As you may have heard, there have been some changes at FiveThirtyEight recently. While it will be strange around here without our founder Nate Silver, his models and his oddly strong opinions about states, there are also some things that aren’t going away: namely, our commitment to rigorous data journalism and our mountains of polling data and trackers.
In fact, we’re planning on adding even more of those polling trackers. Today, we’re releasing new polling averages of three Republican presidential candidates’ favorability ratings. We’re also launching brand-new versions of our trackers of approval polls for President Joe Biden and Vice President Kamala Harris; national polls of the 2024 Republican presidential primary; and favorability polls for former President Donald Trump, former Vice President Mike Pence, former U.N. Ambassador Nikki Haley and Florida Gov. Ron DeSantis.
You might notice a few changes to the way these averages look. First, we’ve added error bands to all of the trendlines. (Previously, just our presidential approval tracker had them.) This is to emphasize that even an average of all the publicly available polls that meet our standards for inclusion isn’t a foolproof way of measuring public opinion — there is still some uncertainty. Trump might be polling at 52 percent in the Republican primary, but he might also be as low as 45 percent, or as high as 59 percent.
Another visible difference is that the lines, especially for our favorability ratings, are smoother. That’s because the new model we’ve written to calculate these averages optimizes each average for the specific thing being measured (favorability, approval, the horse race, etc.), rather than applying one approach to multiple sets of polling data. As a result, the trends are less noisy. However, at the same time, these models will ensure that the averages are more responsive to major campaign and news events when lots of new surveys are released around the same time.
There are also a host of smaller, technical differences between our old and new averaging models that I won’t get into here but that you can read all about on our detailed methodology page. Let us know if you have any feedback; we plan on fine-tuning the methodology in the coming months.
Despite these differences in approach, the outputs of our new polling averages look pretty similar to the old ones. That’s a good sign; if two independent models arrive at roughly the same number, we can be more confident that each is accurately summarizing public opinion. For posterity, here is where both the old and new averages stood as of June 27 at 8 p.m. Eastern. As you can see, all the shifts were smaller than 2.5 percentage points, except in Pence’s favorable rating, which was unusually low in our old average because he just got a bad poll (and our new averages are smoother).
While we’ve gone back and recalculated all our active1 polling averages according to the new methodology, you’ll still be able to access the old data if you wish. You can download spreadsheets from our GitHub page that contain the old FiveThirtyEight polling average for each day from the date we launched each tracker through June 27, 2023.
Finally, as I mentioned at the beginning, we’re also debuting some completely new polling averages today. We now have averages for the favorable and unfavorable ratings of former Arkansas Gov. Asa Hutchinson, businessman Vivek Ramaswamy and Sen. Tim Scott. Have fun obsessing over these averages! We’ll continue to roll out more in the coming months. And as always, if you have any feedback or requests, don’t hesitate to drop us a line.