Between October 28, the day Comey wrote his letter to Congress—publicized in a tweet at 12:57 pm by Jason Chaffetz (R-Utah), chairman of the House Oversight and Government Reform Committee—and November 3, Clinton’s chances of winning Florida and North Carolina fell from 66% and 65% respectively, to a coin-toss, about 50%, in each state, according to Nate Silver’s FiveThirtyEight blog. Much the same happened in Nevada, though she clung to a 52% chance there.

Meanwhile, nearly even races in Arizona, Iowa, and Ohio, where the site rated her chances at 49% 51%, and 50% last Friday, now seem to be long shots for her, with Trump given 71%, 72% and 67% chances of victory.

This data does not prove that Comey’s unprecedented action caused the shift—there are a bunch of important provisos I’m about to list—but it it is consistent with such a view. But the longstanding confidentiality rules that Comey breached are designed to avoid even the appearance that the FBI might be impacting an election. This data does leave that appearance. Voters have a right to be incensed about it.

The data I’m using, FiveThirtyEight’s “Chance of Winning” metric, involves a weighted averaging of available poll numbers. (Silver himself wrote about whether there was any Comey effect on Monday, and concluded that while Comey’s disclosure might not be a “game changer”—the site’s numbers were still predicting a Clinton victory— to suggest that it had “no impact” would be “going to far in the other direction.” His colleague Harry Enten, writing about the five very tight Senate races this year, noted that one reason Comey’s action could be “important” was that “even a small shift could have a big effect in a lot of races.”)

Now for the key caveats.

• In all cases, the trend lines show Trump gaining ground even before Comey’s letter so it’s possible it had no additional effect.
• I don’t know precisely how long it took for the Comey letter to begin to be reflected in the site’s numbers, and its representative told me its analysts were too “slammed” by the election frenzy to be interviewed, only referring me instead to Silver’s and Enten’s stories on the topic, linked above.
• The New York Times Upshot‘s “chance of winning” numbers show less dramatic swings than FiveThirtyEight’s do, with only two of these states—Ohio and possibly Iowa—showing likely movement from the Clinton to the Trump column.
• As my colleague Geoff Colvin noted yesterday, there is a strong argument to be made that “recent swings in polling data probably don’t reflect similarly wide swings in voter sentiment.” He contends that the FBI’s announcement and, indeed, even the surfacing of the Access Hollywood tape three weeks ago, “probably haven’t changed the race much.”

One last proviso is that all of these “chance of winning” figures (not just FiveThirtyEight’s) fluctuate far more than the actual poll numbers, which only seem to stray up and down within a narrow range, a few percentage points deep. A shift of just a few percentage points makes a big difference in one’s chance of winning, as these metrics are computed.

That may reflect, in part, a psychological reality: For many of us, a 25% chance sounds less likely—and a 75% chance more likely— than either really is. Incidentally, on Monday, Nate Silver put Trump’s overall chance of winning the election at about one in four—or “about the same chance as the Chicago Cubs have of winning the World Series.”

The bottom line is that, as Enten put it, with numbers this close, “even a small shift can have a big impact.” Though I don’t question James Comey’s integrity for a moment, the fact that this election (and multiple senate races) could be determined by his having publicized an inquiry into emails that, as he admitted, for all he knows have no significance, just gets more infuriating by the moment.