More on the latest polls by way of a quick lesson on public opinion. Over the past weeks as polls have tightened, it has been tempting to assume that people are changing their minds and moving their support from Clinton to Trump. It’s intuitive to believe that two candidates are moving closer to each other because one is picking up support from the other. Public opinion can in fact move like this if enough people undergo a change of heart about the candidates.

But public opinion also has a characteristic called intensity: how strongly people feel about something. In election polling, how strongly people feel about a candidate can be a good proxy for how likely they are to vote. For that matter, low intensity voters may be less likely to sit through a pollster’s questions, keeping their preferences from being recorded in the first place. And that may explain what we’ve been seeing in the September polling.

There’s good reason to believe Hillary’s September swoon is about changes in the intensity of her support, not wholesale movement from her camp to Trump. Two of the constants in this election season are a high degree of polarization and unprecedented levels of disaffection with the major candidates. These characteristics have combined to provide this election with an unusually high degree of stability, with Clinton maintaining a lead ranging from narrow to comfortable. In this environment, it is extremely difficult to change minds and get large numbers of voters to jump from one team to the other. But it is fairly easy for candidates going through a rough stretch to turn weak supporters into undecided voters and demoralize their partisans.

From late August, when she was fundraising rather than campaigning, through late last week, Clinton has faced negative press about her integrity and health, was the subject of intense scrutiny for her “basket of deplorables” comment, and lost the better part of a week recuperating from pneumonia. In a campaign where both candidates are widely disliked, periods of extended press attention correlate with falling poll numbers as voters are reminded of what they dislike about the candidate under scrutiny. Conversely, Donald Trump was exhibiting an unprecedented stretch of relative self control, giving uncertain Republicans reason to reconsider their discomfort with him.

During this period, pollsters also began reporting likely voter samples rather than (or along with) registered voter samples. Every polling outfit has its own way of estimating who is likely to turn out in November, but the ingredients in the likely voter sauce typically include some combination of past voting history (if you voted before you’re more likely to vote again), general knowledge about the election (for instance, knowing the location of your polling place), and intensity. If Hillary’s voters are demoralized, likely voter estimates based on voting intentions would produce a snapshot of the electorate with more Trump voters—a whiter, older electorate where Trump would understandably have a big advantage. This is reflected in polls where Clinton still leads comfortably among registered voters but is in a close contest among likely voters.

The thing to know about likely voter estimates is just that—they are estimates. As we move closer to Election Day we would expect these estimates to become more accurate, but they are only as good as their assumptions (they were close to the mark in 2008 but appreciably underestimated Obama’s support in 2012). I don’t mean to suggest these estimates are wrong. It is quite possible that Hillary will have difficulty reassembling the Obama electorate, and to that end recent polling is a clear signal to the Clinton campaign that it has to organize to win. In an election defined by voters swinging between both candidates, you win by changing minds, but in an election defined by partisan mobilization, you win by turning out your supporters. Clinton has made a massive investment in voter mobilization efforts and she will need them if she continues to struggle to excite her voters.

With the news agenda shifting last week to Trump’s birther comments and this week to possible fraudulent activity with the Trump Foundation, and with the undisciplined Donald Trump making a return appearance, we may be entering a period when Hillary’s supporters experience a renewed sense of enthusiasm. There is anecdotal evidence of this in new national and state likely voter polls showing wider Democratic margins reminiscent of late August. Regardless, the lesson for the Clinton campaign is clear. Surveys tell us that there are more potential Clinton voters than Trump voters, but it is up to the Clinton campaign to figure out how to deal with its intensity problem and get them to the polls.