Today’s newsletter discusses recent polls and how many undecided voters there might be around Labor Day, which is often thought of as the traditional start of the campaign season. It encourages you to keep in mind both how polls ask their questions and what the pollsters are trying to show. I hope you find it interesting.
Thank you for reading,
Polls conducted at the end of August, prior to what many observers think of as the traditional time when voters start paying attention to election campaigns, showed Democrat John Fetterman having an edge over Republican Mehmet Oz in the race for Pennsylvania’s open US Senate seat. Despite agreeing that Fetterman had the lead, these polls differed about the size of that lead.
Franklin & Marshall’s August poll showed the largest advantage for Fetterman (45% to 36%) with nine percent of voters undecided and nine percent opting for one of the third-party candidates on the ballot. Emerson College and Susquehanna Polling & Research (SP&R) surveys conducted and released around the same date showed a four or five point race, with the undecided at two or three percent and the “others” at five percent. Why the differences?
Different questions, different answers
The three polls released at the end of August in Pennsylvania are quite different methodologically, which includes having different sampling and interviewing methods, which might, in part, explain the different results. But the most easily recognizable difference is in how these surveys ask about voting intentions. The F&M Poll question about voting intention includes a list of all the candidates on the ballot and the explicit option of saying “don’t know,” which is a different approach than SP&R and Emerson. Table 1 shows how the polls asked about voter preferences.
The consensus among survey practitioners is that survey participants’ responses to a question are frequently bound not just by the question asked but the responses offered. This means survey respondents usually answer within the response categories they are given. Consistent with this finding, the general sense is that respondents will use a “don’t know” response more frequently when it is provided as an option although there is no real consensus about when a “don’t know” response should be offered explicitly. These polls’ differences in the proportion of people who said they were either undecided or voting for a third-party candidate is probably the reason they showed different sized leads.
How many undecided voters should there be around Labor Day?
Those interested in the US Senate race, including the reporters covering it, are understandably uncertain about how to use these different poll results to characterize the race. While a nine point lead compared to a four or five point lead is not inconsistent given the sampling error associated with these polls, these different sized leads suggest races that are qualitatively different—a nine point win is a blowout while a four point win is a tight race. Which is it?
It seems likely that more than two or three percent of voters were undecided at the end of August. Post-election surveys of voters conducted as part of the American National Election Surveys between 1948 and 2004 show that an average of 17% of voters made their final decision about who they would vote for in those presidential races during the last two weeks of the election. Exit polls from 2016 put that number at 15%, although they also suggested it was much lower in 2020 at around six percent. If the norm in a presidential election is for a large share of voters to make up their mind in the last few weeks of a race, it seems unlikely that voters will be more certain about their choices in a Senate race that has no incumbent running for re-election.
The other reason it seems more likely than not that there is a larger share of undecided voters is because many voters are not sure how they feel about the candidates. The SP&R survey found about 17% of voters “didn’t know enough” to judge how favorably they feel about Fetterman or Oz, which is similar to the proportions in the F&M Poll. More to the point, data from our poll shows that 10% of voters did not rate either candidate favorably or unfavorably, meaning they were undecided about how they felt about them, and 11% had an “unfavorable” opinion of both candidates. So, about one in five voters doesn’t know enough about or doesn’t feel favorably toward either candidate, and they’ve already made up their minds?
Prediction, description, and uncertainty
US Senate races in Pennsylvania have been decided with an average Democratic margin of 2.7 points since 2000 (Figure 1 shows the final vote margins for the gubernatorial (G), US Senate (S), and presidential (P) races held in Pennsylvania since 2000), so it is likely that the 2022 race will end up as a close race that may in fact look similar to the race being predicted by the SP&R and Emerson surveys. But an important issue these surveys raise is whether they are describing the race as it is, prior to Labor Day, or predicting how the race will turn out. The low numbers of undecided voters suggests the latter, and also raises an equally important question about how many voters are unsure about their decision. There is no correct answer to these questions, except to say that those who are interested in the race should keep both of these ideas in mind whenever they try to assess the state of the race.
 See for example this story: https://www.inquirer.com/politics/election/pennsylvania-polls-senate-governor-fetterman-oz-shapiro-mastriano-20220825.html