democrats, in Townsend and Boeschenstein, Illinois Democracy, op. cit.

²⁴ See Wooddy, op. cit., pp. 159–162.

²⁵ Pro-wet sentiment was inferred from referendum figures in the Blue Book of the State of Illinois 1923–24 (Springfield, 1923), pp. 772–773. Percent Catholic was calculated from the Census of Religious Bodies: 1936 (loc. cit.).

²⁶ The regression slope of Jones minus Davis vote on La Follette vote is only +.42.

²⁷ For this calculation, the La Follette vote was corrected by subtracting the Farmer-Labor and Socialist votes in 1920 for each county, these having probably gone largely to La Follette.

²⁸ The fact that the Democrats were gaining strength during this period means that we need to look mainly for transitions toward a Democratic vote, rather than away from it.

²⁹ See Key, V. O. Jr., “Partisanship and County Office: The Case of Ohio,” this Review, Vol. 47 (1953), p. 526 Google Scholar; Moos, Malcolm, Presidents, Politics, and Coattails (Baltimore, 1952), pp. 13n., 110.Google Scholar

³⁰ Campbell, See A. and Miller, W. E., “The Motivational Basis of Straight and Split Ticket Voting,” this Review, Vol. 51 (1957), pp. 293–312.Google Scholar

³¹ See Lubell, op. cit., p. 141. The point is also supported by Bluetein, op. cit., pp. 50–54.

³² “A Theory of Critical Elections,” pp. 8, 10.

³³ The first election at which a vote occurred for Trustees of the University of Illinois, this election was also selected in order to allow a sufficient interval before 1896.

³⁴ The first principal component for 1914–58 correlates –.71 with per cent rural-farm in 1930 (Table II); while that for 1888–1922 correlates –.51 with a measure of ruralism by counties based on the 1890 census (“farms” divided by “farms” plus “homes”). Data for the latter measure were taken from Report on Farms and Homes: Proprietorship and Indebtedness in the United States at the Eleventh Census: 1890 (Washington, 1896), pp. 254–255, 333–334.

³⁵ Key, V. O. Jr., American State Politics, (New York, 1956), p. 224.Google Scholar

³⁶ This correlation does not mean, however, that “Bryanism” simply reflects traditional Democratic strength; when the average per cent Democratic (1888–1922) is controlled, the partial correlation between Bryanism and proportion of Southern migrants is still +.50. Bryanism is definitely correlated with Democratic strength (r = +.59), but Democratic strength in this period is less closely linked to Southern migration (r = +.35). Data on state of birth were calculated from Statistics of the Population of the United States at the Tenth Census (Washington, 1883), pp. 56–57, 504–505.

³⁷ The relation between Altgeld and the reform movements of the 1890s is discussed in Destler, Chester McA., American Radicalism: 1865–1901. Essays and Documents (New London, 1946), pp. 167–168, 189–190.Google Scholar

³⁸ Kohlsaat, H. H., From McKinley to Harding (New York, 1923) pp. 13–137 Google Scholar, wrote that in 1904 “the Catholics are 85 per cent Democratic, and when they change to the Republican ticket their influence is greatly felt. Roosevelt had over 2,500,000 plurality. It was the religious issue which largely contributed to his great victory.” The Bull Moose ticket in 1912 was also strong in urban areas.

³⁹ Townsend and Boeschenstein, op. cit. vol. I, p. 269. Sullivan also ran for U. S. Senate in 1914. Although Stevenson had been Bryan's running mate in 1900, he did not even mention this fact in his memoirs: Stevenson, Adlai E., Something of Men I Have Known (Chicago, 1909).Google Scholar

⁴⁰ Hutchinson, W. T., Lowden of Illinois (Chicago, 1957), vol. I, p. 258.Google Scholar

⁴¹ Among these is the response of the German vote to World Wars I and II.

⁴² This procedure has been used by M. Woodbury and W. Turanski in preliminary analyses of the Pennsylvania vote.

⁴³ Analysis of the vote for the Illinois House of Representatives (1902–1954) has shown that the variance of per cent Democratic among districts decreases in presidential years, to increase again in the off years. See Sawyer, J. and MacRae, D. Jr., “Game Theory and Cumulative Voting in Illinois,” Proceedings of the American Statistical Association, Social Statistics Section, 1959 Google Scholar (in press).

⁴⁴ Saunders, D. R., “Practical Methods in the Direct Factor Analysis of Psychological Score Matrices,” Ph.D. Thesis, University of Illinois, 1950.Google Scholar

⁴⁵ Hotelling, H., “Analysis of a Complex of Statistical Variables into Principal Components,” Journal of Educational Psychology, Vol. 23 (1933), pp. 417–441, 498–520CrossRefGoogle Scholar.

⁴⁶ The eigenvectors of the residual matrix are also eigenvectors of the initial matrix. See Bodewig, E., Matrix Calculus (Amsterdam, 1959), ch. 8.Google Scholar

⁴⁷ The eigenvalue measures the variance accounted for by the principal component; see Kendall, , A Course in Mullivariaie Analysis, p. 15.Google Scholar