The authors gratefully acknowledge the helpful suggestions of Henry J. Aaron, Richard E. Caves, John E. Jackson, John F. Kain, and William E. McAuliffe. Special thanks are due to Susan E. P. Bloom for her continual assistance throughout the development of the paper.
1 Kramer, Gerald H., “Short-Term Fluctuations in U.S. Voting Behavior, 1896-1964,” APSR, 65 (03, 1971), 131–143 .
2 Stigler, George J., “Micropolitics and Macroeconomics: General Economic Conditions and National Elections,” American Economic Review, 63 (05, 1973), 160–167 .
3 Arcelus, Francisco and Meltzer, Allan H., “The Effect of Aggregate Economic Variables on Congressional Elections,” American Political Science Review, 69 (12, 1975), 1232–1239 .
4 Jordan, Nehemiah, “The Asymmetry of Liking and Disliking. A Phenomenon Meriting Further Reflection and Research,” Public Opinion Quarterly, 29 (Summer, 1965), 315-22, at. p. 315 .
5 Campbell, Angus, Converse, Philip E., Miller, Warren E. and Stokes, Donald E., The American Voter (New York: John Wiley and Sons, 1960), pp. 554–555 . In addition, John Mueller presents evidence of this asymmetry with respect to the impact of an “economic slump” on presidential popularity. See Mueller, John E., War, Presidents and Public Opinion (New York: John Wiley and Sons, 1973), pp. 213-16.
6 Campbell, Angus, Converse, Philip E., Miller, Warren E., and Stokes, Donald E., The American Voter, p. 387 .
7 Similar evidence is provided by results of Gallup Polls taken during the 1937-38 recession. Approximately 63 per cent of the respondents indicated that they had noticed a “decline in business” in their community during the past two months. Of those who noticed a decline, 58 per cent indicated that they held the current Roosevelt Administration at least partly to blame for it. The variation in response by party, however, was quite striking: only 37 per cent of the Democratic respondents held the administration at least partly to blame, whereas 89 per cent of the Republican respondents did so. Gallup, George, The Gallup Poll: Public Opinion 1935-71 (New York: Random House, 1973), p. 78 .
8 Campbell, Angus, Converse, Philip E., Miller, Warren E., and Stokes, Donald E., The American Voter, p. 391 .
9 Converse, Philip E., “The Concept of a Normal Vote” in Elections and the Political Order, Campbell, Angus, Converse, Philip E., Miller, Warren E., and Stokes, Donald E. (New York: John Wiley and Sons, 1966), pp. 9–39 .
10 Campbell, Angus, Converse, Philip E., Miller, Warren E., and Stokes, Donald E., The American Voter, Figure 7.1, p. 154 .
11 In 35 of the 37 elections studied, the major two parties obtained 94 per cent or more of the total vote. Their median percentage during this period was 97 per cent. Details of the effect of including third party votes are discussed later.
12 Philip E. Converse, “The Concept of a Normal Vote.”
13 One of the earliest attempts to measure party ID from aggregate data is: Stokes, Donald E., “Party Loyalty and the Likelihood of Deviating Elections,” Journal of Politics (11, 1962), 689–702 . Stokes estimates the division of party loyalty by taking the mean vote in all presidential elections from 1892-1960, and experiments with breaking this into subperiods for 1892-1928 and 1932-60.
14 Tufte, Edward, “a,” American Political Science Review, 69 (09, 1975), 812–826 .
15 We recognize that individuals, regardless of their true attitudes toward each party, may have incentives to register with the locally dominant party. We do not feel, however, that this poses a serious problem for our analysis because (1) we use aggregate statewide data, in which specific local biases are probably largely cancelled out; (2) remaining net systematic biases will be picked up in the intercepts of our regressions and thus will not affect our coefficient estimates; and (3) remaining net random biases will be picked up in the disturbance terms of our regressions and thus will not bias our coefficient estimates. For an extensive analysis of registration changes over time, see Sundquist, James, Dynamics of the Party System (Washington: Brookings Institution, 1973), pp. 204–233 .
16 Stigler, George J., “Micropolitics and Macroeconomics,” p. 163 .
18 For a general discussion of the effect of omitted variables on the estimates of regression coefficients, see Rao, Potluri and Miller, Roger L., Applied Econometrics (Belmont, California: Wadsworth, 1971), pp. 60–67 .
19 Because VOTE is defined in terms of the two-party vote, the percentage of the vote lost or gained by Republicans exactly equals the percentage gained or lost by Democrats.
20 The two major parties received only 80 per cent of the congressional vote.
21 The difference between the significance of the results for the two subsamples is even more striking if one considers that the increasing income estimates are based on almost twice as many observations as the declining income estimates.
22 This point is demonstrated by results of the regression of VOTE on ID only. For economic upturns, ID “predicts” 61 per cent of the variation in VOTE. In other words, past trends “predict” the current vote fairly well. On the other hand, during economic downturns, ID “predicts” only 9 per cent of the variation in VOTE, since economic conditions cause voters to deviate from their past behavior.
23 Equation 2 is equivalent to Equation 1 with the value of the coefficient for ID constrained to equal one. This constraint is consistent with our inclusion of REG in the model to represent the “expected” vote, around which short-run economic conditions cause the observed vote to fluctuate. Empirical justification for this assumption is provided by the fact that the coefficients for ID in the appropriately specified asymmetric estimates of Equation 1 are not significantly different from one.
24 Results of the model (Equation 2) for elections preceded by rising income were subjected to the following sensitivity test: The rising income subsample was split in two, with one half containing the elections preceded by the largest income increases and the other half containing the elections preceded by the smallest income increases. Equation 2 was then estimated separately from each half of the rising income subsample. The results for both estimates were quite similar to those reported for the rising income subsample as a whole.
25 The state estimates do not provide a completely independent replication of the national analysis. The two analyses are linked to the extent that the characteristics of the three states are imbedded in the national data. If the model had been estimated from data for the nation minus the three states, the results would have been completely independent from the state-level analysis. This was not desirable, however, since the state time-series are shorter than the national ones. Nevertheless, the interdependency between the two levels of analysis is probably not serious since the three states represent only 25 per cent and 28 per cent of the total national vote in the 1930 and 1970 congressional elections, respectively. Also, state estimates cover little over half the time period included in the national analysis.
26 See Ackley, Gardner, Macroeconomic Theory (New York: Macmillan, 1961), pp. 196–198 .
27 See Appendix A for sources of state income data published by the Department of Commerce for 1929 to the present. For 1919-21 see: Leven, Maurice, Income In the Various Slates, Its Sources and Distribution, 1919, 1920 and 1921 (New York: National Bureau of Economic Research, 1925). Note that the Leven data are not strictly comparable to the Department of Commerce data.
28 “Personal Income in Metropolitan and Non-metropolitan Areas,” Survey of Current Business, 50 (05, 1970), 22–36 .
B1 Oregon increased from two to four House districts during this period.
B2 The unadjusted R 2 for the California equation is 0.75. Because they incorporate degrees of freedom, the adjusted R 2's shown in the table are underestimates of the actual proportion of the registration variation “predicted” by each regression. For further discussion of adjusted R 2, see Johnston, J., Econometric Methods, 2nd edition (New York: McGraw-Hill, 1972), pp. 129–130 .
B3 The standard errors of estimate of the regressions for each state were statistically significantly different from each other, according to the appropriate F-fest. Thus to obtain the most efficient estimators, observations for each state in the pooled sample were weighted in inverse proportion to the standard error of the separate regression for that state.
B4 Estimates also indicate that the weights attached to additional elections decrease rapidly as the time between the election and registration figures increases.
* The authors gratefully acknowledge the helpful suggestions of Henry J. Aaron, Richard E. Caves, John E. Jackson, John F. Kain, and William E. McAuliffe. Special thanks are due to Susan E. P. Bloom for her continual assistance throughout the development of the paper.
Email your librarian or administrator to recommend adding this journal to your organisation's collection.