ch3

THE PARTISAN PATTERN of economic performance documented in chapter 2 raises a perplexing political question. Real income growth has historically been much stronger under Democratic presidents than under Republican presidents, especially for middle-class and poor people. So why do so many people—including middle-class and poor people—persist in voting for Republicans? Even allowing for some class bias in turnout, it is clear that most voters have been substantially worse off in economic terms under Republican presidents than under Democrats. Yet Republicans have won most of the presidential elections of the postwar era and received 50.7% of all the votes cast for major-party candidates from 1948 through 2012.² If we suppose that voters want, or should want, to use their power at the ballot box to advance their own economic fortunes or those of their fellow citizens, the fact that Republicans have won even a bare plurality of votes seems baffling.

First, voters are myopic, responding strongly to income growth in presidential election years but ignoring or forgetting most of the rest of the incumbent administration's record of economic performance. As Ostrogorski surmised more than a century ago, the American voter “sees only present advantages.” It is by no means obvious that this quirk of voter psychology should have had significant implications for the balance of power between Democrats and Republicans over the past several decades, but it has. That is the result of a striking disparity in the timing of income growth under Democratic and Republican presidents in the postwar era. The tabulations presented in figure 2.1 show that Democrats have generally presided over more robust income growth, especially for families of modest means. However, that partisan pattern of income growth has disappeared in presidential election years, with most families experiencing faster growth—and affluent families experiencing much faster growth—under Republican presidents than they have under Democratic presidents.

Second, election-year income growth for affluent families turns out to be much more consequential at the polls than income growth for middle-class and poor families. Voters—even middle- and low-income voters—seem to be much more sensitive to the economic fortunes of the affluent than to the fortunes of their own income class or, for that matter, the nation as a whole. Again, it is not obvious that this odd sensitivity should have had significant partisan implications, but it has. As we saw in figure 2.1, Republican presidents in the postwar era have presided over much more income growth for affluent families than for middle-class and poor families. Democrats, on the other hand, have presided over slightly less income growth for affluent families than for middle-class and poor families—and much less in presidential election years, when income growth translates most powerfully into electoral support. Thus, the fact that voters tend to respond much more strongly to the economic fortunes of affluent families than of middle-class and poor families has provided a substantial electoral boost to Republican incumbents and their successors.

One of the primary attractions of this perspective on electoral accountability is that it does not seem to require too much from ordinary citizens. According to another influential theorist of retrospective voting, Morris Fiorina, voters “typically have one comparatively hard bit of data: they know what life has been like during the incumbent's administration. They need not know the precise economic or foreign policies of the incumbent administration in order to see or feel the results of those policies. … In order to ascertain whether the incumbents have performed poorly or well, citizens need only calculate the changes in their own welfare.”⁶

Analysts have routinely treated this feature of the retrospective theory of political accountability as unproblematic. Individual voters' economic perceptions may reflect substantial partisan biases, rationalization, and sheer randomness, but the electorate as a whole is assumed to respond systematically and sensibly to actual economic experience under the incumbent administration.⁷

Is that assumption warranted? Figure 3.1 summarizes the relationship between cumulative income growth and the incumbent party's electoral fortunes in the 16 presidential elections from 1952 through 2012. Election years are arrayed along the horizontal axis based on the total growth in real disposable personal income per capita in the second, third, and fourth years of each administration—the years in which the incumbent president's policies would most plausibly have some impact on the state of the economy.¹⁰ The vertical axis shows the popular vote margin (in percentage points) for the incumbent party's presidential candidate.

Figure 3.1 displays a strong, though not overwhelming, connection between cumulative income growth and presidential election outcomes. On one hand, the incumbent party lost five of the six elections (1960, 1976, 1980, 1992, 2008, and 2012) in which cumulative income growth was less than 5%. On the other hand, the incumbent party's average vote margin was about 7 percentage points in the five elections in which cumulative income growth exceeded 10%. However, the range of outcomes in those elections reflects the imprecision of the overall relationship between cumulative income growth and voting behavior. Incumbent presidents won reelection by landslides in two of those cases (1964 and 1972), but incumbent vice presidents were narrowly defeated in two others (1968 and 2000, albeit with a slim popular vote plurality in the latter case), and in 1952 the incumbent party's candidate was trounced despite much-higher-than-average cumulative income growth.

By comparison, the relationship depicted in figure 3.2 is a good deal stronger and more consistent. The difference here is that elections are arrayed along the horizontal axis not on the basis of cumulative income growth over the president's second, third, and fourth years in office but rather on the basis of income growth in the presidential election year only. Knowing how the economy fared during the election year turns out to be much more helpful in accounting for the incumbent party's fortunes than knowing how it fared over the entire three years leading up to the election. There are still substantial deviations from the overall pattern in specific elections. For example, incumbent presidents (or their successors) did much better than election-year income growth alone would have suggested in 1956, 1964, 1972, and 1996, and significantly worse in 1952, 1968, 1992, and 2000. However, the average deviation is almost 25% smaller than in figure 3.1.

The analyses reported in the third and fourth columns of table 3.1 include an additional explanatory factor—the incumbent party's tenure in office. There is a fairly strong tendency for the incumbent party's electoral fortunes to decline with each additional year that it has held the White House. Presumably this pattern reflects the cumulative effect of exhausted policy agendas, personnel turnover, and accumulating scandals on voters' desire for a change in leadership. Over the course of a typical four-year term, these forces reduced the incumbent party's popular vote margin by about six percentage points. However, allowing for this negative effect of tenure on the incumbent party's electoral fortunes leaves the comparison of long-term and short-term economic factors essentially unchanged. The results presented in the third column of table 3.1 suggest that each additional percentage point of cumulative income growth translated into a gain of about 1.7 percentage points in the incumbent party's expected vote margin, whereas those in the fourth column suggest that each additional percentage point of election-year income growth boosted the incumbent party's vote margin by more than four percentage points. Again, the analysis focusing on election-year income growth does a significantly better job of accounting for actual election outcomes, with an average error of less than six percentage points. The most notable outliers in figure 3.2 are all much better accounted for in this version of the analysis (since 1956, 1964, 1972, and 1996 were all elections in which the incumbent party had held office for just four years, while 1952, 1968, 1992, and 2000 were all cases in which the incumbent party had held office for eight years or more).

These results shed important light on the nature of the economic accountability provided by presidential elections. Election outcomes are sensitive to real changes in the nation's economic fortunes under the incumbent party—but only if those changes occur in close proximity to Election Day. Voters reward their elected leaders for some good times and punish them for some bad times. Does this myopic focus on “present advantages” have significant political consequences?

More than a quarter-century ago, political scientist Edward Tufte noted that the electorate's short time horizon with respect to economic evaluations could produce “a bias toward policies with immediate, highly visible benefits and deferred, hidden costs—myopic policies for myopic voters.” Tufte worried that political manipulation of economic policy could generate significant social costs due to wasteful government spending and other forms of “economic instability and inefficiency” aimed at making the economy flourish around election time. He provided statistical evidence of electoral cycles in transfer payments, income growth, unemployment, and inflation, as well as considerable qualitative evidence of specific efforts by incumbents to produce those cycles. Richard Nixon in 1972 was a particularly energetic manipulator of everything from the money supply (through his erstwhile political ally Arthur Burns, the chairman of the Federal Reserve Board) to effective dates of increases in Social Security benefits and payroll taxes; as Tufte delicately put it, “The extremes of 1972 were special because Richard Nixon was special.”¹²

Subsequent research on political business cycles has produced less clear results. According to one observer, “while the general logic behind the theory is quite persuasive, the empirical evidence for electoral-economic cycles is spotty at best.”¹³ Figure 3.3 redeploys the Census Bureau's figures on income growth from chapters 1 and 2 to explore whether presidents have produced unusual income growth in election years. Separately for families at the 20th, 50th, and 95th percentiles of the income distribution, the figure shows average levels of real income growth in each year of the election cycle, beginning with the year after a new (or reelected) president is inaugurated and continuing through the subsequent inauguration year.¹⁴

FIGURE 3.3 Electoral Cycles in Income Growth Under Republican and Democratic Presidents, 1948–2014

Another possible explanation for the partisan difference in electoral cycles is that the characteristic economic strategies of Democratic and Republican presidents in the “honeymoon” periods at the beginning of each four-year term have predictable spillover effects at the time of the next election. The tabulations presented in figure 2.2 show that the largest partisan differences in income growth, by far, were concentrated in the second year of each four-year presidential term—the first year in which new post-election policies would be likely to influence income growth. Democratic presidents have routinely produced extremely strong income growth in these “honeymoon” years; real disposable income per capita grew by an average of 3.2%, while family incomes in every part of the income distribution grew by an average of 2.9% to 3.4%. However, it seems that these expansionary bursts cannot be sustained indefinitely; by the time of the next presidential election, income growth under Democratic presidents typically slowed to a crawl, especially for families near the top of the income distribution. In stark contrast, the second years of Republican terms have often seen significant economic contractions; in fact, middle-class and working poor families experienced negative income growth in seven of the nine postwar Republican “honeymoon” years.¹⁶ But contractions, like expansions, have a finite duration, and by the time of the next presidential election income growth had typically rebounded significantly.

Since partisan differences in average income growth are greatest for middle-class and working poor families, those families are the most obvious economic losers from the partisan bias in accountability produced by voters' short time horizons. The result, obviously, is to exacerbate economic inequality. As if that were not enough, middle-class and poor families are also distinctly disadvantaged by a different kind of bias in economic voting—a significant class bias in sensitivity to income growth. American voters, regardless of their own place in the income distribution, seem to be quite sensitive to the economic fortunes of high-income families but much less sensitive to income growth among middle-class and poor families. Here, too, the political result is to bolster the Republican Party, and the economic result is to further exacerbate income inequality.

The statistical analyses of presidential election outcomes reported in table 3.1 show that the incumbent party's electoral fortunes are strongly affected by election-year income growth and by how long the incumbent party has held office. The first column of table 3.3 presents the results of a parallel analysis employing ANES survey data from the same 16 elections, relating the presidential vote choices of almost 19,000 voters to these same factors, plus the voters' own partisan attachments. The results are quite consistent with those presented in table 3.1. They imply that each additional percentage point of median income growth during the election year increased an otherwise undecided voter's probability of supporting the incumbent party by almost five percentage points, while each additional term in office reduced her probability of supporting the incumbent party by almost six percentage points.²¹

Surprisingly, the results presented in table 3.4 provide no indication of class-specific economic voting patterns. Rather, the clear pattern for high-, middle-, and low-income voters alike is that high-income growth mattered a great deal at the polls, while low-income growth conveyed no electoral advantage at all to the incumbent party, even among low-income voters. The analyses reported in table 3.5 provide even more remarkable evidence of class biases by comparing the effects of high-income growth and overall income growth (as measured by election-year changes in real disposable personal income per capita). Again, the results indicate that voters in every income group were highly sensitive to high-income growth, but much less sensitive to overall income growth. For the electorate as a whole, each percentage point of income growth for affluent families seems to have produced as much additional support for the incumbent party as five points of growth in overall real disposable income per capita. If anything, this bias was strongest among low-income voters, who were highly responsive to income growth among the affluent but attached no apparent weight at all to overall income growth.

Although the statistical results presented in tables 3.3, 3.4, and 3.5 are based on nearly 19,000 individual vote choices, they represent only 16 distinct configurations of election-year contexts. Thus, it is not impossible that the markedly disproportional electoral significance of high-income growth in tables 3.4 and 3.5 is merely a statistical fluke. However, if it is a fluke, it is a very persistent one. It certainly is not attributable to any one anomalous election, since the effect holds up strongly when each election is dropped from the analysis in turn.²² Nor is it limited to years in which Republicans, or Democrats, held the White House, or to the first half or the last half of the postwar era.²³

Moreover, class-biased economic voting does not seem to be limited to the United States. Timothy Hicks, Alan Jacobs, and J. Scott Matthews have used the framework outlined here to examine the electoral significance of high-income growth in a variety of affluent democracies. Their comparative study included detailed analyses of survey data from the United States (replicating and extending the analyses presented in the original edition of this book), the United Kingdom, Sweden, and Canada, as well as broader cross-national evidence from over 200 elections in 15 countries from 1945 to 2010. They concluded that low- and middle-income voters “seem remarkably poor at defending—and, indeed, seem to systematically undermine—their own distributive interests. We see widespread demand for inequality across the advanced democracies—an electoral bonus to governments that preside over a concentration of gains at the top. And we find no evidence whatsoever that governments pay a penalty among any income group or in the electorate as a whole for upwardly biased distributions of growth.”²⁴

In the case of the United States, as we saw in chapter 2, “upwardly biased distributions of growth” have occurred rarely under Democratic presidents but routinely under Republicans. The stark implication of the analyses presented here is that the dramatic shortfalls in income growth for middle-class and working poor families under Republican presidents evident in figure 2.1 have had little or no electoral impact. If economic voting is viewed as a mechanism by which middle- and low-income voters might “defend … their own distributive interests”—or, for that matter, the economic interests of the American public as a whole—then the class biases documented here have rendered that mechanism utterly ineffectual.

If biased perceptions of economic conditions are not the answer, what else might account for the odd sensitivity of voters at every income level to election-year income gains among affluent families? Perhaps income gains for affluent families get translated into campaign contributions to the incumbent party, which in turn influence the behavior of other voters in ways that are not directly captured in the statistical analyses presented here. Certainly, most campaign contributions come from relatively wealthy people.²⁶ If they are inspired to show their gratitude to the incumbent party by giving a bit more generously when their election-year pocketbooks are flush, that would produce an indirect but potentially potent connection between high-income growth and presidential voting behavior.

The statistical analysis reported in the first column of table 3.6 relates the incumbent party's spending advantage in each election to real election-year income growth for families at the 95th percentile of the income distribution—the sorts of people who are most likely to be contributors to political campaigns. My measure of each presidential candidate's campaign spending is scaled in (inflation-adjusted 2015) dollars per voter. By that measure, spending has increased substantially over the period covered by my analysis, from about $1.75 per voter (for both parties combined) in the 1950s to almost $9.40 per voter in 2012. The spending differential between the two candidates in each election ranged from a few cents (in 1960 and 1976) to more than two dollars (in 1972, 1980, and 1984).²⁷

There is indeed a substantial relationship between the incumbent party's spending advantage and income growth at the top of the income distribution, as the results presented in the first column of table 3.6 show. However, this analysis makes no allowance for the fact that election-year income growth rates, especially at the top of the income distribution, have generally been higher under Republican presidents than under Democratic presidents. And since campaign contributors are likely to have other, ideological reasons to prefer Republican candidates to Democrats, the relationship between incumbent partisanship and high-income growth may produce a spurious statistical association between high-income growth and the incumbent candidate's campaign spending advantage.

Table 3.7 elaborates the statistical analyses of presidential voting behavior presented in table 3.5 to allow for the impact of differential campaign spending on voters' choices. The results for the entire electorate suggest that campaign spending did have a substantial effect on voters' choices. For a voter who was otherwise equally well disposed toward both candidates, they imply that each additional dollar of campaign spending increased the probability of supporting the candidate who spent the money by about three percentage points. The implied effect is, of course, much smaller for voters who were strongly predisposed on other grounds to favor one candidate or the other. Nevertheless, this effect is large enough to suggest that differential campaign spending has probably had a significant electoral impact in presidential elections over the past half-century.³⁰

Surprisingly, campaign spending seems to have had a considerably stronger effect among affluent voters than among voters of modest means. The separate analyses for high-, middle-, and low-income voters in table 3.7 suggest that campaign spending was about twice as effective among high-income voters as among middle-income voters—but utterly ineffective among voters in the bottom one-third of the income distribution. And since the incumbent party's edge in campaign spending has been strongly correlated with election-year income growth for affluent families (r = .73), the variation across income groups in the apparent effect of campaign spending produces offsetting variation in the apparent effect of high-income growth on support for the incumbent party's presidential candidate. For the electorate as a whole, the estimated effect of high-income growth is about 25% smaller in table 3.7 than in table 3.5. For high-income voters, it is about 55% smaller, while for low-income voters it is slightly larger.

The statistical analyses presented in table 3.7 suggest that presidential campaign spending has had a significant impact on voting behavior over the past half-century. However, they do not speak directly to the political consequences of that fact. Those consequences are spelled out more clearly in the first column of table 3.8, which reports the estimated impact of differential campaign spending on the outcome of each of the past 16 presidential elections.³¹ Since Republican candidates outspent their Democratic opponents in 13 of those elections, it is not surprising to find that they would generally have done less well if campaign spending had been equal. The estimated impact of differential campaign spending on the average Republican vote share over all 16 elections amounts to 1.5 percentage points, more than enough to account for the entire net popular vote margin for Republican presidential candidates over the past several decades. In three cases—1972, 1980, and 1984—their share of the two-party vote was more than three points larger than it probably would have been with equal spending, and in two cases—1968 and 2000—Republican candidates won close elections that they very probably would have lost had they been unable to outspend their Democratic opponents.

The estimated effects of economic myopia are reported in the third column of table 3.8, which compares the actual Republican popular vote in each election with the projected outcome based on cumulative economic growth rather than election-year growth.34 The projections suggest that economic myopia inflated the average Republican vote by about 1.7 percentage points, benefiting Republican candidates substantially in six of the past 16 elections and marginally in four more. Richard Nixon in 1968 and George W. Bush in 2000 probably owed their accessions to the White House to the fact that voters forgot (or simply ignored) strong periods of income growth early in the terms of their Democratic predecessors.

Finally, the fourth column of table 3.8 addresses the combined effects of the three distinct partisan biases examined in this chapter—economic myopia, class biases in sensitivity to election-year income growth, and differential campaign spending.³⁵ The results suggest that, in combination, these three distinct partisan biases increased the average postwar vote for Republican candidates by almost four percentage points. Since the actual average Republican vote share over this period was 51.2%, the combined estimated effect of these partisan biases in economic accountability is large enough to account for the average Republican plurality three times over. In four instances—1968, 1972, 1980, and 2000—partisan biases were very probably essential to Republican presidential victories. These elections include every case in the past half-century in which Republican challengers have managed to defeat Democratic incumbents or their successors. Dwight Eisenhower in 1952 was probably the only Republican in the postwar era who would have won the job in his own right, without the benefit of incumbency, myopic voters, and differential campaign spending.

On the other hand, Barack Obama in 2008 turned the usual partisan pattern of biased accountability on its head by winning an election that he might well have lost had voters been more “accessible to long views,” as Ostrogorski put it.36 In one sense, Obama's election was unsurprising given the economic and political conditions at the time of the election. The economy was already plunging into recession as the election year began, and by Election Day it was in a full-scale meltdown. Real disposable income per capita grew by about half a percentage point in 2008, thanks in part to a spring tax cut aimed at bolstering the slumping economy. However, annual median pre-tax income plunged by 3.5%, and even the incomes of affluent families fell by 2.4%. Moreover, George W. Bush had clearly worn out his welcome after eight years in the White House; his Gallup job approval rating did not reach 35% at any point in the election year and often fell below 30%. The 2008 result was quite consistent with projections taking account of these economic and political circumstances.³⁷

In another sense, however, Obama's victory was remarkable. Only once in the postwar era—at the end of Dwight Eisenhower's second term—had election-year income growth under a Republican president been as anemic as it was in 2008. The substantial declines in pre-tax income at every income level contrasted dramatically with the usual pattern of robust income growth at the end of Republican terms. In 2004, George W. Bush had been the first Republican president in more than 70 years to preside over an election-year dip in the incomes of affluent families, but that 0.6% drop was dwarfed by the 2.4% decline in 2008. In part for that reason, John McCain was even more soundly outspent by Obama than Bush had been by John Kerry in 2004.

¹ Ostrogorski (1902, 302–303).

² Republican presidential candidates won a total of 730 million popular votes over this period, Democratic candidates 709 million. All of the election results cited in this chapter are based on the tabulations reported in Dave Leip's Atlas of U.S. Presidential Elections (http://uselectionatlas.org/).

³ Frank (2004, 45). For a detailed critique of this thesis, see Bartels (2006b; 2008, chap. 3).

⁴ This literature is too vast to cite in detail. Kramer (1971) and Tufte (1978) made important early contributions. Erikson (1989; 1990) analyzed presidential and midterm congressional results, respectively. For presidential elections, Bartels and Zaller (2001) compared a variety of alternative measures of economic performance and probed the robustness of the statistical results to variations in model specification. Achen and Bartels (2016, chap. 4) reviewed the logic of retrospective political accountability and the empirical evidence for retrospective voting on the basis of economic conditions.

⁵ Kramer (1971, 140); Key (1966, 7).

⁶ Fiorina (1981, 5).

⁷ On idiosyncratic economic experience, rationalization, and partisan bias, see Kramer (1983); Conover, Feldman, and Knight (1987); Bartels (2002a); Erikson (2004).

⁸ My analysis here builds upon joint work with Christopher Achen (Achen and Bartels 2004; Achen and Bartels 2016, chap. 6).

⁹ Hibbs (2006, 7). Hibbs's own analysis is rare in focusing on the extent of temporal discounting in voters' reactions to economic conditions. He found relatively modest discounting of past economic performance; however, other analysts employing similar models have found that recent economic performance is much more relevant than previous performance (Bartels and Zaller 2001; Erikson, Bafumi, and Wilson 2002; Achen and Bartels 2004).

¹⁰ Real disposable personal income (BEA, NIPA, table 7.1, line 12) includes wages and salaries, investment income, and government transfers (including, for example, Social Security and unemployment benefits but not in-kind government services such as public education) minus taxes paid. Thus, it arguably provides a better gauge of economic well-being than the pre-tax income figures for households or families reported by the Census Bureau. It may also better capture the direct impact of government policies intended to bolster incomes in the short run, such as tax cuts or benefit increases. Of course, per capita income is an average value, and changes in the average may not reflect the economic fortunes of most individuals when income is distributed unequally, as it increasingly has been over the decades analyzed here.

¹¹ Indeed, even the “short-term” measure of election-year income growth employed here may exaggerate voters' attention spans. Using quarterly rather than annual data, Achen and Bartels (2016, chap. 6) showed that income growth in just the middle two quarters of each election year accounted for election outcomes even better than income growth over the entire election year.

¹² Tufte (1978, 143, 63). Tufte argued that Nixon's enthusiasm for political manipulation of the economy arose in significant part from his unhappy experience with an unmanipulated economy in his first presidential campaign in 1960. According to Nixon (quoted by Tufte 1978, 6): “In October, usually a month of rising employment, the jobless rolls increased by 452,000. All the speeches, television broadcasts, and precinct work in the world could not counteract that one hard fact.” Nixon lost the election by fewer than 120,000 votes.

¹³ Schultz (1995, 79).

¹⁴ Here, as in chapter 2, I lag partisan control by one year to allow time for a president's policies to affect income growth. Thus, each president's influence is assumed to run from the year after his inauguration through the year after his election (the following inauguration year).

¹⁵ Galbraith, Giovannoni, and Russo (2007, 22); Clark and Arel-Bundock (2013, 1).

¹⁶ The real incomes of middle- and low-income families fell in 1954, 1958, 1970, 1974, 1982, 1990, and 2002. In two of those cases—1982 and 2002—incomes had already begun to fall in the first year of the new Republican president's term, before his policies were likely to have had much effect. However, even leaving aside first terms, the onset of recessions within 20 months following three of the five postwar Republican reelections (August 1957, November 1973, and July 1990) contrasts with the record of spectacular income growth in the corresponding portions of Democratic administrations.

¹⁷ The largest one-year decline in the real incomes of affluent families, 5.5%, occurred in 1948. The average income growth rate in the other seven election years with Democratic incumbents was positive (.51), but still far below the average growth rate in non-election years (2.46).

¹⁸ Data and documentation are publicly available from the ANES website (http://www.electionstudies.org/). In presidential election years, the ANES surveys have included pre-election interviews conducted between Labor Day and Election Day and post-election reinterviews with 85–90% of the pre-election respondents. Most of the surveys were conducted in respondents' homes by interviewers employed by the University of Michigan's Survey Research Center. The survey samples are intended to be representative of citizens of voting age living in households in the continental United States (excluding Alaska and Hawaii). My analyses employ the survey weights provided in the ANES Cumulative Data File (VCF0009x) but are additionally weighted by the ratio of national turnout to surveyed voters in each election year (in order to avoid over-weighting or under-weighting years with unusually large or small survey sample sizes).

¹⁹ I use survey respondents' self-reports of family income to create three income groups of roughly equal size in each election year, excluding those (about 7%) who declined to report their families' incomes. The differences in reported sample sizes for the three groups reflect differences in election turnout rates across income groups.

²⁰ The classic scholarly explication of party identification and its effects is by Campbell et al. (1960). Bartels (2000) tracked the relationship between party identification and voting behavior in presidential and congressional elections over the second half of the 20th century.

²¹ The statistical results presented in table 3.3 are not directly comparable with those presented in table 3.1, since they are based on probit analyses of the probability of voting for the incumbent party's presidential candidate rather than ordinary regression analyses of the incumbent party's aggregate vote margin. One implication of the probit model is that the implied effects of the explanatory factors are non-linear, reaching their maximum levels among voters who are otherwise equally likely to support either party. (For example, the impact of economic conditions is assumed to be greater among political independents than among strong partisans, since the latter are likely to vote for their party's candidate regardless of whether the economy is booming or slumping.) That the estimated effects reported in the text, for an otherwise undecided voter, are almost twice as large as the average effects for the electorate as a whole implies that an additional percentage point of election-year income growth would increase the incumbent party's vote margin by about 5.5 points and that an additional term in office would decrease the incumbent party's vote margin by about 6.6 points. The corresponding estimates derived from the fifth column of table 3.1 are 3.5 points and 5.2 points, respectively.

²² The estimated effects of high-income growth for the full sample range from .088 (omitting the 1968 election) to .113 (omitting the 2008 election). The estimated effects of overall disposable income growth range from −.016 (omitting the 1980 election) to .037 (omitting the 2008 election). The analysis omitting 2008 is the only one in which the estimated effect of high-income growth is less than four times as large as the estimated effect of overall disposable income growth.

²³ In the nine elections with Republican incumbents, the estimated effects of high-income growth and overall growth were .082 and .003; in the seven elections with Democratic incumbents, the estimated effects were .141 and .028. In the eight elections from 1952 through 1980, the estimated effects were .123 and .010; in the eight elections from 1984 through 2012, the estimated effects were .106 and −.022.

²⁴ Hicks, Jacobs, and Matthews (2015, 31).

²⁵ ANES surveys since 1980 have regularly included a question asking whether “over the past year the nation's economy has gotten better, stayed the same, or gotten worse.” A statistical analysis of responses to that question indicates that they are sensitive to overall election-year income growth, as one might expect, but also to the specific income growth rate for families at the 95th percentile of the income distribution. The latter effect is about half as large as the former effect, suggesting that the specific economic fortunes of affluent families had a significant impact on perceptions of the state of the national economy, over and above the impact of general income growth, even among people who were themselves far from affluent. However, much of the statistical impact of high-income growth rates on electoral support for the incumbent party documented in table 3.5 seems to persist even when voters' perceptions of the national economy are included in the analyses as a separate explanatory factor. Thus, the electoral significance of high-income growth does not appear to be entirely, or even primarily, mediated by biased perceptions of national economic conditions.

²⁶ For example, Verba, Schlozman, and Brady (1995, 194, 565) found that people in the top quartile of the income distribution accounted for almost three-quarters of the total campaign contributions in their 1989 Citizen Participation Study. The broad middle class accounted for almost all the rest; people in the bottom quintile of the income distribution accounted for only 2% of total contributions.

²⁷ The system of financing presidential election campaigns has changed markedly over the past half-century, most notably with the institution of public funding following the Watergate scandal in the early 1970s. In principle, public funding equalized spending by the two major-party candidates in general election campaigns from 1976 through 2004. (Barack Obama in 2008 was the first general election candidate to decline public funding in order to avoid associated spending limits.) However, a rising tide of spending by parties and other groups during this period undoubtedly had important spillover effects on voters in presidential elections. I attempt to allow for these changes by measuring campaign spending somewhat differently in the pre- and post-Watergate eras. For 1952 through 1976, I use estimates of general election spending compiled by Herbert Alexander (1980, 5). For 1980 through 2000, I count the public funds allocated to each presidential candidate for the general election campaign (ignoring spending during the primary season) plus half the total spending by their parties reported by the Federal Election Commission. For 2004 through 2012, I use the detailed estimates of total general election spending compiled by Candice Nelson (2014, 125), which include spending by the presidential candidates, coordinated expenditures by the parties, and independent expenditures by parties and other groups in support of each presidential candidate (or against his opponent).

²⁸ This result may help to account for the fact that Democratic presidential candidates significantly outspent their Republican opponents in 2004 and 2008—something that had not happened previously in the postwar era. However, it would be a mistake to attach too much weight to the result given the crudeness of the measurement on which it is based. I compared the absolute values of Poole and Rosenthal's (2007) DW-NOMINATE scores for each party's congressional delegation (averaging the median scores for party caucuses in the House and Senate). The scores generally range from −1 (for the most liberal members) to +1 (for the most conservative members) in each Congress, although additional restrictions intended to harmonize the scores across Congresses sometimes produce values outside that range. By this measure, the relative extremism of the Republican Party has increased steadily since the 1980s; thus, any other factor that eroded the Republican fund-raising advantage over time would contribute to the apparent effect of extremity in table 3.6.

²⁹ The measure of election-year income growth in figure 3.5 combines the figures for real disposable income per capita and 95th-percentile income growth, each weighted by its estimated effect on relative spending from the third column of table 3.6.

³⁰ There is reason to worry that the estimated effects of campaign spending in table 3.7 might be exaggerated owing to unobserved features of the election context that influence both the incumbent party's spending advantage and the incumbent party's electoral success. Replacing the observed spending advantage in each election year with predicted values (instrumental variables) derived from the statistical analyses in table 3.6 produces even larger estimated effects than those reported in table 3.7. Nevertheless, the estimated effects of campaign spending in table 3.7—and their inferred impact on the outcomes of specific elections in table 3.8—must be considered no more than rough estimates.

³¹ My projections are based on the separate results for high-, middle-, and low-income voters presented in the second, third, and fourth columns of table 3.7. Thus, they allow for greater sensitivity to campaign spending on the part of high-income voters. They also allow for greater sensitivity on the part of voters who were otherwise equally likely to support either candidate (most notably, those who claimed that they did not identify with, or lean toward, either party). Having estimated the impact of unequal campaign spending on the probability of supporting the incumbent party's candidate for each major-party voter in the ANES surveys, I aggregated these estimated effects in each election year to produce the projected impact of unequal spending on the Republican share of the two-party vote.

³² The projections are constructed along the same lines as those reported in the first column of table 3.7. Voters are allowed to be myopic (responding only to election-year income growth), and they are allowed to be responsive to campaign spending and other factors to the extent implied by the statistical results reported in table 3.7. Differences between actual and projected voting behavior are computed by comparing each voter's probability of casting a Republican vote based on the parameter estimates and data in table 3.7 with the corresponding probability calculated by substituting the growth rate of median family income in each election year for the observed growth rate among families at the 95th percentile of the income distribution.

³³ The median family's real income grew by 4.6% in 1968, with working poor families gaining 6.2% and even those at the 80th percentile gaining 3.9%, but affluent families' real incomes declined by 0.6%.

³⁴ The estimated effects of myopia are based on the assumption that non-myopic voters in each election year would have weighed income growth in the second, third, and fourth years of the incumbent president's term equally. This assumption implies that each additional percentage point of income growth would have increased the incumbent party's popular vote by one-third of the amount implied by the corresponding election-year estimate in table 3.7, regardless of whether that growth occurred in the election year or in either of the two preceding years.

³⁵ The relevant counterfactual for these calculations is how each election would have turned out if the two major-party candidates had spent the same amount of money on their campaigns and if economic voting had been based on cumulative growth in median family income and real disposable income per capita over the second, third, and fourth years of each president's term. The projections are based on the income-specific probit parameter estimates reported in table 3.7. The estimated combined effects of the three biases reported in the fourth column of table 3.8 are not equal to the sums of the separate estimated effects in the preceding columns because the adjustments for class bias and myopia in the fourth column are applied simultaneously (replacing election-year high-income growth rates with cumulative median-income growth rates) rather than separately (replacing election-year high-income growth rates with election-year median-income growth rates to estimate the impact of class bias and with cumulative high-income growth rates to estimate the impact of myopia).

³⁶ Ostrogorski (1902, 302).

³⁷ Obama's popular vote margin in 2008 was 7.3 percentage points. The analysis presented graphically in figure 3.2 suggests that he “should” have won by about six percentage points solely on the basis of low election-year disposable income growth. The analysis reported in the second column of table 3.1 incorporating incumbent party tenure suggests that he “should” have won by about 5.8 percentage points. An analysis incorporating both election-year disposable income growth and high-income growth (along with incumbent party tenure) suggests that he “should” have won by about eight percentage points. An analysis based on incumbent party tenure and disposable income growth in the middle two quarters of the election year suggests that he “should” have won by about 9.3 percentage points. By any of these benchmarks, the 2008 outcome was well within the range of expected outcomes based on postwar electoral history.

³⁸ Key (1966, 7; 1964, 568). Although the colorful phrase “rational god of vengeance and reward” is frequently (mis)quoted in connection with the optimistic view of retrospective accountability that Key set out in The Responsible Electorate, its actual use in his earlier textbook, Politics, Parties, and Pressure Groups, is more specific and negatory: “The Founding Fathers, by the provision for midterm elections, built into the constitutional system a procedure whose strange consequences lack explanation in any theory that personifies the electorate as a rational god of vengeance and of reward.”

TABLE 3.1
Myopic Economic Voting in Presidential Elections, 1952–2012
Ordinary least squares regression parameter estimates (with standard errors in parentheses) for incumbent party's popular vote margin (%).
	(1)	(2)	(3)	(4)	(5)
Cumulative disposable income growth (%)	1.55 (.80)	—	1.71 (.57)	—	.49 (.66)
Election-year disposable income growth (%)	—	4.81 (1.19)	—	4.16 (.92)	3.49 (1.30)
Incumbent party tenure (years)	—	—	−1.66 (.44)	−1.22 (.36)	−1.30 (.38)
Intercept	−6.38 (5.97)	−8.79 (3.72)	4.12 (5.08)	1.47 (4.14)	.56 (4.39)
Standard error of regression	10.05	7.68	7.19	5.82	5.92
Adjusted R²	.16	.51	.57	.72	.71
N	16	16	16	16	16
Sources: Dave Leip's Atlas of U.S. Presidential Elections; Bureau of Economic Analysis.

TABLE 3.2
The Presidential Election Cycle and Partisan Income Growth, 1948–2014
Annual real pre-tax income growth (%) for families at various points in the income distribution (with standard errors in parentheses). Partisan control measured from one year following inauguration to one year following subsequent inauguration.
	Democratic presidents	Republican presidents	Partisan difference
Presidential election years
20th percentile	.83 (1.32)	1.08 (1.27)	−.25 (1.83)
40th percentile	1.02 (.89)	1.87 (1.07)	−.85 (1.41)
60th percentile	1.21 (1.17)	1.67 (1.02)	−.46 (1.55)
80th percentile	.66 (.97)	1.91 (1.04)	−1.25 (1.43)
95th percentile	−.24 (.98)	2.46 (.97)	−2.69 (1.39)
N	8	9	17
Non-election years
20th percentile	2.37 (.79)	−.09 (.61)	+2.47 (.98)
40th percentile	2.22 (.61)	.14 (.47)	+2.08 (.76)
60th percentile	2.24 (.48)	.64 (.43)	+1.60 (.65)
80th percentile	2.45 (.46)	.93 (.39)	+1.52 (.60)
95th percentile	2.46 (.60)	1.31 (.45)	+1.15 (.74)
N	23	27	50
Source: Census Bureau, Historical Income Tables.

TABLE 3.3
Economic Voting by Income Group, 1952–2012
Probit parameter estimates (with standard errors in parentheses) for probability of incumbent party vote. Weighted data clustered by election year; major-party voters only.
	All income groups	High income	Middle income	Low income
Cumulative disposable income growth (%)	−.026 (.027)	−.034 (.029)	−.022 (.032)	−.023 (.026)
Election-year disposable income growth (%)	.124 (.049)	.132 (.062)	.146 (.042)	.080 (.057)
Incumbent party tenure (years)	−.037 (.013)	−.039 (.015)	−.033 (.014)	−.051 (.015)
Incumbent party identification (−1 to +1)	1.600 (.076)	1.678 (.085)	1.589 (.084)	1.531 (.083)
Intercept	.161 (.183)	.259 (.219)	.013 (.169)	.334 (.223)
Log likelihood	−7,582.9	−2,567.7	−2,409.9	−2,090.7
Pseudo-R²	.42	.43	.42	.42
N	18,975	6,580	5,947	5,244
Source: 1952–2012 ANES surveys.

TABLE 3.4
The Electoral Impact of High- and Low-Income Growth by Income Group, 1952–2012
Probit parameter estimates (with standard errors in parentheses) for probability of incumbent party vote. Weighted data clustered by election year; major-party voters only.
	All income groups	High income	Middle income	Low income
Election-year high-income (95th percentile) growth	.126 (.013)	.128 (.017)	.139 (.015)	.115 (.018)
Election-year low-income (20th percentile) growth	−.026 (.015)	−.031 (.016)	−.030 (.015)	−.021 (.017)
Incumbent party tenure (years)	−.049 (.011)	−.054 (.014)	−.041 (.011)	−.058 (.010)
Incumbent party identification (−1 to +1)	1.649 (.065)	1.721 (.083)	1.647 (.070)	1.586 (.067)
Intercept	.252 (.102)	.337 (.136)	.155 (.098)	.321 (.104)
Log likelihood	−7,414.3	−2,516.0	−2,354.2	−2,038.2
Pseudo-R²	.44	.45	.43	.44
N	18,975	6,580	5,947	5,244
Source: 1952–2012 ANES surveys.

TABLE 3.5
The Electoral Impact of High-Income Growth by Income Group, 1952–2012
Probit parameter estimates (with standard errors in parentheses) for probability of incumbent party vote. Weighted data clustered by election year; major-party voters only.
	All income groups	High income	Middle income	Low income
Election-year high-income (95th percentile) growth (%)	.096 (.016)	.093 (.021)	.097 (.018)	.100 (.014)
Election year disposable income growth (%)	.018 (.028)	.023 (.039)	.038 (.027)	−.022 (.035)
Incumbent party tenure (years)	−.046 (.012)	−.050 (.015)	−.037 (.011)	−.059 (.011)
Incumbent party identification (−1 to +1)	1.640 (.064)	1.716 (.082)	1.634 (.068)	1.577 (.065)
Intercept	.203 (.151)	.273 (.195)	.057 (.127)	.382 (.170)
Log likelihood	−7,425.5	−2,521.6	−2,356.8	−2,039.8
Pseudo-R²	.43	.45	.43	.44
N	18,975	6,580	5,947	5,244
Source: 1952–2012 ANES surveys.

TABLE 3.6
Sources of the Incumbent Party's Campaign Spending Advantage, 1952–2012
Ordinary least squares regression parameter estimates (with standard errors in parentheses) for the incumbent party's spending advantage (in 2015 dollars per voter).
	(1)	(2)	(3)	(4)
Election-year high-income (95th percentile) growth (%)	.385 (.097)	.303 (.090)	.223 (.085)	.145 (.091)
Election-year disposable income growth (%)	—	—	.291 (.125)	.307 (.116)
Republican incumbent	—	1.28 (.47)	1.25 (.41)	1.55 (.42)
Incumbent party extremity (difference in absolute DW-NOMINATE scores)	—	—	—	−3.33 (1.91)
Intercept	−.81 (.29)	−1.31 (.33)	−2.03 (.42)	−2.16 (.40)
Standard error of regression	1.00	.86	.74	.69
Adjusted R²	.50	.63	.72	.76
N	16	16	16	16
Sources: Alexander (1980); Federal Election Commission; Nelson (2014).

TABLE 3.7
The Electoral Impact of Campaign Spending by Income Group, 1952–2012
Probit parameter estimates (with standard errors in parentheses) for probability of incumbent party vote. Weighted data clustered by election year; major-party voters only.
	All income groups	High income	Middle income	Low income
Incumbent spending advantage ($ per voter)	.078 (.038)	.157 (.037)	.082 (.051)	−.023 (.040)
Election-year high-income (95th percentile) growth (%)	.071 (.021)	.042 (.026)	.072 (.023)	.108 (.020)
Election-year disposable income growth (%)	−.003 (.035)	−.019 (.037)	.013 (.039)	−.017 (.036)
Incumbent party tenure (years)	−.049 (.010)	−.055 (.011)	−.040 (.009)	−.058 (.012)
Incumbent party identification (−1 to +1)	1.650 (.064)	1.722 (.082)	1.647 (.071)	1.572 (.065)
Intercept	.330 (.183)	.526 (.214)	.197 (.177)	.347 (.181)
Log likelihood	−7,410.4	−2,500.6	−2,351.8	−2,039.5
Pseudo-R²	.44	.45	.43	.44
N	18,975	6,580	5,947	5,244
Source: 1952–2012 ANES surveys.

TABLE 3.8
The Effects of Unequal Campaign Spending, Class Bias, and Economic Myopia on Presidential Election Outcomes, 1952–2012
Projected impact on Republican share of two-party popular vote. Italic entries represent years in which actual and projected election winners differ.
Election year (Republican vote %)	Estimated effect of unequal campaign spending (%)	Estimated effect of differential sensitivity to high-income growth (%)	Estimated effect of differential sensitivity to election-year growth (%)	Combined effect of myopia, class bias, and unequal spending (%)
1952 (55.5)	+0.4	−0.0	−2.4	+1.6
1956 (57.8)	+0.6	−0.5	+4.5	+4.9
1960 (49.9)	+0.1	+2.7	+0.4	+2.4
1964 (38.7)	+1.2	+1.1	−1.6	+2.1
1968 (50.4)	+2.2	+7.8	+6.5	+9.0
1972 (61.8)	+4.3	+2.8	+6.1	+12.6
1976 (48.9)	−0.0	−0.4	+4.3	+4.9
1980 (55.3)	+3.9	−1.6	+5.0	+8.3
1984 (59.2)	+4.7	+0.0	+0.6	+8.0
1988 (53.9)	+2.9	+2.6	−0.4	+2.7
1992 (46.5)	+1.0	+1.9	+2.4	+3.5
1996 (45.3)	+2.1	+0.9	+1.3	+3.9
2000 (49.7)	+2.2	+0.9	+4.9	+5.8
2004 (51.2)	−2.0	−0.8	−0.6	−2.3
2008 (46.3)	−2.1	+1.7	−2.6	−5.2
2012 (48.0)	+1.8	−0.4	−1.2	−0.1
Average (51.2)	+1.5	+1.2	+1.7	+3.9
Source: Calculations based on parameter estimates in table 3.7.