York: Key state polls: Another mess in the making?

Who is leading the presidential race in the seven states considered most critical to victory in 2024: Pennsylvania, Michigan, Wisconsin, North Carolina, Georgia, Arizona and Nevada?

The answer is we don't really know. If you look at the RealClearPolitics average of state polls, you'll see that Pennsylvania is exactly tied between Vice President Kamala Harris and former President Donald Trump; in Michigan, Harris is up by 1.2 points; in Wisconsin, Harris is up 1.5 points; in North Carolina, Trump is up by 0.1 points; in Georgia, Trump is up by 0.3 points; in Arizona, Trump is up by 1.6 points; and in Nevada, Harris is up by 0.6 points.

Obviously, all those numbers are within the polls' margins of error. So anyone who cites them is advised to be cautious. But even more caution is called for when you look at the performance of state polls in the 2020 presidential election.

The purpose of a polling average is to avoid some of the errors of individual polls by grouping a bunch of them together. So instead of looking at one or two particularly accurate or terrible polls, look at how the group performed. But some averages are better than others, and all can be wrong if the underlying polls they include are wrong. The following numbers from the 2020 race were compiled in a new article from the liberal Brookings Institution and apply to two of the averages, RealClearPolitics and FiveThirtyEight.

In Pennsylvania in 2020, the final RealClearPolitics average had Joe Biden up by 1.2 points. The FiveThirtyEight average had Biden up by 4.7 points. The final result was Biden winning by 1.2 points — exactly what the RealClearPolitics average said. You can't get better than that. On the other hand, FiveThirtyEight was pretty far off.

In Michigan, the final RealClearPolitics average had Biden leading by 1.5 points, while the FiveThirtyEight average had Biden up by 7.9 points. The final result was Biden winning by 2.7 points — OK for RealClearPolitics but again way off for FiveThirtyEight.

In Wisconsin, the final RealClearPolitics average had Biden up by 6 points, while FiveThirtyEight had Biden up by 8.4 points. The final result? Biden won by 0.7 points. Both averages were terribly wrong, although FiveThirtyEight was more terribly wrong than RealClearPolitics. Both averages were polluted by a late-in-the-race Washington Post-ABC News poll that showed Biden leading by 17 points. It was the kind of error that should make pollsters retire in embarrassment. In any event, it contributed to a really inaccurate average of a critical state's polls before the election.

In North Carolina, the final RealClearPolitics average had Trump leading by 0.2 points, while the FiveThirtyEight average had Biden up by 1.8 points. The final result was Trump winning by 1.4 points. Neither average was miles off, except the FiveThirtyEight average got the winning candidate wrong.

In Georgia, the final RealClearPolitics average had Trump leading by 1.0 points, while the FiveThirtyEight average had Biden up by 1.2 points. The final result was Biden winning by 0.9 points — quite close for FiveThirtyEight, while the RealClearPolitics average got the winner wrong.

In Arizona, the final RealClearPolitics average had Biden up by 0.9 points, while the final FiveThirtyEight average had Biden leading by 2.6 points. The final result was Biden winning by 0.6 points.

Finally, in Nevada, the final RealClearPolitics average had Biden leading by 2.4 points, while the final FiveThirtyEight average had Biden up by 5.3 points. The final result was Biden winning by 2.4 points — another exact hit for RealClearPolitics but significantly off for FiveThirtyEight.

In the end, according to the Brookings calculation, the combined RealClearPolitics average missed a state's final result by an average of 1.1 points, while the combined FiveThirtyEight average missed the final result by 2.1 points. (After all this, you can see why my column relies on the RealClearPolitics average, and not others, in analyzing various races.)

Now look at those numbers and apply them to today's Harris-Trump race in the same seven states. Almost all of the poll averages are closer than the average error, even in the more accurate RealClearPolitics average and certainly in the less accurate FiveThirtyEight average. There's just no safe way to say who is leading right now, much less who will win in November.

Byron York is chief political correspondent for The Washington Examiner.