Aggregate electoral change in United States presidential elections, 1828--1992: A reevaluation of realignment theory

Since the time V. O. Key introduced the notion of realignment theory, it has been the dominant conceptual picture for understanding the history of American electoral politics. Recently, however, that dominant picture has come under attack from critics who argue that the theory is predictively inaccurate or is no longer relevant in an era of dealignment and should, therefore, be abandoned. Rather than abandoning the theory, I argue that realignment theory is a good case study of inductive science gone bad. There has been too much emphasis on using aggregate electoral returns in an attempt to identify critical elections and then building a theoretical structure to explain the findings. I attempt to cut through this dense theoretical structure and place the theory on a more deductive footing in four ways. First, I clarify the concept by identifying the necessary and sufficient conditions for a realignment to occur. Second, I demonstrate that the major criticisms of realignment theory do not apply to a clarified, more deductive version of the theory. I argue that the realignment process is stochastic and that, as a result, realignments cannot occur with any regularity. I also argue that realignment and dealignment are not inversely related phenomena but can and do occur together. Third, I identify methodological problems with the major studies that provide evidence of the occurrence of realignments. These problems stem primarily from the fact that all of them conduct atheoretical searches through series of aggregate election returns, asking of each and every election, Was this a realignment? Once these problems are corrected, the conclusions drawn by these scholars about the timing of realignments are no longer supported by the evidence. Therefore, the theoretical structure built upon these facts is tenuous. Finally, because a hypothesis testing approach is warranted, I test the conventional wisdom about the timing of realignments in a more appropriate way, one that recognizes the special problems of using aggregate historical election returns: pooled time-series intervention analysis. Indeed, the method provides efficient tests of the hypotheses because it corrects for both autocorrelation and heteroskedasticity and reduces the amount of aggregation bias significantly.U of I OnlyETDs are only available to UIUC Users without author permissio