Normalcy and the Erratic Shift of Seats and Party Control in the United States Senate,

Transcription

1 Normalcy and the Erratic Shift of Seats and Party Control in the United States Senate, Andrew W. Podob April 26, 2017 Abstract Congressional Democrats went from holding 44 Senate seats before the 2006 election to 60 Senate seats after the 2008 election, to 46 Senate seats after the 2014 election, for a net shift of 30 seats in less than 10 years. While seemingly hectic, this paper investigates whether that shift what many call surge and decline has occurred in other eras too, namely throughout the 20th Century. Using a dataset going as far back as the election of 1900 and the resulting 57th Congress, I show how erratic shift in party control, operationalized through number of Senate seats held by the Democratic and Republican parties, is not necessarily as erratic as it seems; instead, perhaps it is the normal, as parties that are more overextended naturally lose more seats. This analysis advances and clarifies the surge and decline literature, distinguishes between election-to-election and over-time surge and decline, and finds that surges are cancelled out by roughly equal declines. Prepared for Presentation in Vlad Kogan s Federalism Seminar. Very preliminary draft. Please ask before citing. The Ohio State University, 2001 Derby Hall, 154 N Oval Mall, Columbus, Ohio 43210; podob.1@osu.edu 1

2 1 Opening Feeding off of intense frustration with the progress of the wars in Iraq and Afghanistan, and intense displeasure after six years of the Republican presidency of George W. Bush, in the 2006 midterm elections, Democrats stormed back into the majority of the United State Senate and United States House of Representatives after not controlling the former since November 2002, and not controlling the latter since the Republican Revolution of Senate Democrats would solidify their majorities even further after the 2008 elections, securing a 60-seat filibuster-proof majority in the Senate and 258 seat majority in the House of Representatives 1. Many commentators even went so far as to call the Republican Party dead, asserting they lost their congressional majorities for a generation relegated to an almost permanent minority status 2. But congressional Republicans came roaring back, feeding off Tea Party fever and the backlash by middle-america against President Barack Obama, picking up 63 House seats in the 2010 midterm elections and the majority with it, and 6 Senate seats. Overall, congressional Democrats went from holding 44 Senate seats before the 2006 election to 60 Senate seats after the 2008 election, to 46 Senate seats after the 2014 election, for a net shift of 30 Senate seats in less than 10 years. Was this drastic shift, what many call surge and decline, normal? While almost no one saw it coming, should it have been predictable? 1 By comparison, in the 114th Congress, congressional Republicans controlled 247 House seats, their largest House majority since the 71st Congress ( ). In the 111th Congress, Democrats held 11 more House seats than this. 2 Consider this article from Esquire Magazine: republican-party-strategy-post-election / or this article: article/1109_mark_shields_the_republican_party_is_dead. 1

3 2 Literature Review 2.1 Angus Campbell s Surge and Decline Much of the original literature surrounding congressional elections descends from the Michigan Four and their seminal 1960 book The American Voter 3. Surge and decline, as we understand the term, originates in the mid-1960s with social psychologist Angus Campbell (1966). As he defined it, surge and decline was a mixture of three sets of forces. The first, high-stimulus vs. low-stimulus elections, the second, core vs. periphery voters, and the third, short-term vs. long-term circumstances. In high-stimulus elections, voters with low levels of political interest and low levels of partisan attachment are driven to the polls by the excitement (i.e. stimulation) generated by the campaign usually from the candidate at the top of the ticket. These periphery voters gave the electoral advantage to the in-party, but are no longer there after the surge. Surge and decline occurs, Campbell argues, because midterm elections are low-stimulus elections made up of core voters from each party s base, who tend to follow party lines. The drop-off of periphery voters in low-stimulus elections then causes the decline for the majority party 4. Describing the cycle of surge and decline, Campbell writes: fluctuations in the vote reflect the passing impact of contemporary events, and the subsequent decline toward the underlying division of partisanship after these events have lost their salience. Even more specific, he concludes: swings away from the basic division of party loyalties in high-turnout elections tend to swing back in the low-turnout elections which follow. In this sense, the theory of surge and decline as defined by Campbell is a function of motivation to vote in a particular election (the high-stimulus vs. low-stimulus elections), party ID, turnout (core vs. periphery voters), and time. Short term circumstances, like a charismatic presidential candidate, might swing large numbers of voters in an election, but eventually these voters will return to their usual partisanship, causing a surge 3 The Michigan Four are Angus Campbell, Philip Converse, Warren Miller, and Donald Stokes. 4 See Rogers (2015) for an investigation of rainfall and how it moderates (and weakens) the relationship between presidential and congressional voting, muddying surge and decline. 2

4 and then a decline. 2.2 Beyond Angus Campbell In the decades since, surge and decline has come to encompass many concepts within American electoral politics. One branch of the literature focuses on coattails effects, how winning presidential candidates carry other members of their party into federal and state offices, but how a dropoff for the president s party occurs in the subsequent midterm election when the president (or popular figure) is no longer on the ballot (Miller, 1955; Campbell, 1960; Key, 1964; Hogan, 2005; Meredith, 2013). From this perspective, surge and decline is a result of congressional candidates having someone to prop them up in one election, but no longer having that person in the next election i.e. they can no longer ride the coattails. Another branch of the literature focuses on the electorate itself, arguing that different persons turn out to vote in presidential vs. midterm elections, which naturally alters the electoral outcomes from a presidential to a midterm election, and vice versa (Campbell, 1960, 1985; Denardo, 1987; Erikson, 1988; Mattei and Glasgow, 2005). Tufte (1975) refines this by arguing that the decline for the president s party occurs because voters treat midterm elections as a referendum on the president, hence the loss in congressional seats for his party 5. This may also be influenced by how quickly voters tire of the in-party, as voters lose faith in the competence of the in-party over time (Abramowitz et al., 1986; Lewis-Beck and Rice, 1992) 6. Others refine it further, arguing voters use midterm elections to balance the legislative and executive branches, thereby checking the president s party, preventing it from enacting legislation too extreme (Erikson, 1988; Alesina and Rosenthal, 1989; Fiorina, 1996; Scheve and Tomz, 1999; Bailey and Fulmer, 2011). Finally, others clarify or question Campbell s (1996) original theory, as Cover (1985) does when he fleshes out who exactly core and periphery voters are, and examining their partisan 5 See Kernell (1977) s negative voting model for more work on voters treating midterm elections as a referenda on the president and his party. 6 This likely explains the 6 year itch of a two-term presidency, when the president s party usually sustains more severe losses in the midterm election of their second term. 3

5 loyalties. Furthermore, James Campbell (1987) finds that while the original theory remains intact, the surge seen in presidential elections is not from periphery voters who Angus Campbell assumes are Independents or loosely affiliated voters but from partisans of the winning presidential party who only vote in high-stimulus elections. Most severe, Erikson (1988), who tests similar data to this paper from 1902 until the mid-1980s, finds that the type of surge and decline we would expect to see is not seen. Not long after, Campbell (1991) expands on these tests, testing similar data from , finding strong evidence for surge and decline. His major conclusion being, the decline of seats for the president s party in midterm elections is proportional to the vote margin the president and his party received in the presidential election preceding the midterm election. Overall, the literature spends its time explaining why surge and decline happens (Campbell, 1997), examining how it occurred in one particular election, usually a midterm election (Burmila, 2014; Campbell, 1997; Herrnson and Morris, 2007), or building forecasting models to predict when it will occur and how severe it will be (Coleman, 1997). But if surge and decline is a midterm election theory that is, it explains electoral outcomes for midterm elections in relation to the presidential elections that preceded them can it explain seat gains and losses over time more broadly? Moreover, studies like Campbell (1991) only include elections for the House of Representatives, not the Senate. 2.3 The Exposure Thesis Parallel to the literature on surge and decline, Oppenheimer et al. (1986) focus their study of shifts in congressional seats on deviations from the normal vote. As defined, the normal vote is the usual and customary or baseline vote division expected, usually judged by population, demographical, and party ID data, or from survey data (Converse, 1966). Taking this normal vote concept and applying it to congress, they write: In any alignment era there will be a normal seat split in Congress, not unrelated to its normal vote (Converse, 1966) in the electorate. When a party has more 4

6 seats than would be expected from this normal seat split, it is overexposed and will tend to lose seats in a congressional election. And the more overexposed the party is, the greater the number of seats it is likely to lose. Conversely, when a party is underexposed has fewer seats than its normal level it will gain seats. And those gains will be greater the more the party is underexposed. The tendency in congressional elections is to return to a normal seat split or equilibrium point (Oppenheimer et al., 1986). While Oppenheimer et al. (1986) treat their theory on overexposure as a separate one from surge and decline, arguing it requires fewer assumptions and is a better predictor of midterm election outcomes, I treat it as part of the surge and decline phenomenon. They operationalize overexposure as the excess or deficit number of seats a party holds measured against its long-term norm. This measure, and their theory more generally, is not entirely antithetical to the surge and decline literature. Both address a surge and a decline in seats each party holds, albeit in somewhat differing ways. For this reason I nest their theory into surge and decline because I believe it is the overexposure itself what I hereby term overextension that leads to the surge and decline. As I will lay out in the coming sections of this paper, it is the overextension of seats one party holds that predates and is a necessary precondition for that party losing so many seats later on. When one party controls more Senate seats than what common political sense would call normal or usual, that party is overextended. Without this overextension, such swings in seats like Democrats exhibited between 2006 and 2014 would be impossible. 3 Hypothesis To my knowledge, no one has examined the surge and decline phenomenon across time, at least since Oppenheimer et al. (1986) and James Campbell (1991) did more than two decades ago. Many have investigated the causes of surge and decline, and many have examined 5

7 particular instances of surge and decline, but few since them have mapped the surge and decline itself. To update a mostly dormant literature, I revisit it here quantitatively, and over dozens of congresses the United States Senate specifically. I do so under the guise of one hypothesis and three alternative hypotheses, each drawn from the literature to explain why surge and decline occurs. The strongest of which I expect to be true, I hypothesize that surges will be cancelled out by roughly equal declines. When one party over-performs in an election or consecutive elections, like Democrats did in 2006 and 2008, that party is thereby overextended in the following election or elections, like Democrats were in 2010 and Essentially, the nowoverextended party won Senate seats in states they would not normally have won them in, all else equal. The coattail and wave election effects that swept seemingly out of reach seats into their party s column are no longer present when these Senators stand for reelection six years later, and the opposing party quickly captures these normally-safe seats back. A textbook example of flukes like this is the 2008 election of Democrat Mark Begich in overwhelmingly Republican Alaska 7, who unseated damaged Republican incumbent Ted Stevens by little more than 1 point and with less than 48% of the vote. When Begich stood for reelection in 2014, six years later, he lost by more than 2 points and only received about 46% of the vote. Democrats over-performed in Alaska in 2008, which resulted in Democrats being overextended when they controlled the Senate seat Begich held. The decline echoed the surge six years later, when Republican Dan Sullivan won the seat from Begich. Three alternative hypotheses for erratic shifts in Senate seats include better funded challengers in more recent congresses, a decrease in the value of incumbency advantage (Jacobson, 2015), and because the electorate is different in midterm vs. presidential elections (Campbell, 1960, 1985; Denardo, 1987; Erikson, 1988; Mattei and Glasgow, 2005). It proves nearly impossible to test the better funded challengers hypothesis, as federal campaign finance records from the Federal Election Commission (FEC) are only publicly available for 7 Republican presidential candidate Donald Trump won Alaska by almost 15 points in the 2016 election. 6

8 a few congresses in the time series I consider. Additionally, the campaign finance regime has changed drastically over the last few decades, from being virtually nonexistent when the time series I consider begins, to being fairly robust now. Comparing funding of challengers in the 2008 election to the 1964 election, if even possible, would be like comparing apples and oranges. Also, if we find (which I do) that other eras without FEC campaign finance data have been as hectic in terms of surge and decline as the eras in which we have FEC campaign finance data for, we can assume campaign contributions had similar or negligible effects for the earlier eras. The same would hold true for the value of incumbency advantage. If we find (which I do) that surge and decline was prevalent in multiple eras, it would suggest that the value of incumbency advantage is not declining, making it somewhat irrelevant to surge and decline; or it could mean that the value of incumbency advantage is stronger than the forces of the decline in some instances, and weaker in others. Fowler and Hall (2015) find that Senate elections today are not independent of previous elections for those same seats (i.e. a Senate race in year X is not independent of the election for the same seat in year X + 6 or year X 6.) 8 Seats held by one party are more likely to remain in that party s hands. This finding, that elections have medium to long-term effects on subsequent elections, suggests that incumbency advantage should affect surge and decline by weakening the severity or decreasing the likelihood of decline. I treat the final alternative hypothesis on differing electorates as a branch of the overperformance hypothesis for surge and decline mentioned above, as differing electorates in midterm vs. presidential elections is one of the causes of surge and decline, among other causes. I test this sub-hypothesis indirectly, operationalizing differing electorates by testing if average Senate seat gains are different in different types of elections (midterm vs. presidential elections). Similar average seat gains or seat losses across both types may indicate either the electorate is similar; or even if the electorate is different, the effects of differing electorates cancel out across time. 8 They find that Senate elections for a particular seat are influenced by Senate elections up to three decades in the past for that same seat. 7

9 4 Data 4.1 Building the Dataset The data were obtained from the website of the office of the Senate historian 9. The dataset includes observations on each congress from the 57th through the present 115th ( ), the number of Democrats, Republicans, and Independents/others serving in each congress and which party they caucused with (if they did), and the total number of Senate seats in each congress. I augmented this data by adding variables indicating which party was in the majority 10, whether the congress was meeting following a midterm or presidential election 11, whether the congress was meeting during a Democratic or Republican presidency 12, the net number of Democratic seat changes from the previous congress to that congress 13, and streaks in the net number of Democratic seat changes over multiple congresses in a row. The latter two variables will be necessary to access how many net Senate seats changed parties from one congress to another (really from one election to another), and how these net Senate seat changes compounded or did not over multiple or consecutive congresses and elections. I began with the 57th Congress for three main reasons. The first reason was a judgment call. The election of 1900, and the 57th Congress that follows from it, heralds in the modern era, separating congresses of the 18th and 19th Centuries from those of the 20th and 21st Centuries. Additionally, with the ratification of the 17th Amendment to the Constitution in time for the 1914 elections, the 64th Congress is the first with any directly elected senators 14. There is reason to believe these congresses were inherently different from earlier congresses, as Senators were elected by state legislatures prior to the 17th amendment and by the people A dummy variable for the Majority Party variable, in which D=1, R=0 11 For the Midterm variable: A dummy variable for congresses meeting following a midterm election, in which Yes=1, No=0. For the Presidential variable: A dummy variable for Congresses meeting following a presidential election, in which Yes=1, No=0 12 A dummy variable for congresses meeting during a Democratic administration, in which Yes=1, No=0. 13 Positive numbers indicate positive net seat changes and negative numbers indicate negative net seat changes 14 Only the 1/3 of senators serving in the 64th Congress those in Class 1 were directly elected, as Class 2 and Class 3 senators did not stand for reelected until the 65th and 66th Congresses, respectively. 8

10 afterwards; though, the same forces that affect surge and decline at the national level may affect surge and decline on the state level i.e. state and national vote shares and election outcomes are correlated. Surge and decline could have occurred indirectly though state legislatures before it occurred directly through voters. Beginning with the 57th Congress allows us to examine surge and decline for a few congresses before the direct election of senators, and this serves as an unofficial control group or baseline group. The second reason was a party call, as this analysis concerns the Democratic and Republican parties only; meaning it is necessary to exclude congresses with Federalists, Whigs, and other nominal parties, as these parties cannot be compared to the Democratic and Republican parties across time, which is what this paper seeks to do. The third reason was a numbers call. Congresses of the 18th and 19th Centuries had fewer members than those of the 20th and 21st Centuries, as there were fewer states in the union earlier on. Including a congress like the 33rd Congress, which had 62 seats, is difficult to compare to the 60th Congress, which had 92 seats, as a net change of 4 seats, for instance, seemingly means a lot more in the former than the latter. I believe these three reasons justify beginning the dataset with the 57th Congress, and make for both smoother and internally valid comparisons across time. 4.2 Descriptive Statistics The data include 65 observations for 59 congresses. This is because some congresses, specifically the 103rd, 104th, and 107th, are included more than once, once for each time there was a net shift in Senate seats that resulted in a majority party switch from Democrats to Republicans, or vice versa. There were also a few congresses where special elections did not change the majority party, but changed the number of seats the majority party held, affecting streaks in the net number of Democratic seat changes over multiple congresses in a row. These extra congresses needed to be included in the dataset for these reasons. For example, the 107th Congress, which met from , is included four times. Twice 9

11 because the outgoing and incoming Vice President of the United States broke the seat deadlock, giving majority control first to the Democrats and then to Republicans, respectively 15. Once because Republican James Jeffords of Vermont switched parties, becoming an Independent and handing majority control to Democrats in the process. And once more because a death and a special election returned majority control to Republicans. Though there are technically only 59 congresses in the time series, for ease of explanation, I will hereby refer to them as the 65 congresses. The unit of analysis is each congress, and the subunit of analysis is senate seats. The plurality of Senates in the time series 36 had 100 total seats, followed closely by 96 total seats in 24 Senates. Five Senates, the first five in the time series, had less than 96 total seats, which had 90 and 92 total seats, respectively. This means we can relatively easily compare seats held across time, as the variance in total number of seats is quite small, almost nothing after the 61st Congress. The plot in Figure 1 below shows the total number of Senate Democrats seated in each Senate, from the 57th through 115th Congresses, spanning the entire dataset 16. Democrats hit their high water mark of 76 seats held in the 75th Congress, which met during the first two years of President Franklin D. Roosevelt s second term. Democrats hit their low water mark early on, of 31 seats held in the 60th Congress, which met during the last two years of President Theodore Roosevelt s presidency 17. The blue line shown in the figure is a lowess line, which plots local regressions, as an OLS regression line for the entire dataset would be mostly flat. The lowess line shows that Democrats steadily increased their number of seats held over the first half or so of the Senates in the time series, before steadily decreasing their number of seats held. Overall, Democrats had an average caucus size of 51 seats, meaning on average, they held the (slim) 15 Democrat Al Gore served as president of the Senate from January 3, January 20, 2001, and Republican Dick Cheney served as president of the Senate from January 20, 2001 onward. 16 The vertical grey line plotted for the 64th Congress indicates the first congress after ratification of the 17th Amendment, for which United States Senators became directly elected by the citizens of their respective states. Note that in the 64th Congress only 1/3 of the Senators were directly elected at that point, as only 1/3, the Class 1 senators, stood for election or reelection in the election directly preceding the 64th Congress. 17 In the post-17th amendment era, with direct election of senate seats, Democrats hit their low water mark of 37 seats held in the 67th Congress. 10

12 majority of the seats, which they did in 36 of the 65 congresses. Figure 1: Democrats in each Senate 70 Number of Senate Democrats Congress (57th 115th) The plot in Figure 2 below, like Figure 1 for Democrats, shows the total number of Senate Republicans seated in each Senate, from the 57th through 115th Congresses, spanning the entire dataset. Republicans hit their high water mark of 61 seats held in the 60th Congress 18, when the Democrats hit their low water mark. Republicans hit their low water mark of 16 seats held in the 75th Congress, when the Democrats hit their high water mark. The red line is a lowess line, which shows the reverse of what the blue lowess line in Figure 1 shows for Democrats. It shows a Republican conference that gradually held fewer and fewer seats before gradually growing larger. Overall, Republicans had an average conference size of In the post-17th amendment era, with direct election of senate seats, Republicans hit their high water mark of 59 seats held in the 67th Congress. 11

13 seats, meaning on average, they held the minority of seats. They held the majority of seats in only 29 of the 65 congresses. Figure 2: Republicans in each Senate Number of Senate Republicans Congress (57th 115th) The lines plotted in Figure 3 below show the density of seats held by Democrats and Republicans, respectively, for all congresses in the time series. This can be interpreted as the frequency, expressed as a density, that each party held a particular number of seats. Both parties, as expected, had their most dense seat numbers between 45 and 55. The skew of the Democratic plot is more jagged, indicating Democratic seat densities varied more than Republican ones. Also, the Democratic plot has a fatter left tail, as well as longer tails, indicating the variance in Democratic seats held was larger. The range of Republican seats held was while the range of Democratic seats held was These densities will be 12

14 fleshed out further in Figure 4, expressed as histograms. The histograms in Figure 4 below show the frequencies of holding a particular number of seats. The mode minority number of seats held by Democrats was 47 while the mode majority number of seats held by Democrats was 57. The mode minority number of seats held by Republicans was 44 and 47, while the mode majority number of seats held by Republicans was 53 and 55. The small difference in scale between the two plots reflect this. When Democrats were in the majority their majorities tended to be larger. Figure 4 shows that Democrats most frequent number of seats held was slim minority, meaning they were only a few seats shy of a majority. Republicans, on the other hand, were most often with a slim majority, meaning they were only a few seats shy of losing their majority. One thing not shown in Figures 3 or 4 are the numbers of Independents or members of other third parties holding seats in each Senate. Though not plotted, around half of the congresses in the time series had a Senate with at least 1 Independent or third-party member. While 33 Senates had zero, 17 had 1, 13 had 2, and 2 had 4. It ballooned mostly between the 74th and 77th Congresses, when New Deal Democrats had their largest majorities (and Republicans had their smallest minorities) in the time series. The aforementioned figures show Democratic and Republican seats held in each congress, allowing for straight comparisons between congresses. Figure 5 below shows the Democratic net seat changes in consecutive congresses. This can be interpreted as, if Democrats had a net gain (or loss) of Senate seats in the election preceding the congress, it would be plotted as the size of the net gain (or loss). For instance, Democrats had a net gain of 6 seats in the 57th Congress resulting from the Election of 1900 right after the 56th Congress, necessitating the plotting of a point for the numerical value of 6 for the 57th Congress (the leftmost dot in the figure). Democrats had their largest net gain of 16 seats in the election preceding the 86th Congress, taking them from a bare-bones majority to a robust majority. Democrats had their largest net loss of 12 seats twice, in elections preceding the 80th and 97th Congresses, sinking them from the majority to minority both times. The volatility in 13

15 Democratic net seat changes prohibits us from drawing too many conclusions from the blue lowess line plotted, other than to say that the extreme surge and decline epitomized by the tremendous numbers of ups and downs between sets of consecutive dots means net seat losses were soon cancelled out by net seat gains, and vice versa. This will be further explored later on in the next section. Figure 6 shows a histogram of the frequencies of these net changes in seats from the previous Congress. The range of net losses to gains is -12 to 16, with net changes in Democratic seats held from the previous congress having a positive mode of 1, and a negative mode of -2. Figure 6 obscures though the wide variance in net gains and losses, which will be further explored in terms of streaks. While Figure 5 shows the Democratic (net) seat changes in a particular congress resulting from the election preceding that congress, Figure 7 below takes these net changes and aggregates them, allowing us to examine streaks across multiple elections and multiple congresses. Important to note, streaks are either positive or negative, meaning a positive streak is cancelled out (and reset to zero) after a loss of seats, and a negative streak is cancelled out (and reset to zero) after a gain of seats. For example, Democrats (net) gained one seat in the election preceding the 61st Congress, for a streak of +1. After net gaining 12 more seats in the election preceding the 62nd Congress, their streak climbed to +13. After net gaining 7 more and 5 more seats in elections preceding the 63rd and 64th Congresses, respectively, their streak climbed yet again to +20 and then to +25. But after having a net loss of 2 seats in the election preceding the 65th Congress, their streak of +25 consecutive net seat gains ended, instead becoming a streak of -2. The largest positive streak for Democrats was a net of +37 seats in a row, from the 72nd through 75th Congresses. The greatest negative streak for Democrats was a net loss of -19 seats in a row, which they incurred twice. The mode positive steak for Democrats was a gain of +2 seats, which they incurred 7 times, and the mode negative streak for Democrats was a -1 and -2, which they incurred 8 times. Figure 8 helps flesh these frequencies out. But to examine this volatility more closely, and to examine the surge and decline phenomenon over shorter intervals of time, I will now break down the 14

16 net seat changes. 5 Tests 5.1 Matching Surges and Declines For the overextension/exposure thesis to be true more generally, and for my overarching hypothesis that surges will be cancelled out by roughly equal declines to be true specifically, the surges should be matched by near-equal declines. Note that I do not make a time judgment here. At least for the moment, I am not testing surges and declines in consecutive elections, rather I do so overall 19. If the decline portion of surge and decline is caused, generally speaking, by a return to the equilibrium point, declines should mimic the surges 20. This indeed is borne out by the data in two forms. Across the time series, for all elections in which Democrats net gained senate seats, they averaged a gain of about five seats (4.97). For all elections in which Democrats net lost senate seats, they averaged a loss of about five seats (-4.71). Granted, this is across the entire time series, not election to subsequent election, but it shows a consistent return to a static equilibrium point. Additionally, streaks of Democratic seat pickups are roughly mimicked by streaks of Democratic seat losses. While the average positive streak was a gain of between 8 and 9 seats (8.65), the average negative streak was a loss of between 7 and 8 seats (-7.46). Implications of this are twofold. First, at least over time, surges equal declines. Second, and more specifically, the surges equal the declines in terms of both net seat gains to losses and streaks of net seat gains to losses. Because these results are time irrelevant, meaning they could be election to subsequent election (net gains) or groups of elections to groups of elections (streaks), they hold irregardless of whether one gravitates to the Angus Campbell midterm-based version of surge and decline or Oppenheimer et al. s exposure thesis. 19 See Section 5.5 for this type of test. 20 See Section 5.4 for a discussion and test of mimicking surges and declines across the same class of senate seats. 15

17 5.2 Different Types of Elections Similar average seat gains (for Senate Democrats) across both types of elections midterm and presidential indicates either the electorate is similar in midterm and presidential elections, or even if the electorate is different, the effects of differing electorates cancels out over long periods of time. A basic bivariate regression of election type on Democratic (net) seat changes from previous congresses shows that there is no statistically significant relationship between net Democratic seats in congresses after midterm or presidential elections 21. Additionally, the mean Democratic (net) seat change from the previous congress is similar for midterm and presidential elections, around 0.3 Senate seats 22, meaning the average Democratic pickup of seats is similar for midterm and presidential elections, all else equal. Additionally, in line with the theory of surge and decline, whereby the president s in-party loses congressional seats in subsequent elections, the mean Democratic (net) seat gain when the president is a Democrat is a net loss of seats, while the mean Democratic (net) seat gain when the president is a Republican is a net gain of 1.09 seats. This means Senate Democrats, on average, lose Senate seats under Democratic presidents and gain them under Republican presidents 23. These numbers are shown in Table 1 below (Mean 1). But the former measure does not control for the party of the incumbent president, and the latter measure does not distinguish between presidential and midterm elections. Controls of the sort are necessary to parse out party and election-type effects simultaneously. Controlling for both election type and party of the president (the last four rows of Table 1) produces expected results, different from the ones above without controls. Senate Democrats have 21 This is an ordinary least squares (OLS) model of the NetDem variable as the dependent variable and the midterm and presidential dummy variables as the independent variables. 22 The mean Democratic (net) seat change from the previous congress for midterm elections is seats. The mean Democratic (net) seat change from the previous congress for presidential elections is seats, for an almost negligible difference of These averages are also consistent with the idea of midterm elections as referenda on the president s party, midterm elections as the electorate balancing the legislative and executive branches, thereby checking the president s party, and the absence of presidential coattails in midterm elections. With a Democratic president, on average, Senate Democrats lose seats in midterm elections. With a Republican president, on average, Senate Democrats gain seats in midterm elections. 16

18 their largest losses (-1.67 seats won) in midterm elections during Democratic administrations, and largest gains (3.12 seats won) in midterm elections during Republican administrations. Furthermore, in line with the coattail effects discussed in the literature, Democrats gain Senate seats (0.625 seats) in presidential elections in which a Democrat is elected or reelected president, and lose Senate seats (-0.94 seats) in presidential elections in which a Republican is elected or reelected president. Though, except for gains of about 3 seats in midterm elections under Republican presidents, and losses of about 1 or 2 seats in midterm elections under Democratic presidents, average gains and losses are rather small, most less than 1 whole seat. Given that these are averages across the entire time series, spanning dozens of elections, large surges could be cancelled out by almost equal declines, resulting in tiny averages. Dropping the seven congresses before implementation of the 17th amendment (shown in Mean 2) results in similar averages for all except both election types with Republican presidents, where the signs switch. This is likely because the large gains Senate Democrats made in the few congresses directly before the 64th Congress were not included in the average, while the large losses Senate Democrats sustained in the few congresses directly after the 64th Congress were 24. Table 1: Mean Democratic Seat Gains Election Type President s Party Mean 1 Mean 2 Midterm Presidential Democratic Republican Midterm Democratic Midterm Republican Presidential Democratic Presidential Republican Note: Mean 1 averaged across entire time series, Mean 2 after 17th amendment. 24 Senate Democrats gained 20 seats in a row preceding the 64th Congress. These substantial gains were not included in the Mean 2 averages. 17

19 5.3 A Measure of Overextension Looking to the literature for methods of measuring what it means for parties to be overextended, Converse (1966) uses survey data to gauge expected vote share for each party in congressional elections, allowing him to examine how far above that expected vote share a party received in an election. But he does not estimate the exact number of Senate or House seats each party should have. He says Democrats are expected to get X% of the vote in year Y, not that Democrats should control 53 Senate seats after the elections of year Y. Converse also examines only a short time series across the mid- to late-1950s. Even if we wanted to replicate Converse s study across our time series, the survey data he uses is not available dating back to Instead of this method, Oppenheimer et al. (1986) use Converse s concept of the normal vote to generate a measure of overexposure by subtracting out the equilibrium number of seats from the actual number of seats. The difference between the two is the overexposure, expressed as a number of seats in which the party is overexposed going into an election. This measure will not work here either though, as the actual equilibrium number, which is used as the normal vote from which to gauge the overextension by, is generated once across their entire data. This assumes the equilibrium level, and thereby the overextension measure generated from it, is universal to the entire time series in the analysis. As my time series includes more than double the years in their time series, I am not comfortable making this assumption. Consequently, we cannot necessarily directly measure what it means for a party to be overextended. It is a mostly subjective measure after all. As Supreme Court justice Potter Stewart once said, it may be one of those things that I know it when I see it. For instance, holding 76 of 96 seats in the 75th Congress, it is safe to say Senate Democrats were overextended. Over the next five congresses their majority slimmed down to a minority of 45 seats. On the other token, holding only 31 of 92 seats in the 60th Congress, it is fair to say Senate Democrats were not overextended. But what about when one party controls only a slim or modest majority, say 54 seats? At quick glance one might suggest that is not 18

20 overextension, as the party only controls a bare majority, not a 56 or 58 or 60 seat Senate majority. Upon further examination, if one knew the party (net) gained 10 seats in the two elections preceding the congress in which they controlled 54 seats, it could mean the party is actually moderately overextended and would be ripe for surge and decline to bring gravity to their majority in the next few elections. Looking at streaks and volatility over short time periods saves us from making a one-size-fits-all judgment call about what precisely it means to be overextended, as overextension is relative 25. What is considered overextension in one era of the time series may not be considered overextension in another. Punting on the question of pinning down a specific and arbitrary number to gauge overextension by helps, not hurts, our understanding of surge and decline because it prevents an arbitrary number from obscuring some cases of surge and decline. To address this reality I turn first to congresses by class and then to volatility in seats over short time periods. 5.4 A Story of Classes? Unlike the House of Representatives, whose seats are up for election every two years, Senate seats are up for election every six years. This results in only 1/3 of all Senators (called classes) standing for reelection in each election cycle. Consequently, the surge and decline Angus Campbell predicts from one election to the subsequent election is not comprised of the same batch of senate seats; rather, it is either Class 1, Class 2, or Class 3 senate seats, in a threecycle rotation. The senators who stand for reelection in a presidential election are not on the ballot again until a midterm election after the next presidential election. Absent being elected in a special election, senate seats in each class are contested in a rotation of presidential and then midterm elections, or vice versa. For instance, North Carolina Democrat Kay Hagen 25 We could look at variance around the mean to check for overextension, but this is a difficult comparison to make across the entire time series. An example demonstrates this. Over the entire time series, Democrats held a mean number of 51 seats while Republicans held a mean number of 46 seats. If we look at the 76th Congress, for instance, Democrats held 76 seats, 25 seats more than their mean. But relative to the mean for the few congresses before and after the 76th Congress, which has a much higher mean, they may not be as overextended as they seem though they certainly are. For this reason we may need to establish local means to distinguish over-extension. We could then look at variable around that local mean. But how many congresses we included before and after the congress of interest is a judgment call. 19

21 won her senate seat in the 2008 presidential election and then stood for reelection (and lost) in the 2014 midterm election. If different forces are at play in presidential vs. midterm elections, as the literature suggests, and a six-year gap exists between a senator s election and reelection, wild swings in senate seats may exist within classes. To be slightly more explicit in modeling the party change for the same class of senate seats, I compare party gains and losses in a Congress with those from the last time a particular set of seats was up for reelection. Figure 9 plots Democratic (net) seat gains and losses, first for each of the three classes of senate seats, along with lowess lines to show trends over time, followed by a plot of the trends over top of each other for direct comparison. The trends indicate that all three classes exhibited mostly similar periods of surges and declines, albeit with a range of ups and downs at different periods 26. Looking closely at the plots reveals wild swings in Senate seats, even within a particular class. Most consecutive dots cross the zero line, meaning they are gains followed by losses, or losses followed by gains. For each of the three classes, in the majority of elections, a Democratic (net) seat gain follows a loss the last time that class stood for reelection, and vice versa. In the few cases with gains followed by further gains, or losses followed by further losses, the further gain or further loss usually followed a very large gain or large loss, and is much smaller (i.e. is closer to the mean), indicating a return to the equilibrium. The results of these comparisons are consistent with my theoretical expectations. Table 2 below shows means of Democratic (net) seat gains broken down by class, controlling for type of election and party of the president. Regardless of class, in midterm elections under a Democratic president, Senate Democrats lose seats. On the opposite token, in midterm elections under a Republican president, Senate Democrats gain seats, regardless of 26 Class 1 Senate seats appear to be in a long term surge for Democrats while Classes 2 and 3 appear to be in a slight decline. This is likely due to the nature of which states have seats in these classes. As of the 115th Congress, Class 1 is more than twice as Democratic as Class 2 and almost twice as Democratic as Class 3. Having gained all but 8 of the 33 seats in this class, Democrats are likely overextended, holding seats in conservative or Republican states including Indiana, Missouri, Montana, North Dakota, and West Virginia. Incumbent Democrats holding these seats last stood for election or reelection in 2006 and 2012, strong years for Democrats overall, and will next stand for reelection in 2018, a midterm election. Due to this, Class 1 is ripe for a Democratic decline. 20

22 class. In presidential elections in which a Republican is elected (or reelected), Democrats lose senate seats, again regardless of class. This is in line with the coattails theory of a Republican presidential candidate pulling Senate Republicans up with him. The most puzzling result is presidential elections with Democratic presidents. Democratic presidential candidates do not appear to have the coattail pull that Republican ones do, as one class gains seats, one loses seats, and the other breaks even. It is not clear why this is the case. It could be due to the nature of the seats themselves, both the geographic distribution of seats in each class, as well as the partisan realignment of the South in the Jim Crow era. Overall, Democrats have their largest average gains almost 5 senate seats in class 1 midterm elections featuring Republican presidents, and their largest average losses almost 2.5 senate seats in class 2 midterm elections featuring Democratic presidents. Now I will turn to the volatility of seats over shorter intervals of time. Table 2: Mean Democratic Seat Gains by Class Election Type President s Party Class Mean Midterm Democratic Midterm Democratic Midterm Democratic Presidential Democratic 1 0 Presidential Democratic Presidential Democratic 3-1 Midterm Republican Midterm Republican Midterm Republican Presidential Republican Presidential Republican Presidential Republican Note: Mean 1 from Table 1 broken down by classes. 5.5 Volatility of Short Time Periods The figures below, Figures 10-13, show the volatility of seat changes over short time periods by breaking down Figure 5 above into smaller time intervals (i.e. fewer Congresses). The 21

23 most recent six congresses are shown in Figure 10, which details the more-than-modest Democratic gains of 6 and 8 seats heading into the 110th and 111th Congresses, respectively, and then the losses of 6 and 9 seats heading into the 112th and 114th Congresses, wiping out the gains. The gains were made in a midterm election during a Republican presidency, during two presidential elections in which a Democrat (Barack Obama) was elected and then reelected president, and during a presidential election in which the Democratic candidate won 3 million more popular votes than the Republican candidate (though the Republican candidate won the electoral college). The loses were sustained in a presidential election in which a Republican (George W. Bush) was reelected, and in two midterm elections under a Democratic president. Additionally, the largest gain during the period, entering into the 111th Congress, was matched by an almost equal loss, entering into the 114th Congress, when the same class of senators stood for reelection 27. All of these losses and gains are entirely consistent with the theories of the causes of surge and decline, and are further confirmation for my hypothesis. A loss to gains is shown in Figure 11, where Democrats lost 17 seats in a row going into the 96th through 98th Congress, but then made modest gains of 2 seats and 8 seats going into the 99th and 100th Congresses, respectively. These gains were not enough to nullify the most immediate losses though, as the severity of the decline was not cancelled out by the surge. Losses entering the 96th Congress were sustained during the midterm election of Democratic President Jimmy Carter and a presidential election in which he lost the presidency to Republican Ronald Reagan. The loss of 12 Senate seats for Democrats in 1980 gave Republicans the Senate majority in the 97th Congress, their first majority of either chamber since Democratic gains for the 100th Congress were made during the second midterm election of a Republican president, and they regained control of the chamber until Two elections surge and decline cannot directly parse out are the 27 Of the five Democratic incumbents that lost reelection, three were first elected to the Senate in 2008 beginning their terms in the 111th Congress and three of the five held seats in states Barack Obama lost in both of his presidential elections, and the fourth held a seat in a state Obama won in 2008 but lost in