In at least two ways, the outcomes of the 2012 presidential election across the American states were quite predictable. First, estimates based on pooling publicly available polls in advance of election day were extremely accurate.1 Second, as shown in Figure 1, the correlation between Obama’s vote margins across the states in 2012 and 2008 was nearly perfect, at 0.98.2 However, what neither of these approaches provides is an explanation for why Obama did better in some states while Romney performed better in others.
To explain variation in outcomes across states, I rely on a standard spatial voting model applied to the contemporary American party system that features two major parties and candidates with contrasting positions on economic and cultural policy. Empirically, I proceed in three steps. First, relying on a host of public opinion surveys, I locate the states on the economic and cultural policy dimensions. Second, I relate those positions to the 2012 presidential election results. Third, I replicate the analysis for each presidential election back to 1972 in order to identify long-term trends in the determinants of outcomes.
The results suggest that the state outcomes in 2012 (and 2008) were more strongly driven by state differences in cultural policy locations than economic ones. Moreover, the importance of cultural issues for producing variation in outcomes across the states has been steadily growing since 1976. In an absolute sense, economic issues have not receded in importance, but relative to cultural issues, they have. Thus the “party cleavage” or “cutting” line separating the parties has steadily rotated over time.
Parties, Policy, and Issue Voting
Most accounts of national party politics in the contemporary US note that among political elites and candidates for elective office, there are substantial ideological and policy differences between Democrats and Republicans, and that these have been growing larger in recent decades (McCarty Poole, and Rosenthal 2006; Fiorina and Abrams 2009; Abramowitz 2010; Layman et al. 2010; Aldrich 2011; Fiorina, Abrams, and Pope 2011; Goren 2012).3 Even if the extremism of the major parties falls short of the extremism observed in democracies with many parties, the proposition that the US has a liberal party and a conservative one is generally accepted.
On the longstanding divide over economic or social welfare issues, Democrats support governmental activism to provide a more extensive and generous social safety net and economic redistribution, while Republicans prefer less governmental involvement, a greater reliance on market forces, and are more supportive of business interests. On the cultural or moral dimension, the Democratic Party is more “progressive” or “liberal,” while the Republican party is more “traditional” or “conservative.” Apart from labels, on the two issues most commonly associated with the cultural dimension in recent American politics, Democrats have become more supportive of abortion and gay rights; Republicans prefer more restrictive abortion policy and are less inclined to support policies like gay marriage.
The implications for voting behavior of divergent parties and candidates on the economic and cultural policy dimensions follow from the spatial model of voting. Suppose that a voter cares about policy in both policy domains and that her preference for one party’s candidate over the other party’s candidate is influenced by her relative proximity to the candidates’ positions. The closer a voter is to one candidate relative to the other, the more likely that she will vote for that candidate. Across the electorate, where voters are arrayed along both dimensions, differences in voting behavior will manifest to the extent that the candidates’ positions differ. For example, if there is no difference between the candidates on the cultural dimension, then culturally liberal and culturally conservative voters would be equally distant from both candidates on this dimension and differences in cultural policy preferences across voters would not be correlated with ballot choices. But, if in subsequent elections the candidates’ positions grow increasingly different, the more liberal candidates will develop a growing proximity advantage with culturally liberal voters as more conservative candidates build an advantage with culturally conservative voters. A growing correlation between voters’ cultural policy preferences and ballot choices will become evident.4
At the level of state electorates, differences in outcomes are produced as a result of the individual-level connections between issue preferences and voting and the geographic distribution of voter preferences across the states (Ansolabehere, Rodden, and Snyder 2006). Assessing the strength of the relationships is a two-step process. First, separate measures of state locations on each issue dimension must be constructed. Second, the locations must be used as predictors of state presidential outcomes.
State Issue Positions on Economics and Cultural Policy
To measure state policy locations, I rely on seven datasets that together include 158 issue or issue-related items. The General Social Survey (1972–2006) includes 21 items. The Pooled Senate Election Study (1988–1992) includes 23 items. The 2000, 2004, and 2008 National Annenberg Election Studies have 48, 24, and 10 questions, respectively. And, the 2006 and 2008 Common Content Cooperative Congressional Election Studies both have 16 items. For each item, I computed the average response by state (pooling across years for the GSS and PSES). Because the sampling frames of the surveys did not always include all the states, there is missing data for some items in some states.5 To address and assess the effects of missing data, I use Honaker, King, and Blackwell’s Amelia II to impute values for the missing data.6
With the complete dataset, I use principal factor analysis to assess the dimensionality of the data and create policy scales. A screeplot of the eigenvalues (Figure A2) strongly suggests a two-factor model. Because there is no substantive reason why the dimensions should be uncorrelated, I allow them to be correlated by using a promax rotation. Table 1 lists the indicators with the highest rotated factor loadings for the two dimensions. Clearly there are separate “cultural” and “economic” dimensions. On the cultural dimension, items about Christian fundamentalists, school prayer, abortion, and gay rights are among those with the highest loadings. On the economic dimension, items about health care, housing, and aid for children and the uninsured load highly.
|Patient rights||NAES 2000||0.92|
|Government vs. business||GSS||0.91|
|Spending – social security||NAES 2000||0.86|
|Prescription drugs||NAES 2000||0.85|
|Spending – medicare||NAES 2000||0.82|
|Universal healthcare||NAES 2000||0.81|
|Environmental protection||NAES 2000||0.81|
|Government housing assistance||CCES 2008||0.80|
|Health insurance for children||CCES 2008||0.79|
|Spending – uninsured||NAES 2000||0.7|
|Christian fundamentalists||NAES 2000||0.97|
|Christian groups||NAES 2004||0.96|
|School prayer||NAES 2000||0.92|
|Same-sex marriage||NAES 2008||0.91|
|Tolerance of gays||GSS||0.91|
|Abortion ban||NAES 2000||0.90|
|Underpunished criminals||NAES 2000||0.89|
|Abortion restrictions||NAES 2000||0.89|
|Spending – defense||PSES||0.89|
Shifting from issue items to states, the factor scores on the two dimensions are consistent with conventional notions of states’ locations. Based on the factor analysis, the most culturally conservative states are Mississippi, Alabama, Arkansas, West Virginia, and Louisiana. The least conservative (most liberal) are Rhode Island, Hawaii, Massachusetts, Vermont, and the District of Columbia. On the economic dimension, the most conservative are Wyoming, Idaho, Utah, Nebraska, and Montana while the least conservative are Massachusetts, Hawaii, Rhode Island, New York, and the District of Columbia.
Across all the states, the correlation between scores on the two dimensions is 0.30. If the District of Columbia – a liberal outlier on both dimensions – is excluded, the correlation is reduced to just 0.16. Figure 2A (all states including the District of Columbia) and 2B (excluding the District of Columbia) show the joint distribution of state economic and cultural issue locations.
Explaining State Presidential Outcomes in 2012
To measure the relationships between state issue locations and voting in the 2012 presidential election, I relate the former to the latter through a series of OLS regression models. In every model, the dependent variable is Obama’s vote margin: Obama’s vote percentage minus Romney’s vote percentage. This variable is regressed on the measures of state economic and cultural issue locations.7 Rather than estimate a single model, I estimate several in order (a) to assess whether the historically distinctive nature of the South – defined as the 11 former Confederate states – persists and influences the estimates, and (b) to determine the effects on the parameter estimates of the extreme values for the District of Columbia. In all, I estimate six models, though as demonstrated below, there is not much variability in estimates across them.
Model 1 is based on all states and the District of Columbia and includes the policy measures as independent variables. Model 2 adds an indicator for southern states, and Model 3 is estimated for non-southern states only. Models 4–6 are the same as Models 1–3, respectively, except that the District of Columbia is excluded. All the parameter estimates, standard errors, and fit statistics are reported in the Appendix. The key results are displayed in Figure 3, which shows for each model the estimated difference in Obama’s vote margin between a liberal (75th percentile) and conservative (25th percentile) state for each issue domain.
Across all the models for both issues there are sizable substantive effects.8 The estimated differences in Obama’s vote margin in an economically liberal state compared to an economically conservative one range from 15.1 (Model 6) to 16.7 (Model 1) percentage points, with an average across the six models of 15.8 points. The estimates are larger for culturally liberal versus culturally conservative states, ranging from 20.1 (Model 4) to 23.4 (Model 3) percentage points, with an average of 22.0 points.
Simply put, differences across the states with regard to economic and cultural policy preferences are associated with substantially different electoral outcomes, with the effect of cultural preferences averaging about 40% larger in magnitude than the effect of economic preferences. Together state policy locations do an excellent job of explaining vote margins in 2012, accounting for between 87 and 90% of variance, depending on the model.
State Presidential Outcomes from 1972 to 2012
A useful way to put the estimates for 2012 in perspective is to compare them to estimates for earlier election years. Doing this makes it possible to determine whether there were unique features of the 2012 presidential election along with identifying long-term trends – perhaps shifting “party cleavage lines” (Miller and Schofield 2003, 2008) – that are impossible to observe with estimates from a single election. As a preliminary step, consider the correlations between election results from one election year to the next. Figure 1 has already shown remarkable continuity between state vote margins in 2012 and 2008. Figure 4 displays the correlations (including and excluding the District of Columbia) between successive pairs of elections.9 Clearly the 1972 election is distinctive. The correlations between the 1972 and 1968 outcomes and, especially, the 1976 and 1972 ones are notably lower than the correlations across the rest of the election pairs. Beginning with the 1980 election, there is much more election-to-election continuity and it steadily increases over time. The increase is from a correlation of 0.89 in 1980 to 0.98 in 2012 when the District of Columbia is included. When it is excluded, the increase is from 0.83 to 0.98.
The continuity in state outcomes over time masks a transformation in how state issue locations relate to those outcomes. To see this, I treat state issue positions as fixed and regress state presidential margins on them for every presidential election from 1972 through 2008, in each case estimating the same six models as I did for 2012.10 Figure 5 presents the results, which shows the estimated differences in Democratic vote margins between liberal (75th percentile) and conservative (25th percentile) states for each issue in each election year.
With the estimates for all 11 presidential elections from 1972 through 2012, several important findings become apparent. First, while the differences in outcomes associated with state cultural policy preferences are larger in 2012 than in any of the other election years, they clearly represent the continuation of a long-term trend that extends over nearly 40 years. From 1976 through 2012 there has been a steady – almost election-by-election – increase in the relationship between state cultural policy preferences and presidential vote margins.11
A second important result is that while the influence of state cultural preferences has steadily increased, the influence of economic preferences has not. Over the 11 elections, the average effect of state economic preferences has ranged from about 5 percentage points (1972) to nearly 20 points (2000), with no long-term trend evident.12 As a result, the effect of economic policy relative to cultural policy has been declining. The magnitude of the cultural policy effect estimate trailed that for economic policy from 1976 through 2000, was roughly equal to it in 2004, and surpassed it in 2008 and then again in 2012.
The 1972 presidential election is also worthy of note as – among other things – a very leading indicator of the direction of American electoral politics. The average estimated effect of cultural policy on state presidential voting margins in 1972 is 11.6 points and the average estimated effect of economic policy is 5.1 points. Not only does this election match the 2008 and 2012 elections as the only ones where the estimated cultural policy effect exceeds that of the economic issue effect, but in relative terms, the cultural effect in 1972 – which was more than double the economic issue effect in 1972 – has not yet been matched.
A clear way to observe the evolution in how state issue locations relate to differences in election outcomes is by focusing on the dividing line between so-called Blue (Democratic) states and Red (Republican) states. Specifically, the yearly parameter estimates may be used to produce a set of state economic and cultural locations where the estimated presidential margin is predicted to be the average state margin in that year.13 For each year, the set of locations defines a line, commonly referred to as the “party cleavage” or “cutting” line. With a Democratic candidate’s economic position to the left of the Republican candidate’s economic position and with economic locations ranging from liberal to conservative on the x-axis (as in Figure 2), a vertical party cleavage line suggests Democratic and Republican states as being divided exclusively along the economic policy dimension. States to the left of the line are expected to provide a Democratic vote margin that exceeds the average Democratic vote margin while those to the right of the vertical line are expected to provide a Democratic margin that falls short of the average state Democratic margin. With cultural positions on the y-axis ranging from liberal to conservative (also as in Figure 2) as the cutting line rotates counter-clockwise, the cultural dimension is increasingly engaged with state electorates responding as if the Democratic candidate is more culturally liberal and the Republican candidate is more culturally conservative.
Turning from the abstract to the actual estimated party cleavage lines, Figure 6 shows the estimated cutting lines for 1972 and four groups of elections, 1976–1984 (Carter/Reagan), 1988–1996 (H.W. Bush/Clinton), 2000–2004 (W. Bush), and 2008–2012 (Obama).14 The near-vertical party cleavage line for the Carter and Reagan elections suggests states were divided primarily based on differences on the economic policy dimension. With each successive set of elections the cutting line rotated counterclockwise, indicating the growing role of the cultural dimension relative to the economic dimension for dividing the states.
The H.W. Bush and Clinton elections engaged the cultural dimension more than the Carter and Reagan elections, and the W. Bush elections engaged it still more. Finally, the Obama elections, while still engaging the economic dimension, engaged the cultural dimension relative to the economic dimension even more than the W. Bush elections. Notably, the cutting line for the Obama elections is not as flat as that for the 1972 election, indicating that cultural dimension was engaged even more strongly in the contest between Nixon and McGovern than in the most recent presidential elections.
Discussion and Conclusion
Differences across the American states with regard to preferences on cultural policy were more strongly related to presidential outcomes in 2012 (and 2008) than differences in preferences on economic policy. The multi-decade trend revealed in this paper (Figure 5) suggests that while the Obama elections were the first since 1972 where cultural issues differentiated states more than economic ones, they were less about unique features of those elections (e.g., the presence of a major-party African-American candidate) than about an ongoing transformation in the nature of party locations and cleavages in the US.15
The findings reported in this essay help to assess and put into perspective claims in other recent analyses of state presidential outcomes. Consider Ansolabehere, Rodden, and Snyder (2006), which finds that economic issues matter as much or more than cultural ones for explaining interstate variation in presidential election outcomes. When the time-period on which the claim is based (1992–2000) is taken into account, the finding fits with what is reported here; it is only in subsequent elections that cultural policy preferences matter more than economic preferences for sorting the states. Thus Ansolabehere, Rodden, and Snyder (2006) analyzed the American electorate while it was in the midst of the long-term changes identified in this essay. In the presidential years before the Ansolabehere, Rodden, and Snyder (2006) analysis (1976–1992), cultural issue differences mattered even less for interstate differences in outcomes than for the period considered by Ansolabehere, Rodden, and Snyder (2006).
In another line of research, Miller and Schofield have offered a theoretical account of party positioning and interpretation of the evolution of party and candidate issue positions to explain cross-sectional and over-time differences in state presidential election outcomes (Miller and Schofield 2003, 2008; Schofield and Miller 2007). Key findings reported in this article are consistent with claims by Miller and Schofield, especially with regard to the distinctiveness of the 1972 presidential election and the ongoing “redefinition of party cleavages” [Miller and Schofield (2003), p. 246]. Where the accounts differ is with regard to role of economic policy preferences for producing variation in outcomes across the states.
As interparty and intercandidate differences in cultural policy locations have increasingly grown, Miller and Schofield see the possibility of the parties moving closer on the economic policy dimension, as some economic conservatives who are culturally liberal migrate to the Democratic party and some economic liberals who are culturally conservative migrate to the Republican party. However, if this process was going on, then the relationship between interstate differences in economic policy preferences and state presidential outcomes would have declined as differences in electoral outcomes associated with state cultural preferences increased. Yet this has not been the case, as shown in Figure 5. The party cleavage line has rotated due to increasing cultural effects, not declining economic ones, which appears more consistent with the “conflict extension” theory advanced by Layman and his collaborators (Layman and Carsey 2002; Layman et al. 2010).
To conclude, the results of the 2012 presidential election fit nicely with long-term patterns that began after the distinctive 1972 presidential election. Increasingly, states are sorted into “Red” and “Blue” based on cultural policy preferences. Economic preferences continue to sort the states, too, but relative to cultural ones there is a long-term trend toward their declining significance.
I. Missing Data
For the measures of state policy preferences, Table A1 shows that eleven states had at least one instance of missing data, five states had missing data for more than 10% of the items, and Alaska and Hawaii had missing data for more than half the items. As described in the main text I used Amelia II to impute 500 values for each instance of missing data. To observe the variability in the state locations produced by the imputation process, I factor analyzed each of the 500 datasets produced by Amelia II. Figure A1 displays the mean factor score for each state along with ± two standard deviations on each dimension. The vertical lines depict the range of scores on the cultural dimension and the horizontal ones depict the range for economic preferences. The five states with the most missing data are identified. Clearly the locations produced vary across the imputations, but not so much as to alter fundamentally the relative locations of the states, even in the cases of Alaska and Hawaii.
|State||# of items missing||% of items missing|
|ME, NM, NH, ID, UT, RI||1||0.6|
II. Dimensionality of State Policy Preferences
After addressing the issue of missing data, I used principal factor factor analysis with a promax rotation to assess and identify the dimensions of state policy preferences. Two clear dimensions emerged (Figure A2 is a screeplot of eigenvalues) that were modestly correlated. Across all the states, the correlation between scores on the two dimensions is 0.30. If the District of Columbia – a liberal outlier on both dimensions – is excluded, the correlation is reduced to just 0.16. See text for more details, including Table 1, which shows the items with the largest factor loadings on each dimension.
III. Models of State Presidential Election Results
The main text describes the six models of state presidential election results. Table A2 reports the full regression results for the six models of state presidential election results, which are summarized in Figure 3.
|State economic policy preferences||12.8a (1.1)||12.0a (1.2)||11.8a (1.3)||12.4a (1.2)||11.8a (1.3)||11.5a (1.4)|
|State cultural policy preferences||14.8a (1.1)||16.4a (1.4)||16.9a (1.7)||14.5a (1.2)||16.1a (1.5)||16.7a (1.8)|
|Southern state||–5.7b (3.3)||–5.5b (3.4)|
|Constant||0.2 (1.1)||1.4 (1.3)||1.6 (1.4)||0.4 (1.1)||1.5 (1.3)||1.7 (1.4)|
|Includes southern states?||Yes||Yes||No||Yes||Yes||No|
Abramowitz, Alan I. 2010. The Disappearing Center: Engaged Citizens, Polarization, and American Democracy. New Haven, CT: Yale University Press.
Aldrich, John H. 2011. Why Parties? A Second Look. Chicago, IL: University of Chicago Press.
Ansolabehere, Stephen, Jonathan Rodden, and James M. Snyder, Jr. 2006. “Purple America.” Journal of Economic Perspectives 20 (2): 97–118. [CrossRef]
Bawn, Kathleen, Martin Cohen, David Karol, Seth Masket, Hans Noel, and John Zaller. 2012. “A Theory of Political Parties, Groups, Policy Demands and Nominations in American Politics.” Perspectives on Politics 10 (3): 571–597.
Cook, Rhodes, Alice V. McGillivray, and Richard M. Scammon. 2010. American Votes 28: Election Returns by State. Washington, DC: CQ Press.
Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper & Row.
Erikson, Robert S., Gerald C. Wright, and John P. McIver. 2006. “Public Opinion in the States: A Quarter Century of Change and Stability.” In Public Opinion in State Politics, edited by J. Cohen. Palo Alto, CA: Stanford University Press.
Fiorina, Morris P., with Samuel J. Abrams. 2009. Disconnect: The Breakdown of Representation in American Politics. Norman, OK: University of Oklahoma Press.
Fiorina, Morris P., with Samuel J. Abrams and Jeremy C. Pope. 2011. Culture War? The Myth of a Polarized America. 3rd ed. New York, NY: Longman.
Goren, Paul. 2012. On Voter Competence. New York, NY: Oxford University Press.
Layman, Geoffrey C., and Thomas Carsey. 2002. “Party Polarization and ‘Conflict Extension’ in the American Electorate.” American Journal of Political Science 46 (October): 786–802. [CrossRef]
Layman, Geoffrey C., Thomas M. Carsey, John C. Green, Richard Herrera, and Rosalyn Cooperman. 2010. “Activists and Conflict Extension in American Party Politics.” American Political Science Review 104 (2): 324–346. [CrossRef] [Web of Science]
McCarty, Nolan, Keith T. Poole, and Howard Rosenthal. 2006. Polarized America: The Dance of Ideology and Unequal Riches. Cambridge: MIT Press.
Miller, Gary, and Norman Schofield. 2003. “Activists and Partisan Realignment in the United States.” American Political Science Review 97 (2): 245–260.
Miller, Gary, and Norman Schofield. 2008. “The Transformation of the Republican and Democratic Party Coalitions.” Perspectives on Politics 6 (3): 433–450. [Web of Science]
Schofield, Norman, and Gary Miller. 2007. “Elections and Activist Coalitions in the United States.” American Journal of Political Science 51 (3): 518–531. [CrossRef]