Department of
History & Political Science



Receive regular updates
of this forecast until
Election Day
via Twitter

 

Presidential Election Forecasting

 

DeSart Long-Range Forecast Model

Dr. Jay DeSart, Associate Professor of Political Science - Utah Valley University


August 3, 2016 - FBI Report and GOP Convention give Trump an edge
This month's update gives Donald Trump a signficant edge over Hillary Clinton in the general election in November. Unlike last month, the model projects a clear victory for Trump in both the two-party popular vote (51.88 to 48.12) and Electoral College Vote (321 to 217). Overall, the model gives Donald Trump a 93.8% probability of winning an Electoral College majority.

There are two significant factors that help explain significant shift towards Trump since last month's update. First, the model uses all polls conducted during the previous month to generate it's forecast. July, by and large, was not a very good month for Hillary Clinton, until the very end. The first two weeks were dominated by the FBI's report about the investigation into her email server, and FBI Director Comey's testimony before the House Oversight Committee. While the decision not to indict her was good news for her, Director Comey was sharply critical of her, and this was brought into sharp focus in the news coverage of the Oversight Committee's hearing. In the first two weeks of July, the 6+ point lead that Clinton held in June had dwindled down to around 2 points. Then came the Republican Convention and the significant bounce that Trump received from that. By the time the Democratic Convention happened during the final week of July, Clinton's lead over Trump in the July poll average was just 1.4%. The significant bounce that Clinton received from her own convention was only reflected at the very end of the month. Only 6 of the 39 polls taken in the month of July came after the Democratic Convention. The result is that Clinton's lead over Trump for the month overall only increase to 3.9% . This was not enough to overcome the pro-Republican context of 2016.

Second, it is important to note that this is the earliest that the Conventions have been held since 1992. The sample of election data that this model is based on only relies upon data going back to 1996. As such, the model may be overly sensitive to the Convention bounces. The deeper lesson here is that Trump is shown to do substantially better when the public's focus is on Clinton's email scandal, as was pretty much the case for the first three weeks of July.

The charts below show the projected outcomes, as well as the distribution of possible popular vote and Electoral College vote outcomes based on this projection. As can be seen, the very negative month for Clinton shows up quite clearly in these distributions. Trump was very clearly advantaged by the events of July. In the popular vote distribution, there are no scenarios within the 95% confidence interval where Clinton wins more than 50% of the two-party popular vote, and only a very narrow sliver of outcomes within the 95% confidence interval of the Electoral College distribution gave Clinton an Electoral College majority. Whether that can be sustained, remains to be seen.
The bottom line here is that it must be remembered that this model merely projects what the expected outcome would be given the expected electoral context in a year when a party is defending the White House for a third consecutive term. Under normal circumstances, this should be a very tough uphill climb for Clinton and the Democrats. This year, however, is shaping up to be anything but normal.


     




July 3, 2016 - Model Predicts Electoral College Misfire
This month's update of my long-range forecast model is the first in which both parties have presumptive nominees pitted against each other. It shows Hillary Clinton continuing to swim against the pro-Republican context that, under normal conditions, 2016 would ordinarily represent. Interestingly, the model suggests an Electoral College misfire is possible, with Secretary Clinton losing the two-party popular vote by a very narrow margin to Donald Trump (49.5 to 50.5), but winning enough states to secure a slim Electoral College majority 272-266. Even so, in 100,000 Monte Carlo simulations, Trump wins at least 270 Electoral Votes 65.9% of the time. Given all this, the model continues to suggest that this race is a lot closer than perhaps would have otherwise been the case had a different candidate won the GOP nomination.

Presented below are the full details of the matchup, including the distributions of the projected outcomes for both the national two-party popular vote and the Electoral College vote. Clinton's popular vote point-estimate forecast of 49.5 has a margin of error of 1.21, which means that a popular vote victory is definitely within the 95% confidence interval. The bottom line is that this model continues to suggest that the election is still too close to call.



     


June 3, 2016 - General Election is virtual toss-up
This month's update of my long-range forecast model shows a slight tightening of the race. While Bernie Sanders still holds a slight advantage over Hillary Clinton in the matchups against Donald Trump, the model shows that both races are virtual toss-ups.

Once again, even though both Clinton and Sanders both held leads over Trump in the polls in May, the context of 2016 continues to drive the projections. The model does show that Sanders is more likely to defeat Trump than Clinton, but his probability of winning an Electoral College majority is only 54.3%. The overall conclusion remains the same: The November election will likely be very close regardless who the Democrats eventually nominate.

The full matrix of this month's projected outcomes is presented below:





May 3, 2016 - Context still favors Republicans
This month's update of my long-range forecast model shows very little change from the past month's. 2016 is still a change context year which should favor the Republicans. The model shows all three remaining Republican candidates defeating the Democratic frontrunner, Hillary Clinton. The Republican frontrunner, Donald Trump, continues to fare less well against Clinton in the model than either Ted Cruz or John Kasich, but it projects that he will win in a close election.
Bernie Sanders continues to perform better in the model's projections against the Republican field, but it still gives him a less than 40% chance of winning an Electoral College majority against any Republican candidate.

The Regime Age variable continues to be the driving force behind these projections, suggesting again that this year should be a good opportunity for the Republicans to take back the White House. Whether or not they will be able to capitalize on that remains to be seen.

The full matrix of this month's projected outcomes is presented below:





April 3, 2016 - Trump fades, Kasich most likely to win
This month's update of my long-range forecast shows that Donald Trump's star appears to be fading. The model shows that the current Republican front-runner would be defeated handily by Bernie Sanders and would lose to Hillary Clinton in a virtual toss-up and possible Electoral College misfire. On the other hand, the model indicates that John Kasich would easily defeat Hillary Clinton with a 99.99% probability of getting an Electoral College majority if he were to somehow manage to secure the nomination at a contested Republican Convention in July.

If Ted Cruz were to become the nominee, the model suggests that his fate would be in the hands of Democratic primary voters. If Hillary Clinton goes on to win the nomination, the model indicates that he would have a 90.1% probability of winning an Electoral Collge majority and becoming president. However, if the Democrats were to nominate Bernie Sanders, the model projects that Sanders would likely win that matchup with an 83.5% probability.

The complete details of the head-to-head matchups projected by the model are in the table below:




March 3, 2016 - Model update shows Republicans heavily favored
The latest update of my long-range forecast shows that the Republicans are heavily favored to win the Presidential election this Fall. The model shows that the each of the remaining GOP candidates are more likely than not to defeat Hillary Clinton, should she go on to be the eventual nominee. The model deems Marco Rubio and John Kasich as the two candidates who are most likely to win a majority in the Electoral College against Hillary Clinton, with Win Probabilities over 99.9%.

The picture is slightly brighter for Bernie Sanders, but the model still projects that he would more likely than not be defeated by every Republican except Ted Cruz. Marco Rubio has over a 95% probability of winning an Electoral College majority against Bernie Sanders. On the other hand, the model suggests that Sanders would possibly benefit from an Electoral College misfire against John Kasich and Donald Trump.

A few key notes about this month's update: You'll note that this forecast is significantly more favorable to the Republican candidates than other recent updates. There are two reasons for this. First, it's not so much that the Republicans' chances have improved as much as the model has changed. The forecasts prior to this month were based on coefficients derived from only the last 4 elections. Since the last update, I have obtained data for the 1996 election as well and have included them in the analysis to obtain the coefficients for this month's forecast. One major consequence of this was that the Regime Age variable has taken on slightly greater weight, which works to the benefit of the Republicans.

The other reason there appears to have been such a significant shift in the forecast towards the Republicans is that I discovered some issues in the data which led to a misspecification of the model. Simply put, the model had overstated the Democrats' chances in the December, January, and February updates because of problems in the data, not because their fates were improving.

At the very least what these results suggest is that this year is a "change election" context. This typically appears after a party has occupied the White House for at least two consecutive terms. Given that, this year's context will likely present a significant challenge to the Democratic candidate, regardless of which one eventually wins the nomination.

Below is the projected outcome matrix for this latest update:




February 9, 2016 - Sanders surges, model still projects Electoral College misfire for Clinton
The February update of my long-range forecast shows Senator Bernie Sanders surging, with the model projecting for the first time wins for him over all Republican contenders. Hillary Clinton, on the other hand, has shown no real change. The model still projects close Electoral College victories for her over most Republican candidates, but with a discrepant result between the popular vote and the Electoral College vote. Even so, the probability of a Republican defeating her in the Electoral College remains above 50% for all contenders except New Jersey Governor Chris Christie. Ben Carson is the only Republican candidate who is projected to defeat Clinton outright, with an 84.7% probability of winning an Electoral College majority should he become the eventual nominee facing Clinton.

Bernie Sanders, however, is projected by the model to have a much clearer path to victory, with only a discrepant popular vote/Electoral College vote result projected against Donald Trump. While both Clinton and Sanders continue to lead Republican candidates in the national head-to-head matchup polls, the question remains whether or not those leads will be enough to overcome what is likely to be a "change election" context. What these results here suggest is that Sanders may be better positioned than Clinton to take advantage of the level of voter discontent with the status quo that is likely to be present as a term-limited president leaves office. It is, of course, difficult to say at this point if this is in fact true. What is true, however, is that Sanders' lead in the polls over the Republican field in January was slightly higher than that for Clinton. Either way, the model still projects a fairly close election regardless of who the eventual nominees will be.

The other development is Florida Senator Marco Rubio fading back to the middle of the pack again. Last month's update showed him as the Republican most likely to win in November. That is no longer the case with this latest update. One should note that this projection is based on data that was out prior to the February 6th New Hampshire Republican Debate from which Rubio was criticized for parroting the same line about President Barack Obama multiple times. Whether that will have any further effect on his standing in the polls remains to be seen in next month's update.

Below is the projected outcome matrix for this month's update:





January 17, 2016 - Rubio Emerges as Republican Most Likely To Win in November

The January update of my long-range forecast shows that Senator Marco Rubio is the Republican who would be most likely to win the general election later this year. Even though Hillary Clinton led all Republican contenders in the polls in December, her lead over Rubio was the smallest by less than 1%. Given the influence of the Regime Age variable and Rubio's home state advantage in the all-important battleground state of Florida, the model projects that should Senator Rubio be the Republican nominee he would have an 84.7% chance of defeating Hillary Clinton and a 96.8% chance of besting Bernie Sanders.

We also see an apparent reversal in Chris Christie's fortunes. Last month, Governor Christie held the lead spot among Republicans, but in the December polling data he's fallen well back again. There is, however, a caveat to reading too much into this. Not many pollsters include a Clinton v. Christie matchup in their surveys, most likely because they assume Christie has little chance of actually winning the GOP nomination. Therefore, it's not clear whether this movement from last month's projection was due to actual movement in the polls or simply the vaguaries of sampling error from one poll to the next.

Below is the complete matchup matrix generated by the model:



December 8, 2015 - Carson Fades. Sanders Surges, Christie on Top but Democrats' Overall Chances Improve

With 11 months to go before the election, my long-range forecast model shows a shift in the landscape. Last month, Ben Carson was the clear frontrunner according to the model. He was projected to have over a 90% likelihood of defeating both Clinton and Sanders. November polls suggest that Carson's standing has faded. The latest projection now shows that while Carson still leads both of the Democrats in head-to-head matchups of the popular vote, his win probability is substantially lower, resulting in a virtual coin-toss with Clinton and just over a 60% probability of defeating Sanders.

The other noticeable development is Sanders' rise in the polls. Last month, the model only projected that Sanders would be able to defeat Ted Cruz. With this update, the model projects that Sanders would also be able to defeat Jeb Bush and Carly Fiorina as well. In addition, the model also presents scenarios that Sanders could benefit from an Electoral College misfire against Carson, Marco Rubio, and Donald Trump. Last month's projection only suggested that Hillary Clinton could possibly be the beneficiary of the technicality of the Electoral College where a candidate can lose the popular vote but win an Electoral College majority.

The final new development in the model is the inclusion of New Jersey Governor Chris Christie in the matchups. There was not sufficient data for Christie to include him in last month's projection. This month, the data show that he is the only Republican candidate with a clear chance of defeating either Clinton or Sanders. However, the overall picture that this update shows is that the probability of a Republican victory is now substantially lower than it was a month ago.

Below is the full matchup matrix:


As before, there are the same caveats to trying to read too much into this data. The bottom line in all of this is that the race remains very close. Regardless of who the final matchup ends up being, the model suggests that it will be a fairly contentious election.




November 6, 2015 - My newly developed long-range presidential election forecast model is designed to predict the outcome of presidential elections up to a year in advance of the election.

The model, which is based on state electoral histories, national polling data as well as a variable which attempts to estimate next year's election context, predicts that it is more likely than not that a Republican candidate will win next year's presidential election. Given that we do not know yet who will be the nominees, the model generates an expected outcome matrix for different possible matchups amongst the leading candidates for both parties.

I recently presented this model and forecast at the Iowa Conference on Presidential Politics. The model is fairly straightforward. It generates forecasts of state-level outcomes based on 4 variables:
  • The state's result from the previous election
  • The average of all national head-to-head matchup polls taken in October of the year prior to the election
  • A home state dummy variable
  • A regime age variable measuring how many terms the party currently occupying the White House has done so
Based on the state-level predictions, I can then aggregate up to the national-level to create predictions of the national popular vote and Electoral College vote totals. The results of those aggregations are presented in the outcome matrix below:


Caveats
It is important to note that this is a long-range forecast. A lot can happen between now and Election Day. It is more indicative of what the election context will be like next year, rather than how a specific election matchup will end up. A major factor in these results is the regime age variable. The basic premise behind this variable is the well-documented "time for change" effect (Abramowitz, 1988; Norpoth, 2013). The basic premise behind this variable is that the longer a party holds on to the White House the more difficult it becomes for them to hold on to it. Looking at history, you can see that this is very clearly the case. Since 1952, a party has only been able to win a third consecutive term once when George H.W. Bush won in 1988, and then was unable to win a second term in 1992. Given that it seems more likely than not that a Republican will be successful 2016, regardless of who that candidate will be. It is interesting to note that without the regime age variable in the model, it projects that Hillary Clinton easily wins against every candidate. Ultimately, one key factor in determining the outcome of the election next year will be whether or not the Democrats will be able to successfully make the case that they deserve a third term in the White House

Another factor that this model does not take into account is the presumed effect that a Hillary Clinton candidacy might have as the first female candidate for President. As the clear front runner for the Democratic nomination at this point, it is worth discussing the possible impact that her gender might have. If the historic nature of her candidacy can generate the same level of enthusiasm that surrounded Barack Obama's candidacy as the first African-American candidate, that could possibly counteract the effect of the regime age variable.

Note, however, that despite the fact that the model projects that most of the GOP candidates will win a plurality of the popular vote, and have a better than 50-50 chance of winning the Electoral College vote, it also projects that the most likely outcome in most matchups is a split outcome where the Republican candidate wins the popular vote, but Hillary Clinton wins the Electoral College by a very small margin.  The most certain thing we can conclude from this is that no matter who wins it will likely be a very close election.
You can read more about the model and prediction here:
State Electoral Histories, Regime Age, and Long-Range Presidential Election Forecasts: Predicting the 2016 Presidential Election (Note: The figures presented above differ slightly from those presented in the paper due to the fact that additional poll data were released following the preparation of this manuscript.

References
Abramowitz, Alan I. 1988. "An Improved Model for Predicting Presidential Election Outcomes." PS: Political Science and Politics. 21: 843-847.
Norpoth, Helmut. 2013. "Time For Change: A Forecast of the 2016 Election." Presented at the 2013 Annual Meeting of the American Political Science Association. Chicago, IL. August 29 September 1, 2013.