After over two months of growing poll numbers, Donald Trump has suddenly experienced a notable drop-off — visible in the Pollster polling averages as well as the Real Clear Politics average. My colleague Philip Bump has more.

So why are Trump’s polling numbers declining? As I’ve argued before, changes in people’s preferences almost always require changes in the information they’re receiving. For Trump, media coverage has been crucial in motivating his surge.  Now, media coverage is arguably motivating his decline.

Here’s an updated graph (through Sept. 24) of Trump’s national polling numbers (via Pollster) and his share of news coverage of the Republican presidential candidates, based on the social analytic tools of the firm Crimson Hexagon:

You can see the correlation. An initial surge in news coverage after Trump announced his candidacy on June 16 creates an initial surge in polling.  Then, further increases in news coverage coincide with increases in Trump’s polling numbers — likely in a self-reinforcing cycle. But right before his apparent decline in the polls, there was a decline in his share of news coverage.

The parallel trends in media coverage and polls are even clearer if I smooth out the daily wiggles in media coverage to isolate longer-term trends.

The correlation between those lines is 0.93 (the maximum possible is 1.0).

So why is Trump’s coverage declining? At least partially responsible is a series of events that have brought other candidates into the spotlight. For example, Fiorina’s share jumped on Sept. 10 when Trump criticized her appearance, and again after the Sept. 16 debate. Rick Perry’s and Scott Walker’s shares jumped when they each dropped out of the race. Carson’s share of coverage increased after his controversial remarks about Muslims. Rubio’s share of coverage shot up when he took on Trump.

Another theory would be that it’s not so much the decline in coverage but that the coverage itself became more critical. For example, both Matthew Yglesias and Jack Shafer have criticized my emphasis on news coverage by noting that coverage of Trump has often been negative. To them, it seems odd that negative coverage could help a candidate.

Here’s some data that speaks to this. Crimson Hexagon codes each news story as positive, negative or neutral.  So we can figure out whether stories that mention Trump are positive, negative or neutral. The graph below subtracts the percentage of stories that are positive from the percentage that are negative, comparing Trump to the average of other GOP candidates.

The graph poses a few problems for the notion that the negativity of coverage should be hurting Trump.  One problem, as I noted before, is that coverage of Trump isn’t so “starkly negative” (Shafer’s phrase) compared to the other candidates. It’s only somewhat more negative, at most.

The second problem is that the tone of coverage mentioning Trump hasn’t changed much in the past three week. He hasn’t faced increasingly critical coverage, but his poll numbers are declining anyway.

The third problem is that during this time period, the tone of coverage appears to matter much less than the volume.  Statistical models that try to account for the potential inter-relationships between media coverage and polls show that the volume of Trump’s coverage is helping to drive his poll numbers, and vice versa.  But the tone of coverage has no apparent relationship to poll numbers, once you account for volume.

That is, it matters more how much coverage Trump is getting than how favorable that coverage is. And, by the way, you know believed that back in 1987?

I learned a lot from that experience: good publicity is preferable to bad, but from a bottom-line perspective, bad publicity is sometimes better than no publicity at all. Controversy, in short, sells.

That is Donald Trump, in “The Art of the Deal.”

To be clear, my view isn’t that media coverage is the Sole Factor that Explains Everything about the Primary, of course. I’ve looked at the role of voter anger, for example, and how it may help candidates like Trump. And I’m not suggesting that the tone of media coverage is always irrelevant. Sometimes bad publicity is worse than no publicity.  Ask Herman Cain.

But voter anger hasn’t suddenly changed in the last few weeks and can’t help us understand the change in Trump’s poll standing. Instead, we have to look at something that is changing.

Just as news coverage helped create Trump’s surge in the polls, it appears to be helping create his decline. We’ll see if that continues.