Trump-vs.-Biden Polls: Don’t Be Fooled by Outliers!
The latest Rassmussen national poll on the 2024 presidential contest shows Donald Trump leading Joe Biden by ten points in a two-way contest, and by 12 points in a five-way race. Polls are not predictions, to be sure, but to place this data into an appropriate context, the last time a presidential candidate won the national popular vote by ten points was 30 years ago, when Ronald Reagan won 49 states. Everything we know about contemporary politics indicates the 2024 election will be close, just like all but one of the presidential elections in this century. So the odds are very high that the Raz poll is an outlier, which would hardly be a novel finding; FiveThirtyEight dropped Rasmussen Reports from its polling averages and even from its pollster rankings this year after the firm failed to answer questions about its methodology.
To cite a more generally applicable example: The last two batches of presidential-battleground-state polls were from Bloomberg–Morning Consult (mid-April) and The Hill–Emerson (late April). The former showed Trump leading in six of seven states (all but Michigan) in both two-way and five-way polls, and the latter showed Trump winning everywhere. If you weren’t paying close attention, it might appear that a modest trend toward Biden that began shortly after the State of the Union address had stopped and even reversed; there was some chatter along those lines from Trump-friendly parts of the punditocracy.
In reality, it was the identity of the pollsters rather than any fundamental trend that created this perception; the previous battleground-state surveys from Bloomberg–Morning Consult and The Hill–Emerson in March showed very similar findings. That doesn’t mean either set of data is “biased” or “inaccurate;” we don’t actually know what the “real” numbers are. It does mean that these polls should be assessed alongside similar polls from other outlets (in this case, CBS News) showing Biden doing significantly better. And instead of changing one’s perspective every time a new poll comes out, it makes more sense to look at polling averages that diminish the exaggerated importance of outliers. That’s particularly important in horse-race polling. Staring intently at a single poll can be beneficial in terms of understanding underlying trends, demographic-group preferences, and issue landscapes. But if the question is “Who’s ahead?” and the race is close enough to make precision matter, stick with the averages.
Polling averages aren’t infallible, of course. Their accuracy may depend on the database from which they are drawn, and some may well be “skewed.” The Rasmussen poll mentioned above, for example, is included in RealClearPolitics averages but excluded from those at FiveThirtyEight.
Overall, it’s reasonably clear that the Biden-Trump race remains very close in terms of the national popular vote with Trump maintaining a small advantage in winning enough battleground states to secure the 270 electoral votes necessary for victory. That’s also what other metrics like candidate favorability, partisan affiliation, and above all recent history would suggest. It should require some pretty clear indicators suggesting otherwise before anyone fairly concludes Trump or Biden is “winning,” particularly this far away from November. When you see someone leaping on individual polls or even short-term trends in averages to make bold and conclusive predictions, you should be aware you are in the presence of spin rather than analysis.