Verified by 
Key points
- Uncertainty in political races is exacerbated by misinterpretations of margin of error and small differences.
- Presenting political analysis in distinct categories misleads us about the complexity of voter preferences.
- Composite political polls can provide reliable predictions for upcoming elections.
In political polling, when is a “statistical tie" not a statistical tie? How can a small percentage in the polls mean a large difference in the final results? What popular generalizations mask underlying influences that can affect the outcome of elections?
Most political commentators are not trained in psychology or statistics, so they are prone to perpetuate common misinterpretations of polling results and political preferences. In the next month, here is what we should watch out for.
1. The most frequent and frustrating mistake political commentators make is labeling any poll as a statistical “dead heat” if the difference between candidates is within the margin of error (MoE).
In fact, a dead heat is suggested only when the candidates are actually tied in the poll. If Candidate A is ahead of Candidate B by 49 percent to 45 percent in a poll with a MoE of 4 percent, that means the poll’s best estimate is that Candidate A leads by 4 percentage points. It’s that clear. The 4 percent margin of error does not mean the candidates are in a dead heat. It’s just as likely that Candidate A leads Candidate B by 8 percentage points as it is that the candidates are tied.
Measurement error should be noted but not misinterpreted.
2. Small differences in the polls are often dismissed as insignificant.
A small and consistent difference in the polls can lead to large differences in the voting booth. If Candidate A leads Candidate B by 3 percentage points over a period of two months, the 3 percent lead is meaningful. In a two-way presidential election, that percentage would result in a convincing popular vote win for Candidate A. Extrapolating from the turnout in the 2020 presidential election, a 3 percent national lead in the 2024 election would translate into a difference of more than 5 million votes.
3. Proportions are often presented as categorical and absolute—when they are actually marginal and relative.
The most dramatic example of this mistake is the depiction of “red states” and “blue states.” Although straightforward, this mistake warrants emphasis. If 54 percent of the voters in a state voted Republican, then that means 46 percent did not. The same can be said of demographic information. If Candidate B leads among voters without a college education, that doesn’t mean all such people support Candidate B. In general, human beings have a tendency to think and analyze categorically, which only emphasizes the need for guarding against conceptualizing a majority as the entire group.
4. Even with noticeable differences across polls, appropriate treatment of polling information can provide useful predictions.
One virtue of political polling is that there is a direct test of the polling, otherwise known as an election. Over a recent time frame, polls can be evaluated by comparing predictions to actual results. A corollary is that a composite of the most accurate polls in the last two election cycles can yield reliable predictions. Such a composite is like a meta-analysis in psychology, summarizing key findings across a set of peer-reviewed studies.
5. Ignoring data can lead to statements of false equivalence.
For example, political commentators often talk about the favorability and unfavorability of a candidate, subtracting unfavorable ratings from favorable, which then gives the net favorability rating. If Candidate A has a net favorability rating of –5, that is not good. But if Candidate B has a net favorability rating of –12, then that is worse. To dismiss this difference by saying that voters have unfavorable views about both candidates is to mislead by throwing away data.
6. A particular level of aggregation can be meaningful while also encouraging faulty understanding.
.jpg?itok=aGBMs7JB "Clay Banks/Unsplash")
States are important in elections because they form the basis of the Electoral College, but they are usually not appropriate groupings for understanding political preferences. Again, we do not have red states and blue states. Each state has a mix of Democrats, Republicans, and Independents, with a majority of Democrats in urban areas, a majority of Republicans in rural areas, and a mix in the suburbs. In terms of understanding different voting groups, the categories of urban, rural, and suburban are more informative than state boundaries.
7. Maps represent land mass and not people.
Since the introduction of the red-state/blue-state map in the year 2000, most people have learned about the deception of painting states red or blue and then showing a color-coded map of the United States to illustrate how majorities in each state voted. But it is still visually overwhelming to represent the two major parties in this country with two different colors: red and blue. Mountains, prairies, and fields of corn and wheat do not vote. Showing a map with these two colors superimposed on the different states can mislead viewers about the dominance of one party over another.
Final Words
These are not partisan matters. They are mistakes of numeracy and data representation that basic statistical reasoning can correct.
