Compared to recent presidents, only George H. W. Bush was as unpopular when standing for re-election as Donald Trump is today. Bush lost.
Recent opinion polls, especially ones taken after the government shutdown had worn on for a few weeks, show Donald Trump considerably “underwater” in job-approval polls. On December 21st, the day before the shutdown, the running average on Trump’s job approval numbers at FiveThirtyEight put him at 42.2 percent approving, and 52.7 percent disapproving, for a “net approval” score of -10.5 (=42.2-52.7).
Since then the gap between Trump’s approval and disapproval scores has substantially widened. Today, January 27th, FiveThirtyEight reports his approval score has fallen three points to 39.3 percent, while his disapproval score has increased substantially to 56.0 percent. His net approval score has now fallen to -16.7.
Few recent Presidents have plumbed these depths of public opinion as Donald Trump. This chart presents the net approval scores of recent Presidents as measured by Gallup at three different times in their first terms: when inaugurated, at the time of the first midterm election, and once again at the time of the next Presidential election.
Every President except Trump began his term of office with a positive net approval score that deteriorated over the next two years. George W. Bush showed the smallest decline because his approval had received a thirty-point boost after the 9/11 attacks. Being popular upon inauguration is no guarantee of continued popularity as Ronald Reagan and, especially, Barack Obama discovered. However they were both able to recover and reach positive territory in the Presidential election that followed.
Only George H.W. Bush has stood for re-election with a net approval score below negative twenty like Trump’s current figure. Bush, of course, lost the 1992 election to Bill Clinton. George W. Bush managed to be re-elected with a net approval score of zero. The remaining Presidents in the chart all had positive approval scores at the time of re-election and were sent back to Washington for a second term.
Donald Trump has not seen a net approval score above zero since a week or two after he was inaugurated in 2017. It is hard to fathom how he can recover even to a net zero score like George W. Bush had in 2004, never mind reaching into positive territory. Both Clinton and Obama managed to win re-election with net approval scores in the single digits. Even that relatively low hurdle seems pretty distant for Donald Trump at this point.
One reason we are unlikely to see a substantial movement in Trump’s direction is the large number of people who report “strongly” disapproving of Trump’s performance. The newest Washington Post poll shows that most peoples’ opinions of Trump fall into either the strongly approve (28 percent) or strongly disapprove (49 percent) categories. Only nine percent report they “somewhat” approve of his performance, and another nine percent “somewhat” disapprove.
Obama could recover from his 2010 “shellacking” because fewer people chose the “strongly disapprove” option when asked about him prior to the 2010 midterms. Trump has fewer chances to bring people back to his side because opinions about him are much more locked in stone.
Donald Trump could, of course, confound all his observers as he did in 2016. However it is unlikely he will be able to run against as unpopular a candidate as Hillary Clinton turned out to be.
Observers of the 2020 primaries expect a huge crop of Democratic candidates to declare before next February’s Iowa caucuses. History tells us that few of those hopefuls will still be contending for the nomination by mid-March.
Pundits have routinely suggested that the field for the 2020 Democratic Presidential nomination might range upward of a dozen candidates. As of today, January 22, 2019, a total of seven candidates have announced their intention to run — Julian Castro, John Delaney, Tulsi Gabbard, Kirsten Gillibrand, Kamala Harris, Richard Ojeda and Elizabeth Warren. Wikipedia lists another twenty people who have “publicly expressed interest” in running including Joe Biden, Cory Booker, Sherrod Brown, Amy Klobuchar, Michael Bloomberg, Jay Inslee, Beto O’Rourke and Bernie Sanders. The field could certainly exceed a dozen candidates before the Iowa caucuses take place on February 3, 2020.
The Democrats saw eight contenders in the 2004 primary race that was eventually won by John Kerry. Michael Dukakis in 1988 and Barack Obama in 2008 fended off six other competitors in their paths to winning the nomination.
Most of these competitors fell by the wayside early. In 1992, 2004, and 2008 just two candidates remained by March 15th. In 1988 Al Gore and Paul Simon continued their contests against Dukakis until April, while Jesse Jackson maintained his symbolic quest into the Convention .
This drastic winnowing of candidates reflects a variety of forces. All the campaigns compete for funds and staffers. Candidates who fail to make a strong showing in the early races see their access to contributions and staff dry up.
These resource limitations are exacerbated by the need to attract and maintain media attention. While it might be possible to cover a dozen candidates at the outset, media organizations inexorably must target their resources to a much smaller number of candidates deemed “viable” by the press and party professionals.
So while we may see long lists of potential Presidential candidates lining up in Iowa and New Hampshire, it is hard to imagine there being more than three or four candidates remaining in the race after Super Tuesday, March 3rd.
For comparison, here is the equivalent chart for the 2012 and 2016 Republican primaries. Fully seventeen candidates took to the stage in the Republican primary debates in late 2015. By March 15th just three candidates remained standing — Donald Trump, Ted Cruz, and John Kasich. A month later Trump stood alone.
There have been eleven midterm elections when House retirements by one party outnumbered those of the other party by six or more seats. In all but one election the party with the greater number of retirements lost seats.
In the months before the 2018 election forty Republican House Members chose to give up their seats rather than pursue re-election, by far the greatest Republican exodus since the New Deal. The previous Republican record of twenty-seven was set in 1958 during the Eisenhower recession. Democrats once saw forty-one of their Members choose to depart the House in 1992 when Clinton was first elected.
However it is not the volume of a party’s retirements that matter as much as the excess of retirements from one side of the aisle or the other. To be sure, Members of Congress retire for many reasons. Age and illness catch up with the best of us. Some Members give up their House seats to seek higher office like Kirsten Sinema and Beto O’Rourke did this year.
Still, Members also pay close attention to the winds of politics for fear they might be swept out of their seats. Some choose to retire rather than face an embarrassing defeat in the next election. Such “strategic retirements” might prove a plausible bellwether for future elections. If many more Members of one party are leaving their seats than the other, that might bode ill for the party’s results at the next election.
One thing is certain, retirements prove useless for predicting House results in Presidential election years. Presidential politics overwhelms any effect we might see for strategic retirements in House elections.
The picture looks different in midterm elections. Years that saw more Republicans retiring compared to Democrats were also years where more seats swung from Republican to Democratic hands. This past election joins 1958 as years when an excess of Republican departures from the House foretold a substantial loss of seats at the next election.
The horizontal axis measures the difference between the number of Republican Members who left the House before an election and the number of Democrats who gave up their seats.* The vertical axis shows the swing in House seats compared to the past election. For instance, in 2018 forty Republicans and eighteen Democrats left the House, for a net retirements figure of +22 Republican. The “blue wave” swung forty seats from the Republicans to the Democrats, about nine fewer than the best-fit line would predict.
Some readers might ask whether that nine-seat deficit reflected Republican gerrymandering in the years since the 2010 Census. I simply cannot say. The likely error range (the “95% confidence interval”) around the prediction for any individual year averages about a hundred seats.** With that much variability, detecting things like gerrymandering effects is simply impossible.
As a bellwether, then, retirements seem pretty useless. They appear to have so much intrinsic variability that any effects of strategic decision-making by Members remain hidden. Suppose we group elections by the difference in retirements. Will we see any stronger relationship with the election result than we have so far?
In the six elections where the number of retiring Republicans outnumbered retiring Democrats by six or more Members, the Republicans lost seats in five or them. The same held true for elections when six of more Democrats retired compared to their Republican colleagues. The Republicans gained seats in all five of those elections.
So retirements can prove a useful predictor of future election results if we limit our attention to the more extreme years where one party’s retirements outnumber the other by six or more. The party with the excess of retirements has lost ten of the eleven elections fought in such circumstances.
**The height of the bars depends on the overall “standard error of estimate,” in this case 23.8 seats, the size of the sample (21 elections), and the difference between the number of retirements in a given year and the mean for all years. The confidence intervals average about plus or minus fifty seats for any given election.
Americans saw their real disposable personal incomes grow by a fairly average two percent over the past year. With that mediocre rate of income growth, a “normal” President would have had an approval rating around forty-eight percent at the time of the election. Donald Trump’s forty percent favorability probably cost his party at least four percent of the popular vote, or about half the Democrat’s popular-vote margin of just over eight percent.
The state of the economy was considered one of the few items on the positive side of the ledger for Congressional Republicans in 2018. The stock market accelerated after Trump’s election in 2016; the new tax bill was expected to put cash in peoples’ pockets; and, incomes overall continued to rise as they had since 2010. All of these should have helped Republicans this year. The question is how much.
The effect of the stock market rise is somewhat easy to dismiss as only half of American adults own stocks. And the market has experienced considerable volatility this year compared to the steady march upward during the first year of the Trump Administration. So while the half of Americans with stock portfolios are certainly better off today than they were in November of 2016, the future looks more dicey.
However the evidence for a relationship between stock prices and presidential popularity is, at best, mixed. Back in 2009 Gallup found little correlation between the Dow-Jones Industrial Average and the popularity of recent Presidents. Nate Silver at FiveThirtyEight also expresses skepticism about a stock-market effect.
The benefits of the Republican tax cut also have had limited effects when it comes to the broad voting populace. Some forecasts expected the cuts to stimulate investment and consumption and grow the overall economy. However, even supporters of the plan do not expect those broader benefits to appear for some years to come. Meanwhile, critics of the plan focus on how most of the gains from the tax cut have been funneled into stock buy-back plans that boost companies’ share prices and executive compensation instead of investments in the real economy. “Perhaps that’s why voters aren’t enthusiastic about the tax cuts,” writes a columnist in Forbes. “People just aren’t getting any real economic benefits from the tax cuts and they know it.”
As I’ve discussed here before, political science research often treats changes in real per-capita disposable personal income as a useful shorthand for the economic welfare of an “average” American. That measure turns out to have a weak, though measurable, influence on Presidential popularity. This chart presents Gallup’s job-approval measure in the week or two before an election and the one-year change in real per-capita disposable income as reported by the Bureau of Economic Affairs. I use the third-quarter figure since it covers the period closest to an election.
The scatter around the line in this chart testifies to the weakness of the relationship. The R2 value measures the percentage of the variance in approval that can be accounted for by income changes; here it is about eleven percent. The slope of the regression line, two, suggests that a one-percent increase in real disposable personal income is associated with, on average, a two-point improvement in a President’s popularity.
One thing made immediately clear by this chart is that average Americans saw no extraordinary growth in their incomes over 2018. Real per-capita disposable income grew from $42,866 in the third quarter of 2017 to $43,718 at the end of September. That gain of $852 was just short of a two-percent increase over the year before, and a bit below the 1944-2018 historical average of 2.25 percent.
Still, the chart shows that Trump was less popular than economic conditions would predict. Approval for a “typical” President presiding over an economy showing two-percent in growth should run a bit over 48 percent, not the 40 percent for Trump reported by Gallup in the week before the election.
Earlier this year I presented some results that tied together Presidential popularity and support for the President’s party on the so-called “generic-ballot” question.* We can use that model to imagine how the election might have transpired had Trump been a “normal” President and an economy with two-percent personal income growth. In that model a one-percent change in “net approval,” the difference between the percent approving versus that disapproving of the President, improves the Democrats’ margin on the generic-ballot measure by about 0.3 percent.
Trump’s net-approval rating just before the election stood at 40-54, or -14, according to Gallup. A President with 48 percent approval will likely have an identical disapproval rating, 48-48, after accounting for the four percent or so who report having no opinion. That makes the net approval for this hypothetical President zero, meaning Trump’s net approval is fourteen points below a normal President’s. Trump running fourteen points behind a normal President on net approval probably expanded the Democrats’ margin in the popular vote for the House by about four percent (4.2 = 0.3 X 14). That accounts for half the Democrats’ margin of victory in November, 2018.
*The generic-ballot again did a pretty good job of predicting the actual margin of victory in the election for the House of Representatives. The RealClearPolitics average of generic-ballot polls showed the Democrats with a lead of 7.3 percent. Omitting the obvious Rasmussen outlier, which predicted a one-point Republican victory, brings the average up to 8.2 percent, nearly identical to the actual margin of 8.5 percent.
Democrats have won, on average, about eight fewer seats in each election since 2010 than we would expect given their popular vote. The surge in Democratic votes this year might have cut that deficit down to two, but it is more likely there was no effect at all.
Before the November election, some commentators argued that a surge in turnout could negate the effects of Republican gerrymandering after 2010. Of course, this argument only makes sense if there were a larger increase in Democratic turnout than Republican turnout. A proportional increase for both parties would leave the seat results unchanged.
It is certainly true that the Democratic vote for the House of Representatives was considerably greater in 2018 than it was in 2014. In fact, Democrats cast nearly as many votes this month as they did for Hillary Clinton two years ago. Compared to the 2014 midterm, the Democrats increased their vote by over fifty percent. Republicans also turned out in higher numbers, recording a vote for House candidates some 23 percent above their 2014 totals. (Figures for 2018 from Dave Wasserman of Cook Political Report.)
To test this, I added a term for the 2018 election to my standard model of seats and votes described here and here. I use the “logits” of Democratic seats and votes won with “dummy variables” to represent reapportionment periods. The basic model, with 2018 included, produces this chart showing the number of Democratic seats won or lost compared to what we would expect based on the national popular vote won by that party. Some periods, like 2002-2010, show no significant excess gains or losses. Others like 1942-1950 and 2012-2018 show substantial effects.In the five elections beginning in 1942, Democrats routinely won nearly nine more House seats than their popular vote would predict. Republicans picked up a number of state legislatures in the 1952 election and erased this deficit for the decade to follow. From 1962 through 1990, Democrats were again advantaged, but by a diminishing margin over time. The elections fought between 1992 and 2010 showed no systematic bias for either party After the 2010 Census and the “shellacking” of Democrats in both national and state elections that year, Republicans were able to draw district maps that gave their party just short of eight “excess” seats in the House.
By adding another variable to represent just the 2018 election, it does indicate a diminished effect compared to the 2012-2018 average. However this effect fails to reach any conventional level of statistical significance (t = 1.07).
One other question we might ask is what the 2018 outcome would have been had the neutral results for 1992-2010 continued on into elections held since the 2010 Census. While the chart above shows that Democrats lost on average about eight seats to gerrymandering beginning in 2012, the estimated effect for this past election is just short of fourteen seats, the result of the Democrats’ substantial victory in the popular vote.
Democratic governors in seven “red” states, and Republicans in two “blue” ones, will help insulate 81 likely Congressional seats from gerrymandering after 2020. Redistricting for another 61 seats will likely remain entirely in Republican hands compared to just seven seats in states with unified Democratic control.
Yesterday’s election helped limit potential gerrymandering after the 2020 Census in a half-dozen states but not, unfortunately, in the largest prizes. Democrats appear to have failed in their bids to win the gubernatorial elections in Florida, Georgia, and Ohio, and in all three states Republicans maintained their control over the state legislatures as well. Barring Democratic legislative victories, all three of those states will remain prospects for Republican gerrymanders in 2021.
Democrats did win or retain the governorships in Colorado, Connecticut, Illinois, Maine, Minnesota, Pennsylvania, and Wisconsin and will likely face either Republican or split legislatures when redistricting maps are redrawn after 2020. Together those states will probably encompass 64 Congressional districts after reapportionment. Two states, Maryland and Massachusetts, with a likely total of seventeen seats, will see Republican governors facing off against Democratic legislatures in 2021. I would not be surprised to see a new Republican representative sent to Washington after the 2020 Census from both these states which now have uniformly Democratic Congressional delegations.
Because the Democrats failed to win the governors’ races in Florida, Georgia, and Ohio, all three states will be prime targets for Republican gerrymanders in 2021. (Iowa, with its four Members of Congress, matters much less.) Ohio and Florida accounted for three to five “excess” Republican seats after the 2010 Census, and Georgia may have added another. Because the Democrats fared less well in these larger states, the GOP will be drawing district lines for 61 of the 149 seats in “trifecta” states where they control both the governor’s mansion and the two houses of the state legislature.
1Both Michigan and New York appeared in the earlier version of this chart. However both states will be using nonpartisan redistricting commissions in 2021 and have been excluded from the analysis here.
Elections this fall may limit the extent of gerrymandering for some 200 House seats after the 2020 Census.
Americans will elect thirty-four governors to four-year terms this fall, giving them all a say in how their states’ Congressional and legislative districts will be drawn after the 2020 Census.1
In 2010, Republicans took control of many state legislatures and governors’ offices, which offered them the opportunity to draw district maps that favored their party. During 2011-2012 when those maps were drawn, Republicans controlled both the executive and legislative branches in nineteen states and appear to have won fourteen percent more House elections in those states than we might expect based on historical data. Democrats controlled just eight states and won four percent more seats than expected. In thirteen states partisan control was split between the branches, and there the partisanship of the governor appears to have been the controlling factor. In split control settings, the governor’s party won about six percent more House seats after 2010 than expected.
The election in November has the potential for influencing how dozens of Congressional district lines will be drawn in the aftermath of the 2020 Census. In a number of states Democrats are poised to break the Republicans’ lock on control of government by taking back the governor’s mansion.
I have categorized the governors’ races by their competitiveness using the most recent ratings for those races as compiled at Wikipedia. Since the next redistricting will involve the results of the 2020 Census, I have used state-level population projections to estimate the number of seats each state will be awarded after reapportionment. Most states’ representation in Congress will not change, but a few states like Texas are projected to add seats, while Rhode Island may lose one of its two representatives in the House. Forecasting the results of state legislative races this fall, and more importantly two years hence, is obviously a dicey proposition. I have instead assumed that all legislatures will be controlled by the same parties that control them now. Based on these data I estimate that redistricting for some 188 seats, or 43 percent of the House, may be affected by the results of this year’s gubernatorial elections.
An “S” (“split”) code in the legislative column indicates that the two houses of the state legislature are held by different parties.
The first two columns of this table present the likely outcome of this fall’s race for governor in each of these states. The “Consensus Rating” is based on translating each prognosticator’s ratings like “Safe Republican” or “Likely Democrat” into a numerical score and averaging them. I also present the most recent ratings for each race from the well-known site, FiveThirtyEight.com.
In these more competitive states Democratic candidates for governor appear to be well-positioned to win back these offices from the Republicans. Only in Massachusetts and Maryland are we likely to see Republican governors winning re-election while their states’ legislatures remain in Democratic hands. In ten states from Georgia to Michigan to New Mexico, Democratic candidates are poised to oust Republican governors even if their states’ legislatures do not change hands. As my results from the post-2010 redistricting showed, governors appear to have most of the clout in redistricting battles, reducing the chances of gerrymandering in places like Michigan, Ohio and Florida, all of which had an “excess” number of Republican seats beginning with the 2012 election.
1Governors in New Hampshire and Vermont serve two-year terms.
In “split-control” states, Republicans won 6.2 percent more seats than expected when they held the governorship; when Democrats held that office, Republicans won 6.5 percent fewer seats than expected.
Americans will elect thirty-four governors to four-year terms this fall. They will still be in office after the 2020 Census and will have a say in how states redraw their Congressional and legislative district plans. All states where legislatures draw district lines except North Carolina grant the governor the power to veto any plan. In states where control over the branches is split between the parties, this process should lead to compromises acceptable to both parties. As well see, however, the evidence from the redistricting after the 2010 Census suggests governors hold all the cards.
In nine states, politicians play no direct role as redistricting is left up to nonpartisan commissions. Courts, too, can override the lines drawn by legislatures. This year’s dramatic redrawing of the lines in Pennsylvania follows similar judicial interventions in Florida and New York. The New York decision affects my analysis since it applied to elections beginning with 2012. (The other decisions have yet to come into force.) I have added New York to the commission list, but have analyzed it, and California, separately as well.
I have also excluded the seven states which have only one Congressional district like Wyoming and Alaska since gerrymandering is not possible with no lines to draw. That leaves 43 states which can be categorized as follows:
- maps drawn by nonpartisan commissions or courts (8 states);
- maps drawn by Republican legislatures facing Republican governors (16 states)
- maps drawn by Democratic legislatures facing Democratic governors (6 states); and,
- maps drawn when either the houses of the legislature were held by opposing parties, or where the legislature had unified control but faced a governor of the opposite party (13 states).
To avoid relying too heavily on a single year, I have added together the votes cast for Republican and Democratic House candidates in each group of states for the 2012-2016 elections. I have applied the same method to seats won, again summing up the number of Republican and Democratic seats won across all three elections. That method produces these results:
House Votes and Seats Won 2012-2016 by Redistricting Method
The first column reports the Republican percent of the total two-party popular vote summed across the three elections, 2012, 2014, and 2016. In the sixteen states where Republicans held both houses of the state legislature and the governorship, they won 56.5 percent of the two-party House vote and 71.7 percent of the seats. In solidly Democratic states the Republicans won both a minority of the popular vote and of the seats awarded. The results for commissions and courts is complex; I will deal with it in a later article.
In various articles here I have described the natural inflation of the proportion of seats won due to the operation of our first-past-the-post electoral system. As parties win larger and larger proportions of the vote, they gain an ever-increasing share of seats. I have estimated this inflation factor using both biennial election results back to 1946, and across states in 2012. Both methods produce equivalent results, for instance.
To estimate the share of seats awarded you need only square1 the value of the ratio (Republican Votes)/(Democratic Votes) to get the ratio (Predicted Republican Seats)/(Predicted Democratic Seats). This approach gives rise to the third column in the table, the proportion of seats that are predicted to be won by the Republicans after applying this “square law” rule.2 In the entry for Republican control, that party’s 56 percent share of the House vote should produce a share of about 63 percent of seats. In practice, the Republicans won nearly 72 percent of the seats. The final column measures the over- or under-representation of the Republicans in the House as a percentage gain or loss compared to the predicted share. In this case, the Republican’s 72 percent is about 14 percent higher than the expected 63 percent. This figure provides a criterion for evaluating how over- or under-advantaged a party was compared to expectations.
The normal expectations for states with unified control are confirmed: Republicans win a disproportionate share of seats in states where they controlled the redistricting process, and won disproportionately fewer seats than expected in states where the Democrats were in control. Notice that the size of Republican advantage in states that party controlled is larger than the disadvantage the Republicans faced in states controlled by Democrats, +14 percent versus -4 percent.
By this measure states with some form of split party control show hardly any partisan advantage at all. Republicans won a share of the seats awarded in these states nearly equal to their expected share. However, it turns out this overall result hides a lot of significant variation.
We can identify two different forms of split control:
- ones where both chambers of the state legislature are held by one party but the governor is of the opposite party; and,
- ones where the chambers of the state legislature are held by different parties.
As it turns out there are nearly equal number of each type of split control; in seven states unified legislatures faced an opposition governor, while in six states the chambers themselves were split. Once we break out these various patterns, the power of governors becomes clear. In both types of split control, the governor’s party is disproportionately advantaged during redistricting. In fact, if we group these split-control states together simply by the partisanship of the governor, Republicans were over-represented in seats awarded by 6.2 percent where they held the governorship; when Democrats held that office, Republicans were under-represented by 6.5 percent.
These results are rather striking. They suggest that opposite-party governors can force a redistricting map that is actually more favorable to the governor’s party than to the legislature’s. Similarly when the two legislative chambers are held by opposite parties, it is again the governor who appears to determine which map wins approval. It appears the governor’s veto is a more powerful weapon in the fight over Congressional district lines than the legislature’s control over drawing the lines themselves. This fact could weigh heavily over redistricting fights in states like Colorado, Michigan, Florida and Georgia where Democratic governors may win election and end up facing Republican legislatures. In Massachusetts and Maryland the reverse will likely hold true.
1The longitudinal estimate was 2.04; the cross-sectional estimate was 2.08. For simplicity I have rounded down to two, which is well within the confidence intervals for each estimate of beta.
2The tendency for first-past-the-post systems to disproportionately advantage the winning party was first observed in elections in the United Kingdom. There the coefficient reached three, giving rise to the name “cube law” rule, since cubing the ratio (Labour Votes)/(Conservative Votes) does a good job of predicting the ratio (Labour Seats/Conservative Seats). Following this tradition, I have named the US version of this relationship the “square law” rule.
I have compiled the results of generic-ballot polls taken near to an election and compared them to the actual division of the Congressional vote. The table below presents the margin between support for the President’s party and support for the opposition. For each election I have used about half-a-dozen polls from different agencies taken just before voting day. Averaging the differences between these two quantities shows that these polls have fared rather well since 2002. The average deviation in the four midterm elections is 0.6 percent; in Presidential years that falls to 0.1 percent.
Still these averages hide some fairly wide fluctuations. In four of the eight elections the difference between the polls and the election results exceeds two percent. The error was especially egregious in 2006 when the polls predicted nearly a fourteen-point Democratic margin compared to 8.6 percent in the election itself.
In the most recent election, 2016, the polls predicted a slight positive swing in favor of the Democrats, but the outcome went slightly in the opposite direction. All the cases where the polls erred in picking the winner occurred in Presidential years and usually when the polling margin was close. The four polls taken during midterm years all predicted the correct winner, though the size of the victory was off by more than three points in two of those elections.