Even Odds for Trump & Clinton: Is Trump a 'Spoiler' or a 'Game-Changer'?
It’s now inevitable that Hillary Clinton and Donald Trump will meet in the general election this fall. Who will win? This, of course, is the million-dollar question.
To begin, let’s be wary of pundits or forecasters making definitive predictions at this point in time. Regardless of their experience or expertise, they are seriously understating the uncertainty of this electoral cycle. We believe a healthy dose of skepticism is fundamental right now.
As election forecasters, we at Ipsos typically analyze two key pieces of information when assessing election odds: (1) a “base-rate model,” which is an aggregation of past election outcomes paired with simple variables; and (2) a model based on polls. We do of course utilize other pieces of information, but these two are central to our thinking about an election (for a further explanation, see page 22 here).
Base-rate models normally include the aggregation of multiple past elections and give us an idea about outcomes in elections similar to the one at hand; this is our starting point. Poll-based models, in turn, aggregate the existing polls at hand; we also often adjust them by other factors such as the confidence we have in the methodology employed or in the polling firm conducting the survey.
Both types of models can be expressed in probabilities. Normally, we start off assessing the relative odds with the base-rate model at the early stages of the electoral cycle with little weight to polls (which are very poor predictors far out from Election Day). And then as we get closer to the election, polls take on a greater weight in our overall assessment. Simply put, our assessment or forecast is a weighted average of the two inputs, which can be adjusted over the course of the election cycle.
Well, it should be, but this year these two key pieces of information are materially at odds with each other. This reduces our confidence in our own or any other prediction. Let us explain.
On the one hand, base-rate models, including our own, point towards a Republican victory; so strongly in fact that we wrote a piece in October 2015 titled “Two simple reasons a Republican will likely win in 2016” – and made an earlier point in May 2015 based on similar reasoning. Indeed, ours and other base-rate models suggest, on average, a 70 percent (or more) probability of victory for the party out-of-power (Republicans).
Such base-rate models don’t typically consider the specifics of candidates but rather the underlying political and economic fundamentals and focus on the probability of the government-linked candidate winning versus others. It is a very de-personalized look at an election, which normally is a very strong starting point for our assessment of an election outcome. Ipsos’ own base-rate model (which uses just two variables: incumbency and government approval rating) has accurately predicted elections around the globe on dozens of occasions.
Conversely, the poll-based models (our own included) show a clear Clinton victory.
Indeed, of the 164 polls conducted in May of 2015, just 18 have shown Trump in the lead! In probabilities, the models put a Clinton victory at 80 percent to 90 percent (see table below). This perspective is reinforced by a belief that Trump’s strong negatives and a favorable Electoral College map make Clinton the clear favorite.
So how do we reconcile this difference?
Normally, we would simply take the average of the two, trusting the base-rate model slightly more than the poll-based model. In our experience, base-rate models outperform the polls. As such, we typically place more weight on the base-rate model because we trust them more than the polls.
So this approach lands us somewhere around 50 percent. At first blush, this might seem like a very middling prediction without “teeth.” However, it does fly in the face of most forecaster predictions out there right now, which give Clinton far greater odds of success.
Even so, it still seems simple, right? The odds still point to a probable, if closer than anticipated, Clinton victory.
Not so fast! In our strong opinion, this is an atypical election cycle that does not follow the normal “rules” or norms of election prediction, and which undermines models reliant on historical data to make predictions. Such a disruptive election means that many of our base assumptions go out the window, and in this case, we believe it yields a situation in which Trump still has a clear path to victory under certain conditions. Let us again explain.
A Disruptive Election
Our experience paired with an aggregated database of 500+ elections means we know that most elections (circa 85 percent) are fairly “cut and dried.” They fall along a simple and predictable continuum of public sentiment: “throw the bums out” (we call these change elections) versus “more of the same” (we call these continuity elections). Our models are simple: If people are relatively, measurably happy, they vote for the government-linked candidate; when not, they vote for the opposition.
However, in 15 percent of cases, this simple election dichotomy just does not work. We call these “disruptive elections.” In these cases, the political sands are shifting underfoot and the populace is highly agitated; these situations often occur in times of economic or population turmoil. In these cases, our ability to project an outcome is notably impacted because past behavior does not necessarily predict future behavior. And, correspondingly, our benchmarks say little our future outcomes.
Sometimes such turmoil might be only a momentary instability that opens the space for disrupters. Here a good example is the Five Star Movement led by Beppe Grillo in Italy 2013 or Marina Silva in Brazil 2010; they didn’t break the system, but they were “spoilers” in the end. In other cases, such scenarios can have long-term consequences for party and preference realignment. Here an example is Evo Morales in Bolivia in 2005 – an outsider “game-changer” who basically overturned the existing political order.
We believe the 2016 US presidential election is one of these “disrupter” elections, falling into the 15 percent of cases that are very difficult to predict.
So what is a forecaster to do?
Instinctively, we’d assume we should give greatest weight to the most recent, independent information (the polls) and down-weight the base-rate models. It seems that the market in general has coalesced around this point of view.
The problem is that, in these disruptive elections, voters are so angry that they are willing to blow up the system rather than work within it. Critical here is that analysts and pundits typically fail to see this coming, since we are already part of the system (as educated elites) rather than outside it (as populists).
Possible Trump Scenarios: Spoiler vs. Game-Changer
The key here, of course, is whether Trump is a spoiler or a game-changer.
- If Trump is a “spoiler,” he upends the Republican Party by creating ideological rifts, but he does not win the general election, appealing only to a substantial minority of voters. Put differently, the Republican Party selects a suboptimal candidate in Trump. Analytically, he “disrupts” our base-rate model, but the polls will fairly accurately reflect his likelihood of success (and so our calculations must down-weight the base-rate model vs. the poll-based models).
- If Trump is a “game-changer,” he will win the presidency. Contrary to pundit and analyst opinion, the Republican Party actually has selected the optimal candidate given the present political environment: an environment of angst and fear driven by nativism and belief that the system is broken and cannot be fixed within the traditional framework. Analytically, the base-rate model stands and the polls will catch up as we near Election Day (and so our calculations must down-weight polls and trust our base-rate model).
So Which Is It? Trump’s Milestones to Victory
In our opinion, in order to be more than a spoiler, Trump has to do three things:
- He must adjust his “nativism” rhetoric by watering it down to have broader appeal
Our comparative analysis of drivers shows conclusively that “nativism” or “America first” sentiment is the core driving feature of Trump supporters. What is critical to note, however, is that Trump can broaden his appeal among Independents by adjusting his rhetoric on this notion only a bit.
As illustrated below, over half of Republicans (56 percent) agree with more hard-line statements like “Immigrants take jobs away from real Americans,” and 70 percent agree with a more moderate version of this sentiment: “When jobs are scarce, employers should prioritize hiring people of this country over immigrants.”
It isn’t just Republicans who are on board this idea though! Two in five (41 percent) Independent voters agree with the more “hard-line” statement, and just over half (56 percent) agree with the “nativism lite” statement, which reframes the issue from a more protectionist standpoint. Watering down his hard-line nativism rhetoric could broaden Trump’s appeal beyond its current boundaries to improve his support and share of vote.
- Trump must pivot to being more credible than Clinton on the “system is broken” narrative
In many ways, Bernie Sanders’ and Trump’s messages are not dissimilar: Both champion the notion that the current system is rigged and unfair, and that left alone it will inevitably give rise to further inequality and unjustness. Their difference is framing, with Sanders positioning it on the liberal end as an issue with the banks, corporations and Wall Street, and Trump on the conservative end as an issue with immigrants. But underpinning both is the “broken system” narrative, which has great resonance with the electorate.
If one of the tenets of the American Dream is for our children to be better off than we were, Republicans are despondent and demoralized: three quarters (74 percent) think the next generation will be worse off. Critical to note is that two-thirds of Independents also agree with this idea. This fear of the future and conviction that the system is broken almost beyond repair has been successful for both Trump and Sanders.
If Trump “doubles down” on this idea, which runs counter to Clinton’s message that the country is “already great,” he could improve his standing among more moderate Republicans and even Independents (especially once Sanders is out of the race). If this comes along with leaning back slightly from blaming the problem squarely on immigrants (per Point 1 above), he’ll be sailing.
- Trump’s favorability ratings need to rise above 45 percent -- ideally closer to 50 percent
Candidate favorability or approval is a good way to get a sense of overall popularity. Trump currently has fairly low favorability ratings (high 30s to low 40s). In our view, the key question is whether this is a crystalized dislike led by groups of voters Trump has angered (e.g. women, Latino voters) or if this is only a partially formed perception driven by an overwhelmingly negative and acerbic primary cycle. Inevitably it is a mix of both, but our view is that the campaign effect is stronger than most realize, and that Trump’s favorability ratings will begin to recover. Indeed, while Clinton’s ratings have remained fairly steady in recent weeks, Trump’s have been climbing since April 1.
Independent of the above three points, we think this is a near-even election, with Clinton having a slight advantage. But, if the above three milestones are met, we think Trump could win. In order to continually update our prediction as the campaign unfolds, the daily Reuters/Ipsos poll will be tracking him against these milestones throughout the campaign:
- Tracking public perceptions of which candidate is seen as better on the nativist statements as well as the “protectionist” aspects of nativism
- Tracking public perceptions of which candidate is seen to be most in touch with the population on the “broken system” metrics, especially of interest once Sanders drops out.
- Tracking Trump favorability
Everyone else seems to think he’s a spoiler. We think that with a few adjustments to his approach, he could be a game-changer.