By Berwood A. Yost

No candidate with negative personal popularity has won an election for U.S. Senate, governor, or president in Pennsylvania. Not one. That Donald Trump and Hillary Clinton are the most unpopular candidates ever to run for president is one more example of how unusual and unpredictable this campaign is and will be.

When people who follow politics closely talk about a candidate's popularity, they are normally speaking about that candidate's favorability rating. The unimaginatively named favorability rating comes from a simple question asking whether a voter has a "favorable" or "unfavorable" opinion about a candidate.

In campaign terminology, candidates viewed unfavorably by more voters than view them favorably have negative favorability scores. They are politically underwater.

The dual unpopularity of Trump and Clinton is unusual. In Pennsylvania, only two presidential candidates since 1996 have had negative favorability ratings in the spring prior to the election: Bob Dole and Mitt Romney. In fact, only three candidates have been underwater when considering all U.S. Senate, gubernatorial, and presidential races since 1996. In addition to Dole and Romney, Tom Corbett had a negative favorability score in spring 2014.

Dole, Romney, and Corbett all lost, of course, and this is why political watchers care about favorability ratings: They help predict winners and losers. Normally, candidates with negative ratings lose.

But there is another, less-remarked-on fact about favorability ratings to keep in mind: They are neither fixed nor permanent. Favorability ratings almost always change during the course of elections.

Both Dole and Romney saw their favorability ratings increase during their fall campaigns, although they never made it to positive territory. George W. Bush and John McCain both saw their favorability ratings decline, while John Kerry and Barack Obama's positive ratings increased during their campaigns.

Favorability ratings change for two main reasons. First, campaigns will do their best to change their tone, moderate their positions, and attract more voters, in part by broadening their appeal to less partisan voters.

The second reason favorability ratings change is that the electorate changes. Few people vote during primary elections, so many voters aren't paying close attention to the candidates. And those who do vote in primaries tend to be more strongly ideological and partisan. It is because the electorate changes that campaigns change their messaging and approach.

It also happens that in contested primaries such as what we've just witnessed, partisans change their minds when a candidate they didn't support becomes their party's choice. The presidential election in 2008 provides one example. After the primary election in April, Obama had a 53 percent favorable and 21 percent unfavorable rating among Democrats in Pennsylvania. By October, his ratings among Democrats were 78 percent favorable and 13 percent unfavorable.

So there is some reason to believe that both Clinton and Trump are going to see their popularity improve, even though current thinking suggests otherwise - that Trump is too polarizing, and Clinton too well-known and opinions about her too firm, for either to improve.

The truth of this intuition won't be known until Election Day, but the history of Clinton's favorability rating in Pennsylvania is instructive. Her net favorable rating moved from minus 17 in February 2007 to minus 1 in August 2007, plus 27 in October 2013, and minus 11 in April 2016. Clearly, voters' perceptions of Clinton have changed despite her being well-known.

Perhaps what we need to keep in mind is that favorability ratings at the end of the campaign in early November are more important for predicting winners and losers than favorability ratings at the end of the primary season. One unanswered question at this point is what the fall favorability ratings will look like. Another unanswered question is exactly how many voters will still have unfavorable opinions of both candidates by Election Day, and what will those voters do?

Berwood A. Yost is the chief methodologist for the Franklin and Marshall College Poll, director of the Floyd Institute for Public Policy Analysis, and director of the Floyd Institute's Center for Opinion Research. byost@fandm.edu